Bum

1!}? 1.. I .

.... H.1m..flkdw+“ \WJ.’
I. ,

.. “a.”
«an. ;x..1:,,....,fi 3am.
it; 15-1.}:
1:25;: .ga .

an an;

1.4. B

is ,

awn»...

”Inna-:3. .3
. . P5312. i... V .. .

any... .
g

8‘.-

 

H
3.13.33. . ‘ 2: .. «w
.iififinranh rams. v

. ‘ ‘ufltfi’ .3.

$111,. 5‘
3313.352... . in)... .
4.0.11: . a. ..... .u
ashaafififih
l 3‘ ‘1

 

xi .1!!!

51.)! girl?! In

a...) «VII. - ”iising:

|5520nxkp¥iiflnnvxavl r

.. 5 . ..n\.l.v:9.uur!o it"
1:35;;

 

 

  

0.)". a.
\Jiii‘iifllst:
stilt-8.13.1.3!
J.tiiisu.1.-s..fl.tl§
51.35.}; «7!.
.(ilit-ixzqithHV

y. I

 

 

 

 

 

 

 

"ES" LIBRARY
A s. MIcnIgan State
L’s-3‘0 University

 

 

 

This is to certify that the
dissertation entitled

A COMPARISON OF ORDINARY
LEAST SQUARE AND SPATIAL
REGRESSION MODELS FOR
ESTIMATING RECREATIONAL BOAT
OWNERSHIP IN FLORIDA

presented by

Yue Cui

has been accepted towards fulfillment
of the requirements for the

Community, Agriculture,
PhD degree In Recreation and Resource
Studies

 

 

M/I w
flu ]////c/ DIM
Major F‘rofessor’ 3 Signature

f0] C. «C; ‘UVfl/

 

Date

MSU is an Affirmative Action/Equal Opportunity Employer

 

 

 

 

 

PLACE IN RETURN Box to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE

DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

5108 ICIProj/Acc8PrelelRC/DateDue.indd

 

A COMPARISON OF ORDINARY LEAST SQUARE
AND SPATIAL REGRESSION MODELS FOR
ESTIMATING RECREATIONAL BOAT OWNERSHIP
IN FLORIDA

By

Yue Cui

A DISSERTATION

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILSOPHY

Community, Agriculture, Recreation and Resource Studies

2010

 

ABSTRACT
A COMPARISON OF ORDINARY LEAST SQUARE AND SPATIAL REGRESSION

MODELS FOR ESTIMATING RECREATIONAL BOAT OWNERSHIP IN
FLORIDA

By
Yue Cui

Recreational boating continues to be one of the nation’s most popular
recreational activities and it contributes substantially to the United States’ economic
well-being. However, the future growth in boating is threatened by various factors,
including changing demographics, the increasing cost of waterfront property for boating
facilities and services, and the conversion of facilities that provide public boating access
to other land uses and private access. Developing a better understanding of the current
and expected relationship between the demand for and supply of recreational boating
access in the context of social and demographic changes occurring in Florida, and better
understanding of the relationship with the capacity of government and private
businesses to develop and maintain boating access, is crucial for developing new
boating opportunities and maximizing the returns, including boater and public benefits,

from both private and public investments in boating access.

Over the course of the last fifty years, researchers have utilized “a variety of
different methods and approaches for describing, analyzing and forecasting recreation
demand, including the demand for recreational boating. The majority of recreational
boating demand studies have focused on participation, travel cost, benefit and value.
This research developed a zonal demand model to explore factors that affect boat

0Wnership and to develop models for predicting boat demand in Florida. A primary

 

purpose was to identify and test factors, including socioeconomic variables,
urbanization and boating opportunities that affect the ownership of boats in general and
of various types and sizes of boats at three different zonal levels. Zonal regression
models were developed to estimate recreational boat ownership, simulate changes in
boat ownership associated with demographic, economic, and policy changes, and
predict trends in boat ownership at county, zip code and census tract levels. Based on
OLS regression models, spatial regression models were subsequently developed to
understand spatial effects on zonal models and improve the predictability of the models.
The models demonstrated the importance of population-level socioeconomic factors,
urbanization and boating opportunity predictors in explaining variation in boat
ownership rates. The results show that population-level socioeconomics, urbanization
and the availability of boating opportunities are important factors in determining boat
ownership rates. In terms of R square and AIC, the county model has a better goodness-
of-fit than either of the two smaller zonal units models: zip codes and census tracts. This
SUpports the hypothesis that the large scale model predicts better than the small scale

models.

This study also determined that employing a spatial regression model to improve
model performance by partially removing the spatial effects. Spatial regression models,
including spatial error, spatial lag model, SAR and CAR models, were developed with
the same independent variable combinations as the corresponding OLS models (boat
type, size and zonal unit). The spatial regression models resulted in an overall

imPIOVement of the original OLS models, and a removal, to some extent, of the spatial

effects and associated heteroskedasticity.

 

Acknowledgement

First and foremost, I extend my profound gratitude towards my academic
advisor Dr. Ed Mahoney for his excellent academic guidance, and constant supporting
throughout the course of my doctoral program. I also appreciate my committee
members Drs. Daniel Stynes, Ashton Shortridge, Jon Batholic for their significant
suggestions and sincere encouragement throughout my study.

I deeply appreciate support from Florida Fishery and Wildlife Commission
(F WC) for providing boat registration data and National Marine Manufacture
Association (NMMA) for providing literature and data. Special thanks proffered to Dr.
David Harding from FWC and Mr. Carl Blackwell from NMMA.

I would also like to thank friends in Recreation Industry Research Center at
Michigan State University for their assistance during my doctoral program.

My family members’ unconditional support has been invaluable in my graduate
study. I express my deepest gratitude to my parents for their support and inspirations in
addition to their help in taking care of my children during this work. My dearest girls,
Tianying and Qianyun deserve special thanks for being wonderfiilly patient with me
during the whole program.

Finally, I am eternally indebted to my husband, Ren Zhang, for his persistence,

constant encouragement, and inspiration in and beyond this dissertation.

 

 

TABLE OF CONTENTS

LIST OF TABLES ..................................................................................................... ix
LIST OF FIGURES ................................................................................................... xi
LIST OF ABBREVIATIONS ................................................................................ xviii
CHAPTER ONE INTRODUCTION .......................................................................... 1
1.1 Introduction ........................................................................................................... 1
1.2 Problem statement ................................................................................................ 7
1.3 Research objectives ............................................................................................ 12
1.4 Study hypotheses ................................................................................................ 13
1.5 Organization of the dissertation .......................................................................... 14
CHAPTER TWO LITERATURE REVIEW ............................................................ 15
2.1 Recreational boating studies ............................................................................... 15
2.2 Factors affecting boat ownership and boating participation ............................... 17
2.2.1 National and Florida demographics and socioeconomic factors ..................... 18
2.2.2 Urbanization .................................................................................................... 24
2.2.3 Water and water access .................................................................................... 25
2.3 Models to estimate boat ownership demand and boating participation .............. 27
2.3.1 Recreation demand and recreation participation models ................................. 28
2.3.2 Issues associated with zonal regression models to estimate recreation

demand ...................................................................................................................... 30
2.3.3 Selection of functional form when developing zonal recreational demand
models ....................................................................................................................... 31
2.3.4 Zonal influence on the regression model to estimate recreational demand ..... 32
2.3.5 Spatial regression models ................................................................................ 34
2.3.5.1 Exploring regional variations ....................................................................... 34
2.3.5.2 Exploring spatial dependency ....................................................................... 36
2.4 Spatial and GIS approaches in tourism and recreation studies ........................... 38
CHAPTER THREE DATA AND METHODS ........................................................ 40
3.1 Identification and review of potential modeling approaches to estimate boat
ownership demand .................................................................................................... 41
3.2 Evaluation of alternative zonal structures ........................................................... 43

3.3 Identification of potential model variables and data sources, and data

preparation ................................................................................................................ 48
3.3.1 Dependent variables ......................................................................................... 48
3.3.2 Independent variables ...................................................................................... 57
3.3.2.1 Sources of data for socioeconomic and demographic variables ................... 58
3.3.2.2 Data source for urbanization ......................................................................... 59
3.3.2.3 Data source for proximity to water ............................................................... 60
3.3.2.4 Data source for boating access facilities ....................................................... 62
3.4 The modeling process ......................................................................................... 67

3.4.1 Describe the current recreational boat ownership patterns in the State of
Florida, including the spatial distribution of boat owners ........................................ 68
3.4.2 Identify and test variables that are statistically related to the ownership of all
boats and to various types and sizes of boats aggregated at county, ZCTA and

census tract level ...................................................................................................... 70
3.4.3 Develop and test the zonal OLS model to estimate boat ownership ............... 71
3.4.4. Development of alternative spatial regression models to estimate boat

ownership rates ......................................................................................................... 75
CHAPTER FOUR RESULTS .................................................................................. 79

4.1 Describe the current recreational boat ownership patterns in the State of

Florida, including the spatial distribution of boat owners ........................................ 79
4.1.1 Boat ownership by boat type and size ............................................................. 80
4.1.2 Boat ownership by gender ............................................................................... 80
4.1.3 Age distribution of registered boat owners ...................................................... 81
4.1.4 LifeMode segments of registered boat owners ................................................ 82
4.1.5 The proximity of registered boat owners to the coastline ................................ 85
4.1.6 Urbanization segments by registered boat owners .......................................... 86
4.2 Identify and test variables that are statistically related to the ownership of all

boats and to various types and sizes of boats ........................................................... 89
4.2.1 Age ................................................................................................................... 93
4.2.2 Gender .............................................................................................................. 93
4.2.3 Race ................................................................................................................. 94
4.2.4 Household structure ................................................. . ....................................... 95
4.2.5 Household income ........................................................................................... 96
4.2.6 Seasonal homes ................................................................................................ 97
4.2.7 Urbanization .................................................................................................... 97
4.2.8 Proximity to water ........................................................................................... 98
4.2.9 Marinas ............................................................................................................ 99
4.2.10 Boat Launch Sites ........................................................................................ 100
4.2.1] Examine MAUP and its effects on correlations .......................................... 101

vi

4.3 Development of OLS models that predict boat ownership in Florida at the

county, ZCTA and census tract levels .................................................................... 102
4.3.1 Model significance ......................................................................................... 104
4.3.2 Model fitness ................................................................................................. 104
4.3.3 Assessment of model residuals’ bias ............................................................. 105
4.3.4 Access of residual heteroskedasticity ............................................................ 106
4.3.5 Test for spatial autocorrelation ...................................................................... 106
4.3.6 Identification and evaluation of the OLS model coefficients ........................ 110
4.3.6.1 Age .............................................................................................................. 110
4.3.6.2 Gender ......................................................................................................... 110
4.3.6.3 Race and ethnic factor ................................................................................ 111
4.3.6.4 Household structure .................................................................................... 111
4.3.6.5 Income ........................................................................................................ 112
4.3.6.6 Seasonal housing units ................................................................................ 112
4.3.6.7 Urbanization ............................................................................................... 113
4.3.6.8 Boating facilities and water resources ........................................................ 113
4.4 Develop spatial models based on OLS models ................................................. 115
4.4.1 Model fitness ................................................................................................. 118
4.4.2 Examination of residuals ............................................................................... 118
CHAPTER FIVE SUMMARY AND IMPLICATIONS ........................................ 121
5.1 Summary of the research hypothesis and objectives ........................................ 121
5.2 Limitations ........................................................................................................ 125
5.3 Research recommendations .............................................................................. 129
5.3.1 Recreational boating demand study ............................................................... 129
5.3.2 Zonal structure and design ............................................................................. 130
5.3.3 Spatial weight matrix ..................................................................................... 130
5.3.4 Extend the spatial regression modeling method to other studies ................... 131
5.4 Implications ...................................................................................................... 132
5.4.1 Management and economic implications ...................................................... 132
5.4.2 Methodological implications ......................................................................... 134
APPENDICES ........................................................................................................ 13 7
A ............................................................................................................................. 138
B .............................................................................................................................. 146
C .............................................................................................................................. 155
D ............................................................................................................................. 158
E .............................................................................................................................. 161
F .............................................................................................................................. 166
G ............................................................................................................................. 170

H ............................................................................................................................. 175

I ............................................................................................................................... 182
J ............................................................................................................................... 192
BIBLIOGRAPHY ................................................................................................... 223

viii

Table 3-1
Table 3-2
Table 3-3
Table 3-4
Table 3-5
Table 3-6

Table 3-7
Table 3-8
Table 3-9

Table 3-10

Table 4-1
Table 4-2
Table 4-3

Table 4-4
Table 4-5
Table 4-6
Table 4-7

Table 4-8

Table 4-9

Table 4-1 0

Table 4-11

LIST OF TABLES

Census zonal units used in this study ....................................... 46
Number and types of registered boats in Florida by use types .......... 49
Number of registered boats by vessel (registration) type. . . . . . . . . ..50
Number of registered boats by types and sizes ............................. 54
The dependent variables used in this study ................................. 56
Independent variables employed to develop the boat ownership

models ........................................................................... 57
Dimensions of dependent variables .......................................... 70
The list of ZCTAs with the number of households less than 50 ........ 72
The list of census tracts with the number of households less than

50 ................................................................................. 72
The semi-natural logarithm transformed dependent variables used in
model building ............................................................. ..73
Gender of registered boat owners in Florida ................................ 80
Age distribution of registered Florida boat owners ........................ 81
Average age and median age of owners of different type and size of
boats ............................................................................... 82
Boats distribution by LifeMode segments .................................. 83
Proximity of the boat owners to the coastline ............................. 86
Boats distribution by Urbanization Segments ............................. 88

Correlation between boat ownership and their related
factors ............................................................................ 90

Summary of the results of the OLS models intended to predict the
rate of boat ownership for Florida counties, ZCTAs and census
tracts ........................................................................... 103

Moran’s I and p value for OLS models ................................... 107

Summary results of different types of Spatial Regression Models
predicting boat ownership .................................................. 1 16

The correlation coefficients between observed values and predicted
values of the OLS and spatial regression models at the county,
ZCTA and census tract levels .............................................. 1 19

ix

Table A-1
Table B-1

Table C-l

Table D-l
Table H-l

Table H-2

Table H-3

Table I-l
Table I-2
Table I-3
Table I-4
Table I-5
Table I-6

Table [-7

Table I-8

Table I-9

2009 TapestryTM LifeMode Segmentation Summary Table .......... 139
2009 TapestryTM Urbanization Segmentation Summary

Table ............................................................................ 147
Geocode Match Rates by County at Street Level and ZCTA

Level ............................................................................ 156
Number of Marinas and Boat Launch Sites by Counties ............... 159
Ordinary Least Square Models’ results for different type and size of
boats at county level ......................................................... 176
Ordinary Least Square Models’ results for different type and size of
boats at ZCTA level .......................................................... 178
Ordinary Least Square Models’ results for different type and size of
boats at census tract level ................................................... 180
Spatial Regression Model Results - LNABH (all boats) ............... 183
Spatial Regression Model Results - LNAPBH (all power boats). . . 184
Spatial Regression Model Results - LNASBH (all sail boats) ......... 185
Spatial Regression Model Results - LNCANOEH (canoes). . . . . . . . 186
Spatial Regression Model Results - LNPWCH (PWCs) ............... 187
Spatial Regression Model Results - LNSPBH (small power boats

with length less than 23 feet) ............................................... 188
Spatial Regression Model Results - LNLPBH (large power boats

with length 23 feet plus) .................................................... 189
Spatial Regression Model Results - LNSSBH (small sail boats with
length less than 23 feet) ..................................................... 190
Spatial Regression Model Results - LNLSBH (large sail boats with
length 23 feet plus) .......................................................... 191

Figure 3-1
Figure 3-2
Figure 3-3
Figure 3-4
Figure 3-5
Figure 3-6

Figure 3-7
Figure 3-8
Figure 3-9
Figure 3-10
Figure 3-11
Figure 3-12

Figure 3-13
Figure 3-14
Figure 3-15
Figure 3-16
Figure 3-17
Figure 3-18
Figure 3-19
Figure 3-20
Figure 3-21
Figure 3-22

Figure 3-23
Figure 4-1

LIST OF FIGURES

(Images in this thesis/dissertation are presented in color)

Elements of the Research Process ............................................ 40
Florida County Boundaries .................................................. 46
Florida ZCTA Boundaries ................................................... 47
Florida Census Tract Boundaries ........................................... 47

Distribution of Street-level Geocode Match Rates by County. . ..51

Point Map of the Distribution of Registered Boat Owners Residing

in Florida ........................................................................ 52
Distribution of Registered Boats by Counties in Florida... . .. . . . . .......52
Distribution of Registered Boats by ZCTAs in Florida .................. 53
Distribution of Registered Boats by Census Tracts in Florida .......... 53

Distribution of All Boats Per 1000 Households at County Level. ......55

Distribution of All Boats Per 1000 Households at ZCTA Level. ..55
Distribution of All Boats Per 1000 Households at Census Tract

Level ............................................................................ 56
Urban Areas in Florida ....................................................... 59
Coastline of Florida ........................................................... 60
Misalignment ofCoastline and County Boundary ..61
One Mile Buffer to Counties in Florida ..................................... 61
Distribution of Marinas in Florida .......................................... 62
Distribution of Boat Launch Sites in Florida ............................... 63
Number of Marinas by County .............................................. 63
Number of Boat Launch Sites by County .................................. 64
Distribution of Marinas and One-Mile Buffer to ZCTA 33884 .......... 66
Distribution of Boat Launch Sites and Three-Mile Buffer to ZCTA

33884 ........................................................................... 66
Model Building Diagram ..................................................... 68

LISA Cluster and Significance Maps for All Boats OLS Mode] at
Census Tract Level (Florida State) .......................................... 108

xi

Figure 4-2

Figure 4-3

Figure 4-4

Figure 4-5

Figure E-l

Figure E-2

Figure E-3

Figure E-4

Figure E-5

Figure E-6

Figure E-7
Figure E-8
Figure F-l
Figure F-2

Figure F-3
Figure F-4

Figure F -5
Figure F-6
Figure 6-1

Figure G-2

LISA Cluster and Significance Maps for All Boats OLS Model at

Census Tract Level (Ft. Lauderdale and Miami Region) ................ 108
Cluster and Significance Maps for All Boats OLS Model at Census
Tract Level (Jacksonville Region)...............................................109
LISA Cluster and Significance Maps for All Boats OLS Model at
Census Tract Level (Cape Coral Region) .................................. 109
LISA Cluster and Significance Maps for All Boats OLS Model at
Census Tract Level (Panama City Region) ................................. 109
Characteristics of boat owners by LifeMode Segments- All Power
Boats ........................................................................... 162
Characteristics of boat owners by LifeMode Segments- Small

Power Boats (<23') ............................................................ 162
Characteristics of boat owners by LifeMode Segments- Large

Power boats (23'+) ............................................................ 163
Characteristics of boat owners by LifeMode Segments- All Sail

Boats ........................................................................... 163
Characteristics of boat owners by LifeMode Segments- Small Sail
Boats (<23') ..................................................................... 164
Characteristics of boat owners by LifeMode Segments- Large Sail
Boats (23' +) ................................................................... 164
Characteristics of boat owners by LifeMode Segments- Canoes. 165

Characteristics of boat owners by LifeMode Segments- PWCs. ........165
Proximity of boat owners to coast line - All Power Boats... .............l67

Proximity of boat owners to coast line - Small Power Boats (<23')

and Large Power Boats (23'+) .............................................. 167
Proximity of boat owners to coast line - All Sail Boats. . . . . . . . . . 168
Proximity of boat owners to coast line - Small Sail Boats (<23') and

Large Sail Boats (23'+) ....................................................... 168
Figure F-S Proximity of boat owners to coast line - Canoes ............ 169
Proximity of boat owners to coast line - PWCs ........................... 169

Characteristics of boat owners by Urbanization Segments - All
Power Boats .................................................................... 171

Characteristics of boat owners by Urbanization Segments - Small
Power Boats (<23') ............................................................ 171

xii

Figure G-3

Figure G-4

Figure 0-5

Figure G-6

Figure G-7

Figure G-8

Figure J-l

Figure J -2

Figure J-3

Figure J-4

Figure J-5

Figure J-6

Figure J-7

Figure J-8

Figure J-9

Figure J-10

Figure J-ll

Characteristics of boat owners by Urbanization Segments - Large
Power Boats (23'+) ............................................................ 172

Characteristics of boat owners by Urbanization Segments - All Sail
Boats ............................................................................ 172

Characteristics of boat owners by Urbanization Segments - Small
Sail Boats (<23') ............................................................... 173

Figure G-6 Characteristics of boat owners by Urbanization
Segments - Large Sail Boats (23'+) ......................................... 173

Characteristics of boat owners by Urbanization Segments —
Canoes .......................................................................... 174

Characteristics of boat owners by Urbanization Segments — PWCs... 174

The distribution of residuals of Ordinary Least Square Model for
number of all boats per 1000 households at county level ................ 194

The distribution of residuals of Spatial Regression Model for
number of all boats per 1000 households at county level ................ 194

The distribution of residuals of Ordinary Least Square Model for
number of all power boats per 1000 households at county level. . . 195

The distribution of residuals of Spatial Regression Model for
number of all power boats per 1000 households at county level. . . 195

The distribution of residuals of Ordinary Least Square Model for
number of all sail boats per 1000 households at county level. . . . 196

The distribution of residuals of Spatial Regression Model for
number of all sail boats per 1000 households at county level .......... 196

The distribution of residuals of Ordinary Least Square Model for
number of canoes per 1000 households at county level .................. 197

The distribution of residuals of Spatial Regression Model for
number of canoes per 1000 households at county level. . . . . . . . . .. . . 197

The distribution of residuals of Ordinary Least Square Model for
number of PWCs per 1000 households at county level .................. 198

The distribution of residuals of Spatial Regression Model for
number of PWCs per 1000 households at county level .................. 198

The distribution of residuals of Ordinary Least Square Model for
number of small power boats (<23’) per 1000 households at county
level ............................................................................. 199

xiii

Figure J-12

Figure J-13

Figure J—l4

Figure J-15

Figure J-l6

Figure J-17

Figure J-18

Figure J-19

Figure J -20

Figure J-21

Figure J-22

Figure J—23

Figure J -24

Figure J-25

Figure J-26

The distribution of residuals of Spatial Regression Model for
number of small power boats (<23’) per 1000 households at county
level .............................................................................. 199

The distribution of residuals of Ordinary Least Square Model for
number of large power boats (23’+) per 1000 households at county
level .............................................................................. 200

The distribution of residuals of Spatial Regression Model for
number of large power boats (23’+) per 1000 households at county
level ............................................................................. 200

The distribution of residuals of Ordinary Least Square Model for
number of small sail boats (<23’) per 1000 households at county
level .............................................................................. 201

The distribution of residuals of Spatial Regression Model for
number of small sail boats (<23’) per 1000 households at county
level .............................................................................. 201

The distribution of residuals of Ordinary Least Square Model for
number of large sail boats (23’+) per 1000 households at county
level .............................................................................. 202

The distribution of residuals of Ordinary Least Square Model for
number of large sail boats (23’+) per 1000 households at county

level .............................................................................. 202
The distribution of residuals of Ordinary Least Square Model for
number of all boats per 1000 households at ZCTA level ................ 204
The distribution of residuals of Spatial Regression Model for

number of all boats per 1000 households at ZCTA level ................ 204

The distribution of residuals of Ordinary Least Square Model for
number of all power boats per 1000 households at ZCTA level ....... 205

The distribution of residuals of Spatial Regression Model for
number of all power boats per 1000 households at ZCTA level ....... 205

The distribution of residuals of Ordinary Least Square Model for

number of all sail boats per 1000 households at ZCTA level ........... 206
The distribution of residuals of Spatial Regression Model for
number of all sail boats per 1000 households at ZCTA level ........... 206

The distribution of residuals of Ordinary Least Square Model for
number of canoes per 1000 households at ZCTA level... . .. . .. ..........207

The distribution of residuals of Spatial Regression Model for
number of canoes per 1000 households at ZCTA level. . . . . . . . .......207

xiv

Figure J-27

Figure J-28

Figure J-29

Figure J-30

Figure J-31

Figure J-32

Figure J-33

Figure J-34

Figure J-35

Figure J-36

Figure J-37

Figure J-38

Figure J-39

Figure J-4O

The distribution of residuals of Ordinary Least Square Model for
number of PWCs per 1000 households at ZCTA level .................. 208

Figure J-28 The distribution of residuals of Spatial Regression
Model for number of PWCs per 1000 households at ZCTA level ...... 208

The distribution of residuals of Ordinary Least Square Model for
number of small power boats (<23’) per 1000 households at ZCTA
level ............................................................................. 209

The distribution of residuals of Spatial Regression Model for
number of small power boats (<23’) per 1000 households at ZCTA
level ............................................................................. 209

The distribution of residuals of Ordinary Least Square Model for
number of large power boats (23’+) per 1000 households at ZCTA
level .............................................................................. 210

The distribution of residuals of Spatial Regression Model for
number of large power boats (23’+) per 1000 households at ZCTA
level ............................................................................. 210

The distribution of residuals of Ordinary Least Square Model for
number of small sail boats (<23’) per 1000 households at ZCTA
level ............................................................................. 211

The distribution of residuals of Spatial Regression Model for
number of small sail boats (<23’) per 1000 households at ZCTA
level ............................................................................. 211

The distribution of residuals of Ordinary Least Square Model for
number of large sail boats (23’+) per 1000 households at ZCTA
level ............................................................................. 212

The distribution of residuals of Spatial Regression Model for
number of large sail boats (23’+) per 1000 households at ZCTA
level ............................................................................. 212

The distribution of residuals of Ordinary Least Square Model for
number of all boats per 1000 households at census tract level. . . . . . ....214

The distribution of residuals of Spatial Regression Model for
number of all boats per 1000 households at census tract level .......... 214

The distribution of residuals of Ordinary Least Square Model for
number of all power boats per 1000 households at census tract
level .............................................................................. 215

The distribution of residuals of Spatial Regression Model for

number of all power boats per 1000 households at census tract
level ............................................................................. 215

XV

Figure J-41

Figure J-42

Figure J-43

Figure J-44

Figure J-45

Figure J-46

Figure J—35

Figure J-36

Figure J-37

Figure J-38

Figure J-39

Figure J-40

Figure J-41

Figure J-42

Figure 143

The distribution of residuals of Ordinary Least Square Model for

number of all sail boats per 1000 households at census tract level. . ..

The distribution of residuals of Spatial Regression Model for

number of all sail boats per 1000 households at census tract level. . ..

The distribution of residuals of Ordinary Least Square Model for

number of canoes per 1000 households at census tract level. . . . . .

The distribution of residuals of Spatial Regression Model for

number of canoes per 1000 households at census tract level. . . . . .

The distribution of residuals of Ordinary Least Square Model for

number of PWCs per 1000 households at census tract level ..........

Figure J-46 The distribution of residuals of Spatial Regression
Model for number of PWCs per 1000 households at census tract

level ...........................................................................

The distribution of residuals of Ordinary Least Square Model for
number of large sail boats (23’+) per 1000 households at ZCTA

level ...........................................................................

The distribution of residuals of Spatial Regression Model for
number of large sail boats (23’+) per 1000 households at ZCTA

level ...........................................................................

The distribution of residuals of Ordinary Least Square Model for

number of all boats per 1000 households at census tract level. . . . .

The distribution of residuals of Spatial Regression Model for

number of all boats per 1000 households at census tract level ........

The distribution of residuals of Ordinary Least Square Model for
number of all power boats per 1000 households at census tract

level ............................................................................

The distribution of residuals of Spatial Regression Model for
number of all power boats per 1000 households at census tract

level ...........................................................................

The distribution of residuals of Ordinary Least Square Model for

number of all sail boats per 1000 households at census tract level..

The distribution of residuals of Spatial Regression Model for

number of all sail boats per 1000 households at census tract level. . ..

The distribution of residuals of Ordinary Least Square Model for

number of canoes per 1000 households at census tract level. . . . . .

xvi

216

216

217

...217

218

218

212

212

214

214

215

215

..216

216

217

Figure J-44

Figure J-45

Figure J -46

Figure J—47

Figure J-48

Figure J-49

Figure J-50

Figure J -51

Figure J-52

Figure J-53

Figure J-54

The distribution of residuals of Spatial Regression Model for
number of canoes per 1000 households at census tract level. . . . . . 217

The distribution of residuals of Ordinary Least Square Model for
number of PWCs per 1000 households at census tract level ............ 218

Figure J-46 The distribution of residuals of Spatial Regression
Model for number of PWCs per 1000 households at census tract
level ............................................................................. 218

The distribution of residuals of Ordinary Least Square Model for
number of small power boats (<23’) per 1000 households at census
tract level ...................................................................... 219

The distribution of residuals of Spatial Regression Model for
number of small power boats (<23’) per 1000 households at census
tract level ...................................................................... 219

The distribution of residuals of Ordinary Least Square Model for
number of large power boats (23’+) per 1000 households at census
tract level ...................................................................... 220

The distribution of residuals of Spatial Regression Model for
number of large power boats (23’+) per 1000 households at census
tract level .......................................................................... 220

The distribution of residuals of Ordinary Least Square Model for
number of small sail boats (<23’) per 1000 households at census
tract level ....................................................................... 221

The distribution of residuals of Spatial Regression Model for
number of small sail boats (<23’) per 1000 households at census
tract level ....................................................................... 221

Figure J-53 The distribution of residuals of Ordinary Least Square
Model for number of large sail boats (23’+) per 1000 households at
census tract level ............................................................. 222

The distribution of residuals of Spatial Regression Model for
number of large sail boats (23’+) per 1000 households at census
tract level ....................................................................... 222

xvii

AIC
CAR
DMA
FIPS
FWC
GIS
LISA
MAUP
MSA
NMMA
NOAA
OLS
PWC
SAR
VIF
ZCTA

LIST OF ABBREVIATIONS

Akaike Information Criterion

Conditional Autoregressive Model

Designated Market Area

Federal Information Processing Standard

Florida Fish and Wildlife Conservation Commission
Geographic Information System

Local Indicators of Spatial Association

Modifiable Areal Unit Problem

Metropolitan Statistical Areas

National Marine Manufacture Association
National Oceanic and Atmospheric Administration
Ordinary Least Square Model

Personal Water Craft

Simultaneous Autoregressive Model

Variance Inflation Factor

ZIP Code Tabulation Areas

xviii

Dependent Variables

ABH
LNABH

APBH
LNAPBH

ASBH
LNASBH

CANOEH
LNCANOEH

PWCH
LNPWCH

SPBH
LNSPBH

LPBH
LNLPBH

SSBH
LNSSBH

LSBH
LNLSBH

Number of all boats per 1,000 households

Natural Logarithm transformation of number of all boats per 1,000
households

Number of all power boats per 1,000 households

Natural Logarithm transformation of number of all power boats per
1,000 households
Number of all sail boats per 1,000 households

Natural Logarithm transformation of number of all sail boats per 1,000
households
Number of canoes per 1,000 households

Natural Logarithm transformation of number of canoes per 1,000
households
Number of PWC per 1,000 households

Natural Logarithm transformation of number of PWC per 1,000
households
Number of small power boats (<23 ft.) per 1,000 households

Natural Logarithm transformation of number of small power boats (<23
ft.) per 1,000 households

Number of large power boats (23 ft. +) per 1,000 households

Natural Logarithm transformation of number of large power boats (23
it. +) per 1,000 households

Number of small sail boats (<23 ft.) per 1,000 households

Natural Logarithm transformation of number of small sail boats (<23
ft.) per 1,000 households

Number of large sail boats (23 ft. +) per 1,000 households

Natural Logarithm transformation of number of large sail boats (23 ft.
+) per 1,000 households

xix

Candidate Independent Variables

AGEHP

MALEP
WHITEHP

WHITEHHP

BLACKHP

ASIANHP

HIHP

INClP

INC2P

INC3P

FAP

Percentage of households with householders between the ages of 35 &
64 maintaining households in a given geographic area (only one person
is designated from each household).

Percentage of male population in 2007.

Percentage of White households in 2007. A White household is defined
as separate living White quarters that are occupied by an individual or
group of people having origins in any of the original peoples of Europe,
the Middle East, or North Africa.

Percentage of White Non-Hispanic households in 2007. A White Non-
Hispanic household is defined as separate living quarters that are
occupied by an individual or group of people who indicate their race as
“White” but do not have any “Hispanic,” “Spanish,” “Latino,” origins.

Percentage of Black households in 2007. A Black household is defined
as separate living quarters that are occupied by an individual or group of
people having origins in any of the Black racial groups of Afiica.

Percentage of Asian households in 2007. An Asian household is defined
as separate living quarters that are occupied by an individual or group of
people having origins in any of the original peoples of the Far East,
Southeast Asia, or the Indian subcontinent.

Percentage of Hispanic households in 2007. A Hispanic household is
defined as separate living quarters that are occupied by an individual or
group of people who reported their origin as “Hispanic,” “Spanish,”
“Latino,” or other variations of Hispanic general terms without
identifying a specific country of origin.

Percentage of households with household income greater than $50,000 in
2007.

Percentage of households with household income greater than $75,000 in
2007.

Percentage of households with household income greater than $100,000
in 2007.

Percentage of family households in 2007. Family household is defined as
a householder and one or more other persons living in the same
household who are related to the householder by birth, marriage, or
adoption.

XX

Candidate Independent Variables (con.)

FAMP

FAMCP

FAFEP

POP_DEN

POP_RP

URBAN P

HOSP

WATERP
MARINA
MARINA 1 M

Percentage of married family households in 2007. A married family
household is defined as the householder and his or her spouse are
enumerated as members of the same household with or without children.

Percentage of married family households with children in 2007. A
married family household with children is defined as the householder
and his or her spouse are enumerated as members of the same household
with one or more children under 18.

Percentage of female headed households in 2007. This category includes
a family with a female maintaining a household with no husband of the
householder present.

Population Density. Population per square mile in 2007. Population
density is computed by dividing the total population within a geographic
entity (for example, United States, state, county, place) by the land area
of that entity measured in square kilometers or square miles. Density is
expressed as both “people per square kilometer” and “people per square
mile” of land area.

The percentage of rural population in 2007. Rural area consists of all
territory, population, and housing units located outside of urbanized
areas (UAs) and urban clusters (UCs).

The percentage of urban area in 2007. Urbanized Area (UA) is an area
consisting of a central place(s) and adjacent territory with a general
population density of at least 1,000 people per square mile of land area
that together have a minimum residential population of at least 50,000
people. The Census Bureau uses published criteria to determine the
qualification and boundaries of Urbanized Areas.

Percentage of vacant housing units seasonal in 2007. A seasonally vacant
dwelling is one that is used or intended for use only in certain seasons or
for weekend or other occasional use throughout the year. It includes
those used for summer or winter sports or recreation, such as beach
cottages and hunting cabins. Seasonal units also may include quarters for
such workers as herders and loggers. Interval ownership units,
sometimes called shared-ownership or time-sharing condominiums, also
are included here.

Percentage of areas with one mile’s distance to the coast line.
Number of marinas in 2007.

Number of marinas within 1 mile’s distance to the zone in 2007.

xxi

Candidate Independent Variables (con.)
MARINA3M Number of marinas within 3 mile’s distance to the zone in 2007.

MARINASM Number of marinas within 5 mile’s distance to the zone in 2007.
BOLA Number of boat launch sites in 2007.

BOLAlM Number of boat launch sites within 1 mile’s distance to the zone in 2007.

BOLA3M Number of boat launch sites within 3 mile’s distance to the zone in 2007.

BOLASM Number of boat launch sites within 5 mile’s distance to the zone in 2007.

xxii

CHAPTER ONE INTRODUCTION

1.1 Introduction

Year-round enjoyment of the waterways, bays, lakes and rivers makes Florida one
of the most exciting and vibrant boating locations in the country. It has been estimated
that more than one million recreational boats utilize Florida waterways each year, while
an additional 300,000 boats visit from other states and countries (Carver et al., 2007).
Recreational boating continues to be one of the state’s most popular recreational

activities. It contributes substantially to the state’s economic well being.

Recreational boating participation has experienced steady growth until very
recently. The National Marine Manufacturers Association (NMMA) estimates that almost
a quarter of all American adults participated in recreational boating at least once in 2005,
increasing to about 31% in 2008 (NMMA, 2008). According to national boating surveys
conducted annually by the NMMA, recreational boating participation in Florida has also
grown significantly. Florida has been among the high-participation states since 2004.
NMMA’s annual national recreational boating participation studies from 2004 to 2008
show that the recreational boating participation rate in the South Atlantic region, which
includes Florida, increased from 14.0% in 2005 to 19.5% in 2008. The number of
registered recreational boats is one indicator of the “demand” for recreational boating,
and generally parallels participation trends. Florida has led the nation in the number of
registered and titled boats since 2004. In 2007, the number of registered boats in Florida
reached 884,560; Florida’s p0pulation was 18,270,899 (U .8. Census Bureau, 2008a).
This means that, on average, approximately one in eighteen Florida residents owned a

registered boat, compared with one in twenty one people nationwide. Obviously, boat

1

ownership is one of the factors most strongly associated with boating participation.
According to the NMMA’s 2008 National Boating Participation Study, 38% of current
boating participants own at least one recreational boat. Over 60% of those participants

who did not own boats went boating on watercraft owned by their relatives or friends.

The purchase, upkeep (e.g., maintenance, insurance, storage) and use of
recreational boats generate spending and related economic effects (i.e., direct and indirect
sales, income and employment), which contribute substantially to Florida’s economy.
Recreational businesses, including boat and accessories manufactures, boat dealers,
marinas, and boat yard industries, are not only major employers in Florida, but they also

buy supplies and services from other Florida businesses.

According to a recently-completed recreational boating study in Florida,
conducted by the Recreational Marine Research Center (RMRC) at Michigan State
University (Mahoney et al., 2009), spending on recreational boating trips and craft-
related spending generated 97,000 jobs, $3.1 billion in labor income, $726 million in
indirect business taxes, and $5.3 billion in value added revenue in Florida in 2007. In
total, Florida registered boat owners spent $3.384 billion on trips and $5.16 billion on

craft-related expenses in 2007.

Changes, however, in the demographic and socioeconomic composition of the
population, the current economic and fiscal crisis, the limited supply and high cost of
acquiring water access, and the reduced participation and further-reduced boat ownership
in Florida and many other states are creating challenges for boating agencies and

businesses alike. The growth in boat ownership has slowed and, because of the current

recession, boat sales for 2009 are expected to be down significantly. The NMMA’s first-
quarter New Boat Sales Report (N MMA, 2009) indicated that power boat retail sales
were down 27% for the rolling 12-month period of the first quarter, compared to a year

ago .

A more troubling situation is that the United States population is becoming much
more diverse, while boat owners and participants are not. Women and ethnic minorities
are significantly under-represented among boating participants and owners, including
new entrants. Many studies show that higher income, middle-aged white males represent
the majority of boat owners. For example, the NMMA’s recently-released 2008 Boating
Statistical Abstract shows that 90% of boat owners are White vs. a 66% White population
at the national level. The household incomes of sixty-four percent of boat owners is over

$50,000, compared with 52% of nationwide household incomes over $50,000. Hispanic
boat owners comprise 6%, compared with a 15% Hispanic population nationwide. While
Florida is one of the fastest growing states in terms of overall population, it is also one of
the most diverse. For example, Smith (2005) characterizes Florida as a large, rapidly
growing state with relatively high proportions of Blacks, Hispanics, older people,

migrants from other states, and immigrants from abroad.

In addition, Florida, like many other states, has also become much more
urbaltlized. This has significant implications for boat ownership and participation. Florida
is now a leading state when it comes to urban sprawl. In 2000, 89% of Florida residents
resided in urban areas, compared with 79% nationwide. Geographic Research, Inc.
(2008) has forecasted that by 2013, Florida’s urban population would grow to 96%,

compared to 82% nationwide.

E1

Urbanization is a recognized factor in reducing involvement in more traditional
outdoor recreational activities such as boating, fishing and hunting (Arlinghaus and
Cooke, 2008). Persons living in highly urbanized areas have less exposure and fewer
opportunities to participate in resource-based outdoor recreation, such as boating. Boat

owners in urban areas often find it more costly to participate in boating, due to increased
travel time and restrictions/costs associated with storing their boats. Limited space and
regulations on boat storage (e. g., parking boat trailers on streets or in driveways) means
that oftentimes, boats cannot be stored where the boat owner resides. Additionally, the
high cost of land in urban areas, coupled with the high cost of doing business, limits the
amount of boating access available within and near to urban centers, forcing urban
residents to travel on congested roads to launch their boats. In some instances, the rapidly
increasing cost of waterfront property in urban areas, driven by competition between
alternative land uses (e. g., residential, commercial), is causing boat yards and repair

ShOps to go out of business, making it even more difficult for urban residents to get their

boats repaired and maintained.

Another challenge confronting boating-related businesses and those agencies that
SUpply boating access (e. g., launch sites) is the spatial imbalance between the “demand”
for boating and the location of current facilities. Although Florida has 8,426 miles of
SaltWater tidal coastline, many coastal counties find it difficult to provide an adequate
Supply of boating access. According to Carver et a1. (2007), despite the fact that the
nunlber of boating facilities in Florida has increased over the past ten years, it still does
not adequately meet the needs of boaters. A July 4, 2004 article in the Miami Herald

l“epol‘ted that "... in Miami-Dade County, there are more than 50, 000 registered boats

competing for the use of fifty-six boat ramps at six marinas. " The article describes how

“ . . .on weekends and when the weather is nice there are situations with boats hitting

I

boats, and sometimes people hitting people. ’

The phenomenon referred to as “ramp rage” occurs when too many boat owners
want to launch their boats at the same time. Complicating the situation is the fact that
boats are now larger and more powerful than they were, even ten years ago, putting more
stress on boat (launch) ramps and creating more boater-boater and boater-other
recreational user conflicts. In many areas, boat ramps are not only over capacity,
especially on the weekends, but the waters near the ramps are highly congested. It is not
possible to physically relocate the existing boat ramps and move them to areas of greater

need, and closing existing facilities for the purpose of reallocating operating monies to
other areas is most often politically impossible. Unfortunately, the spatial distribution of

existing boating access was often done without scientific tools for projecting future
den‘land.

The other problem facing boat owners is that not all ramps are open to the public.

Many ramps attached to private marinas are not available to the public—their use is

limited to'their owners or to members of exclusive marinas and yacht clubs. The Florida
Boating Access Inventory Study (Mahoney et al. , 2009) reported that Florida had
approximately 3,000 boat launch sites in 2007, but only about 1,466 boat launch sites

Statewide were operated by public agencies, and were explicitly for public use.

The gap in supply and demand is not limited to boat ramps. Bell (1990) indicated

that Saltwater marinas will not be able to absorb projected boater demand both because of

environmental constraints on wet slip expansion and the competition for land generated
by condominiums and other non-water-dependent economic activities. This places added
pressure on other water access points. Even in areas where there is affordable waterfront
land for marina development, many contend that the combination of federal, state and
local policies, which are often inconsistent, makes it difficult and very costly to gain

approval for proposed marinas.

Waterfront docks at permanent and seasonal homes provide another source of
boating access. However, riparian access receives additional regulatory scrutiny. For
exarnple, environmental, navigation and even visual regulations and standards are making

it much more difficult to build and even to maintain private docks in many areas of

Florida.

Creating a better understanding of the current and expected relationship between
the demand for and supply of recreational boating access in the context of social and
demo graphic changes occurring in Florida, in relationship with the capacity of
government and private businesses to develop and maintain boating access, is crucial for:
(1) developing and providing access to new boating opportunities that are responsive to
the Changing tastes and preferences of both existing and future boaters; (2) maximizing
the returns, including boater and public benefits, from both private and public

inVeStInents in boating access; and (3) evaluating existing and future needs for boating
acCess, as well as the implications of alternative spatial distributions of access. Both
agencies and industry would benefit from a tool that utilizes existing sources of data and

SPatial models to forecast boat ownership and boating participation.

Over the course of the last fifty years, researchers have utilized a variety of
different methods and approaches for describing, analyzing and forecasting recreation
demand, including the demand for recreational boating. The majority of recreational
boating demand studies have focused on participation, travel cost, benefit and value, etc.
Few scholars have addressed boat demand, boat ownership and the factors affect boat
demand. This study aims to develop a zonal demand model to explore factors that affect

boat ownership and to predict boat demand in Florida.

1.2 Problem statement
The majority of recreation demand-related studies focus primarily on estimating

the economic value of recreational opportunities/resources based on willingness to
trave l/pay. There are very few studies that begin by estimating the demand for equipment
and then look at how the equipment ownership affects participation. Factors that
influence boat ownership have not been the focus of rigorous academic research. Many
untested assumptions on the importance and influence of different factors are the basis of
indUStry marketing campaigns and programs designed to recruit and retain boaters. A
comprehensive review of the academic literature, dating back more than 40 years, failed
to identify a model for predicting boat ownership. The last identified study of boat
0Wl'lership was conducted by Bell (1995), in which he employed a regression framework
to Predict registered boats in Florida based on population and per capita income.
Obviously, boat ownership is a complex phenomenon involving not just the

ChaIacteristics of boat owners, but also their spatial relationship to boating opportunities

(3%» Water access). New advances in statistical analysis and technology have created

0PPOl’tunities for a more robust analysis and modeling of boat ownership.

Recreational boating participation is heavily influenced by the size and type of
boat a person owns. This includes what boating-related activities they participate in,
where they can keep their boats and, therefore, where they generally boat, and how much
they spend on their boats. Thus, it seems logical that models to estimate boating
participation would recognize and incorporate factors related to boat ownership.

However, this is often not the case.

The scope at which the demand for boating is estimated and predicted can vary,
ranging from national, to multi-state regions of the nation, to states, counties, and other
sub-regions of a state, to local communities, sites and projects (Haas et al., 2007). Boat
manufacturers and dealers regularly attempt to forecast demand for products and services
for specific market areas. For example, the market area could be a region, a state, a

Niel son DMA (Designated Market Area) or an area defined by the business itself.

Demand for proposed boating facilities and services (e. g., marinas, launch ramps) are
Often estimated as part of feasibility and retum-on-investment analyses. The problem
With many of these efforts to estimate and forecast demand is that they are based on a
relatively naive understanding of the multitude of different factors that influence boat

ownership.

In terms of spatial context, recreation demand models can be classified as either
diselggregate models or aggregate (zonal) model. Zonal modeling depends on the
availability and selection of homogeneous data and is commonly used in market
forecasting, economic analysis, resource mapping and impact assessment. Disaggregate

modeling is commonly used as part of recreation planning where the variables are more

easily defined and obtained (Baud-Bovy and Lawson, 1998).

8

Some of the earliest efforts at recreation demand modeling involved zonal
modeling. Clawson (1959), and Trice and Wood’s (1958) initial work on travel cost
demand models were estimated using zonal data (i.e., aggregate visit rates from
population zones at varying distances from the recreation sites). The difficulty of this
type of model, especially when it came to early efforts, is in (1) obtaining sufficient up-
to-date socioeconomic information about the recreationists and related regions; (2)
scientifically defining and mapping “demand regions”; and (3) combining/overlaying
different data to develop a holistic view of demand. Burt and Brewer (1971) provided the
first application of the travel cost method to site-level data when they estimated a system

of demand equations for recreation on a specific lake. Later, scholars, such as Hanemann
(1978, 1984, 1985), introduced the random utility model (RUM) as a method for

describing choice decisions among multiple individual sites.

Since the 1970’s, the majority of recreation demand studies have focused on
disaggregate demand estimations on the benefit—cost implications of management
Changes at particular sites. Typical disaggregate recreation demand models include the
TITrlvel Cost Method, Hedonic Price Analysis and the Contingent Valuation Method
(Loomis and Walsh, 1997; Freeman, 2003). Random Utility Models, the extension of

traVel cost models for single sites, are used to estimate the demand for a set of recreation

Sues based on users’ discrete choices at those sites.

Although they have received the most emphasis over the last 30 or so years,
disaggregate level models are limited in a number of important ways. First, disaggregate
modeling require the developers to collect information about users of the site(s), usually

t‘hrough some type of on-site survey(s), that become increasingly expensive to conduct.

9

Surveys are required both to develop and to recalibrate the models. To avoid various

biases, surveying must cover all relevant seasons and all kinds of combination for uses of
the site. Second, the recreational demand estimates (e. g., participation rates) produced by
individual-level models are “snapshots” (depictions) of the current demand for the site,
that don’t consider latent or future demand. Furthermore, it is difficult to transfer
disaggregate recreational demand to aggregated sub-regional, regional, or state-level
recreational demand, and this limits the capacity of the site-level model’s ability to be
integrated into the larger scale plans of recreation providers (e.g., agencies, organizations
and businesses). Although some researchers have developed and applied methods for
benefit transfer that generalize benefits from an individual-level estimate to the regional
level (Rosenberger and Loomis, 2001), there is typically a loss of detail/accuracy (i.e.,
user response to site changes captured in individual studies) in the benefits transfer

Process. The reliability of extending demand estimation from individual sites to larger

Spati a1 regions is problematic (Garber-Yonts, 2005).

In practice, recreation agencies require multi-scale levels of information for the
purpo ses of setting facility and budget priorities. While useful for specific purposes,
disaggregate models, in and of themselves, do not provide all the information needed to
maIlage complicated recreation delivery systems comprised of many different types of

Sites that are spatially distributed over regions or entire states. Different spatial
aggregations are necessary because of different issues and questions that arise for
dii:ferent areas, so planners generally require recreation-demand forecasts at different
1eVels of spatial aggregations. For example, the Florida Fish and Wildlife Conservation

C0Inmission (F WC) and Florida State Legislature require information regarding boating

10

participation on a state-wide level, a politically-defined spatial aggregation. In contrast, a
marina manager in Ft. Myers (Florida) cares very little about forecasts of recreational
boating demand for any political region, but rather wants information on his/her market
area. There is a benefit to developing a regional recreational boating demand model that

can aggregate data into multiple scales spatially.

Zonal (aggregate) demand models provide a solution for the above issues. In a
recreation participation study, a typical zonal model would aggregate recreation usage by
zone to estimate the proportion of the regional population that participates in a given
activity on an annual or seasonal basis (Ziemer and Musser, 1979). Using zonal-level
demographic information, trends in participation rates among population segments are
tracked. These trends are modeled using contemporaneous trends in demographic
Characteristics of the population, to facilitate projections of changes in participation rates
on the basis of regional demographic projections. This is not only done to provide the
measurement of recreation demand by multiple scales, but also to give some
measurement of the responsiveness of recreation demand to changes in other dynamic
faCtOrs. It not only reflects regional differentiation of recreation demand, but also
PYOVides greater integration of recreation demand models with other resource models,

6.8-, the economic impact model.

There are some important issues related to zonal models which must be identified
and addressed. For example, Sutherland and Ronald (1982) have raised concerns
regal‘ding the degree to which the size of the zone affects demand and benefit estimates.
Further, Bockstael et al. (1989) were troubled that heteroskedasticity is a major concern

in the development of zonal demand model. The majority of efforts to improve the

11

reliability of zonal demand models have focused on the choice of functional form,
selection of the independent variables that form the models, and the proper specification
0 f the variables. For example, the use of per capita or per household boat ownership,
rather than the numbers of boats in an area, e. g., visits, (Moeltner, 2003) and
transformations of the regression equations (e.g., Adamowicz et al. , 1989). The problem
is that these researchers did not realize that zone definition and the violation of regression
assumptions are all spatially related. Limitations in traditional statistics may not be a

good solution for the issues discussed above. Expanding the traditional regression model

to a spatial model could be a solution for the above problems.

1.3 Research objectives

The primary purpose of this study is to understand how factors affect boat
0VVIlership at different spatial scales and to develop models to estimate and predict

recreational boat demand in Florida, at various spatial scales.

The following more specific research objectives guided the design and
implementation of this study:

(1) Describe the current recreational boat ownership patterns in the State of

Florida including the spatial distribution of boat owners;

(2) Identify and test variables that are statistically related to the ownership

of boats in general, and to various types and sizes of boats at the county,

zipcode and census tract level;

(3) Develop and test alternative recreational boat ownership zonal models
using the Ordinary Least Square (OLS) technique for simulating changes

12

in boat ownership associated with demographic, economic and policy

changes, and for predicting the trends in boat ownership at multiple scales;

(4) Develop a spatial regression model to understand spatial effects on
zonal models, and to improve the ability to estimate and predict

recreational boat demand at various spatial scales in Florida.

1 .4 Study hypotheses

The following collection of hypotheses were tested as part of the process of
developing and testing models for estimating and forecasting recreational boat

ownership:

(1) Boat ownership propensity is influenced significantly by socioeconomic

variables including age, gender, race, household structure and household income;

(2) Urbanization (i.e., population density, ratio of urban population, ratio of urban
area) is negatively related to boat ownership because it limits boating

opportunities and increases the cost of owning a boat and going boating;

(3) The availability of and distance to boating access, including miles of coastline
and the number of launch sites, marinas, and seasonal homes, is significant in

determining boat ownership;

(4) The larger the size of the zones, the stronger the degree of correlation between
the rate of boat ownership and its explanatory variables. The scale of the zones
affect the performance and predictability of zonal boat ownership demand

models;

13

(5) The application of spatial regression can improve the ability to predict boat

ownership rates, if the spatial dependency exists in the residuals of corresponding

(i.e., same independent variables) OLS models.

1 .5 Organization of the dissertation

This dissertation is presented in five chapters. The second chapter is a review of
previous recreation boating studies and literature pertaining to factors affecting boat
ownership and participation, recreation demand analysis and spatial modeling issues. The
third chapter describes the research methods, with special attention to secondary sources
and the process for developing boat ownership demand models. The fourth chapter
presents the results of the model-building process and a comparison of the different zonal
boat ownership demand models that were formulated. The fifth chapter provides an
0Vel‘View of the model development process, a summary of the study hypotheses and
research objectives, recommendations and further development of boat ownership

de“land models, and conclusions regarding the utility and generalizability of the models

to other recreational activities.

14

CHAPTER TWO LITERATURE REVIEW

This chapter reviews pertinent literature dealing with (1) studies related to
recreational boating participation; (2) factors that affect demand and boat ownership; (3)
alternative approaches for modeling recreational demand; and (4) GIS and other methods
of spatial analysis for describing and analyzing recreation and tourism. The review
uncovered a few studies that have modeled the demand for recreational equipment or

employed spatial analytics and statistics to understand recreation participation.

2. 1 Recreational boating studies

Recreational boating has been studied from many different perspectives and
disciplines, with over half of all recreational boating-related literature focused on its
environmental impact (e. g., Johnson, 1994; Lipton, 2003; Minnesota DNR, 2004;

Widrner et al. , 2002; etc), boating safety (e.g., Howland et al. , 1996 and Molberg et al.,

1993), boating carrying capacity (e.g., Ashton, 1971; Sowman, 1987; Tarrant and
English, 1996), etc.

In terms of socioeconomic studies, a large proportion of published recreational
boating research is related to (1) boating expenditures and their related economic impact
(e-gu Lipton and Miller, 1995; Mahoney et al., 2009; Stoll, et al., 1988; Stynes, et al.,
1983); (2) boating and facility/site demand analysis (e.g., Bell 1990, 1995; Carver et al. ,
2007; Thomas 2002, etc.); (3) boat usage (e.g., Wu 1995); (4) boating trends and boater
demographics (e. g., NMMA’s boating-related studies and Responsive Management

COrnpany’s boating studies); and (5) boater behavior and boating characteristics (Lentnek

9’ al., 1969; Sidman and Fik, 2005, 2004, 2002; Sidman et al., 2000).

15

In many of the recreational demand-related models that were reviewed, boat
ownership is often used as a dummy variable in models designed to estimate travel costs
and choice sets. For example, Hunt and Ditton (2002) use boat ownership as an
exploratory variable to examine freshwater fishing patterns by different racial and ethnic
groups. Smith (1988) employs boat ownership as a dummy variable to explore methods
of travel demand models. Greene et al. (1997) also use boat ownership as a dummy

variable in a random utility model to predict recreational fishing demand in Tampa Bay,

Florida.
Literature on the demand for recreational equipment ownership and the
relationship between equipment ownership—especially boat ownership—and
Participation behaviors is limited. Kaiser (1970) conducted a study of multiple-boat
Ownership in Michigan. Hakim (1981) developed theoretical models to forecast the
demand for recreational boats in a metropolitan area and their expected usage in various
re:gional water bodies and marinas, and concluded that statistical regression analysis is a
theoretically sound method for projecting the facilities required for recreational boats.
Bell (1995) developed a demand model to estimate the number of registered boats in

F10I'ida counties and concluded that population and per capita income are positively

related to boat ownership. He also explored how boating-related expenditures affect boat

0Wnership.
Various research projects conducted to support the National Marine
Manufacturers Association’s Grow Boating Project (2007) explore factors affecting boat
0Wnership. NMMA has conducted a national study on recreational boating participation

and boat owners’ characteristics and behaviors each year since 2004.The results are

16

published annually in NMMA’s Recreational Boating Statistical Abstracts. Responsive
Management Co., a natural resource-focused consulting firm, has conducted a number of
surveys regarding outdoor recreation participation, including boating, over the course of
the last 18 years. For example, in 1995, they conducted a quantitative study about
“Factors related to hunting and fishing participation” which also provided boaters’
demographic information (Responsive Management, 1995). In 1998, they studied
women, Hispanics, and African-American boating participation, and their attitudes

towards boating and fishing (Responsive Management, 1998).

Other organizations, such as NOAA (National Oceanic and Atmospheric
Administration), RBFF (Recreational Boating & Fishing Foundation), USDA (United
States Department of Agriculture), the US. Forest Service, and the Recreational Marine
Re Search Center (RMRC) at Michigan State University, have also conducted studies on
reel‘eational boating participation and demand. F edler et al. (1998) conducted a study
regarding “Factors Influencing Recreational Fishing and Boating Participation” for the
Spol'tfishing and Boating Partnership Council. Fedler (2000) also produced a literature
review concerning “Participation in Boating and Fishing” for RBFF. Green and Cordell
(2003) produced a report on “Boating Trends and the Significance of Demographic
Change” for the USDA Forest Service in 2003. Leeworthy et al. (2005) has developed

‘Pl‘ojected Participation in Marine Recreation: 2005 & 2010” for NOAA.

2‘2 Factors affecting boat ownership and boating participation
The reviewed literature identified a great variety of factors that influence boat

OWTlership and boating participation. They include demographics, socioeconomic status

and economic factors.

17

2.2.1 National and Florida demographics and socioeconomic factors
Societal trends influence boating recreation, but none have as great an effect as
population growth and the changing composition of the US. population (McCool and
Clark, 2008). Overall, the US. population is growing but at a slower rate. Some regions
of the country, such as the Great Lakes and Midwest, are experiencing declining
populations caused by worsening economies, i.e., in 2008, Michigan lost 46,368
people—0.5% of its population—due to its bad economy (Tanner, 2008). On the other
hand, populations in other states continue to grow. Coastal areas continue to experience
significant population growth, which has resulted in the urbanization of the coastlines.
According to estimates released by the US. Census (2008a), among the nation’s ten
faste st—growing states, five were coastal: Georgia (fourth), Texas (sixth), Florida
(Seventh), North Carolina (eighth) and South Carolina (tenth). In addition, the population
is aging, the size and composition of families and households are changing, and the
retirement age is increasing (US Census, 2008b). Changes in the structure of the US.
e‘301'10my are, and will continue to have, a major impact on who can afford to own

reel’eational boats and on their ability to participate in recreational boating-related

actiVities.

The aging of the US. population is expected to have a significant effect on boat

ownership and participation in the future. Generally, age is negatively related to
recreational participation, particularity in high intensity activities such as water skiing.
On the other hand, America’s younger generation is not as involved in outdoor
reel‘eation, including water-based activities, as their predecessors (Minnesota DNR,

2006). Green and Cordell (2003) report that boaters and anglers tend to be concentrated

18

in the middle age ranges (31 to 50 years of age). A smaller proportion of boaters and
prospective boaters/anglers are 60 or older. Research conducted for the Recreational
Boating & Fishing Foundation (2003) found that anglers and boaters are younger than the

general population. Almost half (48%) of the American adult population is 45 years and

older, compared to 39% of anglers and 41% of boaters.

The NMMA (2008) National Boating Participation Study found that 67% of
recreational boat owners are between the ages of 31-64. A study by Responsive
Management (2002) revealed that 52% of boaters are between 35 and 64 and a study by

Milon (2000) found that 60% of the boaters in California and Oregon are between 36 and
65 years old.

However, the age distribution of owners is not the same across different types of
boats- Research conducted on behalf of Grow Boating (2007) shows, not unexpectedly,
that ovmers of inboard ski boats and PWCs are considerably younger than that of the
average boat owner. This may be due to the greater physical demands and reduced
paSsenger capacity corresponding to these boat types. Conversely, sail boat owners tend
to be older (51.3 vs. 47.5 years old for the average boat owner). In part, this is due to the

relatively low recruitment of new sailors and the aging of the existing owners.

A number of studies verify that recreational boating participants are comprised of

a higher proportion of males (Duda et al. , 2000; Fedler et al., 1998; Leeworthy et al. ,
2005). For example, Milon (2000) reported that 80% of boaters in the Pacific region were
male. Several authors suggest that the reason fewer women participate in fishing, boating

and other outdoor recreation activities in general is that they have been away from these

19

typically male-oriented activities (Henderson et al., 1988; Henderson, 1991). Lenskyj
(1 98 8) indicates that existing patterns of recreational participation are not necessarily
indicative of girls’ and womens’ so-called “natural interests or abilities,” but may well
refl eet a socialization process that streams females, from an early age, into activities
viewed by parents, teachers, and community members as conducive to femininity, or that

women become involved in organized outdoor programs designed specifically for them

(Lueck and Thomas, 1997; Thomas and Lueck, 1996).

Maj or barriers to boat ownership are the inability to afford the initial purchase
price, and the cost of boat storage (e.g., marina slips) and maintenance. Income, and
especially disposable income, is significant in determining how many persons own boats
and, therefore, participate in boating. Again, research for Grow Boating (2007) concluded
that household income is one of the most important considerations when targeting
Prospective first-time boat owners. There appears to be a minimum income threshold

Starting above $50,000. Nearly three out of four apparent first-time boat owners have an
ant111a] income of $50,000 or greater. Hagler Bailly, Inc. (1997) found that the incomes of
boating households are higher than those of non-boating households, and that relatively

few households with incomes below $40,000 per year can afford to own a boat.

Research suggests that while income is important, there is often a complex group
of Correlated characteristics (e.g., marital status, family structure, education) that
infllJence boat ownership propensity. According to the US Census Bureau (2000b),
households can be divided into quintiles according to their gross income, where each

quil‘ltile represents 20%, or one fifth, of all households. Married couples are

ChgPl'oportionately represented in the upper two quintiles, compared to the general

20

population of households, due to multiple income streams. Households headed by single
females are concentrated in the bottom three quintiles. Education, as expected, is another
factor correlated to household income. The US Census Bureau (2004) published
educational attainment and income data for all households with a householder aged
twenty-five years or older. The greatest income differential was between those with some

college education and those who had a Bachelor’s degree, with the latter making, on

average, $23,874 more annually.

Studies on behalf of Grow Boating (2007) showed that boat owners’ household
incomes varied considerably across different types and sizes of boats. Sail boat, inboard
Ski and stem drive/inboard boat owners had the highest median household incomes—
375 ,000 to $99,000—and approximately 80% of each of these groups earned $50,000 or
more annually. Alumintun fishing boat owners, on the other hand, were the least affluent
With a median income of $50,000 to $74,000, and only two-thirds (67.6%) earned over

$50,000 per year. The median incomes of those who owned runabout boats, fiberglass

boatS, and personal watercrafts were between $64,000 and $68,000, while the median

il’lcotne for owners of aluminum boats was $53,000.

Increasing racial and ethnic diversity is another major element of population
change that is likely to profoundly affect future boat ownership and boating participation.
Generally, the boating industry has chosen to ignore the low rates of boat ownership
alno11g minorities, focusing their recruitment and retention, instead, on “the low hanging
fruit”; in other words, wealthier middle-aged white families. Many studies have reported
differences in the recreation patterns of various racial and ethnic groups, particularly

an“Orig African Americans, Caucasians, and Hispanics. Research confirms that boat

21

 

owners and participants are disproportionately Caucasian (e. g., Duda, et al., 2000; F edler
et al., 1998; NMMA, 2008; Recreational Boating & Fishing Foundation, 2003; etc.) and
that relatively few non-white households own boats. Caucasian households are
sigrn'ficantly more likely than African American households to participate in water-based
or water-enhanced recreational experiences such as non-pool swimming, motor boating,
river canoeing, and primitive camping (Dwyer and Hutchinson, 1990). A demographic
study of boating conducted by the Responsive Management (Duda et al. , 2000) reported
that 86% of boaters are white, non-Hispanic. Milon (2000) similarly reported that 88% of
Pacific-region boaters are Caucasian. NMMA (2008) national boating participation
Studies reported that 87.5% of boat owners are white and only 6% are Hispanic. Although
research has shown that ethnic minorities, such as African Americans and Hispanics,
have positive perceptions of recreational boating and fishing, there are significant barriers
to participation and ownership, including knowledge of and access to opportunities
(Recreational Boating & Fishing Foundation, 2003). Hispanics, who live mainly in urban
ar- eas, are perceived as having less access to desirable boating and fishing locations.

These ethnic minorities generally earn lower incomes and many live in urban areas,

Where the cost of boating is higher.

The level of education appears to have little bearing on the amount of interest in
boat ownership. According to Peyton and Gigliotti (1989), and Stamps and Stamps
(198 5), there do not appear to be any major differences between the educational levels of
recreational boaters and non-boaters. However, in Duda et al. ’3 (2000) research, those
with higher levels of education were slightly more likely to participate in recreational

boating. Those with graduate or professional degrees (32%) and college graduates (36%)

22

were more likely to participate in recreational boating than those with some college
(28%), high school diplomas (24%), or no high school diplomas (28%). On the other

hand, education achievement is always associated with income. Education itself may not

be a good predictor for boat ownership.

Traditionally, recreational boating has been a family-related activity in the United

States- Families with children tend to be more engaged in various outdoor recreational
activities like boating, camping and fishing. Research for Grow Boating (2007) showed
that about 70% of households interested in buying a boat were families with children. Of
great concern to boating agencies and the industry is that family structure has changed
ominously in recent years. More and more American adults are opting to remain
unmarried. The tendency for people to marry at an older age and an increasing divorce
rate are contributing to an overall drop in the number of married couples. According to
the US Census Bureau (2008a), the percentage of married households decreased from

5 5 - 1 % in 1990 to 50.5% in 2008. Among married households, the number of families

With children is down from 25.6% in 1990 to 22.8% in 2008. Households are also

beCorning more diverse (i.e., childless couples, blended families, unrelated persons living

In a household), and this will certainly have an impact on future boat ownership.

Florida continues to be one of the fastest growing states in the nation. During the
1 990’3, the population rose by 3 million, nearly a 24% increase (Smith, 2005). During the
period from April 1, 2000 to April 1, 2006, the natural increase in Florida’s population
6 -e- , the difference between births and deaths). The US Census Bureau estimates that
10I‘ida’s population is expected to grow to 19,620,532 by April 1, 2010 (US Census

Bureau, 2008a). Florida’s population is aging at a rate greater than the country as a

23

whole because of migration. In 1900, the median age of Florida’s population was 20.4
years, younger than for the United States as a whole (22.9 years). According to the US
Census (2009a), by 2008 the median age had reached 30.9 years for Florida and 30.2
years for the United States. In 1980, 1,687,705 (17.3%) of Floridians were 65 or older.
The 2000 census reported 2,807,598 elderly, comprising 17.6% of the state’s population.
Florida’s 65+ population is projected to grow to 3,481,271 by 2010, an increase of 24.0%
over 2000. Florida’s population is also becoming increasingly Hispanic. The Hispanic
population increased by 70.4% between 1990 and 2000, and is projected to represent

22 - 0% of Florida’s population in 2010.

2 -2 -2 Urbanization

Urbanization can impact boat Ownership and boating participation in many
different ways. Generally, participation in boating activities decreases with increasing
urbanization, in part due to reduced exposure and opportunities. This is coupled with
hi gher costs of ownership and participation in urban areas. Households in less populated
towns/rural areas are far more likely to own a boat than the average household. In
c0l'ltrast, people in urban areas are far less likely to own a boat. Two national studies,
conducted by the American Red Cross (1991) and Hagler Bailly, Inc. (1997), found that a
g1-eater proportion of households in rural areas, small towns and cities contained boaters
thall households located in metropolitan areas. About 28% of the boaters lived in
njet-ropolitan areas, 44% resided in small town and cities, and 28% were in rural areas.
I{eSearch from Grow Boating (2007) reported that, for nearly every boat type evaluated,
individuals living in towns/rural areas were far more likely to own boats than those in

Inore populated areas (e. g., major cities in the US). Over half (57%) of first-time boat

24

owners resided in towns/rural areas, but they represented only about 36% of the total US

population.

The literature strongly suggests that urbanization creates issues when it comes to
boat ownership. Urban populations are more diverse, generally have lower incomes, and
must spend more time traveling to participate in boating. Also, waterfront development is

leading to conversion of more public boating access facilities to private uses (e. g.,

condos, retail development). This is especially true in Florida.

Florida is a highly urbanized state. Eighty-nine percent of its population lived in
urbaxl areas in 2000, compared to 79% nationally (US Census, 2009b). The conversion of
land from rural to urban use is more pronounced in Florida than in many other states.

LaIld in urban areas in Florida increased from 1.2 million acres in 1964 to over 5 million
acres in 1997. Today, the majority of the United States’ populace lives in Metropolitan
Statistical Areas (MSA). In Florida, there are 34 counties that comprise 20 MSAs; about

93 0/0 of the state’s population lives in counties classified as MSAs (Geographic Research,

1110-. 2008).

2'2-3 Water and water access

Accessibility to water, places to launch and to store boats are all constraints to
boat ownership and participation. There. is a growing concern that the supply of boating
aceess is inadequate to meet the future demands of recreational boaters. In many areas of
the country, there is insufficient in—water (e. g., slips, moorings) and on-land (e. g., dry
Statek) storage for boats; there are too few launch ramps; and there are increasingly

Corrlplex regulations and policies that restrict or limit boating on lakes, rivers and

25

 

offshore zones. According to the Water Access Alliance, recreational boating faces a
serious threat from limitations to adequate public access to waterways. Thom Dammrich,
president of the National Marine Manufacturers Association (N MMA), maintains that
“Americans’ ability to enjoy the many positive benefits of the boating lifestyle depend on
their access to the water. A stronger commitment to providing and maintaining
appropriate levels of access to the water begins with better information about the
avai lability and economic and social benefits of providing access to the water.” Ditton et
al. (2 003) indicated that inaccessibility, congested waterways, and poor water quality
were three of the main factors boaters take into consideration when deciding whether or

not to boat in Texas.

There is a great deal of variation in the types and sizes of boats kept in different
tYPes of storage (e. g., marina, waterfront home, etc.). The types of storage significantly
influence behavior, including the amount of use and spending. Stynes et al. (1998) argued
that Storage locations are the best predictors of where boats were used and explained the
tyI>es and amounts of use. Wu (1995) classified boats into marina, second home,
waterfront homes and non-waterfront homes as the basis for her models to estimate
bGating activity in Michigan at the county level. She determined that storage type is
highly correlated with crafi size. Boats stored at non-waterfront sites are primarily
Small ler craft that can be trailered to launch sites, while marinas provide storage for larger
DO"Vet boats and sail boats. Thomas and Straitis (2002) also used marina, ramp and

Waterfront home to categorize boats and boating trips in their boating demand study in

1:lol‘ida.

26

 

In addition, it is quite obvious that boating requires boating waters. The travel
distance, cost of, and accessibility to boatable water enters into decisions of whether to
own a boat, where to boat and how much to boat. Stynes (1982) found that 50% of Great
Lakes residents traveled less than 30 minutes to boat. The Florida Recreational Boating
Study (Mahoney et al., 2009) determined that 90% of boating trips occured within the
county where the boater resided Leeworthy (2001) discovered that participants in marine

recreation were much more likely to live in coastal counties and closer to water.

2-3 Models to estimate boat ownership demand and boating participation

In economic studies, demand is a function of prices for certain goods. However,
in the social sciences, and in recreation management and planning, the concept of
del'nand is somewhat ambiguous; it is defined differently in different contexts (Garber-

Yonts, 2005). According to Rosenberger and Loomis (2001), there are two quantifying

e1el'nents of recreational demand: participation and willingness to pay (WTP). When it
col”Tues to participation, recreation/activities days, visitor days/user days and user hours,
nutnber of trips/visits, entrance permits, licenses and tickets are used to estimate
rec31‘eation demand (Loomis and Walsh, 1997). Willingness to pay is more likely to be
ernli>loyed in cost and benefit analyses of recreation facilities and opportunities.
According to Garber-Yonts (2005), “economic values, political pressures from providers
and Users of (a) resource, public opinion and preferences expressed through public
involvement and in terms of sense of place, are variously referred to in shorthand as
demand.” Because recreation demand studies are done for many different uses no single
ruethod can address the full range of recreational demand issues, thus giving rise to a

Variety of approaches for estimating demand.

27

Price, or some proxy for price, is usually the principal variable in recreational
demand models; other conditions, such as other attributes of the recreational resource,
substitute opportunities and user demographics, are assumed to be constant. However,
fliese other conditions, particularly in the case of recreation, are neither constant nor
consistent (e. g., across geographic areas), and may actually be more important than price
in determining demand. In recent years, economic studies of non-market resource values

have concentrated on the implications of non-price attributes on

participation/consumption.

2-3- 1 Recreation demand and recreation participation models

As already mentioned, when it comes to scale, recreational demand models are
Classified as disaggregate individual level models and aggregate (zonal, regional, macro-
01‘ population-specific) models. Currently, the majority of recreational demand studies
inVolve the development of disaggregate models (e.g., participation rates at a particular

reCreation site, travel cost models and household production models).

One issue for disaggregate demand models is that sometimes there is only
aggregate data available, or there is data that can only be used in an aggregate form. For
exii-triple, census data are the most widely used source of demographic and
Socioeconomic data. Because of confidentiality considerations and other restrictions on
the availability of individual-level data, census data in the US are zone-based aggregate

data, reported for census-defined geographic entities, such as census blocks, census block

groups, and census tracts (U .S. Census Bureau, 2000a).

28

Many researchers have developed zonal-level (aggregate) models to estimate and
predict recreational demand. The US. Forest Service has developed a number of applied
national and regional assessments, detailing population-level trends and their relationship
to recreation behavior (Cordell, 1990; Cordell and Overdevest 2001; English et al. , 1999;
Gartner and Lime, 2000). The US. Forest Service’s 1989 assessment model estimated the
quantity of trips for a given activity, generated in aggregate at the regional level, as a
fiincti on of regional demographics, cost per trip, average suitability of sites available to
the community for that activity, and availability of substitute recreational activities
(Cordell, 1990). Projections of recreational demand up to 2040 were made, formulated on
U - S - Census projections of demographic change. These projections were used to analyze
alternative scenarios under which recreational demand and supply equilibrium would be
met in what was referred to as a “gap analysis”. Bowker et al. (1999) developed a series
of Statewide recreational demand models using both county and individual data. They
utilized coefficients from activity models, along with census-based projections of
dernographic characteristics and population growth, to develop participation and

consumption projections for 76 different activities.

Zonal models are not without their own issues and problems. A major concern is
that zonal models predict individual behavior based on aggregate data, and that
aggregation may distort or mask the true exposure/response relationship for individuals,
theI‘eby failing in systematic ways to take underlying heterogeneous behavioral reactions
into account. This is referred to as the ecological fallacy; it makes it difficult to assign a
rneaningful interpretation to estimated model parameters (Moeltner, 2003). Ecological

f . . . .
allaey may occur for many reasons, such as when exposure rs inhomogeneous wrthrn

29

areas and effects are nonlinear, or when a confounding factor leads to non-uniform
baseline rates or modifies the exposure/response relationship across areas (Greenland,

1 992; Richardson, 1992; Robinson, 1950;).

Sutherland (1982) raised considerations relating to the degree to which the size of
the zones affected demand and benefit estimates, and whether it was more appropriate to
use concentric zones or population centroids. His study revealed that consumer surplus is
usually overestimated when concentric zones were used, as compared to population
centroids. He suggested that benefit estimates obtained from a travel cost model would be
sensitive to the zone specification (i.e., size, shape). Smith and Kopp (1980) pointed out
that including zones fi'om the site being evaluated would likely violate some basic

aSSmnptions implicit in the travel cost model.

2-3 .2 Issues associated with zonal regression models to estimate recreation demand

There are various problems associated with developing zonal regression models.
I«o‘Vine (2007) indicated that non-linearity, negative predictions, non-consistent
summations, and large residuals could all be significant problems associated with
ordinary Least Square models. Demetriades (2002) raised the concern with extremely
complicated regression analyses procedures, as they are performed under the premise of
dealing with three problems: collinearity or multicollinearity, the existence of outliers,
and autocorrelation. They often mask but do not alleviate theses problems. Also,
researchers sometimes violate the basic assumptions of OLS, including linearity,

hol‘noskedasticity, independence of residuals, and normality of residuals.

30

A number of researchers involved in recreation study have warned that
heteroskedasticity in the residuals can result in non-valid zonal demand models. Bowes

and Loomis (1980) were among the first to warn of the potential heteroskedasticity
problem created by zonal data. They suggested that correction procedures could be
applied by weighting all of the variables by the square root of the zone’s population.
Vaughan et al. (1982), commenting on the Bowes and Loomis article, argued that an
alternative to assuming a linear demand equation and heteroskedasticity is to test for both
in the data, rather than to impose them as assumptions. He further argued that the
appropriate functional form for the data appeared to be non-linear, and that with a non-

linear forrn, heteroskedasticity did not appear to be a concern.

2-3-3 Selection of functional form when developing zonal recreational demand
models

Traditionally, researchers have applied transformations (e. g., exponential,
quéldratic, reciprocal, logarithmic and semi-logarithmic, power) to “smooth” the shape of
the curve, to achieve the best-fitting model for prediction, and to meet the assumptions

underlying the regression models.

The sensitivity of benefit estimates to alternative functional forms has frequently
been an issue cited in the literature. In a study of warm water fishing in Georgia, Ziemer
8’ C11. (1980) assessed the importance of the functional form on the size of consumer
Sul‘13)lus estimates. They found that surplus estimates varied significantly across linear,
Sell'li-log and quadratic forms. Vaughan et al. (1982) tested for appropriate functional
forth and heteroskedasticity simultaneously, using the Lahiri-Egy estimator, which is

belScd on the Box-Cox transformation, but also incorporates a test for non-constant

31

variance. They concluded that both the linear heteroskedastic and the linear
homoskedastic models were inappropriate. The semi-log form, which did not exhibit
heteroskedasticity, was found to be preferable. Strong (1983) compared the semi-log
model with the linear model, based on the mean squared error, when predicting

recreational trips. She also found that the semi-log function performed better.

Many of the studies appear to point to the semi-log as a desirable functional form,
yet the evidence is far from conclusive and there is no reason to believe that one form is
appropriate for all situations. For example, as discussed in Stynes et al. (1986), there is a
bias associated with retransforrning the log estimates back. The majority of the
recreational demand studies that involved performing transformations were done twenty
Years ago. New functional forms, such as fractional polynomials, spline and local
regression, which have become popular among biometric statisticians, have not yet been

applied in recreational demand studies.

2-3.4 Zonal influence on the regression model to estimate recreational demand

The concept of a zone of influence has only recently appeared in the literature
(Booth et al., 2001; Northridge et al., 2003; Pickett and Pearl, 2000; Sampson, et al. ,
2002). Pickett and Pearl (2000) argued that findings may vary across studies that employ
different zonal schemes. This is because the so-called areal units presume very different

Zones of influence for the contextual variables.

Zonal aggregation implies that variations at the individual level are lost as a result
0f the transformation from individual to aggregate data. There are two key problems

associated with analyzing aggregate data: (1) the modifiable areal unit problems

32

(MAUP); and (2) the ecological fallacy (Wong 1996). Cressie (1996) noted that the
MAUP could not be resolved until the spatial aspect was incorporated into the problem
formulation. However, spatial data have distinctive characteristics dependent upon the

level and mode of aggregation.

Modifiable Areal Unit Problems (MAUP) occurs when statistical results based

upon the analysis of spatial data are sensitive to the nature of the areal units anor which
the data are reported (Openshaw and Taylor, 1979). There are two problems related to
MAUP. First, a scale problem occurs when different statistical results are obtained from
the same set of data, when the data is aggregated at different levels of spatial resolution
(Wrigley, 1995). For example, Florida has 67 counties and 934 ZIP Code areas. The
Pearson Correlation Coefficient between the number of registered boats and the
Proportion of households with incomes over $50,000 at the county level is 0.23. But at
the ZIP Code level, the correlation coefficient increases to 0.4. This means that

generalizations about patterns and processes at one level may not hold at another level.

The second MAUP/aggregation (zoning) problem occurs when different statistical
reSults are obtained, when the number of spatial areas (e. g., zones) is the same, but the
20ties have different boundaries. When zonal schemes overlap (are not nested), different
combinations of spatial neighbors can yield constructed variables with very different
Statistical properties (Ali etal., 2005; Greenland, 2002; Tatalovich er al. , 2006). Both, the
Signs and the significance levels of regression coefficient estimates can be affected, and
Considerable variability has been found across small-area studies examining essentially
the same phenomenon (Cockings and Martin, 2005; Fotheringham and Wong, 1991;

Waller and Gotway, 2004).

33

2.3.5 Spatial regression models

Spatial data exhibits two properties that make it difficult to meet the assumptions
and requirements of the non-spatial statistical methods of OLS, which is the most widely
known and used regression method. First, geographic features are likely to be spatially
autocorrelated. Features near to each other tend to be more similar than features that are
more distant, which creates an “overcount” type of bias when using traditional (non-
spatial) regression methods (Anselin, 1988). Second, geography is important since the
model is non-stationary, meaning that processes behave differently in different regions or
parts of the study area. This characteristic of spatial data can be referred to as a regional
variation or spatial drifi (Fotheringham et al. , 2002). Being non-stationary is related to
the coefficients of the model (Fotheringham et al., 2002) while spatial autocorrelation is
related to the residuals (Anseiln, 1988). Recognizing these issues, the ideal spatial
re gre ssion model should identify a full set of independent variables that capture both
re gional variations inherent in the dependent variable and spatial autocorrelation
associated with different regression elements. However, no currently available regression
method can do both. Some spatial regression methods deal effectively with the spatial

allltocorrelation, while others deal effectively with the model being non-stationary.

2. 3. 5.] Exploring regional variations

Observed geographical patterns and relationships in recreational demand tend to
be Spatially variable (“non-stationary”). Even when the underlying processes are
universal, the realized patterns will vary with local conditions. There are two common

Ways of dealing with spatial variation in regression models. The first method is to include

a Variable in the model that explains the regional variations. This approach for dealing

34

with spatial variability is referred to as the spatial expansion model. Casetti (1972, 1986
and 1997) developed a spatial expansion modeling approach which (1) permits estimated
coefficients for structural characteristics to vary/drift across space; and (2) incorporates
spatially lagged values of the dependent variable. Using this approach, regression

parameters themselves are explicit functions of spatial locations, written as functions of

the position coordinates. This technique is restricted to displaying trends in relationships

over space. The form of the “expansion” equations must be assumed a priori and the

expansion equations must be deterministic. Another major difficulty with this method is

that secondary function specifications are subjective and calibration is challenging when

the expanded model parameters become non-linear (Paez, 2005).

An alternative method is to incorporate regional variations into the regression
model, such as Geographically Weighted Regression (GWR), which was first developed
by Brunsdon et al. (1996). GWR is one of several spatial regression techniques,
increasingly used in geography and other disciplines. It is a non-parametric locally-linear
model that provides a local model of the variable or process by fitting a regression
eQUation to every feature in the dataset. GWR builds a local regression equation for each
fe"flute in the dataset. The GWR method’s main feature is its use of distance-weighted
Sub'Samples of the data to produce locally-linear regression estimates for every point in

Space. Each set of parameter estimates is based on a distance-weighted sub-sample of

“ -
nGilghboring observations.”

Although GWR has been used increasingly for applied and policy-oriented
reSearch to analyze local variations, the spatially varying coefficient technique is prone to

multiCOIlinearity of local coefficients (Griffith, 2008; Wheeler and Tiefelsdorf, 2005),

35

such that the coefficients can be correlated even when there is not collinearity among
variables. Additionally, coefficients may demonstrate strong positive spatial
autocorrelation (Griffith, 2008). Issues with multicollinearity and spatial autocorrelation
among coefficients can lead researchers to arrive at incorrect conclusions about the
meaning of each coefficient. While there are proposed diagnostic tools, or remedial or
alternative methods for identifying and overcoming parameter collinearity in GWR, most
reseaIchers agree that GWR can be reliably used as an exploratory technique for

understanding how models may function differently across regions. The use of GWR in

this study reflects this exploratory strategy.
2- 3- 5. 2 Exploring spatial dependency
Spatial dependency is the extent to which the value of an attribute in one location
depends on the values of the attribute in nearby locations (Fotheringham et al., 2002).
Spatial autocorrelation is the correlation among values of a single variable strictly

attributable to the proximity of those values in geographic space (Griffith, 2003).

Spatial autocorrelation of residuals violates the assumptions of OLS. The
traditional approach is to treat it as a violation of the assumptions of OLS and, therefore,
to employ various methods to remove it from the data (e. g., transformation, removing

“Outliers”, re-sampling). When it comes to spatial modeling, however, autocorrelation is
an integral aspect of the data as well as evidence of important underlying spatial
pl'OGeSses at work. While traditional statistical theory is based on an underlying
assulnption of independent observations, geographers expect to find stronger
relationships among nearby variables than among spatially distant variables (Anselin and

Getis, 1992). According to Tobler’s First Law of Geography “everything is related to

36

everything else, but near things are more related than distant things” (Tobler, 1970). The
smaller the spatial units used in the analysis, the greater the probability that nearby units
will be spatially associated (Anselin and Getis, 1992). Unfortunately, the results of many

traditional statistical analyses conducted on geographic areas are often invalid if spatial

dependence is ignored (Anselin and Griffith, 1988).

There are various strategies for dealing with spatial autocorrelation in regression

model residuals. One way is to resample until the input variables no longer exhibit
statistically significant spatial autocorrelation. While this does not ensure that the analysis
is free of spatial autocorrelation problems, they are far less likely to occur when spatial
autocorrelation is removed from the dependent and explanatory variables. This traditional
stati stician’s approach to dealing with spatial autocorrelation is only appropriate if spatial
autocorrelation is the result of data redundancy. When it comes to recreation, researchers
Prefer to test and select among different functional forms (e.g., semi-log transformation
to “Smooth” the residuals). Unfortunately, space itself has important explanatory value in

geographic models. Removing space when space is a relevant factor will result in weaker

Predictors and the risk of misspecification.

Another solution is to isolate the spatial and non-spatial components of each input

Variable, using a spatial filtering regression method. Space is removed from each
Variable, but then entered back into the regression model as a new variable, to account for
Spatial effects/spatial structure. The idea is to identify which component of each variable

IS due to spatial autocorrelation and then split that variable into its spatial and non-spatial

components (Getis .1995 and Griffith, 2002)-

37

 

Bivand et al. (2008) summarized the use of a spatial regression model to handle
the spatial autocorrelation of residuals. This includes simultaneous autoregressive models
(SAR), conditional autoregressive models (CAR), mixed-effects models, generalized
additive models (GAM), etc. Their book also covered spatial econometric regression
methods (Anselin, 1988, 2000; Anselin et al., 2004), which incorporate spatial
dependence in two distinct ways: (1) as an additional regressor, in the form of a spatially-
lagged dependent variable, which is known as a spatial lag model; or (2) in the error

structure, which is called a spatial error model.

2-4 Spatial and GIS approaches in tourism and recreation studies
Spatial context is important in understanding tourism and recreation. GIS has
been widely applied by researchers and practitioners in tourism and recreation
management (e.g., Lee et al. 2003; Leung et al., 2002; Harris et al., 1995; Walker and
CUff, 1997). GIS maps can simultaneously and holistically show the distribution of
recreation resources, recreation participants’ activity patterns, and other factors such as
demographic information. This multi-dimensional perspective allows researchers to
describe and analyze complex interactions and relationships, which has encouraged the
Wide implementation of GIS as a primary decision-making tool among recreation
reSource planners and managers (N icholls, 2001). For example, Confer and Graefe (1992)
Studied boaters’ attitudes and activity patterns regarding recreation sites by using GIS
maPS. Carver et al. (1996) produced a “wilderness continuum map” to show wilderness
designated areas in the United Kingdom. Distance is an important factor, influencing
VisitOI‘S’ recreational behaviors (e. g., Debbage, 1991; Fesenmaier etal., 1980), and GIS

i . . . . .
mp1» 0Ves the accuracy of distance calculations usmg road network analySIS 1nstead of

38

 

straight line calculations (e. g., Zawacki and Marsinko, 1999; Souleyrette, 1993; Bateman
et al., 1999). Spatial analysis functions in GIS—such as buffer, intersection, network
analysis and spatial queries—can be employed to compute accessibility by radii buffer
techniques. This involves drawing a line around a feature at a given distance to find out

the number of facilities and proportion of the population in the selected area (N icholls

and Shafer, 2001).

So far, GIS functions have focused mainly on spatial data management, data
integration and visualization, but many recreation- and tourism-related studies require the
analysis of complex patterns of interrelated social, behavioral, economic, and
environmental phenomena (Oliveira and Fuks, 2007). There is both opportunity and need
for utilizing spatial statistics and employing spatial regression to more adequately
describe and analyze recreational and tourism demand, including recreational boating.
This study will attempt to employ these spatial methods to better understand and predict

boat demand and ownership in Florida. They will be described in Chapter Three, which

diagrams and describes this dissertation’s research process.

39

CHAPTER THREE DATA AND METHODS

This chapter describes the research process, including the different types and
sources of data, data cleaning and preparation, and the modeling procedure. The
modeling approaches were designed to achieve the research objectives and to test this
study’s hypotheses. The secondary data employed in this study came from a wide variety
0 f sources. Steps that comprise the research process are shown in Figure 3-1. Chapter
One presented the research objectives and hypotheses. This chapter focuses on the

development of the models.

Figure 3-1 Elements of the Research Process

 

Research objectives and hypotheses
[1] (Chapter One)

i

Identification and review of potential modeling approaches
[2] (Chapter 3.1)

l

Evaluation of alternative zonal structures
[3] (Chapter 3.2)

l

Identification of potential model variables, data sources and data
preparation
[4] (Chapter 3.3)

l

Build the model
[5] (Chapter 3.4)

l

Hypotheses testing
[6] (Chapter 4.1 and 4.2)

l

Model estimation and evaluation of the alternatives models
[7] (Chapter 4.3 and 4.4)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

40

Various software programs were employed to organize data, build models and
visualize results, e.g., mapping. Non-spatial statistical analyses (e.g., frequencies
correlation analysis) were performed using SPSS Software for Windows (SPSS, 2009).
Variables to be included in the OLS models were selected using R code (R development
core team, 2008) procedures developed by the author. The locations of the owners of
different types and sizes of boats were determined using ESRI’s Address Coder (ESRI,
2008a). Spatial data were mapped using Shape files (*.shp) in ArcGIS 9.3 (ESRI, 2009)
and images file (*.jpg) in GeoDa (Anselin et al., 2006). GeoDa was employed to estimate
the OLS models and to perform the specification tests. Spatial autocorrelation tests,
including global spatial autocorrelation and LISA, were also performed using GeoDa.
Spatial regression analysis was performed to develop the spatial lag, and spatial error
models were performed using GeoDa. SAR (Spatial Autoregressive) models and CAR

(Conditional Autoregressive) models were created using R code. Images in this

thesis/dissertation are presented in color.

3- 1 Identification and review of potential modeling approaches to estimate
bOat ownership demand

One of the main objectives of this study is to develop a series of multi-scale
Inc>dels to estimate and project boat ownership (“boat demand”) in Florida. As indicated
in the literature review, two types of models have been used to estimate recreational
demand: (1) disaggregate models; and (2) aggregate models, called zonal models in this
Study. Both types of models were carefully reviewed before a model was selected.

Disaggregate models are primarily employed to estimate recreational demand for specific
recreational facilities and site(s). However, their use and scales are somewhat restricted

due to the cost and difficulty (e. g., data collection biases), and are often limited to

41

specific site(s). They are less applicable for projecting aggregate demand on a regional or

state level.

Zonal models provide a more practical approach for estimating and projecting
boat ownership in larger, multi-site geographic areas. Zonal modeling usually employs a
regression framework. A zonal regression model for a study area with k zones can be

expressed as follows:

Y = BX + e (3.1)

where Y is the vector of dependent variable, e. g., the number of boats per 1000
households in Florida counties, X is the vector of independent variables, e.g., the number
of boat launch sites in Florida Counties, B is the vector of coefficients and a is the vector

of errors.

One basic assumption underlying a zonal regression model is that zonal attributes
(e.g., water area, household income) are uniform in their spatial distribution throughout
the zone. The data related to dependent variables (Yi) and independent variables (X 1 i,
X2i,. . ., Xm‘)are associated with/related to the zone, rather than individuals (e.g., boat
owners). Data pertaining to individuals (e. g., a boat owner) residing within the zone are
aggregated for the zone, regardless of where individuals reside within the zone. For
example, in 2007, there were 1,948 boat owners residing in Franklin County (Florida),
and if the zone unit were the county, it would not matter where in the county they lived;

the number of owners was 1,948.

The literature review included a discussion of some problems associated with

zonal regression models. For example, zonal models don’t take variations of attributes

42

across individuals within a zone into consideration, creating problems such as ecological
fallacy. In addition, zonal structure, the scale of zones, and the spatial arrangement of
aggregate data will have a significant influence on the models’ results. Another concern
with zonal models is that they do not take spatial autocorrelation of model residuals or
spatial variation of relationship between dependent variable and independent variables
into account. This study developed a zonal model of boat ownership that employs

alternative ways of addressing the above weaknesses of this modeling approach.

3.2 Evaluation of alternative zonal structures

The initial choice of zonal structure is crucial to the validity and reliability of
zonal regression models. An essential element in formulating zonal regression models is
aggregating data within the zone, to the centroid of the zone. The zonal structure has
implications for not only the model but for the mapping and reporting of spatial
information as well. Irregularly-shaped zones and zones with significant size differentials
cause problems, such as ecological fallacy and the MAUP in the context of a zonal
regression model. This study will assess MAUP in relationship to zonal structure, scale

effects, and aggregate effects.

The two primary issues related to selecting a zonal system are: (1) how to fashion
a relevant and usefiil set of zones; and (2) how to cope with the same data being
aggregated to multiple different geographies, each of which often shows different map
patterns and yields different data or results due to problems such as MAUP. Cliff et al.
(1975) suggested three design requirements when structuring zones: (1) zones should be
simple and kept to as few as possible; (2) zones should be as similar as possible, with a

high degree of homogeneity within the zone; and (3) they should be

43

compact/compressed. Wise et al. (1997) additionally suggested a balance across these
three output zone criteria when creating: (l) homogeneity, (2) equality of size, and (3)
compactness in shape. Levine (2007) contends that the selection of a zonal system entails
necessary tradeoffs among: (1) the size of zones; (2) consistency in zone size (less
variability is better); (3) distortion due to shape (i.e., the more irregular the shape, the

greater the distortion); and (4) the availability of data for the zones that are created.

Researchers recommend and have employed two types of zonal systems: (1)
census geography and (2) grid cells. Census geography is consistent with the zones
employed by the US. Census Bureau. US. Census Bureau data remains the most widely
used source data, as well as the best available data for demographic analyses in social
science. The majority of public health- and crime-related research and other social
science utilize census geography zones, such as the census block, census tracts, ZIP Code
Tabulation Areas (ZCTA), county, state, etc., as their zonal system. The original
obligation of the US. Census Bureau was to provide the most complete and accurate
population count for apportionment of the seats in the US. House of Representatives.
Over the years, other purposes and requirements have evolved, such as the measurement
of social and economic trends. Census data is reported and displayed for different
geographic areas for different purposes: (1) legal, e. g., congressional districts; (2)

administrative, e. g., ZIP Codes; and (3) statistical, e. g., census tracts (U .S. Census Bureau

2000a).

Census data is zone-based aggregate data, which utilizes census-defined
geographic entities including census blocks and block groups, and census tracts. Many

analytical projects and research studies have gone on to define their own special-purpose

44

zonal units based upon modifications of census geography. For example, Traffic Analysis
Zones are employed in transportation studies and Medical Statistical Areas are often used

to present analyses done by health and medical researchers.

Grid cells, which are uniform zones imposed on geographic areas (e. g., regions),
are a widely-used alternative to census-defined zones. Each zone in a grid system is a
square of equal size and shape, and area effects are uniform for all zones, which are
highly desirable for statistical properties. The other advantage to grid cells is that the
zones and zonal structure will not be changed over time. This is not true for census zones
or other politically-influenced zonal units (e. g., congressional districts). The grid cells’
consistency and uniformity are especially beneficial for longitudinal studies. The
difficulty in obtaining accurate, up-to-date grid-level data limits the use of grids. With the
current development of GIS technology, however, it is the compilation of demographic
data for geographic areas of any size and shape much more attainable. For example,
ESRI’s Business Analyst (ESRI, 2008b) offers the ability to report demographic

information on polygons of any size and shape.

This study employs three different census zonal structures to test zonal influences

in the models. The three zonal units and their relationships are shown in Table 3-1.

45

Table 3-1 Census zonal units used in this study
Areal Units Definition

 

County Counties are the primary legal divisions of most states

_______. ' " ‘ " — XZéTXis all statistical ‘geagra‘pmc' any that appioximates sedentary "sea '
for a US. Postal Service five-digit or three-digit ZIP Code. ZCTAs are

ZCTA aggregations of census blocks that have the same predominant ZIP Code
associated with the addresses in the US. Census Bureau’s Master Address
File
_ H T ” '1' fig—“Censusitract—swargsmall, ré'léiivéEBengfififitiéisibdiviéian'd‘f‘é—T' A ‘ 1
Census Tract county. Census tracts generally have between 1,500 and 8,000 people, with an

optimum size of 4,000 people.

 

The boundary maps of Florida’s counties, ZCTAs and census tracts were
downloaded from the Florida GIS Library (httpz/lwww.fgdl.org/, retrieved on May 15

2009). There are 67 counties, 934 ZCTAs and 3,168 census tracts in Florida. The

boundary maps for the different zonal structures are shown as Figures 3-2, 3-3 and 3-4.

Figure 3-2 Florida County Boundaries

it;

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

1:] Counties (2000)

 

 

 

 

my"

 

46

Figure 3-3 Florida ZCTA Boundaries

 

 

 

 

 

 

 

C3 ZIP Code Tabulation Areas (2007)

 

 

 

Figure 3-4 Florida Census Tract Boundaries

 

 

 

 

 

 

C] Census Tracts (2000)

 

O

 

 

 

Among these three zonal units, the census tract is the smallest; these are nested

within counties. A ZCTA unit is smaller than a county, but overlaps with census tracts

47

and county units. ZCTAs and census tracts have a similar scale in low population density
areas. However, in a high population density-area, there are more census tracts than
ZCTA units because census tracts have fixed populations, usually between 1,500 and
8,000 persons (US. Census Bureau, 2000a).
3.3 Identification of potential model variables and data sources, and data
preparation

As already mentioned, the data used in this study was, of necessity, obtained from
many sources and then aggregated to county, ZCTA and census tract level. The agencies
and organizations that collect and maintain the data do so for different geographic areas

and zones. The aggregated data/variables were employed to build the models.

3.3.1 Dependent variables

The dependent variables in the models are boat ownership, defined as the number
of registered boats owned by persons residing in Florida counties. The most up—to-date
and complete database of 2007 registered boat ownership was provided by the Florida
Fish and Wildlife Conservation Commission (FWC). The boat registration records that
comprise the database include boat owners’ address, county of residence, gender, date of
birth, boat type and boat size. According to the FWC (2007), there were 884,560 boats
registered in Florida, including pleasure vessels (96.0%), in 2007. About 96.4% were
owned by Florida residents. Table 3-2 shows a breakdown of 2007’s registered boats by

different use types.

48

Table 3-2 Number and types of registered boats in Florida by use types

 

 

 

Mutation Use Number of Boats Percent
_C-9_rane-r_9ia!_Char_t-§r_-_ - - -- - ‘- - _-5.,4_1_-9 __v 06%
-C_qm-mer-c-ial-Fish- - - - _ - 8.899%.- - , ,1-.0%
Esra-Lngrgial-HHCW- --_- - - - -.-__-4-,-9_§3 -_ ,_ ,- - - 0:699-
..Exsmrztbatialflsssel- .. _ - 3317 414%
Ears-mast}: Y9§s-e!__ -,_ _ -_ - -_ 3,369.” - - 04% -
Jilgasurflesgl- __ - - - --84-8r81L- - W- -._9§-Q‘1/2.-
_0__ther W. -_ , W -.-__9,_8§2___-_W 1_-_1_%__
Total 884,560 100.0%

 

Since the purpose of this study is to develop models to estimate recreational boat
ownership in Florida, only registered boats used for recreation and owned by Florida
residents who reside in counties of Florida were included in the database. In many
previous boating studies, recreational boats were classified as: power boats, sail boats,
personal water craft (PWC), and canoes/kayaks (e.g., Mahoney et al., 2009; NMMA
2008; Stynes et al. , 1998; Wu, 1995). These classifications were developed based on the
characteristics, behaviors and spending profiles of their users. However, the Florida
Registered Boat Database includes a much more detailed description of vessel type, such
as airboat, auxiliary sail boat, cabin motorboat, canoe, houseboat, inflatable, open
motorboat, pontoon, and personal watercraft (PWC). These more detailed classifications
produce a number of segments with a very small number of boats, especially when
viewed geographically. For the purpose of this study, registered vessels were re-classified
and recoded into five types: power boat, sail boat, PWC, canoe, and “other”. These boat
types were the focus of the models. The vessel (registration) types and the

reclassified/recoded boat types are shown in Table 3-3.

49

Table 3-3 Number of registered boats by vessel (registration) type

 

 

 

 

 

 

Aggregate

Number of Number of

Registered 1 Registered 2
Vessel Type Boats Percentage Boat Type Boats Percentage
Airboat 6,048 Power Boat 0.7%
Cabin Motorboat 71,096 Power Boat 8.7%
Houseboat 808 Power Boat 0.1%
Inflatable 14,361 Power Boat 1.8%
Open Motorboat 478,511 Power Boat 58.6%
Pontoon 29,347 Power Boat 3.6% 600,171 83.4%
Auxillary Sailboat 10,626 Sail Boat 1.3%
Sailboat 4,499 Sail Boat 0.6% 15,125 2.1%
Canoe 8,151 Canoe 1.0% 8,151 1.1%
Personal Watercraft 95,81 8 PWC 1 1 .7% 95,8 1 8 13.3%
Other 97,637 Other Boats 12.0% -
Total 816,902 100.0% 719,265 100.0%

 

1. Percent of total number of registered boats including other type of boats
2. Percent of total number of registered boats not including other type of boats

The database used to develop and test the models consisted of 719,265registered
boats, which represents 78% of all registered boats in Florida in 2007. The addresses of
the owners of these boats were then geocoded using ESRI Address Coder (ESRI, 2008a),
which employs the most recent commercial street data from Tele Atlas as the street
matching file. The latitude-longitude coordinates of the owners’ residence, Federal
Information Processing Standard (F IPS) codes, and ESRI’s community tapestry data
(ESRI, 2008c) were appended to the registered boat data. The summary of the
demOgraphics and description of LifeMode and urbanization segmentations in ESRI’s

Comnmnity tapestry are in Appendix A and Appendix B.

In this study, the geocoding match rates at the county level ranged from 5%
(DiXie County) to 94% (Broward County). Appendix C shows the distribution of geocode
match rates at street level and ZCTA level. Figure 3-5 shows the street level match rate

for different counties in Florida. The match rate disparity is consistent with patterns of

50

 

development in Florida, with newer and more rapidly changing counties in Florida faring
worse than more established counties. The total match rate at street level is 86.5%
exceeding the 85% threshold that is the minimum acceptable geocoding rate for address-
based point pattern datasets, as demonstrated by Ratcliffe (2004). Only 0.3% of the

database records had bad addresses that could not be geocoded.

Figure 3-5 Distribution of Street-level Geocode Match Rates by Counties

 

  
    

Geocode Rate at Street Level
;’ under 50%
L4, 50% - 64%
- 65% - 79%
- 80% - 89%

- 90% and higher

 

 

W”

 

In addition to attaching longitude and latitude coordinates to the addresses of
registered boat owners residing in Florida, the ESRI Address Coder also builds a point
map for each geocoded boat registration record as a shape file. Figure 3-6 shows the
diStribution of registered boat owners residing in Florida, using the data created by the
ESR1 Address Coder. The ESRI Address Coder also attaches F IPS County codes,

ZCTAs, and census tracts to the boat registration records, permitting the aggregation of

51

registered boats at the county, ZCTA and census tracts levels. Figures 3-7, 3-8 and 3-9

show the number of registered boats by county, ZCTA and census tracts.

Figure 3-6 Point Map of the Distribution of Registered Boat Owners
Residing in Florida

 

 

0 Registered Boat Owners - . ..l‘ /

 

 

 

Figure 3-7 Distribution of Registered Boats by Counties in Florida

 

  
 

Total Number of Registered Boats
1, '1 under 2,500

n 2,500 - 4,999

- 5,000 - 12,499

- 12,500 - 24,999

-25,000 and higher

 

 

 

52

Figure 3-8 Distribution of Registered Boats by ZCTAs in Florida

 

 

  
 

Total Number of Registerd Boats
D under 199

- 200 - 499

- 500 -999

-1,000 - 1,999

- 2,000 and higher

 

Figure 3-9 Distribution of Registered Boats by Census Tracts in Florida

 

 

  
 

Total Number of Registered Boats
E under 199

[:3 200 - 499

- 500 - 999

- 1,000 -1.999

-2,000 and higher

 

53

 

 

 

To test the hypotheses regarding the relationship of different factors with the type
and size of boats, registered boats were segmented into type and size segments (see Table
3-4). Since the majority (85.6%) of Florida’s registered boats were power and sail boats,
these two types were further segmented into small power boats (<23 feet), large power

boats (23+ feet), small sail boats (<23 feet), and large sail boats (23+ feet).

Table 3-4 Number of registered boats by types and sizes

 

Boat types and sizes Number of Boats Percentage Percentafl
“Tel-1|.- __ -, _ -W---§1§,992W _-_ -
_-P<2!v_§r-399t- W- - - 600,171- -73-5%]- --

-..S-ai1§9at - -, - - - _- M 15.125 19%],

Canoe- -- - . 8,151-- _- 170%1

..PECW _. - 95,818-, 1_l-7%1_, - _. _
_-13_9v_vgr-B-0§L€2-3fi--_--- - 477,677 . 5.8-5.7; ,_-_79-§%2-__
,fiwflgatzérfi- - - ~,_12_2,_494-- .----A15-.0_%‘ - 20.4%?
_saWn-Iagaiszam- .______ __ - Wag-33W _9572L- .._2-.8s6-‘-’43-,
Sail Boat 23m 10,792 1.3%1 71.4%3

 

1 .Percentage of total number of boats
2. Percentage of Power Boats
3. Percentage of Sail Boats

To avoid aggregation bias, the ownership rate—the number of registered boats per
1 ,000 households—served as the dependent variable, rather than the number of registered
to Persons residing in various zonal units. Figures 3-10, 3-11 and 3-12 show the number

0f registered boats per 1000 household by county, ZCTA and census tracts.

54

Figure 3-10 Distribution of All Boats Per 1000 Households
at County Level

 

 

   
 

Number of Boats per 1000 HHs
_ .. under 100

m 100- 149

-150-199

-200-299

-300+

 

Figure 3-11 Distribution of All Boats Per 1000 Households
at ZCTA Level

 

 

  
 

Number of Boats per 1000 HHs
:under 100

100- 149

-150-199

-200 - 299

-300+

 

55

 

 

Figure 3-12 Distribution of All Boats Per 1000 Households
_at Census Tract Level

 

  
 

Number of Boats per 1000 Hl-ls
Sunder 100

[23100-149

—150- 199

-200-299

-300+

 

 

 

If the number of boats was the dependent variable, the number of households and
population would explain most of the variance and it would be difficult to understand
how other factors, e. g., boating access, affect ownership rates. Table 3-5 presents the

dependent variables used in this study.

Table 3-5 The dependent variables used in this study

 

 

 

 

 

Variables Definition

ABH Total number of boats per 1,000 households
I APBH r Rumber of power boats E; 1,000 households 7*? 7 fl
7 ASBH 7 fl 7 ANumberiof sail boats per 1,000 households, 7 7 '—
PWCH 7 7 FTTNumber of PWC per 1,000 households A
CANOEH W Number of canoes per 1,000 households 7 7777 7 _ii
[SPBHT H r Number of small power boats (<23 ft.) per 1,000 households ‘
LPBH r i 7 fl Number of large power boats (23 fi+) per 1,000 households 7 i
ssi3ri 7 7 N W 7 Number of small sail boats (<23 a.) per 1,000 agent]; 7
TLSBHT r r 7 Number of large sail boats (23 1114-) per lbfidseholds TT— W

 

56

3.3.2 Independent variables

The data measuring the independent variables were obtained or derived from
various sources such as SimplyMap (Geographic Research, Inc., 2008), the Florida
Recreational Boating Access Inventory and the Florida GIS Library. Socioeconomic,
urbanization and water access (e. g., marinas, launch sites) variables were formed. The
potential independent variables, the sources of data about the variables, and zones for

which data on the variables are available are listed in Table 3-6.

Table 3-6 Independent variables employed to develop the boat ownership models

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Source of Availability of Data
Factors Variable Definition for Different Zonal
Data .
Umts
Socioeconomic and demographic variables
Age Percentage of population with . .
age between 35 to 65 in 2007 SimplyMap A" Zonal Umts
Gender lz’gggentage of male population in SimplyMap All Zonal Units
Race Percentage of white non- . .
Hispanic in 2007 SimplyMap All Zonal Umts
Ethnic Percentage of Hispanic SimplyMap A11 Zonal Units
Group population
Income Percentage of households with . .
income at least $50,000 SimplyMap A“ Zonal Umts
Percentage of households with . .
income greater than $75,000 SimplyMap A" Zonal Umts
Percentage of households with . .
income over $100,000 SimplyMap A11 Zonal Units
Family Percentage of married family . .
Structure households with children SimplyMap A“ Zonal Umts
Percentage of non-family . .
households SimplyMap All Zonal Umts
Percentage of female head of . .
household SimplyMap All Zonal Umts
manhation
Population Population per square mile . .
Density SimplyMap All Zonal Units
Urban . The percentage 0f urban SimplyMap A11 Zonal Units
Population population
Urban Area The percentage of urban area Florida GIS All Zonal Units
Library

 

57

Table 3-6 Independent variables employed to develop the boat ownership models (Con.)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

. . ' Source of Availability of Data
Factors Variable Definition for Different Zonal
Data .
Units
Water accessibility
' 1
Marina Number of marinas FRBAIP All Zonal Units
Number of marinas within 1 1
mile distance to the zone FRBAIP All Zonal Units
Number of marinas within 3 1
miles distance to the zone FRBAIP All Zonal Units
Number of marinas within 5 1
miles distance to the zone FRBAIP All Zonal Units
1
25:; Launch Number of boat launch sites FRBAIP All Zonal Units
Number of boat launch sites
within 1 mile distance to the 1
zone F RBAIP All Zonal Units
Number of boat launch sites
within 3 miles distance to the 1
zone FRBAIP All Zonal Units
Number of boat launch sites
within 5 miles distance to the 1
zone FRBAIP All Zonal Units
Seasonal Percentage of seasonal housing
Home units that were owner-occupied Simply Map All Zonal Units
Ownership in 2007
Len h of Percenta e of areas within one Florida GIS .
Coa§iline mile’s disgtance to the coast line Library A" Zonal Umts

 

1 F RBAIP: Florida Recreational Boating Access Inventory Project (2007)

3.3.2.] Sources of data for socioeconomic and demographic variables

There are many sources of socioeconomic and demographic data, including the

US. Census Bureau, ESRI, SimplyMap, and DemographicNow. The last US Census that

provided socioeconomic data for counties, ZCTAs and census tracts was conducted in

2000. It was considered necessary to match socioeconomic variables with the 2007 boat

registration data. The US. Census Bureau does not, however, provide projections on all

variables of interest to this project, so other sources of data were utilized.

It is possible to obtain 2007 socioeconomic data for various geographic areas

through America Fact Finder (US. Census Bureau, 2009b). SimplyMap provides 2007

estimates/projections for a greater number of demographic and socioeconomic variables

58

than any other provider of this type of data, including ERSI. Thus, based on the variety of
variables and the credibility of the data, it was decided to obtain 2007 data relating to
population size, households, gender, age, race and ethnicity, income and family structure

from SimplyMap.

3. 3. 2.2 Data source for urbanization

It was hypothesized that urbanization is a significant factor influencing boat
ownership, in that it affects many aspects of it—including availability and cost of storage,
regulations concerning storage locations and time to access boating waters. Population
density (population per square mile), the pr0portion of urban areas and the proportion of
rural populations are often used as measures of urbanization. Population density and rural
population data are available for 2007 from SimplyMap. The proportion of urban areas
was calculated through the urban boundary map in Florida. The urban boundary map,
available as a download from the Florida GIS Library, was used to identify urban
boundaries in Florida. The ratio of urban area to total area for different zonal units was
calculated using the ArcGIS 9.3 (ESRI, 2008a) desktop version. Figure 3-13 shows urban

areas in Florida.

Figure 3-13 Urban Areas in Florida

0. _ . "Ti :7! ‘ if. ‘.'0 r0.
‘. 4.1%.. .l. .2. I":. .I. .
. v "'79. -.-a_:;: , . .
I. . . ' 'I O 7‘

. .

 

 

   

 

Urban Areas

 

 

 

59

3. 3. 2.3 Data source for proximity to water

The proximity to water was calculated based on measurements of the length of
coastline downloaded from Florida GIS Library. Coastline is shown as line segments,
which include tidal line, canals and inland waterways and streams. Figure 3-14 shows a
map of the Florida coastline. However, because different sources of data are utilized,
there is not an exact alignment of coastline with county, ZCTA and census tracts
boundaries. Figure 3-15 shows the extent of the misalignment of county boundaries and
coastline for Lee County. The degree of misalignment reduces the accuracy of estimates
of the aggregate length of coastline for different geographic zones. To reduce the
inaccuracies, a one-mile buffer was added to each zonal unit (See Figure 3-16). The ratio
of the area of a one-mile buffer to the coast line within a zone vs. total zone area was
calculated and served as an indicator of the proximity of residents living in different
zones to boating waters.

Figure 3-14 Coastline of Florida

 

 

 

 

 

 

—Coast Line

 

 

 

6O

Figure 3-15 Misalignment of Coastline and County Boundary

 

     
   

 

 

 

—Coast Line
— County Boundaries

 

 

 

 

 

  

-One-mile Buffer to County Boundary

  

 

 

 

61

3. 3. 2.4 Data source for boating access facilities

Boating facilities data were obtained from the Florida Recreational Boating
Access Inventory Project. This state-wide boating access inventory was conducted in
Florida, in 2007, as part of a larger boating study that includes a demand analysis and an
economic impact assessment. The Recreational Marine Research Center at Michigan
State University and the author of this dissertation played a major role in developing the
inventory methods and databases. The longitude and latitude coordinates of marinas and
boat launch sites were collected as part of this inventory. Marinas and boat launch sites
were digitized into point data in a shape file, and the number of marinas and boat launch
sites were aggregated into county, ZCTA and census tract zones using the ArcGIS’s
intersect function. Figure 3-17 and Figure 3-18 show the distribution of marinas and boat
launch sites in Florida. Appendix D lists the number of marinas and the number of boat
launch sites by county. Figures 3-19 and 3-20 show the number of marinas and boat

launch sites at the county level.

 

Figure 3-l7 Distribution of Marinas in Florida

If

    
       

  

.
x .
~ I

 

 

 

62

Figure CH 8 Distribution of Boat Launch Sites in Florida

 

    

0 Boat Launch Sites

 

 

 

Figure 3-l9 Number of Marinas by County

 

  
 

Number of Marinas by County
F Ll 0

- under 15

- 15 - 29

- 3O - 49

- 50 and higher

..fl"

 

 

 

63

Figure 3-20 Number of Boat Launch Sites by County

 

  
 

Number of Boat Launch Sites by County
, 15 - 14

~' ; 15 - 34

1:135 - 69

L] 70 - 99

- 100 and higher

 

 

 

According to various recreational boating studies, there are three major types of
boating access: marinas, boat launch sites and private docks. The Florida Recreational
Boating Access Inventory Project (FWC, 2009) provides an estimate of boating facilities,
the majority of which are private docks, attached to waterfront home. However, due to
the lack of information at the ZCTA and census tract levels, private docks will not be
included in this study. Alternatively, the proportion of seasonal vacant housing units was
used to measure waterfront houses, including those with private docks. Boating facilities
capacity data, such as the number of slips attached to marinas and the number of launch
ramps attached to boat launch sites, are normally used to measure boating opportunities
and accessibility. Nevertheless, more than half of the marinas and ramps didn’t report
their slip and ramp information in the Florida Recreational Boating Access Inventory

Project. Therefore, the use of wet slip and ramp data may be questionable for this study.

Alternatively, counts of marinas and boat launch sites counts were employed as the

measures of boating access in this study.

As already discussed in the literature review, a marina or a boat launch site within
a zonal unit may not be located nearest to a boat owner. Recreational demand theory
assumes that all things being equal, boaters—including boat owners and non-boat
owners—seek to reduce their time and travel costs, and so will travel to the nearest
facility, in this case a marina or launch site. Boaters, including boat owners, are not
limited to utilizing marinas or launch facilities located in the same zone where they
reside, especially where zones, as is the case for census tracts, are small in size. Launch
sites and marinas in other zones may actually be closer to boaters’ residences than
marinas and launch sites located in the zones where they live. Recognizing this issue,
one-, three-, and five-mile buffer zones along zonal boundaries were created; marinas and
boat launch sites were aggregated for these buffers, using the ArcGIS’s spatial join
function. As an example, Figure 3-21 shows a one-mile buffer for ZCTA 33884. There
are two marinas within ZCTA 33884 and thee more marinas within one mile of its
boundaries. Figure 3-22 shows a three-mile buffer to ZCTA 33884. Adding a three-mile
buffer to ZCTA 33884 increases the number of launch sites to 41. This dramatically
changes the water accessibility. This study will explore the threshold buffer distance for

different zone units.

65

Figure 3-21 Distribution of Marinas and One-Mile Buffer to ZCTA 33884

  
 
   
 

.
I s
tan
“a
_ ...L

 

. Marinas Located in 33884

Marinas Located in One- 7
. mile Buffer to 33884 F

C Other Marinas

fCJ Zipcode 33884
a One Mile Buffer to

 

 

 

 

 

Zipcode 33 884 .1
Li W; Water Bodies ’
Figure 3-22 Distribution of Boat Launch Sites and Three-Mile Buffer to
ZCTA 33884
'f
f.
.

 

Boat Launch Sites
Located in 33884

. Boat Launch Sites Located
in Three-mile Buffer to 33884

C Other Boat Launch Sites

D Zipcode 33884

Three-mile Buffer to Zipcode
33884

_ mi“, Water Bodies

 

 

 

 

66

3.4 The modeling process

Figure 3-23 is a detailed diagram of the model-building process. The first step
was to conduct a preliminary analysis of boat registration data, to identify and explore
variables that may affect boat ownership rates. Based on the literature review and the
preliminary analyses presented above, hypotheses were formulated and tested concerning
the relationships and influences of demographic, socioeconomic, urbanization and water
accessibility factors on boat ownership. This includes where the owners reside, their ages
and genders, all of which were obtained directly from the Florida boat registration data.
The income, race, household structure of boat owners were inferred from the ESRI’s
Community Tapestry LifeMode and Urbanization designations (ESRI, 2008c). Next, boat
owners were identified with, and aggregated to, county, ZCTA and census tract units.
Hypotheses concerning the relationships and influences of demographic, socioeconomic,
urbanization and water accessibility factors on boat ownership were tested. Then a
conventional ordinary least squares (OLS) regression was utilized to build models to
estimate the ownership of different types and sizes of boats, including all boats, at the
county, ZCTA and census tract level. A series of specification tests were then conducted
to assess whether there was evidence of spatial autocorrelation. Finally, alternative spatial
regression models, e.g., the spatial lag model, the spatial error model, and SAR and CAR
models were employed; whether or not they were an improvement on the OLS model was

evaluated.

67

Figure 3-23 Model Building Diagram

 

 

Describe the current recreational boat ownership patterns based on individual boat
owners
(3.4.1)

 

 

ti

 

 

Identify and test variables that are statistically related to the ownership of all boats
and to various types and sizes of boats

(3.4.2)

 

 

Vi

 

 

Develop OLS models that estimate boat ownership in Florida at county, ZCTA, and
census tract level

(3.4.3)

 

 

V

 

 

Develop alternative spatial models based on OLS models
(3.4.4)

 

3.4.1 Describe the current recreational boat ownership patterns in the State of
Florida, including the spatial distribution of boat owners

A preliminary analysis of boat registration data was conducted to identify and

explore variables that affect boat ownership. As described previously, during the

 

 

 

 

geocoding process, ESRI’s Community Tapestry Lifestyle designations were attached to

the data on the owners of registered boats.

According to ESRI’s white paper (ESRI, 2008c), ESRI’s Community Tapestry

Segmentation system provide that structure, a system for classifying consumers using all

the variables that can distinguish consumer behavior, from household characteristics like

income and family type to personal traits like age, education, or employment and even to

housing choices. Community Tapestry combined the “who” of lifestyle demography with

the “where” of local neighborhood geography to create a model of various lifestyle

classifications or segments of actual neighborhoods. Tapestry Lifestyle helped to identify

68

and approach the diversity of American neighborhoods by segmenting US households
into one of 65 distinct market segments, nested within 12 overall LifeModes, along with
11 Urbanization categories reflecting the degree of urbanization based on income,
employment, home value, housing type, education, household composition, age, and
other key determinants of consumer behavior. This was based on an idea that people who
live in close proximity to one another share characteristics that make them similar and, at
the same time, different from those who live elsewhere.

Data sources used to segment American population in Community Tapestry
include Census 2000 data; ESRI's demographic updates; Acxiom Corporation's InfoBase-

XTM consumer database; and consumer surveys, such as the Survey of the American

Consumer from Mediamark Research & Intelligence LLC. Selection of the variables used
to identify consumer markets are mainly from Census 2000, which includes household
characteristics, income, relationships, and personal traits such as age, sex, education,
employment, and marital status; and housing characteristics. In terms of geographic units,
Tapestry profiles enable comparison of consumer markets across the country by state,
metropolitan area, county, place, census tract, ZIP Code, and census block level.

The statistical methods used there is cluster analysis, including the combination of
iterative partition K-means algorithm and Ward's hierarchical minimum-variance method
to group the clusters. Moreover, ESRI developed and incorporated these data mining
techniques to complement and strengthen traditional methods to work with large

geodemographic databases.

69

This study utilized LifeMode and urbanization segments to profile boat owners. A
detailed description of the LifeMode and urbanization segmentation is found in Appendix
A and Appendix B.

3.4.2 Identify and test variables that are statistically related to the ownership of all

boats and to various types and sizes of boats aggregated at county, ZCTA and
census tract level

Correlation analyses were applied to investigate whether associations exist
between boat ownership rates and the hypothesized independent variables, the strength of
these associations, and the form of the relationships for county, ZCTA and census tract
level aggregations. There are two dimensions in correlation analysis: (1) boat type and

size; and (2) different zone units for testing, as shown in Table 3-7, below:

Table 3-7 Dimensions of dependent variables

 

Dimension One
Zonal Structure

 

(1) County
(2) ZCTA
(3) Census Tract

 

All Boats Per 1000 Households (ABH)

All Power Boats Per 1000 Households (APBH)
All Sail Boats Per 1000 Households (ASBH)
Canoes Per 1000 Households (CANOEH)
PWCs Per 1000 Households (PWCH) .
Small Power Boats (<23’) Per 1000 Households (SPBH) owne’Sh‘p

Large Power Boats (23’+) Per 1000 Households (LPBH)

Small Sail Boats (<23’) Per 1000 Households (SSBH)
Large Sail Boats (23’+) Per 1000 Households (LSBH)

 

 

 

Factors Affecting Boat

 

 

 

 

 

 

Dimensmn Two
Ownership by Boat Type and Boat Size

 

 

r

 

 

The scale effect was additionally tested by comparing the models on different
zonal units. According to Wong (1996), when the underlying micro data values are not
uniformly distributed across the spatial landscape, aggregating data using different zonal

Partitioning schemes involves different combinations of spatial neighbors (Wong, 1996).

70

Scale effect is the tendency within a system of modifiable areal units (MAUP) to provide
different statistical results from the same set of data, when the information is grouped at
different levels of spatial resolution (e.g., census tracts, cities, counties). The scale effect
is assessed by comparing the sign (+ or -) and p value of the correlation coefficients when

moving from larger to smaller zones (e.g., county to zipcode and zipcode to census tract).

3.4.3 Develop and test the zonal OLS model to estimate boat ownership

A conventional ordinary least squares (OLS) technique was utilized first to build a
model, so as to estimate the ownership of different types and sizes of boats—including all
boats—at the county, ZCTA and census tract level. The model building process required
many implicit and explicit assumptions and decisions, including variable selection, the

approach for dealing with missing values, and the choice of functional form.

Since the Florida boat registration data was the source of data on boats and their
owners, the number of missing values was not a concern here, as it might have been for
other sources of data (e. g., surveys). However, the registration data still required some
cleaning and modifications. In this study, the numbers of households for certain zipcodes
and census tracts were extremely low, while there were still some boats in these areas.
For example, the census tract “CT007100” in Palm Beach only has one household, while
there are seven boats registered in this census tract. By the current measurement of boat
ownership rates, the number of boats per 1,000 households was 7000. This caused
extreme values (outliers) of boat ownership, and further distorted model estimation. To

avoid such dramatic inflation of boat ownership rates, zones with fewer than 50
households were excluded from the analysis. Table 3-8 provides a list of ZCTAs with

fewer than 50 households. Table 3-9 lists census tracts with fewer than 50 households.

71

Table 3-8 The list of ZCTAs with fewer than 50 households excluded
from the analysis

 

ZCTAs Number of Households
_31629,LT3,mp_a,,,11LWW _, W ,_._,__,W9,_,W W
£§_55L§§1111_M£k§f¥ . ,0 ___ ,
1413,13,Qhokploskgerflum , 0 , W
-3139. _EYEE81§§1€$_C it». ..F L, _ 0 , _
.,_,3,41,4,01,099d1andLFL , V, , ,0
._§,3W1,2_2,LMiaI_niaf_I: W _ __ 0

32013, Day, FL 0

 

Table 3-9 The list of Census Tracts with less than 50 households excluded fi'om the analysis

 

 

 

 

 

 

Census Tracts County Description Number of HHs
W129899901LQOW ,_, __W12986 W ,, CTOOQOOOzMiEIIDi-Déée _WW W W,1_,..
,,___..,_l_298.6910,1,2_5__,, _ , ,_12086_ CI0,10125,Mjami_-,1?a£1§,,, ,, ,, 1

.12123299209. _. ,W1_2,12_3,,,_ ,, ........ CT9902,00,,Iay10L. . 1
___._____1398£5991592______ ,,,,__,,J.ZO8_6W.,_,_, SI 0915,92, MEIPi-QEdL__-__ ,, W 1
___--1299299810L-W _.,_..1,20_-92,__-,W, £1,110,819; 139111133191! W -1.
___WIZQQQOWOHQQWW _W,,_._,,.__,!_2_Q?,9W_ .,.--,W CT007100, Palm Beach WW ,W,.,WL

12119990900 .___ ___ «_lgllgW __ _____,, CT990900, Sumter“ _W, , W W _l__
12073001300 _W, _________13013 __ CT0013QQ,WI_,§9n W W2
12086010702 ,_ _,,,,.1208§WW , 910,197,021MEamLDQdeWW W9.

 

 

 

__-__1_2_,02_79_19290W , WJ 2981 _ ,, 91910900, H1115b010ughWW _, ,, W _ __-JO. .

___,__12,Q9_5911092_W,.___,,,,-, ..,__12925.____ _,WQI_91_3992;Q@£8§W ,__..18

 

 

 

,-1298,7970_1_00,,,_ , ,,,12,08Z ,9127919021‘499199 ,, , , 24,
.-W1_,2,05,1,9_,‘fl1190,,_. _ -12951 ,. , ,__,,-,_,,-_,,,,9T94111002flsnéty . 39
12086004800 12086 CTOO4800, Miami-Dade 46

 

This study utilized a combination of different methods and criteria to identify the
independent variables. First, Adaptative Regression by Mixing with Model Screening
Function (ARMS) (Morfin and Makowski, 2009) was applied to choose the relevant
independent variables from the twenty-eight candidate variables. The selection criteria
used by ARMS is to minimize AIC (Akaike Information Criterion). The following is

sample code in R project to select independent variables:

72

armsCOUNTY<-
arms(data =IC OUNT Y, family=gaussian( ‘identity ’), nbest=5, criterion=b0th, weight=aic)

Then, the stepwise selection method was utilized, with the independent variables
selected by ARMS. All independent variables comprising the models were assessed
examining coefficients, probabilities or robust probabilities, and Variance Inflation
Factors (V IF ). Variables with a p-value over 0.05 and a VIP over 7.5 were excluded from

the model, based on the rules of thumb. The OLS models were built by GeoDa.

The natural logarithm (1n) functional form was utilized to transform dependent
variables. This functional form is frequently used in recreational demand studies (e.g.,
Strong, 1983), in order to satisfy the assumption of normally distributed errors and the
model assumption of homogenous error variance. The other reason for utilizing semi-log
(ln) was to avoid predicting negative (-) values of boat ownership rates. Table 3-10 lists

the transformed dependent variables used to build the OLS models.

Table 3-10 The semi-natural logarithm transformed dependent variables used in model building

 

 

 

 

 

 

 

 

 

Variables Definition

LNABH Natural Logarithm transformation the number of all boats per 1,000 households

LNAPBH _ Natural Logarithm transfoifnat—idnhof the—number of all power boats per {0% W W
households W_ , W,., W _ __W W W__ .WWW,

LNASBH Natural Logarithm transformation of the number of all sail boats per 1,000
households

LNCANOEH Natural Logarithm transformation of the number of canoes per 1,000 households

LNPWCH Natural Logarithm transfbtrhation the number of PWCs per 1,000 households

 

 

 

 

 

J LNSPBH Natural Logarithm transformation of the number of small power boats (<23 ft.) per
W W ,_ W W1,,9901199,S?119'51_5WW W - W , _, W
LNLPBH Natural Logarithm transformation of the number of large power boats (23 it. +) per
_WWWWWWW __13900 households _ _W WWW W WWWWW
LNSSBH Natural Logarithm transformation of the number of small sail boats (<23 fl. ) per
_ l, 000 households _ _
LNLSBH Natural Logarithm transformation of the number of large sail boats (23 ft. +) per

1,000 households

 

73

The next step after model construction is to evaluate model fit and performance
using criteria, including R-Square, Adjusted R-Square and AIC. Model significance is
assessed using F-tests. The degree of model bias was assessed using the (Jarque-Bera
statistic) normality test, which indicates whether or not the model’s residuals are
normally distributed. The diagnostics for heteroskedasticity include the Breusch-Pagan
test and the Koenker-Bassett tests. It is commonly found in zonal regression models that
the errors are not distributed normally and hetroskedasticity exists. This may be due to

the effect of spatial autocorrelation.

Spatial autocorrelation exists when adjacent observations of the same
phenomenon are correlated. In an OLS model, when there is spatial clustering of the
under/over predictions in a certain area, it introduces an over-counting type of bias,
which makes the model unreliable. According to Anselin (1988), ignoring the spatial
autocorrelation of residuals can result in inefficiency or inconsistency of regression
models. Testing for spatial autocorrelation of residuals is now common practice. In this
study, Moran’s I is used to measure the global spatial autocorrelation of the residuals of

the different models._The Moran’s 1 statistic for spatial autocorrelation is given as:

= N ?=12?=1 VVij(xi —X—)((Xj —X)
Biz} WU i1=1<xi " X32

 

I

where N is the number of cases, xi is the variable value at a particular location, xj is the

variable value at another location, if is the mean of the variable, and Wij is a weight
applied to the comparison between location i and location j.

Moran’s 1 provides a measure of the overall pattern of spatial

dependence/autocorrelation or of clustering in the data In order to calculate the Moran’s

74

I it is necessary to construct a spatial weights matrix. This study utilizes a matrix of first

order queen contiguity spatial weights.

The global Moran’s I primarily measures average local variations across all zonal
units. It does not indicate which areas show spatial autocorrelation patterns or the
strength of the correlation in a specific zone. In comparison, LISA (Local spatial
autocorrelation test) is used to describe how residuals are clustered in space and the
statistical significance level of such clusters in each individual zone within a study area.

The formula to calculate local Moran’s I (LISA) is

Ii = Zi Z Wiij
I

where z,- and zj are standardized scores of attribute values for units i and j, and j is among

the identified neighbors of i, according to the weights matrix W I]

This study utilized GeoDa to test and measure global and local spatial

autocorrelation of models’ residuals.

3.4.4. Development of alternative spatial regression models to estimate boat
ownership rates

This study started with a OLS estimation. A series of specification tests were
applied to assess whether there was evidence of spatial autocorrelation, then alternative
spatial regression models—e.g., the spatial lag model, spatial error model, and SAR and
CAR models—were employed to make improvements on the OLS model. The first-order
queen contiguity spatial weight matrix served as a spatial weight matrix for building

Various spatial regression models.

75

As discussed in the literature review, the spatial autocorrelation of residuals
violates the assumptions of OLS and decreases the predictability of the models. So far,
none of the spatial models could identify a fill set of independent variables that captured
both regional variations inherent in the dependent variable and spatial autocorrelations
associated with different regression elements. Different strategies could be used to treat
the spatial autocorrelation of residuals in regression models. The basic regression model
is expressed in equation (3.1). The spatial regression models developed in this study
focused on the treatment of spatial correlation on the error (a) variable. Specifically, this
study formulated a spatial lag model, a spatial error model, and SAR and CAR models
based on the OLS models developed in the previous section. The formulas of these
models were summarized in the book by Bivand et al. (2008). The R code used to derive

the models were also modified from the sample code in the book by Bivand et al. (2008).

There are two types of spatial autoregressive models commonly employed in
spatial statistics: the conditional autoregressive model (CAR), and the simultaneous
autoregressive model (SAR). The major difference stems from how they specify their

inverse covariance function.

The Simultaneous Autoregressive Model (SAR) uses a regression on values from
the other areas to account for spatial autocorrelation. This means that the error terms a are

modeled so that they depend on each other in the following way:

m
8i = z bijei + 8i
i=1

76

where a; are used to represent residual errors, which are assumed to be independently
distributed according to a Normal distribution with a zero mean and a diagonal
covariance matrix 28 with elements 0' £2 , i = 1, . . . , m (the same variance 0' £2 is often

considered, though). The b if values are used to represent spatial dependence between

areas. b it must be set to zero so that each area is not regressed on itself.

The Conditional Autoregressive Model (CAR) specification relies on the

conditional distribution of the spatial error terms. In this case, the distribution of e i
conditioning on e- i (the vector of all random error terms minus e ,- itself) is provided.
Instead of the whole e- ,- vector, only the neighbors of area 1', defined in a chosen way,
are used. A simple way of representing the conditional distribution of e i is
2
ei l ej ~i~ «4.2" 2'

i~i Cii j~i Ci}

where Cij are dependence parameters similar to b 1]

Spatial error models and sometimes spatial lag models are referred to as the
simultaneous autoregressive model (SAR). Bivand et al. (2008) separated the spatial
error model and spatial lag model as being spatial econometrics approaches; this is

authoritatively described by Anselin (1988).

77

The general expression for a spatial lag model can be expressed as:

y = pWy + xfl + 8
where Wy is an N by 1 vector of lagged dependent observations, ,0 is a spatial

autoregressive parameter, and e is a N by 1 vector of independent disturbance terms. In

spatial lag models, the spatial autocorrelation is modeled by a linear relation between the

response variable (y) and the associated spatially lagged variable (Wy). Spatial lag is

suggestive of a possible diffusion process: events in one place predict an increased

likelihood of similar events in neighboring places.

The spatial error model can be expressed as:

y=xfi+£ with £=1lW8+u
where y is an n by a vector of observations of the dependent variable; x is the n by k
matrix of observations on the explanatory variables; ,6 is the associated k by a vector of
unknown parameters; a is the n by one spatially correlated error vector; A is the spatial
autoregressive coefficient; W is the spatial weight matrix, indicating how the locations in
the dataset are spatially interconnected; y is a spatially uncorrelated error term that
satisfied the normal regression assumption. If there is no spatial correlation between the
errors for connected observations i and j, the spatial error parameter I. will be 0, and the

model reduces to the OLS model in the conventional manner.

The predictive improvement associated with using the spatial regression models,
compared to OLS models, was evaluated by comparing R square, log likelihood and AIC

coefficients and their significance levels.

78

CHAPTER FOUR RESULTS

This chapter first provides a descriptive overview of boat ownership in Florida,
focusing on the spatial distribution of ownership of different type and size boats.
Correlation analyses are utilized to identify and test variables that are statistically related
to boat ownership for three different zonal levels. Next, Ordinary Least Square (OLS)
regression models were developed to estimate boat ownership at county, ZCTA and
census tract levels. Spatial regression models, including spatial lag, spatial error, spatial
autoaggressive models (SAR) and conditional autoaggressive models (CAR) were then
built by using combinations of the independent variables comprising the OLS models.
The OLS models for different types and sizes of boats were then compared. The final step
in the analysis was to compare the OLS models with the spatial regression models.

4.1 Describe the current recreational boat ownership patterns in the State of
Florida, including the spatial distribution of boat owners

This section examines the characteristics of Florida’s registered boat owners,
including their genders and ages. These two variables were included as part of the
information that owners provide when registering boats in Florida (FWC, 2007). The
addresses of registered boat owners were geocoded in order to attach ERSI’s LifeMode
and urbanization segment membership (see Chapter Three). Geocoding also provided a
means for spatially distributing boat owners and determining the distance from registered
boat owner residences to the coastline. All of these variables were considered for possible

inclusion in the OLS and Spatial Regression models.

79

4.1.1 Boat ownership by boat type and size

As already described, power boats of various sizes comprise three quarters of all
boats registered in Florida. About 80% of registered power boats are less than 23 feet in
length. PWCs comprise almost 12% of the registered fleets. This indicates that a very
high percentage of boats registered in Florida are capable of being transported by trailers
to public and private launch sites. Many of them were also stored at non-waterfront
residences, meaning that public access to boating waters is an important factor in

determining ownership and participation.

4.1.2 Boat ownership by gender

The State of Florida allows boats to be registered to one male owner, one female
owner, or joint male and female owners. It is not surprising that the vast majority of all
boats were registered to a male owner. Table 4-1 shows the gender of the owners of
different types and sizes of boats. About 18.6% of the boats were registered to joint
owners, whereas only about 9% of registered boats were registered to women. A much
higher percentage (16%) of PWCs was registered to women owners. However, it must be
noted that many of the boats registered to a single male or female owner were, in fact,

family-owned and -used boats.

Table 4-1 Gender of re istered boat owners in Florida

 

 

Boat Type All Small Large All Small Large
, . . . All
nd Size power power power sail sail sail Canoes PWCs boats
Gender boats <23’ 23’+ boats <23: 23:4.
_Male Owners W70.6%W12W5% ”63.3%W 67.2% 77.4%_»___63.l% 83.9% 72.3%W 72.1%

Male Female Jointly

 

Owners 21.0% 18.9% 29.3% 25.0% 13.5% 29.7% 6.9% 11.5% 18.6%
Female Owners 8.3% 8.6% 7.3% 7.7% 9.1% 7.2% 9.3% 16.1% 9.3%
Total 100% 100% 100% 100% 100% 100% 100% 100% l00%

 

 

8O

4.1.3 Age distribution of registered boat owners

Boat owners were older than general population in Florida. The median age of
registered boat owners was 51 for all boats, which was 12 years older than the median
age (39) of all Florida residents in 2007. Table 4-2 indicates that almost three quarters

(72%) were between the ages of 35 and 64 compared to 40% of Florida’s population.

Table 4-2 Age distribution of registered Florida boat owners

 

 

 

_Age % of boat owners % of Florida Population in 2007
24595. _ __-_ _W. _- 194%--.. _- W _ WEE-9?:
35-64 71.9% 39.1%
'33 f h ”136% 7 “77.0%
total 1 00.0% 100.0%

 

The age distribution of owners varies across different types and size boats. Table
4—3 shows the average and median ages of owners across different type and size of boats.
On average, the owners of registered power- and sail- boats were older. The owners of
sail boats, and especially larger sail boats (23’+), were much older on average (57) than
the owners of other types and sizes of boats. As would be expected, PWC owners were
younger. The fact that boat owners were generally older than the general population is
causing concern over the ability to create boating growth in the future. According to a
recent study conducted by the FWC (2009), boating participation is projected to remain

constant or decrease for most counties in Florida through 2025.

81

Table 4-3 Average age and median age of owners of different types and sizes of boats

 

 

 

 

 

 

Type Size (feet) Average—Age Median Age

Power Boats A1901? Boats”. .. __w_,-5.2;_2,-_m,_ ._____._.5_2._.
_§m@_ll_novst<,2§1 ___“ ...__5.2_-.L _ ___._____ -52 _
Large power 23’+ 52.7 52

Sail Boats gilljaflfiati ,_ 363m 57
.§m§l|§aiL<2}?_ a -ééfi- ,,_‘ -56
Large sail 23’+ 57.1 58

Canoes 52.4 52

PWCs 45.2 45

Total ' 51.6 51

 

4.1.4 LifeMode segments of registered boat owners

Registered boat owners were identified using ESRI’s Community Tapestry’s
LifeMode segments, based on the geocoded location where they resided. Again, this was
the owner’s residence and not necessarily where the boat was stored during the boating
season. As described earlier, these LifeMode segments include the following
characteristics: age, race diversity, income and household structure. A description of
ESRI’s LifeMode segments is provided in Appendix A. Table 4-4 shows the percentage
of owners of different types and sizes of registered boats assigned to each LifeMode
segment and, for comparative purposes, the percentage of the Florida population

associated with these segments. Appendix B shows the figures associated with Table 4-4.

82

3.5% 2:50 5 88 fiesta S $.23 2:03 moss . .
. . . Egon m D 05 52w :omono onvSt 0303 95 35 b. _ a ca
. . . . . . :88on
8 mumméwwumwsmmwwfihfi 0.58 .05in 2: WEE?» hon. Abigale 82953 02 2 9%va 05 o Eat mow—HE xoufiffldgofi 25W ”cmwwmwmwwmmwm
5 ob E85: 3 58:0 magnum 95 $5 coca—8:— 05 @597. :09: 2:. 553:. 0:58 Ea Eon... Ewing—Ea mow:— bfiozo 2: ..N
mEocomso: go 68:12 ”I: ._

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

  

 

 

$5.3m _v . £2 . . . . . .
_ am :25: $_ 2 $o 2 $o o $m 2 $N 2 $m 2 $SN $N.: $2: $3. :50 §EE<
22: oz?
vmmdvm mm 33 8.95 o . . . . . . .
35% i. _ $o N $o _ $_ : $9 _ :5 o $« N $3 $nN $N.: was“: a 3:92;
oNo 2% mm mm H52 $3 $3 $2. $3 $5 $v.m $2 $3 $N.:. $_ .N was: 3:36“;
:3, . . . .
$33 on N: 3 95 o . . . . . . .
8"va i w $0 o $N M $» v $N m $_ n $o o $3 $3. $3 gate: 3E5
m I I I
”WWW“ 8 0 MN :2 NEE“. $5 m $m N $N.: $o.N $3 $o.N $3 $3 $wN $5 38: _820
. 8: . . . . . Ii
2§Nm MN m 3 u .2 $N m $_ m $v N $o N $m N $N.: $hN $nN $oN $3 3%: =9:
”INN OX— 0 I I I I I I I I I
2%: Nu? \% o $0 o $c o $5 o $0 0 $N o $v o $v o $§ $_ ._ 32%: a. £35m
camwvm wm 33 we use a . . . . . . .
Bram 3 MN $m mm $m a. $5 mm $o NV :5 mm $m Nm $2M $mNm $2; 33m 82%
0:..va e . Baa . . . . .
o o mm .835 $_ m $v N $_ 0 £5 v $.u. m $9M $N.N $m.N $9N $2 92 28
83a . 2: . . . . .
S w x mph”: $v _ $v : :6 N $m _ $0 : $2 $2 $3 $3 $3 £8852
ONE; S N. 0.95 . . . . . . .
am vwEwwc $o 2 $N : $2: $m : $c : $« 9 $2: $:.: $N.: $2: 8:82 283:
855E
25.35 am m. o 95 . . . . . .
ow LESN $v : $m : $c N: $N : $N : $m : $N.: $m.N_ $5.2 $3 968 E:
2:85
I: 5?: uw< 0% +3 m auto—m nae—5m
_ bNEoZQ 5:52 F _:m 623 8980 a mmv Eon +m~ mmv anon 589 E m I: owe—2&5
Emu”: . . _. m _. m 1“” hogan MOBOQ hOBOL =< . — DNmm “VF—d
mo o\e 09C. 8cm
355 a 08205 Easing 38m 3. 29¢

 

 

83

It is no surprise that the greatest percentage (30.1%) of Florida residents fell into the
ESRI’s Senior Styles LifeMode segment. An even higher percentage (42.3%) of large
(23’+) sail boat owners fell into this segment, followed closely by the owners of large power
(23’+) boats (3 8%). The median age of this segment was 49; they were generally empty
nesters with median incomes of around $45,400. The low race diversity score indicates that
boat owners were more likely to be white non-Hispanics, which is consistent with previous

boat-owner studies.

A relatively high percentage (20.1%) of power boat owners were members of the
American Quilt segment. This was double the percentage (9.8%) of Florida households
categorized as American Quilt in 2007. Table 4-4 and Appendix A reveal that the average
age of American Quilt members was 39, with median incomes of $45,729 and mixed family
types. They are more likely to either be workers in manufacturing, agriculture, local
government, service, construction, communications, or utility workers, or part of an older
population that is retiring to seasonal vacation spots. Relatively fewer boaters were members
of the Metropolis and Global Roost LifeMode segments. The Metropolis Segment includes
those living in larger cities; they tend to own fewer vehicles (and presumably fewer boats)
and to rely more on public transportation. Their lifestyle is also uniquely urban and media-
oriented. The Global Roots segment is characterized by high ethnic diversity, and their
households are younger. Global Roots have modest incomes and tend to rent in multiunit
dwellings. The life-style and -stage of Metropolis and Global Roots members greatly

influence how likely they are to own a boat and participate in boating activities.

84

4.1.5 The proximity of registered boat owners to the coastline

It is logically assumed that proximity to water would encourage boat ownership for
many reasons, including exposure to recreational boating activities, easier access, and
reduced time and travel costs. As already described, the distance was calculated from every
registered boat owners’ residence to the Florida coastline. The coastline includes Florida’s
tidal shoreline and rivers, streams, and waterways such as canals (see Figure 3-14). The
results of these analyses, reported in Table 4-5, show that half of all boat owners resided
within 2 miles of the coastline, and almost two-thirds (64.7%) within 6 miles, in 2007.
Appendix F shows the figures associated with Table 4-5. As might be expected, the owners
of larger power- and sail-boats resided closer to the coastline in large part because their
boats could not easily be trailered, and are usually stored in marinas or yacht clubs. About
two-thirds of the owners of 23’+ power boats lived within two miles of the coastline. The
owners of more transportable/trailerable power boats, less than 23’, resided further from the

coast. About 44% resided more than four miles from the coast.

85

Table 4-5 Proximity of boat owners to the coastline

Distance to water

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

(Miles) 0.5 1 2 4 6 8 10 10 +
Boat Type and Size
% 32.8% 7.8% 9.2% 3.8% 5.6% 3.4% 2.2% 29.7%
All Boats "TH—TM” - -~—-4~ ..---_.._- ~~—_ .4%—4..-.-- -_-- ..... “---“--.M- ..___----
% 32.8% 40.5% 49.7% 59.1% 64.7% 68.1% 70.3% 100%
P b % 33.3% 7.6% 9.0% 3.8% 5.6% 3.4% 2.2% 29.5%
ower oats ~ — -- -—«~—— ———~—~— - ~ '———--—— ~~~ --——————~— —-~
0/01 33.3% 40.9% 49.9% 59.3% 64.9% 68.3% 70.6% 100%
P <23 % 29.2% 7.8% 9.2% 4.0% 5.8% 3.6% 2.4% 32.4%
ower ~—— «— - —— 4 .-.. -_ ._ ..--._.
0/01 29.2% 37.0% 46.2% 55.9% 61.7% 65.3% 67.7% 100%
P 23 % 49.0% 7.0% 8.2% 3.4% 4.9% 2.8% 1.6% 18.3%
ower + ---- .. . -.. ---. --_. -..-- -. m--- . --.- .. . -- . _.-.._..__ A..-“ _._ -.....-. _. . .-.__..,.- .- -.fl ._ M.--—
0/01 49.0% 56.0% 64.3% 72.4% 77.4% 80.2% 81.8% 100%
S 'lb % 51.5% 8.3% 7.5% 2.6% 3.5% 1.8% 1.0% 19.9%
a] oats -7—-_ 2 4 4-4 - ..._ .4 -___- -2--. ....,.A -_. ._ .--..-_.-_-._. -._.--- «gm—.2 *._ -41- n--- - — - ...
0/01 51.5% 59.7% 67.3% 73.7% 77.2% 79.1% 80.1% 100%
Sail % 43.2% 9.3% 8.6% 3.1% 4.2% 2.3% 1.5% 23.5%
<23 %1 43.2% 52.5% 61.2% 68.6% 72.9% 75.1% 76.6% 100%
Sail % 54.8% 7.9% 7.1% 2.4% 3.2% 1.7% 0.8% 18.6%
23+ 0/01 54.8% 62.6% 69.7% 75.8% 79.0% 80.6% 81.4% 100%
C % 27.6% 8.6% 10.5% 4.4% 6.9% 3.9% 2.8% 28.8%
anoes _- _ .. . ___..- ___- ___. ___..-.. ....-_. -_..--- ___..- -_---.--_.
%I 27.6% 36.2% 46.7% 57.5% 64.4% 68.3% 71.1% 100%
PWC % 32.5% 8.1% 9.9% 4.1% 6.2% 3.8% 2.5% 26.8%
s . .. .. _. -_--.__ --- ,. ...-_ .. --_-_--..-- -..- _-- _,-- -- --_
0/01 32.5% 40.6% 50.6% 60.8% 67.1% 70.8% 73.3% 100%
{2:32:33 0f %1 - 14.6% 25.7% 52.4% 54.6% 59.5% 61.8% 100%
Percentage of 1
POP % - 25.5% 52.2% 76.1% 77.1% 79.2% 80.2% 100%
Percentage of
0/01 - 9.7% 20.9% 30.9% 31.4% 32.3% 32.7% 100%

HH

 

 

 

1. Cumulated percentage
2. The Florida population information is for year 2007
3. The Florida household information is for year 2007

4.1.6 Urbanization segments by registered boat owners

It was assumed, based on the literature, that urbanization would be a deterrent to

boat ownership because it is more difficult for persons living in urban areas to store and

transport boats. Also, urban residents generally have less exposure and fewer opportunities

86

to participate in outdoor recreation activities such as recreational boating (Arlinghaus et al. ,
2008). To examine the relationship between boat ownership and urbanization, registered
boat owners were assigned to one of eleven ESRI urbanization segments, based on the
location of their primary residence. A brief description of these segments is found in both
Table 4-6 and Appendix B. Appendix G shows the figures associated with Table 4-6. The
table also reports the percentage of Florida’s population that comprised these same

segments.

The results show that boat owners generally resided in suburban and rural areas, and
much less so in urbanized areas. About 2% of registered boat owners were located in one of
two principal urban area segments, compared to 5.4% of all Florida households in 2007.
About a quarter (24.1%) of Florida households were in Metro Cities (1 and II), in contrast to
14.5% of registered boat owners. A higher proportion of large power- and sail-boat owners
resided in metro areas. This was probably because they stored their boats in large, nearby,
urban waterfront marinas (e. g., Miami, Fort Meyers and Tampa), including ones associated
with condominium residences. In contrast, about 11% of Florida households resided in rural
areas, compared to around 21% of the boat-owning households. Almost a quarter of small
power boat owning households resided in rural areas. By far the highest percentage (34.6%)

of registered boat owners resided in the suburban periphery.

87

3:8» 0.5.00 :0 008 20:05.0 8 w:o.0n 0.:95 5.3.2.9. .w.D 05 :8... 58:0 5.8005: 0.000: :3. .0... b.3002: 300.0: .0 0 m. 0:0... 0.50:.
5...? ..0 m. .0: 0... 8.. 080m b.0030 0:. 6.9536 8... 00.0030 0.0.058. oo. 8 3.0020 05 0 E00 mow—a. .805 0:... 3:8» 2:50 :o 000.. 50.0.0.0 8

90.0: .003 0:5 0... ES. E005: .0 :0mo..0 .8800 0.5 .05 085.00... 05 9.6:... .305 0:5 5.0030 0.500 0:: 3.00.. 0.8.8683 50:. 3.235 0:... .N
0.9.8.5.. .0 50:52 “I... ..

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

355:...
500.50% om wdm 0.0.50 $3.m $.00 $3.. $.0m $.o.N $.N.N $.m .5 $3.0 $.50 $3.m .. .33.
00.5.3.
8...:5“.
80.30% 5N 003 0.950 $.0N. $.0m. $.56 $.N.N. $.00. $.00. $.00. $6.0... $3.3. $.05 . .33.
00.5.3.
.33ng 35 ”.03 00.0.). $6... $.N.m. $.05 $.00. $3... $.05 $. . .m. $0.2 $.o.N. $.06. 330% ..0Em
30. mm 03 3.... 00.2.). $.50. $.03. $...m. $.03. $3.2 $.56 $.0N. $..... $.5... $...m. 5:000:00 50:5...“
SEEN...
050.2% .3 ..o.. 0.950 $...3m $.o.0N $.N...m $3.3 $6.0m $5.0m $3.0N $.. .3 $.o.wN $.0.N . 5.0.3.0.. c.3055
00.5.3.
08de 05 ...m 005.). $5.. $.o.N $5.. $6.. $..... $3.. $.. .N $6.. $.o.N $.00 .. 8.0.830 :35
005.50% m0 3m 00.22 $.m.w $.05 $.. .0 $.00 $.00 $.00 $3.0 $3.0 $.00 $3.0 . 3.0.8.5 :35
350N3m N5 .m 00.02 $.52» $.0.5 $.N.o. $..; $.00 $.. .w $3.0 $.00 $.0.5 $.00. .. 80.0 000.).
m0:05...
08...; mm ..am 0.9.50 $.N.o $6.0 $.06 $.0.5 $.00 $.N.. . $.00 $.05 $.05 $.N... . 8.0.0 8.05.
00.53.
. . .22 . . . . . . . . . . .. 88:00
0mm. 5Nm on v ”N 30:0... $.. . $3 o $... . $5 0 $6 0 $0 0 $0 0 $5 0 $... 0 $.m m 55:: 30.05:.
. . . . . . . . . . . . 22:0
0.5 30% .0 a 0m 00.22 $.5 . $.m o $... m $.m . $.m N $.N N $... 0 $.. . $.m . $.. N :35 “8.0:.“
058:. 0 00.8.... 3:0Ew0m
x :
1.. N 0 . 0w< 0%... I: no.5... 80:00 +MN va 3:2. +MN va $8.. $09“. :. m I: 5.33555
. 5.8020 5.00.). . ..0m ..0w ..am 030.. .0304. .039. ..< . 03m 0:0
5.003. . . .o $. 0%... Sam

 

 

3:2: 0m 5.03.53...

 

 

 

. N... :00...

Ea... 380 0-.. 6.2.:

88

4.2 Identify and test variables that are statistically related to the ownership of
all boats and to various types and sizes of boats

The analyses in section 4.1 are based on individual boat owners, using data from
the 2007 Florida boat registration records. This section provides the results of the
statistical analyses of relationships between various independent variables and boat
ownership aggregated at county, ZCTA and census tract levels. The variables examined
include: age, gender, race, household structure, income, seasonal home ownership,
proximity to water, and number of marinas and boat launch facilities. Information about
registered boat owners was used, together with population estimates, to derive the
number of boats per thousand households in counties, zipcodes (ZCTAs) and census
tracts. The results are reported for each of the following boat type- and size-segments for
each of the three zonal units: all boats (ABH), all power boats (APBH ), small power
boats ( <23’) (SPBH), large power boats (23’+) (LPBH), all sail boats (ASBH), small sail
boats (<23’) (SSBH), large sail boats (23’+) (LSBH), PWCs (PWCH) and Canoes

(CANOEH). Table 4-7 reports the results of the statistical tests.

89

Table 4-7 Correlation between boat ownership demographics and boating supply for counties,
ZCTAs and census tracts

 

 

 

 

 

 

 

 

 

 

 

Boats by 33:31 Counties
Factors ABH APBH ASBH QANOEHPWCHISPBH LPBH §§BH LSBH
Age ‘ 1 1 E 1
AGEP .582 .5521 .512 08‘ .512. .492‘ .512i .532? .492
AGEHP .332 ,27‘ .17 -.11 ,27' ,231 .13‘ .22‘ .15
Gender
MALEP _352 .261 -.05 -.04 -.02 .332 -.03 -.06 1 -.05
Race ‘ ‘ 1
WHITEHHP .3421 .362‘ .20 .322} .3111 .372, .19, .3223 .17
BLACKHP -.16 -.23 -.261‘ -.16‘ .3321 -.18} -2911 .28" -.251
ASIAHP .472 -,422‘ -.01 .261 .031 -.492‘ -.07 -05 .00
HIHP ..301 -251 -.03 ..261 ; -.08 ..322 } .021 -.l6. .00
Family . ‘
FAP .04 -.01 __342 .01, -.06‘ .12‘ -.342: __zg‘i .342
FAMP .09 .11 -.06 .17‘ .19 ‘ .16 -.06 -.03 i -.07
FAMCP -.03 -.10 -.24 -.22‘ -.00 -.02. -291 -.19 -.241
FAFEP -.12 -.23 -.362 ‘ -.24 -.382 -.13 -392 __332 -.362
Income 1
INCIP 4301 -.22 .332 -.231 .29: 437: .29: .251: .34:
INCZP .29]: -.20 .33 ‘ -.23 .25 1 -.362. .312 .24 .342
INC3P -.23 -.19 .31 -.22. .221 __35 .32 .22 _33
Seasonal House 2 2 2 2 2 2‘ l ‘ 2
HOSP .45 .47 .34 1 .03, .42 .40 i .47 .30 1 .34
Urbanization 1 ‘
URBANAP -.502‘ ..442 .04 -.223 -.00 __542 -.02 .02. .05
POP_DEN .45 -.412. .01, -.20‘ ~06 -.492‘ -.041 -.o4‘1 .02
POP RP .47 .362 .251 .322 -.19 _522 -.23 -.09 ..281
Proximity to Water ‘ .
WATERP .30l ; .382 .662 .07 .522 ‘ .22 .652 .592 .662
Marina . 2. l: .‘ 2‘ 21 2
MARINA .01 .101 .57 ‘ -.21: .29 . -.11‘ .63 1 .40 z .60
MARINAIM .04 .13 ,592 -.20‘ .301 -.08 .652 .412 .622
MARINA3M .06 .15 .612 -.13 .301 -.06. .662 .422 .632
MARINASM .03 .16 .622 -.17 .322 -.04 .672 .442 .642
Boat Launch Sites ‘
BOLA .14 .18 .372‘ -.08‘ .382: .091 .362 .352 .362
BOLAIM .16 .20, .33: 403‘ .39: .11‘ .39: 35: .38:
BOLA3M .17 .21 .40 —.05 .37 .11 .40 36 ,40
BOLASM .17 .21 ,4o2 -.05 _372 .11 _402 .362 .402

 

 

1. at a 95% confidence level 2: at a 99% confidence level

90

Table 4-7 Correlation between boat ownership demographics and boating supply for counties,

ZCTAs and census tracts con.)

 

 

 

 

 

 

 

 

 

 

 

 

Boats by Zonal
Units ZCTAs
Factors
ABH APBH ASBH CANOEHPWCH SPBH LPBH SSBH LSBH
Age 2 2 2 2 2 2
AGEP .42 .41 .32 .261‘ .43 .362 .41 y .28 .30
AGEHP .15 .13 -.01 .08 ‘ .25 ‘ 13 .06 03 -02
Gender ‘
MALEP .16 .142 .04 .04 ‘ .08] 17 02 .00 02
Race 2 2 2 2 2 2 2 2‘ 2
WHITEHHP .362 .372 .232 .29 .282 .372 .232 .282 .202
BLACKHP -.222 -2321 -.162 -.19 “29" -.202‘ -.232 41621 -.152
ASIAHP -.322 -.322 —.122 -.19 -.082 -.342 —.14] -.122 -.102
HIHP -.26 -.26 ‘ -.14 -.21 ‘ -.13 1 -.29 -.08 ‘ -.15 1 -.12
Family 2 2 2‘ 2 1‘ 2
FAP .122‘ .11 -14 .09 ‘ .152 .162 -072‘ -06: ~16
FAMP .25 .24 .03 .17 1 .312. .26 122 .0720 0(2)
FAMCP -03 -03 ..202 -.03 .102 -03 -.162 -.122 -.20
FAFEP -.28 -.29 -.26 -.19 -.33 1 -.24 1-.33 -.21 -.25
income ‘
2 2 2
INClP .02 .03 .142 -.01 .34 -.061 .262 10 13
INC2P .03 .03 _162 -02 .34 -06 ,312 11 16
INC3P .03 .03‘ .17 -04 .32 - 07 .33 11 17
Seasonal House .
2 2 21 2
HOSP .33 .34 .32 .11 .24 27 .45 1 22 32
Urbanization 2 2 1 2
URBANAP -.45 -.44 .01 -.32 -.11 -.53 -.02 -09 05
1 2 2
POP_DEN -.392 -.382 --08 -.27 -.21 —.43 -.09 -.13 -.05
POP RP .45 .43 -.03 .33 .11 53 -0.02 10 -.08
Proximity to Water 2 2 2 1 2
WATERP - 1o —.09 .16 -.12 .01 - 16 15 .05 18
Marina ‘ ‘ 1
2 2 2 2
MARINA -071 -.06 .192 -.15 .01 -.15] .242 05 .222
MARINAIM 0% 03 .262 -11 .03 -.o7l‘ .302‘ 10 .292
MARINA3M .18 .192 .412 -.04 17 .082 .4621 .20 1 .442
MARINASM .29 .30 ‘ .50 .02 23 19 54 .26 53
Boat Launch Sites 2 2 2 1 2 2 2 2 1 2
BOLA .22 ‘ .21 12 .06 .14 .22 .11 .11 .11
2 2 2 2 2 2 2 2 2
BOLAIM .30 .28 .16 .11 .17 1 .29 .16 .15 .14
2 2 2 2 2 2 2 2 2
BOLA3M .40 .38 .22 .15 .21 .39 .23 .19 .20
2 2 2 2 2 2 21 2
BOLASM .40 ‘ .39 .24 .14 .21 39 .27 19 23

 

 

l. at a 95% confidence level 2: at a 99% confidence level

91

Table 4-7 Correlation between boat ownership demographics and boating supply for counties,

ZCTAs and census tracts Scon.)

 

 

 

 

 

 

 

 

 

 

 

 

 

Boats byZonal
Units Census Tracts
Factors
ABH APBH ASBH CANOEHPWCH SPBH LPBH SSBH LSBH
Age 2 2 2 2 2 2‘ 2 2‘ 2
AGEP .44 .432 .29 .282 .452‘ .39 .392‘ .29 ‘ .27
AGEHP .20 .18 .02 .14 .28 .19 1 .11 .05 .01
Gender 2 2‘ 2 2 2 2 2 2‘ 2
MALEP .23 .22 .10 .12 .17 .23 .13 .08 .10
Race 2 2 2 2 2 2 2 2 2
WHITEHHP .382 .402 .252 .382 .302 .422 .242 .282 .222
BLACKHP -.282 -.282 -.l621 -.262 -.3011 -.2721-.232 -.19 1 —.142
ASIAHP -.17 -.17 1 -09 -09 ‘ -.04 -.18 ‘-.10 ‘ -.09 ‘ -.09
2 2 2‘ 21 2 21 2, 1 2
HIHP -.21 1 -.22 -.15 1 -.24 -.09 -.27 -.O6 -.17 -.13
Family 2 2 2 2 2 2 2 2
FAP .162 .152 -.1311 .1323 .2121 .192‘ 0%? -.062] -.14
FAMP .342 .332 .042 .272 .382 .352 .21 .10 ‘ .03
FAMCP .08 .06 -.14 .05 ‘ .20 1 .09 1 -.031 -.08 1 -.15
2 2 2 2 2 2
FAFEP -.33 -.33 -.24 -.26 -.33 -.31 -.31 -.24 1 -.22
Income 2 2 2 2 2 2 2‘ 2 2
INCIP .23 .23 ‘ .13 .16 1 .39 .16 , .30 .1421 .122
INC2P .242 .232 .162 .132 .40 .15 .34 .152: .142
lNC3P .22 1 .22 .17 .09 1 .39 .12 1 .36 ‘ .15 ‘ .16
Seasonal House 2 ‘ 2 2 1 2 2 2 2 2 2
HOSP .28 .29 .26 .07 .17 .22 1 .36 .21 .25
Urbanization 2 2 2 2 2 2 2
URBANAP -.39 ‘ -.39 -.02 -.31 -.132 -.50 :-.06 -07 .03
POP_DEN -.372 -.37 -.12 -.312 -.25 -.43 -.16 -.17 1 -.09
POP RP .39 ‘ .38 -00 .33 ‘ .14 1 .49 ‘ .05 .05 1 -.02
Proximity to Water 2 ‘ 2 ‘
WATERP -03 -.03 15 -.05 -.02 - 10 13 .18 14
Marina 1 1
‘ ‘ 2 2‘ 2. 2
MARINA 04; .05; .19: -.13 .03 -.06 242 .1021 .192
MARINAIM .12 .12 .25 -09 .08 .02 31 .15 1 .26
2 2 2 2 2 21 2
MARINA3M .29 1 .30 1 .39 001 .19 .19 45 .25 1 .39
2‘ 2 2‘ 2‘ 2 2 2
MARINASM 44 45 48 10 .25 36 54 34 48
Boat Launch Sites
2‘ 21 2 21 2‘ 2
BOLA .28: .28: .15: .152 .172 .312 .132 .192 132
BOLAlM .36 .35 .17 .18 .22 39 1 .18 .22 : .15
2 2 2 2 2 2 2 2 2
BOLA3M .442 .4321 .192 .232 .2421 .482 .222 .222 .162
BOLASM 43 ‘ 43 19 22 .21 47 22 21 16

 

l. at a 95% confidence level

2: at a 99% confidence level

92

4.2.1 Age

The proportion of residents between the ages of 35 and 64 (AGEP) and the
proportion of households headed by persons between 35 and 64 (AGEHP) are the two
age-related variables that were examined. The variables are significantly positively
correlated with boat ownership for most of the types and sizes of boats in each of the

three zonal units.

The proportion of persons in the 35 to 64 age group (AGEP) has a stronger
correlation (i.e., size of the correlation coefficients and confidence levels) to boat
ownership than the proportion of the heads of household (AGEHP) within this age
bracket. Other than canoe ownership (CAN OEH) at the county level, the proportion of
residents between the ages of 35 and 64 (AGEP) is positively correlated with the
ownership of all types and sizes of boats, with a 99% confidence level. In contrast, canoe
(CANOEH) and sail boat (ASBH, SSBH, LSBH) ownership are only weakly correlated
to the proportion of households headed by persons between 35 and 64 (AGEHP) for all

three zonal levels.

4.2.2 Gender

As can be seen in Table 4-7, while there is a correlation between the proportion of
the population that is male and owners of (all) boats in all three zonal units, the degree of
correlation varies across boat type- and size-segments. The proportion of males in the
population is most highly correlated with the ownership of smaller boats and, not
surprisingly, small powerboats, which are most often used for fishing. There is either
weak or no correlation between sail boat ownership and gender. The results indicate that

the ownership rates for all sail boats (ASBH), small sail boats (SSBH), and large sail

93

boats (LSBH) show no statistically significant correlation with the proportion of the male
population at the county level or the ZCTA level. The lack of correlation may be due, in
large part, to the fact that most boats were owned jointly by women and men (i.e.,
spouses or significant others) living together. The registration data did not indicate who
selected the boat or even who operated it most often. Therefore, the data concerning

gender on the boat registration data may be inexact.

4.2.3 Race

The proportion of White Non-Hispanic households (WHITEHHP) is positively
related to the ownership of most types and sizes of boats, at a 99% confidence level.
Surprisingly, the ownership of all sail boats (ASBH), and especially large sail boats
(LSBH), as well as large power boats (LPBH) show no statistical relationship with the
proportion of White Non-Hispanic households in Florida counties. This is not consistent
with the literature, which indicates a disproportionately high incidence of boat ownership
among white persons. As already reported, large boat owners who are predominately
White Non-Hispanics, are more likely to live in urban areas. Since this analysis involved
aggregating boat owners by various zones, the ownership rates, in terms of number of
boats per 1000 households, decreased when population increased in an urban area. This
may cause the inconsistent relationship between the proportion of White Non-Hispanic
households and boat ownership rates. Other races and ethnic groups (e. g., Black, Asian,
Hispanic) are either negatively correlated or have no significant correlation to boat
ownership at the county level. As would be expected, small power boats (SPBH), many
of which are used for fishing, are the most strongly correlated with the proportion of

White Non-Hispanic households. Sail boats, especially small sail boats, are not correlated

94

to White Non-Hispanic (WHITEHHP), Hispanic (HIHP) or Asian (ASIAHP) households
at the county level. Boat ownership is inversely related to the proportion of Asian and

Hispanic households at the ZCTA and census tract levels.

4.2.4 Household structure

Four different variables are employed to represent household structure. They
include: proportion of family households (FAP) in the zonal units; the proportion of
married family households (FAMP); the proportion of married family households with
children (FAMCP); and the proportion of female-headed households (FAFEP).

Not surprisingly, the proportion of female-headed households (FAFEP) is
negatively related to ownership of all kinds of boats at all zone levels. This was expected,
given that female-headed households, defined as “a family with a female maintaining a
household with no husband of the householder present,” generally have lower incomes
and less historic involvement with boating than married households.

Proportion of family households (FAP) and proportion of married family
households (FAMP) do not show any relationship with boat ownership. The results also
show that the relationship between FAMCP (the proportion of married family households
with children) and boat ownership rates is not uniform across different types and sizes of
boats. The definition of a married family household with children states, “the householder
and his or her spouse are enumerated as members of the same household with one or
more child(ren) under 18.” The proportion of married family households with child(ren)
is negatively correlated with sail boat ownership (ASBH, SSBH,LSBH) and large power
boat ownership (LPBH) at both the ZCTA and census tract levels. The proportion of

married family households with child(ren) (FAMCP) shows no significant correlation

95

with the ownerships of sail boats in general (ASBH) or small sail boats (SSBH) at the
county level. Other types and sizes of boat owners, especially PWCs (PWCH), are
positively related to the proportion of family with children (FAMCP) at census tract
levels. This is in large part due to the younger age of PWC users, as compared with the
owners and operators of sail boats. According to Table 4-3, the average sail boat owner’s
age is about 57. A high proportion of them are likely empty nesters. PWC owners, on the
other hand, have an average age of 45, and a higher proportion of them reside in

households with children.

4.2.5 Household income

Previous research reveals that the initial purchase price and the cost of boat
storage and maintenance are a barrier to boat ownership. The research also shows that the
household income of boat owners varies considerably across different types and sizes of
boats (Growing Boating, 2007). Three different income classifications were utilized to
test the relationship between income and boat ownership: > $50,000 (INClP), >$75,000
(INC2P) and >$100,000 (TNC3P).

The results of the correlation analyses conducted for the three different
zonal/aggregation levels show that the relationship between household income and boat
ownership rate is not consistent across different types and sizes of boats. The ownership
of small power boats (SPBH) is negatively correlated with the proportion of households
having these three income levels at the county level and there is no significant
relationship at the zip code level. However, ownership of small power boats is
statistically positively related to income levels for census tracts. On the other hand, large

power boats (LPBH) and large sail boats (LSBH) show a positive correlation with all

96

income categories, at a 99% confidence level. As would be anticipated, the highest
income category (IN C3P, $100,000) has a stronger degree of correlation with large
powerboat ownership than the low income category (INClP, $50,000). The ownership of
PWCs (PWCH) shows a positive correlation with the three income categories for all

zonal levels. However, the degree of correlation decreases with an increase in income.

4.2.6 Seasonal homes

In Florida, many boats, both large and small, are kept at permanent waterfront and
non-waterfront homes (Mahoney et al., 2009). There are many waterfront residential
developments that include canals leading to the ocean, docks and/or boat access sites. So,
predictably, the ownership of most types and sizes of registered boats is positively
correlated to the proportion of seasonal homes (HOSP), many of which are on the
waterfront. The exception is that the ownership rate of canoes in all three zonal levels
shows no relationship with the proportion of seasonal homes. This is probably due to the
fact that canoes and kayaks are easy to store and to transport, and finding a place to

launch them is usually not much of a problem.

4.2.7 Urbanization

As already discussed, it was considered likely that urbanization would be a
deterrent to boat ownership for reasons previously mentioned, including the difficulty in
finding places to store a boat and the relatively high cost of storage. This hypothesis was
tested using three alternative measures of urbanization: population density (POP_DEN),
the proportion of the rural population (POP_RP) and the proportion of the urban area

(URBANP).

97

The ownership rates of small power boats (SPBH) is negatively correlated to the
proportion of urban area (U RBANAP) and population density (POP_DEN), but
positively correlated to the proportion of rural population (POP_RP). Ownership of all
boats and all power boat models are also negatively correlated to the proportion of urban
area (URBANAP) and population density (POP_DEN), but positively correlated to the
proportion of rural population (POP_RP) because of the predominance and influence of
small power boats. The ownership rates of large power boats (LPBH) and sail boats
(LSBH) are either not correlated or are only weakly correlated with the proportions of
urban areas (U RBANAP), population density (POP_DEN) and rural populations
(POP_RP). This is most likely because there are many large marinas located in urban
coastal cities, and there are also a larger number of persons with high enough incomes

and the ability to afford and maintain large boats.

4.2.8 Proximity to water

The percentage of area within one mile of the coastline, including tidal shoreline
and rivers, streams and waterways, is positively correlated with boat ownership at the
county level, with the exception of the ownership of canoes (CAN OEH), which are
highly portable and are rarely used in saltwater. However, the magnitude and direction of
the correlation changes to negative at the ZCTA and census tract levels. For these two
zone levels, the relationship is the same for power boats (APBH) and all boats (ABH), as
well as for sail boats (ASBH, SSBH, LSBH). There is an inverse relationship between the
area within one mile of the coastline and the ownership rate of PWCs (PWCH) and small
power boats (SPBH). This may be due in part to the easy transportability of these boats,

and the likelihood that they are stored at non-waterfront homes and not marinas.

98

4.2.9 Marinas

As was anticipated, the number of marinas is positively and statistically
significantly correlated to ownership rates of sail boats (ASBH, SSBH, and LSBH),
PWCs (PWCH), and large power boats (LPBH) for all zonal levels. The majority of these
boats are stored at marinas and waterfront homes; their owners are likely to reside near
the marinas. Also, many marinas in Florida are associated with housing communities
(e.g., condominiums, apartments, mobile homes). The ownership rates for small power
boats (SPBH) and, therefore, all boats (ABH) and all power boats (APBH), most of
which are small power boats, are not correlated to the number of marinas.

Not surprisingly, adding larger buffers to the zonal units, thereby increasing the
likelihood of the presence and numbers of marinas, increases the magnitude of the
correlation with the number of boats registered to persons living in each of the zones.
Table 4-7 shows that the highest correlation is between marinas and the number of boats

registered in the buffer-added zones.

The ownership rates for some types and sizes of boats are not consistently
correlated to the number of marina(s) across the three zonal units. For example, the
ownership of all boats (ABH) is not correlated to the number of marinas, primarily
because the majority of boats are small power boats that are not usually stored in marinas.
At the ZCTA level, there is a negative correlation with the number of marinas, but the
correlation is positive when a five-mile buffer is added. This may be caused by the wide
variance in the numbers of marinas by extending the buffer to 5 miles. For example, there

is no marina within the boundary of zipcode 33311, F ort Lauderdale; the number of

99

marinas increased to 212, however, by extending to 5 miles outside of the boundary. This

dramatic variation could change the direction of the correlation.

4.2.10 Boat Launch Sites

The relationship of the number of boat launch sites (in the zonal areas and
buffers) to the rate of boat ownership was inconsistent across types of boats and zonal
levels. The analyses produced a number of unexpected results. There is a relatively strong
correlation (0.275), as would be expected between the rate of small powerboat ownership
and the number of launch sites in census tracts. The strength of the relationship increases
when ownership rates are correlated with the number of launch sites within the census
tracks and three (0.433) and five (0.428) mile buffers. The relationship is strongest for the
ownership rates for all powerboats but surprisingly it is greater for large powerboats than
small powerboats. Part of the reason for this incongruity, may be that larger power boats
are now being trailered and launched, and possibly a large number of smaller powerboats
may be stored at waterfront permanent and second homes. The correlation between the
ownership rates of small power boats in census tracks and the number of launch sites
(within the census tracks and buffer areas) is greater than for large powerboats. However,
unpredictably there is no significant relationship between all powerboats and small
powerboats and the number of launch sites at the county level although a significant
relationship exists with large powerboats. This was not anticipated since small
powerboats, which comprise most powerboats, are the ones that are believed to use
launch sites most frequently. Again it may be the case that more boats larger than 23’

(e. g., 23-26’) are using the launch sites.

100

Another unexpected result is the significant correlation between the ownership rates of
small, large and all sail boats and the number of launch sites across all three zonal levels.
Sail boats are rarely trailered and launched and therefore it is difficult to explain why

there is a consistent significant correlation.

4.2.11 Examine MAUP and its effects on correlations

As discussed in Chapter Three, increasing the zone size led to an increase in the
degree of correlation. However, this study didn’t validate this general conclusion for all
pairs of correlations. For example, in terms of the statistical significance of correlation
tests, 232 of 243 bi-variate correlation tests are significant at the census tract level.
However, this number decreases to 200 at the ZCTA level and 136 at the county level.
Moreover, there were no consistent patterns of correlations, in terms of direction (+/-)
and magnitude, across different types and sizes of boats, zonal units or different
independent variables. In most cases, the absolute value of correlation coefficients is
lower for smaller zonal units. However, for some independent variables, such as the
proportion of male population (MALEP), the correlation with boat ownership is stronger
for census tracts than it is in ZCTAs. Also, the direction of the correlation coefficients is
not constant across different levels of zonal units for some independent variables. For
example, household income is negatively correlated with the ownership of all boats at the
county level, but there is no correlation at the ZCTA level, and there is a positive

correlation at the census tract level.

101

4.3 Development of OLS models that predict boat ownership in Florida at the
county, ZCTA and census tract levels

Ordinary Least Squares (OLS) is the best known of all regression techniques. It is
also the proper starting point for all spatial regression analyses. OLS regression is a
straightforward method, has well-developed theory behind it, and has a number of
effective diagnostics to assist with interpretation and troubleshooting. OLS is only
effective and reliable, however, if the data and regression model meet/satisfy all the

assumptions on which this method is based.

A series of full OLS models were fit to boat owner rates adjusted for age, gender,
race, household structure, household income, urbanization, and boating accessibility.
Boat ownership rates were transformed using a natural logarithmic form, in order to
satisfy the assumption of the normality of residual distribution and to avoid negatively

predicting the rates of boat ownership.

The detailed results of the OLS models are in Appendix H. Table 4-8 below
provides the summary model statistics and performance indicators, including: R squares,
adjusted R squares, AIC, number of independent variables, whether the model supports
the hypothesis in this study, and whether the model residuals are normally distributed or
not. The different OLS models are evaluated based on: (1) model significance; (2) model
fitness; and (3) the distribution, heteroskedasticity and the spatial autocorrelation of

model residuals.

102

Table 4-8 Summary of the results of the OLS models intended to predict the rate of boat ownership

for Florida counties, ZCTAs and census tracts

LNABH R2

(All boats) 14191119614143.2111 -11

1.6118311189191170 grits-r1199,

.1Numbsr-9erriab1les1-

_ SyppgrtHyPO-thesjs -1

Normality of Residuals

LNAPBH R2

. ___-2 _,---_1
(All power _Awfiflsteggm __11

boats)

_Nur_nber of Variables”

jimmnflypgthssis._1 - 1

Normality of Residuals

_Akgiksinfg-critcrion-1-

County Level
1-0-811911 1 -

. .19-8.05111... 1
1. 1191409.- .

Y

1- 9.1985111- 11 .1
_. 1.1-10116.7 0111

231-1519 1

ZCTA Level

N

0.6.7131. _.

11110-971011. 1
1111214131989 _1-

Census Tract Level
*1. 9167111 .121 1. 1‘
-19-1.6.68.1.-. -

11 9101-1099 1
81 .11
N

“0.566
1.1-1119-156511
_ 611191-1290
.5.-

N
-19-28.4..
.1 ___-_91-12321
_ 61151112118991

z -9
- 1 4 _111Y-1

 

LNASBH R2

(All sail
boats)

N,l1lmbsr._on.-V-ariable.s

1519999“ Hypothesis. 1- ._ 1

Normality of Residuals

14011189418121-1-

Alsailse info criterign

.11-91.3.1316 1111-. 1

_. 191304-111. _ ,1 _
253.482 __ f

f 3
N

.11-191.248

1 1411014917 .710.

11101-21551 - _

1 1101-3012.-

1-1-191310.11-

1 6,736.300
V Y
N

 

LNCANOEH £2 1...- 111- . 1

(Canoes) 11589198161411: -

LAB-136.19% 191116-1199-

1Nu1mbgngeiables-V11_1,11 V

13111139991 111/99.11195111511- 1.
Normality of Residuals

1 1913-91-51.
_ 11.11.19.134].-

. .11-91119.02 _1

N

1. 1.912187-

__V1_0-1218-1--11 _-
1:11.9113869-1-11-.. 1

__N
N

__-_9-112-9_§
01219.11-
1 V. _1- -1 7198131599-
1 .811

___Y

N

 

LNPWCH __RZ _

(PWCs) -69.19§_t§£1_13i11111 __

.16-191159-191’99181911199 _

Number of Variables 1

 

_S-yppgwypothas-is. - _

Normality of Residuals

0.559

, 0523
_ 142-704.-.-

.12-13.95.11-111 _

1 191392
1 12.133.711.520. 1

0917.13.

_1-1101-3171611
-11-1339.311920-

5
N

 

LNSPBH . R2

-.Afiiusteéflz... _

(Small power 1 _ -
.- 51159131111119 criterion

boats <23’)

1.1519111199111911’3/8-111812165

1§IPPQEt1Hyp2tb19§i1L-_-1111 ,

Normality of Residuals

103

__ _ 91172-1. §

_ 9171-8.
1 1242.200

_ 1 Y1111V11 1

N

_ ”94588,,

, 01.5.871-
f 6.871790
8
v

N

Table 4-8 Summary of the results of the OLS models intended to predict the rate of boat ownership
for Florida counties, ZCTAs and census tracts (con.)

County Level ZCTA Level Census Tract Level
2

 

 

 

 

 

 

LNLPBH 181-111-1- 1 . 1.112 080151 -19-3179111 .11 10.3219
(mgepowe, 1Adj1u151951_R:-11 107-82 19.13115 63181
boats 23’+) 1Wf9191r-i1ts1ri1619 111.156.3111.- _ 2 2152714150,- 12.192.59.91
1N91mb§E0112Y811128b1815 . ,w 7 .1 16 _. 5
1-Sur33651Hynot-hssis2 .1 1N11 ,,,,, 11 Y1 1 1 1_Y
Normality of Residuals Y N N
LNSSBH 15: 121 1 1 - 1-10.2518- 11220161111 _ 119,291”.
(Smallsail 12491111115199-1821111-1 2 .1 ..11_.__912-18_11 1911919.- 1111 ._.-1119129_61-
boats <23’) 16331320135119.1191}..-.121. 111267-098 ..4565239 12 17317112117001
VNumbergLVfiab19-81__11111 V311 2 7 6
23111112129“ prothcsis 22 Y 1 2 N2 .11 -..2Y
Normality of Residuals N N N
LNLSBH R2 __ ‘ _0.393 _ __ 0.318flwg 0216
(mg... sail now-Isiw .. _ 013-61411. 923112 1 102-12752
boats 23’+) Akaiksinfopriterion 272511-1034 4.391.960 _1 17785-2190
-NuVmbcrof-Yariables 1 3 7 .1 6-
--SUPportHypoth8-S.is N N 1 . Y
Normality of Residuals N N N

 

4.3.1 Model significance

F-statistics provided the measure of the overall model significance. The
confidence levels for all models, including both the OLS and spatial models, are all over
95%. This test result implies that the independent variables selected in the zonal OLS

models effectively predict ownership rates among various types and sizes of boat.

4.3.2 Model fitness

R square and adjusted R square were used to evaluate fitness of the OLS models.
AIC was used to compare models in terms of precision and complexity of the model.
The small power boats (LNSPBH) model has one of the highest adjusted R square

and lowest AIC for all three zonal levels. In part because of the disproportionate

104

influence of small power boats, the power boat (LNAPBH) model and the all boats
(LNABH) model have relatively high adjusted R square and low AIC. Conversely, the
sail boat models, including all sail boats (LNASBH), small sail boats (LNSSBH) and
large sail boats (LNLSBH), have the lowest four R square, adjusted R square and highest
AIC for all zonal levels. The canoe (LNCANOEH) model has a similarly poor
performance. The large power boats (LNLPBH) county level model has the best
performance, with an adjusted R square of 0.747. The R square is reduced by about half
at the ZCTA and census tract levels. The fitness of the PWC model is always in the
middle of the models: the R square is less than the small power boat (LNSPBH) model,
but higher than the sail boat models and the canoe model, at all zonal levels.

Not surprisingly, the models’ R square decreases as the size of the zonal units
decrease. County level models have the highest R square and census tract level models
have the lowest R square. This result supports the conclusion of previous studies (e. g.,
Zou 2005), which state that the aggregation to larger zone sizes will smooth much more

variance and further increase R squares.

4.3.3 Assessment of model residuals’ bias

The Jarque-Bera statistic indicates whether or not model residuals are normally
distributed. The test results (Table 4-8) show that county level models (small power
boats, all boats, all power boats, PWCs) have good fitness with normally distributed
residuals, with a confidence level greater than 95%, while the canoe and sail boat models
have biased distributions of residuals. On the other hand, none of the ZCTA or census

tract models have normally distributed residual patterns at a 95% confidence level.

105

4.3.4 Access of residual heteroskedasticity

In this study, Breusch-Pagan is employed to determine whether the independent
variables comprising the models have consistent relationships with the dependent
variables. The results (see Appendix H) show that there was no residual
heteroskedasticity for the all boats (LABH), PWC (LNPWCH), small power boat
(LNSPBH) and larger power boat (LNLPBH) models at the county level, at a 95%
confidence level. The ZCTA and census tract level boat ownership models have
statistically significant residual heteroskedasticity, above a 95% confidence level. This
indicates that the relationship between the independent variables comprising the models

and the rate of ownership is not consistent.

There are two problems associated with the zonal OLS model related to residual
heteroskedasticity: the assumption of constant variance and independence for residuals
are violated. A possible cause for violation of the assumption when it comes to the zonal
OLS models may be inconsistencies in geographic space and in data space. This study
mainly focuses on exploring inconsistencies in geographic space. Recognizing this, a
spatial autocorrelation test for residuals was applied to assist in determining whether the

heteroskedasticity and distribution bias are caused by spatial autocorrelation.

4.3.5 Test for spatial autocorrelation

Global Moran’s I test was utilized to check whether the model residuals were
statistically significantly autocorrelated in space, and to determine the degree of spatial
autocorrelation for residuals. The results of this test were applied to the boat ownership

models for all three zonal levels, as shown in Table 4-9.

106

Table 4-9 Moran’s l and p value for OLS models

 

 

County ZCTA Census Tract
Dependent
variables Moran's p Moran's p Moran's p

1 Value 1 Value 1 Value

12151116115111 221-1. 1 212191-9612- 221 191-1921 2 -11-19:23.421591902122229.1311;121115919911‘
215N12A1’1-3111122 __1__.. 1 -20-11,391- 2119-.0319. 1110-21512 2411-109112-1.111__19282211111111<Q10_0211.
112-NA2SB11111 21 .10-01712 1112 0-1517-72 1 01-21-212 2259-0101 1111 _ _11191-V1f1171 22501-10011-
1-1-53195119113111.1- -. 11 0.26151 2159-0101111 .10-9194-11V1_f0-001111 ,_ 10.114411. 111 $080.01.-
12151883291122 1, 101-10.342- 2119-313161- 11191-19772121<1Q:OQ-111111 910713-. 2150100111-.
21511811113112- - 11 21 .29-1912181 221 _9-‘111572 21 .20-21722 251019012. 119213312111 5121-90111.
215N121’11éfl22 21 11V 11112129111629.1222 291-910-611 _21 01-1121 1112241290112 1 .2 11202117021 51212091111
212-12188111311122 221-1211919515222 11117-221111 22 1119-11724- 2592100112121 2291112511212 $20921...
LNLSBH 0.073 0.230 0.089 <0.001 0.129 <0.001

 

The test results show that the power boat (LNAPBH), canoe (LNCANOEH) and
large power boat (LNLPBH) county level models exhibit spatial autocorrelation. The test
statistics reveal no autocorrelation for the other six county level models. There is spatial
autocorrelation exhibited by all type and size ownership models at the ZCTA and census
tract levels. The Moran’s 1 statistic increases with smaller aggregation/zone levels, except
for the canoe (LNCANOEH) and large power boat (LNLPBH) models. This clearly
shows that when the zonal units being modeled are smaller (e. g, census tracts), the
neighboring zones have a stronger influence on the models that are created. This is

consistent with the conclusions of other studies cited in the literature review.

Global Moran’s 1 tests only tell the degree of spatial autocorrelation for residuals.
LISA (Local Indicators of Spatial Association) was used next to identify where the “hot
spots” or clusters of individual zones with similar residual values are located. For
example, Figure 4-1, 4-2, 4-3, 4-5 and 4-5 present a series of maps that graphically
display the results of LISA for the total boats (LNABH) OLS model at the census tract
level. The maps shows different spatial grouping:

107

Central Florida, Miami, Cape Coral and Panama City regions are characterized by
clusters of high ownership rate, census tract level, whereas Ft. Lauderdale and

Jacksonville exhibit a cluster of low ownership rates in neighboring census tracts.

Figure 4-1 LISA Cluster and Significance Maps for All Boats OLS Model at
Census Tract Level (Florida State)

 

   
 
 

LISA Cluster Map

Not Significant LISA Significant Map

 

 

    

 

 

IHigh~H1gh .. Not Significant

I High-Low : I p=0.05

I Low-High - t . '3 I p=0.01 3 (m
I Low-Low I p=0.001

 

Figure 4-2 LISA Cluster and Significance Maps for All Boats OLS Model at
Census Tract Level (Ft. Lauderdale and Miami Region)

 

ISA Cluster Ma
Not Significant ISA Significant Map
High-High I Not Significant
High-Low

Low-High
Low-Low -.

 

 

 

 

 

 

 

 

 

108

Figure 4-3 LISA Cluster and Significance Maps for All Boats OLS Model at
Census Tract Level (Jacksonville Region)

 

LISA Cluster Map

[I Not Significant

 

 

7

LISA Significant Map
DNot Significant
Ip=0.05

Ip=0.0l
Ip=0.00l

 

 

 

 

 

\/\

 

Figure 4-4 LISA Cluster and Significance Maps for All Boats OLS Model at
Census Tract Level (Cape Coral Re gion)

 

 

 

 

 

 

ISA Significant Map
I Not Significant

 

 

 

Figure 4-5 LISA Cluster and Significance Maps for All Boats OLS Model at
Census Tract Level (Panama City Reg

gion)
f'

 

     
 

LISA Cluster Map
|:| Not Significant
I High-High

I High-Low

I Low-High

I Low-Low

 

 

     
   
 

LISA Significant Map
DNot Significant

I p=0.05

I p=0.01

I p=0.001

 

 

 

 

109

4.3.6 Identification and evaluation of the OLS model coefficients

One of the purposes of this study was to identify the best model in terms of model
performance, parsimony, interpretability, stability, avoidance of bias in inference, and
effectiveness for predicting boat ownership in Florida. Predictor variables comprising the
models should be causally reasonable, practically available, and statistically significant.
Only effective variables that are statistically significant (p value < 0.05) in both OLS and
spatial regression models, and show a low degree of multicollinearity (VIF < 7.5) are

included in the models.

In general, the county level ownership models have fewer independent variables
than either the ZCTA or census tract level models. On average, the county level boat
ownership models are comprised of four independent variables, in contrast to seven for
the ZCTA level models and six for the census tract level models. This suggests that the
county level models are more efficient than the other two zonal levels. The next sections

discuss each of the independent variables individually.

4.3.6.1 Age

The relationship of age to boat ownership rates is consistent across different types
and sizes of boat ownership models for the three zonal levels. Ownership rates are larger
in areas with higher proportions of total population and head of household between the

ages of 35 and 64.

4. 3. 6.2 Gender

The relationship between gender and boat ownership rates is not consistent.
Gender is only effective as a predictor variable in 8 of the 27 models for different zonal

levels. It is interesting that ownership rates of large boats, both large power (LNLPBH)

110

and sail boat (LNLSBH) decreases as the proportion of males in the population increases.
Nothing in previous research explains this inverse relationship. As was expected, the
predicted ownership of all power boats, small power boats and PWCs are higher in

census tracts with higher proportions of males in the population.

4. 3. 6.3 Race and ethnic factor

Race and ethnicity are effective predictors for all the ZCTA and census tract level
ownership models. Four of the nine county level models also incorporate race and
ethnicity. Consistent with many boating studies, higher ownership rates were found
where there was a greater proportion of white non-Hispanic households and white
households. The rate of boat ownership decreases as the proportion of Black households
increase. The dominance of whites among boat owners, and the declining proportion of
whites in the US. and Florida populations, is a major reason for the overall slowing, and,

in some instances, decline in boat ownership seen in some states.

4. 3. 6.4 Household structure

The inclusion of household structure and its relationship to ownership rates is not
consistent across the three zonal units. For example, at the county level, the ownership
rates for large sail boats (LNLSBH) and PWCs (LNPWCH) increases as the proportion
of married family households (FAMP) increases. Household structure is not a significant
factor in most of the ZCTA level models. The ownership rates for small sail boats
(LNLSBH) and canoes (LNCANOEH) is lower in areas with a high proportion of
households of married families with children. Conversely, the ownership rates of all boats
(ABH) and power boats (LNAPBH) are positively associated with the proportion of

household of married families with children. At the census tract level, the higher the

111

proportion of households of married families with children, the higher the ownership
rates of small and large power boats, and small sail boats (LNSSBH) as well. Ownership

of PWCs (LNPWCH) is higher in areas with more family households.

4.3. 6.5 Income

The different household income categories/variables are significant in the
ownership models for all three zonal levels. The more households with incomes over
$50,000, the greater the ownership rates for small sail boats (LNSAIH), but there is a
decrease in ownership rates for total boats (LNABH) in a county. The ownership rates for
large boats, including power boats (LNLPBH) and sail boats (LNLSBH), increase with
the increase in the proportion of number of households with income over $100,000 for a
zipcode. Conversely, the ownership rates for the total number of boats (LNABH), power
boats (LNAPBH), and small power boats (LNSPBH) decrease for a zipcode, with an
increase in the proportion of the number of households with income over $50,000. The
ownership rates for PWCs are positively related to the proportion of the number of
households with income over $50,000. At the census tract level, the ownership rates for
large boats including power boats (LNLPBH) and sail boats (LNLSBH), increase with an
increase in the proportion of the number of households with income over $100,000 for a
census tract. Conversely, the rates for small power boat ownership for a census tract

decrease with an increase in the proportion of the number of households with income

over $100,000.

4. 3. 6. 6 Seasonal housing units

A large number of seasonal housing units (e.g., cottages, condominiums) are

either oceanfront residences or they are located on canals or rivers that provide access to

112

 

l7

the ocean. Only the canoe models predict lower ownership rates with an increased
percentage of seasonal housing units as a percentage of all housing units. At the county
level, the large power boat (LNBPOZ) ownership rates increase with an increase in the

proportion of seasonal housing units.

4. 3. 6. 7 Urbanization

Urbanization was hypothesized to be an important factor in determining boat
ownership rates. Urban areas generally have fewer places to store/keep boats;
additionally, the cost of owning and maintaining a boat is believed to be higher in urban
areas. Population density, proportion of rural population and proportion of urban area are
measures of the degree of urbanization in this study. The denser the population in a zonal
area, the lower the boat ownership rates, which supports the hypothesis that the boat
ownership rates are negatively related to population density in an area. However, some of
the associations are unexpected or counterintuitive. The proportion of the rural population
was negatively associated with sail boat ownership rates at the ZCTA level, which
violates the hypothesis but makes sense, because most sail boats and large boats are kept

at marinas, which are primarily located in urban areas.

4. 3. 6.8 Boating facilities and water resources

The availability of boating opportunities (e.g., marinas, public boat launch sites,
fishing areas) is a crucial factor in determining boat ownership rates. This study utilized
different measures of boating access as a proxy for boating opportunities. These measures
include: (1) opportunities by water resource (the proportion of a one-mile coast line
buffer); (2) number of seasonal homes, many of which are waterfront homes; (3) the

number of boat launch sites and marinas in a given area). The more water resources

113

found in an area (county, ZCTA and census tract), the higher the boat ownership rates,
which supports both our hypothesis and previous studies, Stynes (1982). However, the
number of marinas has no significant relationship with the ownership rate of sail boats.
This may simply reflect the imprecision and inadequacies of the boating facilities

measurement.

In conclusion, the results obtained through the process of developing and
evaluating the OLS models generally confirm the importance of population-level
socioeconomic variables as predictors of boat ownership rates. However, there were a
number of anomalies and unexpected relationships between various demographic
variables and the ownership of different types and sizes of boats. Urbanization was also
hypothesized to be an important factor in determining boat ownership rates. The findings
suggest that urban areas generally have fewer places to store/keep boats and that the cost
of owning and maintaining a boat is likely to be higher in urban areas. The findings also
reveal that zonal units with higher population densities have lower boat ownership rates.
The availability of boating opportunities (e.g., marinas, public boat launch sites, fishing
areas) is a crucial factor in determining boat ownership rates. The more water resources
there are for boating waters in an area (county, ZCTA and census tract), the higher the
rate of boat ownership, which supports our assumptions and the findings of previous

studies.

The diagnostics tests show that small power boats (LNSPBH), all boats
(LNABH), large power boats (LNLPBH) and PWCs (LNPWCH) at the county level OLS
models exhibit good fitness and effectiveness. However, the all sail boats (LNASBH),

small sail boats (LNSSBH) and large sail boats (LNLSBH) and canoes (LNCANOEH)

114

county level OLS models perform poorly in terms of model bias, residual
heteroskedasticity and model performance. All of the ZCTA and census tract level
models have residual distribution bias and heteroskedasticity, and all of the models show
spatial autocorrelation of residuals. As a result of these findings, spatial regression
models, which have been shown to reduce spatial autocorrelation of residuals were

developed and are evaluated in the next section of this chapter.

The goodness fit of the census tract models is better than it is for the ZCTA
models for the same types and sizes of boats, even though most of the census are smaller
in area than ZCTAs. A possible reason for the better goodness-of-fit is that the census
tract boundaries are established to create more or less homogenous areas (e.g., equal
population), while ZCTA boundaries are delineated by mail routes with no concern for

spatial homogeneity.

The combination of independent variables and the value of their coefficients vary
across different zones and across the different types and sizes of boats. There is no
pattern in the absolute value of coefficients across different types and sizes of boat
models developed for different zonal levels. According to Zou (2005), if there is no
uniform shift pattern for the absolute value of coefficients when the zonal scale changes,
it suggests that there are no systematic scale impacts on the magnitudes of the model

coefficient estimates.

4.4 Develop spatial models based on OLS models

The spatial autocorrelation test for OLS models (Table 4-9) reveals an

increasingly significant spatial dependency among the residuals, as the size of the zonal

115

area decreases. There are a number of potential factors (e. g., zone size) that might
contribute to this spatial effect. To gain a better understanding of the spatial effect, the
spatial models—including spatial lag and spatial error models, SAR (Simultaneous
Autoregressive Model) and CAR (Conditional Autoregressive Model)—were run with
the same combination of independent variables comprising the OLS models. The models
with the highest R square and lowest AIC were selected as the “best” for each type of
model. The complete results of the model building are shown in Appendix I. Table 4-10

presents a summary, for discussion purposes.

Table 4-10 Summary results of different types of Spatial Regression Models predicting boat

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

ownershig

County ZCTA Census Tract
LNABH R2 increasesI (%) 0.5% 8.8% 18.7%
(All boats) AIC decreases2 (%) 10.0% 10.0% 10.7%
Spatial model 53160th Spatial Error Spatial Error Spatial Error

Spatial autocorrelation (OLS-Spatial) N - N Y — N Y -N

Heteroskedasticity (OLS-Spatial) N - N Y —- Y Y - Y
LNAPBH Kilnffiéseglw- ., ,-_ . 2 .21-83411 .. , 11191-181‘79 __ 1114-1994!-
(All power 75119 5130198538? (%)-11122, ,, 13.0% 10.9% ___-281.5%
boats) Spatial model selected Spatial Error Spatial Error Spatial Error
131191-18131111199911?18119112(,021:§§P§1181)1_ 1 V Y -_N 111121121Y1-1N _1 1. Y :N
Heteroskedasticity (OLS-Spatial) Y - Y Y — Y Y - Y
LNASBH R2_ir5=rease§1l(%) _1 __-1_._ . ._ 171143111 15.4%- ,1 __ _1 18.5%
(All sail boats) Algdecreasesz (%) .1 _l.6%v V _ 0.6%W _ "71.1%
33811281111109?!15.811191016191211. V 1. .. CAR- _. CAR-12 _CAR
.32311811181119901261181120111 1(01L2S2-Spati181)-1 1 111211512182 -21. Y1-1N1 V 21. Y1-2Y
Heteroskedasticity (OLS-Spatial) Y - na Y — na Y - na

 

1. Percentage of R square increases from OLS model to spatial regression model
2. Percentage of AIC decreases from OLS model to spatial regression model

116

Table 4-10 Summary results of different types of Spatial Regression Models predicting boat
ownership (con.)

 

 

 

 

 

 

 

 

County ZCTA Census Tract

LNCANOEH Rii-ereésesl-(W . , ,1 _ 111-3.41% V, _,-!9-5-% 2 1117-1%
(All canOCS) AleiecreaSCSZM), 1 1 1 , 1 1 81.17?ng 19.61% 1 10.9%
Spatial model selected SMA Spatial Lag CAR
Spatiala9t0¢9r1r¢|at1i99(QLS+Spa:ial) 1N-1N 1 .V Y :N, ,Y :Y
Heteroskedasticity (OLS-Spatial) Y — na Y — Y Y - na

LNPWCH 53-ingrggscs:(%) ,,,,, 1.11%11” 15.61% 3.17%
<A11PWCs> 521999<=r¢235652£%). - 1__2 _.,_1_121._3%_11 _ 04/1 9.4%
Spatial model selected Spatial Lag SAR Spatial Error
Spatial,autqqorrs-latioy,(OLS-Spatial) -1 _ 1 N -2N , --Y -,._N ...Y.-.N
Heteroskedasticity (OLS-Spatial) N - N Y — na Y - Y

LNSPBH ,5 ins-r6555 1(-%). _ 05% -29-2% 112116572
(Small power ,é!C,£le1¢r_¢9595_2,,(°/9), _ 125.3% _ 112-2% 160/
boats (23,) Spatial model selected Spatial Lag SAR Spatial Error
,SpaLialaqtocorrelation(OLS:SpatiaD,. NN Y-N - Y-N
HeteroskedasticitflOLS-SpatiaI) N - N Y — na Y - Y

LNLPBH 1133119061564012) . 35% ,1_-.8-7°/91_1 125-0%
(Large power élcéécxéase_sz(%)._...__1 125/ - 0.7% .21-.7%
boats 23"“) Spatial model selected Spatial Lag CAR CAR
Spatial any-tocorrclation (OLS-Spatial) . .. Y _-_N__,, Y-N Y:N
Heteroskedasticity (OLS-Spatial) N - N Y — na Y - na

LNSSBH 53110995356192)“ __ .___ _, _ 55% -15-512% __ 309/.
(Small sail AICVdecreaseszg%)V , V , . V 1.5% ,, 1.1% , 1.0%
boats <2?) 1515,8191 moéglgclss-t-QCL-1,22,-, ,, CAR . CAR, _ _CAR
§Patialautocorrelati991(QL_S1-Sfla1tia2|)1111__1 _12N_:1N11 ._ . __Y17111‘J1 ,11 _ Y;Y
Heteroskedasticity (OLS-Spatial) Y - na Y - na Y - na

LNLSBH 1’32 i299¢ases1lj1°é>21 22111,,,-3-_6%_ __ . ___-619%,- __-1121§-11%.
(Large sail 113-Icad2e-c1mases2,(%> 121 , ,1 12111.94», 9.3%. _ 29.18%
boats 23”) Smalgpslsl $919994 _______ .. EAR. CAR CAB-
Spatial autocorrelation (OLS-Spatial) N - N Y — Y Y - Y
HeteroskedasticityLOLS-Spatial) Y - na Y — na Y - na

 

1. Percentage of R square increases from OLS model to spatial regression model
2. Percentage of AIC decreases from OLS model to spatial regression model

117

4.4.1 Model fitness

In general, the R squares of spatial regression model residuals are larger than OLS
models’ R squares. The R squares for spatial models are an average 17.6% greater than
the R squares for OLS models at the census tract level. In comparison, R squares for the
spatial models are, on average, only 6.5% greater than the R squares for OLS models at
the county levels. On the other hand, the degree of spatial autocorrelation of the OLS
model residuals for census tract level models is much stronger than that for the county
level model. This indicates that the spatial regression models have a better goodness-of-

fit than the OLS models, when spatial autocorrelation exists.

4.4.2 Examination of residuals

In an effort to examine whether or not spatial models reduce the deviation of
residuals compared to the OLS models, the percentage of residuals vs. observed values
for both the OLS model and the spatial regression models were mapped. Appendix J
presents maps comparing the magnitudes of the percentage of residuals vs. observed
values for the OLS and spatial regression models that predict the ownership rates (all
boats model) for Florida’s counties, ZCTAs and census tracts. A comparison of the
residual maps shows that the range of spatial regression model residuals is less than that
of the OLS models. The number of negative and positive residuals is fewer and less
extreme for the spatial models. However, a further examination of the residual maps
reveals that the fitness of the all boats (LNABH) spatial regression model at the county
level is not much better than that of the OLS model, since there is no spatial

autocorrelation in the residuals.

118

The autocorrelation of the residuals of the spatial regression model were then
examined and compared with the spatial autocorrelation for the OLS models’ residuals.
Table 4-10 shows that there is spatial autocorrelation of residuals for nineteen of the OLS
models, compared to five for the spatial regression models. Autocorrelation is
significantly less prevalent among the spatial regression models. Furthermore, the
stronger the degree of spatial autocorrelation in the OLS models, the greater the degree of
improvement in model fitness for the associated spatial regression models.

To further compare the predictability of the OLS and spatial regression models,
the observed values of the dependent variables are compared with the fitted values for
both models. The stronger the correlation between the observed, the better the model fit.
Table 4-11 shows that for most of the model pairings, the OLS and spatial regression
models performed at almost the same level. This indicates that using spatial regression
does not significantly improve the predictive capability, even though the spatial

regression models remove the spatial autocorrelations of the residuals.

Table 4-11 The correlation coefficients between observed values and predicted values of the OLS
and Spatial Regression Models at the County, ZCTA and Census Tract levels

 

 

 

 

 

 

 

 

County ZCTA Census Tract
OLS Spatial OLS Spatial OLS Spatial
Model Model Model Model Model Model
29141231111211.__121229-905212111111191-199911 ,, 93121211211111-119-filfi11121-91522-112111197472
1135-6391412- 2-20-1152181- _ 1-1982711111219323 _,1191311162121110-1764 ., 2129-76112
”11111318211 1 0.580__V ,1.._20_-§00 0.503 1 0.541 0.549 0.599
12L1N1§AUQEE11__219-622211.11-29.72811 -21-9151326111 ,01-544 0-541 119-5892
LNPWCH 21- .22-148121211 193152111_--111_1Q-51-5§1,.11_..1- 119-156.91111.119-26111512-111 10115111511
21.311113131311321- 211L875 _ 9-81191111111 035111111111 1,_11Q.-_§_9_11_-1,-1 0797 201611..
11.512991113112112 .29-81927221211292.1321, 112026118121 __2Q-16144 0-565 10-613132,
112N§§B1H222 121129504 , ,, 119-151113 112104-514111-11.111-0:5?—2111_ 1_ 014-515-1- 1 05211
LNLSBH 0.627 0.638 0.564 0.582 0.525 0.569

 

The Breusch-Pagan test used for testing heteroskedasticity is then applied to both

the spatial error and spatial lag models developed using GeoDa. This test could not be

119

employed for the SAR and CAR models developed using the R Statistics Package
because the Breusch-Pagan test for heteroskedasticity of residuals for SAR and CAR
models are not available as part of the R Statistics Package. The results in Appendix I
show that spatial regression does not eliminate the heteroskedasticity of residuals for the
spatial lag and spatial error models. However, the spatial models do exhibit an improved
goodness-of-fit as a result of reducing the herteroskedasticity associated with spatial

autocorrelation.

Finally, the coefficients of the spatial regression models were examined and
compared with their paired OLS model. The results of these comparisons (Appendix I)
show that there is no difference in the signs (+ or -) of the coefficients between the OLS
and spatial regression models. However, there are some differences in the absolute values
of the coefficients, even though there is no consistent pattern of differences between the
OLS model and spatial regression models for different types and sizes boats for the same
zonal level. These findings show that the effects of independent variables are not

different in the spatial and OLS models.

In summary, this study finds that employing spatial regression to estimate boat
ownership could remove the spatial effects to some degree, but not entirely. Moreover,
the results of the Breusch-Pagan test imply that even after incorporating spatial terms,
residual heteroskedasticity still exists. Therefore, it is necessary to go back to examine

the theoretical bases that built the models.

120

CHAPTER FIVE SUMMARY AND IMPLICATIONS

This dissertation is presented in five chapters. The first chapter provides an
overview of boating participation, boat ownership and related industries in Florida, as
well as the research problem, and this study’s research objectives and hypotheses. The
second chapter is a review of previous recreational boating studies and literature
pertaining to factors affecting boat ownership and boating participation, as well as
recreational demand analysis and spatial modeling. The third chapter describes the data
sources used to develop the boat ownership demand models and the research process
employed to build the models. The fourth chapter presents the results of the model-
building process and a comparison of both OLS and spatial regression boat ownership
models. Finally, this the fifth and final chapter provides an assessment of the degree to
which the study objectives were accomplished, summarizes the results of the tests of
hypotheses, reviews the model development process, offers an assessment of the
performance of the different OLS and spatial regression models, recommends possible
refinements and further development of boat ownership demand models, and finally
presents overall conclusions regarding the utility and generalizability of the models for

predicting the ownership of other types of recreational equipment.

5.1 Summary of the research hypothesis and objectives

This study was designed to accomplish four objectives: (1) describe the current
recreational boat ownership patterns using boat registration data; (2) identify and test
variables that are statistically related to the ownership of boats in general, and to various

types and sizes of boats at different zonal levels; (3) develop and test zonal OLS models

121

to estimate recreational boat ownership, simulate changes in boat ownership associated
with demographic, economic and policy changes, and predict the trends in boat
ownership at multiple scales; and, (4) develop spatial regression models to understand
spatial effects on zonal models and improve the predictability of the models. Five

hypotheses associated with these research objectives were tested.

Registered boat owners were profiled by gender, age, LifeMode, proximity of
their permanent residence to water and the degree of urbanization where their principal
residences are located. The principal residence of registered boat owners was geocoded,
and then their LifeMode and urbanization segmentation were determined based on the
coordinates of their residences. The results revealed that boat owners are predominately
male, middle aged (35-64) and reside with their families. They are more likely to live in
affluent suburban and rural neighborhoods. Their household incomes, age and how far
they live from the Atlantic Ocean and Gulf of Mexico coast line vary considerably,
depending on the types and sizes of the registered boats they own. These findings are
very consistent with the profile of boaters developed from Grow Boating’s study (2007).
This study also produced information (e. g., maps) displaying the spatial distribution of
boaters in Florida’s counties, zipcodes and census tracts (See Figure 3-6, Figure 3—7,

Figure 3-8 and Figure 3-9).

In order to identify and test variables that are statistically related to the ownership
of boats in general, and to various types and sizes of boats, boat owners were first
aggregated to a county, zipcode and census tract, based on where they reside. Data were
then obtained about the demographics, urbanization and boating opportunities of thes

zonal levels and organized using various GIS, database and statistics software.

122

Correlation analyses were then conducted to identify those variables statistically related
to rates of boat ownership. The results showed differences in the variables related to the
ownership of different sizes and types of boats in counties, zipcodes and census tracts. In
general, the results of the correlation analysis support the hypothesis that boat ownership
propensity is influenced by socioeconomic factors, including age, gender, race, household
structure and household income. However, the degree of the association varies across
different types and sizes of boats. For example, the proportion of the male population is
positively correlated to boat ownership rates at various zonal levels for most types and
sizes of boats. However, the rate of sail boat ownership is only weakly correlated to the
proportion of males in the population. The results also show that boat ownership rate is
negatively related to the degree of urbanization for the majority of different types and
sizes of boats that were modeled. The results supported the hypothesis that the
availability of, and distance to, boating access——including miles of coastline and the
number of launch sites, marinas, and seasonal homes—are positively associated with boat

ownership rates.

Correlation analyses were also used to test the hypothesis concerning the MAUP
especially for scale effect in MAUP. The results are not consistent with the previous

studies: the larger the zonal unit, the stronger the degree of correlation.

To accomplish research objective three, zonal OLS models were developed to
estimate boat ownership rates at the county, zipcode and census tract levels. The models
clearly demonstrate the importance of population-level socioeconomic factors,
urbanization and boating opportunity predictors in explaining variation in boat ownership

rates. Again, the models supported most of the research hypotheses even though there are

123

variations across different types and sizes of boats, and different zonal units. However,
the factors that affect boat ownership in these aggregated models are not totally
consistent with the findings of a “dose-response” relation for individuals in research
objective one. For example, there was a lack of association with the percentage of the
male population with boat ownership rates at various zones, where our disaggregate

analysis shows that most of the boats were titled to males or jointly by males.

Moreover, the results show that, in terms of R square and AIC, the county model
has a better goodness-of-fit than the either of the two smaller zonal units: zipcode and
census tracts. This supports the hypothesis that the performance of the large scale models
is better than the performance of the small scale model. Tests of the spatial
autocorrelations of model residuals indicate that the smaller the zone scale, the stronger
the degree of spatial autocorrelation. This also supports Hypothesis : the scale of the
zones affects the performance and predictability of zonal boat ownership demand models

in this study.

Finally, to achieve the last research objective, spatial regression models, including
spatial error, spatial lag model, SAR and CAR models, were developed with the same
independent variable combinations as the corresponding OLS models (boat type, size and
zonal unit). All of the spatial regression models were constructed using one first order
Queen Contiguous Weight Matrix. The spatial regression models resulted in an overall
model improvement of the original OLS models, and a removal, to some extent the
spatial effects and associated heteroskedasticity. The sign (+/-) of the coefficients of the

independent variable were unchanged, maintaining the same effect of independent

124

variables as seen in the OLS models. Thus, spatial regression improves the ability to

predict boat ownership if there are residuals from spatial dependency OLS models.

5.2 Limitations

This study classified/segmented recreational boats according to their type and size
in a similar manner to previous studies (e. g., Mahoney et al. 2009). There are two
principle limitations associated with this method of classification. First, types of boats
were predominately classified as power or sail. About 10% of the recreational/pleasure
boats registered in the state of Florida did not fit into any of these pre-determined
segments. Second, the number of registered sail boats was too small in many zonal units,
especially census tracts. This resulted in models that overestimated the ownership of

these boats.

Even though demographic, socioeconomic and boating related data is more
widely accessible than ever before, the lack of availability of some data was still a
limitation. Various demographic data and measures of urbanization could be readily
obtained for different zonal units using different sources, including the US. Census
Bureau and Simply Map. However, data relating to the supply and accessibility of
boating opportunities continues to be a problem. Florida is one of a very few states that
has conducted a comprehensive inventory of boating facilities (F WC, 2009), which
makes it much more advanced and capable in this area than most other states. However,
Florida’s boating supply information is still far from complete. For example, the number
of spaces (e. g., slips and moorings) available in marinas is questionable. The capacity
(e.g., number of possible launches) and congestion of boat launch facilities in Florida is

not known. Also, there is no comprehensive inventory of waterfront permanent or

125

second/vacation homes with boating access, only an estimate (F WC, 2009) at the county '
level. This creates a problem because waterfront home access—both for permanent and

second homes—is the primary form of boating access in Florida.

Data relating to the “accessibility” of boating opportunities is even more
problematic. This is especially true, because it is very important for determining boating
participation and, therefore, boat ownership. Accessibility of boating opportunities is a
function of many different factors, including the supply of boating facilities, travel time
to reach the facilities, and the congestion/availability of the facilities. This study used a
number of different proxies of accessibility, including length of coastline (proportion of
one-mile buffers to waterfront vs. total zone area), numbers of marinas and ramps, and
seasonal homes. It is not clear whether these proxies provide a reliable representation of

accessibility.

The third limitation of this study is that the boat registration data obtained from
FWC only provided limited information for profiling boat owners. The registration data
only includes the age and gender of the person(s) who have title to the boat. There is no
information regarding boat owners’ race, income or household structure. The coordinates
of the owners’ principal residence was used to identify the LifeMode and Urbanization
segments, as opposed to using information obtained directly from and about the owners.
Another problem is that the addresses included in the registration data used to locate the
boat spatially were the boat owners’ residential addresses, which is not necessarily where
the boat is stored (e. g., marina, second waterfront home). This may especially hold true

for large boats, which are most often kept in marinas and boat yards. The lack of

126

information concerning where boats are kept have produced errors in aggregating boats in

different zones.

Geocoding might be another problem in this study, especially when it comes to
zipcode level estimations. The boat registration data included the county where the boat
owners reside. The county level aggregation was based on the county code contained in
the registration data. However, boats were allocated to zipcodes and census tracts based
on the address of the owner, which was geocoded using ESRI’s Address Coder. The
problem is that previous studies that have utilized geo-coded addresses have found that
the geocode match rates are much lower in rural areas than in urban areas (e.g., Cayo and
Talbot 2003; Kravets and Hadden 2007). In rural areas, because of the common use of
rural routes and PO Boxes, a higher proportion of the addressees cannot be reliably
geocoded. In this study, rural area PO Boxes accounted for the majority of un—geocodable
addresses. The census enumeration units of county and census tract have not been
changed since 2000. There was no problem associating demographic and other
socioeconomic variables by looking up tables from the US. Census Bureau and
SimplyMap. However, the zipcodes for some boat owner addresses have been changed
with the development of the regions. The “new” zipcodes in recently developed areas
were not on the list of ZIP Code Tabulation Areas (ZCTA), which was developed in
2000. The zipcode level socioeconomic information provided by US Census Bureau and
other data providers (e. g., SimplyMap) was associated with ZCTA instead of the current
zipcodes. Therefore, point-in-polygon overlays in ZCTA were used to aggregate boat
owners at zipcode level and to lookup zipcode-level socioeconomic information.

However, the point-in-polygon overlays do not always provide accurate results, in part

127

due to inaccuracies in the cartographic boundary files and in part due to errors associated
with geocoding (Zandbergen, 2009). This could be one reason why zipcode level models

have the lowest performance in this study.

The limitation of utilizing geodemographic segmentation to profile boat owners
has been a concern for many scholars. Community Tapestry is utilized in this study to
characterize the life style of boat owners. As a type of geodemographic segmentation,
Community Tapestry is a combination of factors include life stage, household
composition, education, income, employment, home value, housing type and other key
determinants of consumer behavior. Boat owners are associated one of 65 possible life
styles based on the location of their residences. The concern with this method is that a
person/household might be misidentified just because of the location where they reside —
I am not like my neighbor. This is referred to as the ecology fallacy. Other critiques of
this method include the representational (Goss, 1995), discriminatory (Burrow er al. ,
2005; Graham, 2005) and intrusive (Monmonier 2002; Curry 1998) effects of
geodemogaphic practices. Another potential problem is that there might be multiple boat
owners in the same household (e.g., a child who returns to live with their parents).
Persons in the same household, may have very different lifestyles as a result of
differences in their ages, life stages and incomes. However, Geodemographics are based
on the premise that “birds of a feather flock together” and that “the successive generation
of those birds flocking similar fashion”. Unlike the traditional univariate marketing
research, geodemographic schemes that identify consumer types by neighborhoods
involve multivarate cluster analysis of spatial referenced demographic and psychographic

data. Homogeneity, as used in gemodemgraphic cluster analysis, simply means that all

128

neighborhoods within a given cluster will share highly similar neighborhood lifestyles
and as a result predictable consumer behaviors. It is very evident that in many
neighborhoods, residents exhibit very different lifestyles. This is especially true for
example in transition neighborhoods. There is no currently available method to predict
whether a particular dataset is going to yield results, which are close to the individual
values. However, even if ecology fallacy exists in geodemographic segmentation, the
inference from aggregate statistics to individual characteristics is strengthened if the
distribution of the characteristics in the underlying population is known (Openshaw and

Blake, 1995).

5.3 Research recommendations

This section presents recommendations for similar future research based on the
findings from this study. They include recommendations related to: zonal structure and
design, the spatial weight matrix and extending the spatial regression modeling to other

studies.

5.3.1 Recreational boating demand study

In terms of recreation demand, there is a question regarding whether boating
participation (e. g., amount, types of boating activities) drives boat ownership (e.g., type,
size of boats), or boat ownership drives boating participation (e. g., people own different
types and sizes of boat will participate different kinds of activities and do different
amounts of boating). Various research suggests that water related recreational interests
(e.g., fishing and skiing) plays a major role in determining the size and types of boats
people buy as well as the amenities and features added to boats. While this study focused

on the factors affect boat ownership, using secondary sources of data, it did not include

129

boating participation as a factor in the models. However, it would be beneficial to attempt
the development of a nested discrete choice model to investigate whether and to what
extent the demand for boats (ownership) is a function of the amount and types of boating
participation. This would require a survey which explores how boating behaviors affects

ownership preferences.

5.3.2 Zonal structure and design

Literature reveals that census geographic units and grids are the predominant
types of zonal systems utilized in spatial and regional studies in the United States. Census
geographic imits such as counties, census tracts and block groups are widely employed
because of the availability of the demographic data. However, the size and shape of
census tracts and blocks are not consistent, causing problems associated with MAUP and
ecology fallacy. Conversely, the grid zone system provides consistent zonal unit size and
shape, minimizing these two problems. The grid units remain constant over time, unlike
census tracts and blocks, allowing for time series analysis. Difficulties obtaining
information (e.g., population size, demographic characteristics) specific to the grid units
has been a problem in the past. However, with the development of GIS and spatial data
processing techniques, the availability of grid level data, e.g., population, is no longer a
barrier. Future research should consider employing grid units and/or possibly comparing
the performance of models developed with them, with models developed for census

geographic units.

5.3.3 Spatial weight matrix

Spatial weight matrices are used to test for spatial autocorrelation and to build

spatial regression models. Spatial weight matrices were discussed in Chapter Three.

130

Different types of spatial weight matrices will reflect spatial patterns differently. This
study utilized the simplest spatial weight matrix, the First Order Queen contiguous
matrix, to construct the models and test residuals. Other spatial weight matrices were not
employed in this study. In future studies, it might be beneficial to form methodological
and theoretical perspective to employ different spatial weight matrices and compare the

results.

5.3.4 Extend the spatial regression modeling method to other studies

First, the current boat ownership research could be extended to spatial-temporal
research by better understanding how population changes, migration patterns and
environmental changes affects boat ownership. Spatial-temporal modeling is widely
applied to environmental and health studies (e. g., Calhoun et al., 2009). This method of
modeling has not been applied in the recreation and tourism field very much. Similar to
adding a spatial component to the model’s residuals in this research, a time component
could also be added to the model’s residuals to form a spatial-temporal model for

predicting boat ownership.

Next, the methods for modeling boat ownership could be utilized in other
recreational activities research, such as fishing, hunting, etc. This study focused mainly
on recreational boat ownership estimation and prediction. In the future, a similar spatial
modeling method could be employed to estimate recreational activities that require
licenses and/or equipment registrations, such as fishing, hunting, ORVs and

snowmobiles.

131

A concern related to this study is the use of the location where boat owners reside
to count where their boats are kept. This is not accurate since most large boats are usually
kept at marinas and boat yards. Some states ask boat owners to report where they live and
where they keep their boats. Most states, however, including Florida, do not. In the
future, conducting a survey regarding this issue and adjusting the aggregation of boat
owners would improve the accuracy of the information and further improve the quality of

the models.

5.4 Implications
5.4.1 Management and economic implications

The results of this research indicate that is possible to develop models, utilizing
different types of aggregating secondary data, which can be used to estimate boat
ownership rates and therefore the need for boating facilities and services. The results of
this research suggest that demographics will be an increasingly important influence on
the number of boats owned in Florida and also the distribution of boat ownership.
Florida’s population is becoming older and much more racially and ethnically diverse.
The findings of this study imply that unless more non-whites are recruited as boaters, it
can be anticipated that boat ownership and participation will decrease in the coming
years. Until very recently efforts to recruit new boaters have been targeted at middle and
higher income whites with a tradition of boating. Focusing primarily or only on recruiting

traditional (i.e., white, rural) boaters will not be adequate to stem the expected decline.

A decrease in boat ownership and boating participation will not only reduce the
amount of revenues to fund the development and operation of boating facilities, it will

also decrease the size of the future market for boating businesses in Florida. This in turn

132

will have important consequences for Florida’s economy. A recently released study
(Mahoney et al., 2009) estimated that registered Florida boat owners spent $3.38 billion
on their boating trips and $5.15 billion on craft storage and upkeep in 2007. The $3.38
billion in trip spending had a direct effect of $697 million labor income, $194 million in
indirect business taxes, $1.18 billion value added and approximately 26,000 jobs.
Including secondary effects the total contribution was over 38,000 jobs, $1.08 billion
labor income, $284 million in indirect business taxes and $2.04 billion value added. The
$5.15 billion in craft-related boater expenses in 2007 directly supported over 39,000 jobs
and $1.9 billion value added. Including secondary effects, the total economic contribution
from craft- related spending was almost 59,000 jobs, $2.0 billion labor income, $442
million indirect business taxes and $3.3 billion value added. The combined contribution
of trip and craft-related spending to the Florida economy is over 97,000 jobs, $3.1 billion

labor income, $726 million indirect business taxes and $5.3 billion value added.

A high percentage of Florida’s population reside in urban areas and the results
indicate that urbanization is an impediment of boat ownership and participation for
various reasons including less places to store and launch boats, higher costs of storage
and longer travel times to access boating opportunities. This is why efforts to maintain
and increase boating access (e. g., development of boating facilities on brownfields,

integrating recreational boating as part of commercial shipping harbors) are important.

Access to water and availability of boating boat launch facilities and marinas
affects the rate of boat ownership. Unfortunately, the cost of acquiring waterfront land,
permitting and operations is increasing at the same time that the budgets of boating

agencies including Florida Fish and Wildlife Conservation Commission are declining.

133

These agencies will need to be more scientific in how they go about deciding where to
acquire new boating access as well as facilities that they might discontinue because of
reduced demand. Models that are able to predict ownership rates of different type and
size boats could be important in these decisions. The ability to simulate the combined
affections of changes in demographics, urbanization and supply of boating opportunities
would also be valuable in planning amounts and locations of boating access. Utilizing
ownership prediction/estimation models in combination with economic impact and Logit
models

(http://myfwc.com/docs/AboutFWC/Economic/About_Econ_BAF I_09_Signif.pdf) will
assist the Florida Fish and Wildlife Conservation Commission understand the economic
implications of changes in boat ownership and required changes in the amount and

location of boating facilities.

5.4.2 Methodological implications

This study has exposed some interesting implications and the potential associated
with the use of spatial statistics to analyze problems in recreation and tourism, including
forecasting. It also confirms the importance of including spatial variables as part of
recreational demand analysis. Unfortunately, spatial models (including spatial regression
models and GWR models), which are commonly used to analyze public health, crime,
environmental, and demographic related issues and phenomena, have rarely been used in the

field of recreation and tourism.

Boat ownership rate data is spatially relative. Spatial data are inherently
autocorrelated. In nature, zonal features either influence neighbors, are influenced by

neighbors, or both. The level of neighbor influence may vary by factors such as distance.

134

In general, the closer the neighbors, the greater the level of correlation, and this, in turn,
distorts statistical tests of significance in analyses such as correlation, regression, or
analysis of variance (Cliff and Ord 1975). Ignoring significant spatial correlation between
zonal units in the regression analysis can result in misleading estimates of the parameters.
For example, in Florida, the model-predicted large power boat ownership rate in Duval
County was overestimated (22 observed vs. 30 predicted) because its neighbor, Nassau
County’s has high predicted values (44). This is important because over-prediction of
ownership rates may result in decisions that over-invest in boating facilities in some
counties and, therefore, under-invest in others. The use of spatial models in combination
with the visual display of the data may highlight unexpected relationships that would not

be noticed in a standard regression analysis.

The results of this study demonstrate the usefulness of spatial analysis for
predicting boat ownership rates, when compared with classical statistical analysis. The
significant data requirements of spatial analysis and the creation of geo-referenced and
geographically disaggregated databases have limited the ability of recreation researchers
to apply these tools. This is no longer a significant barrier, given the continued
development of more sophisticated computer technology, database software and

accessibility to more and different forms of spatial data.

Another problem related to polygon spatial analysis is, as was already discussed,
commonly referred to as MAUP - the modifiable areal unit problem (MAUP). This
problem exists because areal units (political boundaries) are arbitrary groupings and the
data within each can be aggregated in an infinite number of ways (Nelson, 2001; Bigman

and Deichmann, 2000). Different methods of aggregation produce different results, and

135

variables, parameters and processes that are important at one scale or zonal unit are
frequently not important or predictive at another scale or unit. No definitive solution has
been found for minimizing MAUP, so different approaches are commonly employed.

This study demonstrates this once again.

136

APPENDICES

137

Appendix

ESRI Community Tapestry LifeMode Segments

138

.3:ko 2:50 8 8A.. E2025. 3 mac—on 2.53 Iota—smog .m.D 05 Bob 58:0 588:8 Bacon 9.3 A2: £22305 “=2.th G a A. 9.2: A52:

2023 S A. .m.D 2: .8 28.4. $236 2: .2955 SI 60.23% 203508 co. 9 coat-$6 05 c Bob Aowqfi .82: 2E. .33..» 2:50 .o 88 EEAEG 2
wee—2. .83 2:8 2: Bob 822.8 8 58:0 .2522. o>$ A2: coofifix: 2: A327. .82.. 2:. 5.23% 2:50 .28 38.. AouthEIA .82: 9235 2:. ..

 

 

 

 

 

 

OA AAA. A. .A A40 AAA lAAA. .1A4A .4 . .A-A- AAA AA. . 1.....A AAA AA l-A4..A..AAA o. ...5 52.25. .A...
AA - - AAW AAA» AAAAAA 4AA A4Al - AA . 1.0 .AA. loA-A; .. A4. A AAA. AAA -AAl l.A1Al.A A8 .-WAEE. A. A2883. .....
AA ‘ .AA.AA.A-OAO.A1AA --111A,-AA ..AA - AA-_ .. . .1 . 0. AA -. A.4.A . 4A. A -- .4A .AA-A. lAA 9.5.. A. .o... Ass: 38565 cm...
- AA. .. AAA AA..A AAA AA.A.- AAA. AAA; - AA - A. .A - A... .A 1 A4.A Aom .AA-AAAA.l A-A-A-.A-A.-. A law-Eon MAE... .3
. AA A4A AA.A 4AA A.A -AA4. AAA-W .AA ‘ A AA A. A w - AA... 1 Azw 2A AA AAA omA A. A63. .80.? A--.
A4. 0A4.AA.A1AA. .AA-A AA. 04A - 4A. 4 .Al 44 A AA... - AAA .AAA- .-.-! .AA 4AA 4 lAaloI ...A-E- A-.
OA AAA A..A.A A; 2A A1.4 AAA- - AA A. AA1 AAA - A4... A0... .44A A- .AA. .AAA . ,. AAA-.AA. A. ago-AAA .3 .
AA - AA... AA.A 2A.. 42A. AAA A4A AA. A. A4 1:4. .A - 4A.... . - 8... AA. AA! .AAA. AAA .4. A23 seem A...
AA AA-A .A4AA AAA. AAA WA4-A 4AA- 8 - .AA AA. 1 - 4A A .AAo.A-AA A. - ..... .AA-o AAA A Ae< o-am 4..
AA_ 1 AAA. A4.A 48 AA A... A4A - AA A.4A AA. A .. A- - AAA A.A 2.1:- AS. A. . .0 2.89.2». A..-
AA .AA AAA -..1-o.4.A, AA.-.-Al-. AAA. AAA - A4 AAA - _A-- A 1. .A. . AAA.A8.A41 AAA AA. .2; AAA-5? 0.32.: A...
AA - 8.. AAA A .AA AA4A 8....Ao.A - AA ‘ A64 - 2. A AA. . l O4A.A.A-A.-A.4 -l-oAAdAA 4. 1328A Ali 3
3.5.5 25285
AA AAA.44.A A85... A A.4.oAA 6 .4 A4.A A... 238.2 vo.A4A.A 3.5....
A... AAA-ASA 4AA.AAA A .A.4AA .A AAA AAA 4.... AAA. . AAAoA A.A..AAA..A: . Aasm 3.5
Mfimhwo MAW—MD 5.53 “oz ofiomfi W22: 09.. 02m 2: mm.“ “MA—”....W Iona—39m A2283:
AeoI 5.82 5.3.). 5.34,. 5.25 5.3.2 AAEAE ASA - ASA .

 

2an $583 538:2:me 2402054

3% - o m
2...? .382... ...A

139

Appendix A: 2009 TapestryTM LifeMode Segmentation Description

Source: www.csri.com/library/brochures/pdfs/tapestry-segmentation.pdf

Ll High Society

The markets in High Society are affluent and well educated. As a result, the
median household income for this group, $104,934, is almost twice that of the national
median. Most households are married—couple families residing in affluent neighborhoods
Although this is one of the least ethnically diverse groups in the United States, it is one of
the fastest growing, increasing by 2 percent annually.
L2 Upscale Avenues

Success has been earned from years of hard work. Similar to the High Society
segments, many in this group are also well educated with above average earnings. The
median household income for the group is more than $70,504, and their median net worth
exceeds $178,285. Their leisure activities include sports such as golf and bicycling and,
of course, domestic vacations.
L3 Metropolis

They live in older, single-family homes or row houses built in the 19403 or
earlier. Those living in larger cities tend to own fewer vehicles and rely more on public
transportation, but the majority of markets in Metropolis feature commuters to service-
related jobs. The median value of their homes is $166,249. Employment status also varies
from well-educated professionals to unemployed. The median household income of the
group is approximately $41,099. Their lifestyle is also uniquely urban—and media

oriented.

140

L4 Solo Acts

The Solo Acts summary group features singles who prefer city life. Many are
young, stattup households located in America’s more densely populated neighborhoods;
some are well established singles who have eschewed homeownership and child-rearing
responsibilities. Second only to High Society, this group tends to be well-educated,
working professionals who are either attending college or already hold a degree. Their
incomes reflect their employment experience, ranging from a low median of $40,400
among the newest households to approximately $91,000 among established singles. With
considerable discretionary income and few commitments, their lifestyle is urban,
including the best of city life—dining out, attending plays, and visiting museums—and,
for a break from constant connectivity, extensive travel domestically and abroad.
L5 Senior Styles

Nearly 14.2 million households in the nine Senior Styles segments comprise one
of the largest LifeMode summary groups. As the US. population ages, two of the
fastestgrowing American markets are found among The Elders and the Silver and Gold
segments. Senior Styles segments illustrate the diversity among today’s senior markets.
Although incomes within this group cover a wide range, the median is approximately
$44,094, attributable mostly to retirement income or Social Security payments. Younger,
more affluent seniors, freed of their child-rearing responsibilities, are traveling and
relocating to warmer climates. Settled seniors are looking forward to retirement and
remaining in their homes. Some of the older, less privileged segments live alone and
collect Social Security and other benefits. Their choice of housing depends on their

income. This group may reside in single-family homes, retirement homes, or high-rises.

141

Their lifestyles can be as diverse as their circumstances, but senior markets do have
common traits among their preferences. This is the most politically active market group,
from voting to participating in election campaigns. Golf is clearly their sport of choice,
from playing to just watching the Golf Channel. They read the newspaper daily and
prefer to watch news shows on television. Although their use of the Internet is nearly
average, they are more likely to shop through QVC than online.
L6 Scholars and Patriots

This summary group is unique in the Community Tapestry system. Their shared
traits include youth, with the attendant lower incomes, and atypical environments such as
college life or military service. Because of their transient lifestyle and lifestage, their
homeownership is low. Most live in townhouses or apartments, although one-quarter
reside in single-family homes. One segment, Military Proximity, is dominated by military
life; the other two, College Towns and Dorms and Diplomas, are predominantly students
who are pursuing college degrees. Although most of the military market is either on
active duty or employed in civilian jobs on military bases, the students tend to work part-
time at low-paying jobs to support themselves while attending school. However, low
personal income does not inhibit their lifestyles. Scholars and Patriots residents are the
most active participants in a wide variety of sports—from swimming and snorkeling to
skiing and ice skating. They are style conscious; well connected with PCs, cell phones,
and MP3s; and just beginning to acquire household fumishings.
L7 High Hopes

High Hopes includes Aspiring Young Families and Great Expectations. The High

Hopes group seeks the “American Dream” of homeownership and a rewarding job. Most

142

live in single-family houses or multiunit buildings; approximately half own their homes.
Many are willing to move to a new location to seek better opportunities. The residents in
the summary group are young and college educated; one-third of the householders are
younger than 35 years. Their median net worth is more than $58,793—nearly 76 percent
of the US. median. Households in this group include a mix of married couples, single-
parent families, or single persons.
L8 Global Roots

The common thread among the markets in Global Roots is ethnic diversity. Las
Casas and Ne West Residents represent a strong Hispanic influence in addition to a broad
mix of cultural and racial diversity found in Urban Melting Pot and International
Marketplace. Typical of new households, Global Roots’ households are young with
modest incomes and tend to rent in multiunit dwellings. The youth of this group reflects
recent immigration trends; half of all households have immigrated to the United States
within the past 10 years. The households range from married couples, typically with
children, to single parents to individuals who live alone. Because households with
children dominate this marketplace, it is not surprising that spending for baby goods,
children’s apparel, and toys is higher here. Residents of Global Roots are less likely to
have home PCs but just as likely to use cell phones. They maintain ties with friends and
relatives in their countries of origin with foreign travel.
L9 Family Portrait
Family Portrait is LifeMode’s fastest-growing population. The growth is driven
primarily by the rapid increase in the Up and Coming Families segment. Youth, family

life, and the presence of children are the common characteristics across the five markets

143

 

in Family Portrait. The group is also ethnically diverse: nearly 30 percent of the residents
are of Hispanic descent. The neighborhoods are predominantly composed of homeowners
who live in single-family homes. The majority of households include married couples
with children who contribute to the group’s large household size averaging more than
3.11 persons per household. Their lifestyle reflects their youth and family orientation—
buying infant and children’s apparel and toys. Visits to theme parks and zoos are popular.
Their vehicle of choice is typically a minivan or a full-size SUV.
L10 Traditional Living

Traditional Living includes four markets that convey the common perception of
middle America: hardworking, settled families. The group’s higher median age of 37.8
years also conveys their lifestage—a number of older residents who are completing their
child-rearing responsibilities and looking forward to retirement. The aging of the
population has not slowed their participation in the labor force. They work hard to earn a
modest living and typically own single-family homes in established neighborhoods that
are experiencing slow population growth. Residents in Traditional Living’s segments buy
standard, four-door American cars; belong to veterans’ clubs and fraternal organizations;
take care of their homes and gardens; and rely on traditional information sources, such as
newspapers, for their news.
L11 Factories and Farms
Factories and Farms segments represent rural life—from small towns and villages to
farms. Employment in manufacturing and agricultural industries is typical in these small,
settled communities across America’s breadbasket. Population change is nominal, and the

profile is classic. Most households are families, either married couples or married couples

144

 

with children. By age, the residents of Factories and F arms mirror the US. distribution,
with slightly more retirees. Median household income is a bit lower, almost $39,699, but
so is home value, almost $100,002. Most own their homes. Their lifestyle reflects their
locale, emphasizing home and garden care, fishing and hunting, pets, and local clubs.
L12 American Quilt

Location also links the four segments in American Quilt—America’s small towns and
rural areas. Unlike Factories and Farms, this group represents a more diverse microcosm
of small-town life, including the largest segment of Community Tapestry, Midland
Crowd. Manufacturing and agriculture workers remain part of the local economy, but
American Quilt also includes local government, service, construction, communication,
and utility workers. In addition to farmers, American Quilt includes the Rural Resort
Dwellers segment, an older population that is retiring to seasonal vacation spots, and the
Crossroads segment, a younger, family population that favors mobile homes. Households
in American Quilt are also more affluent, with a median income of $44,478, and more are
homeowners. However, the rural lifestyle is also evident, with fishing and hunting (and

power boats) and a preference for pickups and country music.

145

Appendix

ESRI Community Tapestry Urbanization Segments

146

.8:on 0:28 No 34.. 22456 8 mac—on 2:33 32333303 .m.D 2: ENE :82? b45338 2383 23 .45 $3383 35203.5 N N_ 22: N52: £033
G N_ .m.D 05 N8 208 495456 05 6.3858 5 1N Ab??? 323808 2: 3 606333 05 c Sat Nomad. 535 2:. .8:on 3:50 8 SE 2.95va 9
wee—on .424 2:3 05 ENE 80352 :4 58:0 .2883 92 345 coca—ox: 4.5 N265 4835 2:. 53.533 0:58 4:4 .468 NoNtNEEa 483:— on>5 2:. .—

 

 

 

- 4N1 NNN 1NNNJ NNN.NNN No.4..NNN1NN N.NN 1-4.NN NN. o NNN N4N mN NNN NNN. N = .443. 3 _3
- a - 124 .4431 :2: ..N11 3.1.421 4.4 1 N.N..-- 4.4. N. -- -44.. - , 4.4444. 3 NNN1. : 2. ..... 4.44.445
- . 4N NNN-314$ NNN _NN ...144N1N.1.NN1 ‘ N4 N44 ..... NNN -1 NN. o -- NN41.NN1N1N1_ NNN1N1N4N - - 14:34:45 .N31
4N- 1-4.NN NNN N44 4_ _N 1NNN.. _-1NN .. N4. 4. _ 4 NNN - 4NN NNN. N414. NN -NN_-.N.NN _ ..-1-4334fi51N14495-94NNN3
- 4N -1 1NN4 NNNN N.N.NN.NN HNNN.131N.N _4. .34 NNN - - NNN ..NNN. N41N. 4N -o_N _NN.N_. _NNNNNBN 4.483% N3.-
- N4 -1 1.4.NN. N.NN NN_.4-_N111-NNN.1N-NN NN _ _N . NNN NNN _NN NNN-.N_ NNN. NNN 1N 1.1 _1_ 45450 44443 .43,
- N4 N.N-N1 44_N14N_.-NNN11..-14NN.1NNN _, . N4 4.4m -- NNN - N4 _ - N.N.o NNN.1N.1.N 4NN. 832153340 4483- .N3
NN--1114N41..4N_N1NNNNNN-114NN NN-N , . NN... SN - NNN : NN. N 1 NN.4-. NNN 1% N.N. NNN N._... -. 1: 44.90 9442 4.13.
4N - NNN1. 4N.- N NNN. NNN-N NNN _NN. NN _NN NNN - N. o - NWNNN 4NN. N... :NNN 4N_. N1. 1:25 244: HN3
NN1. NNN _N._N _N_.N_N.1--1.N.NN 1N..NN. _ NN- .4..NN NNN 1 44.4 -- N-NN NNN N. 6.4. NN4 N11: e44441o1w4e34Naa1-4N .N3-
441. . NNN.NNNN _NN NNN .4NN..N.4N - NN N.4N - NNN ..... - NNN -- NNJN .NN-N. 1N..N..1 1N44. 141NN N1- 4443an 4.3.3 _4N5514N. _3-
- - . -1- 111-1 , 1 ..... - 1N.—3.3a ...c1:w.u..-.=-B.N1D1-1 -
.mN. -NN-N..w4.__N NNN .43-N-- .1-N._41..N1-NN 1 _4 _4 . N.N..N N._..‘ 2141. No.3 NNN..NwN.-N144:2N
44 NNN.N&N 4NN NNN. N.N. 4NN 1 G1- 1 . N.NN - NNN -1431 -1 .NN.N._NN .NNN- 4N_..NN N: 1 -11 14.44.45 444243
4 o o: 4 2:8: 4 owes
afiwomflmo canon :HMM‘VBZ I: _ _xouE ow< 0M5 1: MM.“ Each—MW 32323304— Nu_o:omaom
0:5: 54:52 .3 2 .8522 35:35 5622 o Eo>< moon- ooom

 

0341—. bwfifism couficoamom cormfigfiD szNNoQNPmooN Tm 034;.

147

Appendix B: 2009 TapestryTM Urbanization Segmentation Description

Source: www.esri.com/library/brochures/pdfs/tapesay-segmentation.pdf

U1 Principal Urban Centers I

Principal Urban Centers 1 represents the most affluent populations of the
country’s largest metropolitan areas, those with populations of 2.5 million or more.
Residents of the big cities share a lifestyle that favors apartments to single-family homes,
public transportation to cars, and cats to dogs. High population density personifies big-
city life with its attendant inconveniences such as high rents and higher mortgage
payments and opportunities such as high-paying jobs. Households are younger and as
likely to be singles as married couples. Professional employment is typical but so is
diversity. The Principal Urban Centers 1 summary group is home to urbanites who
embrace the amenities of city living from the Starbucks on the corner to museums,
dancing, and dining out. They own the latest in electronics and use the Internet for
everything. If they use the Yellow Pages at all, it is to find a taxi or a locksmith. Because
going out is more popular than staying in, home improvements or fiirnishings are not
popular here.
U2 Principal Urban Centers 11

Principal Urban Centers 11 represents the aspiring populations of the country’s
largest cities. This is the youngest (median age of 28.3 years) and most diverse
population among the Urbanization groups including many recent arrivals in large
“gateway” cities such as New York City, Los Angeles, and Chicago. Although the
population density is second only to Principal Urban Centers 1, it is still significantly
lower. The search for affordable housing has moved these residents away from high-rises

148

and into row houses, duplexes, and relatively lower-density buildings. Their lifestyle is
characterized not only by their locale but also by their youth and nascent socioeconomic
status. Their median household income is $26,999. They are more likely to use public
transportation and less likely to own their homes. Families are also more common in
Principal Urban Centers 1]. Residents are more likely to buy baby goods and groceries
than electronic gadgets and to visit a theme park than a museum.
U3 Metro Cities I

Upscale homeowners living in densely populated cities characterize the eight
segments in Metro Cities 1. Their distinction lies in the single-family homes in
metropolitan cities. They embrace city living with the benefits of suburban single-family
homes. Metro Cities 1 and Suburban Periphery I residents have the highest income
among the Urbanization groups, but Metro Cities 1 residents are second to none in wealth.
Both their median net worth and median home value are twice that of the national level.
Most householders are older than 35 years. Nearly 60 percent of the households are
married couples, both with and without children. These well educated residents are avid
readers, particularly of novels. They are very active in financial investments, health
conscious, and enjoy traveling—both domestically and abroad. They are also world-class
shoppers, from home furnishings to women’s shoes.
U4 Metro Cities 11

Metro Cities 11 segments are found in larger cities and in densely populated
neighborhoods, ranking third in population density behind Principal Urban Centers I and
II. The eight markets in Metro Cities 11 are neighborhoods in transition including young,

starter households; retirees; single-person households; and families. Most householders

149

rent in multiunit buildings. The young population remains mobile. Many are enrolled in
college; most are still trying different jobs. The median household income of this group is
$41,272. But household wealth varies from $10,235 (Dorms t0 Diplomas) to $170,490
(Retirement Communities), reflecting the wide range of age and lifestage in Metro Cities
11. Consumers in this group share a neighborhood with an emphasis on economy and
convenience. Their preferences include compact or subcompact cars, fast food, and
convenience stores. With the high concentration of renters, tenant insurance is common
while home improvement projects are not.
US Urban Outskirts I

The segments in Urban Outskirts I reside in higher-density suburban
neighborhoods spread across metropolitan areas. Many of these neighborhoods are part of
the main hub of social, cultural, and economic activity within the metro area. The
proximity of higher-density suburban areas to places of employment and entertainment
venues combines the convenience of access with the advantage of affordable suburban
living. The median household income of Urban Outskirts I residents is $55,592, on par
with the national median, although the population is slightly younger with a median age
of 33.9 (compared to the national median of 36.7 years). As in established suburban
communities, the housing stock is dominated by single-family dwellings but includes
rental apartments to accommodate younger households with growing incomes. “Do-it-
yourself” (DIY) projects are popular here, with owners tackling home improvement
basics such as patios, fencing, flooring, and lawn care. Residents enjoy an active life that

includes a variety of sports from bowling to roller-blading. Televisions are ubiquitous,

150

with as many as four television sets in many homes, but residents are as likely to read a
newspaper or listen to the radio.
U6 Urban Outskirts II

The settlement density and housing preferences of Urban Outskirts II are similar
to Urban Outskirts I——-high-density suburban neighborhoods in metropolitan areas.
However, here the homes are older and the population is younger, with a median age of
31.1 years. Homes can be single-family or multiunit dwellings, but nearly half of the
housing units were built before 1960. Less than 10 percent of the housing is less than 10
years old. Homes are affordable, with a median home value of $81,472. Half of the
households own their own home, although the younger population is less affluent, with
household income approximately half that of the national median ($53,154). This group
includes a variety of household types ranging from the ethnically diverse family
households of Southwestern Families to the shared and single-person student households
found in College Towns. Their lifestyle preferences include Folgers coffee to Starbucks,
current consumption to saving, going to the movies, participating in recreational football
or basketball games, and attending these collegiate or professional sporting events.
U7 Suburban Periphery I

Moving away from the epicenters of city living, peripheral suburban expansion
represents lower-density housing development located in metropolitan and micropolitan
Statistical areas throughout the United States. Suburban Periphery I is the largest
Urbanization group of Community Tapestry, with the most population and households, in
addition to the highest annual growth, 2.] percent annually. Married-couple families

dominate, approximately half with children, primarily living in their own single-family

151

 

homes, with two cars. They tend to employ a lawn and gardening service, own a security
system, and invest in home remodeling and improvements. This well-educated group is
second to Metro Cities 1 in household wealth, but second to none in conspicuous
consumption. They track investments on the Internet frequently and use a financial
planner. They enjoy golfing, skiing, hiking, water sports, and regular exercise at a club.
Travel is part of their lifestyle but more domestic than foreign. At home, The West Wing
and CNN are TV favorites.
U8 Suburban Periphery II

Suburban Periphery II incorporates a population density similar to Suburban
Periphery I but is more likely to be found in the smaller cities within metropolitan
areas—in urban clusters. Housing is still predominantly owner-occupied, single-family
homes but older and closer to employment. Residents here have the shortest commute to
work. Households are a mix, similar to that of the United States as a whole. More than
half are married~couple families, and one third are householders who live alone.
Although the median household income and home value are below the US. median, their
median net worth is slightly higher. This is the oldest Urbanization group of Community
Tapestry, with the highest median age of 41 .1 years and the highest concentration of
householders who are older than 65 years. They are more inclined to watch sports than to
participate, with the exception of a little golf or fishing. They prefer Folgers
decaffeinated coffee and enjoy gambling, watching QVC and the Game Show Network,

and frequenting family restaurants such as Bob Evans Farms and Perkins.

152

 

U9 Small Towns

Small towns represent the ideal in American communities—affordable, close-knit,
and apart from the hustle and bustle of city life. The Small Towns Urbanization group is
typical. Residents are active members of the community including membership in social
clubs and church boards and participation in local politics. Households earn a modest
living, with a median household income of $37,599, but their earnings are sufficient to
afford a single-family or mobile home. Most of the labor force is employed in
manufacturing, construction, or retail sectors; many are already retired. Heartland
Communities is well settled, but Small Towns welcomes the ongoing migration of
younger Crossroads and older Senior Sun Seekers. With retirement still looming for
many, they invest conservatively in certificates of deposit and annuities instead of the
stock market.
U10 Rural I

Small, nonfarm settlements, some of which are developing in suburban fringe
areas, characterize the neighborhoods of Rural 1. Married-couple families, many with
grown children who have left home, work hard in blue-collar occupations. Some are self-
employed with small businesses or farms. Their median age of 40.4 years is slightly older
than that of the United States median. With a median household income of more than
$53,210, they enjoy the comforts of large single family homes with ample land. As do-it-
yourselfers, they take pride in their homes and gardens, investing in major home
remodeling projects and the tools to get the job done. Residents of Rural 1 may not be
farmers, but they embrace the country lifestyle, from their gardens and pets to their

pursuits, hunting and fishing. The vehicle of choice is the pickup—domestic, of course.

153

 

U11 Rural 11

Rural 11 represents the countryside of the extremes in urbanization. Low
population density characterizes the country with its inconveniences such as the need for
multiple vehicles to get around and advantages such as affordable single-family homes
with land. Most of the population resides in rural farm areas; the rest live in the country
or in small villages and work in mining or manufacturing. Residents are slightly older

than the US. median, with a median age of 39.5 years; some are already retired. Most are

 

homeowners. Few are movers; rural residents are settled. Family and home are central in
their lives. Their lifestyles reflect a preference for comfort and practicality—western or
work boots to dress shoes, kerosene heaters to espresso/cappuccino makers, recliners to

patio furniture, garden tillers to trash compactors.

154

Appendix

 

Geocode Match Rates by County at Street Level and Zipcode Level

155

L’leoosL L
..LL12091_LLBrLaLdfoquL
_ __12099

Table 01 Geocode Match Rates by County at Street Level and ZCTA Level

County F [PS
LL12L0LO 1L, L
L .1200}

County Name

.Alachua
LL Bake! , L
Bay

Brews! LL. .. L

12011 Brgwarg L

 

 

 

 

 

__ LL 129.8}; L

 

 

’12085 ‘Marti‘n

LEDLsBL‘LMFSni-Lag; " "

 

12087 Monroe

L_LMari9Lr1LLLLL L.

Geocode Rates at

LL L73-L2L7%L
L 50.65% L. L
L 3Z-06%LL

Street Level

L L L693L1L%LL
.. . 2.15% LL

‘ _V_.L94-i4%L

 

156

73.56%

72.25%
._ L 38.78% L. _A .

Geocode Rates at ZCTA Level
L LL2613L%LL

L _LL49-L3L5% _L

LIZ-94%
..... L L _ LL33L-6L9‘V9
L L ______ LL7L-4L5L%L

Lise/o- f

_ 12013 CalhgunLL _ LLL37L71_%L L LLL§2-29L%LL ..
LLLL_129_1L5LLLL_CLhaLrL19LtLtsL L L LLLL9Ll-4Ll°/:>L L L LL8-_59L% -L L,
12017 CitrpLsLL LL _ L L 81-27‘V9LLLLL L L L _ LL12L‘13%LLLL
12019 ClaLLL _ L L ,,,L§8L7LS%L L ,,,,,, L L_1,1L22%L
L LIZOZLLLLLCQllisrL _..A.L8L9L-.17°L/<LL L _,L19:8L3%LL
LLL12023LL LColuLmbfiaL __ ___L2L1-94,‘L’L/g_. _7L8_-96Lf’_/9
L LL2927L LDeLsotoL L ..... 76-1L5L% L 23.85%
L. ..L1L202_9L LLQiLriLe L L 4:L6L8%LL_ L 2.5-32% ‘L
L _LL12931L LQuLvalLL,. L _L L L2L3L-L3ft%L _ _ ._ _Lé-L66‘1/o.
LL1L2913L3LLLLE§§Lamfl. _L L _ LL22L-Q9L‘V0LL ,,,,,, L . ._ L L7-9L1L‘72 L
LLL1293L5LLLEIasLler LLL L LL§6L-§9L°/2 _ L LLlL3L-L2LO‘VLL .,
_ LLLL12L0L3ZLLLLFLrankLlLinL L 53-46%LLL, L L .49-5.4%...
L LL12932LLLL92L1d149n LL L L .L85-L2L1L‘1/2L. _LL .. _LL1%}L79L°/9L
..LLLZLthLlLLLLLQiLlsLerLigLL LLL .LLLLLL7L‘1L-Li89/0L L .LZLSL52"/g ‘
LLLILZQL4§LLLL9LESIES LL L6_L9-2_L1%LL L LL3LQ-7L9f’ZoLL. L
__LL2055LLLLQulf LL __ ,L L _L___.__.64L-2L2°/9L L, L _LL3LSL-78"/oLL
1294.7 Hamilton. , LL55-63‘3/9L LL _Lflfl-3L7%LL,
1294L9LLLLHaLd2 _ L 69-86%LL L L, .L L39-1_4°/2L
L LL1L205LI L __LLHendo/LL 58-93%LL L Lil-92%LL
L LLILZOSC: L HemandoLL L LL91L-L7L3L%L , ,L -15-27% . L
. _LZOLSSLLL..LLHigh1LaLnds ._ L L 8L2LQ8L°/2LL L -17-22%- L
L__L120LSZL L BillswrgughLLLL LL L _ Lw92_-§§%LL _ L, _LL7-_1_3_2%LL
LL 12952LL LHQILrueLs L LL 156-81% ‘ ‘ , .LLL3L3-_1.9%LL _, .
L ..LLLZQLQIL LLLlnLCEaLnLBLiyeLLL L §L7LL4Q%L _. 112L60% ,,
L -LL1L29L62LL LittCkL-m L_ __ , LLLLL7§:93% 31.97% L .
LLL1L20L6_5L LfifferseuL LLflzL-LZ3L%.LL L L57-77%LL _
LLLLIZLQL6L7 LL Lafayette. L 13L-29‘72LL H L8L6-71%L_
LLL12969LLLLLLak€=LL L84-54% .LL1_§.-46%L L
L LLLILLZLOZLI L LL59 L§5L-83% _LLlflL-17%L__ .
LJLZQBLLLLLLeLoLnLL L9L2L3LQ‘f/o .V LL7L.L61% L
. LL12LOL7_5L LL_L,LI:9vxLLLLLLLLL . 92e.7_1L‘3/2__ .L.LL_.I:Z-L2_9%
__ ..LLL12LO7L7LL Liberty- _ L LLLleL-4O%LL .,LLLL§ZL§9% _, .. L.
LL2972L _LMLaLdLiLqun L L292L5%LLL L L LIB-23% .
L L120}! Manatee-.. _ _ L 87.93349L _ _. L_L1LZL.§7_%L _L

20.25% g,

11.22%
9.17%

26.44%

 

Table C-l Geocode Match Rates by County at Street Level and ZCTA Level (con.)

 

 

 

County FIPS County Name Geocode Rates at Street Level Geocode Rates at ZCTA Level
#12089 t fiNassau A, H _“ 374.40%, A, '* ' ”762.60%
______u12091fl W Okaloosa V _ .4 82.75%» _ 17.25%
LL12093LL _LleeEhQDee §0-63°/L° _ .19-3.7%L
712095 _ Orange V__92.75% i __‘_7_k7.25%41_
f 12097 *Osceola 7 792.32%,” _ w768%
_ 12099 2 Palm Beach 93.14% 6.86%__
#12101 M Pascofi 7 93.68% 26.32%
p 12103 A Pinellas «91.48% 8.52%, a
#12105 flPolk fl 82.13% 7 A 17.87% i
__ 7512107A7%_ Putnam ”68.90% W 7 “31.10%
_LL12LIL92LL _LSantaLliosa LL,83L-L72f’L/9_ V LL _VLLLl6-23%L
L VLleLlLlVl LLLSaraLSOLtaL _V L L_§L8-1_3L% L L _ Lil-37°41-
.-.L12LL13L LL LsemiansL -.LL?Q-§4%LLLL L9-36‘V9L.
fl Ml_2_11§ _ _rSt._ilohn+si ‘_ W_92.14% W _A 7.86%_‘
--.VLLlZLUZ LL51; 1:991? L L _L_93:02°/9LL _L 6-98%L
_ V12LL11L9L LSuLmterL V LLL6§26%L __ ..-LL_33~743/2._.
712121; v Suwannee _73.84% p _ 26.16% __
12123 2 _Taylor 59.08% _ 40.92%
12125 7 Union A 18.51% _V “81.49%“ .
v 12127 _H Volusia . 91.40% 7 ‘_ 8.60%
_V 12129 _ Wakulla 77779.62%7 220.38%
_121317 ##Waltonfl’ _ 281.63%Aw _ ___l8.37% _
12133 Washirgton 75.70% 24.30%
Total Match Rate 86.50% 13.20%

 

157

 

Appendix

 

Number of Marinas and Boat Launch Sites by County

158

Table D-l Number of Marinas and Boat Launch Sites by Counties

 

 

County County Total Total Number Total Total Number
FIPS Name Number of of Boat Number of All Types
Boat Launch Launch Sites of 1 of Marinas
Sites for Public Use Mannas
1151120051155A13c5hu1a5 1155 27155 5 17 15 2 4
5 5120013155 5131151156155 1 755 77777777 57 77 077 7 77 70 7'
5 51250155535165 _ 5 825 7 77367 7157 7 7 95 7'
5 512007551 Bradford 1 510 7 7 7 6 77 0 7 7 777 177
_1 71200971577 1 Brevard 84 7 777747 7307 7 7 7751
5571201175 Broward7 61___ 34 737 292
17712013 5 7Ca1hou11 20 7 7 0 _50
55 55120157 _55C115a5r1oue 26 5 _ 7_ 5175 82
55 12017 01171175 5* 535 351 _ 615 1 287
1551207195115 Clay 5 _5 5555 555 251 5 1 10 5 5 8
51757120251 Collier 55 541 _ 18 22 5 55107
LLLL129L2L3LLLLC°LILHmbIa L LL 21L, L 13 _ .0, L L 0 _
_ 5_551202’57155515I5)e5o1to5 555 9 5 6 1 1 3 1
15.211291551315111: 55 6075 20 5 13
1__ 152051551 D1155val 5 507 7 17 _ 29 5 2555 _ 5 66 __
155112031315 Fscambia 55 48 2615 20 5 5 63 5
11 5 12035 F lagler 15 _____ 5 7 _ l 8
12037 Franklin 72 26 5 10 5 35
__ 7512039751 _7Gadsd7en 14 9 2 5554
511511_2O415LGil1<:hrist51 20 1 15 0 5 5 50
5 512014555 5G1lades55 5555_ 551 335 1 ll 6 5 153
LLLILZLQfiéL LL 59”. . ,43 L 19L- 1.V. L L, __?_—L
5 512104171_ Hamilton 5 5 33 55 1 9 0 5 __ 0
LL13949LLLLH¢L<17¢€LL 10... L7 ._ . 0,, L lL
LLLL1L295LIL Lyeogry L 7L 7L .4. 2- L L 6.
55129513151 1115611115111166524 __ 5 12 51 6555 521
51 120555115High1ands5 92 47 5 51155 21
_ _ 51512057 5Hi_llsborough5 84 34 12 451
5 712059 1 Holmes 12 6 5 505 0
551206] Indian River 1 22 14 9 39
5 57120653571751 Jackson 26 19 _1 5 l _5
5 1152065 517Jeff7erson l9 5 0 50
_5 55112062 55 Lafayette 5 115 5515 5 8 0 10
55 511205695 55 5l:a5l5<1e5 127 _ _ 55 64 ll 5 5 33
51 15120751151511.9855 93 42 61 1595
11 120173 Leon 40 1 295 O 5 155
L 1.20.77.3L7 7LLIL¥S7VXL7LL L L 34L L 24 4 .17.
5575120575711 7 55Libe5rty 24 _ 20 50 5 05
551207911 Madison 5 7 1 4555 5 0 5 0
15120815 5 Manatee 5515 5 5 17 25 48
11207813 Mar17on 55 31 1 515 5 5 1 5 1
777120875 7 Martin 28 12 25 89

 

1. including Marinas, Dockominiums, Private Clubs, Hotel/Restaurant

159

 

Table D-l Number of Marinas and Boat Launch Sites by Counties (con.)

 

 

 

 

 

 

 

County County Total Total Number Total Total Number
FIPS Name Number of of Boat Number of All Types
Boat Launch Launch Sites of 5 of Marinas7
Sites for Public Use Mannas
55512087 5 M5onroe5 51956 5 5515 15 5555455 5 111111 75 55 5 53950 _
11250859 555Nassa151 207 _ 5125 f 6 5 145
1_ 512095115 okalgosg 67 f _ 345 13 695_
. 515120935_55Ok€7:57echobee 5 22 5 f 7 _7 _5125 71 28
LLL1292L. 24422-5)- ’ 74 _L _ 10, _ Ll IL e L7
LL1L2<97LLLL9444444LL L 40L L 12 2L _LL-16
___LLL42999LLIzahJLBeEh' ”59- LL L ___.34 L33 L . 165.”
12101 Pa-sco _ 7754-LL ,A L 10 L 11 L 42-
L 12103 P4414944 L'_ L__ 9.87 L__. _LL-49L L __ __5-7... L L_.189L.___
_LIL21-95L LL999<LL ’ " _L241- _ .89 - L4 LLLLLLL LL56-
42-197 LLPutnamL L 7 44 7 L23 5 29 __
512109 581-5156111115 _36 5 241 5 5 3 5 55
5121111 1 55St.Lucie 28 155 8 5 615
LLL_12L1L13LLL349t4_R059 7.2. __ 37 ‘10- 8
4293-3224044 - 41 LL18 2 L24-
1211L7LLLSL€9119919 '20- ,,,,,, _ 10L 42 L47
L _L1L21L19LLLSuny-4r :1 19-LLL LL12 L _0LL -L 7
- 12121 SUW29969_LL 13 ___-LL LL 9- L 9__
L_L12g-Lir4warLi L25.- L-L1L6LLL 4 1L1--.
L_LLL12-L‘LZ5LLL U919“ LL 5 _LL-4L 9L 9
L__.LIL2-127L39usff” '7 118 , __L 47L- L1L9L L .74. L_
L__22L12L9 LL W494? 62 _ ..37 ,_ 4 2L
L-LI2-1L21LL LWalLtonL7 69- L39L .4- 9
12133 Washington 77 42 O 0
Total 3161 1466 669 2820

 

1. including Marinas, Dockominiums, Private Clubs, Hotel/Restaurant

160

 

Appendix

Figures related to characteristics of boat owners by LifeMode Segments

161

Percentage of Boat Owners

Percentage of Boat Owners

Figure E-l Characteristics of boat owners by LifeMode Segments
- All Power Boats

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

35%
30%
25%
20%
15% \ //
10% ’—
o - '0 - /
5% \ I 9‘ ’ A I I
\J
0%
. d 0" \{9 ('4 e5" \9 0 (9 § . qo \"-
o‘} 290 oQ oYSJ ‘9 \ $0 QPQ $90 0&1, V‘A (if Cf»
G" Y'4 €36 0\ -0 Q ‘9 ‘1} \ ‘8' . "09
.\§ \0 V“ ‘3 Q 9‘ ®% \Q \‘X . 06°, 66" (\0
lb 9% ‘56 \ “>0" 5° 0° 0
0Q ° Q <56 0‘
can a «a
-—All Power Boats °°°°°° All Boats -0- Florida Households
Figure E-2 Characteristics of boat owners by LifeMode Segments
- Small Power Boats (<23')
35%
30%
25%
20%
15%
10%
5%
0%
. <73 09 \"J '6’ 0°" . ’5’ 6° \‘3’ '5‘ - °o '\"
000 :90 oQ°\ 0Y9 ‘aé $0 Q‘OQ $90 06¢ ‘90 SK O9}
0‘s VS ‘5'? ‘5‘} ‘9‘ sq 4)? ‘0‘} «sq 6‘9) 9 '0'?
er a? ca? ,5» *8 c>° re a" 0“
<2" 0\ ‘5 ob c‘o Y§
Q .56.“ 4 <55
-—Small Power Boats °°°°°° All Boats -0- Florida Households

162

 

Percentage of Boat Owners

Percentage of Boat Owners

 

Figure E-3 Characteristics of boat owners by LifeMode Segments
- Large Power boats (23'+)

 

 

 

 

 

 

 

 

 

 

_Large Power Boats All Boats -0- Florida Households

Figure E-4 Characteristics of boat owners by LifeMode Segments
- All Sail Boats

 

 

 

 

 

 

 

 

 

 

\09 . 0‘9 0% 0‘9 .l\\' . .&% §
€36 q‘y ‘29:} 5°\%0 ‘3ch \\>4 Qfi #0
e F - 0°” ‘9 '0
*8 (a? (65’ 0“
°\ ‘5» 690 9°
6? &‘ <50"
-—- All Sail Boats ° ----- All Boats -O- Florida Households

163

 

Percentage of Boat Owners

Percentage of Boat Owners

Figure E-S Characteristics of boat owners by LifeMode Segments
- Small Sail Boats (<23')

 

40%

35% IA\
30% '

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

25% v.

20% I! a

15% ' /
10% - O—
5% 1' ’ :.-"‘.A 7. ’
0%

0&6 06,09 0 cg: 0 do 930% 59‘ «29$ $99 05% x. % 6&5" (29%
5'” $4 $6 ‘9‘} 4P" 9‘ «8‘? \s°‘° «sq 09$) . J? be“
~29 cg? " V? «y 89 90° x390
0Q .5659 «4’ <5?"
—Small Sail Boats °°°°°° All Boats -0- Florida Households
Figure E-6 Characteristics of boat owners by LifeMode Segments
- Large Sail Boats (23' +)

45%
40%
35%
30%
25%
20% .
15%
10%
5% f

0%

@§ (:2? ¢0 ‘50 92$.“ 699 ‘3‘? 960% fix .-0& '09 6‘0
Q

0 0
e \‘> N o
0Q $6290 «<5 {is} Y§
— Large Sail Boats °°°°°° All Boats -0— Florida Households

164

Percentage of Boat Owners

Percentage of Boat Owners

Figure E-7 Characteristics of boat owners by LifeMode Segments
- Canoes

 

35%

30%

 

 

25%

 

20%

15% % \ j’j
" l ,»-m 1',”
W V

 

 

 

0' I
0%
4'93 $09 \‘c‘, c)" \09 .906 0Q0 O \. {66“ , 4&oo f 0}“.
e300 $5.. 6669 o\° of? {S «2* f}%' Q0 \9 9f; to
g, ~o

®§ % $6 é 930$ ‘53? x 60 “$90 ”.00be 0.06% 00

0Q ‘90 Q 's to

450 ‘g Q

 

Canoes °°°°°° All Boats -0- Florida Households

Figure E-8 Characteristics of boat owners by LifeMode Segments
- PWCs

35%

 

 

 

30%

 

25%

 

 

20%

 

15%

 

10%

 

 

5%

 

0%

 

-—-PWCs °°°°°° All Boats -0- Florida ouseholds

165

Appendix

 

Figures related to proximity of the boat owners to coastline

166

Percentage of Boat Owners

Percentage of Boat Owners

1 00%

90%

80%

70%

60%

50%

40%

30%

20%

l 00%

90%

80%

70%

60%

50%

40%

30%

20%

Figure F-l Proximity of boat owners to coast line

- All Power Boats

 

 

 

 

 

 

 

 

 

5 10 15
Distance to Coast Line (miles)

— All Power Boats

- - All Boats

Figure F-2 Proximity of boat owners to coast line
- Small Power Boats (<23') and Large Power Boats (23'+)

 

 

 

 

 

 

— Small Power
Boats

— Large Power
Boats

 

- - A11 Boats

 

 

 

5 10

Distance to Coast Line (miles)

167

15

20

Percentage of Boat Owners

Percentage of Boat Owners

1 00%

90%

80%

70°/o

60°/o

50%

40%

30%

20%

l 00%

90%

80%

70%

60%

50%

40%

30%

20%

F -3 Proximity of boat owners to coast line
- All Sail Boats

 

 

___

 

 

”
/l’
rt

 

 

 

 

 

 

 

 

/ ’ I ’ -—All Sail Boats
/ / - -All Boats
I
I
.1
I
5 10 15 20

Distance to Coast Line (miles)

Figure F-4 Proximity of boat owners to coast line
- Small Sail Boats (<23') and Large Sail Boats (23'+)

 

 

‘
—’
‘

 

 

 

 

 

 

 

 

// , A Small Sail Boats
’
’ _ .
I I Large Sall Boats
/ l - -All Boats
I I
tL
I
5 10 15 20

Distance to Coast Line (miles)

168

l 00%

90%

80%

70%

60%

50%

40%

Percentage of Boat Owners

30%

20%

l 00%

90%

80%

70%

60%

50%

40%

Percentage of Boat Owners

30%

20%

Figure F-5 Proximity of boat owners to coast line

- Canoes

 

 

 

 

 

 

Canoes

 

- - All Boats

 

 

 

 

 

 

5 10 15
Distance to Coast Line (miles)

20

Figure F-6 Proximity of boat owners to coast line

- PWCs

 

 

 

—PWCs

 

- - All Boats

 

 

 

 

5 10 15
Distance to Coast Line (miles)

169

20

Appendix

Figures related to characteristics of boat owners by Urbanization Segments

170

 

Figure G-l Characteristics of boat owners by Urbanization Segments
- All Power Boats

 

 

 

 

 

 

 

 

 

 

 

I 1.

 

 

 

 

 

 

 

 

 

 

 

30%
g 25%
g 20%
‘5
O
m 15%
‘0—
O
0 10%
2’2"
G
8 5%
b
O-
0%
(9°
o5?
.@
«9‘ .Qéfi
s“ {O
Power Boats All Boats -¢- Florida Households
Figure G-2 Characteristics of boat owners by Urbanization Segments
- Small Power Boats (<23')
30%
:0
a 25% e‘A'.
E i \
o 20% vr“
0% 150/ ‘ I
as ° I \
O o I \
on 10 /o
s I '
a I. . 0 O O o g \
§ 5% : -°/ \ \
fl E .— ' ,y"
0%
X \ N \ \ N \
05" (5 90°" . 69x -¢f° . °°\ ed C? 46$: 4% 9x
009‘ 00"?) 0C} 04‘" $539 {3&5‘ ¢§9 . g) \\«o $9 $$
i° 50° ‘1‘ V150 ~g§ 9° (5° (1,9 C9
\0 \0k 9‘ 0&0 3‘0 $0
.a‘b . Q‘b $$O $0
69" “090 °’
5 «t .
— Small Power Boats ° ° ° 0 ° All Boats - I- Florida Households

171

 

40%
35%
30%
25%
20%
15%
10%

5%

0%

Percentage of Boat Owners

40%
35%
30%

Percentage of Boat Owners

25%
20%
15%
10%
5%
0%

Figure G-3 Characteristics of boat owners by Urbanization Segments
- Large Power Boats (23'+)

 

 

 

— Large Power Boats - - - - - - All Boats -0- Florida Households

Figure G-4 Characteristics of boat owners by Urbanization Segments
- All Sail Boats

 

 

 

 

 

 

 

 

 

 

-—All Sail Boats All Boats -0- Florida Households

172

 

40%
35%
30%
25%
20%
l 5%
10%

Percentage of Boat Owners

5%
0%

40%
35%
30%
25%
20%
15%
10%

5%

0%

Percentage of Boat Owners

 

Figure G-S Characteristics of boat owners by Urbanization Segments
- Small Sail Boats (<23')

 

 

 

 

 

 

 

 

 

 

 

 

--— Small Sail Boats °°°°°° All Boats -0- Florida Households

Figure G-6 Characteristics of boat owners by Urbanization Segments
- Large Sail Boats (23'+)

 

 

 

 

 

 

 

 

 

 

 

Large Sail Boats (23'+) °°°°°° All Power Boats -0- Florida Households

173

 

Percentage of Boat Owners

Percentage of Boat Owners

Figure G-7 Characteristics of boat owners by Urbanization Segments
- Canoes

30%

 

 

25%

20%

 

15%

 

10%

5% v
r " '
0% ...... ,/

 

 

 

 

Canoes All Boats -0- Florida Households

Figure G-8 Characteristics of boat owners by Urbanization Segments

 

 

 

 

 

 

 

   

 

- PWCS
35%
30%
25%
20%
15%
10%
5%
0%
X . x \ \\ ‘3 \ \\\
096; 066659 {06 C3666 962‘)“ ‘5 (is {@066 @906 \\"o $¢$ 6‘6
o o 0
x, $9 «a 56° 2&0 ¢o (53‘ go $6“
0‘ 05° ‘1‘ €° ‘0“ ‘0 x159
\ \ 0 \3‘ ~o\§ 0‘
.69? 8" 9°“
-——PWCs °°°°°° All Boats -0- Florida Households

174

 

 

Appendix

OLS models Results

 

 

175

 

 

modA o=_w> m .N nodnv o:_m> & _

 

 

 

 

 

 

 

NSmé ~20.— _ wa _ .o _vas— New.— umou 3839.353—

~S_.m1 - Java. - 11”.».2401 _. -1 1 ..1_§.211 1 «£31 ,-11mm.1.1.a.£-amlim

was» 1 1 1 [man :1 mafia; 1 NNN; 11 w. mam 1 1 . 21mm eaflmfiozawmp
.1 1 11 ,,,,, 1 11 1-: -1 . 1 . ,,,,, 1 .11 1 1 1 “35%

m8 3 «mm 2 1 8m : E 2 .3 am 1 ‘ 8:660 @soeamzsz

8502059 zofifixoma

_zmfiomtmoo? _2m2<2.8c.o- _namSS, .6:

Nmm1aom*3m.o- _mmmh<3.$._+ _Zmadofisod-

Emma/$32.: _mmw:§~.~- _émh<3.e$.m+ _émfi‘BaE. _+ 162733. 33%; E85

12<EE$ _na m09..~8.o+ Ema—5&3? _zma no“. :23. _Ezmtm3*§._+

Ragga? 135.2- maozrmn? _mmmm::3.m$.o+ 1mm? 3%.?

$2- 82 83. :3 33”

N _ N _ _

Q 8 mm mm 2- 8:35 £5 $82

33.2 .83 218.2 .292 $3.6m 22:93.

22 $3 33 $5 8: m 8:33

and 32 £3 £3. 23 mm

5521. E52521. :35: 29224 $25 0325 “concave

 

_o>o_ fiasco “a 89518 owmm use 099 “caught 8m £sz .2082 oamzcm $qu 55th “TI mink

176

 

 

modA o:_w> m .N moduv 33> & .—

 

 

 

 

_Omo.c_ N398 N53...» N392 $3 3839.50.53—
-wemz ._Mww.e1w 1 rmmmnm 11 1 11mewme 1 1 - ,1aamwfiywmwm ._
”mama 1 ME? 1 News 1 1%? 11 11 motmwézmpuommoafl
awe- 1 33.11 1 11 a? 1 1 11“ 113% 1 stem.@maae@§@
8502059 zoammxoma

_Ema<z.moo.o+

_Emzaavvfir

_Zmodonrmbrm- _Ema<2..8o.o-

“50:33.? mmmzazflf

_ _ I fling; 868n—
_$_E<B.E.e+ WEBSSESP 135.5- ~23 89:85-
_mimkmfi _+ Na _ ozr: 6+ 11555.0th 187553-
_mmq<2*$w.mm- _Ezmtzaamw.m+ _mzmo<.m$.~+ Exam—:35 355+

E 3 ES- 83‘ :3

m _ _ _
«R SN 8 2 8:85 £5 8:82
.30».— xx _ .h .N_wém __awdm 25:8?“—
vomd 23 S3 :3 mm 8%.?
an o Em... none 83 mm
5321. $324 29521. 5:52; 032:; .5223

 

flcoov _o>o_ @550 E 38: mo onm can 090 :88me 8m $158 .2232 ohm—Em ammo; 3520 ”TE 2an

177

 

 

modA o=_m> K .N modnv o:_~> K .—

 

 

 

 

 

_Po.wm _maw. _ c _mvdm— NBS; Nmmad 33 3339.80.53—
_ $3811.11 .. . Emma-6.1 - 1 _mmwmml, 11_.am_11 1 Amen. 111-1111e1w.a,.m31.mom=am.
swowwwwwmm 1 1 1 Ram-RS _nwaofl 1 1 3.3.2 1111- raw-NR. - 1 - eoumwflmmwflozawWW
1 1 . 1 1. 111.11 1 1.11 111-.11 - 11.1 -1 1 1111111111-, 38:2

:3. mm -111 1me _w 1 ©3100 11 1 wow mm 1 wwo aN-11111 . :11. . 11111:oEv:oU bfiwwfimonsE 1

mUFmOZOsQ ZOEmMM—Omd
_E_<.._Om_._m_o.o+
_<Aom*hno.o+ _2_<40m*3c.o+ 22:509.; 86+ ~2<2*m—o.o+
_ZmQImOm*v1o_*m- _mMm—b<3*mwo.o+ _E<S*N_o.c+ _gmh<3*ammd+
_mmomemma? _mmlmoarfi. _ .N- Ede-<3...” _ ~.o+ _ZmDImOm*v1o_*m-
__Zgjomfl _ o.o+ _m02<n_*$od _ - _mmm<m* 215$- ~ZmOImOmev1o _ __.N- _5027 :1. _1 83.25; 8605
152758.: _m:v_0<qm3mm.©1 HIT—mtg imam: _Eozrmmmg- _m02<1,__..e:._+
_n—IMU<1_m*m~w._- _mm4<2*o§.m_- _mm4<2*3m.m- _LIMU<4m*mmm._- _mmv_0<qm*mwm._-
_mm0<*oam.m+ _mEmO<Lw6+ 1mg}; .2: + _mm0<*:~.o+ _mm0<*oow.o+

$.61 Qmw v3.7 Nmod oood
N _ N _ _
wmflm max” .3611. 1 111- 3m; 81..— 1 1 111coto=cnmu1£wfi§<

~39; _ogdm _ooo. mv “nmmdom _moodmm 25:8?“—
Ncmd _wmd wand awed wood mm 333:2
mend nwmd med «.36 ~36 mm
$02321— :mOZ<UZA Imm<zq 19221— Im<ZA 053.9 Eovzomoa

 

_o>o_ 4.2-UN Hm Enos mo oumm can 09? “costume com 338: 2252 083% ammo: SEED ”N13 2% 1.1

178

 

 

modA 33> & .m modnv on> as .—

 

 

 

 

_mvo.hmm _amdfi mound mvmhd “m3 3889.8.ch

11_§11§11 1 1 mama 1:32...- 1 max, 1 - 1 .. .m....§?§

,_§1.1§- 11- *3.me 11 _anxém . ...wmfim- - 9.11m 319135571. 8 mm

com-mm 1 Ewww . - 1- @81- 31._m1 1 - -1 Emmi@waw-rfimwflowww

mufimozoso zoammmomm

_2_<qom*~_o.o+

_Eomtwo? 3215989? flas).31.8.91

_mmm9<33_m._+ REL-59%”? _E§.mmo.o+ Emmzafimwfi

_mmdofimmvm- madam-.32- _fimh<3&oqo+ _zmodongrm-

2.6253,: EELS? _Zmodonrmé L- 1627816 332.; E85

102558.? _mmmbmaémf 182780? _muz<%o$._+

_Ezm::3$oo.m+ Emigowmfi- _Em::333.~+ 15:33:31

25302.? 153....sz 1553311 _mmofaonf

:3- ”8.? $3- 33

_ N N _

3mm - 1 8mm 1 1 R3 1 1 @1211 1 8:125 @53me

.9323 _33m _de _Snam 339$

2.3 83 2.2. 23 mm 85:?

£2 83 and 8.3 mm

331.24 5524 $31.73 $524 0325 20285

 

Adoov .95— <HUN as $8913 3% ES 090 #:20ch .8.“ 8:52 .flonoz 295m “934 EEEO “N1: 03m H

179

 

 

modA o=3> K .N moduv o=3> & .—

 

 

mmq<2 *Nomd +

_

m1$mti3§ofi>+ AIImEI3RN8+ mIIm—EIB 336+ ZmQIn—Om

_

_

_ _

_mm_ .va _Eb. _ m. _mvvdem _wmvSv _ _w _ .mo amoucom8m18xcoov.
.ow_..11..o$.m w, W818. 1.232 .8133 1132.851 : -.wo.§w~n.1_.-om=em
..5108..12.1 38.3 ...ME? 1 1 1.83560. . .omww-aé warm-1.... 32:02 .5 .13
. -11-.1 1 . 11.11 1 . . - 1 1 1 . 1.1111 ,1 1.111111 1 1 : .111? - 1 1111- 11 - -hODEfiz
an 1% 1 . 1%. mm: . a. 2 E1: . 21:... ammméoéaé=§ei

8.502059 20.35.05.
321.0929?
< . 1
.23. 23:... 321.0929?
. + . +
1 .EmZBKS o .25)... .118 o .23). $81.:

2m A. o s- . - < . + . +
. n. 9.6- ....~ ...momao m .2 2a... o .2.<1.om :8 o .359. .08.?

n2“. :31r - . + .. +
_ _._ _Zmo mom—#133 _mmmhsstm _ 25432 *2. m AIMEE—:3 8:32.; 8305

$559.81 ..muz...§.m1 n.<11..a:..m- $.05 N3? . .

_ _ _ _ *moh o+

 

 

...zmo... .82:- ...:mo<.w€.? ..Emwfwmqf .Zmo1mon..m-o....w- #12:-

SE- ”2...... $3- 8...- 3:

. . . N .

o3... . 98.: £3.11 .83 13.61-; 1 ‘ SE...» a... 8:87.
.838 was. .368 .82? .83... 0.2.93.
Em... am... Sm... 2.2. new... my. 3.3.6...
ME”... 8m... 8m... 3.2. 8m... mm
:uamz. 502521. 5.9.21. ..mn.<z1. :32. 2%.? 20.2.2.8

 

3>2 83.. 95:8 3 333 we on? USN 091. Eobbmv .8 $33.. .20on 031% “mama SEED ”WI 2.31—-

180

 

 

modA o:_m> & .N mcdnv o=_a> & _

 

 

 

 

139mm— _Nm _ em _nmmdf .30.: $8 3839.353.
M21331 1 .831. 111.35%.” 1 1118.5..- - 1 1.mo.._.1w1&...a1=.a
..1§12..~1 .119... 1 3:11.182. 1 .8313...“ 1 m.o:m1..o,b.11_1§_11.,.7.1momp..1
11 1 1.3.- 1 1 _..1..1.1_1m 1 3.1.1.1111 _. €211.13ENE?
8.502050 20.39.05.

.2.<.0m..Ro.o+

.E<2...~81o+
.E<2.2..o+ .2<2....:.o+ 18275...-
1%??36: 1.1.23.5..11 1512?. 3%.? 10253.19
182.33.? 1022.11.13.41 . .82. .21... ..m.<2*w..m..+ 332.; 6.8:.
~2ma1..on...m1o.§- .Zmo1mo..m-orc- 10.25.32? .2mn1mon..m-o.a-
12:51.:31813 1:5:..3..S~.m+ 1:52.523? 15.3.2333:
1mo<_....~.o.+ 1.1.0581? 1m0< .80.? 11.0380?

.3. .- 0E1..- 331W .8...

. . . m
185.1 1 3.: 11111 18.1.12 , Ea 1 1 .1._o..o..1.w°.=1.m.1§<
1:2. .5319“. .ommém .32.? 2.2.93.
«8... 8m... .22. :2... mm 8.3.3
08... SN... 81m... .22. m...
5.32. ..mmmz. 5.1.2. £52.. 0.2.5; 2020.8

 

Adoom 3?... 88. 32.8 E Enos .0 ONE was 090 2.80.1.6 .8 $.32 22.02 2.25m .301. 9520 m1! 0:... 1.1

181

Appendix

Spatial Regression Model Results

 

182

 

modA 2:? m .m modnv 33> & .—

 

 

 

 

 

 

 

25.? 22 H v85- 33 $8.? N85 4 Asturias:
w: l _ - l - ml l l _ ll: m i l... 1m ; ; w 1 - ,
£32. 83$ :3. 33m , $2. 8% m2
_ _ _ _ _ _ N aawnméomaohm
«.33 3:2 , E26 «589:8
_ . _ . N V 35an
2%? ngoulaogo: wflw? bammcaufigo l: ‘ ,, Emma Ema—3530.? : w s. ,, l NE.
__, H , £me 338m
# 3.58.58? _2 21.8.29?
A
:25215? .23238?
_émhs? _ «3+ _&E<B.a-.o+
_zm<qom.~_o.o+ 229309 23: “_ZmodoQJbE- _ZmoumOmJbrm- “1533.83 _namté in:
_E§.~8.c+ _E<2 $8.? ” 16258.7 1627;: sf _Zmodonrsod- _Zmolmogocd- 8§E> H235
_mmofmwmf 1mg .38.? :muz<m.mmm._+ _moz<m.e:._+ W 162: :2. 362722-
_mzzm::3_.:o.o+ 1:553: .momfrmmxofimammv _mmxofimkwm..- :EEpEBfimi _Ezmbz?§._+
_Zmalmonrvorw _Zmoumgwbrh- _mmofaomf 18508.? “153.30.? 15359?
mm: .3: :83 _82 :22 W84“
23 - :l 33 l H36; , o2... W:-- ,1 ,, ._ 3., oi.
58.85.” 838me , Kmmmm- 39E. .Rntl 38~.,:l . 585.8.wa
QB. 83 cm; :3 mm: a: H N:
38285 .3% 30 38235 33an 30 38285 .2an 30 l .082
88,—. £550 (FUN D550 E5 ocoN

 

 

 

A38; =3 55: - 33QO 382 cassava 38m 2 29¢

183

nodA 33> & .N nodnv o=_m> k .—

 

 

 

 

 

 

 

 

n_o.o Nde , cmod- VmNd , coo: £26 , fiche: 9:802
N 1 l l. ll- wN- l l_.- l N l :_- l - ll}, l :
mwN._wN.~ cmo. wmo , $3.2 $5.2 ,. 32: 25.2 _. v.8
_ ~ _ _ W _ _ W ENwE-nomzohm
Emma 5"ng m 532. H$5980
N , _ , N , _wsmmm
Ema-0S» QamscoQ-amuzmvl 4,- EwaNMV-glsfllfiou melon ; :l - # EwElBNNSwmcmflloovzo llJ - l S:
, 5%: Exam
_ _2_<-_om.8o.o+ .2248: 5.? h 7
_E<§*NN0.0+ flax/‘2“, mmo.o+ m _2<2§_o.o+ _E<E*N_o.o+
_E_<-Om*:o.o+ _S:<-_Om ...hNo.o+ _mmm:<3.._ MN.o+ _nam-r<3*w_N.o+
ZmQImOnr. ormfi- Nam-ix: .32.?- Zm—leonr or:- ZmQImOmp CTN- ,
w m- _ , _ v- _ v- ,, moBmtm> 832m
_mzxmtmkrwmo._+ ~mm0<* Nov.m+ W _m_07:*wo\-.o- _L_Uz_*nwm._- v—amhaxaemwwi “Emu-(«3354+ ,
| | ,
_LMA<E*VE..N+ _LEEM-EIB rvmodlv V _QIMU’Nfim—LLNA- _mIv_U<-_m*mmm.- 7 _Zmo mOmLood- _ZmQ mOm Lood- ,
_mm—O<*VNQV+ _ZmQIn—Onr w-orLfi- i _mm0<*mmm.m+ _mmO<L-_N.o+ _mflm-EIBNVNQQ- flamim-EwS/thdf
2N6 N2 .o- dooN NmoN ,s Mmo.v ___.v 7
N N , _ _ _ ,
wood Ode 2:.— SNNJ , ON mN U_<
-_mwlma N- - llmoww-Nw-N N- \- 3% WWW:- ll-qawwm-o- - ENE-ll - flow-.l- : m a wooafizmfi
_hc. o 3%. o 0.2.. o Mbcd , 2K. o 236 NR
.252 Etm _Nummm mqo .25—2 Stu _Ncmmm mAO , _ucoE mAO w 382
, “ohm 358m W
“N.N; £550 <HUN DIEEU “:5 BEN

 

A389 533 :3 mamas-ZA- mzsmom _ouoE commbom _NN—«mam N- H 053-

184

 

nodA M53> & .N

nodnv o=3> & .—

 

 

 

 

 

 

 

_VN_.o- _SS 7 NES 2N3 7 N83 NNS.o- e05: €822
<2 l l -_ ENS» - . Sm - - l l awwnm . Sz _NES 32 38-8305
_
_chd , _rovd N.N..E-md- EocOQEoU 3_3mm
£385 93:50-50:03 l 1 £303 b33500 5me - ll - 7:333 9:»:qu 5010 l- 1st x522 32mg
, __2_<40m*$uo.o+ _E_<-_Om*vvo.o+ 7
Ham—pk?» _va.o+ _mmmh<3umwo.o+

.5732qu— .o+ _E<_2*Nv_.o+ 7 _NEIAON—ath- _AMIAOn—{NEN-

Nani-«57$ _._+ adm-raxat-VN: 7 mun—(Eggs- mmm<m1¢o.h-

_ _ w _ , 333:; 8305
1.3.7.30.” _m<muohv.N- _LIIm—EEBL-wm._+ _LIIMF_:3*mov._+7 _LMMH<3*NVn.m+ _mmmfi-<3«omc.m+,
fimzzm-EIBLZN+ _n—IIm—H_:3*NNN+ Nmmq<2*mw_.m- _mmq<2*v_m.m- 7_mmm<m*mom.w_- _mmn—<m.0Nw.eT

_m:mO<*wE 6+ _mzm0<*wma.o+ _Lm—O<*o.o_+ WED"... _ _ m.o_+ NEUZTcONélT NEUZTmmé+

0mNd- mmmd- vac-OM- v9:- mEd mm. .o-

_ _ N N N N
lowaE 03.x: l .206 _wloé l- - th l - :mmN - - l 07‘
20.8%-; OBS-N.N- - . ESQ: ONE-SN- . NNNNNT- :RNNT l 38:23:33
wmmd Nemd moNd VWNd ocmd wad Nk
m<0 mAO M<U w-HO M<U m-HO 3.82

“on; mam—BU <HUN fiasco NED 28M

 

was :3 _3 Ems/5 - £38m 382 commwewom Exam 2 2a:-

185

nodA 33> k .N modnv o=3> & .—

 

 

 

 

 

 

 

 

 

_NN_.o- _33 7 NES. 2.23 Not; NQNd 9:33.582
7 7
«2 l 1- - 15¢ng- - 11- 11.7 ma? l l l NNNNeN 71<\z SEN 1 - m2
_ _ _ _ :wwmm-comzohm
$83 $85 5.83 accomaoo
_ 7 _ 7 _ 33am
l5fl3>> Qawucow 1.8sz l . l l 1 EwEBlgame-vudoowa .1 l l7lwnw31m) b_=w:=o101:wM:O . 11 - SO
7 x75: :2st
_2n<40m.NNo.o+ _Emfiom-Naf 7
335,733.? 335238.? 140939? 148-Sq? 7
NémSB-Nom? _mmmk<3.Nov.o+ _Zmo-monrvbrvN- _me-Nofivbri
_mmozaga- _mmoztmod 71moz-oNN? 18:38.0- 7_2m2<2-8o.o- _En2<2.8o.o-
NZmo-mo?e-orN.T Ema-326-23- 102553. 1025-593. mew-"75333.: _nEmSB-SI 83233285
_mmozrnNN- ESE-Mai- 715659299 11205983- 7Nmmomafi..- _mmom-SNN- 7
12:553-806- ..Ezmtzazomt 7 1552-3; _- 1332-032- 7 _mm-mo?3m.o+ _mxumofiNNo?
15058.? Emmoffiv? 7 _ mag-RN? 153-5.9+ 7 REE-ammo- Emmi-2.
_Smi- _NSE. 7mm? .N_N.m 7_$_._ _aN._
18$: 1 11 - 11 awe-.21 - 7 $0.1m l - l - l2? 1- 11 .7 £1 1- 11 No- 1 1 7 11 1- -02
51.03.71 gnaw» _bFHNomT - 1 omc.N%.1_-- - 1 793mm- :93. . 1 32333
9.3 8N6 , :2 NNNd _de 83 Na
56 30 , wfiafimm So 77 5% So .252
88,—- mzmcoo «NEON goo E5 ocoN

 

3853 E52524 - flag 6%: Eamoaom Beam .1 037$

186

modA 33> & .N .3de o=3> & _

 

 

 

 

 

 

 

 

mood mod 7 ”no.0 N.N-ed 30.0- 36d 7 Gate: 95222
1N 111 l- _1 1 N 11 1_ 211 l117N1 N11 1 - 1 1
_vomémhfi _owvdmofi 7 «<2 _vovfiog 7 thmfi N26 “m3 cnwmm-numzohm
_aai N6 #5536 NQ*NN_.O EocOmEoU _m_3mm
Ems? b33300 561:0; 1 11 1 .7 “$7103 .03-$260 5on- - l Ewfi? gums—1501530 7. 1 93 £522 373%-
_Zmfllm0m*m-o:©.m- _ZmQILOmso-OTNS-7 _Em<-_Om*moo.o+ _2m<-_Om*moo.o+7
_n—<N_*Nwo.o+ —n_<--_ ..._;.o+ _2m<-_Om*v_o.o+ _Em<-_Om*w_o.o+ NmzlmOm*NCN.O- NLMILOQiNVNd-

QIMU<Am*VS.Y AIXU<JELON41 L_Uzrwo._+ L_UZ—*hao._+ 7 hawk<>¢*mo_._+ gmh<>a§ohf~+
_ _ 7 _ _ _ _ 333?; 860.5
_n—m4<2*oo¢.N+ Ema—<2 *NomN + _m:¥U<-_m*_w._- _mIMU<-m.MNw._- 7 _m2<nr7VN_.v+ _LE<L*MNN.V+
_LImO<*Nmo.v+ _m1m0< *cN_.v+ 7 _mm0<*wmv.m+ _mm0<§wam.w+ 7 _LIMO<*3.N+ _mImO<*wo_.N+

owo._- Nay..- 7 Sid- wm_.o- mvm._- 32.7
_ _ N N _ N
813 71 1 8w: . NNNN $3 11 7 3.1 121. 1 1 . 12%
193.30% 1 248%.“- . 3.32... 8587.71 1 018.3- Nmmfiw 1 8053:9117
Namd wnmd Nde memd mend ommd 7 NM
..otm 3_3mm mJO 7 m<w mAO 7 ma... 3_3mm mAO $602

68,—- msmcoo (CON balzoo 7 “ED ouoN

 

303% EUKLZ-H - 333m 3322 sommmohwum 33% m.” 23H-

 

187

modA o:_~> & .N wodnv o:_m> & .—

 

 

 

 

 

 

 

33 23 ES- NNN1o ”87o NNo1o coca
N _ N _ N N 3,5202
$113018 NSMME1 1 713211 11 92$ 7 SNe 83 magma
_ _ — N N 1:835
$83 $83. SM: ..o- 22258
_ _ N 723%
was? szudou 1:02.01 1 17. Emma}, 70:73:80 mound 11153103 Dszcco 50710 111 1 A26
7 7 5me 3:QO
7221598.? 3240989? 7 7224098371 3750929?
_E<2*NNo.o+ _E<2.N81o+ 7E<2Lo§+ 725238.?
72mo1mo%m-o:§- _Zmolmofimbra- 7.177%23‘7813 _mmmEBLE?
18252..- 1827;”... 72mo1moN.v-o_L.o1 381.726-27.76 _2m2<2*N8.o- _Ema<z.Noo.o1
7muz<m*$N.N+ 102512.? _N 7 02:3”. _1 _N 7 ozraovN- _nEmSBaONNL1 _AEESszm _ .N+ 87%wa
_szmtzakomi _nizmtzaaNmi 12257.25: 7,525.03: 7_ 28133390- _zgnmonrsod- .
155413371 _Nm4<2*wwm._+ 1595333.: 1:353:31 7_mNoerNm.N- 7_mNoerm.N- 7
_Nmofmowvr _mmofNonf 7_n_mo<.7mwm.o+ _mmufoowf _Emm:=3.m$.o+712772.53 as?
327 an; 82 82 N86 :3.
N N _ _ 7 _
vad 1 ES 1 .83 11 1 1 1N3; bN 1 1 he . o?
1NS.1_NN,N- camanfi 7.8.9? 1 11 SN? 77mNmN- .22- . moofieamfi
08.0 ”who 33 NE SE 83 Na
55 Exam So ”Em 30 7mm; Exam So .252
68 H 3550 <HUN $80 55 25M

 

38m mN 52: $2 59:: :23 Swen 8259 52:3 mmmmzq 1 338% $on :Emmuhwom Exam ml 2an

188

 

 

nodA u=_w> & .N moduv 27:; K .—

 

 

 

 

 

 

 

763- 77:77 7 $77.77 777.77 7 No.77 $7.77 7557
N 7 N _ 7 N N 3.77822
<1? 1 1 11 SNmzN 11 1717771712 11 11 1 773.37.” 11 117 mNaNN. N 7 mm? 32.5%;

_ 7 _ 7 N 7 N .5835

7933 $2.8 $32 78:27:78
_ _ .7 — Exam
71.71335 57159771700171.3120 1117771773013 QJEBU 5047717101 1 .711unw71u? bwgioo 78qu 11 1 1 . 1 C51

7 7777722 Exam
77237777218? 72m7~7<2.N8.o+
7 72.77.7238,? E<72...N8.o+ 7 7EmN<37NNvN+ 757N17N<BBSN+
757377.313? 7755:7523? 7EmN<3LN§+ .EmNSsaoo? 772mo1n70n7*v1o7..N.Y 7231779761777?
778252.741 .7827..cN.7+ 7 $018.7. crow- 281.797.. o7..N-7 “5952.741 $97.73.: 3
7 7 77 m- 7 m- 77 7 77725
7A702<ENNnN+ 7,7025%8? 7 7.782738? Nmoz7*oom.N+ 7 777mn7<%NNo.N- 7Nmm<h7.v.N- 8777277
7.777sz7:3*$7.$ 157353.73? 7.7sz7:3.:7.7+ $55333: 7777m7<2$3w1 77Nm7<2*oN¢.m-
77757733? 777mg .80.? 7n7m70<.7.N8.m+ 77353:? 777.7573... 7NN. N+ 77750580.?
NNNN- SEN- 827- 733. 83. 71.8.7.
7 7 N N 7 77
23.77 8171.2 1 .7 SWN :NNm N 71% 1 . on 0121
233.7- 778.30.? 7 299N171 ONN. emN 7- 1 7.mN m7- .3: .ON- 77877778777w3
827 77de N727 Em o 277. 77 877.77 N77
”75 30 ~75 m17o 7w77177~7amm 7 So 70872
78.7 F 27200 «POM 7057700 7:5 25M

 

3:3 H78.“ mN Swan. £73 3509. $309 owhfiv E9375- 33QO 70972 77083qu 793mm N17 03m H

189

wodA 077777> & .N

modnv o=7a> l .7

 

Cotov N M77822

 

 

 

 

 

Vm 7 .9 7m 7 .o 7 wood 37 .o 7 Rod 7 39¢.

.7 11 ‘ \ 7 11 1 1 N 1 111111 1 711111 111 1 N1 1 1 1 . N11? 1 1 .1 1

<2 7 mo 7 .2 <72 7 N86 W7 32 7 Sn .3. 7.87 cmwnaéomsem

77.17577 7.9.806 NFwoNd- 75:27:80 Eamm

Eye? b7=w77771ow 7737701 11111 . EMS? bywzcoo 7702710 \1 . Ems? b7=w77=1o10\ 77187170 7. Q15 x7bn211771w717m77m

77E7<170m773qod+ 7727<17Om7Lbod+ 7 7

757527.877: 7252.71.71 7Em71<3$ood+ 7mmm171<>3wmwd+

7357155397.? 7557275118.: 7mmlm0m*ao.7. 7m~7In7077$wm71

2817077.. 77:72- 251.70% ore- 7 77$..m86- n7<1,7*Nm7.m-

7 m- 7 m- 7 7 3777777773 87787;

7770321737.? N7707>7<nr77v7+ 777:m.7.77.73*$N1e+ 7N7=m7h777>fm3w0+ Ngmh<3*NwN.N+ 7Nm~7mk<3$ooa+

7n77177.7m.::3*v$.m+ 7N7:=m7.7177173*NwN.m+ 7751223377- Haze/733.2- 7Nm7uzrwvwé+ 7NN77UZ7..7.7.m+ 7

777m0<Lde+ 7mmo<*oov.w+ 7n777m70<_&NN.w+ 7mImO<LwNN+ 7.727777771777751 7.717 777:77m77.7773.mmw.m+7

$0.77- «731.7 7- NNoN- ”No.0- 7Nv6- 7 775:0-

7 7 N N 7 77

mmmfi17 1 $71.21 3777371 1 036 1 1 1 _NN 111 1 111$N 1 1 11 1U7<

mw7.N7NN.N- 03677177777- . nw7.7.vN.N- oomiWN- memdN71- 11 111; avwdfl- 11 1uoo7777wfi71wou7

75.77 53.0 ode 086 woNd VmNd NM

x<U m170 ~7<U m170 m<o m170 7 70782
708.71 3200 «PUN 79:60 7775 270M

 

087 N 5.1777337 777M787 77773 3783 77% 7772773 Emmw 217 1 $775.37 707572 =o7mw07w37

77~77a77m 77-7 2773

190

wodA o=7w> & .N moduv 2777~> & .7

 

 

 

 

 

 

 

m7 7.0- mN7d 7707.0 awed emod- mod Coho: 97772072
1711111 7 1111 711‘; 1 7 . N 1 1111111N 1 11 11 1
$2 70377: . <72 7333 7 $2 78277 32 5377177877277
7573077 7.3.927 77 NN*NwN.o 7:0:anoU7a77a77m
777203 797797780 782170 1 11 1171791075 b7=w7171717oo 75270 11‘ 1111 111117117\77w7\8151 x77=w71777717u177mo1=1m11 11 7 SC 777572 77777725

7 <170mikbd+ 7 «7170?. 7 ”0.0+ 7

72.32.7N7d+ 77M<2.._mv7.o+ 7 757m17.<311N.7+ 7N7M7m171<337m7+ 7
7.72m.7.<3*nv.7+ 7am171<3¢vofi7+ 7AMI¢ONTmem1 7mmlm0m*mmv.m-

mmUZ7*no.m+ mmUZ.7*ewN.m+ m7UZ7Lm70é+ N77027§om5v+
7 I 7 I 7 7 7 7 3777777773 87755
_ZmQ N70;*m-o7.._7.n- NZmD monrmbrwfl 77mUE<mL7vé1 7LUE<NTSYY 7am171<37777no+ 7.77am17.<>>*:.71.o+
777717717m717.77173*voo.c+ 7N777717m7171=73§oofio+77N77.7w7m717.777>>*mNh.N+ 7n7717717m7.7.7717\5*eoo.N+717277711737700: 7N72<117..me.77+
7N7m70<77>vfiw+ 7mm0<*wN.o7+ 77mm0<5m§w+ 7mm0<*oo7.m+ 77mm17<2§amNN1 777m..7<72*new.NN- 7
$7.77- 78.77- 7 cNo.m- 77%.? ommfi 77776
7 7 7 7 N N
778N171 1111 18N.N7 7 SQ: 1 133 . NNN NNN: 7. u7<
www.7w7NW 1 89277.77- 7171.71Nm.7777.N-1 23$? 1 11 1 17 72.27- 1 N71102:- . 1 82777187177317
ond ohNd 7 ammd M7727 7 hovd mamd NM
~7<U ma70 ~7<U m170 ~7<U W70 7 7302

88.71 25:00 <7.UN 7074750 77:: BEN

 

727777 787 N 7777757 .7773 3877 77% owe: E7327 - 37337 707572 87327737 733% N17 2773.

191

Appendix

Models residual maps

192

County Level Models

193

Figure J-l The distribution of residuals of Ordinary Least Square Model
for number of all boats per 1000 households at county level

 

   
 
    
 

Dependent Variable: LNABH
% of Residuals/Observed Value

- <=-1o.0% (1)
[:1 3.9%-5.0% (8)
[:1 4.9%-0.1% (26)
[:1 0.0%- 4.9% (26)
5.0%- 9.9% (6)
- >= 10.0% (0)

 

 

 

Figure J-2 The distribution of residuals of Spatial Regression Model
for number of all boats per 1000 households at county level

 

Dependent Variable: LNABH
% of Residuals/Observed Value
<= - 10.0% (O)

-9.9% - -5.0% (8)
-4.9% - -0.1% (25)
0.0% - 4.9% (28)
5.0% - 9.9% (6)
>= 10.0% (0)

IL'IUDDI

 

 

 

194

Figure J-3 The distribution of residuals of Ordinary Least Square Model
for number of all power boats per 1000 households at county level

 

   
 
   
 

Dependent Variable: LNAPBH
% of Residuals/Observed Value

- <=-1o.0% (4)
E -9.9%-—5.0% (6)
[j 4.9%—0.1% (23)
[:1 0.0%- 4.9% (20)
= 5.0%- 9.9% (12)
- >= 10.0% (2)

 

 

 

Figure .14 The distribution of residuals of Spatial Regression Model
for number of all power boats per 1000 households at county level

 

Dependent Variable: LNAPBH
% of Residuals/Observed Value

- <=-1o.0% (3)
-9.9%--5.0% (9)
{:1 4.9%--0.1% (22)
E] 0.0%- 4.9% (21)
u 5.0%- 9.9% (10)

- >= 10.0% (2)

 

 

 

195

Figure J-5 The distribution of residuals of Ordinary Least Square Model
for number of all sail boats per 1000 households at county level

 

   
    

Dependent Variable: LNASBH
% of Residuals/Observed Value

- <=-10.0% (24)
[:3 -9.9%--5.0% (0)
:I 4.9%--0.1% (2)
[3 0.0%- 4.9% (2)
E 5.0%- 9.9% (1)
- >= 10.0% (38) ’1‘;

D7-

 

 

 

Figure J-6 The distribution of residuals of Spatial Regression Model
for number of all sail boats per 1000 households at county level

 

    
 

Dependent Variable: LNASBH
% of Residuals/Observed Value

- <=-10.0% (25)
(:1 -9.9%--5.0% (1)
:1 4.9%-0.1% (2)
E] 0.0%- 4.9% (2)
5.0%- 9.9% (1)
-

>= 10.0% (36)

 

 

 

196

Figure J-7 The distribution of residuals of Ordinary Least Square Model
for number of canoes per 1000 households at county level

 

  
 
   
 

Dependent Variable: LNCANOEH
% of Residuals/Observed Value

- <=-1o.0% (8)
[:1 9.9%-5.0% (4)
:1 -4.9%--0.l% (7)
(:1 0.0%- 4.9% (23)
D 5.0%- 9.9% (15)
- >= 10.0% (10)

 

 

 

Figure J- 8 The distribution of residuals of Spatial Regression Model
for number of canoes per 1000 households at count—y level

 

Dependent Variable: LNCANOEH
% of Residuals/Observed Value

- <=-1o.0% (6)
-9.9%--5.0% (3)
E 4.9%-0.1% (5)
[:1 0.0%- 4.9% (26)
5.0%- 9.9% (15)
- >= 10.0% (12)

 

 

 

197

Figure J-9 The distribution of residuals of Ordinary Least Square Model
for number of PWCs per 1000 households at county level

 

   
 
   
 

Dependent Variable: LNPWCH
% of Residuals/Observed Value
<= - 10.0% (12)

-9.9%--5.0% (12)
4.9%-0.1% (10)
0.0%- 4.9% (14)
5.0%- 9.9% (9)
>= 10.0% (10)

IMUDI

 

 

 

Figure J-lO The distribution of residuals of Spatial Regression Model
for number of PWCs per 1000 households at county level

 

Dependent Variable: LNPWCH
% of Residuals/Observed Value

- <=-1o.0% (12)
C3 -9.9%--5.0% (10)
[:1 4.9%-0.1% (12)
[___—J 0.0%- 4.9% (13)
m 5.0%- 9.9% (10)
- >=1o.0% (10)

 

 

 

198

Figure J-ll The distribution of residuals of Ordinary Least Square Model
for number of small power boats (<23’) per 1000 households at county level

 

   
 
   
 

Dependent Variable: LNSPBH
% of Residuals/Observed Value

- <= - 10.0% (3)
[:3

-9.9%--5.0% (13)
[:1 4.9%-0.1% (16)
[:1 0.0%- 4.9% (24)
I:

5.0%- 9.9% (11)

- >= 10.0% (0)

 

 

 

Figure J-l2 The distribution of residuals of Spatial Regression Model for
number of small power boats (<23’) per 1000 households at county level

 

Dependent Variable: LNSPBH
% of Residuals/Observed Value
<= - 10.0% (3)

-9.9% - -5.0% (13)
4.9% - -0.1% (18)
0.0% - 4.9% (23)
5.0% - 9.9% (10)
>= 10.0% (0)

IHUDUI

 

 

 

199

Figure J-13 The distribution of residuals of Ordinary Least Square Model
for number of large power boats (23’+) per 1000 households at county level

 

   
 
   
 

Dependent Variable: LNLPBH
% of Residuals/Observed Value

- <=-l0.0°/o (14)
-9.9°/o--5.0°/o (7)
E 4.9%-0.1% (12)
I: 0.0%- 4.9% (14)
= 5.0%- 9.9% (12)
- >= 10.0% (3)

 

 

 

Figure J-l4 The distribution of residuals of Spatial Regression Model for
number of large power boats (23’+) per 1000 households at county level

 

Dependent Variable: LNLPBH
% of Residuals/Observed Value

- <=-1o.0% (12)
a -9.9%--5.0% (8)
[:1 4.9%-0.1% (12)
l:l 0.0%- 4.9% (16)
5.0%- 9.9% (11)

- >= 10.0% (8)

 

 

 

200

Figure .l-lS The distribution of residuals of Ordinary Least Square Model
for number of small sail boats (<23’) per 1000 households at county level

 

   
 
   
 

Dependent Variable: LNSSBH
% of Residuals/Observed Value

- <=-10.0% (16)
m -9.9%--5.0% (7)
[:J -4.9%--0.l% (10)
C] 0.0%- 4.9% (7)
E 5.0%- 9.9% (11)
- >= 10.0% (16)

 

 

 

Figure J-l6 The distribution of residuals of Spatial Regression Model for
number of small sail boats (<23’) per 1000 households at county level

 

Dependent Variable: LNSSBH
% of Residuals/Observed Value

- <=-1o.0% (15)
E -9.9%--5.0% (9)
1:] 4.9%-41.1% (7)
:3 0.0%- 4.9% (10)
a 5.0%- 9.9% (11)

- >= 10.0% (15)

 

 

 

201

Figure J-l 7 The distribution of residuals of Ordinary Least Square Model
for number of large sail boats (23’+) per 1000 households at county level

 

   
 
   
 

Dependent Variable: LNLSBH
% of Residuals/Observed Value

- <=-1o.0% (9)
g: -9.9%--5.0% (7)
1:] 4.9%--010/6 (6)
l: 0.0%- 4.9% (16)
E 5.00/..- 9.9% (15)

- >= 10.0% (14)

 

 

 

Figure J-l8 The distribution of residuals of Spatial Regression Model for
number of large sail boats (23’+) per 1000 households at county level

 

Dependent Variable: LNLSBH
% of Residuals/Observed Value

- <=-10.0% (11)
E: -9.9%--5.0% (7)
1:1 4.9%--o.1% (8)
[:1 0.0%- 4.9% (12)
D 5.0%- 9.9% (13)

- >=1o.0% (16)

 

 

 

202

 

ZCTA Level Models

 

203

Figure J-l9 The distribution of residuals of Ordinary Least Square Model
for number of all boats per 1000 households at ZCTA level

 

   
 
    
 

Dependent Variable: LNABH
% of Residuals/Observed Value
<= -10.0% (121)

9.9%-50% (133)
4.9%-0.1% (191)
0.0%- 4.9% (219)
5.0%- 9.9% (138)
>= 10.0% (88)

IflUDlI

 

 

 

Figure J—20 The distribution of residuals of Spatial Regression Model
for number of all boats per 1000 households at ZCTA level

 

Dependent Variable: LNABH
% of Residuals/Observed Value
<=-10.0% (110)

-9.9% - -5.0% (139)
-4.9% - -0.1% (210)
0.0% - 4.9% (243)
5.0% - 9.9% (123)
>= 10.0% (65)

IEUDDI

 

 

 

204

Figure J-Zl The distribution of residuals of Ordinary Least Square Model
for number of all power boats per 1000 households at ZCTA level

 

   
 
   
 

Dependent Variable: LNAPBH
% of Residuals/Observed Value

- <=-10.0% (147)
5:] -9.9%--5.0% (125)
1:] 4.9%--o.1% (170)
E] 0.0%- 4.9% (206)
a 5.0%- 9.9% (130)
- >=10.0% (112)

 

 

 

Figure J-22 The distribution of residuals of Spatial Regression Model
for number of all power boats per 1000 households at ZCT A level

 

Dependent Variable: LNAPBH
% of Residuals/Observed Value
- <= - 10.0% (130)
-9.9% - -5.0% (136)
[:I -4.9% - 0.1% (187)
0.0% - 4.9% (223)
5.0% - 9.9% (133)

- >=1o.0% (81)

 

 

 

205

Figure J-23 The distribution of residuals of Ordinary Least Square Model
for number of all sail boats per 1000 households at ZCTA level

 

    
 
   
 

Dependent Variable: LNASBH
% of Residuals/Observed Value

- <=-1o.0% (274)
D -9.9%--5.0% (11)
1:] 4.9%-0.1% (8)

[:1 0.0%- 4.9% (10)
D 5.0%- 9.9% (18)
- >= 10.0% (569)

 

 

 

Figure J-24 The distribution of residuals of Spatial Regression Model
for number of all sail boats per 1000 households at ZCTA level

 

Dependent Variable: LNASBH
% of Residuals/Observed Value

- <=-1o.0% (309)
[:3 -9.9%--5.0% (12)
1:] 4.9%-0.1% (8)
El 0.0%- 4.9% (23)
5.0%- 9.9% (15)

D
- >= 10.0% (523)

 

 

 

206

Figure J-25 The distribution of residuals of Ordinary Least Square Model
for number of canoes per 1000 households at ZCTA level

 

 

  
 

Dependent Variable: LNCANOEH
% of Residuals/Observed Value

-<=-100% (158)
-99%--50% (41)
49%- -10/1. (80)
0%- 49% (120)
50%- 99% (203)
>=100% (288)

IDDUI

 

 

Figure J-26 The distribution of residuals of Spatial Regression Model
for number of canoes per 1000 households at ZCTA level

 

 

  
 

Dependent Variable: LNCANOEH
% of Residuals/Observed Value
- <= - 100% (159)
E -99% - -50% (58)
l: 49%- -l% (71)
D 0%- 49% (146)
E 50%- 99% (187)
- >= 100% (269)

 

 

Figure 1-27 The distribution of residuals of Ordinary Least Square Model
for number of PWCs per 1000 households at ZCTA level

 

   
 
    
 

Dependent Variable: LNPWCH
% of Residuals/Observed Value
<= - 10.0% (241)

-

D -9.9%--5.0% (76)
C] -4.9°/o--0.l% (101)
D 0.0%- 4.9% (97)
__1- 5.0°/o- 9.9% (99)
- >= 10.0% (276)

 

 

 

Figure J-28 The distribution of residuals of Spatial Regression Model
for number of PWCs per 1000 households at ZCTA level

 

Dependent Variable: LNPWCH
% of Residuals/Observed Value
- <=-10.0% (239)
E: -9.9%- -5.0% (75)

[:1 4.9%-01% (106)
[:1 0.0%- 4.9% (96)

D 5.0%- 9.9% (113)
- >= 10.0% (261)

 

 

 

208

Figure J-29 The distribution of residuals of Ordinary Least Square Model
for number of small power boats (<23’) per 1000 households at ZCTA level

 

   
 
   
 

Dependent Variable: LNSPBH
% of Residuals/Observed Value

- <=-10.0% (157)
E] 9.9%-50% (103)
1:] -4.9%--0.l% (171)
(:1 0.0%- 4.9% (201)
= 5.0%- 9.9% (151)
- >=10.0% (107) /

 

 

 

Figure J-30 The distribution of residuals of Spatial Regression Model for
number of small power boats (<23’) per 1000 households at ZCTA level

 

Dependent Variable: LNSPBH
% of Residuals/Observed Value
- <=-10.0% (140)
[:3 -9.9%--5.0% (114)
[:1 -4.9% ---0.1% (206)
[:1 0.0%- 4.9% (206)
5.0%- 9.9% (139)
- >= 10.0% (85)

 

 

 

209

Figure J-3l The distribution of residuals of Ordinary Least Square Model
for number of large power boats (23’+) per 1000 households at ZCTA level

 

   
 
   
 

Dependent Variable: LNLPBH
% of Residuals/Observed Value

- <=-1o.0% (241)
1:] 99%-50% (95)
[:1 4.9%--0104. (97)
[:1 0.0%- 4.9% (103)
-; 5.0%- 9.9% (116)
- >= 10.0% (238)

 

 

 

Figure J-32 The distribution of residuals of Spatial Regression Model for
number of large power boats (23’+) per 1000 households at ZCTA level

 

Dependent Variable: LNLPBH
% of Residuals/Observed Value

- <=-1o.0% (225)
9.9%-5.0% (109)
:1 4.9%-01% (112)
L__: 0.0%- 4.9% (107)
5.0%- 9.9% (111)
- >= 10.0% (226)

 

 

 

210

Figure J-33 The distribution of residuals of Ordinary Least Square Model
for number of small sail boats (<23’) per 1000 households at ZCTA level

 

 

   
 

Dependent Variable: LNSSBH
% of Residuals/Observed Value

- <=-100% (250)
[:1 -99%--50% (74)
[:1 49%- -1% (74)
[:1 0%- 49% (120)
a 50%- 99% (169)
- >= 100% (203)

 

 

Figure J-34 The distribution of residuals of Spatial Regression Model for
number of small sail boats (<23’) per 1000 households at ZCTA level

 

 

   
 

Dependent Variable: LNSSBH
% of Residuals/Observed Value
- <= - 10.0% (237)
E: -9.9% - -5.0% (74)
[:l -4.9%--0.l% (93)
C] 0.0% - 4.9% (142)
5.0%- 9.9% (156)

- >= 10.0% (188)

 

 

 

211

Figure .l-3S The distribution of residuals of Ordinary Least Square Model
for number of large sail boats (23’+) per 1000 households at ZCTA level

 

 

   
 

Dependent Variable: LNLSBH
% of Residuals/Observed Value

- <=-100% (147)
[:3 -99%--50% (52)
1:] 49%- -1% (91)
[:1 0%- 49% (126)
Q 50%- 99% (187)
- >=100% (287)

 

 

Figure J-36 The distribution of residuals of Spatial Regression Model for
number of large sail boats (23’+) per 1000 households at ZCTA level

 

 

   
 

Dependent Variable: LNLSBH
% of Residuals/Observed Value

- <= - 10.0% (153)
-9.9% - -5.0% (59)
-4.9% - -0. 1% (97)
0.0% - 4.9% (137)
5.0% - 9.9% (192)
>= 10.0% (252)

IIDUD

 

 

212

 

Census Tract Level Residuals

213

Figure J-37 The distribution of residuals of Ordinary Least Square Model
for number of all boats per 1000 households at census tract level

 

   
 
    
 

Dependent Variable: LNABH
% of Residuals/Observed Value

- <=-10.0% (691)
[:3 9.9%-50% (357)
:3 -4.9%--0.1% (444)
E] 0.0%- 4.9% (574)
D 5.0%- 9.9% (484)
- >= 10.0% (590)

 

 

 

Figure J-38 The distribution of residuals of Spatial Regression Model for
number of all boats er 1000 households at census tract level

 

Dependent Variable: LNABH
% of Residuals/Observed Value

- <=- 10.0% (571)

E3 -9.9%--5.0% (367)
[:1 4.9%-01% (590)
[3 0.0%- 4.9% (670)
E 5.0%- 9.9% (492)
- >= 10.0% (450)

 

 

 

214

Figure J-39 The distribution of residuals of Ordinary Least Square Model
for number of all power boats per 1000 households at census tract level

 

   
 
   
 

Dependent Variable: LNAPBH
% of Residuals/Observed Value
- <=-10.0% (735)
[3 -9.9% - -5.0% (303)
E] -4.9%--0.1% (412)
E 0.0%- 4.9% (517)
D 5.0%- 9.9% (500)
- >= 10.0% (673)

 

 

 

Figure J-40 The distribution of residuals of Spatial Regression Model for
number of all power boats per 1000 households at census tract level

 

Dependent Variable: LNAPBH
% of Residuals/Observed Value
- <=- 10.0% (617)
-9.9%- -5.0% (354)
[:1 4.9% - -0.1% (506)
m 0.0%- 4.9% (629)
a 5.0% - 9.9% (513)
- >= 10.0% (521)

 

 

 

215

Figure J-41 The distribution of residuals of Ordinary Least Square Model
for number of all sail boats per 1000 households at census tract level

 

   
  
   
 

Dependent Variable: LNASBH
% of Residuals/Observed Value
- <=-10.0% (777)
:1 -9.9%- -5.0% (16)
[:l -4.9%--O.l% (15)
C] 0.0%- 4.9% (15)
= 5.0%- 9.9% (26)
- >= 10.0% (2291)

 

 

 

 

Figure J-42 The distribution of residuals of Spatial Regression Model
for number of all sail boats per 1000 households at census tract level

 

Dependent Variable: LNASBH
% of Residuals/Observed Value

- <=-10.0% (856)
E3 -9.9%--5.0% (22)
|:] -4.9% - -0.1% (22)
[3 0.0%- 4.9% (39)
n 5.0%- 9.9% (44)
- >= 10.0% (2157)

 

 

 

216

Figure J-43 The distribution of residuals of Ordinary Least Square Model
for number of canoes per 1000 households at census tract level

 

  
  
  
 

Dependent Variable: LNCANOEH
% of Residuals/Observed Value
<= - 100% (738)

-99% - -50% (65)

[j 49% - -o.1% (83)
1:]

0.0%- 4.9% (486)
5.0%- 9.9% (619)
- >= 10.0% (1148)

 

 

 

 

Figure 144 The distribution of residuals of Spatial Regression Model
for number of canoes per 1000 households at census tract level

 

Dependent Variable: LNCANOEH
% of Residuals/Observed Value
- <=-10.0% (738)
E: -9.9%--5.0% (52)
[:1 4.9%-01% (116)
E] 0.0%- 4.9% (625)
5.0%- 9.9% (604)
- >= 10.0% (1004)

 

 

 

217

Figure J-45 The distribution of residuals of Ordinary Least Square Model
for number of PWCs per 1000 households at census tract level

 

- <= - 10.0%

E -9.9% - -5.0%
E -4.9% - —0.1%
E 0.0%- 4.9%
5.0%- 9.9%

- >= 10.0%

 

Dependent Variable: LNPWCH
% of Residuals/Observed Value

   
 

(937)
(180)
(214)
(210)
(233)
(1366)

 

Figure J-46 The distribution of residuals of Spatial Regression Model for
number of PWCs per 1000 households at census tract level

 

<= - 10.0%
-9.9% - -5.0%
-4.9% - -0. 1%
0.0% - 4.9%
5.0% - 9.9%
>= 10.0%

IEIDDDI

 

Dependent Variable: LNPWCH
% of Residuals/Observed Value

   
 

(920)
(182)
(212)
(239)
(252)
(1335)

 

218

 

 

Figure J-47 The distribution of residuals of Ordinary Least Square Model
for number of small power boats (<23’) per 1000 households at census tract level

 

   
 
   
 

Dependent Variable: LNSPBH
% of Residuals/Observed Value
- <=- 10.0% (739)
D -9.9% - -5.0% (285)
1:] -4.9% - -0.1% (422)
0.0%- 4.9% (505)
5.0%- 9.9% (475)
- >= 10.0% (714)

 

 

 

Figure J-48 The distribution of residuals of Spatial Regression Model for
number of small power boats (<23’) per 1000 households at census tract level

 

Dependent Variable: LNSPBH
% of Residuals/Observed Value

- <=-10.0°/o (641)
B -9.9%--5.0°/o (326)
E -4.9%--0.l% (521)
E 0.0%- 4.9% (580)
a 5.0%- 9.9% (507)
- >= 10.0% (565)

 

 

 

219

Figure .149 The distribution of residuals of Ordinary Least Square Model for Number
of large power boats (23’+) per 1000 households at census tract level

 

   
 
   
 

Dependent Variable: LNLPBH
% of Residuals/Observed Value
- <= - 10.0% (869)
E -9.9%--5.0% (211)
[:1 -4.9% --0.1% (254)
[:1 0.0% - 4.9% (237)
= 5.0% - 9.9% (239)

- >= 10.0% (1330)

 

 

 

Figure J-50 The distribution of residuals of Spatial Regression Model for
number of large power boats (23’+) per 1000 households at census tract level

 

Dependent Variable: LNLPBH
% of Residuals/Observed Value
- <= - 10.0% (820)
-9.9% - -5.0% (250)
-4.9% - -0.1% (272)
0.0% - 4.9% (299)
5.0% - 9.9% (249)
>= 10.0% ( 1250)

”[1111]

 

 

 

220

Figure J-51 The distribution of residuals of Ordinary Least Square Model for
number of small sail boats (<23’) per 1000 households at census tract level

 

   
 
   
 

Dependent Variable2LNSSBH
% of Residuals/Observed Value
- <=-10.0% (985)
E: -9.9%- -5.0% (26)

[:1 -4.9%--0.l% (46)

[2 0.0%- 4.9% (833)
a 5.0%- 9.9% (613)
- >= 10.0% (637)

 

 

 

Figure J-52 The distribution of residuals of Spatial Regression Model for
number of small sail boats (<23’) per 1000 households at census tract level

 

Dependent Variable: LNSSBH
% of Residuals/Observed Value
<= - 10.0% (935)

9.9%-50% (31)
-4.9%--o.1% (91)
0.0%- 4.9% (968)
5.0%- 9.9% (494)
>= 10.0% (621)

IMUDDI

 

 

 

221

Figure J-53 The distribution of residuals of Ordinary Least Square Model
for number of large sail boats (23’+) per 1000 households at census tract level

 

   
 
   
 

Dependent Variable: LNLSBH
% of Residuals/Observed Value

- <=-1o.0% (389)
[j -9.9%--5.0% (51)
[:1 -4.9%--0.1% (80)
[:1 0.0%- 4.9% (572)
E 5.0%- 9.9% (756)
- >= 10.0% (1292)

 

 

 

Figure J-54 The distribution of residuals of Spatial Regression Model for
number of large sail boats (23’+) per 1000 households at census tract level

 

Dependent Variable: LNLSBH
% of Residuals/Observed Value

- <=-1o.0% (392)
9.9%-50% (58)

[:1 -4.9%--o.1% (105)
[:1 0.0%- 4.9% (694)
_- 5.0%- 9.9% (734)
- >= 10.0% (1157)

 

 

 

 

222

BIBLIOGRAPHY

223

Bibliography

Adamowicz, W., Fletcher, J ., & Graham-Tomasi, T. (1989). Functional form and the
statistical properties of welfare measures. American Journal of A gricultural
Economics, 71, 414-420.

Ali, M., Park, J ., Thiem, V., Canh, D., Emch, M., & Clemens, J. (2005). Neighborhood
size and local geographic variation of health and social determinants. International
Journal of Health Geographics, 4:12. Retrieved June 10, 2009,
http://www.ncbi.nlm.nih.gov/pmc/articles/PMC1 1 56930/?tool=pubmed.

American Red Cross. (1991). American Red Cross national boating survey: a study of
recreational boats, boaters, and accidents in the United States. Washington, DC:
American Red Cross.

Anselin, L. (1988). Spatial econometrics, methods and models. Boston: Kluwer
Academic Publishers.

Anselin, L., & Griffith, D. (1988). Do spatial effects really matter in regression analysis?
Papers, Regional Science Association, 65, 11-34.

Anselin, L., & Getis, A. (1992). Spatial statistical analysis and Geographic Information
Systems. The Annals of Regional Science, Springer, 26(1), 19-33.

Anselin, L. (2000). Computing environments for spatial data analysis. Journal of
Geographical Systems, 2(3), 201-221.

Anselin, L., Raymond, J ., & Florax, S. (editors) (2004)Advances in spatial econometrics:
methodology, tools and applications. Berlin : Springer-Verlag.

Anselin, L., Syabri, 1., & Kho, Y. (2006). GeoDa: An introduction to spatial data
analysis. Geographical Analysis, 38 (1), 5-22.

Arlinghaus, R., Bork M., & F ladung, E. (2008). Understanding the heterogeneity of
recreational anglers across an urban-rural gradient in a metropolitan area (Berlin,

Germany), with implications for fisheries management. Fisheries research, 92(1),
53-62.

Arlinghaus, R., & Cooke, S. (2008). Recreational fisheries: socio-economic importance,
conservation issues and management challenges. In B. Adams (Ed.), Recreational
Hunting, Conservation and Rural Livelihoods: Science and Practice. Blackwell
Science, Oxford.

224

Ashton, P. (1971). Recreational boating carrying capacity: A preliminary study of three
heavily used lakes in Southeastern Michigan (PhD Dissertation, Michigan State
University, Pages 225).

Bateman, I., Brainard, J ., Garrod, D., & Lovett, A. (1999). The impact of measurement
assumptions upon individual travel cost estimates of consumer surplus: a GIS
analysis. Regional Environmental Change, 1(1), 24-30.

Baud-Bovy, M., & Lawson, F. (1998). Tourism and recreation handbook of planning and
design: handbook of planning and design. Architectural Press.

Bell, F. (1990). Economic impact of bluebelting incentives on the marina industry in
Florida. Florida Sea Grant Report. SGR-99, 1990.

Bell, F. (1995). Projection of boating demand by length (size) of boat and for marina,
boat ramp, and private dock access points at 5- Year intervals, 1991 - 2010, with a
consideration of future facility needs and blue-belting incentives for Charlotte,
County, Florida. Tallahassee, Florida: Special Report, Department of Economics,
Florida State University.

I Bigman, D., & Deichmann, U. (2000). Geographic targeting for poverty alleviation. In
D. Bigman & H. F ofack, eds. Geographic targeting for poverty alleviation.
Methodology and application. World Bank Regional and Sectoral Studies.

Bivand, R., Pebesma, E., & Gomez-Rubio, V. (2008). Applied spatial data analysis with
R. Springer, Heidelberg, Germany.

Bockstael, N., McConnell, K., & Strand. I. (1989). Measuring the benefits of
improvements in water quality: The Chesapeake Bay. Marine Resource Economics,
6(1), 1-18.

Booth, 8., Sallis, J ., Ritenbaugh, C., Hill, J ., Birch, L., & Frank, L. (2001). Environmental
and societal factors affect food choice and physical activity: Rationale, influences,
and leverage points. Nutrition Review, 59(3 Pt 2), 21—39.

Bowes, M. D., & Loomis, J. B. (1980). A note on the use of travel cost models with
unequal zonal populations. Land Economics, 56(4), 465-470.

Bowker, J. M., English, D. B. K., & Cordell, H. K. (1999). Projections of outdoor
recreation participation to 2050. In Cordell, H.K., Betz, C.J., Bowker, C.J. (Eds.),
Outdoor recreation in American life: a national assessment of demand and supply
trends (pp. 320-350). Champaign, IL: Sagamore Publishing.

Brunsdon, C., Fotheringham, A., S., & Charlton, M. E. (1996). Geographically weighted
regression: A method for exploring spatial nonstationarity. Geographical Analysis,
28, 281—298.

225

Burrows, R., Ellison, N. & Woods, B. (2005). Neighbourhoods on the Net: The nature
and impact of internet-based neighbourhood information systems. Bristol, UK: The
Policy Press.

Burt, O. R., & Brewer, D. (1971). Estimation of net social benefits from outdoor
recreation. Econometrica, 39 (5), 813-840.

Calhoun, V. D., Liu, J ., & Adali, T., (2009). A review of group ICA for fMRI data and
ICA for joint inference of imaging, genetic, and ERP data. NeuroImage, 45(1), 163-
172.

Carver, S. (1996). Mapping the wilderness continuum using raster GIS, In: Raster
Imagery in Geographical Information Systems, High Mountain Press, Albuquerque.

Carver, J ., Sargent, W., Sidman, C., Swett, R., & Wiggin, J. (2007). Developing a
comprehensive planning framework for Florida’s waterfronts and waterways: the
public access element. Proceedings of Coastal Zone, 07. Portland, Oregon.

Casetti. E. (1972). Generating models by the expansion methods: applications to
geographical research, Geographical Analysis, 4, 81-91.

Casetti, E. (1986). The dual expansion methods: an application for evaluating the effects
of population growth on development. IEEE Transactions on Systems, Man and
Cybernetics SMC, I 6, 29-39.

Casetti, E. (1997). The expansion method, mathematical modeling, and spatial
econometrics. International Regional Science Review, 20(1-2), 9-33.

Cayo, M. R., & Talbot, T. O. (2003). Positional error in automated geocoding of
residential addresses. International Journal of Health Geographics, 2(10).
Retrieved January 26, 2009, http://www.ij-healthgeographics.corn/content/2/ l/ 10.

Clawson, M. (1959). Methods of measuring the demand for and value of outdoor
recreation. Washington, DC: Resources for the Future.

Cliff, A.D., Haggett, P., Ord, K, Bassett, K., & Davies, R. (1975). Elements of spatial
structure CUP. Cambridge.

Cockings, S., & Martin, D. (2005). Zone design for environment and health studies using
pre-aggregated data. Social Science and Medicine, 60, 2729—2742.

Confer, J ., & Graefe, A. (1992). A Geographic Information Systems Approach to
Analysis of Spatial Relationships between Recreational Boating Users and
Perceived Impacts. Proceedings of the I 992 Northeastern Recreation Research

226

Symposium (pp. 103-110), Bolton Landing, NY: USDA Forest Service General
Technical Report NE-23 2.

Cordell, H. K. (1990). An analysis of the outdoor recreation and wilderness situation in
the United States, 1989-2040: a technical document supporting the 1989 RPA
assessment. Gen. Tech. Rep. RM-189. Fort Collins, CO: US. Department of

Agriculture, Forest Service, Rocky Mountain Forest and Range Experiment Station.
112 p.

Cordell, H. K., & Overdevest, C. (2001). Footprints on the land: an assessment of
demographic trends and the future of natural resources in the United States (pp.
449). Champaign, IL: Sagamore Publishing.

Cressie, N. (1996). Change of support and the modifiable areal unit problem.
Geographical Systems, 3 (2—3 ), 159-180.

Curry, M. (1998). Digital Places: Living with Geographic Information Technologies.
London: Routledge.

Debbage, K. (1991). Spatial behavior in a Bahamian Resort. Annals of Tourism Research,
I 8, 251-268.

Demetriades, E. (2002). Descriptive statistics (Critiki Eds.) Athens.

Ditton, R., Chang, C. F., & Hollin, D. (2003). Boating infrastructure needs in Texas.
College Station, TX: the Texas Parks and Wildlife Department, with additional
funding support from the Texas Agricultural Experiment Station, Texas A&M
University.

Duda, D. M., Wise, V. L., Testerman, W., Bissell, S. J ., & Lanier, A. (2000). Factors
related to recreational boating participation in the United States: A Review of the
Literature. Harrisonburg, VA: Responsive Management.

Dwyer, J. F., & Hutchinson, R. (1990). Outdoor recreation participation and preferences
by black and white Chicago households. In J. Vining. Social science and natural
resource management (pp. 49-68). Boulder, Colorado: Westview Press.

English, D. B. K., Cordell, H. K., & Bowker, J. M. (1999). Implications of this
assessment. In Cordell, H.K., McKinney, S.M. (Eds.), Outdoor recreation in
American life: a national assessment of demand and supply trends (pp. 433-440).
Champaign, IL: Sagamore Publishing.

ESRI. (2008a).Address Coder. (ESRI Inc, Redlands, CA)

ESRI. (2008b).Business Analyst for Desktop. (ESRI Inc, Redlands, CA)

227

ESRI. (2008c).Community Tapestry Data. Retrieved May 20,2009,
http://www.esri.com/data/esri_data/tapestry.html.

ESRI. (2009). ARCGIS 9.3 for Desktop. (ESRI Inc, Redlands, CA)

Fedler, A. J ., Ditton, R. B., & Duda, M. D. (1998). Factors influencing recreational
fishing and boating participation. Alexandria, VA.: Sport Fishing and Boating
Partnership Council.

Felder, A. J. (2000). Participation in Boating and Fishing: A Literature Review.

F otheringham, A., & Wong, D. (1991). The modifiable areal unit problem in multivariate
statistical analysis. Environment and Planning, 23, 1025-1044.

F otheringham, A. S., Brundson, C., & Charlton, M. (2002). Geographically weighted
regression - the analysis of spatially varying relationships. Chichester: Wiley.

Freeman, A. (2003). The measurement of environmental and resource values: theory and
methods. Baltimore, MD: Resources for the Future Press.

F esenmaier, D. R. (1980). Modeling variation in destination patronage for outdoor
recreation activity. Journal of Travel Research, 24(2), 17-23.

FWC. (2007). Florida Boat Registration Data.

FWC. (2009). The Florida boating access facilities inventory and economic study RFP
NO. FWC 04/05-23.

Garber-Yonts, B. E. (2005). Conceptualizing and measuring demand for recreation on
national forests: a review and synthesis. United States Department of Agriculture
Forest Service, Pacific Northwest Research Station General Technical Report
PNW-GTR-645.

Gartner, W. C., & Lime, D. W. (Eds. 2000). Trends in outdoor recreation, leisure, and
Tourism (pp. 458). New York: CABI International.

Geographic Research, Inc. (2008). Map retrieved November 17, 2008, from SimplyMap
database.

Getis, A. (1995). Spatial filtering in a regression framework: experiments on regional
inequality, government expenditures, and urban crime. In L. Anselin & R.J.G.M.
F lorax (Ed.), New Directions in Spatial Econometrics ( pp. 172-88). Berlin:
Springer.

228

Goss, J. (1995). “We Know Who You Are and We Know Where You Live”: The
Instrumental Rationality of Geodemographic Systems. Economic Geography, 71
(2), 171-198.

Graham, S. (2005). Software-sorted geographies. Progress in Human Geography, 29,
562—80.

Green, G. T., & Cordell, K. (2003). Boating trends & the significance of demographic
change. Retrieved October 12, 2008,
http://www.srs.fs.usda.gov/trends/NASBLALV.pdf.

Greene, G., Moss, C. B., & Spreen, T. H. (1997). Demand for recreational fishing in
Tampa Bay, Florida: A random utility approach. Marine Resource Economics,
12(4), 293-305.

Greenland, S. (1992). Divergent biases in ecologic and individual-level studies. Statistics
in Medicine, I I, 1209-1223.

Greenland, S. (2002). A review of multilevel theory for ecologic analyses. Statistics in
Medicine, 21, 389—395.

Griffith, D. (2002). A spatial filtering specification for the auto-Poisson model, Statistics.
Probability Letter, 58, 245—251.

Griffith, D. (2003). Spatial autocorrelation and spatial filtering: gaining understanding
through theory and scientific visualization. Berlin: Springer-Verlag.

Griffith, D. (2008). Spatial-filtering-based contributions to a critique of geographically
weighted regression (GWR). Environment and Planning A, 40(11), 2751-2769.

Grow Boating (2007). Barriers & Competitors Quantitative Study by Left Brain
Marketing. Retrieved September 19, 2008,
http://www.houseboatindustry.org/discoverboating.aspx.

Haas, G. E., Wells, M. D. ,Lovejoy, V., & Welch, D. (2007). Estimating future recreation
demand: a decision guide for the practitioner. Denver, Colorado: United States
Department of the Interior, Bureau of Reclamation, Office of Program and Policy
Services, Denver Federal Center.

Hakim, S. (1981). Forecasting demand for recreational boats. Journal of the American
Water Resources Association, 1 7, 6.

Hagler Bailly, Inc. (1997). The national recreational boating needs assessment survey:
Final report. A report prepared for the international Association of Fish and
Wildlife Agencies. Washington DC.

229

 

Hanemann, W. M. (1978). A methodological and empirical study of the recreation

benefits from water quality improvement (Ph.D. Dissertation, Harvard University,
1978)

Hanemann, W. M. (1984). Discrete-continuous models of consumer demand.
Econometrica, 52 (2), 541-561 .

Hanemann, W. M. (1985). Welfare analysis with discrete choice models. In C. H. Kling,
Valuing Recreation and the Environment. Northampton, MA: Edward Elgar.

Harris, L. K., Gimblett, H. R., & Shaw, W. W. (1995). Multiple use management: using a
GIS model to understand conflicts between recreationists and sensitive wildlife.
Society and Natural Resources, 8, 559-572.

Henderson, K.A., Stalnaker, D., & Taylor, G. (1988). The relationship between barriers
to recreation and gender-role personality traits for women. Journal of Leisure
Research, 20, 69-80.

Henderson, K. A. (1991). The contribution of feminism to an understanding of leisure
constraints. Journal of Leisure Research, 23, 363-377.

Howland, J ., Mangione, T. W., & Minsky, S. (1996). Perceptions of risks of drinking and
boating among Massachusetts boaters. Public Health Reports, 111(4), 372-3 79.

Hunt, K. M., & Ditton, K. R. (2002). Freshwater fishing participation patterns of racial

and ethnic groups in Texas. North American Journal of Fisheries Management, 22,
52—65.

Johnson, S. B. (1994). Recreational boating impact investigations, Upper Mississippi
River System. LTRMP-EMTC Special Report 94-S004. Pool 4, Red Wing,
Minnesota: US. Fish and Wildlife Service.

Kaiser, R. (1970) A study of multiple boat ownership in Michigan (MS. Thesis,
Department of Park and Recreation Resources, Michigan State University, 1970).

Kravets, N., & Hadden, W. C. (2007). The accuracy of address coding and the effects of
coding errors. Health & Place, 13, 293-298.

Lee, B., Graefe, A., & Burns, R. (2003). Analysis of demographic segmentation of local
residents in the Columbia River Gorge National Scenic Area: A GIS approach for
the potential park visitors. 2003 Annual Conference of Population Association of
America, Minneapolis, MN.

Leeworthy, V., & Wiley, P., (2001). Current participation patterns in marine recreation.
Special Projects on National Survey on Recreation and the Environment 2000, US.

230

Department of Commerce, National Oceanic and Atmospheric Administration
National Ocean Service, Silver Spring, Maryland. Retrieved December 28, 2008,
http://marineeconomics.noaa.gov/nsre/nsre_2.pdf.

Leeworthy, V. R., Bowker, J. M., Hospital, J. D., & Stone, E. A. (2005). Projected
participation in marine recreation: 2005 & 2010. Final report; US. Department of
Commerce, National Oceanic and Atmospheric Adminsitration. Silver Spring, MD:
US. Department of Commerce, national Oceanic and Atmospheric Adminsitration,
National Ocean Service.

Lenskyj, H. (1988). Measured time: women, sport, and leisure. Leisure Studies, 7, 233-
240.

Leung, Y. F., Shaw, N., Johnson, K., & Duhaime, R. (2002). More than a database:
integrating GIS data with the Boston Harbor Islands visitor carrying capacity study.
In: The George Wright Forum, 19(1), 69-78.

Lentnek, B., Van Doren, C. S., & Tiall, R. J. (1969). Spatial behavior in recreational
boating. Journal of Leisure Research, 1(2), 103-124.

Levine, N. (2007). CrimeStat: A spatial statistics program for the analysis of crime
incident cocations (v 3.1). from CrimeStat Manual. Retrieved Janurary 20, 2009,
http://www.icpsr.umich.edu/CRIMESTAT/files/CrimeStatChapter. 13.pdf.

Lipton,D. W., & Miller, S. (1995). Recreational boating in Maryland: an economic
impact stuay. UM-SG-MAP-95-02. College Park, Maryland: Maryland Sea Grant
Extension Service, University of Maryland.

Lipton, D. (2003). The value of improved water quality to Chesapeake Bay boaters,
Working Paper, University of Maryland, College Park (2003). Retrieved May 15,
2009, http://www.arec.umd.edu/Publications/papers/Working-Papers-PDFfiles/03-
16.pdf.

Loomis, J. B., & Walsh, R. G. (1997). Recreation economic decisions: comparing
benefits and costs. State College, PA: Venture Publishing.

Lueck, D., & Thomas, C. L. (1997). Effect of becoming an outdoors -woman program on
attitudes and activities. Women in Natural Resources, 19(1), 4-8.

Mahoney, E., Stynes, D., Cui, Y., Herbowicz, T., & Stock, T. (2009). Economic
significance of recreational boating in Florida. Report created for the Florida Fish
and Wildlife Conservation Commission, June 2009.

McCool, S. F., & Clark, R. N. (2008). Water and people: challenges at the interface of
symbolic and utilitarian values. USDA Forest Service Pacific Northwest Research
Station General Technical Report PNW-GTR-729.

231

Minnesota DNR (2004) Shoreline and water quality impacts from recreational boating on
the Mississippi River. Retrieved August 18, 2009,
http://www.dnr.state.mn.us/aboutdnr/reports/impacts_mississippi.html.

Minnesota DNR (2006) Changes in outdoor recreation participation. Retrieved at July 19,
2009,

http://www.dnr.state.mn.us/conservationagenda/trend/outdoor_recreation.html.

Milon, W. (2000). Current and future participation in the marine recreational fishing in
the Pacific U.S. Region. NOAA. Retrived September 12, 2009,
http://www.st.nmfs.noaa.gov/stS/RecEcon/Publications/fspo45.pdf.

Moeltner, K. (2003) Addressing aggregation bias in zonal recreation models. Journal of
Environmental Economics and Management, 45(1), 128-144.

Molberg, P. J., Hopkins, R.S., Paulson, J ., & Gunn, RA. (1993). Fatal incident risk
factors in recreational boating in Ohio. Public Health Reports, 108(3), 340-346.

Monmonier, M. (2002) Spying with Maps: Surveillance Technologies and the Future of
Privacy. Chicago: University of Chicago Press.

Morfin, M., & Makowski, D. (2009) Model selection uncertainty and model mixing
Package “MMIX”. Retrived September 28, 2009, http://www.r-project.org.

Nelson, A. (2001). Analyzing data across geographic scales in Honduras: detecting levels
of organization within systems. Agriculture, Ecosystem and Environment, 85 (1 -3),
107-131.

Nicholls, S. (2001). Measuring the accessibility and equity of public parks: a case study
using GIS. Managing Leisure, 6, 201-219.

Nicholls, S., & Shafer, C. S. (2001). Measuring Accessibility and Equity in a Local Park
System: The Utility of Geospatial Technologies to Park and Recreation
Professionals. Journal of Park and Recreation Administration, 19(4), 102-124.

NMMA (2008). Recreational Boating Statistics Abstracts. Chicago, IL.

NMMA (2009) New Boat Sales Report. Chicago, IL.

Northridge, M., Sclar, E., & Biswas, P. (2003). Sorting out the connections between the

built environment and health: A conceptual framework for navigating pathways and
planning healthy cities. Journal of Urban Health, 80, 556-568.

232

Oliveira, R. P., & Fuks, S. D. (2007). Geo-Inforrnation system modeling for
environmental research applications. Retrieved September 20, 2007,
http://wgbis.ces.iisc.emet.in/energy/HC270799/LM/SUSLUP/Thema5/241/241.pdf.

Openshaw, S. & Blake, M. (1995) Geodemographic segmentation systems for screening
health data. Journal of Epidemiology and Community Health, 49(2), 34-38.

Openshaw, S., & Taylor, P. (1979). A million or so correlation coefficients: three
experiments on the modifiable areal unit problem. In N. Wrigley (Ed.), Statistical
Applications in the Spatial Sciences (pp. 127-144). London: Pion.

Paez, A. (2005) Local analysis of spatial relationships: a comparison of GWR and the
expansion method. Lecture Notes Computer Science, 348, 162-172.

Peyton, R. B., & Gigliotti, L. M. (1989). The utility of sociological research: a re-
examination of the East Matagora Bay experience. Fisheries, 14(4), 5-8.

Pickett, K. E., & Pearl, M. (2000). Multilevel analyses of neighborhood

sociodemographic context and health outcomes: A critical review. Journal of
Epidemiology and Community Health, 5, l 1 1-122.

R Development Core Team (2008). R: A language and environment for statistical
computing. R Foundation for Statistical Computing, Vienna, Austria. ISBN 3-
900051-07-0, Retrieved January 05, 2010, http://www.R-project.org.

Ratcliffe, J. H. (2004) Geocoding crime and a first estimate of a minimum acceptable hit
rate International. Journal of Geographical Information Science, 18(1), 61 -72.

Recreational Boating & Fishing Foundation. (2003). Boating and fishing attitude
segmentation study. Retrieved Febuary 08, 2009,
http://www.rbff.org/up]oads/Research_section/Part_l.pdf

Responsive Management (1995). Factors related to hunting and fishing participation
among the nation ’s youth, Phase I: review of the literature. Harrisonburg, VA:
Responsive Management.

Responsive Management (1998). Women’s, Hispanics and African-Americans’
participation in, and attitudes toward boating and fishing. Harrisonburg, VA:
Responsive Management.

Responsive Management. (2002). Kansas residents ’ participation in, opinions on, and
attitudes toward boating in Kansas. Conducted for the Kansas Department of
Wildlife and Parks.

233

Richardson, S. (1992). Statistical methods for geographical correlation studies. In P.
Elliott, J. Cuzick, D. English, & R. Stern (Eds.), Geographical and environmental

epidemiology: Methods for small-area studies (pp. 181-204). Oxford: Oxford
University Press.

Robinson, W. S. (1950), Ecological correlations and the behaviour of individuals.
American Sociological Review, I 5, 351-357.

Rosenberger, R. S., & Loomis, J. B. (2001). Benefit transfer of outdoor recreation use

values: a Technical Document Supporting the Forest Service Strategic Plan. Gen.
Tech. Rep. RMRS-GTR-72. Fort Collins, CO: US: Department of Agriculture,
Forest Service, Rocky Mountain Research Station.

Sampson, R., Morenoff, J ., & Gannon-Rowley, T. (2002). Assessing “neighborhood

effects”: Social processes and new directions for research. Annual Review of
Sociology, 28, 443-478.

Sidman. C., Antonini, G., Saures, S., Jones, 6., & West, N. (2000). Evaluating
recreational boating patterns at selected sites in Southwest Florida for regional

anchorage management. Florida Sea Grant Publication TP-105. University of
Florida, Gainesville.

Sidman, C. F ., & Fik, T. J. (2002). Enhancements to a regression-based approach to

estimate preferred recreational boating destinations in Charlotte Harbor, Florida.
Florida Marine Research Institute Report, St. Petersburg, Florida.

Sidman, C. F., Fik, T. J ., & Sargent, B. (2004). A Recreational boating characterization
for Tampa and Sarasota Bays. publication of the University of Florida Sea Grant
Program and the Fish and Wildlife Conservation Commission supported by the
National Sea Grant College Program of the United States Dept. of Commerce and

the National Oceanic and Atmospheric Administration under NOAA Grant #NA
16RG-2195, award number NA17OZZ330.

Sidman, C. F., & Fik, T. J. (2005). Modeling spatial patterns of recreational boaters:
vessel, behavioral, and geographic considerations. Leisure Sciences, 27, 1-15.

Smith, V. K., & Kopp, J. R. (1980). The spatial limits of the travel cost recreational
demand model. Land Economics, 56(1), 64-72.

Smith, V. K. (1988). Selection and recreation demand. American Journal of Agricultural
Economics, 70(1), 29-36.

Smith, K. (2005). Florida population growth: past, present and future. Retrieved July 5,
2008, http://www.floridaforboomers.com/population-of-florida/.

234

 

Souleyrette, R. R. (1993). Use of supercomputer for interactive travel demand modeling
through GIS. Center for Transportation Research and Education Project, Retrieved
September 20, 2007, at http://www.ctre.iastate.edu/reports/supercomputer.pdf.

Sowman, M. R. (1987). A procedure for assessing recreational carrying capacity of
coastal resort areas. Landscape and Urban Planning, 14, 331-344.

SPSS 17.0 for Windows. (2009). (SPSS Inc., Chicago IL).

Stamps, S, & Stamps, M. (1985). Race, class and leisure activities of urban residents.
Journal of Leisure Research, 17(1), 40-56.

Stoll, J. R., Bergstrom, J. C., & Jones, L. L. (1988). Recreational boating and its
economic impact in Texas. Leisure Sciences, 10, 51—67.

Strong, E. J. (1983). A Note on the functional form of travel cost models with zones of
unequal populations. Land Economics, 59(3), 342-349.

Stynes, D., & Holecek, D. (1982). Michigan Great Lakes recreational boating: a
synthesis of current information. Michigan Sea Grant Technical Report MICHU-
SG-82-204. Ann Arbor.

Stynes, D., Brother, G., Holecek, D., & Verbyla, D. (1983). Recreational boating
spending patterns and economic impacts of Michigan registered boat owners.
Michigan Sea Grant Technical Report MICHU-SG-83-210. Ann Arbor.

Stynes, D. J ., Peterson, G. L., & Rosenthal, D. (1986). Log transformation bias in
estimating travel cost models. Land Economics, 62(1), 94-103.

Stynes, D.,Wu, T., & Mahoney, E. (1998). 1994 Michigan boating survey. Research
report 549. East Lansing, Michigan: Michigan Agricultural Experiment Station,
Michigan State University.

Sutherland, R. J. (1982). A regional approach to estimating recreation benefits of
improved water quality. Journal of Environmental Economics and Management, 9
(9), 229-47.

Tanner, K. (2008, December 23). Michigan Population Continues to Decline. Detroit
Free Press.

Tarrant, M. A., & English, D. B. K. (1996). A crowding-based model of social carrying
capacity: application for whitewater boating use. Journal of Leisure Research, 28,
155-168.

235

Tatalovich, 2., Wilson, J ., Milam, J ., Jerrett, M., & McConnell, R. (2006). Competing
definitions of contextual environments. International Journal of Health
Geographics. Retrieved April 12, 2009, http://www.ij-
healthgeographics.com/content/S/1/5 5.

Thomas, C. L., & Lueck, D. H. (1996). Training and retraining women in traditional fish
and wildlife-based recreation. Transactions of the 61st North American Wildlife and
Natural Resources Conference, 61, 374-378.

Thomas, M., & Straitis, N. (2002). Compensating variation for recreational policy: a

random utility approach to boating in Florida. Marine Resource Economics, 17, 23-
33.

Tobler, W. R. (1970). A computer movie simulating urban growth in the Detroit Region.
Economic Geography, 46, 234-240.

Trice, A. H., & Wood, S. E. (1958). Measurement of recreation benefits. Land
Economics, 34, 195-207.

US. Census Bureau (2000a). Census Bureau Geography, US. Census Bureau,
Washington DC.

US. Census Bureau (2000b). Earnings by occupation and education, US. Census
Bureau, Washington DC. Retrieved April 15, 2009,
http ://www.census. gov/hhes/www/income/earnings/ call 1 usboth.htm1.

US. Census Bureau (2004). Income quintile and top 5 percent US. Census Bureau,
Washington DC. Retrieved May 22, 2009,
http://pubdb3.census.gov/macro/032005/hhinc/new05__000.htm.

US. Census Bureau (2008a). Annual estimates of the resident population for the United
States, Regions, States, and Puerto Rico: April 1, 2000 to July 1, 2008 (N ST-
EST2008-01), US. Census Bureau, Washington DC. Retrieved July 19, 2009,
http://www.census.gov/popest/states/NST-ann-est.htrnl.

US. Census Bureau (2008b). America’s families and living arrangements, US. Census
Bureau, Washington DC. Retrieved June 08, 2009,
http://www.census.gov/population/www/socdemo/hh-fam/cpsZOO8.html.

US. Census Bureau (2009a). Population division. US. Census Bureau, Washington DC.
Retrieved May 22, 2009, http://www.census.gov/popest/estimates.html.

US. Census Bureau (2009b). American fact finder. US. Census Bureau, Washington

DC. Retrieved May 22, 2009,
http://factfinder.census.gov/home/saff/main.html?_lang=en.

236

Vaughan, W. J ., Russell, C., & Hazilla, M. (1982). A note on the use of travel cost
models with unequal zonal populations: comment. Land Economics, 58(3), 400-
407.

Walker, 8., & Cuff, R. (1997). Location optimisation — what’s the problem? Association
for Geographical Information Systems Conference Proceedings 1 99 7, paper 4.3.

Waller, L. A., & Gotway, C. A. (2004). Applied spatial statistics for public health data.
Hoboken, NJ: John Wiley.

Wheeler, D. & Tiefelsdorf, M. (2005) Multicollinearity and correlation among local
regression coefficients in geographically weighted regression. Journal of
Geographical Systems, 7, 161-1 87.

Wise, 8., Haining, R., & Ma, J. (1997). Regionalisation tools for the exploratory spatial
analysis of health data. In A. Getis and M.M. Fischer (Eds.), Recent Developments
in Spatial Analysis (pp. 83-100). Berlin: Springer.

Wrigley, N. (1995). Revisiting the modifiable areal unit problem and the ecological
fallacy. In Cliff, A. D., Gould, P. R., Hoare, A. G. & Thrift, N. J. (Eds.), Diffusing
geography (pp. 49-71). London: Blackwell.

Widmer, W., Underwood, A. & Chapman, M. (2002). Impacts of recreational boating &
PWC Use. Natural Recourse Management, 5(1), 22-27. Retrieved Apri112, 2009,
http://www.mass.gov/czm/pwcmgntguide2.pdf.

Wong, D. (1996). Aggregation effects in geo-referenced data. In Arlinhaus, S. (Ed.),
Practical handbook of spatial statistics (pp. 83—106). Boca Raton, FL: CRC Press.

Wu, T. C. (1995). A system of models for estimating recreational boating use in
Michigan Counties ( Ph.D. Dissertation. East Lansing, Michigan: Michigan State
University, 1995).

Zandbergen, P. A. (2009). Geocoding quality and implications for spatial analysis.
Geography Compass, 3(2), 647-680.

Zawacki, W., & Marsinko, A. (1999). Using Geographic Information Systems with travel
cost models: a case study. Proceedings of the 2000 Northeastern Recreation
Research Symposium (op. 28 7-292). Bolton Landing, NY: USDA Forest Service
General Technical Report NE-269.

Ziemer, F. R., Musser, N. W., & Hill, R. C. (1980). Recreation demand equations:

functional from and consumer surplus. American Journal of A gricultural
Economics, 62, 136.

237

Ziemer, R. F. & Musser, W. N. (1979). Population-specific recreation demand models

and the consequences of pooling sample data. Western Journal of A gricultural
Economics, 4(1), 121-128.

Zou, L. (2005). The modifiable areal unit problem: empirical evaluation using census
data of the City and County of Honolulu. (Ph.D. dissertation, University of Hawaii
at Manoa, United States — Hawaii). Retrieved March 23, 2010, from Dissertations
& Theses: A&I. (Publication No. AAT 3198390).

238

   

”111111111111111111111111111111111111“
3 1293 03063 6983