MEASURING PREFERENCES FOR
CHANGES IN WATER QUALITY AT GREAT LAKES BEACHES
USING A CHOICE EXPERIMENT
By
Scott Arndt Weicksel

A THESIS
Submitted to
Michigan State University
In partial fulfillment of the requirements
for the degree of
MASTER OF SCIENCE
Agricultural, Food and Resource Economics
2012

ABSTRACT
MEASURING PREFERENCES FOR CHANGES IN WATER QUALITY AT
GREAT LAKES BEACHES USING A CHOICE EXPERIMENT
By
Scott Arndt Weicksel
In two essays, this thesis reports the findings of a choice experiment (CE)
conducted as part of a state-wide web survey of Great Lakes beach goers in Michigan.
Despite the popularity of visiting Great Lakes beaches, little is known about visitors’
preferences for environmental attributes at beaches. Existing literature focuses on
marine beach valuation and the valuation of Great Lakes ecosystems services other
than beaches. In the first essay, to address this gap, we gather preferences for
environmental quality attributes at Great Lakes beaches. We find respondents prefer
beaches closer to home, beaches with less algae in the shore and in the water, and
beaches tested for bacteria. In a second essay, we examine the effect of labeling (i.e.
assigning names to) the alternatives within a CE. Although there is a growing literature
focused on the effects of CE design elements at the discretion of the researcher, little
attention has been given to the effects of different labeling schemes (i.e. whether
alternatives within the choice sets are labeled with a name or are left as generic
alternatives). We employ a split-sample CE where respondents see either labeled,
same-labeled (i.e. labels present but held constant), or unlabeled alternatives (with
Great Lake names used as labels). Although we find some differences in parameters
and marginal rates of substitution from the three labeling schemes, results are highly
similar in rank and magnitude, suggesting a large degree of preference consistency and
a high degree of transferability of the values for use in benefit transfer.

Copyright by
SCOTT ARNDT WEICKSEL
2012

ACKNOWLEDGEMENTS
Funding for this research was provided by NOAA Coastal Oceans, Multiple
Stressors project and the MSU AgBioResearch and multistate project W2133.
This research benefited from assistance provided by Shannon Briggs (Michigan
Department of Environmental Quality), Charles Kovatch (U.S. EPA), Sonia Joseph
Joshi (Michigan Sea Grant), Erin Dreelin and the Center for Water Sciences at MSU,
and numerous County health officials and beach managers who shared their time with
us.
I would thank my committee members, Drs. John Hoehn, Michael Kaplowitz and
Frank Lupi, for their guidance in the preparation of this thesis. To the Co-PI’s of the
project, Dr. Kaplowitz and Dr. Lupi, I would like to express the utmost gratitude for the
opportunity to work on this project—all of the lessons learned and experiences gained
over the past two years are ones I will draw upon regularly as I continue on to new
pursuits. Also, I wish to thank my teammates Min Chen and Kwame Yeboah for their
help in keeping the project on course, and helping me learn along the way: best wishes
to you both as you work to complete your dissertations. Thanks are of course due to
Dustin Kubas, the project assistant who laughs at paper cuts, lunch breaks, and all of
Jody Knol’s jokes; and to Kevin Adams, whose energy and teamwork was greatly
appreciated. And lastly, to my friends and family, especially my parents and Lindsay: I’m
incredibly lucky to continually benefit from all of your encouragement and support-thank you all for your thoughtfulness, energy, and home-cooked meals you provided me
along the way.

iv

TABLE OF CONTENTS

LIST OF TABLES ........................................................................................... vii
LIST OF FIGURES ......................................................................................... x
INTRODUCTION ............................................................................................ 1
CHAPTER 1
Introduction .................................................................................................. 12
Threats to the resource: Human Development ............................................ 12
Threats to the resource: Harmful Bacteria ................................................... 13
Threats to the resource: Nuisance Algae .................................................... 14
Purpose ....................................................................................................... 15
Previous Research: Valuing Changes in the
Great Lakes Environment ............................................................................ 16
Previous Research: Beach Recreation ........................................................ 19
Previous Research: Great Lakes Beach Recreation ................................... 23
Method: Choice Experiment ........................................................................ 24
Random Utility Theory ................................................................................. 27
Survey Development: Pretests .................................................................... 31
Survey Development: Input from experts .................................................... 34
Survey Sample ............................................................................................ 37
Web Survey Implementation........................................................................ 37
Attributes and Attribute Levels ..................................................................... 39
Results......................................................................................................... 45
Conclusions ................................................................................................. 51
CHAPTER 2
Introduction .................................................................................................. 54
Purpose ....................................................................................................... 55
Research Questions .................................................................................... 61
Method: Choice Experiment ........................................................................ 63
Random Utility Theory ................................................................................. 65
Previous Research: Sensitivity to Design Factors ....................................... 69
Previous Research: Effect of Labeling ......................................................... 71
Application: Great Lakes Beaches............................................................... 74
Previous Research: Great Lakes and Beach Valuation ............................... 75
Survey Development: Pretests .................................................................... 76
Survey Development: Input from experts .................................................... 77
Survey Sample ............................................................................................ 78
Web Survey and Choice Experiment ........................................................... 81
Attributes and Attribute Levels ..................................................................... 82
Results......................................................................................................... 86
v

Conclusions ................................................................................................. 97
APPENDICES ................................................................................................ 100
APPENDIX A: Image of screen display from Great Lakes Beaches
Web-survey, choice experiment and follow-up portion,
original size monitor-dependent ............................................. 101
APPENDIX B: Screener Survey (Michigan activities survey) materials,
contact schedule, robodial scripts, and disposition ................ 127
APPENDIX C: Web Survey (Great Lakes beaches web survey) materials,
contact schedule, and robodial scripts ................................... 155
APPENDIX D: Details of Michigan Activities Survey, Robocalls, and
Great Lakes Beaches web survey methods ........................... 176
REFERENCES ............................................................................................... 202

vi

LIST OF TABLES

Table 1.1

Attributes and Attribute Levels Included in the Great Lakes
Beaches Choice Experiment ...................................................... 40

Table 1.2

Number of Respondents and Observations for Great Lakes
beach web survey Choice Experiment ....................................... 45

Table 1.3

Results of random-effects logit model estimating determinants
of Great Lakes beach choice...................................................... 47

Table 1.4

Mean estimate of Marginal Rates of Substitution for Great
Lakes beach characteristics, including lower-bound and
upper-bound estimates of 95% Confidence Intervals from
Krinsky-Robb method, 10,000 draws ......................................... 49

Table 2.1

Attributes and Attribute Levels Included in the Great Lakes
Beaches Choice Experiment ...................................................... 83

Table 2.2

Results of Random-Effects Logit model estimating
determinants of Great Lakes beach choice for “Labeled,”
“Same-Labeled” and “Unlabeled” choice experiments ............... 87

Table 2.3

Mean and Krinsky-Robb 95% confidence intervals
of Marginal Rates of Substitution for marginal changes
in Great Lakes beach characteristics across “Labeled,”
“Same-Labeled,” and “Unlabeled” choice experiments .............. 91

Table 2.4

Results of approximate one-sided significance of
differences between mean MRSs: “Labeled and Unlabeled,”
“Labeled and Same-Labeled,” and “Same and Unlabeled”
choice experiments. ................................................................... 96

Table B.1

Schedule of Michigan Activity Survey Contacts ........................ 152

Table B.2

Summary of Michigan Activity Survey response
and disposition .......................................................................... 154

Table C.1

Schedule of Great Lakes Beaches Survey Contacts ................ 175

Table D.1

Description and number of different record matches used
to assign phone numbers to survey sample members for
robodial implementation ............................................................ 183

vii

Table D.2

Date, Time, cost, and number of calls for robodials during
wave 2 of the Michigan Activities Survey.................................... 183

Table D.3

Date, Time, cost, and number of calls for robodials during
wave 3 of the Michigan Activities Survey.................................... 184

Table D.4

Date of Great Lakes Beaches Survey Contacts ........................ 191

Table D.5

Final mailing, distribution and response sorted by
post-paid incentive .................................................................... 192

Table D.6

3-way Chi-squared test comparing response rates to web
survey based on final communication featuring $20, $10
or no post-paid incentive ........................................................... 192

Table D.7

2-way Chi-squared test comparing response rates to web
survey based on final communication offering $20 or $10
post-paid incentive .................................................................... 192

Table D.8

2-way Chi-squared test comparing response rates to web
survey based on final communication offering $10 or no
post-paid incentive .................................................................... 193

Table D.9

2-way Chi-squared test comparing response rates to web
survey based on final communication offering $20 or no
post-paid incentive .................................................................... 193

Table D.10

Response and disposition for the Great Lakes
beaches web survey ................................................................. 194

Table D.11

Great Lakes Web Survey Item Non-Response table 1 of 6,
section: Front Matter ................................................................. 195

Table D.12

Great Lakes Web Survey Item Non-Response table 2 of 6,
section: Choice Experiment section .......................................... 196

Table D.13

Great Lakes Web Survey Item Non-Response table 3 of 6,
section: Great Lakes Opinion section ....................................... 198

Table D.14

Great Lakes Web Survey Item Non-Response table 4 of 6,
section: Exposure section ......................................................... 199

Table D.15

Great Lakes Web Survey Item Non-Response table 5 of 6,
section: Background information section .................................. 199

viii

Table D.16

Great Lakes Web Survey Item Non-Response table 6 of 6,
section: income question .......................................................... 200

Table D.17

Summary of distances seen in Great Lakes beach web survey
choice experiment (in miles) ..................................................... 201

Table D.18

Summary of item response among choice experiment
respondents, listed by labeling scheme viewed ........................ 201

ix

LIST OF FIGURES

Figure 1.1

Image of example choice set from the Great Lakes beaches
choice experiment, original size monitor-dependent ................ 26

Figure 1.2

Image of diagram of Great Lakes beach attribute:
Algae in the water, original size monitor-dependent ................ 35

Figure 1.3

Image of Diagram of Great Lakes beach attribute:
Algae on the shore, original size monitor-dependent ................ 36

Figure 2.1

Image of example choice set from “Labeled”
choice experiment, original size monitor-dependent ................. 57

Figure 2.2

Image of example choice set from “Same- Labeled”
choice experiment, original size monitor-dependent ................. 58

Figure 2.3

Image of example choice set from “Unlabeled”
choice experiment, original size monitor-dependent ................. 59

Figure 2.4

Image of diagram of Great Lakes beach attribute:
Algae in the water, original size monitor-dependent ................ 79

Figure 2.5

Image of diagram of Great Lakes beach attribute:
Algae on the shore, original size monitor-dependent ................ 80

Figure 2.6

Mean and Krinsky-Robb 95% Confidence Intervals
of MRS in terms of miles for “amount of algae on the shore,”
relative to the baseline level “high” ........................................... 92

Figure 2.7

Mean and Krinsky-Robb 95% Confidence Intervals
of MRS in terms of miles for “amount of algae in the water,”
relative to the baseline level “high” ........................................... 92

Figure 2.8

Mean and Krinsky-Robb 95% Confidence Intervals
of MRS in terms of miles for “frequency of testing for bacteria,”
relative to the baseline level “tested daily” ................................ 93

Figure A.1

Image of screen display from Great Lakes Beaches
Web-survey, choice experiment and follow-up portion,
original size monitor-dependent ................................................ 102

Figure B.1

Image of Michigan Activities Survey Wave 1
Introduction letter, original size 8.5” x 11” ................................. 130
x

Figure B.2

Image of Michigan Activities Survey Wave 1
survey instrument, original size of each page 8.5” x 7” ............. 136

Figure B.3

Image of Michigan Activities Survey Wave 2
Introduction letter, original size 8.5” x 11” ................................. 141

Figure B.4

Image of Michigan Activities Survey Wave 2
survey instrument, original size of each page 8.5” x 7” ............ 143

Figure B.5

Image of Michigan Activities Survey Wave 3
introduction letter, original size 8.5” x 11” ................................. 146

Figure B.6

Image of Michigan Activities Survey Wave 3
survey instrument, original size of each page 8.5” x 7” ............. 151

Figure C.1

Image of Great Lakes Beach Web survey
Wave 1 Introduction letter, original size 8.5” x 11” .................... 159

Figure C.2

Image of Great Lakes Beach Web survey Wave 2 reminder,
half sheet black and white postcard, front,
original size 8.5” x 5.5” .............................................................. 161

Figure C.3

Image of Great Lakes Beach Web survey Wave 2 reminder,
half sheet black and white postcard, back,
original size 8.5” x 5.5” .............................................................. 161

Figure C.4

Image of Great Lakes Beach Web survey Wave 3 reminder,
quarter sheet color postcard, front,
original size 4.25” x 5.5” ............................................................ 162

Figure C.5

Image of Great Lakes Beach Web survey Wave 3 reminder,
quarter sheet color postcard, back,
original size 4.25” x 5.5” ............................................................ 162

Figure C.6

Image of Great Lakes Beaches Web Survey Wave 4 reminder
letter, $20 post-paid incentive, original size 8.5” x 11” .............. 165

Figure C.7

Image of Great Lakes Beaches Web Survey Wave 4 reminder
letter, $10 post-paid incentive, original size 8.5” x 11” .............. 169

Figure C.8

Image of Great Lakes Beaches Web Survey Wave 4 reminder
letter, no post-paid incentive, original size 8.5” x 11” ................ 173

Figure C.9

Image of Great Lakes Beaches Web Survey Wave 4 reminder
Business Reply Mail postcard stating respondent
xi

does not have the Internet, front, original size 4.25” x 5.5” ....... 174
Figure C.10 Image of Great Lakes Beaches Web Survey Wave 4 reminder
Business Reply Mail postcard stating respondent
does not have the Internet, back, original size 4.25” x 5.5” ....... 174

xii

INTRODUCTION:
The Laurentian Great Lakes are one of the nation’s most vast and important natural
resources. In Michigan, the Great Lakes support a broad range of public uses including
fishing, boating, and sightseeing along with diverse private industry. Even though visiting
Great Lakes public beaches is one of the most popular outdoor activities among Michigan
residents, the quality of Great Lakes beaches comes under constant threat from natural
(e.g. nuisance algae in the water and on the shore) and man-made (e.g. elevated levels of
unsafe bacteria, developed shoreline) sources.
When considering how the resource should be managed to maximize public wellbeing, there is a clear gap in required information. While much is known about the effects of
stressors on the health and function of natural systems, little is known about the public’s
preferences and values for different levels of environmental quality at Great Lakes beaches.
Programs that preserve and protect the quality of Great Lakes beaches can be costly: since
2010, the Great Lakes Restoration Initiative has invested over $1 billion into projects
throughout the Great Lakes with an additional $300 million authorized for 2013 (U.S. EPA
2009, 2010, 2011, 2012a). Without information on Great Lakes beach goers’ preferences for
environmental quality attributes, decision makers are unable to find an efficient balance
between cost and benefit of such programs. With each passing year, government spending
on programs to improve Great Lakes beach quality falls under further scrutiny, further
highlighting the importance of gathering scientifically sound information on the preferences
of the public for different conditions at Great Lakes beaches. Funding decisions affecting the
quality of Great Lakes beaches can occur at the local, state, or federal level. For example,
the proposed 2013 budget for the U.S. EPA would eliminate the federal support of the

1

BEACH Grant Program which funds monitoring of beaches for elevated levels of harmful
bacteria (U.S. EPA 2012b).
Previous research has focused on valuing a variety of ecosystem goods and services
provided by the Great Lakes. Lupi et al. (1998) estimate a statewide model of recreational
fishing demand to measure the benefits to anglers that result from changes in fishing site
characteristics such as catch rates. Lupi, Hoehn and Christie (2003) estimate the benefits to
anglers resulting from a program to control invasive sea lamprey in the St. Mary’s River that
would result in a predicted rebound for the trout population in Lake Huron. Kotchen et al.
(2007) use a similar recreational fishing model to show that the benefits of improved habitat
that results from hydropower dam relicensing exceed the cost of the operational changes.
Other studies use the hedonic method relating observed changes in environmental quality
(such as water quality or air quality) to the prices of nearby property to estimate economic
values of changes in environmental quality. Ara et al. (2006) use a hedonic pricing analysis
of the value of houses near Lake Erie and find that changes in water quality in Lake Erie
have significant impacts on the value of nearby houses. Braden et al. (2004) use a hedonic
model as well as a choice experiment to estimate the benefits from cleaning up
contamination in the Waukegan Harbor in Waukegan, IL. Chattopadhyay (1999) used
hedonic pricing to estimate the residents’ willingness to pay for air quality improvements in
the largest Great Lakes coastal population center, Chicago, finding that residents have a
higher willingness to pay for reduction in particulate matter (PM-10) than sulfur dioxide
(S02). As a means of valuing proximity to and aesthetics of Lake Erie, Seiler et al. (2001)
use a hedonic pricing model of homes in Ohio and estimate that all else equal, houses that

2

have a view of Lake Erie are an average of 56% more valuable than houses that do not
have a view of Lake Erie.
Whitehead et al. (2009) combine estimates from revealed preference (travel cost
recreation site choice) and stated preference (contingent valuation of hypothetical
restoration scenarios) to estimate the value of restoring Saginaw Bay wetlands. Hoehn et al.
(2010) use a choice experiment to examine the value of wetlands in Michigan in terms of
economic equivalency of wetland services that are part of the replacement of impaired or
eliminated wetlands required under the Clean Water Act. The authors’ findings describe the
public’s willingness to accept wetlands of varying characteristics as compensation for
impaired or eliminated wetlands. In general, respondents required increased amounts of
wetlands when the restored wetland was of lower quality in terms of habitat than the
impaired wetland.
There also exists a literature studying the value for characteristics and environmental
quality at beaches, but research tends to focus on marine beaches. Bockstael, McConnell
and Strand (1989) use the results of a contingent valuation telephone survey of residents of
the Baltimore-Washington DC metro area to estimate individual’s willingness to pay to
improve water quality in the Chesapeake Bay to levels acceptable for swimming. Smith,
Zhang and Palmquist (1997) employ a contingent valuation survey to measure the economic
value of controlling marine debris (natural or man-made) at recreational beaches in New
Jersey and North Carolina. Parsons, Massey and Tomasi (1999) use a random utility model
to estimate the demand Delaware residents have for marine beaches within a day’s drive
and relate that demand to benefits of beach nourishment programs that keep beaches from
eroding to less preferred widths. Landry, Keeler and Kriesel (2003) use hedonic pricing and

3

a choice experiment to value additional beach width along the beaches of Tybee Island east
of Savannah, GA. Shivlani, Letson and Theis (2003) surveyed visitors to Southern Florida
coastal beaches, including a contingent valuation question aimed at valuing beach
nourishment programs. Lew and Larson (2005) surveyed residents of San Diego County
and estimate a travel cost model and calculate implicit prices for different beach
characteristics including the presence of life guards and water quality measures. Hilger and
Hanemann (2006) use information on beach visits among residents of Southern California
and use the travel cost method to estimate individual’s willingness to pay for improvements
in water quality. Huang, Poor and Zhao (2007) sampled randomly selected households in
New Hampshire and Maine about their preferences for different coastal erosion control
programs.
In contrast, there is only one published peer-reviewed journal article valuing the
changes in characteristics at Great Lakes beaches. Murray, Sohngen, and Pendleton (2001)
valued decreasing incidences of water contamination (beach advisories) at beaches on
Lake Erie using the results of a survey of visitors to Lake Erie beaches in 1998. A separate
technical report stemming from a similar study, Sohngen, Lichtkoppler and Bielen (1999)
use a travel cost model to estimate the value of a trip to beaches at Maumee Bay State Park
and Headlands State Park on Lake Erie. In addition, unpublished research by Egan and
Dwyer (2008) uses a small sample of visitors to the Maumee Bay State Park in northwest
Ohio to investigate the value of wetland restoration that would result in the removal of all
swim advisories at the site. In unpublished findings presented to several conferences,
Shaikh (2005 and 2012) estimates a travel cost model Lake Michigan beaches in Chicago.

4

This overview provides examples of environmental and resource valuation work on
the Great Lakes, as well as numerous studies valuing beach recreation and beach
characteristics. Despite the vastness of the Great Lakes and the popularity of beach
visitation, the literature remains thin in terms of studies valuing Great Lakes beach
characteristics. The few available studies focus on a limited number of sites, often have
small sample sizes, and only one has been peer-reviewed.
To address these gaps in the literature, this thesis provides the first insights into
Michigan resident’s preferences and values for environmental conditions at Great Lakes
beaches. It reports the results of a state-wide web survey of Michigan residents conducted
in the spring of 2012. As part of the web survey, respondents took part in a choice
experiment pertaining to Great Lakes beach characteristics. The results of the choice
experiment provide valuable insights into the public’s preferences for environmental
conditions at Great Lakes beaches. The implementation of the choice experiment also
included an experimental element to examine the effect that design elements that are
chosen by researchers have on the outcome of choice experiments.
This thesis is divided into two chapters. The first chapter discusses the factors that
threaten the health of the natural systems and human enjoyment of Great Lakes beaches,
then describes the implementation of a choice experiment as a way to provide insight about
Michigan residents’ values and preferences for different conditions at Great Lakes beaches
in Michigan. This study builds on previous work within the field of environmental and natural
resource economics that values aspects of the Great Lakes, but is unique in that only one
previously published study focuses on Great Lakes beaches. The preference information will
prove useful to resource managers and decision makers in need of scientifically sound, fact-

5

based research about residents’ economic values and preferences for environmental quality
characteristics that may be protected by costly, budget-limited programs. The results
reported in this thesis stand as a first step to closing a gap left by the existing literature for
information that could be used to prioritize further studies or determine demand for
programs to protect and enhance environmental conditions at Great Lakes beaches.
Results of the choice experiment show that nearly all levels of attributes included in
the experiment (distance from home, measures of algae on the shore and in the algae in the
water, beach length, and frequency of bacteria testing) are statistically significant and have
the expected sign. The results show that all else equal, Michigan beach users prefer
beaches on the Great Lakes that are closer to home, have lower levels of algae in the water,
and have less algae on the shore. Respondents also prefer beaches that are tested for
bacteria at least monthly versus those that are not tested at all, signaling a value for testing
information which is separate from any change in risks for negative health impacts from
exposure to bacteria. These results can help decision makers prioritize areas of restoration
while seeking to balance the benefits of increasing environmental quality at Great Lakes
beaches with the costs of enacting such programs.
The second chapter of thesis reports on a split-sample experiment within the Great
Lakes beaches web survey where respondents were randomly assigned to view choice sets
that featured different labeling schemes (i.e., whether or not alternatives have names or are
unnamed collections of attributes). Within a choice experiment, alternatives can either be
unlabeled or labeled. Unlabeled choice alternatives are commonly denoted by a generic title
or a letter (e.g. “Choice A, Choice B” or “Program C, Program D”). Labels on choice
alternatives can indicate a brand name, location name, or any other unique identifier for that

6

alternative (e.g. “New York, Los Angeles”, “Coke, Pepsi”, or “Train, Car, Bus”). In the second
chapter of this thesis, within a choice experiment pertaining to Great Lakes beach choice,
we use labels to indicate which Great Lake the beach alternative lies on.
When deciding whether or not to include labels in a choice experiment, researchers
should consider the benefit of added familiarity or realism the labels provide (Adamowicz
and Boxall 2001). However, when labels are present in a choice set, respondents may seek
to simplify the decision making process and base their choice of the preferred alternative
solely on the label, ignoring other attributes (Blamey et al. 2000). Respondents faced with
choice sets containing labels may also consider alternatives outside of those appearing in
the choice task (Adamowicz and Boxall 2001). Concerns exist among practitioners of
attribute-based choice or ranking methods that labeled alternatives may contain levels of
attributes that, in combination with the presence of a label, appear unrealistic or infeasible to
respondents, depending on the respondent’s perceptions of the label (Huybers 2005,
Carson et al. 1994). Despite the concerns over how the presence or absence of labels may
affect the results of choice experiments, little empirical evidence exists to characterize those
effects.
Chapter 2 presents the results of a split-sample choice experiment where
respondents either viewed “labeled” alternatives, where the beaches were labeled with a
Great Lake; a scheme with “same-labeled” alternatives, where Great Lake labels were
present but remained constant within choice sets; and a scheme with “unlabeled”
alternatives, where no Great Lake labels were present.
The investigation into the effect of different labeling regimes builds on previous
research which tests the sensitivity of choice experiment results to different design

7

components at the discretion of the researcher. Researchers across disciplines have used
split-sample studies to vary design elements such as the number of choice sets faced by
respondents (Hensher 2001), the number of alternatives (Rolfe and Bennett 2009, DeShazo
and Fermo 2002, Racevskis and Lupi 2008), choice set presentation formats (Hoehn, Lupi
and Kaplowitz 2010), and answer formats (Fenichel et al, 2009). A group of studies have
used a “Design of Designs” format to generate an experimental design where the design
elements are varied within a split sample choice experiment, so as to isolate the effects of
variation in design elements on the resulting data. Caussade et al. (2005), Hensher (2006b),
and Rose et al. (2009) employ this strategy that varies each of the five design elements
(number of attributes, number of attribute levels, range of attribute levels, number of
alternatives within choice sets, number of choice sets faced by respondents) across the
sample such that they can isolate the impact each element has on the choice data.
While the effects of different design elements such as the number of choice sets,
alternatives, attributes, and attribute levels faced by respondents have been studied, there
are only a few studies that test the effects of labels. Much of the research investigating the
effects of labeling on choice experiment results uses labels to represent a means of
program provision or program funding. Rolfe and Windle (2011) test the effects of labels
indicating the method with which a proposed coral reef protection program would be
achieved. Similarly, Czajkowski and Hanley (2009) conduct a study of Polish households’
willingness to pay to protect local biodiversity and test for differences in the results of a
generic choice experiment and a choice experiment where alternatives are labeled as being
provided through the expansion of an existing national park, or by “other” means. Blamey et
al. (2000) implemented a choice experiment among households in Brisbane, Australia,

8

regarding preferences for forest preservation where a portion of the respondents viewed a
choice experiment with generic policy alternatives, while the remaining respondents viewed
a choice experiment with labeled policy alternatives which named minimum percentages of
scarce forest was to be left preserved. A recent study by Brannlund and Persson (2012)
conducted a choice experiment eliciting information on Swedish citizen’s preferences for
CO2 reduction programs. The authors use labels to indicate different payment mechanisms
with half of the respondents shown unlabeled alternatives, and the other half shown
alternatives labeled as being funded through private (i.e. individual) taxes or by “other”
means. While these studies are able to compare choice experiments with labeled and
unlabeled alternatives, the use of labels to represent different program funding mechanisms
or modes of program provision can be viewed largely as a means of gathering preferences
for separate program attributes, which is different from the use of labels as a brand, site
name, or other unique identifier capturing preferences for information not explicitly listed in
the choice set.
Huybers (2005), however, compares the results of generic and labeled choice
experiments pertaining to vacation destination choices where the labels are site names.
Huybers’s study is the only published study we are aware of that uses site names as the
labels of a choice experiment and compares those results to a generic choice experiment.
While certain applications of choice experiments may mandate the use of either generic or
labeled alternatives, in other cases it may be left to the discretion of the researcher. In these
cases, the literature offers little guidance on the expected effects of such a design decision.
The impact of including labels in a choice experiment regarding preference for
environmental quality also has implications for possible benefit transfer. Research supports
9

the notion that similarity in site-context is a significant factor in the transferability of values
from one site to the next (Johnston 2007). The transferability of the values and preferences
elicited in this study would be considered higher (lower) should the results depend less
(more) on the presence of labels. If preferences remain consistent in the presence or
absence of Great Lakes names, we hypothesize the transferability of the preferences from
this study site to other policy sites that do not share the same label would be higher than if
preferences depended on whether alternatives were labeled or not. The author is not aware
of other choice experiments that test for transferability of results based on the degree of
similarity between labeled (alternatives named by site) designs and generic (unnamed)
designs.
Results show that the preferences elicited under the different labeling schemes are
similar in rank and magnitude, while Log-Likelihood Ratio tests reveal significant differences
in parameters across the labeled and unlabeled experiments as well as the same-labeled
and labeled experiments, while no significant difference was found in the parameters
estimated from the unlabeled and same-labeled experiments. However, these LLR tests do
not take into account differences in the scale factor (i.e., the unidentified variance) specific
to each labeling scheme’s different data set. To control for differences in scale, we compare
the MRS calculated from the different labeling schemes and find that in general, the average
MRSs are highest in the labeled scheme. We use a Krinsky-Robb procedure with 10,000
iterations to simulate the 95% confidence interval around the MRSs. Comparison of the 95%
CI of the MRS shows that there is no statistically significant difference in welfare measures
(economic tradeoff information) calculated across labeling schemes. This result is
encouraging for the practice of benefit transfer: similarities in the MRS from labeled and

10

unlabeled choice experiments helps build confidence in the transferability of values from
study sites where labeled choice experiments have been used at policy sites that may be
different from the sites in the original study.

11

CHAPTER 1: Preferences and Values for Changes in Water Quality at Great Lakes
Beaches in Michigan
Introduction
The Laurentian Great Lakes provide Michigan with environmental services that
support a broad range of public human uses including fishing, boating, and sightseeing
along with diverse private industry. With over 3,000 miles of Great Lakes coast along Lake
Superior, Lake Michigan, Lake Huron, Lake St. Clair and Lake Erie, Michigan enjoys the
most Great Lakes coastline of any state (MDEQ 2012b). The Great Lakes coast of Michigan
features approximately 600 public beaches along with countless private beaches (MDEQ
2012a). Visiting Great Lakes public beaches is one of the most popular outdoor activities
among Michigan residents, with more than 50% of respondents to a statewide mail survey
reporting having visited a public beach on the Great Lakes during the two prior swimming
seasons (Lupi et al. 2012).
Threats to the resource: Human Development
Despite their popularity and importance, the quality of the Great Lakes ecosystem
and the associated human uses constantly come under threat (IJC 2011a). The health of
natural systems within the Great Lakes as well as the associated human uses is greatly
affected by human development of Great Lakes shoreline. Shoreline development
eliminates not only public access for swimming, fishing, boating, and other outdoor
recreation, but also eliminates aquatic habitat for biota. Point source and non-point source
pollution from human development near the coast of the Great Lakes also degrades water
quality. Human development in coastal areas causes increases in pollutants from residential
uses (phosphorous, lawn fertilizers, leaky septic systems, and improper hazardous material
disposal), agricultural practices (chemical fertilizers, spread manure), and proximate
12

commercial and industrial activity (Environment Canada and U.S. EPA 2009). A summary of
the complexity of issues in dealing with the impairment to the Great Lakes caused by nonpoint source pollution can be found in IJC (2011a).
Threats to the Resource: Harmful Bacteria
The presence of pathogens and harmful bacteria such as E. coli in the waters of
Great Lakes threatens the public’s enjoyment of Great Lakes beaches. E. coli enters the
Great Lakes through natural processes and manmade sources that are each difficult to
avoid or mitigate (U.S. Policy Committee 2002). Manmade sources such as municipal
sewers or septic systems can leak feces containing E. coli into water ways connected to the
Great Lakes, while waterways receiving runoff from agricultural areas or other land where
manure is spread can also carry E. coli into the Great Lakes. Population centers with sewer
systems that combine waste and storm water can overflow following wet weather events,
leading to Combined Sewer Overflows (CSOs) which can carry untreated wastewater and
effluent containing E. coli into the Great Lakes (MDEQ 2011). Projects across the country
are underway updating sewer systems to avoid CSOs and can be very costly (ibid; U.S.
EPA 1999).
When visitors to the Great Lakes are exposed to high levels of harmful bacteria they
can suffer a range of negative health impacts including skin irritation, cold-like symptoms, or
diarrhea (Rose et al. 1999). Due to the health risks involved with exposure to E. coli at
beaches, federal, state, and county health officials have developed state-wide monitoring
plans that test for harmful bacteria and pathogens and inform the public about
contamination at beaches both on the Great Lakes and on inland lakes in Michigan (MDEQ
2012a). Without testing, beach goers are unable to make informed decisions about the

13

health risks associated with swimming. Information about testing and closures is available
1

online and through local media outlets. As of 2011, 262 of Michigan’s approximately 600
Great Lakes public beaches are tested at least monthly for E. coli, leaving nearly 60% of
public beaches on the Great Lakes untested (ibid). In 2011, 101 of the 262 Great Lakes
beaches monitored for bacteria, or nearly 39%, reported an exceedance of safe bacteria
levels at least once during the peak swimming season. Beach advisories or beach closures
caused by high levels of bacteria are not only an inconvenience to recreators, but can also
prove costly to local communities whose businesses and economies depend on tourists and
visitors to Great Lakes (NRDC 2012).
Threats to the Resource: Nuisance Algae
The presence of nuisance algae in the Great Lakes has been a long-running
challenge for resource managers (IJC 2011b; Higgins et al. 2005). The management of
nuisance algae in the Great Lakes raises a set of difficult questions linked to concerns about
public health, invasive species, and climate change. Large blooms of naturally occurring
species of algae, such as cladophora, can grow throughout the Great Lakes and eventually
wash onto shore. The greenish-brown blooms form decaying mats on the shore sometimes
several inches thick (MI Sea Grant n.d., Harris 2005). These mats of decaying algae,
commonly called muck, can greatly diminish beach aesthetics, obstruct passage along the
shore, and produce powerful and unpleasant odors, thus fouling stretches of beaches for
residents and visitors (Verhougstraete et al. 2010, SBSCPWG 2007, Higgins et al. 2005).
Recent studies have shown increased incidents of muck along the shore of the Great Lakes
1

Michigan’s Great Lakes beach testing and beach action information is available through
the Michigan Department of Environmental Quality’s BeachGuard website:
http://www.deq.state.mi.us/beach/
14

across the Great Lakes basin (Auer et al. 2010, Higgins et al. 2008). Research has also
shown that muck can harbor pathogens and harmful bacteria (Vanden Heuvel et al. 2010;
Verhougstraete et al. 2010; Olapade et al. 2006). The possible negative health concerns
associated with exposure to pathogens and bacteria in muck has caused county health
officials to issue contact advisories recommending avoidance of muck (Bay County Health
Department 2007). Muck also benefits from the presence of invasive zebra mussels that
increase water clarity, which encourages further growth of algae such as cladophora (Auer
et al. 2010, Higgins et al. 2008, Wilson et al. 2006). Algae in the Great Lakes grows at the
highest rates while water temperatures are higher during the summer months
(Verhougstraete et al. 2010, Higgins et al. 2005) leading some to hypothesize that levels of
muck in the Great Lakes could increase with warming trends as the result of climate change,
though current research predicts only marginal increases (Malkin et al. 2008).
Previous Research: Valuing Changes in the Great Lakes Environment
As one would expect, programs designed to protect the environmental health and
public uses of the Great Lakes are costly. While a wealth of information exists characterizing
the stresses placed on health of the Great Lakes and the associated human uses, there is
still the question of how to manage those threats in a cost effective way that still maximizes
the public’s enjoyment of the resource. Only once decision makers can view both the costs
alongside the benefits of programs to protect or restore the environment can an
economically sound course of action be taken. Within the field of environmental and natural
resource economics, a number of studies have sought to value changes in the
environmental conditions on the Great Lakes to better inform decision making (NMI and
NOAA 2001). These studies elicit the values and preferences of the public for changes in

15

the Great Lakes’ environment. These values can then be applied in policy settings where
the costs of programs that protect or improve the environment can be balanced against
public benefits of such programs.
Lupi et al. (1998) estimate a statewide model of recreational fishing demand to
measure the benefits to anglers that result from changes in fishing site characteristics such
as catch rates. Lupi, Hoehn and Christie (2003) estimate the benefits to anglers resulting
from a program to control invasive sea lamprey in the St. Mary’s River that would result in a
predicted rebound for the trout population in Lake Huron. The net present values of several
lamprey suppression programs, each with specific costs and trout recovery scenarios, are
all found to be positive, indicating that there is a net economic benefit to lamprey control in
the St. Mary’s River. Similarly, Kotchen et al. (2007) use a recreational fishing model to
show that the benefits of improved Great Lakes fish habitat that results from hydropower
dam relicensing exceed the cost of the operational changes.
Some studies measure the economic benefits of changes in environmental quality by
relating observed changes in environmental quality (such as water quality or air quality) to
the prices of nearby property, which is called the hedonic method. Ara et al. (2006) use a
hedonic pricing analysis of the value of houses near Lake Erie and find that changes in
water quality in Lake Erie have significant impacts on the value of nearby houses. Their
model predicts that an increase in water quality at Lake Erie beaches comparable to a 1
meter increase in secchi disk depth could increase housing values in that beach’s county
from $221 to $2379 per house, depending on the beach nearest to the home in question
(1996 USD). Ara et al. also estimate that homes in the area of beaches experiencing unsafe
levels of fecal coliform counts would benefit from a reduction in fecal coliform to safe counts

16

(i.e. 200 counts per 100 mL) in the amount of $88 to $2692 per house (1996 USD). Braden
et al. (2004) use a hedonic model as well as a choice experiment to estimate the benefits
from cleaning up contamination in the Waukegan Harbor in Waukegan, IL. The results of
each method are largely convergent, and the authors estimate the value of removing
contamination among residents of Waukegan Harbor to be 16% to 19% of the total values of
residents’ homes. Chattopadhyay (1999) used hedonic pricing to estimate the residents’
willingness to pay for air quality improvements in the largest Great Lakes coastal population
center, Chicago, finding that residents have a higher willingness to pay for reduction in
particulate matter (PM-10) than sulfur dioxide (S02). As a means of valuing proximity to and
aesthetics of Lake Erie, Seiler et al. (2001) use a hedonic pricing model of homes in Ohio
and estimate that all else equal, houses that have a view of Lake Erie are an average of
56% more valuable than houses that do not have a view of Lake Erie.
Whitehead et al. (2009) combine estimates from revealed preference (travel cost
recreation site choice) and stated preference (contingent valuation of hypothetical
restoration scenarios) to estimate the value of restoring Saginaw Bay wetlands. They
estimate the present value of benefits accrued to residents of the area from one acre of
Saginaw Bay marsh to be to be $2421 per acre (2005 USD) with the majority of that value
($1870) being derived from benefits to recreational users, and the remaining from passive
(non-use) values. Lupi et al. (2002) and Hoehn et al. (2010) use a choice experiment to
examine the value of wetlands in Michigan in terms of economic equivalency of wetland
services that are part of the replacement of impaired or eliminated wetlands required under
the Clean Water Act. The authors’ findings describe the public’s willingness to accept
wetlands of varying attributes as compensation for impaired or eliminated wetlands. In
17

general, respondents required increased amounts of wetlands when the restored wetland
was of lower quality in terms of habitat than the impaired wetland.
Previous Research: Beach Recreation
Public beach recreation has received a sizable amount of attention from resource
managers, policy makers and researchers. This review focuses on providing examples from
the literature on studies that examine not only the value of access (or the demand for) beach
recreation, but on the value of changes in beach characteristics or environmental quality
present at public beaches, as that is the focus of our study.
The value of changes in beach characteristics has been investigated at least since
McConnell (1977) studied the effect of congestion on visitation to beaches in Rhode Island.
Bockstael, McConnell and Strand (1989) use the results of a contingent valuation
telephone survey of residents of the Baltimore-Washington DC metro area to estimate
individual’s willingness to pay to improve water quality in the Chesapeake Bay to levels
acceptable for swimming. On average, those who reported having used the Chesapeake
Bay were willing to pay $121 annually and non-users $38 annually (1984 USD) for the
improvement in water quality.
Smith, Zhang and Palmquist (1997) employ a contingent valuation survey to measure
the economic value of controlling marine debris (natural or man-made) at recreational
beaches in New Jersey and North Carolina. While the authors note that study’s sample size
is too small for the results to be used to estimate population parameters, the results do show
that respondents consistently displayed higher willingness to pay and higher choice
probabilities for programs that resulted in lower amounts of debris versus programs that
resulted in higher amounts of debris.

18

Parsons, Massey and Tomasi (1999) use a random utility model to estimate the
demand Delaware residents have for marine (coastal) beaches within a day’s drive. Should
beach re-nourishment programs in the region be discontinued, and all beaches within the
choice set be reduced to widths of less than 75 feet, the authors estimate the loss to each
individual to be $7.25 per trip (assumed to be 1997 dollars, the year the study was
conducted), though the losses are less when considering losses only experienced at sites
reported as “favorites” by respondents. This result seems to indicate that beach width is a
less important to a person visiting his or her favorite beach.
Landry, Keeler and Kriesel (2003) use several methods to value additional beach
width along the beaches of Tybee Island east of Savannah, GA. First, they construct a
hedonic price model and estimate that all else equal a one meter increase in beach width
increases property value by an average of $233 (US$ 1996). Next, the authors implement a
choice experiment where respondents make tradeoffs between beach widths in the study
area and program costs in terms of parking fees. The authors estimate that on average,
households who visit beaches on Tybee Island were willing to pay between about $4 and
$11 per day at the beach for a one meter increase in beach width. Significant determinants
of willingness to pay for an increase in beach width included whether or not the respondent
owned an annual parking pass and the mode of beach management used to achieve
additional width (e.g. respondents exhibited higher WTP for achieving increases in beach
width under nourishment programs versus programs that achieved wider beaches through
2

shoreline retreat ).

2

As described by Landry et al. (2003), shoreline retreat allows for natural shoreline erosion
without nourishment. As the beach erodes, structures nearing the receding shoreline are
demolished or relocated. Beaches would be allowed to migrate inland.
19

Shivlani, Letson and Theis (2003) surveyed visitors to Southern Florida coastal
beaches, including a contingent valuation question regarding whether or not the respondent
would pay additional parking fees for a beach nourishment program. Depending on the
version of the survey, respondents were either told the beach nourishment program would
increase recreational opportunities, or that the program would enhance habitat available for
turtle nesting. The authors found that respondents had a higher willingness to pay to
increase turtle habitat ($2.12 per trip) than for improved recreational opportunities ($1.69 per
trip).
Lew and Larson (2005) surveyed residents of San Diego County, gathering
information on their most recent trip to a coastal beach. Using this trip information, the
authors estimate a travel cost model and calculate implicit prices for different beach
characteristics. The authors find that on average, residents of San Diego County are willing
to pay about $9 per trip for the presence of on-beach lifeguards, and $4.25 per trip to avoid
a beach that has a cobblestoned, or sand-denuded shore (2004 USD). Interestingly, their
3

model found that water quality (whether or not there was a beach water quality posting the
day of the respondent’s most recent trip or during the week before the respondent’s trip) did
not have a significant impact on beach choice. The authors hypothesize that visitors’ beach
choice may not be influenced by water quality since some individuals may not plan on

3

The authors refer to beach water quality postings both as postings of water quality
violations and as postings of beach closures. This would indicate that water quality postings
could signify either a type of contact advisory or a beach closure.
20

coming into contact with the water, or that those who do plan on entering the water will enter
4

the water regardless of conditions during their trip .
Hilger and Hanemann (2006) use a finite mixed logit model to account for
heterogeneity in water quality preferences across groups of individuals and across different
seasons. The authors gathered beach visitation data from 595 households in Southern
California in 2001 and modeled beach choices, using travel costs for each individual as the
price for each beach. Willingness to pay for a one grade improvement in water quality
(grades are based on the results of recent and past bacteria tests) at Southern California
beaches varied from negative values to over $17 per trip, with an average of $5.71 per trip.
Huang, Poor and Zhao (2007) sampled randomly selected households in New
Hampshire and Maine about their preferences for different coastal erosion control programs.
The researchers designed the study to include respondent’s values for both the benefits of
erosion control programs such as maintained beach width, as well as potentially harmful
effects of erosion control programs, such as erosion on neighboring beaches, decreases in
water quality, or disturbances to wildlife habitat. Results show the least desirable coastal
erosion program would be one that caused a disturbance to wildlife habitat, while
respondents preferred programs that saved beaches versus programs that protected beach
front houses. Not all respondents visited coastal beaches, though the number of trips a
4

Additionally, while the authors do not report how aware respondents were of beach water
quality postings, the authors purport the finding of water quality insignificance to “support the
idea that beach users may not have much knowledge about current beach water quality
postings,” perhaps indicating that beach postings did not affect beach choice because
respondents were not aware of them. The authors also comment that water quality postings
are only one aspect of how individuals may form opinions of water quality, adding “Other
indicators, such as the amount of trash on a beach, or objective water quality measures
provided through the media … may provide better insights into how water quality may affect
choices between beaches” (Lew and Larson 2005).
21

respondent took to the beach did not significantly influence preferences for erosion control
programs.
Previous Research: Great Lakes Beach Recreation
While there is some existing literature on the valuation of Great Lakes environmental
resources and more literature on the valuation of marine beach recreation, examples of
which we described above, we know of only one peer-reviewed journal article that studies
values and preferences for different characteristics of Great Lakes beaches. Murray,
Sohngen, and Pendleton (2001) valued decreasing incidences of water contamination
(beach advisories) at beaches on Lake Erie. That study gathered information on Ohio
residents’ trips to beaches on Lake Erie during 1998 and found that reducing the number of
beach advisories at a given beach by one in a given recreation season would result in a
seasonal benefit of $28 for each Great Lake beach goer. In a separate technical report
stemming from a similar study, Sohngen, Lichtkoppler and Bielen (1999) use a travel cost
model to estimate the mean value of a trip to beaches at Maumee Bay State Park and
Headlands State Park to be approximately $25 and $15 respectively (1998 USD).
Unpublished research by Egan and Dwyer (2008) uses a small sample (n= 178) of
visitors to the Maumee Bay State Park in northwest Ohio to investigate the value of wetland
restoration that would result in the removal of all swim advisories at the site. Using a
contingent behavior approach, the authors estimate that on average, respondents would
increase their trips to Maumee Bay State Park by 37% (from 3.8 trips per year to 5.2 trips
per year), with an inferred economic value of $147. The survey also included a contingent
valuation referendum question where respondents were asked whether or not they would
vote for a program that would restore local wetlands, improving water quality on site and

22

thus eliminating swimming advisories. Analysis of the response to the referendum question
yields a median willingness to pay for wetland restoration to eliminate swim advisories of
5

$375 per year .
In findings presented to several conferences, Shaikh (2005 and 2012) estimates a
travel cost model of visitors to Lake Michigan beaches in Chicago and finds that visitors
have an average value per trip of $35. Included in the study was a contingent valuation
question where survey participants were asked whether or not they would vote for an
increase in income tax that would provide better sewage treatment and decrease the total
swimming bans by 50% each year. Depending on the survey version, the increase in
income tax ranged from $10 to $100 per year, and respondents either faced a dichotomous
choice (vote yes or vote no) or a polychotomous choice (five choices: definitely vote no,
probably vote no, not sure, probably vote yes, definitely vote yes). Different coding schemes
were used to attribute polychotomous choices as votes for or against the program as a
means of dealing with varying degrees of respondent certainty. As reported in Shaikh
(2012), an analysis of the responses yielded a willingness to pay for a 50% decrease in
beach closures of approximately $38 to $65 per person per year, depending on
assumptions made.
Purpose
Surprisingly, even with millions of visitors each year and nearly 600 public beaches
on the Great Lakes in Michigan, resource managers lack information on residents’
preferences for different levels of environmental quality at Great Lakes beaches. This paper
5

The formal statistical models used to develop the willingness to pay for wetland restoration
from both the contingent behavior and the contingent valuation were not discussed in the
available manuscript.
23

uses a choice experiment to elicit the values and preferences that Michigan residents have
for different environmental quality characteristics at Great Lakes beaches. Information on
these values and preferences helps fill the gap that exists between knowledge of the
stresses to environmental quality of Great Lakes beaches and how those stresses affect the
enjoyment of that resource by the public. Information on Michigan residents’ values and
preferences for changes in Great Lakes water quality conditions will provide a starting place
for resource managers and policy makers: such information could help prioritize further work
to study the implementation of Great Lakes environmental protection programs such as the
Great Lakes Restoration Initiative, a Federal program that has invested $1.075 billion from
2010 to 2012, with an additional $300 million authorized for 2013 (U.S. EPA 2009, 2010,
2011, 2012a,). Scientifically sound preference information could be used as a basis to justify
revisiting recommended best management practices or current industry standards that are
known to still cause negative impacts to environmental quality in the Great Lakes. Resource
managers could eventually use information on the public’s preferences for different levels of
environmental quality when quantifying the damages caused by environmental disasters or
benefits of projects that restore injured natural resources.
The next section of this paper explains the methods of choice experiments and the
underlying statistical model. It then describes the choice experiment we implemented as part
of a Great Lakes beaches web-based survey. The results are then presented, followed by a
discussion.
Method: Choice Experiment
The choice experiment format is a well-established tool that has been used by
marketing researchers to measure consumer preferences for different levels of attributes in

24

a given product (Louviere, Hensher, Swait 2000). The practice was then popularized within
the field of resource and environmental economics where the method is applied to goods for
which no markets exist (Hanley, Wright and Adamowicz 1998). In a choice experiment,
participants are shown two or more alternatives that are comprised of different levels of
attributes and the respondents are asked to pick their preferred alternative. Figure 1.1
shows an example of a choice used in our study. The underlying assumption of a choice
experiment is that when faced with two or more alternatives containing varying levels of
attributes, the respondent will choose the alternative that leaves them the best off. The
choices made by respondents reveal tradeoffs between the levels of attributes presented in
each of the alternatives. When researchers include cost as one of the attributes of the
alternatives, respondent choices are subject to income constraints, as well as the restriction
that they can only choose one of the alternatives. In this way, the choice experiment mimics
choices made by consumers in a market: respondents in a choice experiment are forced to
tradeoff a finite resource (income) while maximizing their enjoyment (utility) of a product
composed of several attributes. As a result of modeling this constrained utility maximization
problem, researchers can estimate the demand for the attribute levels included in the choice
experiment, as well as measure the tradeoffs individuals make between levels of different
attributes.
The attributes and the levels of those attributes that comprise alternatives presented
in a choice experiment are determined by the researcher. Deciding which attributes to
include is a complicated task. Researchers must be sure to include enough attributes to
describe the good in question so as to present a choice set that is representative of realworld conditions. However, researchers must keep from making the choice task too

25

Figure 1.1: Image of example choice set from the Great Lakes choice experiment, original
size monitor-dependent (For interpretation of the references to color in this and all other
figures, the reader is referred to the electronic version of this thesis.)
Visiting a Great Lakes Beach
Suppose you are taking a trip to the beach and there are only two beaches to choose from.
The beaches have different characteristics as shown in the table, but otherwise, they are the
same. For example, they would have the same amount of litter, the same amount of
crowding, and the same scenery. Please compare Beach A and Beach B in the table and
answer the question below:
Beach A

Beach B

Great Lake

Lake Huron

Lake Michigan

Bathrooms

Flushing toilets
cleaned hourly

Flushing toilets
cleaned hourly

Algae
in the water

Moderate
(occasionally come in
contact with algae)

Low
(rarely come in
contact with algae)

Algae
on the shore

Low
(1-20% of the
shore has algae.)

None

Length of beach

5280 yards (3 miles)

880 yards (1/2 mile)

Testing water
for bacteria

Daily

Weekly

Distance from
your home

48

169

Which of the above
beaches would you visit?

Beach A

26

Beach B

burdensome for the respondent by including too many attributes. Since there may be many
determinants of an individual’s enjoyment of a good included in a choice experiment,
researchers typically conduct focus groups or cognitive interviews with members of the
sample population to be sure the attributes and attribute levels included in the choice
experiment are meaningful to respondents and that information within the survey is not
overly technical or burdensome. Survey forms are also developed carefully and tested to be
sure that the information provided is complete, and minimizes any tendencies for
respondents to make inferences about missing information which could obscure results.
Once the attributes and their levels have been defined, the researcher then
generates a list of all the combinations of attribute levels that will appear within the choice
sets shown to respondents, known as the experimental design. The levels of attributes
included in choice sets is varied in such a way that minimizes unwanted correlation between
attribute levels and allows the researcher to properly identify the effect that each attribute
level has on the likelihood that a respondent chooses an alternative.
Random Utility Theory
Underlying the choice between alternatives is the utility the respondent would
experience from choosing each alternative. The statistical model used to represent the
choices made by choice experiment participants is based on random utility theory
(McFadden 1974). Respondents review the attributes and attribute levels listed for each
alternative in the choice experiment and choose the alternative that maximizes their utility.
The indirect utility function is comprised of two main components: a deterministic
(observable) component and a stochastic (unobservable) component. The deterministic
component is a function of the attributes that comprise each alternative, the individual’s

27

characteristics, and unknown parameters. The stochastic component is an error term. Per
Louviere et al. (2000) and Alberini et al. (2006), this relationship is stated formally as:

̅(

)

(1)

for every i individual and every j alternative; x is a vector of attributes that take on various
levels for each alternative within a choice set, β is a vector of unknown parameters
associated with those attributes, and

is an error term capturing factors specific to the

alternative and to the individual that affect the respondent’s utility but that the researcher
cannot observe. The deterministic portion of the indirect utility function is commonly
specified as a linear function of the attributes of the alternative and the respondent’s income
(y) less the cost of the alternative (C):

(
where the coefficient

)

(2)

represents the marginal utility of income because yi –Cij represents

money person i has left over to spend elsewhere after alternative j was chosen at cost Cij.
In our application, we ask respondents to review the attributes of two Great Lakes
beaches and then choose which beach they would visit. For this study, we do not use a
strictly monetary cost component, but rather use distance from home as the cost to the
respondent to enjoy the beach alternative of their choosing. By using distance instead of
monetary value, the choice situation presented in this choice experiment is consistent with
real trip taking behavior, where each respondent would face a trip from home with an
associated travel distance to the beach rather than a dollar amount they must pay in order to
enjoy the beach. The marginal utility of additional miles away from home could be converted
to a monetary measure by defining a relationship between miles driven and money.
28

However, this paper leaves the “cost” of the alternatives, and therefore the costs of changes
in attribute levels, in terms of miles driven. Future analysis of web survey data will focus on
translating information on actual trips taken into economic values for beach attributes in
terms of dollars.
The appropriate model for the outcome of the choice experiment with two alternatives
is a discrete choice model describing the probability that an alternative, A, is chosen over
alternative B by an individual, i.
(3)
Which can be expanded as:

(4)
And it follows:
(5)
We see in equation 5 that the probability of selecting an alternative does not depend on
attributes that remain constant across the alternatives, such as income (y), the intercept
term (

0),

or other characteristics of the individual. Rather, the probability that a respondent

selects an alternative depends on the differences in the levels of attributes for a given
alternative relative to the levels of attributes that appear for other available alternatives.
When it is assumed that the errors in the model are independently and identically
distributed (IID) and follow a type 1 extreme value distribution, a conditional logit model can
be used to estimate the probability of a respondent’s choice (Louviere, Hensher and Swait

29

2000, Alberini et al. 2006). McFadden (1974) shows that the choice probabilities within the
logit model are equal to:
(6)

Where µ is scale parameter that is inversely proportional to the model’s overall
variance (Ben-Akiva and Lerman 1985). The scale parameter cannot be separated from the
estimated parameters, β, and so it is commonly normalized to 1 in order to allow
researchers to identify the preference parameters.

In the case of choice experiment where respondents are faced with more than one
choice set, researchers can employ a random-effects logit to control for correlation in error
terms across responses from the same respondent (Wooldridge 2010). Our choice
experiment shows respondents three choice sets, thus the estimation of a random-effects
logit is appropriate.
Using the results of the logit model, researchers can estimate the tradeoffs that
respondents make between different levels of the attributes. In economics, these tradeoffs
are referred to as the Marginal Rate Substitution (MRS) and are commonly put in terms of a
cost parameter, or in our case, distance. In general terms, holding all other attributes
constant, the MRS is equal to the change in one attribute, X1, required to compensate the
individual for a one unit change in another attribute, X2 (i.e. the amount of X1 required to
keep the individual at the same level of utility as before the one unit change in X2). In our
case, we calculate the MRS in terms of miles from home. The MRS assumes there is a

30

change in environmental quality at a beach and estimates the additional miles further from
home (if the MRS is positive) or the number of miles closer to home (if the MRS is negative)
that the beach would have to be in order for an individual to be indifferent between the
beach before and after the change in environmental quality. Equation (7) illustrates how to
calculate MRS as the opposite of the ratio of the partial derivative of the indirect utility
function with respect to the first attribute of interest and the partial derivative of the indirect
utility function with respect to the second attribute of interest Equation (8) shows an example
MRS calculation for moderate algae in the water in terms of driving distance specific to our
application.

( )

(7)

(

) (8)

For example, an MRS of 75 for a change in algae in the water from high to moderate is
interpreted to mean that all else equal, a respondent would be indifferent between travelling
some distance to a beach with high amounts of algae in the water, or travelling that same
distance plus an additional 75 miles to a beach with identical characteristics except for
moderate amounts of algae in the water instead of high amounts.
Survey Development: Pretests
This study reports on the findings of a choice experiment implemented as part of the
Great Lakes beach web-based survey in the spring of 2012. The web survey was developed
and questions were tested in the spring of 2012 using an iterative process that was guided
by the results of 57 one-on-one cognitive interviews (Kaplowitz, Lupi and Hoehn 2004). The
31

57 interviews were composed of eight interviews conducted among a convenience sample
of Michigan State University students at a campus food court, 19 interviews conducted in
person among randomly selected adults at local shopping malls, and 30 interviews
conducted remotely with respondents recruited from a web survey panel (Survey Sampling
International) of adults in the lower peninsula of Michigan.
The remote interviews were conducted over the phone using an innovative approach
not previously documented in the literature on the pretesting of web-based valuation
surveys. At the start of each remote interview, the lead researcher called the participant at a
scheduled time into a shared conference call line, and then emailed the participant a link to
6

web-based screen sharing application called “GatherPlace.” The screen sharing application
allowed the participant to view and control the screen of the host computer belonging to the
lead researcher which was configured to show the web survey. The use of the shared
conference call line and screen sharing application allowed other members of the research
team to simultaneously listen to the interview as well as observe the participant navigating
the web survey without creating any disturbance. After calibrating the screen sharing
application to suit the display settings on the participant’s computer, the lead researcher
would grant control of the host computer showing the web survey to the pretest participant.
Once the participant had become accustomed to controlling the host computer, the lead
researcher instructed the participant to work through the survey to the best of their ability.
Following an outline of interview questions while probing as necessary on given
issues as they arose, the lead researcher would gather information related to the
6

The application ran using Java, a program that is frequently used by websites. Since Java
is popular among a wide range of websites, most internet users already have the program
installed. We found that 25 of the 30 remote interview participants had a working version of
Java already installed on their computer.
32

respondent’s perceptions, understandings, and opinions of survey tasks and information
presented in the survey. Pretest participants were asked detailed questions about their
understanding of the definitions of attributes, attribute levels, and their interpretation of
questions pertaining to the attributes to be sure that the survey clearly communicated the
intended themes. Pretests also focused on ensuring the respondents were able to answer
questions according the information provided and, where applicable, their personal
experience, attitudes, or opinions. After each interview was complete, research team
members debriefed as necessary to address issues raised by a participant’s questions or
difficulties, and programming changes were made to the survey instrument as soon as
possible to then test those edits in subsequent pretests.
This remote pretesting approach proved advantageous since it allowed several
research team members to observe survey pretests without “hovering” or “crowding” around
a single participant at a research facility, perhaps making it easier for the participant to focus
on the tasks and honestly answer questions. The method was also inexpensive, with the
only costs being a small monthly subscription to the screen sharing service and low-cost
conference calling. Another key advantage this remote pretesting had over onsite pretests
was that it allowed pretest participants to be drawn from across the Lower Peninsula of
Michigan without requiring research team members to travel. This geographic
representativeness helped capture opinions on the survey from individuals living adjacent to
the Great Lakes who visit Great Lakes beaches nearly every day as well as those who live a
few hours from the nearest Great Lake beach and may visit only a few times each year.
Capturing opinions about the survey from individuals with a range of experiences with

33

visiting Great Lakes beaches was a key to our understanding of how well the survey
instrument functioned.
Pretests were halted once it was clear that respondents comprehended survey tasks
and that survey information was clearly communicated to respondents.
Once pretests and initial web-survey debugging were complete, a pilot group of 85
individuals were mailed invitations to the web survey. In the two weeks following the initial
invitations, pilot study members were mailed two reminders: a half-sheet postcard and a
quarter-sheet postcard, each listing the survey’s web-address, the individual’s password,
and information on who to contact with questions. The pilot study received a total of 22
logins for a response rate of 25.9%. The pilot survey allowed the research team to simulate
the process of actually implementing the survey (e.g. timing of mailing, receiving data) and
to see if any questions arrived via phone or email from pilot survey respondents.
Survey Development: Input from experts
The text, diagrams, and graphics used to describe the attributes to respondents were
developed with the help of state and county health officials, water policy experts, as well as
resource managers from State and Federal agencies including NOAA Great Lakes
Environmental Research Laboratory, Michigan Sea Grant, EPA Great Lakes Program
Office, Michigan Department of Natural Resources, and county officials that perform water
quality tests and conduct beach algal assessments. Parts of the choice experiment needed
to convey scientific information in a way that was understandable to members of the general
population who are not assumed to have advanced knowledge of the topics in the survey.
The resource experts we consulted were able to ensure that the survey communicated
accurate information.

34

Certain attribute descriptions shown to choice experiment participants also contained
diagrams and drawings. Figure 1.2 and figure 1.3 show diagrams that accompanied
definitions of attribute levels for amount of algae on the shore and amount of algae in the
water, respectively. The use of photos to describe attributes and their levels was considered
but research shows that unless closely controlled, photos can communicate unintended
information (Hoehn, Lupi and Kaplowitz 2003). Even though
Figure 1.2: Image of diagram of Great Lakes beach attribute: Algae in the water, original
size monitor dependent.
Amount of
algae in
the water

Definition

View of the Water

None

visitors never come in contact
with algae while swimming or
wading

Low

visitors rarely come in contact
with algae while swimming or
wading

Moderate

visitors occasionally come in
contact with algae while
swimming or wading

High

visitors constantly come in
contact with algae while
swimming or wading

35

Figure 1.3 Image of diagram of Great Lakes beach attribute: Algae on the shore, original
size monitor-dependent
Amount of
algae on the
shore

Definition

View of swimming area shore

None

None of the shore of
the swimming area has
algae.

Low

1 to 20% of the shore of
the swimming area has
algae.

Moderate

21 to 50% of the shore
of the swimming area
has algae.

High

More than 50% of the
shore of the swimming
area has algae.

36

the diagrams are simplifications of real-world conditions, the experts we consulted in the
development of survey materials and pretest participants (i.e. perspective survey
respondents) found the diagrams to represent conditions experienced at actual Great Lakes
beaches.
Survey Sample
Because there is no list of beachgoers to facilitate sampling, a two-step sampling
process was used. First, in the summer of 2011, a general population mail survey on
participation in leisure and recreation activities was conducted using a sample of 32,230
residents of Michigan’s Lower Peninsula, 18 years and older, drawn from the drivers' license
list. The mail survey contacts followed a modified version of Dillman's tailored design
method (Dillman 2009), and the survey achieved a response rate of 38%. Among the
activities asked about in the mail survey was whether or not the respondent had visited a
Great Lakes beach in the last year. Second, in the spring of 2012, invitations to a Great
Lakes beaches web-based survey were sent to 5,434 qualifying respondents from the mail
survey (i.e. those who reported having visited a beach on the Great Lakes in Michigan,)
again following a modified version of Dillman's tailored design method and achieving a
response rate of 59.6%.
Web Survey Implementation
Web survey sample members received a maximum of five contacts over a month
and a half-long field period (from April 14 to June 1, 2012): an invitation letter, a 1/2 sheet
reminder postcard (in black and white), a quarter-sheet postcard (in color), an automated
phone call reminder (robodial) to respondents with publically available land-line phone
numbers (2,469 sample members) were called during the evening on April 27, and a final

37

letter offered a post-paid incentive of $10 or $20 to those who responded to the survey by
7

the end of the field period (May 29) . A small control group was mailed a final letter that did
not offer a post-paid incentive. Each final contact letter included a business reply mail
postcard instructing respondents to send the postcard back only if the individual did not
have access to the Internet. Each mailing listed the survey web-site and provided the
respondent with a unique password to use when logging onto the website, as well as a
phone number and email address to contact if there were any questions.
The web survey included three main sections: a section gathering detailed
information about trips to Great Lakes beaches in the last year for use in revealed
preference modeling, a choice experiment section, and a section of background and
demographic questions. This paper deals with the results of the second section, the choice
experiment.
The choice experiment section began with separate pages explaining each of the
attributes. Attribute descriptions were developed with the help of the aforementioned
resource experts. Thus, each attribute description was followed by questions engaging the
8

respondent with the information presented about the attributes. Follow-up questions asked
about the respondent’s personal experiences and attitudes towards the attributes. Each
respondent’s path through the choice experiment presented information on the attributes in
question in a controlled format, and the follow-up questions allowed respondents to interact
with information on the attributes, a technique that survey research shows increases

7
8

See appendix C for copies of all materials sent to members of the web survey sample.

See appendix A for set of screen captures from the choice experiment portion of the web
survey, complete with informational treatments, warm-up tasks, choice sets and follow-up
questions.
38

respondent’s comprehension of survey tasks (Kaplowitz et al. 2004, Hoehn and Randall
2002, Schwarz and Sudman 1996).
Following descriptions of the attributes, each respondent was shown three different
choice sets. Figure 1.1 shows an example of a choice set that was used in our study. The
choice experiment asked respondents to assume they were taking a trip to a Great Lakes
beach, and the only two alternatives available were the beaches shown in the choice sets.
Respondents were told the alternatives had the different characteristics described in the
table, but otherwise were same. Respondents were then instructed to review the attributes
in the table and then select which beach they would visit. The attribute names in the choice
set were hyperlinked so that respondents could click on the attribute name to open a new
window displaying the same attribute definition shown earlier in the survey for the
respondent to review. Beaches in the choice sets were described using a set of seven
attributes. Attributes were selected for our study based on their salience and relevance to
Great Lakes beach visitors, with a focus on beach attributes that could be tied to
environmental management schemes, not physical characteristics or facilities.
Attributes and Attribute Levels
The attributes used in the choice experiment were as follows: Great Lake (what Great
Lake the beach lies on), bathrooms (the type of bathrooms available at the beach), amount
of algae in the water, amount of algae on the shore, length of beach, the frequency of
bacteria testing, and the distance away from home. Table 1.1 shows the attributes and their
levels.

39

Table 1.1: Attributes and Attribute Levels included in the Great Lakes Choice
Experiment
Attribute Name
Great Lake

Bathrooms*
Algae in the water

Algae on the shore

Length of beach

Testing water for
bacteria

Attribute level
Michigan
Huron
St. Clair
Erie
Flushing toilets, cleaned daily
Flushing toilets, cleaned hourly
None
Low (rarely come in contact with algae)
Moderate (sometimes come in contact with algae)
High (constantly come in contact with algae)
None
Low (1-20% of the shore has algae)
Moderate (21-50% of the shore has algae)
High (more than 50% of the shore has algae)
50 yards
220 yards (1/8 mile)
880 yards (1/2 mile)
1760 yards (1 mile)
5280 yards (3 miles)
Never

Monthly
Weekly
Daily
Distance from your
Individual’s minimum distance to selected Great Lake
home
+ 0 miles
+ 15 miles
+ 40 miles
+ 100 miles
+ 150 miles
* Attribute was held constant within choice sets to control for preferences for different
bathrooms, but no preference or value information was gathered.
For the levels of the Great Lakes attribute, we focused on the Great Lakes that are
adjacent to Michigan’s Lower Peninsula, where we drew our sample from. Bathrooms were
included in the choice sets but were held constant within choice sets. In pretests, we found
40

bathrooms were a salient attribute when choosing the preferred beach, but facilities at the
beaches (including playgrounds, picnic tables, and bathrooms) did not fit into our study’s
focus on environmental quality. Instead of including bathrooms as a variable in our choice
experiment, we included it as an attribute that would remain constant within choice sets.
This kept respondents from making choices based on inferred levels for a missing bathroom
attribute based on existing attributes (e.g. “Lake Michigan has cleaner bathrooms”). With the
bathrooms attribute being held constant, bathrooms would not impact the individual’s
decision. The attribute levels for algae in the water and algae on the shore were chosen to
correspond directly with data currently being gathered by the EPA Great Lake Beach
Sanitary Survey (U.S. EPA 2008). By using the same scale as the EPA sanitary survey, we
can combine information on which Great Lakes beaches people visited and how often along
with what the levels of algae were on the shore and in the water at those beaches (revealed
preference data) with the marginal rates of substitution for changes in algae levels to
estimate the economic costs or benefits associated with changes in the levels of algae.
Within the choice experiment, we define beach length as the distance along the Great Lake
9

within the boundaries of the park or managed area where the beach is located. We chose
the attribute describing the frequency of testing to have the levels of daily, weekly, monthly,
or none to represent examples of how frequently actual beaches are tested. As mentioned
previously, not every Great Lakes public beach is tested for E. coli. The final attribute,
driving distance contained two components. The first component was the minimum distance
from each individual’s home to the Great Lake that was specifically named in the choice
alternative (i.e. the minimum distance from an individual’s home to Lake Michigan when
9

Note, findings from Parsons et al (1999) refer to preferences for changes in beach width,
measured in terms of the distance from dune toe to berm.
41

Lake Michigan was selected for the choice alternative, or the minimum distance to Lake St.
Clair when Lake St. Clair was selected for the choice alternative). Distances were calculated
as the minimum distance from the center of the web survey sample member’s zip code to
one of several beaches along each Great Lake using PC Miler. The second component of
distance was one of five levels of “additional miles” that were added as described in the
experimental design to the minimum distance to create the total distance. This total distance
was the only distance presented to respondents. The levels of the additional distance
variable were designed to allow distance to be estimated as a continuous variable in our
analysis. The distance attribute needed to control for the minimum distance from a
respondent’s home to a specific Great Lake since there is such wide variation in distances
across our sample: certain sample members live adjacent to some Great Lakes (i.e.
minimum distance of zero to the nearest Great Lake) while the same individual may live 150
or more miles from a different Great Lake. By controlling for the minimum distance from a
respondent’s home to the Great Lake within the specific choice alternative, we increase the
plausibility and realism of the choice alternative, and avoid showing the respondent
counterfactual attribute levels, a concern raised previously in literature regarding labeled
choice experiments (Huybers 2005, Carson et al. 1994).
The experimental design, which is the list of each combination of attributes that would
appear within the choice sets featured in the choice experiment, was generated using
Ngene software (Choice Metrics 2011). The experimental design was derived in such a way
that allows researchers to identify the impact each attribute level has on the probability of a
respondent selecting a given alternative (ibid, Johnson et al. 2006.) The study utilized an Fefficient design which minimized the variances of MRS calculations from the parameters

42

estimated from the results (Choice Metrics 2011). To derive the design, a researcher
inputted the number of alternatives in the choice experiment, as well as the attributes and
attribute levels. As it was possible, attributes were also assigned an expected sign, allowing
NGene to minimize the instances of dominated alternatives and further increase design
efficiency. NGene used a swapping algorithm that searched for new choice pairs to swap
into the existing design in order to increase overall efficiency. For the choice experiment
task reported here, 126 choice pairs were generated and randomly sorted into 42 groups of
three, with each respondent viewing one of the 42 groups.

43

Following the attribute levels used in the choice experiment (and described above)
equation 9 shows the logit model we are estimating:

(9)
44

Where each independent variable named with a specific attribute level is a dummy equal to
1 when that attribute level is present in the alternative being evaluated and equal to zero
when that attribute takes on any other level. All coefficients in the model are estimated
relative to the base case for the following beach attributes: algae on the shore equal to high,
algae in the water equal to high, beach length equal to 50 yards, testing for bacteria equal to
daily, and Great Lake equal to Lake Erie.
Results
The web survey obtained a response rate of 59.6% (3,211 individuals out of an
effective sample of 5390). A random subset of respondents to the web survey was assigned
10

the choice experiment task as described above, which included a total of 1040

individuals.

Of the 1040 assigned to the choice experiment task reported here, 988 filled out all three
choice sets presented to them, 43 filled out two, and 9 filled out one, for a total of 3059
observations. Table 1.2 shows the number of respondents and observations for the Great
Lakes beach web survey choice experiment reported here.
Table 1.2: Number of respondents and observations for Great Lakes beach web
survey Choice experiment
1209 individuals logged on to survey
1040 individuals answered at least one choice set within choice experiment
988 individuals with three choice sets answered
43 individuals with two choice sets answered
9 individuals with one choice set answered
3059 total number of choice experiment observations

10

One respondent to the choice experiment task reported here was inadvertently excluded
from the analysis due to a write-error performed by the web survey. The error made it
impossible to link the choice set viewed by the respondent to the respondent’s answers to
the choice experiment until further steps were taken after the analysis was conducted.
45

For our statistical analysis, we first estimate a random-effects logit model, controlling
for the fact that individuals answered up to three different choice sets. The model presented
here includes the main effects for all of the attributes and attribute levels included in the
choice experiment, and results are reported in Table 1.3. Note the baseline attribute level:
all preferences are measured as changes from the baseline level. For example, the
estimated parameter on “algae in the water: low” is for a change in algae in the water from
the baseline level, none, to low. The estimated model had a Wald chi-squared statistic of
705 (p<0.001), and a McFadden’s pseudo R-squared of 0.302 (meaning that on a scale of 0
to 1, the model we estimated has a higher likelihood or fits the data better than a model that
only contains a constant). The model fit, as judged by the predictive ability, is well-balanced
as it correctly predicted 77% of Beach A choices and 76% of Beach B choices (thus, the
model predicted the probability of a person choosing Beach A as being greater than 50% in
77% of the cases a respondent actually chose Beach A). Nearly all model parameters are
estimated to be statistically significant at the p<0.001 level. As expected, the estimated
parameter on driving distance is statistically significant and negative. The model reveals that
Great Lakes beach goers prefer less algae on the shore and lower amounts of algae in the
water, with algae in the water causing more of a nuisance than algae on the shore. We find
that Michigan residents also prefer longer beaches to shorter beaches, with no significant
difference in preferences between three mile long beaches (baseline) and one mile long
beaches. The model also shows that respondents prefer a beach that is tested more
frequently for bacteria versus a beach that, all else equal, is not tested for bacteria at all. We
also see that, all else equal, Great Lakes beach goers prefer beaches on Lake Michigan

46

and Lake Huron to beaches on Lake Erie (the baseline) and that beach goers are indifferent
between beaches on Lake St. Clair and Lake Erie.

Table 1.3: Results of random-effects logit model
estimating determinants of Great Lakes beach choice
Variable
coefficient
p-value
One-way distance
-0.008
<0.001
Algae on the shore: None
1.474
<0.001
Algae on the shore: Low
1.225
<0.001
Algae on the shore:
0.763
<0.001
Moderate
Algae in the water: None
1.768
<0.001
Algae in the water: Low
1.473
<0.001
Algae in the water: Moderate
1.111
<0.001
Length: 50 yards
-0.78
<0.001
Length: 220 yards
-0.421
<0.001
Length: 880 yards
-0.246
0.012
Length: 1760 yards
0.069
0.489
Testing: None
-1.496
<0.001
Testing: Monthly
-0.332
<0.001
Testing: Weekly
-0.381
<0.001
Great Lake: Michigan
1.194
<0.001
Great Lake: Huron
0.498
<0.001
Great Lake: St. Clair
0.004
0.966
*= significant at the 5% level
**= significant at the 1% or less level
Algae on the shore parameters are relative to "High"
Algae in the water parameters are relative to "High"
Length parameters are relative to "3 miles/5280 yards"
Testing parameters are relative to "Daily"
Great Lakes parameters are relative to "Lake Erie"

**
**
**
**
**
**
**
**
**
*
**
**
**
**
**

While we can judge relative preferences for beach characteristics from the estimated
logit parameters (i.e. the β’s), it is also informative to examine the trade-offs between
attributes implied by those parameter estimates. We examine the trade-offs by calculating
the MRSs for each attribute level. Table 1.4 shows the economic tradeoffs (MRSs)
47

estimated for the marginal changes in Great Lakes beach attributes relative to the baseline
attribute level. All trade-offs are in terms of additional miles from home. The table includes
95% confidence intervals around the mean MRS, which were calculated using the KrinskyRobb method with 10,000 draws (Krinsky and Robb 1986). To interpret the marginal rates of
substitution these would need to be translated into dollar values and placed into a realistic
choice set for beaches, that is, a choice set that includes a full range of substitute sites.
Instead, these welfare measures assume that each respondent would take a trip to one of
the two beaches they saw in the choice experiment, By not accounting for the real set of
substitute alternatives, these measures are an upper-bound for the actual economic value of
these marginal changes in beach attributes since respondents were not shown a full range
of substitutes, nor were they given a proxy for this such as a "do not go," or "go to some
other beach" option.
From the trade-offs calculated using our estimated model, we can see that while
higher amounts of algae on the shore and higher amounts of algae in the water are both
undesirable to respondents, the presence of algae in the water is even more of a nuisance
than algae on the shore. While the mean MRSs for testing for bacteria appear to show a
preference for monthly testing over weekly testing, the significant overlap between
confidence intervals indicate that for qualitative purposes, respondents are indifferent
between the two frequencies of testing. Both monthly and weekly testing for bacteria are
preferred to no testing at all. The MRSs reinforce the intuition that respondents would prefer
longer beaches to shorter ones, with respondents being indifferent between beaches that
are one mile versus three miles long, all other beach attributes being equal.

48

Table 1.4: Mean estimate of Marginal Rates of Substitution for Great Lakes
beach characteristics, including lower-bound and upper-bound estimates of
95% Confidence Intervals from Krinsky-Robb method, 10,000 draws
LowerMean
Bound
Algae on the shore: None
154
180
Algae on the shore: Low
125
150
Algae on the shore: Moderate
70
93
Algae in the water: None
188
216
Algae in the water: Low
154
180
Algae in the water: Moderate
111
136
Length: 50 yards
-121
-95
Length: 220 yards
-76
-51
Length: 880 yards
-54
-30
Length: 1760 yards
-15
8
Testing: None
-213
-183
Testing: Monthly
-63
-41
Testing: Weekly
-70
-47
Great Lake: Michigan
121
146
Great Lake: Huron
38
61
Great Lake: St. Clair
-21
0
Algae on the shore estimates are relative to "High"
Algae in the water estimates are relative to "High"
Length estimates are relative to "3 miles/5280 yards"
Testing estimates are relative to "Daily"
Great Lakes estimates are relative to "Lake Erie"
Attribute

UpperBound
210
178
119
248
209
162
-71
-29
-7
32
-157
-18
-25
173
85
23

The MRSs estimated from the model are consistent with what we would expect of
preferences for desirable or undesirable traits: respondents have higher MRS for “better”
levels of the attributes included in the choice experiment. These tradeoffs show that all else
equal, respondents are willing to travel further to beaches that have lower amounts of algae
in the water, less algae on the shore, and are tested more frequently for bacteria.

49

The estimated MRSs for beach length

11

are consistent with the law of diminishing

marginal utility which states that the marginal utility gained by additional units of
consumption is greatest for the first additional unit and decreases for each subsequent unit
consumed. The MRS for length of beach is largest for the initial change in beach length from
50 yards to 100 yards. The marginal utility of additional beach length diminishes on a per
yard basis until we estimate that respondents are indifferent between a beach that is 1 mile
long (1760 yards) and 3 miles long (5280 yards) all else equal. Note that this finding is
consistent with revealed preference studies that show a significant and non-linear
relationship between beach length and site choice (e.g. the natural log of length in Parsons
et al. 2009, or length plus length squared in Lew and Larson 2005) where the marginal
benefits of length decrease as beaches are longer.
The model also indicates respondents’ preferences for specific Great Lakes. The
finding that the parameters for Lake Michigan and Lake Huron are statistically significant
(p<0.001) shows that respondents infer certain characteristics about a beach based on
which Great Lake it lies on, independent of beach attributes and environmental quality
measures such as frequency of testing for bacteria and levels of algae. Compared to Lake
Michigan, the most preferred Great Lake, respondents would next prefer Lake Huron.
Preferences are the lowest for visiting beaches on Lake St. Clair and Lake Erie. The model
predicts that all else equal, a beach on Lake Erie or Lake St. Clair would have to be nearly
150 miles closer to an individual’s home than the same beach on Lake Michigan in order for
11

Note, our findings pertain to beach length, while previously mentioned marine beach
studies such as Parsons et al. (1999) refers to preferences for changes in beach width,
measured in terms of the distance from dune toe to berm. Our study measures preferences
for beach length defined as the distance along the Great Lake within the boundaries of the
park or managed area where the beach is located.
50

the respondent to have an equal probability of visiting the Lake Erie or Lake St. Clair beach
and the Lake Michigan Beach.
Conclusions
This paper provides the first insights into the values and preferences that Michigan
residents have for a popular and important resource, Great Lakes beaches, which has yet to
be fully studied. By capturing tradeoffs between Great Lake beach attributes through a
choice experiment, we are able to estimate the upper-bound of the trade-offs residents who
visit Great Lakes beaches would make for marginal changes in Great Lakes beach
attributes. The study shows that visitors to Great Lakes beaches first and foremost prefer
beaches that are closer to home versus beaches that are further away. Great Lakes beach
goers also prefer lower amounts of algae on the shore and in the water, and all else equal
they prefer longer beaches to shorter beaches.
While the findings themselves may seem intuitive, it is important to consider the
implications. This study represents the first wide-spread collection of preference information
from Michigan residents regarding changes in environmental quality at Great Lakes
beaches. The tradeoff information gathered in this study is consistent with accepted
economic theory, and stands as a first step in studying the values that residents hold for
changes in beach quality.
One important outcome of this study is the ability to compare the MRSs for different
attribute levels. Comparisons of the Marginal Rates of Substitution give us great insight into
the importance of different beach characteristics relative to one another. Interestingly, this
study suggests that Great Lakes beach goers have a greater preference for beaches with
lower amounts of algae in the water than for beaches with lower amounts of algae on the

51

shore. The combinations of stresses to the natural system that lead to increases in the
amounts of algae at Great Lakes beaches are widespread and complicated. The findings of
this study indicate that there is value to seeking controls to the various factors influencing
nuisance algae growth, or mitigating the presence of nuisance algae at Great Lakes
beaches. This study’s results highlight the importance of setting best management practices
and other industry standards to minimize the promotion of nuisance algae in the Great
Lakes. Based on this study’s results, we gain a clearer understanding of how such programs
could have widespread impacts not only on the Great Lake ecosystem but also on the
millions of Michigan residents who enjoy the use of public beaches each year.
This study also shows respondents have a strict preference for beaches that are
tested for bacteria over beaches that are not tested at all. This result shows an example of
the value of information: while testing for bacteria does nothing to affect the frequency of
beach closures due to elevated bacteria, this study shows there is a benefit to having tests
in place. While addressing the large-scale causes of elevated bacteria such as combined
sewer overflows or runoff may seem daunting, it is important to consider that there is still
value to having beaches tested. This result is similar to Krieger and Hoehn (1998) who
found Michigan anglers valued a full disclosure fish monitoring program that would inform
anglers of sites that were tested and found to be safe (as opposed to advisory information
that only listed exceedances). This finding verifying the value that visitors to Great Lake
beaches have for bacteria testing is timely: as this manuscript is in preparation, the draft
budget for U.S. EPA’s fiscal year 2013 (U.S. EPA 2012b, p. 28) calls for the elimination of
Beach Grant Funding. Beach Grant funding helps support testing of public beaches across

52

the country. As evidenced by this study, tests for bacteria in the waters of public beaches
provide vital information and a valuable service to millions of beach goers in Michigan.
In general, this study helps describe not only the values Michigan residents have for
different Great Lakes beach attributes, but by placing the preferences for attributes relative
to one another, gives a sense of the relative importance of each attribute. The values and
relative importance of certain environmental quality measures, such as the importance of
low amounts of algae on the shore versus algae in the water can help decision makers and
resource managers prioritize future studies or the planning of Great Lakes beach
improvements. These results also show justification for further beach testing, even at low
frequencies, as there is a value for having a beach tested at least monthly, even though the
act of testing has no impact on water quality. By gathering information on these preferences,
this study has begun to provide the proper economic context for resource managers and
decision makers to answer questions facing the Great Lakes.

53

CHAPTER 2: Labeled Versus Unlabeled Choice Experiments for Valuing Great Lakes
Beach Characteristics.
Introduction
There is a rich literature that uses choice models to estimate the values and demand
that consumers have for the attributes that comprise goods as well as goods on a whole
(Louviere, Hensher and Swait 2000). Choice experiments have been used in marketing
(Green and Srinivasan 1978), transportation (Hensher 2001), development economics
(Rubey and Lupi 1997), and agricultural economics (Pozo, Tonsor and Schroeder 2012).
Within a choice experiment, a participant is shown two or more alternatives comprised of
varying levels of attributes and asked to select the most preferred alternative. Alternatives
can either be unlabeled or labeled. Unlabeled choice alternatives are commonly denoted by
a generic title or a letter (e.g. “Choice A, Choice B” or “Program C, Program D”). Labels on
choice alternatives can indicate a brand name, location name, or any other unique identifier
for that alternative (e.g. “New York, Los Angeles”, “Coke, Pepsi”, or “Train, Car, Bus”). In
this paper, within a choice experiment pertaining to Great Lakes beach choice, we use
labels to indicate which Great Lake the beach alternative lies on.
In a labeled choice experiment, researchers elicit respondents’ preferences for the
labels as well as the other attributes of the good, with the label serving to capture
characteristics of the alternative that consumers associate with the name apart from the
other attributes in the choice experiment (Louviere, Hensher and Swait 2000). Discrete
choice studies in marketing have examined differences in consumer preferences for goods
in labeled choice experiments as a means of investigating brand equity and market share at
least as far back as the studies of Swait et al. (1993).

54

For many attribute based choice studies, whether alternatives are labeled versus
unlabeled is central to the project’s research goal, such as studies of brand equity, brands’
market shares, or different modes of transportation. Other times the inclusion of labeled
alternatives is at the researcher’s discretion. When deciding whether or not to include labels
in a choice experiment, researchers should consider the benefit of added familiarity or
realism the labels provide (Adamowicz and Boxall 2001). However, when labels are present
in a choice set, respondents may seek to simplify the decision making process and base
their choice of the preferred alternative solely on the label, ignoring other attributes (Blamey
et al. 2000). Respondents faced with choice sets containing labels may also consider
alternatives outside of those appearing in the choice task (Adamowicz and Boxall 2001).
Concerns exist among practitioners of attribute-based choice or ranking methods that
labeled alternatives may contain levels of attributes that, in combination with the presence of
a label, appear unrealistic or infeasible to respondents, depending on the respondent’s
perceptions of the label (Huybers 2005, Carson et al. 1994). Despite the concerns over how
the presence or absence of labels may affect the results of choice experiments, little
empirical evidence exists to characterize those effects.
Purpose
The purpose of this study is to provide further evidence of the effects of different
labeling schemes on the results of choice experiments. In this paper, we use a split sample
web survey to test for differences in the preference information elicited from choice
experiments with different labeling schemes. The study is based on a choice experiment
pertaining to Great Lakes beach recreation site choices wherein the names of Great Lakes
were used as labels to identify which Great Lake that a beach alternative lied on. These

55

labels indicating the Great Lake stand as “brands,” or attributes within the choice
alternatives that capture preferences for inferred characteristics outside of the attributes
listed in the choice alternatives. Respondents were randomly assigned to one of three
choice experiments each with a unique labeling scheme: a scheme using “labeled”
alternatives, where the beaches were labeled with a Great Lake; a scheme with “samelabeled” alternatives, where Great Lake labels were present but remained constant within
choice sets; and a scheme with “unlabeled” alternatives, where no Great Lake labels were
present. Figure 2.1, 2.2 and 2.3 show examples of choice sets from each of the three
different labeling schemes used in this study.
This paper will add to the existing research on the sensitivity of choice experiments to
researchers’ design choices (in the spirit of Rose et al. 2009, among others).The labeling
experiment in this paper will also provide insight about the usability of preference
information from choice experiments in a benefit transfer setting. Benefit transfer is the
practice of applying welfare information gathered from an original study site to a different
site (commonly referred to as the policy site) where no such information exists. For
overviews of benefit transfer methods, see Johnston and Rosenberger (2010), Rosenberger
and Loomis (2003), Boyle et al. (2010), Wilson and Hoehn (2006). A key advantage of
benefit transfer is that it can provide decision makers a convenient estimate of relevant
information without the considerable time and resources required to conduct an original
study. A reasonable application of benefit transfer requires a degree of similarity across the
study and policy sites, as well as a similarity in contexts between the original study and the
focus of the transfer (Johnston 2006). Within the application of benefit transfer, one must
also consider the inclusion of all relevant explanatory variables (Piper and Martin 2001).

56

Figure 2.1: Image of example choice set from “Labeled” choice experiment, original size
monitor-dependent
Beach A

Beach B

Great Lake

Lake Huron

Lake Michigan

Bathrooms

Flushing toilets
cleaned hourly

Flushing toilets
cleaned hourly

Algae
in the water

Moderate
(occasionally come in
contact with algae)

Low
(rarely come in
contact with algae)

Algae
on the shore

Low
(1-20% of the
shore has algae.)

None

Length of beach

5280 yards (3 miles)

880 yards (1/2 mile)

Testing water
for bacteria

Daily

Weekly

Distance from
your home

48

169

Which of the above
beaches would you visit?

Beach A

57

Beach B

Figure 2.2: Image of example choice set from “Same- Labeled” choice experiment, original
size monitor-dependent
Beach A

Beach B

Great Lake

Lake Erie

Lake Erie

Bathrooms

Flushing toilets
cleaned hourly

Flushing toilets
cleaned hourly

Algae
in the water

Moderate
(occasionally come in
contact with algae)

None

Algae
on the shore

None

High
(more than 50% of the
shore has algae.)

Length of beach

1760 yards (1 mile)

220 yards (1/8mile)

Testing water
for bacteria

Weekly

Daily

Distance from
your home

18

33

Which of the above
beaches would you visit?

Beach A

58

Beach B

Figure 2.3: Image of example choice set from “Unlabeled” choice experiment, original size
monitor-dependent.
Beach A

Beach B

Bathrooms

Flushing toilets
cleaned hourly

Flushing toilets
cleaned hourly

Algae
in the water

None

Moderate
(occasionally come in
contact with algae)

Algae
on the shore

Low
(1-20% of the
shore has algae.)

High
(more than 50% of the
shore has algae.)

Length of beach

220 yards (1/8mile)

880 yards (1/2mile)

Testing water
for bacteria

Weekly

Daily

Distance from
your home

118

78

Which of the above
beaches would you visit?

Beach A

59

Beach B

Recent published research has focused on advancing practitioners’ understanding of
the optimal conditions for conducting benefit transfer. Several key factors for a successful
benefit transfer have been identified including similarity between the study and policy sites.
Site similarity can refer to several factors including socio-demographic characteristics of the
affected populations, and physical characteristics of the resources being valued.
Rosenberger and Phipps (2007) review several transfers and conclude that expected errors
in transfers decrease as the similarity in sites and populations increase. Johnston (2007)
examines how similarity in contexts between study and policy sites can improve benefit
transfer. Johnston uses results from an identical choice experiment valuing land
conservation practices conducted in four different Rhode Island communities to test the
transfer of values between similar populations and sites. Results show that transfer validity
is increased for transfers between communities more similar in terms of attributes relevant
to land use policy such as housing density and population density. This finding highlights the
importance of context-similarity between study and policy sites within the practice of benefit
transfer.
Additional contributors to similarity in context across study and policy sites could
include a number of site-specific features, including the presence of name-specific or sitespecific attributes that are brought to bear when measuring preferences. Through the
comparison of the three labeling regimes in the choice experiment, our study allows a view
into the context similarity between our study site and possible policy sites. Our study’s split
sample design allows a unique perspective as to whether labels such as site names matter
when eliciting stated preference information. Transferring preference information from a
labeled study site to a policy site outside of the study region is likely to be more feasible if

60

differences between the preference information elicited in labeled and unlabeled choice
experiments are minimal (i.e. preference information is not affected by labels, and therefore
is more readily transferable).
This paper will also help fill the need for information on the public’s values and
preferences for different environmental conditions at Great Lakes beaches. The Great Lakes
coast of Michigan features approximately 600 public beaches along with countless private
beaches (MDEQ 2012a). Visiting Great Lakes public beaches is one of the most popular
activities among Michigan residents, with more than 50% of residents reporting having
visited a public beach on the Great Lakes during the prior two swimming seasons (Lupi et al.
2012). Despite the popularity and uniqueness of the resource, it comes under constant
threat from natural and manmade stresses, and little is known about the values and
preferences Michigan residents have for different environmental conditions at Great Lakes
beaches to allow for economically efficient management of the resource.
Research Questions
In addition to interpreting the outcome of the basic choice models, this paper seeks to
examine the following research questions:
Question 1: How will the parameters estimated from the “Labeled,” “Same-Labeled,”
and “Unlabeled” choice experiments compare to one another? We expect to see similar
parameters in the “Same-Labeled” and “Unlabeled” models, and to find different parameters
in the “Labeled” model. Under this question, we test the null hypothesis that the parameters
within the three models (each estimated from the results of different labeling schemes) are
identical (equation 10)

61

(10)
Where β is a vector of parameters estimated for Great Lakes beach characteristics included
in a choice experiment employing a given labeling scheme.
Question 2: How will the marginal rates of substitution (MRS) estimated from the
different choice experiments compare to one another? The expectation is that despite
possible differences in parameter estimates, the marginal rates of substitution from the
“Labeled,” “Same-Labeled,” and “Unlabeled” models will all be highly similar and exhibit
similar rankings and relative magnitudes. To compare the MRSs, we use the approach of
Poe et al. (1997) test the null hypotheses that the MRS estimated from the results of each of
the three labeling schemes are identical (equation 11).

(11)
where MRS is a vector of marginal rates of substitution calculated for parameters (beach
characteristics divided by the parameter on distance) included in a choice experiment
employing a given labeling scheme.
This paper will give a brief background on the theory underlying the choice
experiment, including random utility theory and logit model estimation, as well as useful
62

outputs from choice experiments (relative preferences and marginal rates of substitution).
Next, the paper describes past work investigating the sensitivity of choice experiment data
to different study design dimensions. Then, we review other studies that have investigated
the effects of labeled versus unlabeled alternatives. Following that discussion, we will outline
the design, implementation, and results from the choice experiment conducted about Great
Lakes beach attributes. The choice experiment uses the names of Great Lakes as the labels
in different labeling schemes to test the effect of different labeling regimes on the
preferences elicited by choice experiments. The paper closes with a discussion of results,
and conclusions.
Method: Choice Experiment
Choice experiments have long been used in marketing research as a means of
measuring consumers’ values for entire goods or specific attributes of goods being
developed for sale in markets. Green, Krieger and Wind (2001) summarize the evolution of
conjoint analysis (rating or ranking exercises) into discrete choice based methods with the
introduction of McFadden’s random utility theory and application of logit modeling of choice
(McFadden 1974). Choice experiments have since become popular within the field of
resource and environmental economics where the method is applied to goods for which no
markets exist (Hanley, Wright and Adamowicz 1998). A choice experiment presents a
respondent with two or more alternatives comprised of varying levels of attributes and asks
the respondent to choose the most preferred alternative. The underlying economic
interpretation of the choice experiment is as that the choice set represents a constrained
optimization problem where the respondent is assumed to be maximizing his or her utility
subject to his or her budget constraint. By observing many choices made among

63

alternatives with varying attribute levels, researchers can model the tradeoffs respondents
make between different levels of the attributes, and these tradeoffs can be used to calculate
the demand for attributes as well as economic welfare measures.
Measuring brand equity has long been a focus of marketing research (Keller 1993).
The labeling of alternatives within a choice experiment has served as a means of accessing
brand equity since at least Swait et al. (1993) and Park and Srinivasan (1994): both studies
used a discrete choice experiment, labeling alternatives with different brand names, as a
means for uncovering brand equity. Kamakura and Russell (1993) analyze actual market
transactions (consumer’s laundry detergent purchases) to measure brand equity in a
framework similar to Swait et al. (1993). Since labels capture inferred attributes associated
with the label and not with attributes explicitly listed in the choice set, it can be difficult to
identify exactly what is being inferred by the label (Louviere, Hensher, and Swait 2000).
However, as Kamakura and Russell (1993) point out, brand equity, or other measures of
utility related to alternative specific labels can best be calculated when the measurable or
physical attributes of the product are well-accounted for within the choice sets. Practitioners
can use focus groups and pretests to ensure that in addition to labels they have included
pertinent attributes. Within the context of the choice experiment, researchers can vary
alternative’s attributes (and if applicable, labels) to allow the identification of the influence
each attribute (and label) has on respondents’ choices apart from the other attributes
(Louviere, Hensher, and Swait 2000). Including labels in a choice experiment can increase
the realism of the choice alternatives, an overall desirable trait, since choice sets should
mimic actual decisions made by respondents as closely as possible to ensure the accuracy
of results (Harrison 2006).

64

Random Utility Theory
Underlying the choice between alternatives is the utility the respondent would
experience from choosing each alternative. The statistical model used to represent the
choices made by choice experiment participants is based on random utility theory
(McFadden 1974). The choices observed are assumed to consistent with Lancaster (1966)
in that respondents review the attributes and attribute levels listed for each alternative in the
choice experiment and choose the alternative that maximizes their utility. The indirect utility
function is comprised of two main components: a deterministic (observable) component and
a stochastic (unobservable) component. The deterministic component is a function of the
attributes that comprise each alternative, the individual’s characteristics, and unknown
parameters. The stochastic component is an error term. Per Louviere, Hensher and Swait
(2000) and Alberini et al. (2006), this relationship is stated formally as:

̅(

)

(12)

for every i individual and every j alternative; x is a vector of attributes that take on various
levels for each alternative within a choice set, β is a vector of unknown parameters
associated with those attributes, and

is an error term capturing factors specific to the

alternative and to the individual that affect the respondent’s utility but that the researcher
cannot observe. The deterministic portion of the indirect utility function is commonly
specified as a linear function of the attributes of the alternative and the respondent’s income
(y) less the cost of the alternative (C) :

(

)

65

(13)

where the coefficient

represents the marginal utility of income because yi –Cij represents

money person i has left over to spend elsewhere after alternative j was chosen at cost Cij.
For this study, we do not use a strictly monetary cost component, but rather use
distance from home as the cost to the respondent to enjoy the beach alternative of their
choosing. By using distance instead of monetary value, the choice situation presented in this
choice experiment is consistent with real trip taking behavior, where each respondent would
face a trip from home with an associated travel distance to the beach rather than a dollar
amount they must pay in order to enjoy the beach since not every beach has parking fees or
costs for entry. The marginal utility of additional miles away from home could be converted
to a monetary measure by defining a relationship between miles driven and money.
However, this paper leaves the “cost” of the alternatives, and therefore the costs of changes
in attribute levels, in terms of miles driven. Future analysis of web survey data will focus on
translating information on actual trips taken into economic values for beach attributes in
terms of dollars.
Within the choice experiment, the respondent chooses the option from the available
alternatives that maximizes his or her utility. The appropriate model for the outcome of the
choice experiment with two alternatives is a discrete choice model describing the probability
that an alternative, A, is chosen over alternative B by an individual, i.
(14)
Which can be expanded as:

(15)
66

And it follows:
(16)
We see in equation 16 that the probability of selecting an alternative does not depend on
attributes that remain constant across the alternatives, such as income (y), the intercept
term ( ), or other characteristics of the individual. Rather, the probability that a respondent
selects an alternative depends on the differences in the levels of attributes for a given
alternative relative to the levels of attributes that appear for other available alternatives.
When it is assumed that the errors in the model are independently and identically
distributed (IID) and follow a type 1 extreme value distribution, or Gumbel distribution, a
conditional logit model can be used to estimate the probability of a respondent’s choice
(Louviere, Hensher and Swait 2000, Alberini et al. 2006). McFadden (1974) shows that the
choice probabilities within the logit model are equal to:
(17)

Where µ is scale parameter that is inversely proportional to the model’s overall variance
(Ben-Akiva and Lerman 1985). The scale parameter cannot be separated from the
estimated parameters, β, and so it is commonly normalized to 1 in order to allow
researchers to identify the preference parameters. However, logit models estimated from
different data sets will have different model variances, and hence different scale factors.
When comparing the parameter estimates from logit models with different underlying data
sets, without controlling for differences in scale factors, it is unclear whether differences in
parameters are due to scale factors or are due to actual differences in parameters (Swait
and Louviere 1993).
67

In the case of choice experiment where respondents are faced with more than one
choice set, researchers can employ a random-effects logit to control for correlation in error
terms across responses from the same respondent (Wooldridge 2010).
Using the results of the logit model, researchers can estimate the tradeoffs that
respondents make between different levels of the attributes. These tradeoffs are referred to
as the Marginal Rate Substitution (MRS) and are commonly put in terms of a cost
parameter, or in our case, distance from the respondent’s home. In general terms, holding
all other attributes constant, the MRS is equal to the change in one attribute, X1, required to
compensate the individual for a one unit change in another attribute, X2 (i.e. the amount of
X1 required to keep the individual at the same level of utility as before the one unit change in
X2). In our case, we calculate the MRS in terms of miles from home relative to the baseline
attribute level. The MRS assumes there is a change in an attribute at a beach and estimates
the additional miles further from home (if the MRS is positive) or the number of miles closer
to home (if the MRS is negative) that the beach would have to be in order for an individual to
be indifferent between the beach before and after the change in environmental quality.
Equation (17) illustrates how to calculate MRS, while equation (18) shows an example MRS
calculation for moderate algae in the water in terms of driving distance specific to our
application.

68

( )

(

)

(17)

(18)

For example, an MRS of 75 for a change in algae in the water from high to moderate (high
being the baseline level) is interpreted to mean that all else equal, a respondent would be
indifferent between travelling some distance to a beach with high amounts of algae in the
water, or travelling that same distance plus an additional 75 miles to a beach with identical
characteristics except for moderate amounts of algae in the water instead of high amounts.
The above MRS equations implicitly include the scale factor, µ, which is inseparable
from the parameter estimates and specific to the logit model from which the parameters
were estimated. The scale factor, present in both the numerator and denominator of the
MRS is cancelled out through this calculation (see equation 6). This result means that MRSs
estimated from different data sets can be compared without being confounded by
differences in scale factor (Swait and Louviere 1993).
Previous Research: Sensitivity to Design Factors
Data gathered in choice experiments is sensitive to many factors, including the
respondent’s informational capacities and abilities (Ford et al. 1989) but also by aspects at
the discretion of the researcher, such as task and informational complexity (Heiner 1983;
Swait and Adamowicz 2001). Recently there has been a growing literature on the sensitivity
of choice experiment data to the elements of choice experiments that are left to the

69

discretion of the researcher (e.g. Hensher 2006b, Rose et al. 2009). Apart from their own
original contributions, Caussade et al. (2005) provide a detailed summary of studies that use
split sample experiments to test the effects of varying different design elements within a
choice experiment.
Hensher (2001) varied the number of choice sets each respondent faced in a choice
experiment pertaining to travel choices in New Zealand, finding no relationship between the
valuation information elicited across varying numbers of choice sets. Rolfe and Bennett
(2009) conducted a split sample choice experiment on the development of water resources
in Australia using treatments where respondents viewed choice sets containing two or three
alternatives, finding differences in parameter sign and significance between the models.
Serial non-response in the case of the two alternative treatment lead the researchers to
question whether the referendum approach could prevents respondents from making
tradeoffs at all, and instead avoid answering such a format. DeShazo and Fermo (2002) find
that increases in the number of alternatives posed to respondents in each choice set first
decreases model variance then, as the numbers of alternatives continue to rise, the model
variance begins to increase as well. Racevskis and Lupi (2008) found statistically significant
differences between Michigan residents’ preferences for forest management programs
gathered from individuals facing only one choice set versus those who faced four choice
sets.
Other studies such as Hoehn, Lupi, and Kaplowitz (2010) have investigated how
different information presentation formats affect preferences measured in choice
experiments. They found that presenting choice alternative information in a text format
elicited different preferences than from identical information in a table, with preference

70

parameters from text formats having higher variances than those estimated from table
formats.
A group of studies have used a “Design of Designs” format to generate an
experimental design where the design elements are varied within a split sample choice
experiment, so as to isolate the effects of variation in elements on the resulting data.
Caussade et al. (2005), Hensher (2006b), Rose et al. (2009) employ this strategy that varies
each of the five design elements (number of attributes, number of attribute levels, range of
attribute levels, number of alternatives within choice sets, number of choice sets faced by
respondents) across the sample such that they can isolate the impact each element has on
the choice data.
Recent work by Hensher (2006a) studies respondent’s attribute processing strategies
and how those interact with varying design to dimensions to influence perceived task
complexity (and therefore affect possible biases). The research is unique in that it considers
not only how the “amount” of information in a choice experiment can affect the outcome, but
also the perceived relevance of information. While our research does not extend into the
impact of attribute processing, it does fill a gap left in previous literature on the influence of
including (or excluding) information communicated by a label has on the tradeoffs observed
within a choice experiment.
Previous Research: Effects of Labeling
Much of the research conducted investigating the effect of labeling on the results of
choice experiment use labels to represent a means of program provision. Rolfe and Windle
(2011) test the effects of labels indicating the method with which a proposed coral reef
protection program would be achieved. The authors conducted a split sample survey of

71

households in Brisbane, Australia, eliciting household’s willingness to pay to protect the
Great Barrier Reef. One version of the choice experiment featured generic alternatives
describing protection policies and outcomes, and the other featured policy labels describing
how the policy would be provided (e.g. improve water quality, reduce greenhouse gas
emissions, or increase conservation zones). They found WTP to be higher among those
responding to the labeled version, perhaps indicating that respondents were willing to pay
more for Great Barrier Reef protection when they the method of protection. However, a Poe
test (Poe et al. 1997) revealed no statistically significant difference in the WTP from the
labeled and unlabeled experiments.
Similarly, Czajkowski and Hanley (2009) conduct a study of Polish households’
willingness to pay to protect local biodiversity and test for differences in the results of a
generic choice experiment and a choice experiment where alternatives are labeled as being
provided through the expansion of an existing national park, or by “other” means. The
authors find that by controlling for the effects of service provision through labeling choice
experiment alternatives, the resulting WTP measures show greater sensitivity to scope, a
desirable trait.
Blamey et al. (2000) and implemented a choice experiment regarding preferences for
forest preservation, sampling households in Brisbane, Australia. A portion of the
respondents viewed a choice experiment with generic policy alternative, while the remaining
respondents viewed a choice experiment with labeled policy alternatives which named
minimum percentages of scarce forest was to be left preserved. The authors found that on
average, the WTP for forest preservation was four times higher in the generic case than in
the labeled case. The authors also found that certain attributes that were found to be

72

significant influencers of respondent’s policy choices the forest which was to be preserved.
While the results of Blamey et al. do suggest anchoring around the label, the effect of
introducing the label to the choice experiment may be confounded by the quantitative nature
of the label itself.
Recently, Brannlund and Persson (2012) conducted a choice experiment eliciting
information on Swedish citizen’s preferences for CO2 reduction programs. Respondents
were shown two alternatives, each resulting in the same amount of total CO2 reduction. Half
of the respondents saw unlabeled alternatives, while the other half were shown alternatives
labeled as being funded through private (i.e. individual) taxes or by “other” means. Models
estimated from the unlabeled and labeled choice experiments revealed similar results in
terms of statistical significance and sign. The model estimated from the labeled choice
experiment included an alternative specific constant which indicated respondents had a
significantly lower willingness to pay for CO2 reduction programs funded by taxes than
through other means. While these studies are able to compare choice experiments with
labeled and unlabeled alternatives, the use of labels to represent different program funding
mechanisms or modes of program provision can be viewed largely as gathering preferences
for separate program attributes, which is different from the use of labels as a brand, site
name, or other unique identifier capturing preferences for information not explicitly listed in
the choice set.
Huybers (2005) compares the results of generic and labeled choice experiments
pertaining to vacation destination choices where the labels are site names. To the best of
this author’s knowledge, Huybers’s study is the only published study that uses site names

73

as the labels of a choice experiment and compares those results to a generic choice
experiment. The implicit prices calculated from models estimated from responses to the two
versions of the survey reveal that in seven of the ten attribute levels, the implicit prices were
higher in the labeled case. Again, using the Poe et al. (1997) test, the authors found that
fewer than half (four) of the differences in implicit prices were statistically significant at the
10% level or lower.
Application: Great Lakes Beaches
We apply this experiment investigating labeling effects in choice experiments to a
2012 survey of Michigan resident’s use of Great Lakes beaches. Michigan’s 3200 miles of
Great Lakes coastline provides nearly 600 public beaches, with countless more private
beaches, each with varied characteristics (MDEQ 2012a, MDEQ 2012b). Beach
characteristics range from short beaches adjacent to large population centers to expansive
sand dunes in rural areas. Nearly half of Michigan’s residents visit Great Lakes beaches
each year (Lupi et al. 2012).
Despite their popularity and importance, the Great Lakes constantly come under
threat (IJC 2011a). The health of natural systems within the Great Lakes as well as the
associated human uses is greatly affected by human development of Great Lakes shoreline,
yet pressure still persists to build new homes and to locate new industry within Great Lakes
coastal regions. Harmful bacteria and pathogens in the waters of Great Lakes public
beaches pose public health risks (Rose et al. 1999) and threaten the public’s enjoyment of
Great Lakes beaches. E. coli enters the Great Lakes through natural processes and
manmade sources that are each difficult to avoid or mitigate (U.S. Policy Committee 2002),
resulting in persistent health risks at Great Lakes beaches. However, state budgets currently

74

allow only 262 of Michigan’s approximately 600 Great Lakes public beaches to be tested at
least monthly for E. coli (MDEQ 2012a), leaving nearly 60% of public beaches on the Great
Lakes untested. In 2011, 101 of the 260 Great Lakes beaches monitored for bacteria, or
nearly 39%, reported an exceedance of safe levels of bacteria at least once during the peak
swimming season (ibid). Recently, large blooms of nuisance algae such as cladophora have
been occurring with greater frequency than in years past (Auer et al. 2010, Higgins et al.
2008). The blooms can eventually be swept onto beaches and begin to rot, by spoiling the
aesthetics of coastal areas and generating a powerful, unpleasant odor (Verhougstraete et
al. 2010). Studies have linked the rise in nuisance algae to the success of the invasive zebra
mussel (Dreissena polymorpha) (Auer et al. 2010, Higgins et al. 2008, Wilson et al. 2006)
and also shows the mats of decaying algae can harbor bacteria and pathogens (Vanden
Heuvel et al. 2010; Verhougstraete et al. 2010; Olapade et al. 2006) causing public health
concerns among local officials (Bay County Health Department 2007).
Previous Research: Great Lakes and Beach Valuation
Within the field of environmental and natural resource economics, a number of
studies have sought to value changes in the environmental conditions on the Great Lakes to
better inform decision making (NMI and NOAA 2001). Several studies use models of
recreational fishing demand to examine the economic costs and benefits of programs on the
Great Lakes (e.g. Lupi et al. 1998, Lupi et al. 2003, Kotchen et al. 2007). Research has also
described how changes in air quality (Chattopadhyay 1999) and water quality (Ara et al.
2006, Braden et al. 2004) on the Great Lakes can impact property values. Others including
Whitehead et al. (2009) and Hoehn et al. (2010) examine the economic value of changes in
the quality of wetlands near the Great Lakes.

75

Existing literature also provides many examples of studies valuing changes in
environmental attributes related to beach recreation on marine beaches. Researchers have
examined the value of changes in visitor congestion at beaches (McConnell 1977), water
quality (Bockstael, McConnell and Strand 1989, Lew and Larson 2005, Hilger and
Hanemann 2006), debris (Smith, Zhang and Palmquist 1997), and beach width and beach
re-nourishment (Parsons, Massey and Tomasi 1999, Landry, Keeler and Kriesel 2003,
Shivlani, Letson and Theis 2003, Huang, Poor and Zhao 2007).
In contrast, existing literature on Great Lakes ecosystem service valuation includes only one
peer-reviewed published article (Murray et al. 2001) that examines the value of Great Lakes
beaches, in addition to few unpublished studies (e.g. Sohngen, Lichtkoppler and Bielen
1999, Egan and Dwyer 2008, Shaikh 2006, Shaikh 2012). Although there is a growing
literature on valuing changes in environmental quality across various Great Lakes resources
and an established literature on valuing marine beach recreation, there is very little known
about the public’s values and preferences for different levels of environmental quality at
Great Lakes beaches. With this gap in mind, we designed a choice experiment to help fill
provide this information.
Survey Development: Pretests
This study reports on the findings of a choice experiment implemented as part of the
Great Lakes beach web-based survey in the spring of 2012. The web survey was developed
and questions were tested in the spring of 2012 using an iterative process that was guided
by the results of 57 one-on-one cognitive interviews (Kaplowitz, Lupi and Hoehn 2004). The
57 interviews were composed of eight interviews conducted among a convenience sample
of Michigan State University students at a campus food court, 19 interviews conducted in

76

person among randomly selected adults at local shopping malls, and 30 interviews
conducted remotely with respondents recruited from a web survey panel (Survey Sampling
International) of adults in the lower peninsula of Michigan. The remote interviews were
conducted over the phone using an innovative approach not previously documented in the
literature on the pretesting of web-based valuation surveys.
Once pretests and initial web-survey debugging were complete, a pilot group of 85
individuals were mailed invitations to the web survey. In the two weeks following the initial
invitations, pilot study members were mailed two reminders: a half-sheet postcard and a
quarter-sheet postcard, each listing the survey’s web-address, the individual’s password,
and information on who to contact with questions. The pilot study received a total of 22
logins for a response rate of 25.9%. The pilot survey allowed the research team to simulate
the process of actually implementing the survey (e.g. timing of mailing, receiving data) and
to see if any questions arrived via phone or email from pilot survey respondents.
Survey Development: Input from experts
The text, diagrams, and graphics used to describe the attributes to respondents were
developed with the help of state and county health officials, water policy experts, as well as
resource managers from State and Federal agencies including NOAA Great Lakes
Environmental Research Laboratory, Michigan Sea Grant, EPA Great Lakes Program
Office, Michigan Department of Natural Resources, and county officials that perform water
quality tests and conduct beach algal assessments. Parts of the choice experiment needed
to convey scientific information in a way that was understandable to members of the general
population who are not assumed to have advanced knowledge of the topics in the survey.

77

The resource experts we consulted were able to ensure that the survey communicated
accurate information.
Certain attribute descriptions shown to choice experiment participants also contained
diagrams and drawings. Figure 2.4 and figure 2.5 show diagrams that accompanied
definitions of attribute levels for amount of algae on the shore and amount of algae in the
water, respectively. The use of photos to describe attributes and their levels was considered
but research shows unless closely controlled, photos can communicate unintended
information (Hoehn, Lupi and Kaplowitz 2003). Even though the diagrams are simplifications
of real-world conditions, the experts we consulted in the development of survey materials
and pretest participants (i.e. perspective survey respondents) found the diagrams to
represent conditions experienced at actual Great Lakes beaches.
Survey Sample
Because there is no list of beachgoers to facilitate sampling, a two-step sampling
process was used. First, in the summer of 2011, a general population mail survey on
participation in leisure and recreation activities was conducted using a sample of 32,230
residents of Michigan’s Lower Peninsula, 18 years and older, drawn from the drivers' license
list. The mail survey contacts followed a modified version of Dillman's tailored design
method (Dillman 2009), and the survey achieved a response rate of 38%. Among the
activities asked about in the mail survey was whether or not the respondent had visited a
Great Lakes beach in the last year. Second, in the spring of 2012, invitations to a Great
Lakes beaches web-based survey were sent to 5,434 qualifying respondents from the mail
survey (i.e. those who reported having visited a beach on the Great Lakes in Michigan,)

78

Figure 2.4: Image of diagram of Great Lakes beach attribute: Algae in the water, original
size monitor-dependent.
Amount of
algae in
the water

Definition

View of the Water

None

visitors never come in contact
with algae while swimming or
wading

Low

visitors rarely come in contact
with algae while swimming or
wading

Moderate

visitors occasionally come in
contact with algae while
swimming or wading

High

visitors constantly come in
contact with algae while
swimming or wading

79

Figure 2.5: Image of diagram of Great Lakes beach attribute: Algae on the shore, original
size monitor-dependent.
Amount of
algae on the
shore

Definition

View of swimming area shore

None

None of the shore of
the swimming area has
algae.

Low

1 to 20% of the shore of
the swimming area has
algae.

Moderate

21 to 50% of the shore
of the swimming area
has algae.

High

More than 50% of the
shore of the swimming
area has algae.

80

again following a modified version of Dillman's tailored design method (Dillman 2009) and
achieving a response rate of 59.6%.
Web Survey and Choice Experiment
The choice experiment section of the web survey began with separate pages
12

explaining each of the attributes . Attribute descriptions were developed with the help of
the aforementioned resource experts. Each attribute description was followed by questions
engaging the respondent with the information presented about the attributes.
Follow-up questions asked about the respondent’s personal experiences and
attitudes towards the attributes. Thus, each respondent’s path through the choice
experiment presented information on the attributes in question in a controlled format, and
the follow-up questions allowed respondents to interact with information on the attributes, a
technique that survey research shows increases respondent’s comprehension of survey
tasks (Kaplowitz et al. 2004, Hoehn and Randall 2002, Schwarz and Sudman 1996).
Following descriptions of the attributes, each respondent was shown three different
choice sets. Figures 2.1, 2.2 and 2.3 show example choice sets from each of the three
labeling schemes were faced by a respondent. The choice experiment asked respondents to
assume they were taking a trip to a Great Lakes beach, and the only two alternatives
available were the beaches shown in the choice sets. Respondents were told the
alternatives had the different characteristics described in the table, but otherwise were
same. Beaches were described using a set of six or seven attributes, depending on whether
or not the alternatives were labeled. Respondents were then instructed to review the
12

See appendix A for set of screen captures from the choice experiment portion of the web
survey, complete with informational treatments, warm-up tasks, choice sets and follow-up
questions.
81

attributes in the table and then select which beach they would visit. When respondents
logged onto the website, they were randomly assigned to one of the three labeling groups
for the choice experiment. Members of each group saw identical versions of the survey until
programming dependent upon the respondent's labeling group showed different versions of
the choice sets: labeled (alternatives having various Great Lakes names present), samelabeled (Great Lakes names present but not varied within choice sets,) or unlabeled (no
13

Great Lake name present). In total, 1040

respondents were in the labeled group, 665

were in the same group, and 1085 were in the unlabeled group.

14

Attributes and Attribute Levels
Attributes were selected for our study based on their salience and relevance to Great
Lakes beach visitors, with a focus on beach attributes that could be tied to environmental
management schemes, not physical characteristics or facilities. The attributes used in the
choice experiment were as follows: Great Lake (what Great Lake the beach lies on),
bathrooms (the type of bathrooms available at the beach), amount of algae in the water,
amount of algae on the shore, length of beach, the frequency of bacteria testing, and the
distance away from home. Table 2.1 shows the attributes and their levels used in this choice
experiment.

13

One respondent to the “labeled” choice experiment was inadvertently excluded from the
analysis due to a write-error performed by the web survey. The error made it (continued
from previous page) impossible to link the choice set viewed by the respondent to the
respondent’s answers to the choice experiment until further steps were taken after the
analysis was conducted.
14
The unbalanced design occurred due a programming error. Although assignments to
treatment groups were random, the error resulted in fewer people being assigned to the
“same” design group. However, sample sizes are robust and all statistical tests account for
the different sample sizes across treatment groups.
82

Table 2.1: Attributes and Attribute Levels included in the Great Lakes Choice Experiment
Attribute Name
Great Lake

Bathrooms*
Algae in the water

Algae on the shore

Length of beach

Testing water for
bacteria

Attribute level
Michigan
Huron
St. Clair
Erie
Flushing toilets, cleaned daily
Flushing toilets, cleaned hourly
None
Low (rarely come in contact with algae)
Moderate (sometimes come in contact with algae)
High (constantly come in contact with algae)
None
Low (1-20% of the shore has algae)
Moderate (21-50% of the shore has algae)
High (more than 50% of the shore has algae)
50 yards
220 yards (1/8 mile)
880 yards (1/2 mile)
1760 yards (1 mile)
5280 yards (3 miles)
Never

Monthly
Weekly
Daily
Distance from your
Individual’s minimum distance to selected Great Lake
home
+ 0 miles
+ 15 miles
+ 40 miles
+ 100 miles
+ 150 miles
* Attribute was held constant within choice sets to control for preferences for different
bathrooms, but no preference or value information was gathered.

83

For the levels of Great Lakes attribute, we focused on the Great Lakes that are
adjacent to the Lower Peninsula, where we drew our sample from. Bathrooms were included
in the choice sets but were held constant within choice sets. In pretests, we found
bathrooms were a salient attribute when choosing the preferred beach, but facilities at the
beaches (including playgrounds, picnic tables, and bathrooms) did not fit into our study’s
focus on environmental quality. Instead of including bathrooms as a variable in our choice
experiment, we included it as an attribute that would remain constant within choice sets.
This kept respondents from making choices based on inferred levels for a missing bathroom
attribute based on existing attributes (e.g. “Lake Michigan has cleaner bathrooms”). Instead,
this strategy meant that with the bathrooms attribute being held constant, bathrooms would
not impact the individual’s decision. The attribute levels for algae in the water and algae on
the shore were chosen to correspond with data currently being gathered by the EPA’s Great
Lakes Beach Sanitary Survey (U.S. EPA 2008). By using the same scale as the sanitary
survey, we can combine information on which Great Lakes beaches people visited and how
often along with what the levels of algae were on the shore and in the water at those
beaches (revealed preference data) with the marginal rates of substitution for changes in
environmental quality characteristics including the level of algae on the shore and in the
water to estimate economic costs or benefits of changes in the levels of algae. Five different
values for “beach length” were used in the design to represent the approximate range of
Great Lakes public beaches in Michigan (MDEQ 2009): 50 yards, 220 yards, 880 yards (1/2
mile), 1760 yards (1 mile), and 5280 yards (3 miles). For modeling purposes, these lengths
were handled as categorical variables, allowing for a non-linear relationship between beach
length and the respondent’s utility. This mimics other beach valuation work that models the

84

natural log of beach length (e.g. Hilger and Hanemann 2006) or a linear length term as well
as a length-squared term (Lew and Larson 2005) as a means for allowing a non-linear
relationship between utility and beach length. We chose the attribute describing the
frequency of testing to have the levels of daily, weekly, monthly, or none to represent
examples of how frequently actual beaches are tested (MDEQ 2012a). As mentioned
previously, not every Great Lakes public beach is tested for E. coli. The final attribute,
driving distance contained two components. To make choice sets more realistic and avoid
showing respondents counterfactual attribute levels, the first component was the minimum
distance for each individual to reach the Great Lake that was specifically named in the
choice alternative (i.e. the minimum distance for a specific respondent to Lake Michigan
when Lake Michigan was selected for the choice alternative, or the minimum distance to
Lake St. Clair when Lake St. Clair was selected for the choice alternative). Distances were
calculated as the minimum distance from the center of the web survey sample member’s zip
code and to one of several beaches along each Great Lake using PC Miler. The second
component of distance was one of five levels of “additional miles” that were added as
described in the experimental design to the minimum distance to create the total distance.
This total distance was the only distance presented to respondents. The levels of the
additional distance variable were designed to allow distance to be estimated as a
continuous variable in our analysis. The distance attribute needed to control for the
minimum distance from a respondent’s home to a specific Great Lake since there is such
wide variation in distances across our sample: certain sample members live adjacent to
some Great Lakes (i.e. minimum distance of zero to the nearest Great Lake) while the same
individual may live 150 or more miles from a different Great Lake. By controlling for the

85

minimum distance from a respondent’s home to the Great Lake within the specific choice
alternative, we increase the plausibility and realism of the choice alternative, and avoid
showing the respondent counterfactual attribute levels, a concern raised previously in
literature regarding labeled choice experiments (Huybers 2005, Carson et al. 1994).
The experimental design for each of the three labeling treatments of the choice
experiment was generated using NGene software (Choice Metrics 2011). The experimental
design was derived in such a way that allows researchers to identify the impact each
attribute level has on the probability of a respondent selecting a given alternative (ibid,
Johnson et al. 2006.) Designs were optimized for “WTP” or F-efficiency, where designs are
produced such that the variance of the ratio of user-specified model coefficients is
minimized. Designs were also generated taking into account the expected signs of attribute
parameters so as to lessen the presence of clearly dominated alternatives and further
increase overall design efficiency. The number of choice pairs (rows) generated for each of
the three design types was as follows: 126 for “labeled,” 102 for “same-labeled,” and 96 for
“unlabeled.” Since each respondent viewed three choice sets, this translated into the
following number of unique versions of the choice experiment (i.e. unique groups of three
choice sets): 42 for “labeled,” 34 for “same-labeled,” and 32 for “unlabeled.”
Results
We begin by estimating a random-effects logit model for each choice experiment
type. Results are shown in table 2.2. The "labeled" model has a Wald chi-squared statistic of
705 with a corresponding p-value less than 0.001 and a McFadden's pseudo R-squared of
0.302. The fit of the model, as evidenced by its predictive capability, appears well-balanced
as it correctly predicted 77% of Beach A choices and 76% of Beach B choices. As expected,

86

Table 2.2 : Results of Random-Effects Logit model estimating determinants of Great Lakes beach choice for “Labeled,”
“Same-Labeled” and “Unlabeled” choice experiments

Variable
One-way distance
Algae on the shore: None
Algae on the shore: Low
Algae on the shore: Moderate
Algae in the water: None
Algae in the water: Low
Algae in the water: Moderate
Length: 50 yards
Length: 220 yards
Length: 880 yards
Length: 1760 yards
Testing: None
Testing: Monthly
Testing: Weekly
Great Lake: Michigan
Great Lake: Huron
Great Lake: St. Clair

LABELED
coefficient
-0.008
1.474
1.225
0.763
1.768
1.473
1.111
-0.780
-0.421
-0.246
0.069
-1.496
-0.332
-0.381
1.194
0.498
0.004

Goodness of fit measures
McFadden’s R-Squared
Wald Chi-squared (df)
Prob>chi-Squared

0.302
705 (17)
<0.001

P>|z|
<0.001
<0.001
<0.001
<0.001
<0.001
<0.001
<0.001
<0.001
<0.001
0.012
0.489
<0.001
<0.001
<0.001
<0.001
<0.001
0.966

**
**
**
**
**
**
**
**
**
^
**
**
**
**
**

SAME-LABELED
coefficient P>|z|
-0.013 <0.001
1.805 <0.001
1.616 <0.001
0.891 <0.001
2.095 <0.001
1.899 <0.001
1.323 <0.001
-0.816 <0.001
-0.491 <0.001
-0.179
0.152
-0.081
0.530
-1.669 <0.001
-0.534 <0.001
-0.072
0.557

0.269
314 (14)
<0.001

Beach A predicted correctly (%)
Beach B predicted correctly (%)

76%
74%
77%
78%
** significant at the p<0.001 ; * significant at the p<0.01 ; ^ significant at the p<0.05
87

**
**
**
**
**
**
**
**
**

**
**

UNLABELED
coefficient
P>|z|
-0.013 <0.001
1.742 <0.001
1.662 <0.001
0.998 <0.001
2.040 <0.001
1.907 <0.001
1.351 <0.001
-1.060 <0.001
-0.537 <0.001
-0.316
0.002
-0.142
0.156
-1.667 <0.001
-0.624 <0.001
-0.142
0.118

0.317
546 (14)
<0.001
78%
77%

**
**
**
**
**
**
**
**
**
*
**
**

the parameter on driving distance is statistically significant and negative. The
model results for the "labeled" choice experiment reveal that Great Lakes beach goers
prefer less algae on the shore and lower amounts of algae in the water, with algae in
the water causing more of a nuisance than algae on the shore. We find that they also
prefer longer beaches to shorter beaches, with no significant difference in preferences
between three mile long beaches (baseline) and one mile long beaches. Note that this
finding is consistent with revealed preference studies that show a significant and nonlinear relationship between beach length and site choice (e.g. the natural log of length in
Parsons et al. (2009), or length plus length squared in Lew and Larson (2005) where
the marginal benefits of length decrease as beaches are longer). In the labeled model,
we also see that respondents prefer beaches that are tested for bacteria versus
beaches that are not tested at all. The results also show that, all else equal, Great
Lakes beach goers prefer beaches on Lake Michigan and Lake Huron to beaches on
Lake Erie (baseline), and that beach goers are indifferent between beaches on Lake St.
Clair and Lake Erie.
The model results for the "same-labeled" choice experiment were estimated with
a Wald chi-squared statistic of 314 corresponding to a p-value less than 0.001 and a
McFadden's pseudo-R-squared of 0.269 while correctly predicting 78% of Beach A
choices and 74% of Beach B choices. The "same-labeled" choice experiment results
are vastly similar to the "labeled" model in magnitude and significance, with a significant
negative parameter on distance. In the case of same-labeled alternatives, we find that,
all else equal, respondents are indifferent between beaches that are a half mile, one

88

mile, or three miles long. Respondents shown “same-labeled” choice sets were also
indifferent, all other factors equal, between beaches tested daily or weekly for bacteria.
The model estimated using observations from the "unlabeled" choice experiment
has a Wald chi-squared statistic of 546 with a p-value less than 0.001, a McFadden's
pseudo R-squared of 0.317 and correctly predicted 77% of Beach A choices and 78% of
Beach B choices. The significance and magnitudes of estimated parameters for Great
Lakes beach characteristics in "unlabeled" choice experiment are consistent with the
"labeled" and "same-labeled" models. As with the other models, driving distance is
estimated to have a significant and negative coefficient. Unlike the "same-labeled"
model, the "unlabeled" model shows a preference for 1 mile and 3 mile long beaches
over half mile long beaches. Similar to the "same-labeled" model (but in contrast to the
"labeled" model) the unlabeled model shows Great Lakes beaches goers, all else equal,
do not have a preference for beaches tested for bacteria daily versus weekly.
Returning to our first research question regarding the similarity of model
parameters estimated from each of the three labeling schemes, we estimate models
pooling different combinations of the labeling schemes in order to calculate loglikelihood ratio test statistics. The log-likelihood ratios test the null hypothesis that the
parameters estimated in the individual models are equal to one another (equation 10).
The results of a likelihood ratio test comparing the “labeled” and “same-labeled” models
with 14 degrees of freedom yielded a likelihood statistic of 35.457, with p=0.0013, lead
us to reject the null hypothesis that the parameters estimated for Great Lakes beach
attributes in the models from the “labeled” and “same-labeled” experiments are the
same. A likelihood ratio test comparing the “labeled” and “unlabeled” models with 14

89

degrees of freedom yielded a likelihood statistic of 50.21 , with p<.001, again leading us
to reject the null hypothesis that the “labeled” and “unlabeled” choice experiments
yielded identical parameters. A likelihood ratio test comparing the “same-labeled” and
“unlabeled” models with 14 degrees of freedom yielded a likelihood statistic of 6.034,
with p=.96, thus we fail to reject the null hypothesis that the “same-labeled” and
“unlabeled” choice experiments in the Great Lakes Web survey have models with
identical parameters.
One short coming of these LLR tests, however, is that they fail to account for
differences in scale factors that exist between different logit model data gathering
methods (Swait and Louviere 1993). As mentioned previously, the presence of different
scale factors confounds differences in logit models estimated from different data sets.
To control for differences in scale, we can compare the MRS calculated from models
estimated for each labeling scheme.
Table 2.3 shows the welfare measures estimated for marginal changes in Great
Lakes beach attributes relative to the baseline attribute level. All MRS estimates are in
terms of the number of additional miles a respondent would be willing to travel to reach
a beach with the new attribute level versus an identical beach with the former (baseline)
level and still be no better or worse off than before the change in attribute. For a
desirable change in an attribute, the MRS is positive, and represents a willingness to
travel further to reach a beach under the new conditions. For an undesirable change in
an attribute, the MRS is negative, and represents a decrease in willingness to travel
further from home to experience the beach under the new conditions. The table
includes 95% confidence intervals around the mean MRS, which were calculated using

90

Table 2.3: Mean and Krinsky-Robb 95% confidence intervals of Marginal Rates of Substitution for marginal
changes in Great Lakes beach characteristics across “Labeled,” “Same-Labeled,” and “Unlabeled” choice
experiments.
LABELED
SAME-LABELED
UNLABELED
95%
95%
95%
95%
95%
CI
CI
CI
95% CI CI
CI
LowerUpper- LowerUpper- LowerUpperAttribute Level
Bound Mean Bound Bound Mean Bound
Bound Mean Bound
Algae on the shore: None
154
180
210
122
144
169
122
139
159
Algae on the shore: Low
125
150
178
108
129
154
115
133
152
Algae on the shore: Moderate
70
93
119
52
71
92
65
80
96
Algae in the water: None
188
216
248
143
167
196
144
163
184
Algae in the water: Low
154
180
209
128
151
179
134
152
174
Algae in the water: Moderate
111
136
162
84
105
130
91
108
126
Length: 50 yards
-121
-95
-71
-87
-65
-45
-103
-85
-68
Length: 220 yards
-76
-51
-29
-59
-39
-19
-60
-43
-27
Length: 880 yards
-54
-30
-7
-35
-14
6
-41
-25
-9
Length: 1760 yards
-15
8
32
-27
-6
14
-27
-11
4
Testing: None
-213 -183
-157
-160 -133
-110
-152 -133
-116
Testing: Monthly
-63
-41
-18
-62
-43
-24
-65
-50
-34
Testing: Weekly
-70
-47
-25
-25
-6
13
-25
-11
3
Great Lake: Michigan
121
146
173
Great Lake: Huron
38
61
85
Great Lake: St. Clair
-21
0
23

91

the Krinsky-Robb method at 10,000 random draws (Krinsky and Robb 1986). To
interpret the marginal rates of substitution these would need to be translated into dollar
values and placed into a realistic choice set for beaches, that is a choice set that
includes a full range of substitute sites. Instead, these welfare measures assume that
each respondent would actually take a trip to the beach specified in the choice set, and
not to some other beach. As such, these measures are an upper-bound for the actual
economic value of these marginal changes in beach attributes since respondents were
not shown a full range of substitutes, nor were they given a proxy for this such as a "do
not go," or "go to some other beach" option. In future work we will merge the choice
experiment preference estimates with a statewide recreation demand model of beach
use in Michigan that fully accounts for all substitution possibilities.
The results show some variation among the mean MRSs across models, with the
"labeled" choice experiment tending to have larger absolute values of MRS than the
"same-labeled" or "unlabeled" choice experiments. Despite this variation in mean
MRSs, we see that welfare measures are consistently ordered with intuitively "better"
attribute levels (e.g. lower algae, more frequent testing) having higher MRSs than
"worse" levels (e.g. decreases in beach length, or beaches with no testing for bacteria
at all) across all of the labeling schemes, except in several cases where respondents
were shown to be indifferent between attributes (e.g. results for the attribute levels
"weekly testing" versus “daily testing” for "unlabeled" and "same-labeled" model).
Figure 2.6, Figure 2.7, and Figure 2.8 each show graphs of the lower and upper
bounds of the Krinsky-Robb 95% confidence intervals as well as the mean MRSs for
each of the different labeling schemes. The figures reinforce the notion that across the

92

Figure 2.6: Mean and Krinsky-Robb 95% Confidence Intervals of MRS in terms of miles
for “amount of algae on the shore,” relative to the baseline level “high”

Figure 2.7: Mean and Krinsky-Robb 95% Confidence Intervals of MRS in terms of miles
for “amount of algae in the water,” relative to the baseline level, “high”

93

Figure 2.8: Mean and Krinsky-Robb 95% Confidence Intervals of MRS in terms of miles
for “frequency of testing for bacteria,” relative to the baseline level, “tested daily”

three labeling schemes, the relative scale and ordering of MRS is highly similar. The
figures also illustrate the significant overlap between confidence intervals, and how in
each case, the confidence intervals around the unlabeled MRS are the smallest.
Using the estimated MRSs from the Krinsky-Robb iterations, we return to our
second research question (equation 11) regarding the similarity of MRS estimated
across different labeling schemes, and we test for the presence of a statistically
significant difference between MRSs estimated from the different choice experiments.
Following Poe et al. (1997) and as applied by Blamey et al. (2000) and Huybers (2005),
we pair MRS estimations from the Krinsky-Robb iterations, and take the difference of
the paired estimations. We calculate the proportion of the differences that take a
hypothesized sign: following Efron and Tibshirani (1993) this proportion is an

94

approximation for a one-sided significance test. The results of these one-sided
significance approximations are shown in table 2.4. For simplicity, all differences are
tested against the hypothesis that the choice model listed first in the pairing has a
higher MRS for any given change in attribute level than the choice model listed second.
For example, in the first column, one-sided significance is calculated for the MRSs
calculated in the labeled choice experiment are higher than the MRSs calculated from
the unlabeled choice experiment. In the second column, we test for one-sided
significance of MRSs calculated in the labeled choice experiment are higher than the
MRSs calculated from the same-labeled choice experiment. It follows that the third
column estimates the one-sided significance of the MRSs elicited by the same-labeled
choice experiment are higher than the MRSs elicited by the unlabeled choice
experiment. Given these hypotheses, a value of 0.95 or higher indicates an approximate
one-sided significance of 0.05 (or lower) in support of the hypothesis that the MRSs in
the appropriate choice experiment are higher than the other. Similarly, a value of 0.05 or
lower would indicate an approximate one-sided significance of 0.05 (or lower) in support
of the converse hypothesis that the MRSs estimated by the choice experiment in
question are actually lower than the specified model of comparison.
The results of these approximate one-sided significance estimates show no
statistically significant differences between MRSs calculated from the same-labeled and
unlabeled choice experiments. When comparing MRSs from the labeled and samelabeled as well as the labeled and unlabeled, we find that fewer than half of the MRSs
are found have statistically significant differences (5 of 13 MRSs from the labeled choice
experiment are different from the unlabeled choice experiment; 6 of 13 MRSs from the

95

Table 2.4: Results of approximate one-sided significance of differences between
mean MRSs: “Labeled and Unlabeled,” “Labeled and Same-Labeled,” and “Same
and Unlabeled” choice experiments.
Proportion of differences greater than zero

Attribute Level
Algae on the shore: None
Algae on the shore: Low
Algae on the shore: Moderate
Algae in the water: None
Algae in the water: Low
Algae in the water: Moderate
Length: 50 yards
Length: 220 yards
Length: 880 yards
Length: 1760 yards
Testing: None
Testing: Monthly
Testing: Weekly

Labeled
Vs
Unlabeled
0.9924
0.853
0.8234
0.9982
0.9493
0.9622
0.2405
0.2739
0.3614
0.9188
0.0009
0.7451
0.0036

**
**
*

^
^

Labeled
Vs
Same-Labeled
0.9755 **
0.8821
0.9242
0.9895 *
0.9365
0.9592 *
0.0322 ^
0.2135
0.1534
0.8297
0.0044 ^
0.5549
0.003 ^

Same
Vs
Unlabeled
0.6231
0.3959
0.2506
0.5947
0.4733
0.4238
0.9227
0.6128
0.7987
0.6505
0.5059
0.7144
0.6739

* Approximate one-sided significance between p=0.01 and p=0.05 that MRS is greater
in base case than in case model is compared to
** Approximate one-sided significance at the p<0.01 that MRS is greater in base case
than in case model is compared to
^ Approximate one-sided significance at the p<0.01 that MRS is less in base case than
in case model is compared to
labeled choice experiment are different from the same-labeled choice experiment; 0 of
13 MRSs from the unlabeled choice experiment are different from the same-labeled
choice experiment). Only for certain levels of certain attributes are differences in MRSs
found to be statistically significant. We find that respondents in the labeled choice
experiment exhibited higher MRSs for beaches with no algae on the shore and no algae
on the water in the case of labeled choice experiments than they did in the unlabeled or
same-labeled. This could indicate a higher willingness to travel to enjoy a beach with no
algae on the shore (or in the water) when the respondent knows which lake the beach
96

lies on. We also see MRSs for no bacteria testing and weekly bacteria testing are lower
for respondents to the labeled choice experiment than for the unlabeled or samelabeled experiments. This may indicate that respondents’ preferences for levels of
bacteria testing at Great Lakes beaches is dependent on knowing which Great Lake the
beach lies on: respondents may rely on pre-conceptions of water quality at lakes when
determining their willingness to travel to enjoy beaches tested for bacteria at different
frequencies. Never-the-less, the results all show that no matter the labeling scheme,
some testing is preferred to no testing at all.
Conclusions
This paper employs a split-sample Web survey to compare the preference
information gathered from three different labeling regimes. Reviewing the randomeffects logit model outputs, we see that estimated model parameters the three labeling
schemes are all similar in sign, statistical significance, and magnitude. To address our
first research question regarding how model parameters would compare across labeling
schemes, we performed Log-Likelihood ratio tests. The results from the Log-Likelihood
Ratio tests show there is no statistically significant difference in parameters estimated
from the “same-labeled” and “unlabeled” choice experiments. Further, the LogLikelihood Ratio tests suggest that there are statistically significant differences between
preference information gathered from “labeled” and “same-labeled” or “unlabeled”
choice experiments (i.e. we reject the null hypothesis that parameters estimated in
those models are identical in those cases). While useful, simply comparing the
parameters does not account for scale factors unique to each model. To account for
differences in scale factors and to help answer our second research question regarding

97

the MRSs estimated across models, we calculated the mean and 95% CIs around the
MRSs using the Krinsky-Robb method. The mean MRSs elicited from each choice
experiment reinforce the notion that preference information is similar in order and
magnitude across the different labeling schemes. However, Poe tests of the pairwise
comparisons of individual attribute MRS show some significant differences in some of
the MRS from the labeled model. Even though some MRS are significantly different in
the labeled model, the general pattern and signs of the data are similar.
In each labeling scheme, respondents preferred beaches closer to home. The
signs and significance of estimated parameters across all three models are identical in
all but a few cases. Regardless of labeling scheme, respondents preferred less algae
on the shore and less algae in the water compared to higher amounts. In each labeling
scheme and at each attribute level, algae in the water was more of a nuisance than
algae on the shore. Respondents across the three labeling schemes also preferred
longer beaches to shorter beaches, as well as beaches tested for bacteria at least
monthly to beaches that are not tested at all.
This study’s results show that despite some statistical differences in model
parameters (evidenced by LLR test results and some of the MRS based on the Poe
tests), the preference information gathered across different labeling schemes is
qualitatively similar: the ranking and relative magnitudes of the preferences of
respondents are consistent across the different labeling treatments. The similarities in
preferences observed in this study run counter to concerns raised by previous research
about respondents relying solely on labels when answering labeled choice experiment
tasks (Blamey et al. 2000). The preference information gathered in our study could

98

show promise for benefit transfer because in a general sense the values do not depend
on the lake names. That the preferences do not substantially depend on the lake name
show these results may exhibit a degree of consistency across study and policy site
context. Had the preferences gathered from the different labeling regimes showed
significant differences in rank or magnitude, the values would appear to be highly sitespecific, and would have raised concerns about the transferability of values. Instead, the
results show that allowing site names to influence preferences for other characteristics
did not produce results exhibiting any economically consequential differences than
when site names were not present.
Although our results are favorable regarding the similarity and generalizability of
the MRS from our study, before we can confidently conclude the names (i.e., labels)
don’t matter for benefit transfer, we would recommend additional original research to
compare valuation information gathered under named (labeled) and unnamed regimes.
In particular, gaining further insight into cases where labels do and do not affect values
would be especially desirable. Never the less, that there are no economically relevant
differences between the tradeoffs calculated using data from the three different labeling
schemes used here may have ramifications for the practice of benefit transfer: since the
values elicited in the labeled setting are consistent with those elicited in the unlabeled
setting shows that these results are not Great Lake dependent. In general, the
similarities across labeling schemes should provide a firm basis for the application of
such economic information to a policy setting. The preference information gathered here
is a much needed first step to understanding how the public’s values and preferences
can be factored into the management of the Great Lakes.

99

APPENDICES

100

APPENDIX A

Image of screen display from Great Lakes Beaches Web-survey, choice experiment
and follow-up portion, original size monitor-dependent

101

This appendix replicates the information that was displayed in the Great Lakes Beaches
Web survey’s choice experiment and follow-up portions. The content was modified to
meet the requirements of the Michigan State University formatting guide for submission
of Master’s Theses and Doctoral Dissertations. The figure begins with the border that
appeared on the left margin of each webpage viewed throughout the survey and then
continues with the content of the survey.
Figure A.1: Image of screen display from Great Lakes Beaches Web-survey, choice
experiment and follow-up portion, original size monitor-dependent

102

Figure A.1 (Cont’d)
Great Lakes Beach Characteristics
This section of the survey is about characteristics of Great Lakes public beaches.
There are more than 600 public beaches on the Great Lakes in Michigan and many
more private beaches. For this survey, the Great Lakes in Michigan include:







Lake Huron
Lake Superior
Lake Michigan
Lake Erie
Lake St. Clair
and connecting waters (like the Detroit River, and the St. Clair River)

The characteristics of Great Lakes public beaches vary widely. We would like to know if
some characteristics are more important than others when you choose to visit Great
Lakes public beaches.
Next
Save and Return Later

103

Figure A.1 (Cont’d)
Length of Great Lakes public beaches
The length of a beach:




is the distance of shoreline along the Great Lake within the boundaries of the
park or managed area where the beach is located
includes the sandy areas, plus any rocky, muddy, grassy, wetland, and forest
areas along the shore
does not include beach or shoreline outside the boundaries of the park or
managed area

Great Lakes public beaches in Michigan vary in length from 50 yards to over 14,000
yards (over 8 miles). The average Great Lakes public beach length is 2,600 yards (1.5
miles).

Park boundary

Park boundary

Land

Area outside of
park boundary

Area outside of
park boundary

Water

104

Figure A.1 (Cont’d)
Q1. What is the length of the Great Lakes beach you most recently visited?
Less than 100 yards
100 to 300 yards
300 yards to 880 yards (half mile)
880 yards (Half mile) to 1760 yards (one mile)
Over 1760 yards (over one mile)
Don’t know
Q2. How much do you agree or disagree with the following statement?
“The length of a beach is important to my enjoyment of the beach.”
Disagree
Somewhat Disagree
Neither Agree or Disagree
Somewhat Agree
Agree
Next
Save and Return Later

105

Figure A.1 (Cont’d)
Bacteria testing in the water at Great Lakes public beaches
Unsafe bacteria may come from sources such as sewers, septic systems, farms, and
manure. Rivers and streams near these sources can carry unsafe bacteria into the
Great Lakes.
Public agencies can test the water at beaches for unsafe bacteria.
When beachgoers come in contact with high levels of unsafe bacteria they are at risk of
becoming ill. Unsafe bacteria can cause skin, eye, or ear irritation, vomiting, diarrhea,
cramps, or other stomach discomfort.
If the levels of unsafe bacteria are too high, health agencies can post signs at the beach
and publish messages recommending that beachgoers limit or avoid contact with the
water at that beach.
Testing Great Lakes public beaches for unsafe bacteria costs money. Not all public
beaches are tested for unsafe bacteria because of the large number of beaches and the
limited budgets of public health agencies.
Great Lakes public beaches can be tested for unsafe bacteria:





Daily
Weekly
Monthly
Never

106

Figure A.1 (Cont’d)
Q1. Have you heard about a Great Lakes public beach being closed because
of high levels of unsafe bacteria?
Yes
No

Q2. How much do you agree or disagree with the following statement?
“I do not care if the Great Lakes public beach I visit is tested for unsafe bacteria
because…”

… I do not go in the water.*
… the beach has been tested in the past
and no unsafe bacteria were found.*
… the beach is not close to any streams
or rivers.*
… the beach is not close to any shoreline
farms or development.*
Next
Save and Return Later
* The column headings to the right of each sub-question were labeled were slanted text
which read (from left to read): “Disagree, Somewhat Disagree, Neither Agree or
Disagree, Somewhat Agree, Agree.”

107

Figure A.1 (Cont’d)
Algae and Great Lakes public beaches
Algae are organisms that grow naturally throughout the Great Lakes. It can be common
for algae to be present at Great Lakes beaches.
Some types of algae





are invisible to the naked eye
look like a layer of film on the surface of the water
appear plant-like with stems and leafy branches
look like muck or mud

Algae in the water
Some algae grow at the Great Lakes beach where visitors swim or walk. Other algae
may float into areas where visitors swim or walk.
At beaches with…
Amount of
algae in
the water
Definition

View of the Water

None

visitors never come in contact
with algae while swimming or
wading

Low

visitors rarely come in contact
with algae while swimming or
wading

Moderate

visitors occasionally come in
contact with algae while
swimming or wading

High

visitors constantly come in
contact with algae while
swimming or wading

108

Figure A.1 (Cont’d)
Q1. Using the definitions above, how much algae was present in the water
at the Great Lakes beach you most recently visited?
None
Low amounts
Moderate Amounts
High amounts
Don’t know
Q2. How much do you agree or disagree with the following statement?
"The amount of algae in the water affects my enjoyment of a beach"
Disagree
Somewhat Disagree
Neither Agree or Disagree
Somewhat Agree
Agree
Next
Save and Return Later

109

Figure A.1 (Cont’d)
Algae on the Shore
Algae may die or be knocked loose from the place where it grows. Dead algae can be
carried by winds and waves onto the shore of Great Lakes beaches.
When algae washes onto shore, it can accumulate in bunches.
At beaches with:
Amount of
algae on the
shore

Definition

View of swimming area shore

None

None of the shore of
the swimming area has
algae.

Low

1 to 20% of the shore of
the swimming area has
algae.

Moderate

21 to 50% of the shore
of the swimming area
has algae.

High

More than 50% of the
shore of the swimming
area has algae.

.
110

Figure A.1 (Cont’d)
Q1. Using the definitions above, how much algae was present on the shore
at the Great Lakes beach you most recently visited?
None
Low amounts
Moderate Amounts
High amounts
Don’t know
Q2. How much do you agree or disagree with the following statement?
"The amount of algae on the shore affects my enjoyment of a beach"
Disagree
Somewhat Disagree
Neither Agree or Disagree
Somewhat Agree
Agree
Next
Save and Return Later

111

Figure A.1 (Cont’d)
Bathrooms at Great Lakes public beaches
Great Lakes public beaches can have different types of bathrooms:





Outhouses (toilets do not flush), cleaned daily
Outhouses (toilets do not flush), cleaned hourly
Flushing toilets, cleaned daily
Flushing toilets, cleaned hourly

Q1. What type of bathrooms are available at the Great Lakes beach you most recently
visited?
It does not have any bathrooms
Outhouses/portable toilets
Flushing toilets
Don’t know

Q2. Do you know how often the bathrooms at the Great Lakes beach you most recently
visited get cleaned?
Hourly
Daily
Weekly
Don’t Know
Next
Save and Return Later

112

Figure A.1 (Cont’d)
Distance from your home to Great Lakes public beach
The distance from your home to a Great Lakes public beach is the number of miles from
your permanent residence to get to the Great Lakes beach.
This table shows the typical miles of driving distance from the following cities to different
Great Lakes:

Lake Michigan

Lake Huron
Lake St Clair Lake Erie

Lake Superior

Detroit

186

346

61

6

25

Lansing

95

287

100

96

110

Grand Rapids

30

285

130

160

175

Gaylord

37

115

49

235

255

Q1. About how many miles is it from your permanent residence to the
Great Lakes beach you most recently visited?
Number of miles
Q2. How much do you agree or disagree with the following statement?
"The distance from my home to a beach affects how often I visit a beach.”
Disagree
Somewhat Disagree
Neither Agree or Disagree
Somewhat Agree
Agree
Next
Save and Return Later

113

Figure A.1 (Cont’d)
Next, we are going to compare beaches that have some of the characteristics that
we have asked about.
You are 75% of the way through the survey.
Next
Save and Return Later

114

Figure A.1 (Cont’d)
Visiting a Great Lakes Beach
Suppose you are taking a trip to the beach and there are only two beaches to choose
from. The beaches have different characteristics as shown in the table, but otherwise,
they are the same. For example, they would have the same amount of litter, the same
amount of crowding, and the same scenery. Please compare Beach A and Beach B in
the table and answer the question below:
Beach A

Beach B

Great Lake

Lake Huron

Lake Michigan

Bathrooms

Flushing toilets
cleaned hourly

Flushing toilets
cleaned hourly

Algae
in the water

Moderate
(occasionally come in
contact with algae)

Low
(rarely come in
contact with algae)

Algae
on the shore

Low
(1-20% of the
shore has algae.)

None

Length of beach

5280 yards (3 miles)

880 yards (1/2 mile)

Testing water
for bacteria

Daily

Weekly

Distance from
your home

48

169

Which of the above
beaches would you visit?

Beach A

Next
Save and Return Later

115

Beach B

Figure A.1 (Cont’d)
Visiting a Great Lakes Beach
Now suppose you are taking a trip to the beach and Beach C and Beach D are the only
beaches to choose from. The beaches have different characteristics as shown in the
table, but otherwise, they are the same. For example, they would have the same
amount of litter, the same amount of crowding, and the same scenery. Please compare
Beach C and Beach D in the table and answer the question below:
Beach C

Beach D

Great Lake

Lake Huron

Lake Erie

Bathrooms

Flushing toilets
cleaned hourly

Flushing toilets
cleaned hourly

Algae
in the water

Low
(rarely come in
contact with algae)

None

Algae
on the shore

Low
(1-20% of the
shore has algae.)

High
(more than 50% of the
shore has algae.)

Length of beach

220 yards (1/8 miles)

5280 yards (3 miles)

Testing water
for bacteria

Daily

Weekly

Distance from
your home

25

188

Which of the above
beaches would you visit?

Beach C

Next
Save and Return Later

116

Beach D

Figure A.1 (Cont’d)
Visiting a Great Lakes Beach
Finally, suppose you are taking a trip to the beach and Beach E and Beach F are the
only beaches to choose from. The beaches have different characteristics as shown in
the table, but otherwise, they are the same. For example, they would have the same
amount of litter, the same amount of crowding, and the same scenery. Please compare
Beach E and Beach F in the table and answer the question below:
Beach E

Beach F

Great Lake

Lake Michigan

Lake St. Clair

Bathrooms

Flushing toilets
cleaned daily

Flushing toilets
cleaned daily

Algae
in the water

Low
(rarely come in
contact with algae)

None

Algae
on the shore

High
(more than 50% of the
shore has algae.)

None

Length of beach

220 yards (1/8 miles)

50 yards

Testing water
for bacteria

Daily

Never

Distance from
your home

365

178

Which of the above
beaches would you visit?

Beach E

Next
Save and Return Later

117

Beach F

Figure A.1 (Cont’d)
Thank you for your input. You are almost done with the survey.

You are 85% of the way through the survey
Next
Save and Return Later

118

Figure A.1 (Cont’d)
Your opinions about Great Lakes beaches
Please rate the following beach characteristics of each Great Lake.

Q1. Water quality*
(clarity, color,
and odor)
Lake Erie
Lake Huron
Lake Michigan
Lake Superior
Lake St. Clair
Q2. Level of algae*
Lake Erie
Lake Huron
Lake Michigan
Lake Superior
Lake St. Clair
Q3. Debris in the*
water and
on the shore
Lake Erie
Lake Huron
Lake Michigan
Lake Superior
Lake St. Clair

O
O
O
O
O

O
O
O
O
O

O
O
O
O
O

O
O
O
O
O

O
O
O
O
O

O
O
O
O
O

O
O
O
O
O

O
O
O
O
O

O
O
O
O
O

O
O
O
O
O

O
O
O
O
O

O
O
O
O
O

O
O
O
O
O

O
O
O
O
O

O
O
O
O
O

O
O
O
O
O

O
O
O
O
O

O
O
O
O
O

* Column headings to the right of each question were listed in slanting text and read
(from left to right): “Satisfactory, Somewhat satisfactory, Neither Satisfactory nor
unsatisfactory, Somewhat unsatisfactory, Unsatisfactory, Don’t know.”

119

Figure A.1 (Cont’d)

Q4. Natural Beauty*
Lake Erie
Lake Huron
Lake Michigan
Lake Superior
Lake St. Clair
Q5. Quality and
texture of the sand
or beach areas*
Lake Erie
Lake Huron
Lake Michigan
Lake Superior
Lake St. Clair
Q6. Crowding at
the beach and
surrounding area*
Lake Erie
Lake Huron
Lake Michigan
Lake Superior
Lake St. Clair

O
O
O
O
O

O
O
O
O
O

O
O
O
O
O

O
O
O
O
O

O
O
O
O
O

O
O
O
O
O

O
O
O
O
O

O
O
O
O
O

O
O
O
O
O

O
O
O
O
O

O
O
O
O
O

O
O
O
O
O

O
O
O
O
O

O
O
O
O
O

O
O
O
O
O

O
O
O
O
O

O
O
O
O
O

O
O
O
O
O

* Column headings to the right of each question were listed in slanting text
and read (from left to right): “Satisfactory, Somewhat satisfactory, Neither
Satisfactory nor unsatisfactory, Somewhat unsatisfactory, Unsatisfactory,
Don’t know.”

120

Figure A.1 (Cont’d)

Q7. When was the last
time you went to this
Great Lake ^
Lake Erie
Lake Huron
Lake Michigan
Lake Superior
Lake St. Clair

O
O
O
O
O

O
O
O
O
O

O
O
O
O
O

O
O
O
O
O

O
O
O
O
O

Next
Save and Return
Later

^ Column headings to the right of each question were listed in slanting text and read
(from left to right): “Within the last year, 1 to 10 years ago, More than 10 years ago,
Never, Don’t know.”

121

Figure A.1 (Cont’d)
Things you typically do at a Great Lakes beach
Q1. How often do you do the
following things on a typical visit to a
Great Lakes beach?

Never

Sometimes

Most of
the time

I walk on the sand
I play or dig in the sand
I get my feet wet
I go in the water, but not past my
knees
I go in the water, but not past my
shoulders
I go in the water and my entire body
gets wet

O
O
O
O

O
O
O
O

O
O
O
O

O

O

O

O

O

O

Next
Save and Return Later

122

Figure A.1 (Cont’d)
Concluding Section: Background Information
Summaries of these background questions help us accurately represent all
Michigan residents. Individual answers are strictly confidential.
Q1. Who is filling out this survey?
The person the invitation was addressed to
Another household member
Someone else
Q2. What is your sex?
Male
Female
Q3. In what year were you born?
Year
Q4. What is your race and/or ethnicity?
Mark all that apply.
White
Black / African American
Hispanic, Latino, or Spanish
American Indian
Asian
Other

123

Figure A.1 (Cont’d)
Q5. What is the highest degree or level of school you have completed?
Some schooling
High school or equivalent
Some college, no degree
Associate’s degree
Bachelor’s degree
Graduate or professional degree
Q6. What is the zip code of the place you live?
zip code
Q7. How long have you lived in Michigan?
Less than 1 year
1 to 5 years
More than 5 years
Q8. What is your current employment status?
Employed Full Time
Employed Part Time
Unemployed
Stay at home parent
Retired
Student

124

Figure A.1 (Cont’d)
(If Q8= “Employed Full Time” or “Employed Part Time”)
How many hours do you work in a typical week?
Q9. Do any of the following live in your household?
Spouse or significant other
Children age 5 and under
Children age 6 to 17
Other immediate family
Extended family or other adults
None of these
Q10. In 2011, what was your total household income, from all sources, before taxes?
Less than $24,999
$25,000 to $34,999
$35,000 to $49,999
$50,000 to $74,999
$75,000 to $99,999
$100,000 to $149,999
$150,000 to $199,999
$200,000 or more
Next
Save and Return Later

125

Figure A.1 (Cont’d)
If Q10 was left blank:
One last follow up
Q1. Was your total household income in 2011 less than $50,000?
Yes
No

If Q1=Yes:
Q1b. Was it less than $25,000?
Yes
No
If Q1=No:
Q1b. Was it less than $100,000?
Yes
No
Next
Save and Return Later

126

APPENDIX B

Screener Survey (Michigan Activities survey) materials, contact schedule, robodial
scripts, and disposition

127

PLEASE NOTE: The figures in this appendix include images of survey materials that do
not meet the font size and/or margin requirements listed in the Michigan State
University formatting guide for submission of master’s theses and doctoral dissertations.
As such, the text of the materials featured in these figures is entered plainly above each
figure to preserve all information contained within the figures and to meet the standards
in the formatting guide. The images are presented in the figures to preserve a record of
overall formatting of actual materials.
The following is the text for the Michigan Activities Survey Wave 1 Introduction letter,
original size 8.5” x 11”; for an image of this letter see Figure B.1.
Text for Michigan Activities Survey Wave 1 Introduction letter, original size 8.5” x 11”:
DATE
<NAME>
<STREET>
<CITY>, <STATE> <ZIP>
Dear <INITIAL CAPS NAME>,
We need your help with a study of recreation and leisure activities in Michigan.
By answering the enclosed survey, you will help shape future policies and
programs influencing the quality of life of people in Michigan.
You are part of a small scientific sample. We need your answers to ensure that
our results accurately represent Michigan residents.
Whether or not you take part in the activities listed in the questionnaire, your
input is important. Please complete the questionnaire and return it to us in the
enclosed prepaid envelope.
The survey should take less than five minutes to complete. Your answers will
remain completely confidential. Your privacy will be protected to the maximum
extent allowable by law. Your participation is voluntary and you may choose not
to participate at all, or not to answer certain questions.
If you have any concerns or questions about this research study, such as
scientific issues, how to do any part of it, or if you believe you have been
harmed because of the research, please contact Frank Lupi, Department of
Agricultural, Food, and Resource Economics, Michigan State University, 301B
Agriculture Hall, East Lansing, MI 48824-1039; MIstudy@msu.edu, 517-3551692.
If you have any questions or concerns about your role and rights as a research
participant, or would like to register a complaint about this research study, you
may contact, anonymously if you wish, Michigan State University Human
128

Research Protection Program at 517-355-2180, FAX 517-432-4503, or e-mail
irb@msu.edu, or regular mail at: 207 Olds Hall, Michigan State University, East
Lansing, MI 48824.
Thank you very much for helping with this important study.
Sincerely,
Frank Lupi, Professor

129

Figure B.1: Image of Michigan Activities Survey Wave 1 Introduction letter, original size
8.5” x 11”

130

The following is the text for the Michigan Activities Survey Wave 1 survey instrument,
original size of each page 8.5” x 7”; for an image of this survey, see figure B.2
Text for Michigan Activities Survey Wave 1 survey instrument, original size of each
page 8.5” x 7”
Michigan Activities Survey
We need your help!
“By taking five minutes to tell us about your activities, you will help us manage
our state’s cultural and natural resources to better meet peoples’ needs.”
--Rodney Stokes, Director, Michigan Department of Natural Resources.
“Your input will help us better connect people with Pure Michigan’s unique
attractions, renowned destinations, festive celebrations, and pristine outdoors.”
--George Zimmermann, Vice President, Travel Michigan
INSTRUCTIONS
Please answer each question for your own activities that were:
• in the State of Michigan
• since Memorial Day weekend 2010 (since June 1, 2010)
• not work related
Even if you have not done these activities, we need to hear from you.
1. Have you done any of the following in the State of Michigan in the past year, since
June 1, 2010? (YES/ NO); (If YES, how often since June 1, 2010? 1-5 days/ 6-19 days/
20 or more
Eat dinner at a restaurant
Go for a walk or a hike
Attend or participate in outdoor sports
Swim at a pool, lake, or river
Go to a movie in a theater
Attend a music concert
Attend a cultural or arts festival/fair
(page 2) Park Visits
2. Have you visited any of the following public parks in the State of Michigan since June
1, 2010? (YES/ NO); (If YES, how often since June 1, 2010? 1-5 days/ 6-19 days/ 20 or
more)
County, City, or Township Park
State Park or State Campground
State Forest or State Game Area
National Park or National Forest

131

Outdoor Recreation
3. Have you participated in any of the following activities in the State of Michigan since
June 1, 2010? (YES/ NO); (If YES, how often since June 1, 2010? 1-5 days/ 6-19 days/
20 or more)
Camping
Hunting
Fishing
Boating
Picnicking at public parks
Visiting a beach
Driving an all-terrain vehicle (ATV)
Snowmobiling
Skiing or snowboarding
Great Lakes Recreation
4. Have you participated in any of the following activities on the Great Lakes in the State
of Michigan since June 1, 2010? (YES/ NO); (If YES, how often since June 1, 2010? 1-5
days/ 6-19 days/ 20 or more)
Visiting a beach on the Great Lakes
Fishing on the Great Lakes
Boating on the Great Lakes
(page 3) Typical Weekly Activity
5. During a typical week, do you do any of the following activities? (YES/NO); (If YES,
how often during a typical week? 1-9 hours per week/ 10-19 per week/ 20-29 hours per
week/ 30+ hours per week)
Read books
Indoor / outdoor exercise
Watch television
Use the internet
Play video games
Play a musical instrument
Volunteer
Obstacles
6a. Were you able to participate in outdoor recreational activities as often as you would
have liked since June 1, 2010? (YES/NO) If NO Skip to question 7.
6b. How much do you agree or disagree? “Since June 1, 2010, I would have
participated in outdoor recreation more often, except…” (Disagree/ Somewhat Disagree/
Neither Agree or Disagree/ Somewhat Agree/ Agree)

132

… I did not have enough time
… I had too many family or work obligations
… I did not have enough money / cost too much
… I had a personal health issue or disability
… I had no one to go with
… I did not know what activities were available
… I did not know what places to go to
... the places I knew of were poorly maintained
... the places I knew of were too far away
... the places I knew of were too crowded
(page 4) Please provide the following background information
7. Who is filling out this survey?
(The person it is addressed to / Another household member/ Someone else)
8. What is your sex?
(Male / Female)
9. In what year were you born? (19 _ _ )
10. What is your race and/or ethnicity? Mark all that apply.
White
Black / African American
Hispanic, Latino, or Spanish
American Indian
Asian
Other _____________
11. What is the highest degree or level of school you have completed?
No schooling
Some schooling
High school or equivalent
Associate’s or technical degree
Bachelor’s degree
Advanced degree
12. Where do you live?
Lower Peninsula of Michigan
Upper Peninsula of Michigan
Some other state, not Michigan
13. How long have you lived in Michigan?
Less than 1 year
1 to 5 years
More than 5 years
133

14. What is your current employment status?
Employed Full Time
Employed Part Time
Unemployed
Stay at home parent
Retired
Student
15. If employed, how many hours do you work in a typical week?
Hours per week _ _
16. Do any of the following live in your household? Mark all that apply.
Spouse or significant other
Children age 5 and under
Children age 6 to 17 years old
Other immediate family
Extended family or other adults
None of these
17. In 2010, what was your total household income, from all sources, before taxes?
Less than $25,000
$25,000 to $49,999
$50,000 to $99,999
$100,000 and higher
THANK YOU FOR YOUR HELP. If you have misplaced the return envelope, please
return your survey to: Dr. Frank Lupi, 301b Agriculture Hall, East Lansing, MI 488241039

134

Figure B.2: Image of Michigan Activities Survey Wave 1 survey instrument, original size
of each page 8.5” x 7”

135

Figure B.2 (Cont’d)

136

Figure B.2 (Cont’d)

137

Figure B.2 (Cont’d)

138

The following is the text for the Michigan Activities Survey Wave 2 Introduction letter,
original size 8.5” x 11”; for an image of this letter see Figure B.3.
Text for Michigan Activities Survey Wave 2 Introduction letter, original size 8.5” x 11”:
DATE
<NAME>
<STREET>
<CITY>, <STATE> <ZIP>
Dear <INITIAL CAPS NAME>,
Three weeks ago, I sent you a questionnaire for an important research study of
recreation and leisure activities in Michigan. If you have already completed
and returned your survey, thank you very much! If not, please do so today.
Whether or not you take part in the activities listed in the questionnaire,
your input is important.
Your answers will help Michigan ensure the state’s resources are managed for
the betterment of Michigan residents.
Because you are part of a small scientific sample, we need your answers to
be sure that our results accurately represent Michigan residents.
The survey should only take a few minutes to complete and your answers will
remain completely confidential. Participation in the study is voluntary, and you
may choose not to participate at all, or choose not to answer certain questions.
If you have any concerns or questions about this research study, please
contact Dr. Frank Lupi, Department of Agricultural, Food and Resource
Economics, Michigan State University, 301B Ag Hall, East Lansing, MI 48824;
MIstudy@msu.edu, 517-355-1692.
Thank you very much for helping with this important study.
Sincerely,
Frank Lupi, Professor

139

Figure B.3: Image of Michigan Activities Survey Wave 2 Introduction letter, original size
8.5” x 11”

140

Note: the Michigan Activities Survey Wave 1 survey instrument and the Michigan
Activities Survey Wave 2 survey instrument are identical. For full text of Michigan
Activities Survey Wave 2 survey instrument, see text appearing in this appendix before
figure B.2.For an image of Michigan Activities Survey Wave 2 survey instrument, see
figure B.4.

141

Figure B.4: Image of Michigan Activities Survey Wave 2 survey instrument, original size
of each page 8.5” x 7”

142

The following is the text for the Michigan Activities Survey Wave 3 Introduction letter,
original size 8.5” x 11” ; for an image of this letter see figure B.5.
DATE
<Name>
<Street>
<City>, <STATE> <ZIP>
Dear <Name>,
This summer, I have tried to contact you about an important study of recreation
and leisure activities in Michigan. If you have returned your questionnaire,
thank you very much!
I know you are busy, but we need your answers to ensure our results
accurately represent Michigan residents. Whether or not you take part in the
activities listed in the questionnaire, your input matters.
This is the last questionnaire I plan to mail you. Please respond today.
The questions take only a few minutes to complete, and your answers will be
kept completely confidential. Participation in the study is voluntary, and you
may choose not to participate at all, or choose not to answer certain questions.
If you have concerns or questions about this study, contact me at the
Department of Agricultural, Food and Resource Economics, Michigan State
University, East Lansing, MI 48824; 517-355-1692, MIstudy@msu.edu.
Thank you,
Frank Lupi, Professor
Answers to Frequently Asked Questions
Why does this survey matter?
Surprisingly, there is little scientifically sound information on the
recreational and leisure activities of Michigan residents. Yet, decisions
must get made about how to manage the State’s natural and cultural
resources. The information this survey gathers on people’s recreational
and leisure activities will facilitate fact-based decision making about
resource management.
Why do you want me to do the survey?
We need your help because you are part of a small, scientifically
selected sample, designed to be representative of all Michigan
residents. Some people do the activities in the survey, and others do
143

not. Either way, we need to hear from everyone selected to ensure the
accuracy of our results.
How was I selected?
A computer program randomly selected your name and address from drivers’
license and State ID holders in Michigan over the age of 18.
Who sees my answers?
Your responses are scanned directly into a computer by a small team of
researchers. The scanned data does not contain your name or address.
Personal information is only used to manage the mailing and collection of
surveys.
How is my privacy protected?
Your answers are kept separately from our mailing list with your name and
address. Our mailing list is stored on password protected in computers in
locked offices. Everyone who works on the survey has completed training and
signed an oath saying that they will not share any private information they see
as part of their work on the survey.
How can I see the results?
Contact our research team by email (MIstudy@msu.edu) or by phone (517355-1692) or visit our website: www.msu.edu/~mistudy

144

Figure B.5: Image of Michigan Activities Survey Wave 3 Introduction letter, original size
8.5” x 11”

145

Figure B.5 (Cont’d)

146

Note, the Michigan Activities Survey Wave 3 survey instrument differed from the
Michigan Activities Survey Wave 1 survey instrument and Michigan Activities Survey
Wave 2 survey instrument on pages 1 and 4. The following contains the full text ONLY
for those pages of Michigan Activities Survey Wave 3 survey instrument that are
different from Michigan Activities Survey Wave 1 survey instrument and Michigan
Activities Survey Wave 2 survey instrument (the full text of both wave 1 and wave 2
surveys is available above figure B.2). For an image of Michigan Activities Survey Wave
3 survey instrument, see figure B.6.
(page 1) We need your help!
Your answers matter whether or not you do the things in the survey.
You are part of a small, scientific sample chosen to participate
We need your input to ensure the results are accurate.
INSTRUCTIONS
Please answer each question for your own activities that were:
• in the State of Michigan
• since Memorial Day weekend 2010 (since June 1, 2010)
• not work related
Even if you have not done these activities, we need to hear from you.
1. Have you done any of the following in the State of Michigan in the past year, since
June 1, 2010? (YES/ NO); (If YES, how often since June 1, 2010? 1-5 days/ 6-19 days/
20 or more
Eat dinner at a restaurant
Go for a walk or a hike
Attend or participate in outdoor sports
Swim at a pool, lake, or river
Go to a movie in a theater
Attend a music concert
Attend a cultural or arts festival/fair
(page 4) Summaries of the following questions help us represent all MI residents.
Individual answers are strictly CONFIDENTIAL.
7. Who is filling out this survey?
(The person it is addressed to / Another household member/ Someone else)
8. What is your sex?
(Male / Female)
9. In what year were you born? (19 _ _ )

147

10. What is your race and/or ethnicity? Mark all that apply.
White
Black / African American
Hispanic, Latino, or Spanish
American Indian
Asian
Other _____________
11. What is the highest degree or level of school you have completed?
No schooling
Some schooling
High school or equivalent
Associate’s or technical degree
Bachelor’s degree
Advanced degree
12. Where do you live?
Lower Peninsula of Michigan
Upper Peninsula of Michigan
Some other state, not Michigan
13. How long have you lived in Michigan?
Less than 1 year
1 to 5 years
More than 5 years
14. What is your current employment status?
Employed Full Time
Employed Part Time
Unemployed
Stay at home parent
Retired
Student
15. If employed, how many hours do you work in a typical week?
Hours per week _ _
16. Do any of the following live in your household? Mark all that apply.
Spouse or significant other
Children age 5 and under
Children age 6 to 17 years old
Other immediate family
Extended family or other adults
None of these

148

17. In 2010, what was your total household income, from all sources, before taxes?
Less than $25,000
$25,000 to $49,999
$50,000 to $99,999
$100,000 and higher
THANK YOU FOR YOUR HELP. If you have misplaced the return envelope, please
return your survey to: Dr. Frank Lupi, 301b Agriculture Hall, East Lansing, MI 488241039

149

Figure B.6: Image of Michigan Activities Survey Wave 3 survey instrument, original size
of each page 8.5” x 7” (Note: pages 1 and 4 of Wave 3 survey instrument are different
from Wave 1 and Wave 2)

150

Figure B.6 (Cont’d)

151

Table B.1: Schedule of Michigan Activity Survey Contacts
Wave number and contact type
Wave 1 intro letter and survey instrument mailing
Wave 2 intro letter and survey instrument mailing
Wave 2 "pre-notice" call
Wave 2 "reminder" call
Wave 3 intro letter and survey instrument mailing
Wave 3 "pre-notice" call
Wave 3 "concurrent" call
Wave 3 "reminder" call
* the date mailings were post-marked

date
7/12/2011 *
8/5/2011 *
8/5/2011
8/10/2011
9/23/2011 *
9/23/2011
9/25/2011
9/28/2011

Wave 2 Screener Survey robodial script: pre-notice
I’m Professor Frank Lupi from Michigan State University. Your household was recently
sent a short survey about leisure activities. If it was returned, thank you. In case it was
misplaced, I mailed you another copy which should arrive soon.
I know you’re busy, but we need your help. Whether or not you do the activities in the
survey, we need to hear from you to ensure the accuracy of our results.
For survey help, e-mail mistudy@msu.edu or call 517-355-1692
Thanks.

152

Wave 2 Screener Survey robodial script: reminder:
I’m Professor Frank Lupi from Michigan State University. Your household was recently
sent a short survey about leisure activities. If it was returned, thank you. In case it was
misplaced, I mailed you another copy which should have arrived in the last day or two
soon.
I know you’re busy, but we need your help. Whether or not you do the activities in the
survey, we need to hear from you to ensure the accuracy of our results.
For survey help, e-mail mistudy@msu.edu or call 517-355-1692
Thanks.

Wave 3 Screener Survey robodial script: pre-notice:
I’m Professor Frank Lupi from Michigan State University. Your household was recently
sent a short survey about leisure activities. If it was returned, thank you. In case it was
misplaced, I mailed you another copy which should arrive soon. That is the last copy of
the survey we plan to mail you.
I know you’re busy, but we need your help. Whether or not you do the activities in the
survey, we need to hear from you to ensure the accuracy of our results.
For survey help, or for another copy of the survey, call 517-355-1692
Thanks.

153

Wave 3 Screener Survey robodial script: concurrent:
I’m Professor Frank Lupi from Michigan State University. Your household was recently
sent a short survey about leisure activities. If it was returned, thank you. In case it was
misplaced, I mailed you another copy. That is the last copy of the survey we plan to mail
you.
I know you’re busy, but we need your help. Whether or not you do the activities in the
survey, we need to hear from you to ensure the accuracy of our results.
For survey help, or for another copy of the survey, call 517-355-1692
Thanks.

Wave 3 Screener Survey robodial script: reminder:
I’m Professor Frank Lupi from Michigan State University. Your household was recently
sent a short survey about leisure activities. If it was returned, thank you. In case it was
misplaced, I mailed you another copy which should have arrived in the last day or two.
That is the last copy of the survey we plan to mail you.
I know you’re busy, but we need your help. Whether or not you do the activities in the
survey, we need to hear from you to ensure the accuracy of our results.
For survey help, or for another copy of the survey, call 517-355-1692
Thanks.
Table B.2: Summary of Michigan Activity Survey response and disposition
Initial sample size
32,230
Deceased
78
Undeliverable
2,434
Moved
106
Response
11,028
Wave 1 responses
5522
Wave 2 responses
3061
Wave 3 responses
2445
Refusals
61 (51 blank
responses
counted as
refusals)

154

APPENDIX C

Web Survey (Great Lakes beach web survey) materials, contact schedule, and robodial
scripts

155

PLEASE NOTE: The figures in this appendix include images of survey materials that do
not meet the font size and/or margin requirements listed in the Michigan State
University formatting guide for submission of master’s theses and doctoral dissertations.
As such, the text of the materials featured in these figures is entered plainly above each
figure to preserve all information contained within the figures and to meet the standards
in the formatting guide. The images are presented in the figures to preserve a record of
overall formatting of actual materials.
The following is the text for the Great Lakes Beach Web survey Wave 1 Introduction
letter, original size 8.5” x 11”. For an image of this letter, see figure C.1
DATE
<Name>
<Street>
<City>, <STATE> <ZIP>
Dear <Name>,
We need your help with a survey about Great Lakes beaches.
By taking a few minutes to share your thoughts and opinions you will be
helping us out a great deal. The results will be used by local and state
governments to make management decisions that affect Michigan residents.
No matter how often you go to Great Lakes beaches, your input is essential to
ensure the accuracy of our results. Your answers are strictly confidential.
The survey is being conducted on the internet. To access the survey, please
visit the web address below, and log on using your unique password:
Web Address: beaches.anr.msu.edu
Password: <PASSWORD>
We have included a small token of appreciation as a way of saying thank you
for your help.
If you have questions about this study, contact me at 301B Ag Hall, Michigan
State University, East Lansing, MI 48824; 517-355-1692, MIstudy@msu.edu.
Sincerely,
Frank Lupi, Professor
(page 2)
Answers to Frequently Asked Questions

156

How was I selected?
You helped in an earlier survey where a computer program randomly
selected names and addresses from licensed drivers in Michigan. Now, you
are part of a small group selected to participate in final follow-up survey about
Great Lakes beaches.
Will I be contacted about other surveys from you?
No, this is the last survey that we will ask you to take. We know that you are
busy and greatly appreciate your help with this important research project.
Why does this survey matter?
Surprisingly, there is little scientifically sound information on the Great Lakes
beach activities of Michigan residents. Yet, decisions must get made about
how to manage this vast natural resource. The information this survey
gathers on people’s Great Lakes beach activities will facilitate fact-based
decision making about resource management.
Why do you want me to do the survey?
We need your help because you are part of a small, scientifically selected
sample, designed to be representative of all Michigan residents. Some
people go to Great Lakes beaches frequently, and others do not. We need to
hear from everyone selected to ensure the accuracy of our results.
Who sees my answers?
Your responses are saved directly into a database that does not contain your
name or address. Personal information is only used to manage the mailing of
survey invitations.
How is my privacy protected?
Your answers are kept separately from our mailing list. Our mailing list and
data are stored on password protected computers in locked offices.
Everyone who works on the survey has completed training and signed an
oath saying that they will not share any private information they see working
on the survey.
How do I get help with web survey access or other problems?
If you have trouble accessing the web survey or if you have other technical
issues you should contact our research team by email (MIstudy@msu.edu) or
by phone (517-355-1692). One of our research assistants will help you.
How can I see the results?
Contact our research team by email (MIstudy@msu.edu) or by phone (517355-1692) or visit our website in a few months: www.msu.edu/~mistudy

157

Figure C.1: Image of Great Lakes Beach Web survey Wave 1 Introduction letter, original
size 8.5” x 11” (Note: all wave 1 survey invitation were mailed with a $1 bill)

158

Figure C.2 (Cont’d)

159

Figure C.2: Great Lakes Beach Web survey Wave 2 reminder, half sheet black and
white postcard, front, original size 8.5” x 5.5”

We need your help!
Please complete the
Great Lakes beach survey
Website: beaches.anr.msu.edu
Password: <PASSWORD>

<NAME>
<ADDRESS>
<CITY, STATE, ZIP>

Figure C.3: Image of Great Lakes Beach Web survey Wave 2 reminder, half sheet black
and white postcard, back, original size 8.5” x 5.5”
Recently, I contacted you about an internet survey on Great Lakes beaches.
If you have already answered the survey, thank you very much!
If you have not filled out the survey, we still need your help. You are
part of a small scientific sample, and your answers help us represent
all people in Michigan.
To access the survey, please use the web address and password
printed on the other side of this card.
For questions about the survey, email: mistudy@msu.edu
or call: 517-355-1692.
Thank you very much for your help with this important research study!
Sincerely,

Frank Lupi, Professor

160

Figure C.4: Image of Great Lakes Beach Web survey Wave 3 reminder, quarter sheet
color postcard, front, original size 5.5” x 4.25”

Figure C.5: Image of Great Lakes Beach Web survey Wave 3 reminder, quarter sheet
color postcard, back, original size 5.5” x 4.25”

161

The following is the complete text for the Great Lakes Beaches Web Survey Wave 4
reminder letter, $20 post-paid incentive, original size 8.5” x 11”. For an image of this
letter, see figure C.6.
DATE
<Name>
<Street>
<City>, <STATE> <ZIP>
Dear <Name>,
We still need your help!
This spring we contacted you several times about the Great Lakes beaches survey. The
survey ends May 25, 2012, so please respond soon.
You are part of a small, scientific sample and we need your input to help us improve
Michigan’s Great Lakes beaches.
If you complete the survey by May 25, 2012, we will send you a $20 check.
To access the survey, please visit the web address below, and log on using your unique
password:
Web Address: beaches.anr.msu.edu
Password: <PASSWORD>
If you do not have access to the Internet, it is critical that you let us know by sending the
enclosed prepaid postcard so that our records are correct.
If you have questions about this study, contact me at 301B Ag Hall, Michigan State
University, East Lansing, MI 48824; 517-355-1692, MIstudy@msu.edu.
Sincerely,
Frank Lupi, Professor
P.S. This is the last time we will contact you; please respond today.
(page 2)
Answers to Frequently Asked Questions
How was I selected?

162

You helped in an earlier survey where a computer program randomly
selected names and addresses from licensed drivers in Michigan. Now, you
are part of a small group selected to participate in final follow-up survey
about Great Lakes beaches.
Will I be contacted about other surveys from you?
No, this is the last survey that we will ask you to take. We know that you are
busy and greatly appreciate your help with this important research project.
Why does this survey matter?
Surprisingly, there is little scientifically sound information on the Great Lakes
beach activities of Michigan residents. Yet, decisions must get made about
how to manage this vast natural resource. The information this survey
gathers on people’s Great Lakes beach activities will facilitate fact-based
decision making about resource management.
Why do you want me to do the survey?
We need your help because you are part of a small, scientifically selected
sample, designed to be representative of all Michigan residents. Some
people go to Great Lakes beaches frequently, and others do not. We need
to hear from everyone selected to ensure the accuracy of our results.
Who sees my answers?
Your responses are saved directly into a database that does not contain
your name or address. Personal information is only used to manage the
mailing of survey invitations.
How is my privacy protected?
Your answers are kept separately from our mailing list. Our mailing list and
data are stored on password protected computers in locked offices.
Everyone who works on the survey has completed training and signed an
oath saying that they will not share any private information they see working
on the survey.
How do I get help with web survey access or other problems?
If you have trouble accessing the web survey or if you have other technical
issues you should contact our research team by email (MIstudy@msu.edu)
or by phone (517-355-1692). One of our research assistants will help you.
How can I see the results?
Contact our research team by email (MIstudy@msu.edu) or by phone (517355-1692) or visit our website in a few months: www.msu.edu/~mistudy
How do I get paid?
Participants who complete the survey by May 25, 2012, will be mailed a $20
check from Michigan State University about six weeks later. The check will
be made payable to the person whose name appears at the top of this
letter.
163

Figure C.6: Image of Great Lakes Beaches Web Survey Wave 4 reminder letter, $20
post-paid incentive, original size 8.5” x 11”

164

Figure C.6 (Cont’d)

165

The following is the complete text for the Great Lakes Beaches Web Survey Wave 4
reminder letter, $10 post-paid incentive, original size 8.5” x 11”. For an image of this
letter, see figure C.7.
DATE
<Name>
<Street>
<City>, <STATE> <ZIP>
Dear <Name>,
We still need your help!
This spring we contacted you several times about the Great Lakes beaches survey. The
survey ends May 25, 2012, so please respond soon.
You are part of a small, scientific sample and we need your input to help us improve
Michigan’s Great Lakes beaches.
If you complete the survey by May 25, 2012, we will send you a $10 check.
To access the survey, please visit the web address below, and log on using your unique
password:
Web Address: beaches.anr.msu.edu
Password: <PASSWORD>
If you do not have access to the Internet, it is critical that you let us know by sending the
enclosed prepaid postcard so that our records are correct.
If you have questions about this study, contact me at 301B Ag Hall, Michigan State
University, East Lansing, MI 48824; 517-355-1692, MIstudy@msu.edu.
Sincerely,
Frank Lupi, Professor
P.S. This is the last time we will contact you; please respond today.
(page 2)
Answers to Frequently Asked Questions
How was I selected?
You helped in an earlier survey where a computer program randomly
selected names and addresses from licensed drivers in Michigan. Now, you
166

are part of a small group selected to participate in final follow-up survey
about Great Lakes beaches.
Will I be contacted about other surveys from you?
No, this is the last survey that we will ask you to take. We know that you are
busy and greatly appreciate your help with this important research project.
Why does this survey matter?
Surprisingly, there is little scientifically sound information on the Great Lakes
beach activities of Michigan residents. Yet, decisions must get made about
how to manage this vast natural resource. The information this survey
gathers on people’s Great Lakes beach activities will facilitate fact-based
decision making about resource management.
Why do you want me to do the survey?
We need your help because you are part of a small, scientifically selected
sample, designed to be representative of all Michigan residents. Some
people go to Great Lakes beaches frequently, and others do not. We need
to hear from everyone selected to ensure the accuracy of our results.
Who sees my answers?
Your responses are saved directly into a database that does not contain
your name or address. Personal information is only used to manage the
mailing of survey invitations.
How is my privacy protected?
Your answers are kept separately from our mailing list. Our mailing list and
data are stored on password protected computers in locked offices.
Everyone who works on the survey has completed training and signed an
oath saying that they will not share any private information they see working
on the survey.
How do I get help with web survey access or other problems?
If you have trouble accessing the web survey or if you have other technical
issues you should contact our research team by email (MIstudy@msu.edu)
or by phone (517-355-1692). One of our research assistants will help you.
How can I see the results?
Contact our research team by email (MIstudy@msu.edu) or by phone (517355-1692) or visit our website in a few months: www.msu.edu/~mistudy
How do I get paid?
Participants who complete the survey by May 25, 2012, will be mailed a $10 check from
Michigan State University about six weeks later. The check will be made payable to the
person whose name appears at the top of this letter.

167

Figure C.7: Image of Great Lakes Beaches Web Survey Wave 4 reminder letter, $10
post-paid incentive, original size 8.5” x 11”

168

Figure C.7 (Cont’d)

169

The following is the complete text for the Great Lakes Beaches Web Survey Wave 4
reminder letter, $10 post-paid incentive, original size 8.5” x 11”. For an image of this
letter, see figure C.8.
DATE
<Name>
<Street>
<City>, <STATE> <ZIP>
Dear <Name>,
We still need your help!
This spring we contacted you several times about the Great Lakes beaches survey. The
survey ends May 25, 2012, so please respond soon.
You are part of a small, scientific sample and we need your input to help us improve
Michigan’s Great Lakes beaches.
To access the survey, please visit the web address below, and log on using your unique
password:
Web Address: beaches.anr.msu.edu
Password: <PASSWORD>
If you do not have access to the Internet, it is critical that you let us know by sending the
enclosed prepaid postcard so that our records are correct.
If you have questions about this study, contact me at 301B Ag Hall, Michigan State
University, East Lansing, MI 48824; 517-355-1692, MIstudy@msu.edu.
Sincerely,
Frank Lupi, Professor
P.S. This is the last time we will contact you; please respond today.
(page 2)
Answers to Frequently Asked Questions
How was I selected?
You helped in an earlier survey where a computer program randomly
selected names and addresses from licensed drivers in Michigan. Now, you

170

are part of a small group selected to participate in final follow-up survey
about Great Lakes beaches.
Will I be contacted about other surveys from you?
No, this is the last survey that we will ask you to take. We know that you are
busy and greatly appreciate your help with this important research project.
Why does this survey matter?
Surprisingly, there is little scientifically sound information on the Great Lakes
beach activities of Michigan residents. Yet, decisions must get made about
how to manage this vast natural resource. The information this survey
gathers on people’s Great Lakes beach activities will facilitate fact-based
decision making about resource management.
Why do you want me to do the survey?
We need your help because you are part of a small, scientifically selected
sample, designed to be representative of all Michigan residents. Some
people go to Great Lakes beaches frequently, and others do not. We need
to hear from everyone selected to ensure the accuracy of our results.
Who sees my answers?
Your responses are saved directly into a database that does not contain
your name or address. Personal information is only used to manage the
mailing of survey invitations.
How is my privacy protected?
Your answers are kept separately from our mailing list. Our mailing list and
data are stored on password protected computers in locked offices.
Everyone who works on the survey has completed training and signed an
oath saying that they will not share any private information they see working
on the survey.
How do I get help with web survey access or other problems?
If you have trouble accessing the web survey or if you have other technical
issues you should contact our research team by email (MIstudy@msu.edu)
or by phone (517-355-1692). One of our research assistants will help you.
How can I see the results?
Contact our research team by email (MIstudy@msu.edu) or by phone (517355-1692) or visit our website in a few months: www.msu.edu/~mistudy

171

Figure C.8: Image of Great Lakes Beaches Web Survey Wave 4 reminder letter, no
post-paid incentive, original size 8.5” x 11”

172

Figure C.8 (Cont’d)

173

Figure C.9: Image of Great Lakes Beaches Web Survey Wave 4 reminder Business
Reply Mail postcard stating respondent does not have the Internet, front, original size
4.25” x 5.5”

Figure C.10: Image of Great Lakes Beaches Web Survey Wave 4 reminder Business
Reply Mail postcard stating respondent does not have the Internet, back, original size
4.25” x 5.5”

174

Web Survey robodial script:
I’m Professor Frank Lupi from Michigan State University. Someone in your household
was invited to an internet survey about Great Lakes beaches.
If you’ve completed it, thank you. If not, I’ve mailed another post card which should
arrive soon. The survey internet address and your password are on the postcard.
I know you’ve busy, but we need your help. Whether you go to Great Lakes beaches a
lot or rarely, we need to hear from you to ensure the accuracy of our results.
If you need help with the survey, call 517-355-1692. Again that’s 517-355-1692.
Thank you so much.
Table C.1: Schedule of Great Lakes Beaches survey contacts
Web Survey contact 1 invitation letter
(plain paper, $1 token included)

April 12

Web Survey contact 2 reminder postcard
(half sheet postcard, black and white)

April 20

Web Survey contact 3 reminder postcard
(quarter sheet postcard, color)

April 27

Web Survey reminder robodial
(only for sample members with publically
available land line telephone numbers).
Web Survey contact 4 final survey
reminder letter
(printed on MSU bond paper offering $20,
$10 or no post-paid incentive, Business
Reply Mail postcard saying “I do not have
the Internet” was included).

April 27
May 16

175

APPENDIX D

Details of Michigan Activities Survey, Robocalls, and Great Lakes Beaches web survey
methods

176

Note on names of surveys: herein when referring to the screener survey or the mail
survey, I am referring to the Michigan Activities Survey, the first of the two-stage survey
project. When I refer to robocalls, I am referring to a series of automated calls made to
survey respondents as a reminder to respond to either the mail or the web survey.
When I refer to the web survey or the internet survey, I am referring to the Great Lakes
Beaches survey.
Adherence to IRB Guidelines:
All survey responses, sample lists, pretests, information for survey respondents/pretest
participants were conducted and handled according to IRB guidelines, as described in
our IRB application. All project team members handling sample members data or survey
materials signed an oath stating that the signee would not disclose any private
information encountered as a result of working on surveys at Michigan State University.
The oaths were signed by each team member as well as a witness to the signing.
Screener Survey Details
Timing of Screener Survey and Implementation
Mailing schedule for the survey was determined by factors including the time it
took to design survey material, testing of survey, materials production, avoidance of
major holidays (mainly the 4th of July), the timing of subsequent follow-up web survey,
and elements of overall project timelines (e.g. length of assistantships for project team
members, etc).
Materials were mailed using a modified Dillman tailored design to allow for
processing of survey forms, review of data, and possible edits to survey materials.
These steps were taken to maximize response and increase representativeness of
results.
Screener Survey Sample
Our sample population included all individuals on the Michigan Drivers’ license
list (MDLL). We obtained the MDLL from the Michigan Department of State’s list sales
department. The list contained 8,917,678 records as of June 2011 when we obtained
the list. Data was received in the form of a text file and was imported into Access, with
fields set according to a character-per-column key obtained from the Department of
State.
Determining Sample Size
The sample size was determined based on factors such as forecasted response
to the screener and follow-up web survey, expected participation rate in Great Lakes
beach recreation, survey production and postage costs including outgoing and business
reply mail return postage (for both screener and web survey), data needs for robust
analysis for travel cost and stated preference applications from follow up survey, and
project budget. Factors were informed by production quotes from FedEx, Michigan
Wholesale Printing and IPC Print Services (Karen Scovie), previous recreation
177

participation studies (National Survey of Fishing, Hunting, and Wildlife Associated
Recreation (http://www.census.gov/prod/www/abs/fishing.html) and the National Survey
on Recreation and the Environment
(http://www.srs.fs.usda.gov/trends/Nsre/nsre2.html), studies of survey response rates
and past experiences of project team members. 32,000 was deemed an appropriate
sized sample.
Geographic Extent of Sample
The screener survey sample was to be drawn from counties within the more
populous Lower Peninsula of Michigan due to expected response rate and budget
constraint.
Drawing the Screener Survey Sample
A random sample of 200,000 records was drawn from the over 8 million records
in the Michigan Drivers’ License List. We used information on each county’s portion of
the Lower Peninsula’s population in the 2010 Census and applied a desired weighting
of 60% coastal counties and 40% non-coastal to determine the number of records
should be drawn from each county. Random draws within each county were made by
generating a random sequence with a unique value for each member of the 200,000
subset of the MDLL from a specific county (i.e. for Alcona county, with 208 members in
the 200,000 subset, each randomly assigned a number from 1 to 208). Records within a
county were then sorted and the predetermined number of records was selected for the
screener survey sample. Random draws were generated using R.
Screener Survey materials: Format
The form of survey materials was designed given the expected volume of
response (>10,000), postage and production costs, as well as project budget. The
survey instrument, a 1 page legal sized (8.5” x 14”) document folded in half to create a
small survey booklet was designed as such in part to minimize postage and production
costs. It also maximized the efficiency of data processing. The size of the survey
instrument was chosen based on having a survey booklet that fit in a standard #10
outgoing envelope and #9 business reply mail returning envelope, keeping the survey
weight to 1 ounce or less. The survey instrument was designed to be compatible with
and Optical Mark Recognition program (Remark) to expedite data processing and to
increase the accuracy of data processing. As part of designing the scan-able form, the
project purchased high speed scanner to convert hard copy documents into images (jpg
files). Survey images were read by the Remark software, which converted the survey
images into data exported into access database files.
Before finalizing the design of the survey instrument, various scanner and OMR
software settings were tested (i.e. darkness of scan, sensitivity of scoring), as well as
different formats of survey instruments to ensure the instrument was properly designed
and that the scanner and software were properly calibrated. Different colors, font sizes
and layouts were tested to ensure the software correctly scored test surveys (i.e.
avoidance of false “positives/scores” and “correct/accurate” scoring of test survey
sheets). Scanning sensitivities and settings were tested for accuracy against a variety
of marks (crosses, scribbles, lines across circles, words/letters, light marks and overly
178

large dark marks). The OMR software vendor was consulted to judge the score-ability
of the final survey instrument and to ensure formatting or lines on the instrument itself
would not cause false positives (blank circles from being scored as a marked answer).
Each survey sample member was assigned a unique five digit ID number
followed by a hyphen and the wave number by the FedEx production facility as part of
the materials production process (e.g. 12345-1 for wave 1, 23456-2 for wave 2 and
34567-3 for wave 3). For the production facilities matching process, this identifier had to
be included. The ID numbers were assigned by the production facility. We were able to
specify the location and size of the identifier, and that it would be five digits followed by
a hyphen, and then a 1, 2 or 3 depending on the wave of contact. ID numbers were
placed inconspicuously in the upper right hand corner of the survey instrument. This ID
was used by the production facility to match the outgoing envelope, introductory letter
and survey instrument unique to each respondent. For each wave, these IDs were
emailed to the project team by the FedEx production facility. This list of sample
respondent and five digit ID was maintained throughout the processing of responses
and data analysis as a means of tracking which respondents had responded. As
described below, research team members used the IDs from returned surveys to track
which sample members had responded to the survey.
Screener Survey Content
The screener survey was developed with two main purposes: first to identify
Great Lakes beach goers who would take part in the follow-up web survey on Great
Lakes beach recreation, and second to gather information on the general activities of
Michigan residents. For the screener survey, we attempted to “generalize” the contents
to appeal to as many respondents as possible by including questions about general
leisure activities (e.g. going to a restaurant, going to a movie in a movie theater, going
for a walk) and daily activities (e.g. watching TV, using the Internet). By having the
survey content appeal to a broad audience, we hoped to minimize possible response
bias among those who frequently participate in many activities. Any bias towards more
active respondents would skew the study’s estimated participation rates in activities and
decrease the representativeness of our results.
Questions about general activities, weekly activities, “barriers to participation”
and background/demographic questions were included to later summarize possible
relationship between respondent characteristics and participation, or participation
across various activities. Barrier questions were chosen so as to compare our results to
previous studies in other states.
The project team obtained permission from the director of the Department of
Natural Resources, Rodney Stokes and the Vice President of Travel Michigan (i.e. Pure
Michigan), George Zimmerman to place each organization’s logo and a brief appeal to
respondents attributed to each official. These appeals were drafted by our project staff
and approved by G. Zimmerman and R. Stokes. These appeals, logos, and names were
included to legitimize the purpose of the survey.

179

Screener survey instrument pretest:
A total of 26 pretests of the screener survey were conducted from June 1
through June 18th, 2011. Pretesting of the screener survey occurred in two phases. The
first was a convenience sample of 9 Michigan State University students conducted at
the International Center. Respondents were paid $5 for taking the survey and were
asked general questions about how they answered questions in the survey once they
completed it. The second phase of pretesting was a random sample of 17 adults at a
local shopping mall. Participants were paid $10 or $20 for their participation in the
pretesting.
Following each phase of screener survey pretests, edits were made to the
instrument based on respondent feedback. These edits were discussed among the
project team members before being implemented and then were tested among the next
pretest respondents.
Once pretest respondents were able to consistently take the screener survey
without incidents of confusion or uncertainty, the instrument was finalized and put into
production.
An additional subset of pretests was conducted at beaches between Port Huron
and Detroit on June 13. This special pretest was conducted to test whether visitors to
beaches along Lake St Clair and the Detroit River considered those beaches to be
Great Lakes beaches. Our research intended to include beaches along those bodies of
water as Great Lakes beaches. A total of 14 random visitors were asked whether or not
they had been to a Great Lakes beach in the last year, and whether or not they
considered the beach they were currently visiting a Great Lakes beach. 8 of the 14
people considered the beach a Great Lakes beach.
Screener Survey Materials throughout the field period:
Along with the survey instrument, screener survey sample members were also
mailed an introductory letter inviting them to participate in the survey and a business
reply mail envelope.
Waves 1 and 2 were mailed in standard number 10 outgoing envelopes with
either a green block “S” or a green Beaumont Tower icon (depending on the survey
wave) and marked with the return address “Michigan Activities Survey” in the upper-left
hand corner.
In an effort to increase response, the introductory letter for the third wave of the
screener survey was edited to include a set of answers to frequently asked questions as
a means of increasing the response rate and addressing common concerns among
callers to the survey help line. FAQs centered upon the purpose of the study,
assurances of the confidentiality of survey responses, and why and how respondents
were selected to take part in the survey.
The third wave of the survey was also mailed in a flat, 9 inch by 13 inch envelope
plainly marked with the study’s return address as a means to change the feel and
appearance of the survey package (which the respondent had already received twice
prior). The return address on the third wave only listed Dr. Lupi’s name, and did not say
“Michigan Activities Survey,” as the prior two waves had. The return address had
previously listed “Michigan Activities Survey” as a means of legitimizing the mailer as

180

being for some specific purpose, as well as to signal that this was not a solicitation for
fund raising or other purpose.
For the third wave, the survey instrument was slightly modified: the appeals from
MDNR and Pure Michigan were removed and replaced with a bright yellow box stating:
“We need your help!” This edit was made as a last attempt to convert sample members
based on the urgency and importance of the survey.
Additionally, the page containing demographic and background information was
edited to include a header ensuring respondents that their answers would remain strictly
confidential.
The screener survey instrument was also reviewed by 4 MSU grad students who
have experience working on survey research projects. Their comments were taken into
account, though they were not offered an incentive for their participation.
Sample address cleansing:
Before each wave, the project team submitted survey material files and an
address list to FedEx. FedEx’s direct mail production facility would cleanse the address
list based on the current USPS Saturation File which accounted for known changes in
address.

Undeliverable mail
In addition to completed surveys, undeliverable mail was returned to the project
office or picked up at Mail Processing throughout the survey field period. As
undeliverable mail was received, the five-digit ID for the respondent was added to a list
of undeliverable addresses. The list was double entered (as a means of QA/QC) and
IDs on the list were removed from future mailings. The screener survey sample
contained a total of 2,464 sample members that had their surveys returned as
undeliverable out of a total of 32,230 (12.6%).
Receiving and Tracking Responses:
Returned surveys were delivered via campus Mail Processing to Agriculture Hall
or picked up directly from Mail Processing. Returned surveys were then opened. Any
materials aside from surveys that were sent back (notes, comments, etc). were kept
locked in the project office for later review. Returned surveys were then scanned by an
undergrad assistant. To be sure that no surveys were skipped in the scanning process,
the assistant counted the number of surveys in each batch put through the scanner and
would compare that number to the number of image files the scanner produced. If there
were any discrepancies, the image files for that batch would be deleted and scanned
again, and the batch of surveys was recounted. This ensured that each survey was
scanned.
Hard copies of surveys were then sorted numerically (by wave, since each wave
had a different numbering scheme) and filed in locked cabinets in the project office. All
survey image files were saved on department lap top, backed up an external hard drive,
and on the MSU cloud back up system. The department laptop and external hard drive
were stored in a locked cabinet in the project office.
181

This process of receiving and scanning surveys was conducted daily until
response decreased. When fewer surveys arrived the process was repeated as
necessary.
Scanned surveys were then scored using the Remark OMR software. Project
team members tracked responses by scoring surveys using Remark, and then
comparing the software generated value for the ID versus the scanned image of the
survey to ensure the ID was scored correctly.
As surveys marked with IDs were received, the corresponding individual was
marked as having responded to the survey. Individuals who responded were removed
from future mailings.
QA-QC of Survey Data Scoring Using Remark
Once survey images were scored by Remark, the software flagged certain
responses or areas within surveys as “exceptions,” or places where the software could
not definitively code the response. The undergraduate assistant reviewed all exceptions
as identified by Remark and hand code responses.
Before the scanning and scoring of returned surveys began, to be sure that both
the scanner and the scoring software was configured to correctly score surveys, 100
random returned surveys were hand coded in addition to be scanned and read by the
OMR software. The results of the two data entry methods were compared (after
“exceptions” produced by the Remark software were addressed in the scanned and
scored data) and no differences were found in the data sets.
Robocall Methods
A list of publically available phone numbers and addresses were obtained from
Select Phone Data (selectphonedata.com) for a total cost of $200. The list covered the
entire state of Michigan (all zip codes). Members of screener survey sample were
matched with a phone number using 5 different levels of matches. Table D.1 below
shows the level number, description of the match, and the number of each type of
match. Note, the description of match types may include matches for cell phone
numbers were inadvertently included, but removed for further calls and analysis.

182

Table D.1: Description and number of different record matches used to assign phone
numbers to survey sample members for robodial implementation
Match level
Description
Number of records matched in
Screener Survey Sample from
Select Phone Data phone list
1
first name; last name; middle
7257
initial; 9 digit zip code
2
first name; last name; 9 digit zip 1261
code
3
last name; 9 digit zip code
5856
4
first name; last name; middle
508
initial; zip code
5
last name; zip code
606
total
15488
These were the matches used for robocalls that occurred along with the second wave of
the screener survey.
Robocalls throughout the second wave of the screener survey included a pre-notice and
reminder call. Each call type was made at 2:00 PM on the afternoon of the date
indicated herein. Any phone numbers that were not answered or did not reach an
answering machine or voicemail were called again that same day at 5:30 PM.
Table D.2 shows timing, amount of numbers in calls, and cost for calls during wave 2 of
the Michigan Activities Survey. Note, the information in table D.2 may include matches
for cell phone numbers were inadvertently included, but removed for further calls and
analysis.
Table D.2: Date, Time, cost, and number of calls for robodials during wave 2 of the
Michigan Activities Survey.
phone
Date
time
Call type
Cost
numbers
August 5th
2:00 PM wave 2 pre-notice (initial)
$136.4
10492
5:30 PM wave 2 pre-notice (recall)
$37.04
2849
August 10th
2:00 PM wave 2 reminder (initial)
$48.00
3000
5:30 PM wave 2 reminder (recall)
$15.01
938
We also conducted robocalls during the third wave of the screener survey. Before the
robocalls during the third wave occurred, we acquired additional phone numbers
through a phone number appending service, Accurate Append, who appended our list
of sample member names and addresses with phone numbers. From Accurate Append,
we received an additional 4281 phone number matches at a cost of $306.
Robocalls throughout the second wave of the screener survey included a prenotice call, a call estimated to be concurrent with the respondent’s receipt of the wave 3
183

mailer, and a reminder call. Each call type was made at 5:30 PM that day. Any phone
numbers that were not answered or did not reach an answering machine or voicemail
were called again that same day at 7:30 PM. Table D.3 shows call times and totals
which may include matches for cell phone numbers were inadvertently included, but
removed for further calls and analysis.
Table D.3: Date, Time, cost, and number of calls for robodials during wave 3 of the
Michigan Activities Survey.
Date
September 23rd
September 23rd
September 25th
September 25th
September 28th
September 28th

time
Call type
5:30 PM Wave 3 pre-notice (initial)
7:30 PM Wave 3 pre-notice (recall)
Wave 3 concurrent call
5:30 PM (initial)
Wave 3 concurrent call
7:30 PM (recall)
5:30 PM Wave 3 reminder (initial)
7:30 PM Wave 3 reminder (recall)

phone
cost
numbers
$76.38
4774
$26.11
1632
$53.94

3371

$17.63
$39.50
$10.50

1102
2469
656

Web Survey Details and methods
Web Survey Development
The Great Lakes beaches web survey was developed to suit the purposes of two
broad studies: first, a revealed preference travel cost modeling study including
estimates of demand for trips to different Great Lakes beaches; and second a stated
preference study of Michigan resident’s preferences for Great Lake beach attributes.
With those two studies in mind, the content of the survey was crafted based on review
of past literature and study goals. Questions were posed to respondents in such a way
that allowed for robust recreational demand modeling (in the case of questions related
to the revealed preference portion of the survey) and site characteristic preferences to
be estimated (in the case of questions related to the stated preference portion of the
survey). Consideration was also made to keep from overburdening respondents with
lengthy or arduous lines of questioning that could fatigue respondents and induce early
terminations/break-offs.
The text, diagrams, and graphics used to describe the attributes to respondents
were developed with the help of state and county health officials, water policy experts,
as well as resource managers from State and Federal agencies. Parts of the choice
experiment needed to convey scientific information in a way that was understandable to
members of the general population who are not assumed to have advanced knowledge
of the topics in the survey. The resource experts we consulted were able to ensure that
the survey clearly communicated accurate information.
Certain attribute descriptions shown to choice experiment participants also
contained diagrams and drawings. Even though the diagrams are simplifications of real184

world conditions, the experts we consulted in the development of survey materials and
pretest participants (i.e. perspective survey respondents) found the diagrams to
accurately represent conditions experienced at actual Great Lakes beaches.
Experts consulted included: Jan Stevenson (MSU), Erin Dreelin (MSU), Shannon
Briggs (MDEQ), Charles Kovatch (USEPA) and various county health department
officials.
The colors included in figures and diagrams were tested across a number of
available screen displays to simulate the assorted display types and settings sample
members would ultimately view the survey under. These tests ensured contrast and
representativeness were preserved as intended among different settings.
Various aspects of the web survey instrument were programmed to be
customized to the respondent. The web survey was programmed to include skip
patterns following a logic decided by research team members so that survey
respondents would follow a path through the survey based on their answers to
questions. This logic and these skip patterns were all determined based on the study’s
specific data needs. For example, only respondents who reported that they had taken
an overnight trip to a Great Lakes beach would be asked to provide details about an
overnight trip.
The web survey also featured programming that filled text into questions based
on information provided by respondents in earlier questions.
In order to ensure that the tasks included in the web survey met the goals of the
research study and that all programming features (skip patterns, text fills, etc.)
functioned properly, extensive programming debugging and survey pretesting was
performed.
The websurvey’s web address was submitted to several search engines (such as
google) to ensure that if the address was entered into a search engine instead of into an
address bar, that the respondent could reach the webpage. When the websurvey’s web
address was entered into a search engine, the first result was the web survey’s page.
Web Survey Debugging
To ensure that the survey was properly following the determined paths based on
possible answers to key questions, research team members systematically worked
through the survey following all possible paths, using all combinations of answers.
Research team members also tested each question, data field, and answer area
to ensure that information entered by respondents on the web survey was properly
stored for later analysis or recall in later questions.
The text of questions is filled in different ways depending on users’ answers to
previous questions. Research team members tested entering all combinations of
character types and answers into fields to ensure fills and data storage were functioning
properly.
All errors were addressed, corrected, and then tested again.
Care was also taken to ensure that the general appearance of all aspects of the
survey, including formatting, font size, tables, page borders, and diagrams appeared
clear, readable, and understandable under different display settings, browsers, and
operating systems. This was done in order to account for differences between

185

computers used by research team members to design the survey and respondents who
would ultimately take the survey.

Web Survey Pretesting:
Cognitive interviews were conducted to pretest the web survey for several
reasons:
The complexity and length of the survey task depend largely on a user’s specific
responses. In general, respondents who visit Great Lakes beaches more frequently
faced a more complex path through the web survey than those who visit Great Lakes
beaches less frequently. We needed to test whether or not users who took varying
numbers and types of trips could correctly navigate the tasks without feeling excess
recall burden or fatigue. We needed to make sure that the web survey followed a
coherent path regardless of how often the respondent visited Great Lakes beaches.
The flow, ordering, and transition between different tasks can vary depending on
the complexity of the path taken by the user. We needed to be able to test multiple
people’s understanding of the tasks as well as the transitions between them.
Given the broad range of Great Lakes beach going behaviors and statewide
scope of the project (i.e. statewide survey sample,) a pretest mode that allowed for us to
pretest the survey with a wide range of participants was required.
Web Survey Pretest Method:
For the Great Lakes beaches web survey, we chose to conduct pretests using a
remote, internet-based screen sharing application accompanied by a phone interview.
Using this method, pretest participants were called on the phone by a lead researcher
who conducted the interview. The lead researcher would then direct the participant to
join a web-based screen sharing session via a direct link emailed to the participant. The
participant would view the web survey as it appeared on the lead researcher’s
computer. The lead researcher would calibrate the display settings specific to the
pretest participant, and then grant control of the cursor (i.e. web survey) to the pretest
participant.
Throughout the pretest, the lead researcher would observe the participant and
note how well the participant seemed to navigate the survey. At designated points in the
survey, the researcher would stop the participant and ask questions to test how survey
features were functioning and how well the respondent understood the information and
tasks presented in the survey. The goal of the interview was to gather information
relevant to which portions of the survey were functioning as intended or which portions
needed revisions. When the researcher posed questions about the survey to the
participant, the researcher was able to freely navigate through the web survey to
specific pages in the survey and refer to the specific item in question as it appeared on
the screen.
A pre-arranged outline of questions and follow-up probes was developed to seek
out how respondents interpreted survey directions, and what they interpreted certain
questions to be asking.

186

The interviewer also walked through tasks with the respondent, asking questions
along the way such as: “how did you answer this question? What were you thinking
about when you first answered this question? How confident are you in your answer to
this question?” to get a feel for the respondent’s overall grasp of the task, as well as for
the cognitive burden of specific questions.
Advantages of the method:
Screen sharing technology is becoming more common place, and is frequently
used in professional settings by companies so that many parties (often remote) can
view the same computer screen, electronic document, or presentation at the same time.
Based on our contact with over thirty recruits to the survey pretest, screen sharing is
understood by many people and is not seen as risky or out of the ordinary.
The Great Lakes web survey itself draws from a statewide sample of Michigan
residents. We preferred this pretest mode because it made the pretest representative of
the survey population by allowing pretest participants to be recruited from a broader
geographic range than other modes (such as on campus convenience sampling, or
mailing random residents from within a certain radius of Michigan State University to
come to campus for onsite pretesting).
Additionally, though many East Lansing area residents may make frequently take
trips to the Great Lakes, their opinions of the survey may be different from residents of
areas/counties that are adjacent to the Great Lakes. To seek the opinion of a key group
of Web survey respondents, coastal county residents, we required a pretesting mode
that allowed us to reach beyond the East Lansing area. Coastal county residents likely
have different Great Lake visitation patterns (due primarily to proximity), and could have
a different approach to a survey about Great Lakes beach recreation. For example, a
research question we kept in mind was whether or not coastal residents would consider
Great Lakes recreation a routine activity and end up under reporting or mis-reporting
details of recreation? Would non-coastal county residents have more difficulty recalling
details of Great Lakes beach trips that occur less routinely? By drawing from a statewide pretest participant pool, we were able to investigate these questions.
This method of pretesting allowed for many members of the research team to
observe pretests. While one researcher lead the meeting and interviewed the
participant, other researchers could, without disturbing the participant or the ongoing
pretest, join the screen sharing session and the call between the researcher and
participant (which was hosted on a conference call line). Additional researchers called in
to the conference call before the interview began and muted their phone lines.
Researchers also had the option of recording calls that ran along with the pretests for
review at a later time. This method of pretesting allowed for several research team
members to observe a single participant without needing to physically be in the same
room as the single respondent.
Mechanics of the web survey pretesting method
We used the screen sharing application GatherPlace (Gatherplace.net). With a
monthly subscription, we were allowed to set up unlimited screen share sessions.
Monthly subscriptions cost $65. Each session allowed up to 40 guests to observe the
host’s screen at one time. The screen sharing application is run on JAVA, a nearly
187

ubiquitous program commonly used to run many other applications on a variety of
websites. Outside of JAVA, GatherPlace does not require guests of screen sharing
sessions to download any third party programs. Screen sharing sessions first require a
host to begin a session. For pretesting, this meant that only once the lead researcher
had set up and begun the session could a pretest participant see the web survey. This
provided a large degree of control for the researcher and kept participants from viewing
unintended content. Following the start of a session, guests can click direct links to the
specific session (that can be emailed directly by the host) or can visit GatherPlace.net
and navigate to the session by entering a screen share session ID.
We recruited participants through an electronic sample purchased from Survey
Sampling International (SSI). We provided SSI with basic information about our target
population: adults in the Lower Peninsula of Michigan. Next, SSI emailed a generic
invitation to their sample members matching those criteria. The invitation simply stated:
“Dear <Name>, Your Opinion Matters! Click here to participate in a new SurveySpot
survey opportunity or click the button below.”
Users were then brought to a webpage explaining the criteria to be eligible to
participate in the pretests: participants must have visited a Great Lakes beach in the last
year, must speak on the phone while taking the survey on the internet. The interviews
were said to last approximately 45 minutes. We offered recruits a $40 incentive for
completing the interview. The content of the web site appears below:
We attempted to recruit pretest participants of different participation levels who,
over the course of answering survey questions to the best of their ability, would follow
different paths through the survey, testing different functions of the web survey
instrument in the process. In particular, within each batch of pretest completes, we tried
to roughly recruit an equal number of respondents who had taken the following types of
trips: trips lasting a day or less, overnight trips of less than 4 nights, and overnight trips
of 4 nights or more. These trip types each have a separate line of questioning.
Additionally, pretests across participants visiting Great Lakes beaches with different
frequency let researchers check respondents’ comprehension of each of those lines of
questions as well as check respondents’ awareness and understanding of transitions
between those lines of questions—therefore, obtaining a pretest sample of respondents
who had taken a variety of trips was important to the success of the pretest.
SSI sent out an unknown number of emails based on our need for 15 completes.
The research team received approximately 25 emails and 20 phone calls in the week
following the initial invitation. In many emails and phone messages, pretest recruits
listed the Great Lakes beach(es) and when they visited them in the last year as means
of proving they met the criteria listed in the invitation. In cases where a pretest recruit
did not list where they visited (and therefore were not certain to be eligible for the
pretest) a researcher followed up with an email or a call inquiring whether they had
visited a Great Lakes beach in the last year (as was required to participate) as well as
the length of the trip (to help ensure we spoke with respondents with a range of trip
types and frequencies.
In general, it was easier to recruit pretest participants using email than using the
phone. Email offered an easily traceable line of communication and was more
frequently checked than voicemail by pretest recruits. Email contact was established
with every participant regardless of initial form of contact (phone or email).
188

Once a pretest recruit was verified as eligible, an appointment was scheduled.
Most appointments were made fewer than three days in advance to keep from having
pretest recruits forget about appointments.
We scheduled a total of 34 appointments to reach our 30 completes. Four pretest
recruits made appointments and did not keep them. None of the four cancelers
responded to requests to reschedule.
Appointments with pretest recruits were confirmed via an email similar to this:
Hello <NAME>, This is an email to confirm that we have an appointment scheduled for
<TIME> on <WEEKDAY, MONTH, DATE>
At <TIME> I will call you at the number you provided. I will be calling from MSU, but the
phone number will likely show up as: <800 number specific to conference call service
used>
Near our appointment, I will email you a link to follow to take the survey.
In order to take the survey, you will need to have the program Java installed on your
computer. If you're unsure about whether or not you have Java installed on your
computer, you should download it before our appointment. Please visit
http://www.java.com/en/ and click on the "Free Java Download" button. Java is a
program commonly used to run certain parts of websites, and will not be harmful to your
computer. Again, in order to take the survey, you need to have Java on your computer.
Thanks again
The email included a notice about double checking to see if a working version of
JAVA was installed on their computer. The JAVA reminder was used to avoid possible
problems from the JAVA application to allow for screen sharing. Several respondents
would reply to the appointment confirmation email with a message stating, “I already
have JAVA” or “I just installed JAVA.” Only five pretest recruits (out of 30 completes)
needed to install JAVA either just before or at the beginning of a pretest appointment.
None of these respondents had any issues or showed hesitancy with installing JAVA.
Approximately six of the pretest participants were unable to access the meeting
using Internet Explorer as their browser, but did not have any difficulties connecting
through Mozilla Firefox.
It is worth noting that SSI panel members take a variety of surveys, some of
which may require JAVA, and as such, SSI panel members may be more likely to have
JAVA installed and working on their computers than random members of the population.
It’s also important to note that no portion of the Web survey itself required JAVA to run:
only the pretest screen sharing required JAVA.
In a typical day, between two and five interviews were scheduled. This frequency
allowed researchers the flexibility to address issues that might arise from interviews by

189

modifying or adding probes during the next interview, editing the survey instrument, and
to debrief about other occurrences.
Pretests typically lasted between 45 and 60 minutes. Actual interview times
ranged from 30 to 75 minutes depending on whether the respondent encountered any
technical issues, the speed and specific path taken through the survey, and the types of
responses to probes and questions within the cognitive interview).
We secured a conference call line for approximately $0.097 per minute (this
estimate includes taxes and fees). Throughout the course of pretesting, we used
approximately 6100 conference line minutes (this includes minute totals for all members
of the research team and for the participant who took part in interviews). A few minutes
before the appointment, the research team joins a conference call. There is only one
researcher who conducts the interview. All other researchers mute their phone lines so
as not to interrupt the interview. At this time the call recording is begun.
Web Survey pretesting occurred in two phases of 15 pretests each, for a total of
30 completes. Minor edits were made between pretests (e.g. corrections to spelling
errors, spacing issues) while more major edits were made between the two phases of
pretests. Substantive changes in the web survey instrument were tested in subsequent
pretests to be sure edits functioned as intended. As themes arose in comments from
several respondents, or the directions of certain tasks were ignored or misinterpreted,
the research team decided how, if at all, the topic should be addressed. For example, in
one case, a respondent did not follow an instruction to only list the beach they spent the
most amount of time at on a daytrip, instead listing all the beaches they visited. The
research team decided it was an important direction that needed to be followed, and
formatted the text of the instruction to appear in italics for further emphasis. All
remaining pretest participants followed the direction, listing only the beach where they
spend the most amount of time on the day trip in question.
Pilot test:
Following the completion of pretests, we conducted a pilot survey among 85
individuals who responded to the mail screener survey that they visited a Great Lakes
beach, but responded after the field period ended for the screener mail survey.
Members of the pilot survey sample were contacted using 3 mail contacts: an
introduction letter on March 26th, a half-sheet postcard in black and white on April 2nd,
and a quarter sheet postcard in color on April 9th. 22 of the 85 pilot survey sample
members logged onto the web survey for a response rate of 25.9%.
The pilot pretest allowed the research team to simulate the process of actually
implementing the survey (e.g. timing of mailing, receiving data), to have one final round
of programming debugging, and to see if any questions arrived via phone or email from
pilot survey respondents. Data from pilot survey responses were used to develop and
test a script to code responses from the Stated Preference section for use in analysis.
Implementing the Web Survey
The Great Lakes web survey field period began as soon as sample members
received their invitation letters with unique password. Invitation letters were mailed on
April 12th. Data from survey responses was populated in a data file that was integrated
190

into the web survey’s programming but accessible only by research team members.
Research team members periodically downloaded and inspected the data file
throughout the survey field period to search for possible errors as well as monitor
response rates.
Table D.4 below summarizes all contacts made to web survey sample members
throughout the survey field period. The field period lasted (i.e. responses were
accepted) from April 12, 2012 until June 1, 2012. Invitation letters were mailed and
included a $1 token. Wave 2 reminder postcards were half sheet (8.5” by 5.5”) in size
and were printed in black in white, while Wave 3 reminder postcards were quarter-sheet
(4.25” by 5.5”) and printed in color. Concurrent with the mailing of the wave 3 reminder
postcard, a robocall was made to sample member’s we were able to match to a publicly
available phone number.
Table D.4: Date of contacts for Great Lakes beaches web survey
Web Survey contact 1 invitation letter
April 12
(plain paper, $1 token included)
Web Survey contact 2 reminder postcard
(half sheet postcard, black and white)

April 20

Web Survey contact 3 reminder postcard
(quarter sheet postcard, color)

April 27

Web Survey reminder robodial
(only for sample members with publically
available land line telephone numbers).
Web Survey contact 4 final survey
reminder letter
(printed on MSU bond paper offering $20,
$10 or no post-paid incentive, Business
Reply Mail postcard saying “I do not have
the Internet” was included).

April 27
May 16

In the final wave of contacts in the web survey, we mailed a reminder letter
printed on MSU bond paper. The envelope included a business reply mail postcard that
when returned would indicate that the respondent did not have the Internet and
therefore could not take the survey. Of the 2863 recipients of the final reminder, 292
individuals returned the “no internet” postcard. Included in the final reminder letter was
also an offer for a $20, $10, or no post-paid incentive. Details of the distribution and
response associated with each offer are shown in table D.5.

191

Table D.5 Final mailing, distribution and response sorted by post-paid incentive
Post-paid incentive
Sample members
Responses
Response rate
$0
242
28
11.6%
$10
1266
254
20.1%
$20
1274
295
23.2%
Below are the results of Chi-squared tests which test for differences in response rates to
the web survey based on different post-paid incentives offered in the final reminder.
Table D.6: 3-way Chi-squared test comparing response rates to web survey based on
final communication featuring $20, $10 or no post-paid incentive
Post-paid
Group size (n)
Actual response Actual response Expected
Incentive
rate
response (n)
given overall
response rate
$0 (Control)
242
28
11.6%
50.2
$10
1266
254
20.1%
262.6
$20
1274
295
23.2%
264.2
Pooled
2782
577
20.7%
p-value=0.0011
We reject the null hypothesis that response rates to the web survey were identical
among individuals offered $20, $10 or no post-paid incentive (i.e. there is a statistically
significant difference between response rates among different post-paid incentives).
Table D.7: 2-way Chi-squared test comparing response rates to web survey based on
final communication offering $20 or $10 post-paid incentive
Post-paid
Incentive

Group size (n)

Actual response

Actual response
rate

$10
$20
Pooled
p-value=0.094

1266
1274
2540

254
295
549

20.1%
23.2%
21.6%

Expected
response (n)
given overall
response rate
262.6
264.2

We fail to reject the null hypothesis that response rates to the web-survey were identical
among those offered a $20 or $10 post-paid incentive.

192

Table D.8: 2-way Chi-squared test comparing response rates to web survey based on
final communication offering $10 or no post-paid incentive
Post-paid
Group size (n)
Actual response Actual response Expected
Incentive
rate
response (n)
given overall
response rate
$0 (control)
242
28
11.6%
45.3
$10
1266
254
20.1%
236.8
Pooled
1508
282
18.7%
p-value=0.0051
We reject the null hypothesis that response rates to the web-survey were identical
among those offered a $10 post-paid incentive and those not offered a post-paid
incentive.
Table D.9: 2-way Chi-squared test comparing response rates to web survey based on
final communication offering $20 or no post-paid incentive
Group name
Group size (n)
Actual response Actual response Expected
rate
response (n)
given overall
response rate
$0
242
28
11.6%
51.6
$20
1274
295
23.2%
271.4
Pooled
1516
323
21.3%
pvalue=0.00034
We reject the null hypothesis that response rates to the web-survey were identical
among those offered a $20 post-paid incentive and those not offered a post-paid
incentive.

193

Response to Web Survey
Table D.10, below, describing the overall disposition of responses to the web
survey
Table D.10: Response and disposition for the Great Lakes beaches web survey
Sample Members
5434
log-ons to web survey
3211
Logged on and terminated immediately
2
Moved, unable to forward mail
2
Deceased
2
Mail returned as undeliverable
40
Indicated by phone or mail that they did not have
Internet
361
Indicated by phone or mail that they did not have
Internet but go to Great Lakes beaches
4
Indicated by phone or mail that they did not have
Internet but do NOT go to Great Lakes beaches
8
Returned the business reply mail postcard stating
they did not have access to the internet
292
Refusal
6
The following six tables show a question-by-question summary of response and item
non-response for six sections of the survey: The front matter section which includes the
first three questions of the survey, the choice experiment section, the Great Lake
opinion section, the exposure question, the background information section, and the
income section. This analysis of item non-response does not include questions included
in the travel cost portion of the survey. The income question is handled as its own
section since it was asked in three sub-parts.

194

Table D.11: Great Lakes Web Survey Item Non-Response table 1 of 6, section: Front Matter
number
number
number
item nonpage location question asked
who
of nonresponse
question text
of question
number question
responded responses rate
Have you gone to any Great Lakes beaches in
Michigan in the past two years?
q0.aspx
Q0
3211
3200
11
0.3%
Did you go to any Great Lakes beach in Michigan
from October 1, 2010 to May 26, 2011?
Did you go to any Great Lakes beach in Michigan
from Memorial Day weekend, 2011 to September
30, 2011?

q1.aspx

Q1

3200

3017

183

5.7%

q1.aspx

Q2

3193

2994

199

6.2%

195

Table D.12: Great Lakes Web Survey Item Non-Response table 2 of 6, section: Choice Experiment section
number
number
item nonpage location question asked
number
of nonresponse
question text
of question
number
question responded responses rate
What is the length of the Great Lakes public beach you visit most
ce_q7.aspx
ce_q7
2901
2887
14
0.5%
frequently?
How much do you agree or disagree: “The length of a beach is
ce_q7.aspx
ce_q8
2901
2889
12
0.4%
important to my enjoyment of the beach.”
Have you heard about a Great Lakes public beach being closed
ce_q9.aspx
ce_q9
2889
2880
9
0.3%
because of high levels of unsafe bacteria?
How much do you agree or disagree: "I do not care if the Great Lakes public beach I visit is tested for unsafe bacteria because…"
ce_q9.aspx
ce_q10a
2886
2823
63
2.2%
…I don’t go in the water.
…the beach has been tested in the past and no unsafe
ce_q9.aspx
ce_q10b
2885
2754
131
4.5%
bacteria were found.
ce_q9.aspx
ce_q10c
2885
2749
136
4.7%
…the beach is not close to any streams or rivers.
…the beach is not close to any farms or shoreline
ce_q9.aspx
ce_q10d
2885
2748
137
4.7%
development.
How much algae was present in the water at the Great Lakes
ce_q11.aspx
ce_q11
2869
2854
15
0.5%
beach you most recently visited?
How much do you agree or disagree: "The amount of algae in the
ce_q11.aspx
ce_q12
2869
2857
12
0.4%
water affects my enjoyment of a beach"

196

Table D.12 (Cont’d)

question text
How much algae was present on the shore at the Great Lakes
beach you most recently visited?
How much do you agree or disagree: “The amount of algae on the
shore affects my enjoyment of a beach.”
What type of bathrooms are available at the Great Lakes beach
you most recently visited?
Do you know how often the bathrooms at the Great Lakes beach
you most recently visited get cleaned?
About how many miles is it from your permanent residence to the
Great Lakes beach you most recently visited?
How much do you agree or disagree: The distance from my home
to a beach affects how often I visit a beach.”
First choice set
Second choice set
Third choice set

page location
of question

question
number

number
asked
question

ce_q13.aspx

ce_q13

2862

2845

17

0.6%

ce_q13.aspx

ce_q14

2862

2847

15

0.5%

ce_q15.aspx

ce_q15

2857

2846

11

0.4%

ce_q15.aspx

ce_q16

2857

2791

66

2.3%

ce_q17.aspx

ce_q17

2851

2814

37

1.3%

ce_q17.aspx

ce_q18

2851

2835

16

0.6%

ce_q19a.aspx
ce_q19b.aspx
ce_q19c.aspx

ce_q19a
ce_q19b
ce_q19c

2832
2832
2830

2703
2750
2747

129
82
83

4.6%
2.9%
2.9%

197

number
item nonnumber
of nonresponse
responded responses rate

Table D.13: Great Lakes Web Survey Item Non-Response table 3 of 6, section: Great Lakes Opinion section
number number
number
item nonpage location question
asked
who
of nonresponse
question text
of question
number
question responded responses rate
Please rate the following beach characteristics of each Great Lake:
Water quality (clarity, color, and odor) Erie
ce_q20.aspx ce_q20a
2822
2692
130
4.6%
Water quality (clarity, color, and odor) Huron
ce_q20.aspx ce_q20b
2822
2663
159
5.6%
Water quality (clarity, color, and odor) Michigan
ce_q20.aspx ce_q20c
2822
2766
56
2.0%
Water quality (clarity, color, and odor) Superior
ce_q20.aspx ce_q20d
2822
2699
123
4.4%
Water quality (clarity, color, and odor) St. Clair
ce_q20.aspx ce_q20e
2822
2685
137
4.9%
Level of algae Erie
ce_q20.aspx ce_q21a
2822
2680
142
5.0%
Level of algae Huron
ce_q20.aspx ce_q21b
2822
2666
156
5.5%
Level of algae Michigan
ce_q20.aspx ce_q21c
2822
2754
68
2.4%
Level of algae Superior
ce_q20.aspx ce_q21d
2822
2668
154
5.5%
Level of algae St. Clair
ce_q20.aspx ce_q21e
2822
2666
156
5.5%
Debris in the water and on the shore Erie
ce_q20.aspx ce_q22a
2822
2665
157
5.6%
Debris in the water and on the shore Huron
ce_q20.aspx ce_q22b
2822
2666
156
5.5%
Debris in the water and on the shore Michigan
ce_q20.aspx ce_q22c
2822
2754
68
2.4%
Debris in the water and on the shore Superior
ce_q20.aspx ce_q22d
2822
2669
153
5.4%
Debris in the water and on the shore St. Clair
ce_q20.aspx ce_q22e
2822
2665
157
5.6%
Natural Beauty Erie
ce_q20.aspx ce_q23a
2822
2680
142
5.0%
Natural Beauty Huron
ce_q20.aspx ce_q23b
2822
2689
133
4.7%
Natural Beauty Michigan
ce_q20.aspx ce_q23c
2822
2761
61
2.2%
Natural Beauty Superior
ce_q20.aspx ce_q23d
2822
2684
138
4.9%
Natural Beauty St. Clair
ce_q20.aspx ce_q23e
2822
2669
153
5.4%
Quality and texture of the sand or beach areas Erie
ce_q20.aspx ce_q24a
2822
2669
153
5.4%
Quality and texture of the sand or beach areas Huron
ce_q20.aspx ce_q24b
2822
2678
144
5.1%
Quality and texture of the sand or beach areas Michigan
ce_q20.aspx ce_q24c
2822
2765
57
2.0%
Quality and texture of the sand or beach areas Superior
ce_q20.aspx ce_q24d
2822
2675
147
5.2%
Quality and texture of the sand or beach areas St. Clair
ce_q20.aspx ce_q24e
2822
2668
154
5.5%

198

Table D.14: Great Lakes Web Survey Item Non-Response table 4 of 6, section: Exposure section
number number
number
item nonpage location of question
asked
who
of nonresponse
question text
question
number
question responded responses rate
How often do you do the following things on a typical visit to a Great Lakes beach?
I walk on the sand
ce_q25.aspx
ce_q25a
2821
2792
29
1.0%
I play or dig in the sand
ce_q25.aspx
ce_q25b
2821
2684
137
4.9%
I get my feet wet
ce_q25.aspx
ce_q25c
2821
2753
68
2.4%
I go in the water, but not past my knees
ce_q25.aspx
ce_q25d
2821
2670
151
5.4%
I go in the water, but not past my shoulders
ce_q25.aspx
ce_q25e
2820
2664
156
5.5%
I go in the water and my entire body gets wet
ce_q25.aspx
ce_q25f
2819
2737
82
2.9%

Table D.15: Great Lakes Web Survey Item Non-Response table 5 of 6, section: Background information section

question text
What is your sex?
In what year were you born?
What is your race and/or ethnicity? Mark all that
apply
What is the highest degree or level of school you
have completed?
What is the zip code of the place you live?
How long have you lived in Michigan?
What is your current employment status?
Do any of the following live in your household?

page
location of
question
demo_a.aspx
demo_a.aspx

question
number
ce_q27
ce_q28

item
number number
number
nonasked
who
of nonresponse
question responded responses rate
2786
2751
35
1.3%
2785
2722
63
2.3%

demo_a.aspx ce_q29

2785

2734

51

1.8%

demo_a.aspx
demo_a.aspx
demo_a.aspx
demo_a.aspx
demo_a.aspx

2785
2785
2784
2780
2768

2752
2735
2731
2755
2742

33
50
53
25
26

1.2%
1.8%
1.9%
0.9%
0.9%

ce_q30
ce_q31
ce_q32
ce_q33
ce_q34

199

Table D.16 Great Lakes Web Survey Item Non-Response table 6 of 6, section: income question
number
number
page location question asked
who
question text
of question
number question responded
In 2011, what was your total household income,
demo_a.aspx
ce_q35
2768
2464
from all sources, before taxes?

number
item nonof nonresponse
responses rate
304
11.0%

Was your total household income in 2011 less
than $50,000? (Median)

demo_a1.aspx

ce_q35a

304

204

100

32.9%

(if yes: Was it less than $25,000?) (if no: Was it
less than $100,000?) (Quartiles)

demo_a1.aspx

ce_q35b

204

74

130

63.7%

Summary of income information
Respondents with some income information
(who reached income questions)

Respondents without income information (who
reached income questions)

2668 out of 2768 Or 96.4%

100 out of 2768 or 3.6%

200

Table D.17 Summary of distances seen in Great Lakes beach web survey choice
experiment (in miles)
Design
type

beach
left beach total
labeled
distance
right beach total
distance
left beach total
unlabeled distance
right beach total
distance
left beach total
same
distance
right beach total
distance

observations Mean

std.
deviation min

3059

171

90.7

0

436

3059

172.7

91

0

422

3190

173

93.3

0

467

3190

172.8

93.7

0

467

1948

172.1

93.9

0

441

1948

173

92.7

0

451

max

Table D.18 Summary of item response among choice experiment respondents, listed by
labeling scheme viewed

labeling scheme
labeled
same
unlabeled
overall

total
1040
665
1085
2790

choice sets answered (out of 3)
3
2
1
988
43
9
625
33
7
1031
43
11
2644
119
27

201

REFERENCES

202

REFERENCES

Adamowicz, V., and P. Boxall. 2001. “Future Directions of Stated Choice Methods for
Environmental Valuation.” Paper prepared for Choice Experiments: A New
Approach to Environmental Valuation. London, England.
Alberini, A., A. Longo, and M. Veronesi. 2006 “Basic Statistical Models For Stated
Choice Studies.” In: Kanninen, B.J (Ed.), Valuing Environmental Amenities Using
Stated Choice Studies. A Common Sense Approach to Theory and Practice.
Dordrecht, The Netherlands: Springer. pp. 203–227.
Ara, S., Irwin, E., and Haab, T. 2006. “The Influence of Water Quality on the Housing
Price Around Lake Erie.” Selected Paper prepared for presentation at the
American Agricultural Economics Association Annual Meeting, Long Beach, CA,
July 23-26.
Auer, M.T., L.M. Tomlinson, S.N. Higgins, S.Y. Malkin, E.T. Howell, and H.A. Bootsma.
2010. “Great Lakes Cladophora in the 21st century: same algae—different
ecosystem.” Journal of Great Lakes Research. 36(2): 248-255. June.
Bay County Health Department. 2007. “Beach ‘Muck’ Advisory.” http://www.baycountymi.gov/Health/BeachMuckAdvisory.aspx accessed 8/12/12
Ben-Akiva, M., and S. Lerman. 1985. Discrete Choice Analysis: Theory and Application
to Predict Travel Demand. Cambridge, MA: MIT Press.
Blamey R.K., J.W. Bennett, J.J. Louviere, M.D. Morrison, J. Rolfe. 2000. “A test of
policy labels in environmental choice modelling studies.” Ecological Economics,
32(2): 269–287.
Bockstael, N.E., K.E. McConnell, and I.E. Strand. 1989. “Measuring the Benefits of
Improvements in Water Quality: The Chesapeake Bay.” Marine Resource
Economics. 6(1): 1-18.
Boyle, K.J., N.V. Kuminoff, C.F. Parmeter, and J. C. Pope. 2010. The Benefit-Transfer
Challenges. Annual Review of Resource Economics. 2: 161-180.
Braden, J.B., A.A. Patunru, S. Chattopadhyay, and N. Mays. 2004. “Contaminant
Cleanup in the Waukegan Harbor Area of Concern: Homeowner Attitudes and
Economic Benefits”. Journal of Great Lakes Research. 30(4): 474-491.
Brannlund, R., and L. Persson. 2012. “To tax or not to tax: preferences for climate policy
attributes.” Climate Policy. 12(6): 704-721.

203

Brookshire, D.S. and H.R. Neill. 1992. “Benefit transfers: Conceptual and empirical
issues.” Water Resources Research. 28(3): 651-655.
Carson, R.T., J.J. Louviere, D.A. Anderson, P. Arabie, D.S. Bunch, D.A. Hensher, R.M.
Johnson, W.F. Kuhfeld, D. Steinberg, J. Swait, H. Timmermans, and J.B. Wiley.
1994. “Experimental Analysis of Choice.” Marketing Letters. 5(4): 351-368.
Caussade S., J.D. Ortuzar, L.I. Rizzi, and D.A. Hensher. 2005. “Assessing the influence
of design dimensions on stated choice experiment estimates.” Transportation
Research Part B. 39(7): 621-640.
Chattopadhyay, S. 1999. “Estimating the Demand for Air Quality: New Evidence Based
on the Chicago Housing Market.” Land Economics.75(1):22-38. February.
Choice Metrics. 2011. “NGene 1.1 User Manual and Reference Guide.” Version
13/05/2011. http://www.choice-metrics.com/download.html
Czajkowski, M., and N. Hanley. 2009. “Using Labels to Investigate Scope Effects in
Stated Preference Methods.” Environmental and Resource Economics. 44(4):
521-535.
DeShazo, J.R. and G. Fermo. 2002. “Designing Choice Sets for Stated Preference
Methods: the Effects of Complexity on Choice Consistency.” Journal of
Environmental Economics and Management. 44(1): 123–143.
Dillman, D.A. 2009. Internet, Mail, and Mixed-Mode Surveys, the Tailored Design
Method. Third Edition. New York: John Wiley and Sons.
Efron, B., and R.J. Tibshirani. 1993. An Introduction to the Bootstrap. New York:
Chapman and Hall.
Egan, K.J. and D.F. Dwyer. 2008. “The Value of Restoring Wetlands for Maumee Bay
State Park Beach Visitors.” Final Report. January.
http://www.utoledo.edu/nsm/lec/pdfs/MBSP_Beach_Survey_Final_Report.pdf
Accessed August 2012.
Environment Canada and U.S. EPA. 2009. “State of the Great Lakes Report.” EPA
905-R-09-031 Online: http://www.epa.gov/solec/sogl2009/index.html accessed
8/12/2012
Fenichel, E.P., F. Lupi, J.P. Hoehn, and M.D. Kaplowitz. 2009. “Split-Sample Tests of
“No Opinion” Responses in an Attribute-Based Choice Model.” Land Economics,
85(2): 348-362.

204

Ford, J.K., N. Schmitt, S.L. Schechtman, B.M. Hults, and M.L. Doherty. 1989. “Process
tracing methods: problems and neglected research questions.” Organizational
Behavior and Human Decision Processes 43(1): 75-117.
Green, P.E. and V. Srinivasan. 1978 “Conjoint Analysis in Consumer Research: Issues
and Outlook.” Journal of Consumer Research. 5(2): 103-123. September
Green, P.E., A.M. Krieger, and Y. Wind. 2001. “Thirty Years of Conjoint Analysis:
Reflections and Prospects.” Interfaces. 31(3): S56-S73.
Hanley, N., R.E. Wright and V. Adamowicz. 1998. “Using Choice Experiments to Value
the Environment.” Environmental and Resource Economics. 11 (3-4): 413428.428.
Harris, V.A. 2005. “Nuisance Algae on Lake Michigan Shores.” Fact Sheet. University of
Wisconsin Sea Grant Institute. Green Bay, WI.
Harrison, G.W. 2006. “Making Choice Studies Incentive compatible.” In: B.J. Kanninen
(Ed.), Valuing Environmental Amenities Using Stated Choice Studies. A Common
Sense Approach to Theory and Practice. Dordrecht, The Netherlands: Springer.
pp. 67-110.
Heiner, R.A. 1983. “The Origin of Predictable Behavior.” The American Economic
Review. 73(4): 560-595.
Hensher, D. A. 2001. “Measurement of the Valuation of Travel Time Savings.” Journal
of Transport Economics and Policy, Special Issue in Honour of Michael Beesley,
35(1): 71–98.
Hensher, D.A. 2006a. “Attribute Processing in Choice Experiments and Implications on
Willingness to Pay.” In: B.J. Kanninen (ed)., Valuing Environmental Amenities
Using Stated Choice Studies. Dordrecht, The Netherlands: Springer. Pp. 135158.
Hensher, D.A. 2006b. “Revealing Differences in Willingness to Pay due to the
Dimensionality of Stated Choice Designs: An Initial Assessment.” Environmental
and Resource Economics. 34(1): 7-44.
Higgins, S.N., E.T. Howell, R.E. Hecky, S.J. Guildford, and R.E. Smith. 2005. “The Wall
of Green: The Status of Cladophora glomerata on the Northern Shores of Lake
Erie’s Eastern Basin, 1995-2002.” Journal of Great Lakes Research 31(4): 547563.
Higgins, S.N., S.Y. Malkin, E.T. Howell, S.J. Guildford, L. Campbell, V. Hiriart-Baer, and
R. E. Hecky. 2008. “An Ecological Review of Cladophora Glomerata
205

(Chlorophyta) in the Laurentian Great Lakes.” Journal of Phycology. 44 (4), 839–
854.
Hilger, J., and W.M. Hanemann. 2006. “Heterogeneous Preferences for Water Quality:
A Finite Mixture Model of Beach Recreation in Southern California.” Research
Final Reports, California Sea Grant College Program, UC San Diego. February.
Hoehn, J.P., F. Lupi, and M.D. Kaplowitz. 2003. “Untying a Lancastrian Bundle: Valuing
Ecosystems and Ecosystem Services for in Wetland Mitigation” Journal of
Environmental Management. 68 (3): 263-272. July.
Hoehn, J.P., F. Lupi, M.D. Kaplowitz. 2010. “Stated Choice Experiments with Complex
Ecosystem Changes: The Effects of Information Formats on Estimated Variances
and Choice Parameters.” Journal of Agricultural and Resource Economics. 35(3):
568-590.
Hoehn, J.P., and A. Randall. 2002. “The Effect of Resource Quality Information on
Resource Injury Perceptions and Contingent Values.” Resource and Energy
Economics, 24(1): 13-31.
Huang, J., P.J. Poor, and M.Q. Zhao. 2007. “Economic Valuation of Beach Erosion
Control.” Marine Resource Economics. 22(3): 221-238.
Huybers, T. 2005. “Destination choice modelling: what’s in a name?” Tourism
Economics. 11(3): 329-350.
International Joint Commission. 2011a. “Fifteenth Biennial Report on Great Lakes Water
Quality.” January.
http://www.ijc.org/rel/boards/watershed/15biennial_report_web-final.pdf
International Joint Commission. 2011b. “Nuisance and Harmful Algae.” Great Lakes
Water Quality Agreement, 2009–2011 Priority Cycle Report IJC, Special
Publication 2011–3, Windsor, Ontario, Canada.
Johnson, F.R., B. Kanninen, M. Bingham, S. Ozdemir. 2006. “Experimental Design for
Stated Choice Studies.” In: B.J. Kanninen (ed)., Valuing Environmental
Amenities Using Stated Choice Studies. Dordrecht, The Netherlands: Springer.
Pp. 159-202.
Johnston, R.J. 2006. “Context Similarity and the Validity of Benefits Transfer: Is the
Common Wisdom Correct?” Selected Paper prepared for presentation at the
American Agricultural Economics Association Annual Meeting, Long Beach,
California, July 23-26.
Johnston, R. 2007. “Choice Experiments, Site Similarity and Benefits Transfer.”
Environmental and Resource Economics. 38(3): 331-351.
206

Johnston, R.J. and R.S. Rosenberger. 2010. “Methods, Trends and Controversies in
Contemporary Benefit Transfer.” Journal of Economic Surveys. 24(3): 479-510.
Kamakura, W.A., and G.J. Russell. 1993. “Measuring Brand Value With Scanner Data.”
International Journal of Research in Marketing. 10(1): 9-22.
Kaplowitz, M.D., F. Lupi, and J.P. Hoehn. 2004. “Multiple-Methods for Developing and
Evaluating a Stated Choice Survey to Value Wetlands.” In S. Pressor, J. M.
Rothgeb, M.P. Couper, J.T. Lessler, E. Martin, J. Martin, and E. Singer (eds.),
Methods for Testing and Evaluating Survey Questionnaire, Hoboken, NJ: John
Wiley and Sons, pp. 503-524.
Keller, K.L. 1993. “Conceptualizing, Measuring, and Managing Customer-Based Brand
Equity.” Journal of Marketing. 57(1): 1-22. January.
Kotchen, M.J., M.R. Moore, F. Lupi, and E.S. Rutherford. 2006. “Environmental
Constraints on Hydropower: An Ex Post Benefit-Cost Analysis of Dam
Relicensing in Michigan.” Land Economics 82(3): 384-403.
Krieger, D.J., and J.P. Hoehn. 1998. "Improving Health Risk Information for Sport
Anglers: The Value of Information about Chemical Residues." North American
Journal of Fisheries Management, 18(2):411-421.
Krinsky, I. and A.L. Robb. 1986. “On Approximating the Statistical Properties of
Elasticities.” The Review of Economics and Statistics. 68(4): 715–719.
Lancaster, K.J. 1966. “A New Approach to Consumer Theory.” Journal of Political
Economy. 74(2): 132-157.
Landry, C.E., A.G. Keeler, W. Kriesel. 2003. “An Economic Evaluation of Beach Erosion
Management Alternatives.” Marine Resources Economics. 18(2): 105-127.
Lew, D.K., and D.M. Larson. 2005. “Valuing Recreation and Amenities at San Diego
County Beaches” Coastal Management 33(1): 71-86
Loomis, J.B., R.S. Rosenberger. 2006. Reducing Barriers in Future Benefit Transfers:
Needed Improvements in Primary Study Design and Reporting. Ecological
Economics. 60: 343-350.
Louviere, J. J., D. A. Hensher and J.D. Swait. 2000. Stated Choice Methods: Analysis
and Applications. New York: Cambridge University Press.

207

Lupi, F., J.P. Hoehn, H.Z. Chen, and T.D. Tomasi. 1998. “The Michigan Recreational
Angling Demand Model.” Staff Paper 97-58, Department of Agricultural
Economics, Michigan State University. East Lansing, MI. March.
Lupi, F., M.D. Kaplowitz, and J.P. Hoehn. 2002. “The Economic Equivalency of Drained
and Restored Wetlands in Michigan.” American Journal of Agricultural
Economics, 84(5): 1355-1361.
Lupi, F., J.P. Hoehn, G.C. Christie. 2003. Using an Economic Model of Recreational
Fishing to Evaluate the Benefits of Sea Lamprey (Petromyszon marinus) Control
on the St. Marys River. Journal of Great Lakes Research. 29 (Supplement 1):
742-754.
Lupi, F., M.D. Kaplowitz, S.A. Weicksel, and M. Chen. 2012. “Who visits Michigan’s
Great Lakes beaches and do they value Great Lakes water quality?” Presented
at the Great Lakes Beach Association Annual Conference. Mackinaw Island, MI.
October 17.
Malkin, S.Y., S.J. Guilford, and R.E. Hecky. 2008. “Modeling the Growth Response of
Cladophora in a Laurentian Great Lake to the Exotic Invader Dreissena and to
Lake Warming.” Limnology and Oceanography. 53(3): 1111-1124.
McConnell, K.E. 1977. “Congestion and Willingness to Pay: A Study of Beach Use.”
Land Economics. 53(2): 185-195.
McFadden, D. 1974, "Conditional Logit Analysis of Qualitative Choice Behavior." in P.
Zarembka (ed.), Frontiers in Econometrics. New York: Academic Press pp. 10521.
MDEQ. 2009. Summary of Inland and Great Lakes Beach Monitoring, Actions, and
Beach Attributes. Microsoft Excel workbook.
MDEQ. 2011. “Combined Sewer Overflow (CSO) Sanitary Sewer Overflow (SSO) and
Retention Treatment Basin (RTB) Discharge. 2010 Annual Report.” Available
online: http://www.michigan.gov/documents/deq/deq-oea-cso-sso2010_377010_7.pdf
MDEQ. 2012a. “Michigan Beach Monitoring Year 2011 Annual Report.” Michigan
Department of Environmental Quality, Water Resources Division. Staff Report.
MI/DEQ/WRD-12/034. August.
MDEQ. 2012b. “Shorelines of the Great Lakes.” From "Shoreline of the Great Lakes and
Connecting Rivers", U.S. Lake Survey, File No. 3-3284, corrected to 1952.”
Online: http://www.michigan.gov/deq/0,4561,7-135-3313_3677-15959--,00.html
accessed 8/12/2012

208

Michigan Sea Grant. n.d. “Harmful Algal Blooms and Muck: What’s the Difference?”
Fact Sheet. Online: www.glerl.noaa.gov/pubs/brochures/HAB_Muck.pdf
accessed Aug 10 2012.
Morrison, M., J. Bennett, R. Blamey and J. Louviere. 2002. “Choice Modeling and Tests
of Benefit Transfer.” American Journal of Agricultural Economics. 84 (1): 161170.
Morrison, M., and O. Bergland. 2006. “Prospects for the use of Choice Modelling for
Benefit Transfer.” Ecological Economics. 60(2): 420-428.
Murray, C., B. Sohngen, and L. Pendleton. 2001. Valuing Water Quality Advisories and
Beach Amenities in the Great Lakes. Water Resources Research. 37(10): 25832590.
Natural Resource Defense Council. 2012. “Testing the Waters 2012 annual report.”
http://www.nrdc.org/water/oceans/ttw/default.asp accessed 8/12/2012
Northeast-Midwest Institute and the National Oceanic and Atmospheric Administration.
2001. “Revealing the Economic Value of Protecting the Great Lakes.”
Washington, D.C.
Olapade, O.A., M.M. Depas, E.T. Jensen, and S.L. McLellan. 2006. “Microbial
Communities and Fecal Indicator Bacteria Associated with Cladophora Mats on
Beach Sites along Lake Michigan Shores.” Applied and Environmental
Microbiology. 72(3):1932–8.
Park, C.S., and V. Srinivasan. 1994. “A Survey-Based Method for Measuring and
Understanding Brand Equity and Its Extendibility.” Journal of Marketing
Research. 31(2): 271-288.
Parsons, G.R., D.M. Massey, and T.D. Tomasi. 1999. “Familiar and Favorite Sites In a
Random Utility Model of Beach Recreation.” Marine Resource Economics. 14(4):
299-315.
Parsons G.R. and M. Powell. 2001. "Measuring the Cost of Beach Retreat." Coastal
Management volume 29, pp. 91-103.
Parsons, G.R., A.K. Kang, C.G. Leggett, and K.J. Boyle. 2009. “Valuing Beach Closures
on the Padre Island National Seashore.” Marine Resource Economics, 24(3):
213-235.
Piper, S. and W. E. Martin. 2001. “Evaluating the accuracy of the benefit transfer
method: A rural water supply application in the USA.” Journal of Environmental
Management. 63 (3): 223-235. November.

209

Poe, G.L., M.P. Welsh, P.A. Champ. 1997. “Measuring the Difference in Mean
Willingness to Pay When Dichotomous Choice Contingent Valuation Responses
Are Not Independent.” Land Economics. 73(2): 255-267.
Pozo, V.F., G.T. Tonsor, and T.C. Schroeder. 2012. “How Choice Experiment Design
Affects Estimated Valuation of Use of Gestation Crates.” Journal of Agricultural
Economics. 63(3): 639-655.
Racevskis, L.A., and F. Lupi. 2008. “Incentive Compatibility in an Attribute-Based
Referendum Model.” Selected Paper prepared for presentation at the American
Agricultural Economics Association Annual Meeting, Orlando, FL, July 27-29.
Rolfe, J. and Bennett, J. 2009. “The Impact of Offering Two Versus Three Alternatives
in Choice Modelling Experiments.” Ecological Economics. 68(4): 1140-1148.
Rolfe, J. and J. Windle. 2011. "Assessing national values to protect the health of the
Great Barrier Reef." Paper presented at the 55th Annual Australian Agricultural
and Resource Economics Society Conference, 8-11th February, Melbourne.
Rose J.B., R.M. Atlas, C.P. Gerba, M.J. Gilchrist, M.W. LeChevallier, M.D. Sobsey, and
M.V. Yates. 1999. “Microbial Pollutants in Our Nation’s Water: Environmental and
Public Health Issues.” American Society for Microbiology. Washington, D.C.
Rose, J.M., D.A. Hensher, S. Caussade, J.D. Ortuzar, and R. Jou. 2009. “Identifying
differences in willingness to pay due to dimensionality in stated choice
experiments: a cross country analysis.” Journal of Transport Geography. 17(1):
21-29. January.
Rosenberger, R.S. and J.B. Loomis. 2001. “Benefit transfer of outdoor recreation use
values: A technical document supporting the Forest Service Strategic Plan (2000
revision).” Gen. Tech. Rep. RMRS-GTR-72. Fort Collins, CO: U.S. Department of
Agriculture, Forest Service, Rocky Mountain Research Station.
Rosenberger, R.S. and J.B. Loomis. 2003. “Benefit Transfer.” In P.A. Champ, K.J.
Boyle, and T.C. Brown (Eds.), A Primer on Nonmarket Valuation. Norwell, MA:
Kluwer, pp. 445-482.
Rosenberger, R.S. and T. Phipps. 2007. “Correspondence and Convergence in Benefit
Transfer Accuracy: Meta-Analytic Review of the Literature.” In S. Navrud and R.
Ready (eds.) Environmental Value Transfer: Issues and Methods. Dordrecht, The
Netherlands: Springer, pp. 23-44.
Rubey, L. and F. Lupi. 1997. “Predicting the Effects of Market Reform in Zimbabwe: A
Stated Preference Approach.” American Journal of Agricultural Economics.
79(1): 88-89. February

210

Saginaw Bay Science Committee Pathogen Work Group. 2007. “Saginaw Bay Coastal
Initiative: Potential Public Health Risks Associated with Pathogens in Detritus
Material (“Muck”) in Saginaw Bay.” Steven E. Chester, Director. Joan Rose,
Chair.
Schwarz, N., and S. Sudman. 1996. Answering questions: methodology for determining
cognitive and communicative processes in survey research. San Francisco:
Jossey-Bass.
Seiler, M.J., M.T. Bond, V.L. Seiler. 2001. “The Impact of World Class Great Lakes
Water Views On Residential Property Values.” Appraisal Journal 69 (3). July.
Shaikh, S.L. 2005. “Does Uncertainty Matter? An Application to the Willingness to Pay
to Reduce Swimming Bans in Chicago.” Selected Paper prepared for
presentation at the American Agricultural Economics Association Annual
Meeting, Providence, RI. July. Draft. Cited with permission.
Shaikh, S.L. 2012. “The Value of Beach Visits and the Effect of Swim Bans in Chicago.”
Presentation at the Delivering Health Water Workshop, University of Stirling.
October. http://www.deliveringhealthywater.net/wpcontent/uploads/2012/10/Sabina-Shaikh_REVISED.pdf
Shivlani, M.P., Letson, D., and M. Theis. 2003. “Visitor Preferences for Public Beach
Amenities and Beach Restoration in South Florida.” Coastal Management 31 (4):
367-385.
Smith, V.K., X. Zhang, and R.B. Palmquist. 1997. “Marine Debris, Beach Quality, and
Non-Market Values.” Environmental and Resource Economics 10(3): 223–247.
Sohngen, B., F. Lichtkoppler, and M. Bielen. 1999. “The Value of Day Trips to Lake Erie
Beaches.” Ohio Sea Grant College Program, Technical Bulletin Series.
Publication OSHU-TB-039.
Swait, J. and J. Louviere. 1993. “The Role of the Scale Parameter in the Estimation and
Comparison of Multinomial Logit Models.” Journal of Marketing Research. 30(3):
305-314. August.
Swait, J., and W. Adamowicz. 2001. “The Influence of Task Complexity on Consumer
Choice: A Latent Class Model of Decision Strategy Switching.” Journal of
Consumer Research. 28(1): 135-148.
Swait, J., T. Erdem, J. Louviere, and C. Dubelaar. 1993. “The Equalization Price: A
Measure of Consumer-Perceived Brand Equity.” International Journal of
Research in Marketing. 10(1): 23-45.

211

U.S. EPA. 1999. “Combined Sewer Overflow Management Fact Sheet: Sewer
Separation.” Office of Water. EPA 832-F-99-041. Washington, D.C.
U.S. EPA. 2008. “Great Lakes Beach Sanitary Survey User Manual.” Office of Water.
EPA-823-B-06-001. May.
http://water.epa.gov/type/oceb/beaches/upload/2008_05_29_beaches_sanitarys
urvey_user-manual.pdf
U.S. EPA. 2009. “Fiscal Year 2010 EPA Budget in Brief.” United States Environmental
Protection Agency Office of the Chief Financial Officer (2710A) Publication
Number: EPA-205-S-09-001. May.
U.S. EPA. 2010. “Fiscal Year 2011 EPA Budget in Brief.” United States Environmental
Protection Agency Office of the Chief Financial Officer (2710A) Publication
Number: EPA-205-S-10-001. February.
U.S. EPA. 2011. “Fiscal Year 2012 EPA Budget in Brief.” United States Environmental
Protection Agency Office of the Chief Financial Officer (2710A) Publication
Number: EPA-190-S-11-001. February.
U.S. EPA. 2012a. “Fiscal Year 2013 EPA Budget in Brief.” United States Environmental
Protection Agency Office of the Chief Financial Officer (2710A) Publication
Number: EPA-190-S-12-001. February.
U.S. EPA. 2012b. “Fiscal Year 2013 Justification of Appropriation Estimates for the
Committee on Appropriations.” EPA-190-R-12-001.
http://www.epa.gov/planandbudget/annualplan/FY_2013_CJ.pdf
U.S. Policy Committee. 2002. “Great Lakes Strategy 2002, A Plan For The New
Millennium.” A Strategic Plan for the Great Lakes Ecosystem Developed by the
U.S. Policy Committee for the Great Lakes. Final.
Vanden Heuvel, A., C. McDermott, R. Pillsbury, T. Sandrin, J. Kinzelman, J. Ferguson
M. Sadowsky, M. Byappanahalli, R. Whitman , and G.T. Kleinheinz. 2010. “The
Green Alga, Cladophora, Promotes Escherichia coli Growth and Contamination
of Recreational Waters in Lake Michigan.” Journal of Environmental Quality.
39(1):344.
Verhougstraete, M.P., M.N. Byappanahalli , J.B. Rose, and R. L. Whitman. 2010.
Cladophora in the Great Lakes: impacts on beach water quality and human
health. Water Science and Technology 62(1): 68-76.
Whitehead, J.C., P.A. Groothuis, R. Southwick, P. Foster-Turley. 2009. “Measuring the
economic benefits of Saginaw Bay coastal marsh with revealed and stated
preference methods.” Journal of Great Lakes Research. 35(3): 430-437.

212

Wilson, M.A., J.P. Hoehn. 2006. “Valuing Environmental Goods and Services Using
Benefit Transfer: The State-of-the Art and Science.” Ecological Economics. 60
(2): 335-342.
Wilson, K.A., E.T. Howell, and D.A. Jackson. 2006. “Replacement of Zebra Mussels by
Quagga Mussels in the Canadian Nearshore of Lake Ontario: the Importance of
Substrate, Round Goby Abundance, and Upwelling Frequency.” Journal of Great
Lakes Research 32(1): 11-28.
Wooldridge, J.M. 2010. Econometric Analysis of Cross Section and Panel Data (Second
Edition). Cambridge, MA: MIT Press.

213