\full

,3»: I! , villi.
”NJ.

“1...: A. U3: .,
. 7?... to. 3
A, :3: 5a... .6...

4B

39.:
..f.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

rrnwu-1.a.......§.hfi..u|.l ‘

.mrnianfi ‘
:fifitfi .

13:4

‘5‘:
5:.
s‘i‘
f3

z:

 

 

 

THESIS

Date

lllllllflllllllllUIWWIJHIIHIIHWill!!!”HIHIIWI

302050 9935

 

This is to certify that the
thesis entitled

Evaluation of Non—Response Bias
in a Study of Great Lakes Sport
Fish Consumption and Conception Failure

presented by

Eugene Michael Tay

has been accepted towards fulfillment
of the requirements for

Master's Epidemiology

 

 

degree in

W

Major professor

February 15, 2000

 

0-7 639

MS U is an Affirmative Action/Equal Opportunity Institution

 

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

 

DATE DUE

DATE DUE

DATE DUE

 

‘

 

c“.
C'.
Hi

6
Kb
13

A
‘1:

V [k A‘ ‘rl~]
t) [l [U

I.)

 

04

 

 

 

 

 

 

 

 

 

 

 

 

woo chlRC/DdoDuopGS-pfl

T"

EVALUATION OF NON-RESPONSE BIAS IN A STUDY OF GREAT LAKES
SPORT FISH CONSUMPTION AND CONCEPTION FAILURE

By

Eugene Michael Tay

A THESIS

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

MASTER OF SCIENCE

Department of Epidemiology

2000

ABSTRACT

EVALUATION OF NON-RESPONSE BIAS IN A STUDY OF GREAT LAKES
SPORT FISH CONSUMPTION AND CONCEPTION FAILURE

By

Eugene Michael Tay

Research previously conducted by Courval et a1.1 has suggested a modest
association in men of Great Lakes sport fish consumption with risk of conception delay.
However, a low response rate raised concerns about non-response bias as an explanation
for these findings. This study was performed to evaluate whether non-response bias
could have occurred. Telephone interviews were conducted with 230 men and 38 women
who did not respond to the original survey. Non-responders were compared to the
original responders on key demographic, behavioral, and reproductive characteristics.
Non-responders were approximately 1.5 years older at interview, were more likely to be
Caucasian, and reported higher incomes than responders. No differences were found with
respect to education level, marital status, or smoking. Non-responders fished fewer days
in the past year and consumed fewer fish meals than responders. Compared with
responders, non-responders were more likely to have had two or more children and were
less likely to intend to have additional children within the next five years. However,
among both non-responders and responders there was an increased prevalence of a period
of conception failure among men who reported consuming greater quantities of Great
Lakes sport fish. These results suggest that non-response bias is unlikely to have played

a major role in the observed association of sport fish consumption and conception delay.

 

l Courval JM, DeHoog JV, Stein AD, Tay EM, He JP, Humphrey HEB, Paneth N. Sport-caught fish consumption and
conception failure in licensed Michigan anglers. Environ Res 1999;80:8183-8188.

To DAVID CAMERON
whose impending birth finally motivated me
to finish this thesis essay

To LAURA JANE
who had to endure six years of
“Don’t worry — I’m going to finish my thesis”

iii

ACKNOWLEDGMENTS

I would first like to acknowledge Dr. Jeanne Courval and Dr. Aryeh Stein. Dr.
Courval and Dr. Stein were instrumental in involving me with the Fisheaters Family
Health Project and provided much guidance in the development, execution, and data
analysis of this project.

I would also like to acknowledge the role of my committee members, Dr. Nigel
Paneth, Dr. Joseph Gardiner, and Dr. Lynda Farquhar, who provided meaningful
feedback and constructive comments on this thesis essay.

Finally, I would like to acknowledge the Environmental Protection Agency,
which supported me and this research under a Science to Achieve Results (STAR)

Graduate Student Fellowship Award.

iv

TABLE OF CONTENTS

LIST OF TABLES .......................................................................................................... vii

LIST OF FIGURES ......................................................................................................... ix

LIST OF ABBREVIATIONS ............................................................................................ x

INTRODUCTION ............................................................................................................. 1

CHAPTER 1

BACKGROUND ............................................................................................................... 3
Polychlorinated biphenyls ...................................................................................... 3
Toxicity of polychlorinated biphenyls ................................................................... 4
Reproductive toxicity of polychlorinated biphenyls in humans ............................ 6
The F isheaters Family Health Project .................................................................... 6

CHAPTER 2

NON-RESPONSE BIAS ................................................................................................... 9
The postal survey ................................................................................................... 9
Limitations of postal surveys ............................................................................... 10
Handling non-response ........................................................................................ 12
Non-response bias in social science literature ..................................................... l4
Non-response bias in epidemiological and health sciences literature .................. 17

CHAPTER 3

METHODS ...................................................................................................................... 33
Population and setting .......................................................................................... 33
Sampling .............................................................................................................. 35
Identification of telephone numbers .................................................................... 35
Interview .............................................................................................................. 36
Data entry ............................................................................................................. 37
Statistical methods ............................................................................................... 38

CHAPTER 4

RESULTS ........................................................................................................................ 40
Demographic characteristics of men .................................................................... 43
Behavioral characteristics of men ........................................................................ 48
Demographic and behavioral characteristics of women ...................................... 48
Fishing habits and fish consumption ................................................................... SO
Reproductive characteristics ................................................................................ 54
Logistic regression analyses ................................................................................ 57

CHAPTER 5

DISCUSSION .................................................................................................................. 59
APPENDIX A
TELEPHONE CALLING RULES OF REPLACEMENT .............................................. 66
APPENDIX B
QUESTIONNAIRE ......................................................................................................... 68
APPENDIX C
QUESTIONNAIRE SCRIPT AND LOG SHEET .......................................................... 71
REFERENCES ................................................................................................................ 73

vi

TABLE 1:

TABLEZ:

TABLE 3:

TABLE 4:

TABLE 5:

TABLE 6:

TABLE 7:

TABLE8:

TABLE 9:

TABLE 10:

TABLE 1 1:

LIST OF TABLES

Responses to three mailings in a postal survey of North Carolina
peach growers, 1946 ................................................................................... 14

Summary socio-demographic variables and their relationship with
response status ............................................................................................ 16

Age-adjusted death rates for responders and non-responders for the
first five years of the F ramingharn study, 1953-57 ..................................... l7

Age-adjusted comparisons of responders and non-responders to
responses on health status in a population-based study of
cardiovascular disease ................................................................................. 18

Comparison of baseline characteristics of non-responders and
responders to a mailed follow-up survey of 15,440 persons who
attended a preventive medicine center in Dallas, Tx at least once
from 1972-81 .............................................................................................. 21

Summary of studies of non-response bias in the epidemiological and
health sciences literature ............................................................................. 23

Participation status in a survey of Michigan licensed anglers, by
gender and region ........................................................................................ 41

Demographic, socio-economic, and behavioral characteristics among
male non-responders and responders to a survey of Michigan
licensed anglers ............. 45

Demographic, socio-economic, and behavioral characteristics among
male non-responders and responders to a survey of Michigan
licensed anglers, by geographic region ....................................................... 46

Demographic, socio-economic, and behavioral characteristics among
female non-responders and responders to a survey of Michigan
licensed anglers ........................................................................................... 49

Fishing habits and sport-caught Great Lakes fish consumption in past

year among non-responders and responders to a survey of Michigan
licensed anglers, by sex ............................................................................... 52

vii

TABLE 12: Fishing habits and sport-caught Great Lakes fish consumption in past
year among male non-responders and responders to a survey of
Michigan licensed anglers, by geographic region ....................................... 53

TABLE 13: Reproductive history among non-responders and responders to a
survey of Michigan licensed anglers ........................................................... 55

TABLE 14: Reproductive history among non-responders and responders to a
survey of Michigan licensed anglers, by geographic region ....................... 56

TABLE 15: Associations between sport-caught Great Lakes fish consumption
and prevalence of a period of conception failure among male
responders and non-responders to a survey of Michigan licensed
anglers ........................ . ................................................................................ 58

viii

LIST OF FIGURES

FIGURE 1: Basic structure of polychlorinated biphenyls ................................................. 3
FIGURE 2: Target counties for the Fisheaters Family Health Project, Michigan ........... 34
FIGURE 3: Tracking flowchart of study participants ...................................................... 42

ix

LIST OF ABBREVIATIONS

PCB ............................................................................................. Polychlorinated biphenyl

FFHP ............................................................................... Fisheaters Family Health Project

INTRODUCTION

FISHING in the State of Michigan is big business. An estimated two million
Michigan anglers along with 334,000 nonresident tourists fish Michigan waters annually.
Collectively they contribute close to $1.4 billion to Michigan’s economy in pursuit of
sport-fishing (1).

Not only is fishing a great recreational activity for some, but eating Great Lakes
sport fish is a healthy choice as well. Fish are a highly nutritious food - they are a good
source of protein and are low in saturated fat, and they contain many valuable vitamins
and minerals. Certain sport fish contain omega-3 polyunsaturated fatty acids, which can
lower triglyceride and cholesterol levels in the blood (2, 3). Eating fish regularly can
reduce mildly elevated blood pressure and prevent hardening of the arteries as well as
heart disease (2, 4).

These benefits, however, are not without some potential drawbacks. Studies
beginning in the early 1980’s suggested potential developmental delay in children of
sport fish-eating mothers (5-9). The concern has been about the chemicals found in sport
fish of the Great Lakes basin. Organochlorine compounds such as polychlorinated
biphenyls, or PCBs, do not degrade quickly or easily in the environment. Despite the
banning of PCBs from production in 1976, it is estimated that only about five percent of
the 3.4 billion pounds of PCBs made worldwide have been destroyed or degraded (10).
Furthermore, PCBs accumulate in the fatty tissues of fish over their lifetimes in a process
known as bioaccumulation. The resulting concentration of PCBs in fish can be millions

of times greater than the concentration of the chemicals in the water in which they live.

In the Great Lakes region, consumption of contaminated fish has been identified as an
important exposure route (11).

Because of these potential concerns, Michigan’s fish advisory program was
implemented in the 1970’s and is updated annually. Women of child-bearing age and
children under age 15 are advised to limit their intake or avoid certain types of sport fish
altogether, depending on the species and size of fish and the location at which the fish
were caught (12).

Recently, more intensive efforts at characterizing the potential health effects of
sport fish on humans have been made. An emerging area of research has focused on the
effect of sport fish consumption on reproductive and endocrine function. The Fisheaters
Family Health Project at Michigan State University was established to pursue this
research initiative.

Research conducted by Courval et al. (13) has suggested a modest association, in
men only, of Great Lakes sport fish consumption with risk of conception delay.
However, these results were not without their limitations. A low response rate in that
study raised concerns about the validity of these findings. Responders may have differed
from non-responders, and this difference, termed non-response bias (14), may have been
a factor in the observed results. Therefore, the present study was conducted to evaluate
whether non-response bias could explain these findings. The results of the present study

are presented hereinafter and have also been published in the academic literature (15).

CHAPTER 1

BACKGROUND

Polychlorinated biphenyls

Polychlorinated biphenyls (PCBs) are a group of manufactured organic chemicals
comprised of two covalently bonded benzene rings with chlorine substitution at any of
the remaining carbons (Figure 1). They have the empirical formula of C,2H,o,,,Cl,,, where
n=1-10. However, due to steric hindrance and electrostatic factors, generally only four to
eight of the available carbon atoms are chlorinated at one time. There are 209 possible

PCB compounds, called congeners.

 

meta- ortho-
3’ 2' 2 3
para. 4. O O 4
6' 6' 6 6
meta- ortho-

 

 

 

Figure 1: Basic structure of polychlorinated biphenyls.

PCBs were first synthesized in 1881 and manufactured commercially from 1929
until 1977 in the United States by the Monsanto Chemical Company, under the trade
name Aroclor. Aroclors were designated by four digit numbers. The first two digits
specify the number of carbon atoms in the biphenyl group (with the exception of Aroclor
1016), while the last two digits indicate the approximate percentage of chlorine content

by weight in the compound, e.g., Aroclor 1254 contains 54% chlorine content by weight.

Commercial mixtures generally contained between 20 and 60 percent chlorine content by
weight.

The amount of chlorination confers different chemical and physical properties to
individual congeners. In general, however, PCBs are lipophilic and hydrophobic,
resistant to acids, bases, and oxidation/reduction reactions, nonflarnmable,
nonconducting, and heat-resistant. These properties made them well-suited to a variety of
industrial applications, such as insulators in transformers and capacitors, lubricants and
hydraulic fluids, plasticizers, as flame retardants, and in pesticides, paints, sealants, glues,
and carbonless copy paper.

Unfortunately, these very properties that made PCBs ideal in industrial
applications create problems for the environment. Their resistance to degradation causes
PCBs to persist in the environment. Furthermore, their lipophilic nature causes PCBs to
bioaccumulate in the food chain, with the amount of toxic chemicals increasing with each

link up the food chain, a process known as biomagnification.

Toxicity of polychlorinated biphenyls

The toxicity of PCBs, like their chemical and physical properties, is a function of
the structure of the individual congener, which in turn depends on its amount of
chlorination. Individual congeners can assume a coplanar or non-planar conformation
depending on the amount and position of the chlorine substitution. Chlorine atoms can
be substituted at the ortho, meta, or para positions (Figure 1).

Safe (16) summarized that those congeners that assume a coplanar conformation,

namely 3,3’,4,4’-tetraCB, 3,3’,4,4’,5-pentaCB, and 3,3’,4,4’,5,5’-hexaCB, are

approximate stereoisomers of 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) and elicit
many of their toxicological effects through the same mechanism, hypothesized to be
through the binding of the PCB compound to the Ah receptor to induce hepatic
micosomal enzymes. The toxicity of such PCBs can be expressed in terms of TCDD-
equivalency factors (TEFs), which is the potency of a particular congener to elicit a toxic
effect relative to the potency of TCDD. In addition, there has been increasing focus on
the toxic effects of non-coplanar, ortho substituted congeners that elicit their toxicity via
unknown mechanisms (17).

Toxic responses to PCBs have been demonstrated in laboratory animals,
observations in wildlife, and epidemiologic studies in humans. The literature on PCB
toxicology is extraordinarily vast and many reviews have been published (16, 18-27).

In humans, toxicity as a result of PCB exposure came to prominence in 1968
when over 1,300 persons in Japan became ill from eating PCB-contaminated rice oil, in
an incident later coined “Yosho” (28). A similar accidental poisoning occurred in Taiwan
in 1979, called “Yu-cheng” (29). Consumption of PCB-contarninated rice oil resulted in
a severe form of acne called chloracne and hyperpigrnentation of the skin, as well as
fatigue, nausea, and liver disorders (30).

Since then, researchers have studied the possible toxic effects of PCBs, which
may include carcinogenicity (31-38), as well as hepatic (20, 32), derrnatologic (32, 39-
41), immune (42, 43), pulmonary (44), neurologic (45-48), developmental (5, 7, 8, 41,

48-54), endocrine (55), and reproductive (5, 41, 56-60) dysfunction.

Reproductive toxicity of polychlorinated biphenyls in humans

In several studies of reproductive toxicity of PCBs, women exposed to PCBs
prenatally or occupationally gave birth to children who were slightly lighter in weight
than those born to women less exposed (5, 56, 57), although this finding has not been
consistently found (61, 62). In addition, women with higher serum PCB levels were
more likely to have miscarriages than the general population (58), although these results
may be confounded (63). More recently, women who ate more than one sport fish meal
per month from Lake Ontario had shorter menstrual cycles compared to women who did
not eat fish and a slight conception delay, although the latter finding was not statistically
significant (59, 60).

In men, Bush et a1. (64) found that the concentration of certain PCB congeners
was inversely correlated with sperm motility index. Other studies (65, 66) which have
reported a decline in semen quality and sperm count have heightened concern about

possible reproductive effects of organochlorine compounds.

The F isheaters Family Health Project

The Fisheaters Family Health Project (FFHP) at Michigan State University was
established in order to further our understanding of the exposure to and potential adverse
human health effects from environmental PCB exposure. Finding a suitable population to
study these effects, however, can prove challenging. Few large populations are
chronically exposed to detectable levels of PCBs. One group of individuals potentially at

riSk of PCB exposure is anglers of the Great Lakes who consume their catch.

Consumption of contaminated fish from the waters of the Great Lakes has been identified
as an important exposure route (11).

For the FFHP, a postal questionnaire was developed to survey anglers and their
partners about their demographic characteristics, behavioral and fish consumption habits,
and reproductive histories (67). A source population of licensed anglers was selected
from ten Michigan counties with Great Lakes shorelines. These anglers were identified
using fishing license data obtained from the Michigan Department of Natural Resources.
Surveys were mailed to 4,931 reproductive-age male and female licensed anglers aged
18-34 years between 1993 and 1995. Five hundred and five questionnaires were returned
as undelivered; of the anglers who were presumed to have received the screening survey,
1,445 returned the questionnaire, giving an actual response rate of 1,445/(4,931-505) =
33%. Questionnaires were also received from 840 of their partners.

Data obtained from the original F FHP screening survey suggested a modest
association, in men only, of Great Lakes sport fish consumption with risk of conception
delay (13). However, these results were not without their limitations, as the low response
rate and the simultaneous ascertainment of exposure and outcome raised concerns about
bias as possible explanations for these findings.

In order to address some of these limitations, a prospective study was developed.
From the previous postal survey, 375 couples and 1,030 individuals who were consumers
of Great Lakes sport fish and who planned to have children in the near future were
identified. These individuals were most suitable for a prospective study of reproduction
and/or correlates of infertility in relation to PCB exposure. The methods and current

progress of this prospective study have been described (68). Furthermore, a study of

potential non-response bias was also undertaken, the results of which are presented

hereinafter and have also been published in the academic literature (15).

CHAPTER 2

NON-RESPONSE BIAS

The postal survey

The postal survey is a valuable epidemiological research tool. One of the primary
reasons for its high prevalence in epidemiological research today is its cost and time
effectiveness. Personal and telephone interviews demand significant time and human and
financial resources. Multiple telephone calls are often required in order to contact
individuals for interviewing or for setting up appointments for personal interviews.
Personal interviews have the added cost of transportation of the interviewer to the
interview site. If multiple interviewers are used, interviewer variability can become a
concern.

In contrast, postal surveys do not require an interviewer and can be done with
bulk mailings. Thus, data can be procured quicker, cheaper, and more abundantly with a
postal survey. Questionnaires can be mailed to eligible participants, filled out at home at
the participants’ convenience, and returned by mail. In addition, postal surveys tend to be
more valid than personal or telephone interviews because they enable respondents to
check information by verifying their records or consulting with other members of the
family, and because they permit more leisurely and thoughtful reply (69). Also, they
often elicit responses fi'om people too busy for personal interviews and avoids interviewer
or respondent bias for topics that are potentially embarrassing in a personal interview

situation (70). Enrollment of subjects into a prospective epidemiologic study, such as the

one carried out by us for the Fisheaters Family Health Project (67), is often accomplished

using such a self-administered questionnaire.

Limitations of postal surveys

Postal surveys, however, are not without their concerns. In the present society, an
endless onslaught of bulk mailings and solicitations deluge individuals and make them
reluctant to respond or even open before discarding. F urthennore, the rising prevalence
of surveys, including those in the health professions field, may have created a “survey
fatigue” in the population (71). The majority of Americans have participated in at least
one survey and a substantial proportion are asked to do so again each year (72). In many
cases, the experience of being a survey participant is a negative one, leading to an
increased disinclination to participate in future surveys (73). Taken together, these
factors have made it difficult to achieve high response rates.

Indeed, response rates since the 1950’s have decreased markedly. Steeh (74)
found that refusal rates in two ongoing trend surveys carried out by a major university
survey research center increased from only 6-8 percent in the 1950’s to 15-20 percent by
the end of the 1970’s. Refusal rates in one of these surveys that has continued to the
present have leveled off in recent years at 18-20 percent (75). Other authors have found
refusal rates as high as 30 percent in major academic surveys (73) and even higher in
commercial surveys (76). Increasing concerns about privacy and confidentiality and how
the data will be used are primary reasons why some individuals choose not to participate

(77). Some authors have speculated that the decline in participation rates can be

10

attributed to a decline in people’s sense of social responsibility and less belief in the
legitimacy of social institutions (78).

This high refusal rate, or non-response, is a major limitation of survey research
and can seriously challenge the external validity of survey results. Bias can occur when
those who do not respond to a survey (the “non-responders”) differ from those who do
(the “responders”). This bias is termed “non-response bias” (14). When non-response
bias occurs, the study population may not be representative of the total survey population.
Results drawn from studies of this study population are not necessarily generalizable to
the entire group invited to participate and incorrect conclusions may be drawn. Response
that is correlated with exposure levels and disease outcome may artificially increase or
decrease measures of association in cross-sectional, case-control, and cohort studies (79-
83).

Non-response bias is most likely to occur when response rates are low. In this
situation, survey estimates will become biased the more non-responders differ from
responders. A low response rate, however, does not necessarin indicate that survey
estimates are biased by non-response. When responders and non-responders do not
differ, the response rate will have no effect on prevalence estimates. Conversely, high
response rates does not necessarily protect against non-response bias, as this bias can still
be important if the few remaining non-responders differ markedly from responders. For
example, in a postal survey of almost 2,500 economically active and capably employable
disabled persons, Sheikh and Mattingly (84) obtained an 84% response rate; nevertheless,
non-responders were significantly different from responders with respect to employment

status and training. Regardless, it has been suggested that in order to best minimize the

11

potential for non-response bias, response rates of at least 80 percent should be achieved

(14, 85).

Handling non-response

There are multiple ways of dealing with non-response. Ideally, the best strategy
to limit non-response bias is to maximize response rates. Several reviews and meta-
analyses have been published (86-89), and the success of individual methods has been
reported (69, 70, 90-101). These methods involve efforts revolving around the timing
and techniques of survey administration. Some of the important factors include financial
incentives, follow-ups, preliminary notification, questionnaire characteristics (length,
color, subject matter), cover letter characteristics (length, type of appeal, format, status of
signer), sponsoring institution, deadlines, time of posting, type and amount of outgoing
and retum postage, and respondent anonymity (102, 103). However, the particular details
behind these methods are beyond the scope of this discussion. For the Fisheaters Family
Health Project, several of these methods were employed to boost the response rate (67).

Given that increasing response rates often proves challenging, other strategies
have been advocated in order to detect a possible bias. Most commonly, information is
collected on a sample of non-responders. Non-responders can be contacted by telephone
or in person and persuaded to complete a full or shortened survey. Then, a simple
comparison of characteristics and responses of responders and non-responders can be
made to determine if and how non-responders differ from the responders. However, it
should be noted that similarity between responders and non-responders with regard to

certain characteristics does not necessarily imply similarity with regard to exposure and

12

disease status, and the exact influence of any bias on survey variables cannot easily be
determined.

Another strategy for detecting a possible bias is to compare the responses of
“early” responders, or those who respond before a follow-up letter is sent, versus “late”
responders, or those who respond after a follow-up letter is sent, to see if there are any
trends operating. The assumption of this strategy is that responders are ordered on a
continuum, and that non-responders are simply “super-late” responders. Considerable
differences have often (104), but not always (105), been found between early and late
respondents.

A final strategy for detecting a possible bias is to compare responders and non-
responders on characteristics for which complete data are available, such as in the case of
studies of veterans using military records. Seltzer et al. (106) was able to use military
records in a postal questionnaire of smoking habits of US. veterans and found that only
67% of smokers returned their questionnaire within 30 days, in contrast to 85% of non-
smokers.

If a significant difference between responders and non-responders or early versus
late responders is detected, statistical methods for offsetting the bias can be employed.
For example, weighting adjustments can also be employed. Adjustment using data from
a sub-sample of non-responders followed-up intensively, weighted for the original
proportion of non-responders, can be performed. However, weighting can sometimes
produce disastrous results because weighting schemes assume that (1) the respondents
represent a random sample of their subgroup, and (2) all relevant differences between

responders and non-responders have been taken into consideration in defining subgroups,

13

both of which are tenuous assumptions (102). While weighting adjustments are easy to
implement and often employed, they at best will attenuate, but not necessarily eliminate,
non-response bias. For a more detailed discussion of data analysis strategies for taking

non-response into account, see Kessler et al. (107).

Non-response bias in social science literature

The literature on survey response is enormous. Some of the first studies were
published as early as the 1920’s and 1930’s, although the bulk of the literature did not
begin to appear until the 1940’s. The majority of the early research came out of the social
sciences. Comprehensive bibliographies of these early studies have been published (108-
110). An example of one such early study is that conducted by Finkler in 1946 (111) on
commercial peach production by growers in the State of North Carolina. One of the
survey questions asked how many trees the grower owned; the exact number of trees

owned for each grower was previously known. The results are shown in Table 1.

Table 1: Responses to three mailings in a postal survey of North Carolina peach
growers, 1946 (111).

 

 

No. growers Percent of Mean no. of
population fruit trees per
grower

Response to first mailing 300 10 456
Response to second mailing 543 17 382
Response to third mailing 434 14 340
Non-responders after three mailings 1,839 59 290
Total population 3,1 16 100 329

 

The presence and direction of non-response are clear. Growers with a large

number of fruit trees were more likely to respond and more likely to respond in an earlier

14

mailing than growers with fewer trees. Population estimates based solely on the sample
obtained from the three mailings would have greatly overestimated the actual number
fruit trees owned by each grower.

Attempting to summarize the socio-demographic determinants of response is a
daunting task. This is due to the vast literature and the inconsistency within it. Bridge
(102) and Goyder (78), however, have reviewed the literature and have attempted to
summarize the existing evidence based on hundreds of early studies. Some of their main
conclusions are presented in Table 2.

From Table 2, the primary socio-demographic variables that are related to survey
response are education, income, and age. Education and income are positively correlated
with response, while age is negatively correlated with response. Said another way, non-
responders have historically tended to be less educated, have a lower income, and be of
an older age.

It should be noted that the single most important factor in predicting response is
interest in the topic of survey (102). If the topic interests the respondent, and s/he has
something positive to report, then the respondent is very likely to complete the
questionnaire. For example, in a 1939 study by Stanton (112), teachers who used radios
in the classroom were more likely to respond to a survey about the educational use of the
radio. In another study, Pace (113) found that alumni who had graduated from a
university were more likely than dropouts to respond to a questionnaire from the
university. Finally, Edgerton, Britt, and Norman (114) found that winners in a science
fair contest were most likely to respond to a follow-up survey, honorable-mention

recipients were next most likely to respond, and “others” were least likely to respond.

15

Table 2: Summary socio-demographic variables and their relationship with response
status (abstracted from Bridge [102] and Goyder [78]).

 

Variable

Effect

 

Education

Income

Age

Sex

Marital status

Ethnicity

Others

In general, education is positively correlated with response. However, in
some studies, highly educated individuals have been difficult to contact.

Income is positively correlated with response. However, as for education,
in some studies, the wealthy have been difficult to contact.

Age and response are negatively correlated for response once contacted
(older individuals are less likely to respond once contacted). Some studies
have reported that age and response are positively correlated for contact
(older individuals are easier to contact than younger individuals, because of
the frequent shifts in residence and active life style of the young), but this
finding was not consistent.

No relationship exists between sex and response.

Non-married individuals are more difficult to contact than married
individuals, but this may be confounded by age, as non-married individuals
tend to be younger. Once contacted, married and non-married individuals
generally respond at similar rates.

At least one study cited by Bridge (102) found that blacks had a lower
response rate than whites, but this study also noted that ethnicity was
highly correlated with income.

Other variables were evaluated, such as work status, religion, country of
birth, home ownership, dwelling type, family size, and urban vs. rural
habitat, but none had convincing evidence of a relationship with response.

 

l6

Non-response bias in epidemiological and health sciences literature

More recently, studies of non-response began to appear in epidemiological and
health sciences research predominately in the 1970’s and have grown in importance ever
since. An early study was performed by Gordon et al. (115) in 1959. Using the
Framingham cohort, the authors sampled 6,532 individuals for their study and were able
to perform examinations on 4,494 persons, for a response rate of 68.8%. The authors
found that more non-responders than responders died within the first five years of the
study, as depicted in Table 3.

Table 3: Age-adjusted death rates for responders and non-responders for

the first five years of the Framingham study, 1953-57 (adapted from
Gordon et al. [115]).

 

Annual age-adjusted death rate

 

Men Women
Responders (n=4,494) 8.2 3.2
Non-responders (n=1,964) 10.4 6.9
Refused (n=1,464) 11.9 7.3
Moved (n=426) 4.6 3.6
Incapacitated/lll (n=74) 14.8 1 1.9

Died during recruitment period (n=74) - -

 

For both males and females, non-responders had higher age-adjusted annual death
rates compared with responders. In particular, those non-responders who had refused to
participate, or those who “voluntarily” choose not to participate, had even higher age-
adjusted annual death rates. This contrasts with those non-responders who had moved
out of the study site, or those who were “involuntary” non-responders. As would be
expected, those who were incapacitated or too ill to participate had the highest annual

age-adjusted death rate within the first five years of the study. From these observations,

17

the authors suggested that non-responders were on average more likely to be seriously ill
than the responders.

Other studies have reached similar conclusions. In a population-based study of
cardiovascular disease in a planned suburban development in Southern California from
1972-1974, Criqui et al. (116) sampled 6,155 persons, of which 5,052 participated in the
study, for a response rate of 82.1%. Analysis was restricted to those aged 30-79. The
authors characterized the non-responders as less healthy than responders, at least in terms
of past cardiovascular illness, based on the data presented in Table 4.

Table 4: Age-adjusted comparisons of responders and non-responders to responses on

health status in a population-based study of cardiovascular disease (adapted from Criqui
et al. [116]).

 

 

 

 

Males Females
Non- Res- p“ Non- Res- pa
resonders ponders resonders ponders
Personal health history
Hosp. for heart failure % 4.2 1.1 0.0001 3.1 0.8 0.0001
Hosp. for heart attack % 9.0 8.0 0.28 4.1 3.0 0.13
History of diabetes % 4.9 5.7 0.31 6.0 3.0 0.0018
History of stroke % 3.7 3.0 0.28 1.7 1.3 0.27
Family history (1'‘ degree
relatives)
Family history heart attack % 22.8 34.2 0.0001 25.1 40.5 0.0001
1f previous question yes, % 12.9 15.1 0.32 17.8 18.9 0.4
was it at age 50 or under
Family history stroke % 20.3 22.3 0.23 15.6 30.5 0.0001
Family history diabetes % 12.7 15.7 0.10 14.8 17.8 0.09
Risk factors for disease
History of hyperlipidemia % 13.1 17.2 0.03 7.9 14.0 0.001
History of hypertension % 23.4 23.8 0.44 24.4 24.7 0.46
No. eggs eaten weekly # 4.0 4.2 0.25 3.0 3.2 0.08
Current cigarette smoker % 26.9 22.4 0.05 31.6 26.6 0.03

 

a Z-test for differences between proportions.

Both male and female non-responders were significantly more likely than
responders to have ever been hospitalized for a heart failure. Female but not male non-

responders were significantly more likely to have a history of diabetes. Both male and

18

female non-responders were also significantly more likely than responders to be current
cigarette smokers, a finding replicated in other studies (106, 117). Interestingly, both
male and female responders were more likely than non-responders to have a history of
hyperlipidemia or a first degree relative with a previous heart attack. The authors thus
characterized the responders as the “worried well”, or those who had lower prevalence of
disease but had higher prevalence of risk factors for disease, as compared to non-
responders.

Despite the differences found, however, Criqui et al. (116) believed that
responders were generally representative of the target population. They based this
conclusion for several reasons: the response rate was above 80%; subsets of the
population had similar age-sex distributions; and the differences, while statistically
significant, were generally small, suggesting that calculations of prevalence or relative
risk would be relatively unbiased. For example, the variable with the largest percentage
difference between responders and non-responders is the question of a family history of a
heart attack for females, 40.5% vs. 25.1%, respectively. The authors claim that if all non-
responders had participated, this percentage would have been reduced fiom 40.5% to
37.8%, and the effect of this bias on subsequent relative risk calculations would probably
have been minor. However, the authors did note that the potential for significant non-
response bias might increase in a study with larger differences between groups and/or a
higher non-response rate. If the response rate had been only 50%, the true prevalence of a
family history of a heart attack for females would have decreased from 40.5% to 32.8%.

In another health survey carried out by Macera et al. (118), all persons who

visited a free health clinic in Dallas, TX at least once from 1972-1981 were surveyed in

19

1982 to obtain health outcome information, including questions on physical activity,
demographics, and health conditions that had developed since their last visit. Of the
18,806 questionnaires mailed, 3,224 were returned for bad addresses and an additional
142 were not deliverable because of that person’s death. Of the 15,440 persons who
received the questionnaire, 11,972 responded (77.5%). To assess any potential non-
response bias, the authors compared data from the first clinic visit for the responders and
non-responders. This data is presented in Table 5.

Among men, responders were more likely to be older, thinner, and able to perform
on the treadmill longer than non-responders. Women responders were similarly able to
perform on the treadmill longer than non-responders. Both male and female responders
were more likely to exercise more and smoke less than non-responders. Responders
reported higher percentages of all family history of illness than non-responders (except
for women having a father having died of CHD, which was equal). The differences were
statistically significant for a family history of cardiovascular disease and stroke in both
men and women, and additionally for a family history of hypertension in males only.
However, the responders and non-responders did not differ statistically on personal
medical history characteristics or clinical measurements, except for male responders

having significantly lower uric acid levels than male non-responders.

20

Table 5: Comparison of baseline characteristics of non-responders and responders to a
mailed follow-up survey of 15,440 persons who attended a preventive medicine center in
Dallas, TX at least once from 1972-81 (adapted from Macera et al. [118]).

 

 

 

 

Men % Female %
Non- Non-
responders Responders responders Responders
(n=2,624) (n=9,409) (n=844) (n=2563)
Physical characteristics
Age“ 20 25* 22 26
Waist girthb 7 8 5 8
Body Mass Indexc 23 19* 20 18
Treadmill testd 36 27* 36 28*
Lifestyle behaviors
Current non-exerciser 50 37* 54 38*
Current smoker 23 19* 20 12*
Current or past smoker 50 51 44 39
Alcohol usage 2 14 drinks/wk 21 19 7 8
Egg consumption 2 4 eggs/wk 41 38 26 30
Fried food consumption 2 4 times/wk 24 23 10 7
Family medical history
Cardiovascular disease (CVD) 35 40" 35 42
Father died of CVD 19 21 19 19
Hypertension 30 34* 37 44
Obesity 31 34 35 40
Diabetes 20 21 24 27
Stroke 14 18* 15 22*
Personal medical history
Hypertension 16 16 10 1 l
Hyperlipidemia 6 6 4 3
Diabetes 4 4 2 2
Asthma 6 6 6 6
Chest pain 17 18 21 19
Thyroid problems 3 3 13 16
Clinical measurements
Total serum cholesterol > 220 ml 42 38 33 28
Serum glucose > 110 ml 16 15 9 6
Uric acid<3or>7 36 31* 7 6
Diastolic blood pressure 2 90 mm 14 12 6 5
Systolic blood pressure 2 140 mm 10 9 6 5

 

a Percent of men 2 50 years or women 2 49 years
5 Percent of men with waist girth 2 97 cm or women 2 75 cm
C Percent of men with BMI 2 27.7 or women 2 30.5 cm

d Percent of men with treadmill test s 780 sec or women s 495 sec
"' p<0.001

21

Macera et al. (118) thus concluded that responders tend to have better health
practices, such as regular exercise and non-smoking, and also tend to have a family
history of disease of cardiovascular disease, hypertension, and stroke. Therefore they
reasoned that using data to the mail survey to make estimates about the underlying
population of the clinic would introduce bias into their results, insofar as estimates of
prevalence of current non-exercise or smoking habits would be spuriously low and
estimates of prevalence of physical fitness and family history of cardiovascular disease,
hypertension, and stroke would be spuriously high.

These studies are just a few examples of studies conducted to assess the role of
non-response bias in the epidemiological and health sciences literature. A more complete

review of this literature is presented in Table 6.

22

 

flow—=33.
.5: .6 $8 3 9.3 an 55.3
398%». 28.0580: .0 $3 ”9.7.

 

 

 

 

mg.

a. <2 523m 5
6.56 33.2.50 <>
2.. we beam w:_m<
023562 2.. c.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

m=.§&o.§n has: >25.“ .83.. Go: 2.6.
.886va .\.~.mw $5.3 5.25 8.23.0 8a.. 33:8... 3?. $8 30.. 2s: 8....5 ._:..< .ueououu.umo._o r... .o 35.8
.955 .8 0:... a £532 5.3..
86".. «one. .\...... 8 use 6:95 Eaton. 8 03:5 .3. e. <0 cation
86“.. :3 :3 sea... 2.... a =25 .8. .2. 2o 5 23:5 5
x. .2 $3.. 5.8.3.. 9.5. 52.. Ba 23..
$.34. $0.3 56.25.52 noucamaéee .8 ca .59 £2.52:
.\... .2. «3.? toerm 3222:. .2522. 855:5...53. .333 .88“. 82. n3.
86“.. mesa 9.6.9.5 .5 2.2.3.; $.30 Ru... .0 0.95“ E85: 2288.320 :3 .u “830
.3. e. M: o...
a. .523. 5... 293
335882.»? 2.3 v... 53.nov— 9732.9.
:2 x: 2 N. 23.25 36: was»: as... 2.. 3.3.58. so»
:3.» a 6: $0 $5 flue—95.82 3332.? :e 295 3:3: 08...: o. 5.3 $35“ A... 3 So.
So. :. “58m 96.9.5 8: 2.8% $025 Sndw 89% 5525 can :02 15.33320 ...u .9 =8
l R2 5 39. N N
a. as... .= 693.5
been .6 22.9: m a 3 3.3.qu
.5 56. e 2.5.. 8.3.58 8.... 2-2. Em“ 5:. .2 : $2
cozw A. o: n... o... .2. 8.... >558:— 8:ao...u 12:? *3 No.6 .3 9.95m Eco—.5. r... .u .3...
:3.» a o: 0.0 fin 5:53 033....
cozw a e: Z: N.» :02 Nmévo. in“... 3 53933.50
€330 Mao.» m E... wet—6 -on coma 5.3232. .8 Ave... :28
8.32.2. 8.... .2. as 2.5.2. 2.2.... as: gauges. 2.62.8... .2 : one
8.5.... 8.3.3.6». .352 see a... £35: :33 as .356 6353 sunscefiaoa r... .6 8:60
cove 8. refinance 8... Reg
.5: 95.5 9.25 3.58.8.3 SE
3.91. on: so...» 3: Echo 83.59 mean 8.5.. me E 8:88: = 5222.2. beam 5.3.. 33m 353.

 

 

 

.2538: 823.8 5.3.. was 323.352....

 

o a... a. as... snag—85-5: he 8:5.» he Eafifiaw 6 05:.

23

 

 

 

 

 

 

8. en. 88 .n .5. a 3.2.. 28.5.0 .850
8. cl. 9. n ~. n >283 :88 ammo .oz
3...... «3. z .x... a 8.832.... 8 .x:
.851. «as «8.: genetic... .o a:
5.3;.
3. cu: 58.8 .\.v.- 8.25 8.938 .850
R. an: {.56 *3. 3383 :88 an 62
35".. .\.v.n~ .\.w.n~ 5.8.8.3.... .5 .x...
8...". x. .2 .5: ..E»........&.. .o a:
to}.
93.05“ 3... ~36qu&
8.1. 8...: 8...: 8.3.... as“... 2.5....
88...“... .58 $2; 3.2.. as... 3.5...
cum on: 5 g 83
:1. 5a.“. $5.». .8» 5:83.. 85.5.... ..
.853. 8.2 8...... .8... 52. .5. 5:2...
28.5.:
2...". $2. $2. 88...... as... 2......
2.1. $2.” .52 3.2.. as... 2:5.
3% on: a g 8.3
NET... $0.2 X. .m .8» 5.82... 25.35 .=
58...". .33 $2.. .8... 52. .5. 5:5.
2...“. x... .\.m.. 3.2.. .c 592.:
28...". .2... .8... 33¢... 8 bag:
n. .cu. .\._ .v .33 .8... 52. a. :8:
.831. .\.. .m at... 95...... «89.. 5.. .95: 8:8...
28.5: 55830.28
5. cu. $2 $3 3.28... .85: .o 53.
.m. cl. «8... .\....m «833...... >53... 5.. $6.55
.21. .8... :3 .8... 52. a. :8: «23. «Zn .8». 185..
.83". $3 .\.. .. 22.... 52. a. :8: 5288. «.5023 .558 a.» 8:...
.3: .8: 5.: its... 5 89:35.2. 23.88.82» 3. = «a.
88w... 8 as... 2.2.33 s. .2 8 . .o 5.5.5 68.98.33... .._. .o .55
835%». 88.558. 83 3.88.. . .
.5: 95...: 52:: 5.58.8»: 8n.
3...... 3.. no... a; .8...» 8.3.5 52m #135. .a 38: 88.3.. .. @338. .35 5.8.. .38 3&3.

 

 

 

 

 

 

 

 

 

6.5.9.8.. 82.0.3. 5.3.. .3: .ao.uc.c.Eu—....o 2.. a. a... 3:38..-.5: .5 8......» .5 baa—saw Ace—.5350. e 2......

 

24

 

 

 

 

m2 $2.. 32.. 23>
m2 #2.. *8... n 39>
m2 $3.. $3... v 30>
8.9.. $8.. :8... n 5.,
mcdvq $3.. 3%.: N 89>
3.9.. $2.... $2.... . 5..
Qaéczubw 3qu c .853»...
3... as $88... 3.3.... .35... 532...... 84.3. J5 2.. c.
3-32.8 .8 a3» NS 58:. v5 ransom 232 «N 9.25 .99.
m2 x. .n :3 a. $.53... assaésoasafiu a. 3%.... .3 BEE... £8. 5:. =. 3.3...
.35.... .o as. N... .53 £2 8.... you“ 5... 3895.28 .o as. S...
8.1. $1.... x...“ 52.. a. $.53... 889...< .532 539 $13 2...... 8.8.3 258.5. .3... 3.63.8. r... .o 3.51
.\.. s :2: N N
.303. $92” _
$1? $3.“ a
.2.an 3:998 258.:— uo .3552
.\.. .NN gag 8>
$5.: $5.? oz
cows—3:8
mmodnn _a=c_mmo..o=_ 23:9. b.503—
.\.N.: XS: N N
$2». .\.o. . n .
$2; .\.N.Nm o
o _ .oAn 22.6.80 2:25 .550 .3 .oz
$N. _ n $5.3 N N
$5.3 Q; .wN _
.\._ .mv .\.N.nN o
3.83 N .2... =_ coo—8:35
58.9.. “axis? .332; .3 62
.\.¢. _ n XE. 3. 3qu
:2.“ x... . m 2......
:66 $06 883 gonad. Na SN:
52. .2... .o 30.5.... .8. s as. 2.3 .25.. .38.. a... .....
58.qu 2 vanes .23.. commommaéom 2.293.. $5.0m am}: van“ 5803 .32 18.88.82”. .0 afigm
38:83. c858. . 4.: a... 3:88. 4 . , . , .
.5: 96.5. 9.2... . : €53.03.- ofi. . ..
3.2... 8.3%.» 8.30m.» 82...? 6.... 83.... .o 85...... .283. __ 8.3.82. .3... 5.8.. as.” 5......

 

 

 

 

 

 

 

 

 

 

 

6.5.9.8: 82.39» 5.3.— 2:. 39323892..» 2.. a. an... 032—3..-.5: .3 8:53 a: b.5555 “£555..ch e Saab

25

 

 

 

 

 

 

 

 

 

 

 

 

 

 

«.2 $5 $.. 5&9:
m2 $.~ $2 5:.
933...... 3 N 09.3 .28....
8.9.. $2 $2 5:33
8.3.. $2 $2 8:
hog—95m 30.5-U
8.3.. $3 $$ 5:53
8.3.. $8 $2 5:
hoflouoxoég «cot—.0
8.9.. $8 $3 8” 3.. v. 8:53
8...”... $3 $2. 8» 2: w 5:.
.u._.=8.a EN. .8. 558:
3.9.. $8 $2 a... 2: ~ 8.83
8...”... $2 $2 EN N 82 .38. :5... 88 5.... .5; can”
.— . . . .
m2 .\ n . 3.9.8.2. at“... 48 .Wfléom .80. a 5 3.5 2...... were...
.. 9.. m an N 3 a. 528 83...»... 5.3 .93.
m2 .x.» {on Eu 3 N :02 83 32:38.. - eon—.83 5 meta .88.. 8. 3 32
.o__.=8§. EN. 5.... E; a: 0.5.9 E... $2... 83. 2.3 2.2%.. 35.83.35 ..... .o 58.2
.89.. $3 $3 25.... “a.” .5
8..."... $3. $3. .52.... 9.252
.89.. $2.. $3 9....9.8 ”3.52
.89.. $~.R $3.. 9.98 9.8
.333:
m2 $2 $3 26.... 9..» 5 2.828 8.3...
m2 $3. $2. .52.... 9....3: 8&3. a. .282 52. 98.8
3.3.. $3. $3. 99.9.8 9.232 38:8 .8529 c. 9...... a 328... 3. 5..
.83.. $...% $. .3. .395 9.5 82. 9.8... 3.. 3.8 B». 5.83 9.3 233%... 9.... ....
5: .2. 5% 8:35 $2... 5.... 2a 3.8 .39. 52 .o 9:3 .58.. .o 8308.
92.88. £8.89». 83 8:88.. 9 9 , ,
-3: wee—s. 9.95 3.53.88 8,! ,
3.2,... BMW...» 8:389 8.3;...5892 .o .55»: . 3.2.3 .. 9.33.2.8 .35 =9»... can 5......

 

 

 

.2538: 82.32. 5.3.— 2:. 783285023 05 5 33 0332-52. .3 8:53 .3 ran—55m “€025..ch e 035—.

26

 

 

 

 

 

 

 

 

 

 

 

 

m2 0 m E... 3. N mm 3.293
m2 0 n E... co N .5 3.29.5
mz h e 5A.omv2u¢ or:
.2 o o .5 2. A 382. .55...
m2 mm mu .8 can A .9... E28 .8...
22.3.: .......o.:o.a.do.: .3250
m2 o. o .5: 3. N .5 3.8.3
m2 v. N. 5.: 8 N .m 3.835
3.9. on a . A a m v 28 2.:
m2 2 m. .5 2. A882. seam
wz 2. an .5 :9 A .9... 528 do...
to: .....=o.:§o£ 28.5.0
.2 .N o. 8......
mz o o 8:53.
m2 N N 8.215
m2 v m «£023.59»:
m2 c. . . 3.83.25.—
..oES: “.02.”... Sofia: .gfiom
mz : 2 8......
mz o e 8:52
m2 v v 8.3.5
.2 e e .58_._:§:
m2 2 2 8.3.5....
to}. .52»... 28%»... 33.5%
adv. 2 a 3.2..
m2 vm R 8.3.5
m2 2 3 383
m2 2“ 3. 8.2.8.2....
3?. 2 S 8.8... 5888.30
:35: ..o3o...h\o .66.»... 3.33%
3.9.. 2 a. 3.2..
m2 3 .N 8.2.29
.2 a 3. £88
86v: on cm 3.388%... 63:250.
36v: mm 3 8:8... 3.335.an G. .. coo.
to} 838......6 be”... bemé ...: .o 88::
96:88. 5.52.8. as... 8:88. i .
-3: use: :95 €23.83 oi.
8.5.... 8.. .oobo 3.... Soto 3.3.5 in Bis. .0 .658... 8:88.. : . 8.32.2. gm 5.8.. in 853.

 

 

 

6.5.28.— 325.3 5.3.. 6:: .ao.»£2.:o.=..o o... a. as... 8:38..-.5: .o 3......» a: b.2555 .Aeoaazacuv e o......—.

27

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

8-3.2 .23. S...
cows .5..on a. 32..
£25.23 “.53.
28:83. 285.3: 6.3... :3:
.5: .0 30.28:. 53.58 a. wet... mots... .38.. Moo. :3
.89.. x93 .\.. .2 .26.... 2.5 2.9.33 x2: '3 35.3... ..._..< 3.263.320 .u 52.3:
8.1.. _. 2 e2. 3 9.8 s 2... 2.23.52. .558: 8.5.. 9732....
mat—6 28.3.53
88%. b.3833. .833: :5...
9.2... 3 :2. o . 9 2... 2.23.53 .333: .232 8.2.2.5
m2 .2 .v ,5... oi. :3:
m2 .22 .8... Sm. 53:
mz $3 $2. m.....2.e.. 2.5.5 3%
m2 .\._.2 3...: 8.9. .o m an.» 8.5:: :48. _ _ .c :28
m2 .x. . . N :12 .52.... 38.. £258 5 8.3 2.32.8...
mz $5.: $02 5.50 :28 2.. .c 5.... .59 £9.83 22:8 ..2.........8 s. .2: «8.
m2 e\ew.Nm .3— .cv uoonm >>ao= 05—039 59¢ Sun— Xfiw «6v.— u—GE .«O 2ng MO hut—am Emom :3 go 333/
. . 3&3. saws... nose .2 2...». .2: 32......
m2 8.. R. ... 3o... .e .85.. 8.8 .o x... .3.
.33“. .. 2.2.3.3 2325.8 .2.
m2 m3 2.... 23.9.; a. .85.. “.5. .o .2. .3.
Esau; c. 9.3.8 3o> 958:
m2 $3 5.2 525:...u a 22......ch 5...... hi .8:
8...". o... n. 358... 8.: .....
.8". m... o... .58.. .o 25.5:
Noon. 2 n. 8.8 a... .2:
8...". w. I .353 So. ....
2...... n. a. 9.3 .0 52...»:
8.1. 2 N... 038...... .38 5.8: 25m 2 933......
8.3.2.2. 8.... 8.... «385:8 so. 5 :26: 3-32....
.5..— 0E hide—OE 30:8 3:5 5&8 no 3030.— .anno co 9:23.. .30» m. .5230 AvN:
coma—anon coo; £> .8=-_=o>.=_m .0780- Eot ho beam tozoo _8_ .._u
.899. .3. 3. .x. a... $.85... 02.8.... .525. 2.35: $0.: 2.... ES 5... 035...... 2.62.8... .0 55:9:
now—8&6. ”Encamp— 35 852—8.
.5: 9.25 9.2:- msséoon- 2.:
31>-.. 3.“ 88% on.» Soho mu_aata> 3.2.0; 8?: .8 3502 8:83: = cgsanon Exam 5.8—. 33m 55.3.

 

6.59.8: 82.38 5.3.— 6:: _3_uo_o_EuE._o o... 5 mm. onscnflhuuc: no 8:5.» He ban—.55 Hagan—330V e 055—.

 

28

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

8...-36. 8.. 5.. 82.8 3.00
Gm. 3...... 8.. on... .328 3.8585
.8..-2.:$.. 2.. .858 25..
5.78.... 8.. 3....” .85... .82..
G. . 78. .. 8.. 3.... .856 o..m-..<
.3. En. _. 3.. a... 2.3.2.. 23.5.... 3.58.:
.8. To... .. an. a: 3.8.2.. 8.39.2
33.2. 5. to. engage: 8....
.2. 8.... 3333.0? 3232... «>6. .3 him»... 53.. .6 .859
”32:38.. #0. . N X. .n. 3.3 N 5.3: 88m .3 8:95.03
-5: :3. .\.N. . N $0.2 3.n~.¢sé~ Ea... 8.6 3:56... 8.. 85...».5...
gain... NEE. .\.m.o~ wainébflm can .350 new 03. .. .935... 9152.8
2. .e mafia... .6 .\.o. .~ 3.3-2.3 5:25.... 828.. .5 so» o9? .3» n 5.3
82.. turgmnux x: .8 e. ..N v. countamcflh com: 22:93 «>6. .3 .323. .88.. :2. 3o.
3......3 55.9.. 5.2.. 2.: 25m 2. .5535 2...? 3...... 29:3 585. 2.8 38.68.an r... .0 .5me
m2 2.. .2 5.20223.
.89.. .2... 22.... 52. 3.32.50
22...... .2an $530.36....
.2 e2.2.. 253252.. 332:2. 22
Edam
4.8.0.8252. 8 288.3.
.o.EoEo.m 5539...»: Snow... 3:: .8 35.2.. 2.3.8.358
mz $92 3.83722... .5. .o 8:88.. 8528. ..a 8.32... 8.. 38.8 3.2.2...
5?. 9.23 .2. 33m... 8.. 032...: 803 958... 3528.. ..o .8...“ :28 8m : moo.
.8.ov.. 35% $62 3.38.72an .5 .o 8:82.. 2:82 .863. $5... 2... 25.2 582.03. ..< 3.82.8... ....w .o 22.02
mz $3.. $.30 .8» a... £53 2.258.. .859
m2 23.2 8%. 23 .35.. .o 32.8 3.22:.
.2253:
589.. 2&2 .\.m. . m .58.. a. .32. 3282.23
$5... 2x...
22.2.. $~d~ as".
$3... $36 300
.\.~...~ 2%: 22.82.". 9:8 2.2.9.2
8.9.. 5.8.. .20 2. «2.2.23. a 8.2.28
58an .\.. KN .\.n.mn 8.33 v 32>»... a. :2... .EO aotam 2.9.9.0. .33.?
m2 2.92 2.22 ”232.0 52.2.. 2.. 222... 302.8 go 6.5.5 2.32. 2.9.5..
86v; .33.. .\..~.- «3.3 a. 26.3.8... 5.x. .32... 3.. 5 N5.... .020 v5 ..o 33.5.... an:
.89.. .\.. Ne $33 6:8 do... 0.52.0 29: .o 9:0 a... 3...v 05.8.5 <53 «.8... .59. cm cows 9.23. .38.“... m8. 28.8...
«.0388. .5388. . . . a... 022.8. m H .. . .
2.9. $.25 98:... H ”2.2.8.83 .82 . . . . _ .
3.2,... 3.202% 3.82% 3.3.25 .3...st .2. 3.5:. 8:23.. ._ 8.2.2.8...5 5.8.. .3... .25.

 

 

 

.2528.— 826...» 5.3.. E... .33228923 2.. a. ma... 02.2—82.5.— .3 8:5... .9 ESE—=5 ”Ace—.2350. e 035—.

29

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

adv. x: .x... .203... a .38.. .2
.89.. .x... «S. gasesfisea...
m2 :2 .x... 8.205....
50?. $3. :2 395.550
£80....
3.. 3.336.080 00:000....20m
8.9.. x... x... .8. .3. 03.338: 8.0.... .8:
X... .\.N .00.. .3 00.0.3.3 .00....000 >08...
$3 $00 3.3.000 dd .00.. .0 0:0 .0 .038 30.0.3 0....
$3 «8m 30:09.... 00.03.00 5:00.00. 2.0.3.500 30.30.... 0.00..
.89.. 5.8.. .308. 8:83. 83 .32.. 2.. .3... B... 9...... 2208.220 .3... 08.
no.2. 3...; 29.2 a... a... SE0 :8 at». .o 2.58 58.5. 833232.... e... a 83.8.
:2.» a 2. x... 8.... .38
00>.» .. 0.. gm... .\.m.n 0.0350... x: £030.35: ...
00>.» .. 0.. $3. $00 0.03:. 50.9.... 9.0.000. 0.00.80... 2.03.. 03. .253 330.. Ann:
:3.— ..03 30. .0. 03. 0.80.30... 5.3 00.0330. 0.00.. 300.68.00.20 m8. :3 .0
00.3.3800 0030.00-09. 3......< 03.850800 $3 000... 3.x. 002. 8.00.... 000.095.3300.. 00.».00waaam
.vo.~-vo.8 .mv.~-m . . :
00>.» 0 0.. 3.. 2... 0.00.3 0.52.8. E0352...
afinén... 33.0%....
00>.» 0 0.. EN m..~ 8.85
5.13.3 5.78.3
S... a 8 a... a... 8.03.... 3.58.2
8m.~-vw.0. .$.~-~m. :
00>.» .. 0.. .6. «0.. .0.m0.0.=-..05 5.30333.
3.2.73.8 .2808...
00>.w 0 0.. vm.» 3; «BER: 8.0.35
.8...-.....c .363...
00>.» .. 0.. new mm... 0.00.5.2... 80.80....—
Gede. C Gm.0.0~.$
00>.» 0 0.. .0.~ and 00.80.... .08. .93....
2.05.2. 3 $00.00....
00>.» a 0.. 0.3 and 8:03 3.0.8.5003. 8.30. .22 5.500
33.38 .330: 8.2.5 62...;
00>.m .. 0.. m0. 3.. 0.588.... 00.9. 02.00300» 30.30....
$00.20 .8. is... as. on. .23. a2... 28.. 5.3
=0...» 0 0.. 36 0.020 3005.09... 03800... 5.00.. n.00..0..8....0.. 0»: 3:00.00. 03... >039. 3.8.. RM: 30.
«0.... 00:03.6... 00.3..00.0w< .0 .03.... 330.. {can 2...... .0 0.9.3.. 500...... 300.88.00.05 :3 .0 .85..
£00038. 0.0000000. . . 8... 80008.. . . + . .. . .
.00.. 9.0.5 9.0.00 ., «0.0.0003... 03.. ,.
3.2-.. a... g a... .8...» 83...; .3... 3...: .o 2%»: 38.8w , _. , $3.32. 32m 5.8.. .35 ‘ 353.
0.0.0.0... 00000.0» 5.00.. E... Egg—0.0.0.5.... 0... a. an... 032.00.700.— ..0 00:53 .0 b.2555 2000:3000. e 030,—.

30

The results of Table 6 display some remarkably consistent results.
Overwhelmingly, non-responders compared to responders tended to be smokers (106,
116-118, 120, 123, 127, 128, 134), with increased cough and phlegm production (123)
and requiring more frequent hospitalization for respiratory disease and COPD (127).
Most studies that examined differences in mortality between responders and non-
responders found a higher all-cause mortality among the non-responders (115, 122, 124,
131), with increased mortality due to myocardial infarction and all cancers (131),
although one study (119) did not find such an association.

More conflicting results arose when looking at chronic disease. Some authors
found an increased prevalence of congestive heart failure, diabetes mellitus,
hyperlipidemia, and/or previous myocardial infarction and stroke among non-responders
compared to responders (116, 130, 134). Other authors, however, found no increased
prevalence for such conditions (118, 132).

Criqui et al. (116) discussed the “worried-well” phenomena, as discussed
previously, in which non-responders tended to have lower prevalence of disease but had
higher prevalence of risk factors for disease. Macera et al. (118), however, found
somewhat divergent results. Although the authors did find an increased prevalence of
risk factors for disease (increased family history of cardiovascular disease, hypertension,
and stroke), much like Criqui et al., they found that in their study population non-
responders were not as healthy as responders, as evidenced by an increased BMI and
decreased exercise tolerance. Bisgard et al. (131) also found that non-responders tended

to have higher BMI when compared to responders. Furthermore, other authors have

31

determined that non-responders tended to self-report poorer general health (121, 134) and
required more frequent hospitalizations than responders (134).

Interest in the study subject also tended to attract responders. Persons who did not
feel well at the time of survey or were experiencing problems related to the study topic
were more likely to respond (120, 129, 133), a finding consistent with previous literature
(102).

The data presented by the above studies seem to suggest that in general, non-
responders can and do differ from responders with respect to certain morbidity and
mortality endpoints. However, the data are far from consistent. Differences in study
design, study population, and study topic likely play a sizable role in the discrepancies
encountered within the literature. Furthermore, data from non-responder studies are
presently limited, and no study to date has examined the role of non-response bias in

reproductive epidemiology studies.

32

CHAPTER 3

METHODS

Population and setting

The design of the Fisheaters Family Health Project (FFHP) has been described
(67). Briefly, between 1993 and 1995 a questionnaire was mailed to 4,931 licensed
anglers between 17 and 34 years of age and resident in one of 10 Michigan counties
bordering a Great Lake (Figure 2). These counties were selected because of the high
PCB burden in the sport fish of the surrounding waters due to localized industrial
contamination, thereby maximizing the PCB exposure among the study population.
Usable responses were received from 1,445 anglers, who form the study base for the
results reported previously (13,67). Eligible participants for this study of non-response
consisted of the 2,937 individuals (2,371 men, 566 women) who did not respond to the
original FFHP screening survey and whose original survey had not been returned as
undeliverable by the U.S. Postal Service.

Due to the potential for differences in responses between men and women, the
population was first stratified on gender. In addition, because individuals from different
parts of the State are quite diverse and distinct from one another in terms of demographic,
socio-economic, and behavioral characteristics, the population was stratified based on
geographic region as well. Region 1 includes the counties of West Michigan: Allegan,
Muskegon, and Ottawa counties. Region 2 includes the Bay Area counties: Bay,
Midland, and Saginaw counties. Region 3 includes counties of Southeast Michigan:

Macomb, Monroe, St. Claire, and Wayne counties.

33

 

Lake .S'upen‘or

11" w
1... J” I.“

 

 

Lake Huron

 

 

Lea/'21

 

 

 

 

 

 

Counties

l. Muskegon Lake
2. Ottawa M 0,1180” _ ,;:§:§: gig
3.A11egm 5 i
4. Bay 1 5" :
5. MIdlmd "'ozsé 323331233:
55,193,“? E] ”'2: ' f=;§5; iii:
73mm: 3 W?
8. Macomb ' I I iiifiés” 8
9. Wayne .4; y-

10. Monroe " Lake 3*“

 

 

 

 

:gzg' 333333
0..
........

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 2: Target counties for the Fisheaters Family Health Project, Michigan.

Based upon the population size in each region by gender cell, initial estimates
called for a randomly selected sample of 40 men and 40 women from each region, for a
total of 240 participants. These numbers were determined based on power calculations
performed in Epi Info 6.02 (135). This sample size would provide adequate power to
detect, at the 95% confidence level, a 10% difference in proportions between responders
and non-responders within each region (n=80) as well as within each gender separately
(n=120). Given the much larger numbers of participants that would be required to survey
in order to increase our power, and considering the financial and time constraints of this
project, it was felt to be impractical to attempt to detect a smaller difference or to attempt

to detect within-region differences among men and women separately. For example, in

34

order to detect a 9% difference in proportions at the 95% confidence level, at least 100
participants per region would have been required, and in order to detect a 5% difference a

sample in excess of 240 participants per region would be needed.

Sampling

Names and addresses of non-responders were imported into a Microsoft Excel
(136) spreadsheet. After being sorted into their respective gender and region strata, non-
responders were assigned random numbers using the RAND function and sorted by this
number. Non-responders were then selected sequentially from each stratum for
recruitment into this study until either the target sample size for each stratum was reached

or until the population of non-responders for that stratum was exhausted.

Identification of telephone numbers

The original database of angler license applications from the Michigan
Department of Natural Resources was previously obtained. This database, however, did
not provide telephone numbers. Since the names and addresses of licensed anglers listed
in the database were obtained in 1992 and 1993, at the start of the original FFHP survey,
it was felt that a telephone number for the angler would best be located by using a
telephone database from those years rather than using a database current at the time of
this non-response study. It was expected that many would have changed residences
during this intervening time interval, given the young age and relative mobility of the
study population. Furthermore, many women might have married and hence changed

their family names. Use of a 1992 database would, at a minimum, provide a telephone

35

number which could be used to initiate a tracing of the individual, whereas a 1996
database might have no number at all listed for that individual. A 1992 CD-ROM
database of listed telephone numbers (137) was used to obtain anglers’ telephone
numbers.

Some licensed anglers would not be expected to be listed in the telephone
directory, because they may have been children, spouses, partners, or friends of the
telephone subscriber. In addition, because people often move within a city, it was
thought that anglers who moved between 1991 and 1992, and who consequently had
outdated listings and might not match on address, might still live in that city with the
same telephone number. Therefore anglers were matched to a listing if they matched in
either of the following ways: last name and street address; or first and last name and city,
with no other matching name in that city. For example, an angler with a unique name in
a city would be matched to a unique listing of that name in that city, regardless of street
address. However, there was no way to match a common name in a community with a

particular listing without the exact street address.

Interview

The instrument used for this telephone survey was based on the original FFHP
screening survey. In order to increase participant compliance and response rates, the
survey was adapted so that it could be administered by telephone in a five-minute period.
Key questions regarding reproductive history and future productive plans, Great Lakes

sport fish consumption and fishing habits, and demographic, socio-economic, and

36

behavioral characteristics were included. A copy of the survey instrument used is
included in Appendix A.

The administration of the telephone survey took place between November 1995
and April 1996. A protocol on how to handle situations such as encountering wrong
numbers, busy signals, answering machines, and no answers was developed, based on a
protocol developed for the Behavioral Risk Factor Surveillance System (138). In such
situations, effort was made to reach the angler on five separate occasions during different
periods of the day (morning, aftemoon, or evening) and on both weekdays and weekends
before replacement. This protocol was established in order to minimize non-contact with
those individuals who are employed during non-traditional hours. In general, however,
most attempts at reaching eligible anglers occurred weeknights between the hours of 6:00
and 9:00 PM. All attempts were logged for proper book-keeping and tracking. The
complete protocol is included in Appendix B.

At the beginning of each telephone call, identification was made by the
interviewer and the purpose of the call explained. The subjects were informed that their
responses would remain confidential. A copy of the script used is included in Appendix
C. If the subject agreed to participate, responses to the interview were recorded on a

paper form with only the identification number attached in order to ensure confidentiality.

Data entry
Data were entered into a Microsofi Access (136) database concurrently to its
collection and subsequently exported to SPSS for Windows 6.1.3 (139) for analysis.

Participants were identified in the database only by their identification number. This

37

database was separate and distinct from the database matching identification number to
study participant. Both databases were password-protected. These measures were all

undertaken to ensure confidentiality.

Statistical methods

Initial analyses examined the distribution of all variables between non-responders

and responders as a whole using the Pearson )8 test for dichotomous variables and
independent samples t-test for continuous variables. Stratified analyses were also
performed in order to examine the distribution of variables within each gender group and
among the three regions for men.

Previous research by Courval et al. (13) suggested a modest association, in men
only, of Great Lakes sport fish consumption with risk of conception failure, defined as
ever having failed to conceive after 12 months of trying. An attempt to replicate these
results by performing logistic regression analyses on both the responder and the non-
responder population was performed, recognizing that this study was under-powered to
detect modest differences between the two groups. For these analyses, Great Lakes sport
fish consumption was defined as the number of average—sized meals of fish caught from
the Great Lakes or its tributary system in the past year by the angler himself or by
someone that angler knew personally. Sport fish consumption was categorized first as a

dichotomous variable (none vs. any), and second as a categorical variable with four levels

(none, 1-12 meals, 13-24 meals, or 2 25 meals per year). In order to increase

comparability, analyses were restricted to currently-married and ever-married males. It

38

was felt that these individuals were most likely to have attempted to conceive a child and
thus most likely to have experienced conception delay. As in the main study, both
unadjusted analyses and analyses adjusted for age, race, geographic region of residence,
income, education, tobacco use, and alcohol consumption were performed. Individuals
with missing covariate data were included in the adjusted models, with a ‘missing’ code
applied in such cases, in order to maximize our sample sizes. All data analyses were

performed with SPSS for Windows 6.1.3 (139).

39

CHAPTER 4

RESULTS

The final disposition of all eligible study participants is provided in Table 7.
Target cell sizes for men in Region 3 (Southeast Michigan) or for women in any region
were not reached. The primary reason for failure to conduct an interview was the
inability to identify a telephone number in our CD-ROM telephone database for the
angler. This was particularly true in the regions in which target cell sizes were not
reached, where 78% of the anglers were not listed, compared to only 56% of the men in
Regions 1 and 2. Overall, 72% were not listed. Significantly, an additional 10% of the
anglers had telephone numbers that were incorrect or disconnected at the time of this
survey. Smaller percentages of anglers did not retum phone messages left with family
members or on the answering machine, and there were a few anglers who did not answer
the phone on successive attempts, were deaf and could not communicate, or had passed
away. Of the 364 anglers reached, 230 men and 38 women completed the telephone
interview, for a response rate of 74%. A complete tracking flowchart is shown in Figure
3. Because of the inability to reach target cell sizes for women, the three regions of

women were combined into one for further analyses.

40

Table 7: Participation status in a survey of Michigan licensed anglers, by gender and
region.

 

 

 

Males
Region 10 Region 20 Region 30 Females Total

Eligible for non-response study 543 597 1,207 566 2,937
No contact attempted 68 52 0 0 120
No telephone number listed 261 312 945 439 2,030
Listed telephone number incorrect 54 48 68 46 216
Listed telephone number 10 13 33 8 64
disconnected

Did not return phone calls 17 26 47 9 99
Answering machine 5x 3 8 6 5 22
Ring no answer 5x 7 2 8 2 19
Unable to communicate (deal) 0 l 0 0 1
Deceased 0 0 2 0 2
Refused Interview 26 20 3 l 19 96
Completed Interview 82 83 65 38 268

 

a Region 1: Allegan, Muskegon, and Ottawa counties.
5 Region 2: Bay, Midland, and Saginaw counties.
C Region 3: Macomb, Monroe, St. Claire, and Wayne counties.

41

 

 

4,931 licensed Michigan anglers surveyed in 1993-95

1,445 Responders
505 Undelivered
44 Refused

.7

2,937 Non-responders

120 Not selected/
sampling frame
quota reached

2,030 No phone number

identified in 1995—96

 

‘ f

l

787 Non-responders with telephone number

280 Disconnected or
incorrect number
143 No contact

7

364 Non-responders contacted in 1995-96

96 Refused

7

268 Non-responders completed telephone survey

 

Figure 3: Tracking flowchart of study participants.

42

 

Demographic characteristics of men

Demographic, socio-economic, and behavioral characteristics among male
responders and non-responders are presented in Table 8. Region-specific analyses of
these characteristics for men are presented in Table 9. Although the largest percentage
(42%) of male responders resided in Region 3, male non-responders from the same region
comprise the smallest percentage (28%) of non-responders. This is consistent with our
inability to contact eligible participants from and to reach our target cell size for this
region.

On average, male non-responders were approximately 1.5 years older at interview
than were responders, 30.9 years vs. 29.5 years. This, however, may reflect the time
interval between the two studies. The median year of birth for male non-responders was
actually one year later than that for responders, suggesting that, if interviewed at the same
time, male non-responders on average would have in fact been younger than responders.
These trends were consistent within all three regions. Interestingly, male non-responders
tended to be both older and younger than their responder counterparts: higher percentages
of male non-responders were found at ages 21-26 years and at ages 35 years and older,
with lower percentages in the age groups in between. There were virtually no male non-
responders 20 years of age or younger, a finding consistent with the fact that non-
responders were interviewed at a significantly later date.

Male non-responders were more likely to be Caucasian than responders. Almost
99% of male non-responders were Caucasian, compared to 92% of responders. This
trend was consistent within all three regions, and particularly for Region 3, in which this

difference reached significance. In Region 3, no non-Caucasian males were recruited for

43

our non-responder population, despite the fact that this region contained the highest
percentage of non-Caucasians in our responder population.

No significant differences of marital status were found between male non-
responders and responders. Overall, approximately 50-53% of both groups reported
being currently married, 40-43% of both groups reported having never married, and about
7% of both groups reported being divorced, separated, or widowed. Significantly, a
higher percentage of male responders residing in Region 1 (West Michigan) reported
being currently married compared to non-responders of the same region as well as
responders from the other regions.

Non-responders reported higher incomes than responders. Almost 90% of non-
responders reported annual incomes of $20,000 or more, and 40% reported annual
incomes in excess of $40,000. This compares with approximately 75% and 32% of
responders, respectively. These trends were consistent within all three regions.

With respect to education, similar percentages of male non-responders and
responders had earned a high school degree or less. Of those who attended college, male
non-responders were slightly more likely than responders to have obtained a college
degree. Twenty-two percent of male non-responders compared to 18% of responders had
attained a college degree, although this finding was not statistically significance. These

trends were similar among the three regions.

44

Table 8: Demographic, socio-economic, and behavioral characteristics among male non-
responders and responders to a survey of Michigan licensed anglers.

 

Non-responders (n=230) Responders (n=1, 129) pa

 

95% Clb 95% Clb
Region of Residence <0.001
West Michigan % 35.7 (29.6, 42.3) 28.3 (25.7, 31.0)
Bay Area % 36.1 (30.0, 42.7) 29.7 (27.1, 32.5)
Southeast Michigan % 28.3 (22.7, 34.7) 42.1 (39.2, 45.1)
Year of birth (median) year 1965 1964
Age at time of survey (mean) years 30.9 (30.0, 31.9) 29.5 (29.1, 29.8) <0.01“
Age at time of survey <0.001
320 years % 0.4 (0.0, 2.7) 5.6 (4.4, 7.2)
21-26 years % 32.7 (26.8, 39.2) 22.4 (20.0, 25.0)
27-32 years % 21.7 (16.7, 27.7) 42.2 (39.3, 45.2)
33-34 years % 11.9 (8.2, 17.0) 18.8 (16.6, 21.2)
235 years % 33.2 (27.2, 39.7) 11.0 (9.3, 13.0)
Ethnic distribution <0.001
Caucasian % 98.7 (95.9, 99.7) 91.9 (90.1, 93.4)
Other % 1.3 (0.3, 4.1) 8.1 (6.6, 9.9)
Marital status 0.66
Never married % 43.2 (36.8, 49.9) 40.0 (37.1, 42.9)
Married % 49.8 (43.2, 56.4) 53.0 (50.0, 55.9)
Divorced, separated, widowed % 7.0 (4.2, 11.3) 7.1 (5.7, 8.8)
Annual income <0.001
<20,000 % 10.4 (6.9, 15.3) 24.9 (22.4, 27.6)
$20,000 - $39,999 % 49.6 (43.0, 56.2) 43.2 (40.3, 46.2)
>= $40,000 % 40.1 (33.8, 46.8) 31.9 (29.2, 34.7)
Highest education level 0.09
High school degree or less % 47.1 (40.5, 53.8) 44.2 (41.3, 47.2)
Some college, no degree % 30.8 (25.0, 37.3) 38.0 (35.2, 40.9)
College degree or higher % 22.0 (16.9, 28.0) 17.8 (15.6, 20.2)
Tobacco use 0.10
Current smoker % 31.1 (25.3, 37.6) 38.6 (35.8, 41.5)
Prior smoker °/o 20.2 (15.3, 26.1) 18.8 (16.6, 21.2)
Non-smoker % 48.7 (42.1, 55.3) 42.6 (39.7, 45.6)
Alcohol use in past year 0.001
None % 16.4 (12.0, 22.0) 9.1 (7.5, 11.0)
< 1 drink/wk % 38.7 (32.4, 45.4) 48.2 (45.3, 51.2)
2 1 drink/wk % 44.9 (38.4, 51.6) 42.7 (39.8, 45.7)

 

a Significance of chi-square test for differences in the distribution of the characteristic between response

groups, unless otherwise noted.

5 95% confidence interval.

0 Significance of independent-samples t-test for the difference in the mean age at time of survey between

response groups.

45

 

 

 

 

 

3.. ....N n... ...NN 4.... n: .x. .23.. .o 8.»... ””260
9... EN 4.... in man NNN .\. 8.»... 2. $2.8 2.8
9... N.N.. ..N.. 2.. m...“ can .\. a». .c 8.»... .8... .3:
... ... 2.... mm... .2... 5.32.... .92.»...
”.2 .2... EN 3.. .2... 2.. .x. 8.....3 N
2.. 2... 4...... ..Nn 2... S... a 8......» - 25.8.
...N 3 ....m .4... a...N 2 .\. 8...on v
86 5... 86 2:8... .353.
o... E. v... n... ..... S. .\. 838.3 628...... €8.25
2.. 3.. 3... can “.3 4.8 .x. 8.5.2
2... 3.. 3.... m. . .. ....N N... .\. 8.5... .302
3... 2.... 8... 3.8... .852
a... o... a... ...N on N.. .x. .25
Na. ........ N? 93 S... 3... .x. 5.888
S... ...... . .... 8.25.5... 2......—
mN. 2.... no «N N... 2.... .\. a... 2 N
.....N N. ...... 3.. N...N a... .\. as... 3.-....
2.. SN 3... ...NN 2. N... .x. as. N3N
....N NNN a...N .2. “N ..N... .x. as... eN-.N
a. o... 3 o... .1. N.. .\. a8. ..N w
.oo.ov Sodv 599 33:35 «a 03.
.NN... 2.. N. . m .8... 3N 4...... .3... RN N. . m as. .52... .223... 2.... .a on...
So. .8. .6... S... E... 8... 228.... 5.... .o 23>
at"... .31.. 62...... . ,. GT... . $3.13. . .Nar... ,
ha 38:88..— aovconmo. V.— mnvaoaflm 235%.... ha «no—vacancy. Quechua.
.coz , , I .52. l: -52
u... 8.»... 88.8.... . , , 1.5.»...— . g

 

dame».— o... Ewen»

 

 

.3 rho—1w...» 10.3.3: swim—.3:

be rota... a S Eat-3.76.. can refinances—2. 0.2: wee—5. 83......82239 .3823».— caa ESE—30.393 6392389: 5 03¢.

46

 

 

 

 

 

62w... 2.5.3 .853 8:58. 5038.. >35... .5 2.... .a own :8... o... a. 855...... o... 5.. .8.-. 8......8-...o..=o..o..... .5 352.358 m
.35.. 3.3.2.... 80...: 60.8. 5......» .855 3:88. 503.2. 0.35882... 2.. .«o 5.2....me o... a. 8288...... 5.. .8. 253-..? .5 352.25% E
8:58 05...? v.8 .230 am 6855. 5.582 .m 5.8”. o
8.258 3.....me .5. €5.22 Sam .N 5.8... o
8.550 958.0 v.8 £08.82 .5mo..< .. 5&3. a
QC. ”6m 1? mdv Won 09. ..\¢ .325... . N
6.2. ”.mm adv «an ”.9 O? .x. 333...... . v
.d v.2 a... .3. md o6. ..\.. 052
m . .o 8... E... .8» .8.— ... 8.. 559.4.
m. . v New fine N. . m age... :3. ..\.. 85.5-52
..E m. . N v.8 odm ed. m... .x. 8.0:... 5....
m. . .. m.mm Qmm wdm 2h v.3. ..\o 85...... .850
on... an... on... on: 392......
3..."... .8"... San... .9"... as"... .Nmu...
ha 85558.. 355.8. N... 8388.. 9.0.52.8. ha 5....38... .5558.
.52 -52 -52
on 5.8m an 5.8.. a. 5&3.

 

.539. 0.59.50» .3 .829... .8289: awning. .5 8?...»
a 8 85:58.. E... Eo..5..8..-5.. 0...... mean... 8.828555 2.5.2.5.. .z... 6.553.782... 6.59.5989 ӣ85....ch a 2......

47

Behavioral characteristics of men

There was little difference between male non-responders and responders in their
smoking patterns. Approximately 43-49% of both groups reported themselves as non-
smokers and 31-39% of both groups as current smokers. A higher proportion of non-
responders, however, reported drinking no alcohol in the past year, 16% compared to 9%.

These trends were relatively similar within the three regions.

Demographic and behavioral characteristics of women

On the whole, these trends for most of the characteristics described above in
males were similar for females as well. These data are presented in Table 10. Female
non-responders were on average two years older at time of interview than responders,
31.4 years vs. 29.4 years, and a significantly larger percentage of female non-responders
were 35 years of age or older, 32.4% vs. 12.6%. Female non-responders were also more
likely to be Caucasian, be currently married, have higher annual incomes, have a college
degree, and report no alcohol consumption in the past year than their responder
counterparts, although because of the small sample size, only the income and alcohol

consumption characteristics reached statistical significance.

48

Table 10: Demographic, socio-economic, and behavioral characteristics among female
non-responders and responders to a survey of Michigan licensed anglers.

 

 

Non-responders (n=3 8) Responders (n=316) p“
95% Cib 95% ab

Region of Residence <0.01
West Michigan % 42.1 (26.7, 59.1) 19.9 (15.7, 24.8)
Bay Area % 13.2 (5.0, 28.9) 29.4 (24.5, 34.8)
Southeast Michigan % 44.7 (29.0, 61.5) 50.6 (45.0, 56.2)
Year of birth (median) year 1963 1964
Age at time of survey (mean) years 31.4 (29.9, 33.0) 29.4 (28.9, 29.9) 0.020
Age at time of survey 0.02
320 years % 0.0 (0.0, 11.4) 2.9 (1.4, 5.6)
21-26 years % 18.9 (8.7, 35.4) 23.5 (19.0, 28.6)
27-32 years % 32.4 (18.7, 49.6) 43.9 (38.4, 49.6)
33-34 years % 16.2 (6.9, 32.4) 17.1 (13.2, 21.8)
235 years % 32.4 (18.7, 49.6) 12.6 (9.3, 16.9)
Ethnic distribution 0.14
Caucasian % 97.3 (84.5, 99.9) 89.7 (85.7, 92.7)
Other % 2.7 (0.2, 15.5) 10.3 (7.3, 14.3)
Marital status 0.31
Never married % 24.3 (12.5, 41.3) 27.3 (22.5, 32.6)
Married % 70.3 (53.1, 83.4) 59.5 (53.9, 64.9)
Divorced, separated, widowed % 5.4 (1.0, 19.3) 13.2 (9.8, 17.6)
Annual income 0.04
<20,000 % 6.3 (1.3, 20.4) 26.7 (22.0, 32.0)
$20,000 - $39,999 % 46.9 (30.9, 63.5) 38.0 (32.7, 43.6)
>= $40,000 % 46.9 (30.9, 63.5) 35.3 (30.1, 40.9)
Highest education level 0.07
High school degree or less % 43.2 (29.7, 60.1) 37.3 (32.0, 42.9)
Some college, no degree % 27.0 (14.5, 44.1) 45.0 (39.5, 50.7)
College degree or higher % 29.7 (16.6, 46.9) 17.7 (13.8, 22.5)
Tobacco use 0.48
Current smoker % 42.1 (26.7, 59.1) 35.7 (30.5, 41.3)
Prior smoker % 13.2 (5.0, 28.90 21.2 (16.9, 26.2)
Non-smoker % 44.7 (29.0, 61.5) 43.1 (37.6, 48.8)
Alcohol use in past year 0.01
None % 26.3 (14.0, 43.4) 10.3 (7.3, 14.3)
< 1 drink/wk % 55.3 (38.5, 71.0) 70.8 (65.4, 75.7)
2 1 drink/wk % 18.4 (8.3, 34.9) 18.9 (14.8, 23.8)

 

0 Significance of chi-square test for differences in the distribution of the characteristic between response

groups, unless otherwise noted.

b 95% confidence interval.

0 Significance of independent-samples t-test for the difference in the mean age at time of survey between

response groups.

49

Fishing habits and fish consumption

Table 11 presents data on fishing habits and fish consumption for both men and
women. Region-specific analyses of these characteristics for men are presented in Table
12. Among men, non-responders as a whole fished significantly fewer days than did
responders. Non-responder men fished in the Great Lakes on average 26.3 days in the
past year, compared with 33.5 days for responders. In addition, 12.2% of non-responder
men reported no fishing at all during the past year, compared to 4.3% of responders.
Approximately 40% of non-responders and 50% of responders claimed to have fished 25
or more days in the past year. These trends generally hold for men when examined
regionally, particularly for men in Regions 2 and 3. In Region 1, although there were a
higher proportion of non-responder men who reported no fishing during the past year
compared to responders, 12.3% vs. 3.3%, the non-responder men on average fished about
the same number of days as did their responder counterparts, 34.2 and 36.8 days,
respectively. Interestingly, non-responder men from Region 1 fished more days than the
non-responder men from the other regions (22.1 and 21.8 days for non-responder men in
Regions 2 and 3), although this was not statistically significant by the Kruskal-Wallis
one-way ANOVA nonparametric test (p=0.11).

These trends were similar for women, although both non-responder and responder
women fished on significantly fewer days than the their male counterparts. Non-
responder women on average fished in the Great Lakes on slightly fewer days in the past
year than did responder women, 12.9 vs. 15.9 days, respectively, and a higher percentage

of non-responder women reported no fishing at all during the past year, 28.9% compared

50

to 14.1%. Neither of these findings were statistically significant, however, due to the
small sample size of non-responder women.

For men, there was no clear pattern in consumption of Great Lakes sport fish.
Over 40% of both non-responders and responders reported eating 1-12 Great Lakes sport
fish meals in the past year. A higher percentage of non-responders, however, did not eat
any Great Lakes sport fish in the past year compared to responders, 20.5% vs. 13.7%.
For women, significantly more non-responders did not eat any Great Lakes sport fish

meals in the past year compared to responders: about 45% vs. 25%, respectively.

51

Table 11: Fishing habits and Great Lakes sport fish consumption in past year among non-
responders and responders to a survey of Michigan licensed anglers, by sex.

 

Male non-responders Male responders pa
(n=230) (n=1,129)
95% Clb 95% orb

 

Days fished in waters of State of days 26.3 (22.2, 30.4) 33.5 (31.2, 35.8) <0.01“
Michigan in past year (mean)

 

Days fished in waters of State of <0.001
Michigan in past year
None % 12.2 (8.4, 17.3) 4.3 (3.2, 5.7)
1 — 12 days % 33.2 (27.2, 39.7) 28.9 (26.3, 31.7)
13 — 24 days % 14.4 (10.3, 19.8) 18.1 (15.9, 20.5)
2 25 days % 40.2 (33.9, 46.9) 48.7 (45.8, 51.7)
Great Lakes sport fish meals 0.02
eaten in past year
None % 20.5 (15.6, 26.4) 13.7 (11.8, 15.9)
1 — 12 meals % 40.6 (34.2, 47.3) 45.6 (42.7, 48.6)
13 — 24 meals % 20.1 (15.2, 26.0) 25.4 (22.9, 28.1)
2 25 meals % 18.8 (14.1, 24.6) 15.3 (13.3, 17.6)
Female non-responders Female responders p0
(n=3 8) (n=3 16)
95% orb 95% ab

 

Days fished in waters of State of days 12.9 (5.8, 20.1) 15.9 (13.3, 18.5) 046“
Michigan in past year (mean)

Days fished in waters of State of 0.13
Michigan in past year

None % 28.9 (15.9, 46.1) 14.1 (10.6, 18.5)

1 - 12 days % 42.1 (26.7, 59.1) 50.3 (44.7, 55.9)

13 — 24 days % 13.2 (5.0, 28.9) 17.3 (13.4, 22.0)

2 25 days % 15.8 (6.6, 31.9) 18.3 (14.3, 23.1)

Great Lakes sport fish meals 0.05
eaten in past year

None % 44.7 (29.0, 61 .5) 24.6 (20.0, 29.8)

1 - 12 meals % 28.9 (15.9, 46.1) 48.4 (42.8, 54.1)

13 —- 24 meals % 15.8 (6.6, 31.9) 16.3 (12.5, 20.9)

2 25 meals % 10.5 (3.4, 25.7) 10.7 (7.6, 14.8)

 

a Significance of chi-square test for differences in the distribution of the characteristic between response
groups, unless otherwise noted.

5 95% confidence interval.

0 Significance of independent-samples t-test for the difference in the mean number of days fished
between response groups.

52

dawn. £53, 698% 8:88. c8389 85: when no 898:: :88 05 E nag—8&6 05 .8 .83 moan—smacogoaoefi mo 8:85:3m a
.88: 33:28 82:: .532 55; .mqaoa 8:88. 8833 unficofignu 05 .3 5:2:me 2: =_ 80:20.:E 3.. “8. 0333-20 me 3585ch b

.8658 0:83 28 .236 am 68:22 55082 ”m 539: o
.8358 Bafiwmm v5 €5.22 Sam ”N came”. a
8:580 «355 93 acme—82 demo—2 n. :2qu o

 

 

 

v.2 odm v.2 n4: 3.— NNN .x. 288 mm N
Wmm a: fig ES 0.5 Sm .x. £88 vm I m—
5? «.mm ”av 93. fine ndm ..\o £85 2 I _
m2 Ndm _.: Q3 wN- m5 .x. 252
can.» .2:— 5 :88
mod Sac 2d 28:. .7...— team 8:: «8.0
9% m.mm edv 9mm 9% fine o\. 996 3 N
”.2 E. _.om m6. m5. Q! .X. mam—u vm I m—
m.w~ 2% _.mm Eh N.n~ Sam ..\o 99% N- I _
ad Na Na.” WE m.m m.~_ .x. 052
.39» «8.. a. swig:
53. .85 :3 3 33m 3 can: a. 8%... 25
38.5 .39» 32. E sane—9.2
mmod Own a. _ N 85.: wdm _.Nm 8%.: «an Ném 3% .3 88m .3 9.853 E toga..— 9?:
Angus Gen—c Gmmuav ASHE A3 mus Amwnav .
Ba flouceamom Eoucoa8c Va “engined 83:83. Va— ncocaeamox 83°92
.52 .52 .52
on Ewan amas— & some. ‘

 

£3»?— omaaauueou .3 .82»: eon—.3: aqua—32 be

that...» a 8 82.539. can 206:38553 22: «38a .39» .3.— E Egan—3:3 5.: to? 8:3 39.6 can 532. uE-Eh fi— 93:.

53

Reproductive characteristics

Table 13 provides data on reproductive characteristics of the two populations.
Table 14 provides region-specific analyses of reproductive characteristics for males.
Among men, similar percentages of non-responders and responders reported having
fathered at least one child, but non-responders were more likely than responders to have
fathered two or more children, 40.5% vs. 28.1%, respectively. Consistently, non-
responders were significantly less likely to intend to have at least one child in the next
five years, 27.2% compared with 40.4% of responders. These trends were consistent
among the regions. Similar trends persisted for women as well, but statistical
significance was lost due to smaller cell sizes.

A key component in analysis of response bias is the potential for selective
response by those who have experienced the study outcome event. As shown in Table
13, among men, a small excess of responders (8.5% compared to 5.7% for non-
responders) reported a period of conception failure, defined as trying to conceive and not
succeeding for twelve months or more. This difference, however, was not significant.
There was no difference in the proportion who reported a medically diagnosed
reproductive problem with fathering a child; for both groups only 2.2% reported a
medically diagnosed reproductive difficulty.

For women, both non-responders and responders were several times more likely
to report difficulty conceiving or having a medically diagnosed reproductive problem
than men. However, the non-responders did not differ significantly from the responders

for either characteristic. Indeed, female non-responders even reported a higher

54

prevalence of a period of conception failure (15.8%) than did female responders (11.6%).

Thus, there was no suggestion of this response bias.

Table 13: Reproductive history among non-responders and responders to a survey of

Michigan licensed anglers.

 

Male non-responders

Male responders

 

 

 

a
(n=230) (n=1,129) p
% 95% ab % 95% CI!)
Fathered 2 I pregnancy 54.4 (47.7, 60.9) 51.4 (48.4, 54.4) 0.55
Fathered 2 1 live birth 52.4 (45.7, 59.0) 46.3 (43.4, 49.3) 0.09
Number of previous live births <0.001
None 47.6 (41.0, 54.3) 53.7 (50.7, 56.6)
1 11.9 (8.2, 17.0) 18.2 (16.0, 20.6)
_>_ 2 40.5 (34.2, 47.2) 28.1 (25.5, 30.8)
Intends to have 2 1 child within the 27.2 (21-7. 33-5) 40-4 (37.5. 433) <0-001
next 5 years
Has tried, unsuccessfully, to 5.7 (3.2, 9.8) 8.5 (7.0, 10.3) 0.16
father/conceive a child for > 1 year
Doctor has said that s/he would 2.2 (0.8, 5.3) 2.2 (1.5, 3.3) 0.98
have difficulty having children
Female non-responders Female responders a
(n=38) (n=316) P
% 95% ab % 95% crb
Experienced 2 1 pregnancy 65.8 (48.6, 79.9) 63.7 (58.1, 69.0) 0.80
Experienced 2 1 live birth 65.8 (48.6, 79.9) 55.1 (49.4, 60.6) 0.21
Number of previous live births 0.1 1
None 34.2 (20.1, 51.4) 44.9 (39.4, 50.6)
1 13.2 (5.0, 28.9) 19.9 (15.7, 24.8)
2 2 52.6 (36.0, 68.7) 35.3 (30.1, 40.9)
Intends to have 2 1 child within the 3 1-6 (13-1. 48-3) 50-2 (44-6. 55-3) 0-03
next 5 years
Has tried, unsuccessfully, to 15.8 (6.6, 31.9) 11.6 (8.4, 15.8) 0.45
father/conceive a child for > 1 year
Doctor has said that s/he would 10.8 (3.6, 26.1) 11.9 (8.4, 16.1) 0.85

have difficulty having children

a Significance of chi-scTuare test for differences in the distribution of the characteristic between response

groups, within sex.
5 95% confidence interval.

55

:2»... 55.3 anew 8:88.. 80.30.. 0.3.8822: o... .o 8.395%: e... :. 8:88.....: 8.. .8. 2:8».-.8 .o 858585 :
8:88 o??? :8 .230 am 68:02 5:88.). ”m 8&3. o

8:58 BEES :8 .:::_:_2 Sam. ”N 8&3. ..

8:55: 55.80 :8 acme—8.2 .58.? u. 8&3. a

8.5.2.: 9...»...— b.8..=.: 2:...

 

 

::.: S n. 8.: ..~ 5: 3.: :.N n. .x. 2.5: 2. 2.... a: a... .28:
.39. _ A .3. 2.8 a 8......
m _ .: ..:. o... 3.: on m: 5.: 4.: :.... .\. 3 £=a§3=m== .8... 2.:
88.» n «no: 2..
_:.:v .3... :8 8.: an: 3.: 3.: 9:: 3m .x. 5...... 2...0 . N 2:... 3 8.3....
3: In :5 Nov .3: :8 .x. N N
3. 8. 2: o... :.:N .2. .\. .
m... an: 5.: 3:. o. .v n... .x. 282
8.: 3.: :0: 3...... 9... 28.3... .o .382
2.: Sn «.2. X: «.8 NE 8.: 4...: 3.: .x. 5...: 2... . N 3.2.2....
::.: :9. 8.. :m: 8: E: 8.: 0:: 2: .x. .238... . N 3.2.2.:

5.1: .8"... San... .3"... a. mu... awn...

N... m.o:..o..8x 38:88.. 1. 8.82.81 E0888. Va 80888: 88:88.
IcOz IEOZ. ISOZ
um 8&3. :N 8&3. :. 8&3.

 

:28. 2.1.638» .3 ago—w...“ .888: gut—9.2 .: 8?...» a S 80:558.. :5. 88:88.78: M85: Dem... o>.8=:e.:.o~. ”v. «Ben.

56

Logistic regression analyses

Table 15 shows the results of the logistic regression models relating fish
consumption in the past 12 months to the prevalence of a period of conception failure
among both ever-married and currently-married male anglers. Among both responders
and non-responders, there is an increased prevalence of a period of conception failure
among those men who reported consuming any Great Lakes sport fish in the past year.
The unadjusted odds ratio for conception failure among ever-married non-responders was
0.96 for any Great Lakes sport fish consumption in the previous year compared to non-
consumers. Among ever-married responders, the odds ratio was 1.47. Afier adjusting for
age, race, region of residence, household income, education, tobacco use, and alcohol
consumption, the odds ratios were 2.24 and 1.52, respectively. It should be noted that
these odds ratios differ from those presented by Courval et al. (13) because Great Lakes
sport fish consumption of non-responders was recorded only for that within the previous
year and not for lifetime consumption. Analysis for this study was restricted to Great
Lakes sport fish consumption within the past year.

A dose-response relationship between fish consumption and conception failure
was found among ever-married responder men. This relationship was not evident among
the non-responder men; however, the non-responders represent a small sample, and
estimates are inevitably imprecise. On the whole, these trends were similar for currently-

married male anglers as well.

57

Table 15: Associations between Great Lakes sport fish consumption and prevalence of a
period of conception failure among male responders and non-responders to a survey of

Michigan licensed anglers.

 

Ever-married

 

Non-
responders Responders
(n=129) (n=806)

Non-
responders Responders
(n=129) (n=806)

 

Great Lakes sport fish meals
consumed in past year

unadjusted odds ratio

adjusted odds ratio a

 

 

 

Model 1
None (-) (-) (-) (-)
Any 0.96 1.47 2.24 1.52
Model 2
None (-) (-) (-) (-)
1 — 12 meals 1.13 1.35 2.89 1.39
13 — 24 meals 0.72 1.61 1.98 1.63
_>_ 25 meals 0.87 1.71 1.11 1.97
Currently married
Non- Non-
responders Responders responders Responders
(n=1 13) (n=721) (n=1 13) (n=721)
Great Lakes sport fish meals unadjusted odds ratio adjusted odds ratio a
consumed in past year
Model 1
None (-) (-) (-) (-)
Any 1.56 1.89 2.81 1.95
Model 2
None (-) (-) H H
l — 12 meals 1.76 1.68 3.21 1.71
13 - 24 meals 1.20 2.15 2.67 2.21
2 25 meals 1.50 2.40 1.71 2.88

 

a Adjusted for region, age, race, education, income, tobacco use, and alcohol consumption.

58

CHAPTER 5

DISCUSSION

In this study, non-responders were approximately 1.5 years older at interview,
were more likely to be Caucasian, and reported higher incomes than responders. No
differences, however, were found with respect to education level, marital status, or
smoking. Non-responders had fished fewer days in the past year and consumed fewer
fish meals than responders. Compared with responders, non-responders were more likely
to have had two or more children and were less likely to intend to have additional
children in the next five years.

However, these differences, while real, did not impact on the previously observed
association between sport fish consumption and conception failure as published by
Courval et al. (13). Among both non-responders and responders there was an increased
prevalence of a period of conception failure among men who reported consuming greater
quantities of Great Lakes sport fish. These results suggest that non-response bias is
unlikely to have played a major role in the observed association.

There are several limitations of this study. The first limitation was an inability to
recruit a sufficient number of female non-responders. The total of 38 females out of a
desired 120 provides very little statistical power. Indeed, on a nmnber of study variables,
similar trends were observed for both males and females. For example, 28.9% of female
non-responders reported having not fished in the Great Lakes within the past year, as
compared to 14.1% for female responders. For men, these percentages were 12.2% and

4.3%, respectively. While clearly females in general were more likely than males to have

59

not fished within the past year, the trend is similar. While this difference reached
statistical significance for men, it did not for women. This lack of power was also most
notable on the number of previous live births, with both male and female non-responders
reported having more children, being only significant for males.

The second limitation of this study was the roughly two year delay in interviewing
the non-responders. While the data for the original F F HP study was collected between
1993 and 1995, the data for this non-response study was collected in 1996. This two year
delay correlates remarkably well with the non-responders mean age, which was 1.5 to two
years older at the time of interview. They had, however, similar median years of birth.
In fact, for males, median year of birth was 1965 for non-responders and 1964 for
responders; if the non-responders and responders were assessed simultaneously, the non-
responders may have been on average younger than the responders.

This delay in assessment can predictably produce spurious results. Being on
average two years older, non-responders could have been expected to have higher
incomes, to have completed schooling and college, to have had a child, and to have
completed their planned family. Many of the differences found between non-responders
and responders are likely to be confounded by this delay. The differences of many socio-
demographic and behavioral characteristics found in this study are not easily comparable
to those found in the literature due to this critical delay in non-responder assessment.
Ideally, this limitation could have been avoided had the non-responder study been
conducted simultaneously with the main F isheaters study.

The third limitation of this study is the possibility that the results could be invalid

due to the differences in interview method. The responders were surveyed by a postal

60

survey. In contrast, the non-responders were surveyed by telephone interview. This
raises the possibility of differential response between non-responders and responders.

According to Dillman et al. (140), there are several major differences between
mail and telephone survey. First, telephone interviews require the presence of an
interviewer to read questions and record responses, whereas mail surveys can be done
privately, directly, and virtually anonymously. The necessary social interaction of a
telephone interview may effect answers from respondents. Respondents may be looking
for the most socially acceptable response or not want to divulge personal and sensitive
information to a total stranger.

Second, telephone interviews require dependence on visual or aural
communication, whereas mail surveys depend solely on written directions and cues.
Dillman et al. (140) suggested that memory limitations and cognitive processes may
effect responses during a telephone interview. Lengthy questions and the interviewer’s
pace, which is frequently pressured, could make responding to the telephone questions
more difficult. In contrast, mail questionnaire respondents can look back and forth at the
question and answer choices as needed and answer at their own pace.

Third, the context for responding differs between the two modalities. In telephone
interviews, respondents answer one question at a time. In contrast, during mail surveys
respondents can see individual questions as part of a larger set of questions; they can look
ahead and preview questions and answer them in a different order.

Despite the potential differences between mail and telephone surveys, few studies
have directly compared data collected by the two methods. Of the handful of studies

performed, there are conflicting results. In a study by McHorney et al. (141), mail

61

respondents were more likely than telephone respondents to skip questions, report a less
favorable health status, and have a chronic medical condition. However, other studies
have found little or no systematic differences in responses from mail and telephone
respondents (105, 142-144). Mixed results have been reported across survey modes for
tests of reliability and response validity (see McHorney et al. (141) for a brief review). In
short, few generalizations can be gleaned about differential score reliability or response
validity.

A major limitation of this study was an inability to interview a true random
sample of the non-responders to the original survey. Telephone numbers were unable to
be found in a CD-ROM telephone database (137) for 72% of those eligible for this non-
response study, and an additional 10% had disconnected or incorrect telephone numbers.
The latter can be attributed primarily to the time interval between the two studies. Given
our young, mobile population, it is not unreasonable to expect a significant percentage to
have changed residences in this time interval.

The former, however, cannot as easily be explained. A telephone database from
1992 was explicitly used to obtain names and addresses of licensed anglers, rather than
one from a current one, in hopes of identifying more eligible anglers. It was felt that this
would give the best chance of locating an individual, particularly if that individual had
moved between then and the current non-response sub-study. If that individual had
moved, a current telephone database would result in no listing; a previous database
would, at a minimum, provide a telephone number that could be used to attempt a tracing.
In addition, “sofi” matching criteria of either a last name and street address, or unique

name and city, was used, without regard to address, in hopes of increasing sensitivity.

62

However, while these techniques did result in a tangible number of completed interviews,
neither was notably successful in significantly increasing the low telephone number
identification percentage. Furthermore, while a certain percentage of eligible anglers
could be expected to have unlisted numbers, as well as to cohabit with an unrelated
individual who was the telephone subscriber and therefore not be listed, these factors
alone would not be expected to account for all of our difficulty.

Regardless of the causes, the inability to interview a true random sample of the
non-responders may have resulted in a study that continues to be biased, as interviewees
represent the responders among a study of non-responders. Nevertheless, it is felt that the
sample is reasonably valid, as the primary reason for non-participation was not related to
the study population’s awareness of the hypothesis, but to the above-mentioned inability
to identify telephone numbers for a large proportion of the population using publicly
available information. Once an individual was personally reached, participation was
74%.

Finally, this study cannot address the more critical aspect of the main study,
which is the cross-sectional nature of the observed relationship. A time order to the
association can not be assigned: indeed, it is likely that the period of conception failure
was in the more remote past than the past 12 months for which fish consumption was
estimated. Resolution of this limitation requires prospective data, which is currently
being accrued (68).

Despite these limitations, this study has a number of strengths. This study adds to
the growing literature on non-response bias. Several socio-demographic and behavioral

characteristics were assessed. Similarities and differences in results of different variables

63

provides for further intellectual thought and debate. In addition, this study is the first
study to our knowledge to assess non-response bias in a reproductive epidemiological
study. Furthermore, it provides validation to the findings of the relationship between
sport-caught fish consumption and conception delay.

While this study provides useful information, as with any study it could be
improved upon. Future studies should give consideration to alternative recruitment
methods. While in 1996 world wide web telephone directories were rather primitive,
today they are quite advanced, and their use in searching for persons is suggested. Given
that anglers are required to purchase annual fishing licenses, it may be possible to cross-
reference names from previous databases to names on the current license database.
Finally, the use of credit histories and banking information to locate individuals is a
possible, albeit costly, method at locating individuals.

Future studies should also consider the application of more advanced statistical
analyses. The analyses conducted here consisted mainly of Pearson )8 test for
dichotomous variables and independent samples t-test for continuous variables.
Multivariate logistic regression modeling could be done in order to ascertain independent
predictors of response.

Finally, as mentioned above, ideally the non-responder sub-study should have
been conducted simultaneously with the main Fisheaters study. Future studies should

allow for non-response considerations in its planning stages.

64

APPENDICES

65

APPENDIX A

TELEPHONE CALLING RULES OF REPLACEMENT

CODE

01

02

O3

04

05

O6

O7

08

09

10

Completed interview
Wrong number
Non-working number

Ring no answer (five
rings)

Answering machine --
left message

Line busy

Language barrier

Refused interview

Correct number but
angler not home

Angler home but bad
time -- appointment
made

EXPLANATION/RULE
Do not replace.

 

Angler does not live at this residence. Replace.

Usually recognized by a recording or fast-busy signal.
Replace.

Normal telephone ring which no one answers. Replace
after five (5) calling occasions, each consisting of three
(3) attempts, and occurring at varying periods of the day
and on different days (weekday, weeknight, weekend).

Answering machine picks up. Leave name, message
(script), and l-800#. Replace afier five (5) calling
occasions, occurring at varying periods of the day and on
different days (weekday, weeknight, weekend)

Replace only after five (5) calling occasions, each

consisting of three (3) attempts at 2 10 minute intervals,
and occurring at varying periods of the day and on
different days (weekday, weeknight, weekend).

Angler does not speak English well enough to be
interviewed and there are no interviewers who speak
his/her language. Replace.

Replace.

Other family member answers telephone, but angler
him/herself is not home. Attempt to determine when

angler will be home and schedule an appointment.

Angler home, but bad time for interview. Appointment
scheduled.

66

CODE

11

12

13

Terminated within
interview

Respondent unable to
communicate

Angler no longer
resides at this

residence -- left 1-
800#

EXPLANATION/RULE

 

A hang-up or refusal to continue at some point after the
first question has been asked (this does not mean the
respondent refused a particular question). If
disconnection was accidental, phone again and attempt to
complete interview; otherwise, consider completed.

Physical/mental impairment. Example: respondent is
deaf. Replace.

Family (e.g., parents) resides at residence, but angler has
moved. Give 1-800#.

67

APPENDIX B

QUESTIONNAIRE

I would first like to ask you some questions about your fishing and fish eating
habits.

1.

In this past fishing year, from April 1, 1994 to March 31,
1995, how many days did you go fishing in waters in the
State of Michigan?

In April, May, and June of this past fishing year (1994),
how many meals of Great Lakes sport fish that were
caught by you personally or by someone you know did you
eat?

In July, August, and September of this past fishing year,
how many meals of Great Lakes sport fish that were
caught by you personally or by someone you know did you
eat?

In October, November, and December of this past fishing
year, how many meals of Great Lakes sport fish that were
caught by you personally or by someone you know did you
eat?

In January, February, and March of this year (1995), how
many meals of Great Lakes sport fish that were caught by
you personally or by someone you know did you eat?

How do these amounts of fish that you eat during this past
fishing year compare with previous years? Would you say
you ate about the same, more than usual, or less than
usual?

days

meals

meals

meals

meals

same
more
less

Now I would like to ask you some general questions about your health.

7.

Have you smoked more than 100 cigarettes during your

entire life?
NO: GOTO QUESTION 9.

68

yes
no

10.

11.

12.

13.

14.

15.

Do you currently smoke cigarettes?

How often have you drunk alcoholic beverages over the
past 12 months?

1 Never 5 4-6 days/week
2 Once a month or less 6 About once a day
3 > l/mo. but < l/wk 7 More than once a day
4 2 l/wk but < 4
days/wk
How many times have you and any partner conceived a
child together?
NONE: GOTO 12.

How many of these have resulted in a live birth?

Have you ever had a child with a birth defect?

How many children do you intend to have within the next
five years (not including any child if currently pregnant)?

Did you and any partner ever try for more than one year to
conceive a child without being able to do so?

Has a doctor told you that it would be difficult or

impossible for you to have children?
NO: GOTO 15.

What was the reason?

 

Now I would like to ask you some questions about yourself.

16.

17.

In what year were you born?

What is your main racial background?

1 African-American 4 Caucasian
2 American Indian 5 Hispanic
3 Asian/Pacific Islander

69

code

yes
no

preg.

births

19

code

yes
no

kids

yes

no

yes
no

18. What is the highest level of school you have completed?
1 No high school degree 3 Some college

2 High school degree 4 College degree or code
higher
19. What is your present marital status?
1 Never married 4 Separated
2 Married 5 Widowed
3 Divorced code

20. Which category best describes your total family income

for 1994?
1 < $20,000 3 $40,000 or more
2 $20,000 to $39,999 code

21. Do you have a spouse or partner? yes
NO: GOTO END. no
YES: May I ask some questions of him/her?

END:

Thank you very much for taking the time to answer these questions. Like I said before,
your responses will be very helpful in the analysis of our data and will be kept strictly
confidential. I greatly appreciate your help. Do you have any questions or comments?

Would you like to receive any information about our study?
NO: Well, thank you, again. Goodbye.
YES: Can I confirm your address? My records indicate that you live at...

l. Wants information about study B 1. Correct Address
2. Does not want information 2. New Address:
about study

COMMENTS/QUESTIONS:

7O

APPENDIX C

QUESTIONNAIRE SCRIPT AND LOG

Name:
Address:

Hello, may I speak with (First and last name)? Hi, my name is Gene Tay. 1 am a graduate
student at Michigan State University, and I'm working with the Fisheaters Family Health Project,
which is a project that is looking at people's eating habits of Great Lakes fish. Within the past
year or two, we mailed a questionnaire to you on this topic, but we did not receive it back. At
this time, I am not asking you to participate in our study, but 1 would like to ask you a few
questions about your fish eating habits and your family background so that we can learn whether
those who responded to our questionnaire are different from those who did not. Your responses
will be critical in the analysis of our data and will also be helpful for me for my master's thesis
project. I want to let you know that your participation in this telephone survey is voluntary, but if
you do, your responses will be kept strictly confidential. In addition, you may refuse to answer
any question you do not wish to answer, and you may stop this interview at any time. Is this a

good time for you to spend about five minutes to answer my questions?

NO: Is there a better time I could call back? Like I said, your responses are very important to us.
NO AGAIN: Okay. Thank you for your time and I'm sorry to have inconvenienced you.
Goodbye.

YES: Continue.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

LOG
CALL #1 CALL #2 CALL #3
DATE TIME RESULT DATE TIME RESULT DATE TIME RESULT
CALL #4 CALL #5
DATE TIME RESULT DATE TIME RESULT

 

 

 

 

 

 

71

REFERENCES

72

10.

11.

REFERENCES

Department of Natural Resources. Home Page. [Online] Available
http://www.dnr.state.mi.us, 1999.

Roche HM, Gibney MJ. Long-chain n-3 polyunsaturated fatty acids and
triacylglycerol metabolism in the postprandial state. Lipids 1999;34 Suppl:8259-
65.

Howe PR, Clifton PM, James MJ. Equal antithrombotic and triglyceride-lowering

effectiveness of eicosapentaenoic acid-rich and docosahexaenoic acid-rich fish oil
supplements. Lipids 1999;34 Suppl:S307-8.

Mori TA, Bao DQ, Burke V, Puddey IB, Beilin LJ. Docosahexaenoic acid but not
eicosapentaenoic acid lowers ambulatory blood pressure and heart rate in humans.
Hypertension 1999;34:253-60.

Fein GG, Jacobson JL, Jacobson SW, Schwartz PM, Dowler JK. Prenatal exposure

to polychlorinated biphenyls: effects on birth size and gestational age. J Pediatr
1984;105:315-20.

Jacobson SW, Fein GG, Jacobson JL, Schwartz PM, Dowler JK. The effect of
intrauterine PCB exposure on visual recognition memory. Child Dev 1985;56:853-
60.

Jacobson JL, Jacobson SW, Humphrey HE. Effects of exposure to PCBs and
related compounds on growth and activity in children. Neurotoxicol Teratol
1990;12:319-26.

Jacobson JL, Jacobson SW, Humphrey HE. Effects of in utero exposure to
polychlorinated biphenyls and related contaminants on cognitive functioning in
young children. J Pediatr 1990;116:38-45.

Jacobson JL, Jacobson SW. Intellectual impairment in children exposed to
polychlorinated biphenyls in utero. New Engl J Med 1996;335:783-89.

Department of Environmental Quality. Sport-caught fish consumption advisory for
the Saginaw Bay watershed. [Online] Available http://www.deq.state.mi.us/ogl/
sagbaylsprtfishhtm, 1997.

Schwartz PM, Jacobson SW, Fein G, Jacobson JL, Price HA. Lake Michigan fish
consumption as a source of polychlorinated biphenyls in human cord serum,
maternal serum, and milk. Am J Public Health 1983;73:293-96.

73

12.

l3.

14.

15.

l6.

17.

18.

19.

20.

21.

22.

23.

24.

25.

Department of Community Health. Charts of specific advisories. [Online]
Available http://www.mdch.state.mi.us/pha/fishadvi.htm, 1999.

Courval JM, DeHoog JV, Stein AD, Tay EM, He JP, Humphrey HEB, Paneth N.
Sport-caught fish consumption and conception failure in licensed Michigan anglers.
Environ Res 1999;80:8183-8188.

Sackett DL. Bias in analytic research. J Chronic Dis 1979;32:51-63.

Stein AD, Tay E, Courval JM. Absence of nonresponse bias in a study of sport-
caught Great Lakes fish consumption and conception failure. Environ Res
1999;80:289-93.

Safe S. Toxicology, structure-function relationship, and human and environmental
health impacts of polychlorinated biphenyls: progress and problems. Environ
Health Perspect 1992;100:259-68.

Fischer LJ, Seegal RF, Ganey PE, Pessah IN, Kodavanti PR. Symposium
overview: toxicity of non-coplanar PCBs. Toxicol Sci 1998;41:49-61.

Parkinson A, Safe S. “Mammalian biologic and toxic effects of PCBs.”
Polychlorinated biphenyls (PCBs): mammalian and environmental toxicology. Ed.
S. Safe and O. Hutzinger. Heidelberg: Springer-Verlag, 1987. 49-75.

Kimbrough RD. Human health effects of polychlorinated biphenyls (PCBs) and
polybrominated biphenyls (PBBs). Ann Rev Pharmacol Toxicol 1987;27:87-111.

Safe S. Polychlorinated biphenyls (PCBs) and polybrominated biphenyls (PBBs):
biochemistry, toxicology, and mechanism of action. Crit Rev Toxicol 1984;13:319-
95.

Peterson RE, Theobald HM, Kimmel GL. Developmental and reproductive toxicity
of dioxins and related compounds: cross-species comparisons. Crit Rev Toxicol
1993;23:283-335.

Battershill JM. Review of the safety assessment of polychlorinated biphenyls
(PCBs) with particular reference to reproductive toxicity. Hum Exp Toxicol
1994;13:581-97.

Kreiss K. Studies on populations exposed to polychlorinated biphenyls. Environ
Health Perspect 1985;60:193-99.

Lione A. Polychlorinated biphenyls and reproduction. Reprod Toxicol 1988;2283-
89.

Kimbrough RD. Laboratory and human studies on polychlorinated biphenyls
(PCBs) and related compounds. Environ Health Perspect 1985;59:99-106.

74

26.

27.

28.

29.

30.

31.

32.

33.

34.

35.

36.

37.

Huang J. Effects of 3,3',4,4'-tetrachlorobiphenyl on reproductive performance and
gamete fertilizing ability in mice. East Lansing, MI: Department of Animal
Science, Michigan State University, 1995.

Wiggers PA. Withdrawal of polychlorinated biphenyl (PCB) and poly-brominated
biphenyl (PBB) residues from rats using feed restriction and/or mineral oil in the
diet. East Lansing, MI: Department of Animal Science, Michigan State University,
1990.

Kuratsune M, Yoshimura T, Matsuzaka J, Yamaguchi A. Epidemiologic study on
Yusho, a poisoning caused by ingestion of rice oil contaminated with a commercial
brand of polychlorinated biphenyls. Environ Health Pespect 1972;1:119-28.

Chen PH, Chang KT, Lu YD. Polychlorinated biphenyls and polychlorinated
dibenzofurans in the toxic rice-bran oil that caused PCB poisoning in Taichung.
Bull Environ Contam Toxicol 1981;26:489-95.

Masuda Y. Health status of Japanese and Taiwanese after exposure to contaminated
rice oil. Environ Health Perspect 1985;60:321-25.

Rothman N, Cantor KP, Blair A, Bush D, Brock JW, Helzlsouer K, Zahm SH,
Needham LL, Pearson GR, Hoover RN, Comstock GW, Strickland PT. A nested

case-control study of non-Hodgkin lymphoma and serum organochlorine residues.
Lancet 1997;350:240-44.

Longnecker MP, Rogan WJ, Lucier G. The human health effects of DDT
(dichlorodiphenyltrichloroethane) and PCBs (polychlorinated biphenyls) and an
overview of organochlorines in public health. Ann Rev Pub Health 1997;18:211-
44.

Brown DP. Mortality of workers exposed to polychlorinated biphenyls — an update.
Arch Environ Health 1987;42:333-39.

Bertazzi PA, Riboldi L, Pesatori A, Radice L, Zocchetti C. Cancer mortality of
capacity manufacturing workers. Am J Ind Med 1987;11:165-76.

Sinks T, Steele G, Smith AB, Watkins K, Shults RA. Mortality among workers
exposed to polychlorinated biphenyls. Am J Epidemiol 1992;136:389-98.

Bahn AK, Rosenwaike I, Hermann N, Grover P, Stellman J, O'Leary K. Melanoma
after exposure to PCBs. New Engl J Med 1976;295:450.

Loomis D, Browning SR, Schenck AP, Gregory E, Savitz DA. Cancer mortality
among electric utility workers exposed to polychlorinated biphenyls. Occup and
Environ Med 1997;54:720-28.

75

38.

39.

40.

41.

42.

43.

44.

45.

46.

47.

48.

Hardell L, van Bavel B, Lindstrom G, Fredrikson M, Hagberg H, Lijegren G,
Nordstrom M, Johansson B. Higher concentrations of specific polychlorinated
biphenyl congeners in adipose tissue from non-Hodgkin's lymphoma patients
compared to controls without malignant disease. Int J Oncol 1996;9:603-8.

Fischbein A, Thornton J, Wolff MS, Bernstein J, Selikoff IJ. Dermatological
findings in capacitor manufacturing workers exposed to dielectric fluids containing
polychlorinated biphenyls (PCBs). Arch Environ Health 1982;37:69-74.

Chase KH, Wong 0, Thomas D, Berney BW, Simon RK. Clinical and metabolic
abnormalities associated with occupational exposure to polychlorinated biphenyls
(PCBs). J Occup Med 1982;24:109-14.

Rogan WJ, Gladen BC, Hung KL, Koong SL, L.Y. S, J.S. T, Wu YC, Yang D,
Ragan NB, Hsu CC. Congenital poisoning by polychlorinated biphenyls and their
contaminants in Taiwan. Science 1988;241:334—36.

Svensson BG, Hallberg T, Nilsson A, Schutz A, Hagmar L. Parameters of
immunological competence in subjects with high consumption of fish contaminated

with persistent organochlorine compounds. Int Arch Occup Environ Health
1994;65:351-58.

Weisglas-Kuperus N, Sas TC, Koopman-Esseboom C, van der Zwan CW, de
Ridder MA, Beishuizen A, Hooijkaas H, Sauer PJ. Immunologic effects of
background prenatal and postnatal exposure to dioxins and polychlorinated
biphenyls in Dutch infants. Pediatr Res 1995;38:404-10.

Warshaw R, Fischbein A, Thornton J, Miller A, Selikoff IJ. Decrease in vital
capacity in PCB-exposed workers in a capacitor manufacturing facility. Ann NY
Acad Sci 1979;320:277-83.

Humphrey HEB. Population studies of PCBs in Michigan residents. Ed. F.M.
D'Itri and M. Kamrin. Boston, MA: Butterworth, 1983.

Corrigan FM, Murray L, Wyatt CL, Shore RF. Diorthosubstituted polychlorinated
biphenyls in caudate nucleus in Parkinson's disease. Exper Neurol 1998;150:339-
42.

Mergler D, Belanger S, Larribe F, Panisset M, Bowler R, Baldwin M, Lebel J,
Hudnell K. Preliminary evidence of neurotoxicity associated with eating fish from
the Upper St. Lawrence River Lakes. Neurotoxicology 1998;19:691-702.

Rogan WJ, Gladen BC. Neurotoxicity of PCBs and related compounds.
Neurotoxicology 1992;13:27-36.

76

49.

50.

51.

52.

53.

54.

55.

56.

57.

58.

59.

60.

Gladen BC, Rogan WJ, Hardy P, Thullen J, Tingelstad J, Tully M. Development
after exposure to polychlorinated biphenyls and dichlorodiphenyl dichloroethene
transplacentally and through human milk. J Pediatr 1988;113:991-95.

Jacobson JL, Humphrey HE, Jacobson SW, Schantz SL, Mullin MD, Welch R.
Determinants of polychlorinated biphenyls (PCBs), polybrominated biphenyls
(PBBs), and dichlorodiphenyl trichloroethane (DDT) levels in the sera of young
children. Am J Public Health 1989;79:1401-4.

Jacobson JL, Jacobson SW. Evidence for PCBs as neurodevelopmental toxicants in
humans. Neurotoxicology 1997; 1 8:41 5-24.

Foster WG. The reproductive toxicology of Great Lakes contaminants. Environ
Health Perspect 1995;103 Suppl 9:63-69.

Buck GM. Epidemiologic perspective of the developmental neurotoxicity of PCBs
in humans. Neurotoxicol Teratol 1996;18:239-41, 271-76.

Mendola P, Buck GM, Vena JE, Zielezny M, Sever LE. Consumption of PCB-
contaminated sport fish and risk of spontaneous fetal death. Environ Health
Perspect1985;103:498-502.

Koopman-Esseboom C, Morse DC, Weisglas-Kuperus N, Lutkeschipholt IJ, van
der Pauw CG, Tuinstra LGMT, Brouwer A, Sauer PJJ. Effects of dioxins and
polychlorinated biphenyls on thyroid hormone status of pregnant women and their
infants. Pediatr Res 1994;36:468-73.

Taylor PR, Stelma JM, Lawrence CE. The relation of polychlorinated biphenyls to
birth weight and gestational age in the offspring of occupationally exposed mothers.
Am J Epidemiol 1989;129:395-406.

Wassermann M, Ron M, Bercovici B, Wassermann D, Cucos S, Pines A.
Premature delivery and organochlorine compounds: polychlorinated biphenyls and
some organochlorine insecticides. Environ Res 1982;28:106-12.

Gerhard I, Daniel V, Link S, Monga B, Runnebaum B. Chlorinated hydrocarbons
in women with repeated miscarriages. Environ Health Perspect 1998;106:675-81.

Buck GM, Sever LE, Mendola P, Zielezny M, Vena JE. Consumption of
contaminated sport fish from Lake Ontario and time-to-pregnancy. New York State
Angler Cohort. Am J Epidemiol 1997;146:949-54.

Mendola P, Buck GM, Sever LE, Zielezny M, Vena JE. Consumption of PCB-
contaminated freshwater fish and shortened menstrual cycle length. Am J
Epidemiol 1997;146:955-60.

77

61.

62.

63.

64.

65.

66.

67.

68.

69.

70.

71.

72.

73.

Dar E, Kanarek MS, Anderson HA, Sonzogni WC. Fish consumption and
reproductive outcomes in Green Bay, Wisconsin. Environ Res 1992;59:189-201.

Rogan WJ, Gladen BC, McKinney JD, Carreras N, Hardy P, Thullen J, Tinglestad
J, M. T. Neonatal effects of transplacental exposure to PCBs and DDE. J Pediatr
1986;109:335-41.

Leoni V, F abiani L, Marinelli G, Puccetti G, Tarsitani GF, De Carolis A, Vescia N,
Morini A, Aleandri V, Pozzi V. PCB and other organochlorine compounds in blood

of women with or without miscarriage: a hypothesis of correlation. Ecotoxicol
Environ Safety 1989;17:1-11.

Bush B, Bennett AH, Snow JT. Polychlorobiphenyl congeners, p,p'-DDE, and
sperm function in humans. Arch Environ Contam Toxicol 1986;15:333-41.

Carlsen E, Giwercman A, Keiding N, Skakkebaek NE. Evidence for decreasing
quality of semen during past 50 years. Br Med J 1992;305:609-13.

Auger J, Kunstmann JM, Czyglik F, Jouannet P. Decline in semen quality among
fertile men in Paris during the past 20 years. N Engl J Med 1995;332:281-85.

Courval JM, DeHoog JV, Holzman CB, Tay EM, Fischer L, Humphrey HE, Paneth
NS, Sweeney AM. Fish consumption and other characteristics of reproductive-aged

Michigan anglers — a potential population for studying the effects of consumption of
Great Lakes fish on reproductive health. Toxicol Ind Health 1996;12:347-59.

DeHoog JV, Courval JM, Paneth N, Holzman C, Stein A, Sauer H, Mullard A,
Tripathi M, Cook D. Recruitment of males, females, and couples as participants into

a multi-outcome study of human reproductive health. Presented at the Great Lakes
Health Conference '97, Montreal, Quebec, Canada, May 12-15, 1997.

Kanuk L, Berenson C. Mail surveys and response rates: a literature review. J
Marketing Res 1975;12:440-53.

Linsky AS. Stimulating responses to mailed questionnaires: a review. Public
Opinion Q 1975;39:82-101.

Eastwood BJ, Gregor RD, MacLean DR, Wolf HK. Effects of recruitment strategy
on response rates and risk factor profile in two cardiovascular surveys. Int J
Epidemiol 1996;25:763-69.

National Academy of Sciences. Panel on privacy and confidentiality as factors in
survey response. Washington, DC: National Academy of Sciences, 1979.

Schleifer S. Trends in attitudes toward and participation in survey research. Public
Opinion Q 1986;50:17-26.

78

74.

75.

76.

77.

78.

79.

80.

81.

82.

83.

84.

85.

86.

87.

88.

Steeh CG. Trends in nonresponse rates, 1952-1979. Public Opinion Q 1981;45:40-
57.

Luevano P. Response rates in the national election studies, 1948-1992. Ann Arbor,
MI: Institute for Social Research, University of Michigan, 1994.

Wiseman F, McDonald P. Noncontact and refusal rates in consumer telephone
surveys. J Marketing Res 1979;16:478-84.

Wiseman F, Schafer M. Focus group interviews with survey non-respondents.
Presented at the Conference of American Association for Public Opinion Research,
1976.

Goyder J. The silent minority: nonrespondents in sample surveys. Cambridge:
Polity Press in association with Basil Blackwell, 1987.

Greenland S. Response and follow-up bias in cohort studies. Am J Epidemiol
1977;106:184-87.

Criqui MH. Response bias and risk ratios in epidemiologic studies. Am J
Epidemiol 1 979; 1 09:394-99.

Criqui MH, Austin M, Barrett Connor E. The effect of non-response on risk ratios
in a cardiovascular disease study. J Chron Dis 1979;32:633-38.

Greenland S, Criqui MH. Are case-control studies more vulnerable to response
bias? Am J Epidemiol 1981;114:175-77.

Austin MA, Criqui MH, Barrett Connor E, Holdbrook MJ. The effect of response
bias on the odds ratio. Am J Epidemiol 1981;114:137-43.

Sheikh K, Mattingly S. Investigating non-response bias in mail surveys. J
Epidemiol Community Health 1981;35:293-96.

Schlesselman JJ. Case control studies: design, conduct, and analysis. Oxford, New
York: Oxford UP, 1982.

Dillman DA. Mail and telephone surveys: the total design method. New York:
Wiley, 1978.

Church AH. Estimating the effect of incentives on mail survey response rates: a
meta-analysis. Public Opinion Q 1993;57:62-79.

Fox RJ, Crask MR, Kim J. Mail survey response rate: a meta-analysis of selected
techniques for inducing response. Public Opinion Q 1988;52:467-91.

79

89.

90.

91.

92.

93.

94.

95.

96.

97.

98.

99.

100.

101.

102.

Yammarino FJ, Skinner SJ, Childers TL. Understanding mail survey response
behavior: a meta-analysis. Public Opinion Q 1991;55:613-39.

Armstrong JS, Lusk EJ. Return postage in mail surveys: a meta-analysis. Public
Opinion Q 1987;51:233-48.

Asch DA, Christakis NA. Different response rates in a trial of two envelop styles in
mail survey research. Epidemiology 1994;5z364-65.

Campbell MJ, Waters WE. Does anonymity increase response rate in postal
questionnaire surveys about sensitive subjects? A randomised trial. J Epidemiol
Community Health 1990;44:75-76.

Camunas C, Alward RR, Vecchione E. Survey response rates to a professional
association mail questionnaire. J NY State Nurses Assoc 1990;2127-9.

Choi BC, Pak AW, Purdham JT. Effects of mailing strategies on response rate,

response time, and cost in a questionnaire study among nurses. Epidemiology
1990;1 :72-74.

Herzog AR, Dielman L. Age differences in response accuracy for factual survey
questions. J Gerontol 1985;40:350-57.

Kaplan S, Cole P. Factors affecting response to postal questionnaires. Br J Prev
Soc Med 1970;24:245-47.

Newland CA, Waters WE, Standford AP, Batchelor BG. A study of mail survey
method. Int J Epidemiol 1977;6265-67.

Rimm EB, Stampfer MJ, Colditz GA, Giovannucci E, Willett WC. Effectiveness of
various mailing strategies among nonrespondents in a prospective cohort study.
Am J Epidemiol 1990;131:1068-71.

Schillmoeller EA. Tackling declining response rates. J Advertising Res
1987;27:RC10-11.

Shiono PH, Klebanoff MA. The effect of two mailing strategies on the response to
a survey of physicians. Am J Epidemiol 1991;134:539-42.

Spry VM, Hovell MF, Sallis JG, Hofsteter CR, Elder JP, Molgaard CA. Recruiting
survey respondents to mailed surveys: controlled trials of incentives and prompts.
Am J Epidemiol 1989;130:166-72.

Bridge RG. Nonresponse bias in mail surveys: the case of the Department of
Defense post-service survey. Santa Monica: Rand, 1974.

80

103.

104.

105.

106.

107.

108.

109.

110.

111.

112.

113.

114.

115.

116.

Couper MP, Groves RM. “Household-level determinants of survey non-response.”
Advances in survey research. Ed. M.T. Braverrnan and J .K. Slater. San Francisco:
Jossey-Bass, 1996. 63-80.

Paganini-Hill A, Hsu G, Chao A, Ross RK. Comparison of early and late
respondents to a postal health survey questionnaire. Epidemiology 1993;4z375-79.

Siemiatycki J, Campbell S. Nonresponse bias and early versus all responders in
mail and telephone surveys. Am J Epidemiol 1984;120:291-301.

Seltzer CC, Bosse R, Garvey AJ. Mail survey response by smoking status. Am J
Epidemiol 1974;] 002453-57.

Kessler RC, Little RJ, Groves RM. Advances in strategies for minimizing and
adjusting for survey nonresponse. Epidemiol Rev 1995;17:192-204.

Goudy WJ. Nonresponse effects: studies of the failure of potential respondents to
reply to survey instruments. Monticello, IL: Council of Planning Librarians, 1977.

Hawkins DF. A bibliography of studies of nonresponse in survey research. Chapel
Hill, NC: Institute for Research in Social Science, University of North Carolina,
1978.

Daniel WW. Statistical adjustment for nonresponse in sample surveys: a selected
bibliography with annotations. Monticello, IL: Vance Bibliographies, 1979.

F inkler AL. Methods of sampling for estimating commercial peach production in
North Carolina. North Carolina Agricultural Experimental Station Technical
Bulletin 1950;91.

Stanton F. Notes on the validity of mail questionnaire returns. J Appl Psychol
1939;23:95-104.

Pace CR. Factors influencing questionnaire returns from former University
students. J Appl Psychol 1939;232388-97.

Edgerton HA, Britt SH, Norman RD. Objective differences among various types of
respondents to a mailed questionnaire. Am Soc Rev 1947;12:433-44.

Gorton T, Moore FE, Shurtleff D, Dawber TR. Some methodological problems in
the long-term study of cardiovascular disease: observations on The Framingham
Study. J Chron Dis 1959;10:186-206.

Criqui MH, Barrett Connor E, Austin M. Differences between respondents and
non-respondents in a population-based cardiovascular disease study. Am J
Epidemiol l 978; 1 08 2367-72.

81

117.

118.

119.

120.

121.

122.

123.

124.

125.

126.

127.

128.

129.

130.

Doll R, Hill AB. Mortality in relation to smoking: ten years’ observations of
British doctors. Br Med J 1964;] :1399-1410.

Macera CA, Jackson KL, Davis DR, Kronenfeld JJ, Blair SN. Patterns of non-
response to a mail survey. J Clin Epidemiol 1990;43:1427—30.

Paul O, Lepper MH, Phelan EH, Dupertuis GW, MacMillan A, McKean H, Park H.
A longitudinal study of coronary heart disease. Circulation 1963;28:20-31.

Oakes TW, Friedman GD, Seltzer CC. Mail survey response by health status of
smokers, nonsmokers, and ex-smokers. Am J Epidemiol 1973;98:50-55.

Brambilla DJ, McKinlay SM. A comparison of responses to mailed questionnaires
and telephone interviews in a mixed mode health survey. Am J Epidemiol
1987;126:962-71.

Walker M, Shaper AG, Cook DG. Non-participation and mortality in a prospective
study of cardiovascular disease. J Epidemiol Community Health 1987;41:295-99.

Jacobsen BK, Thelle DS. The Tromso Heart Study: responders and non-responders
to a health questionnaire, do they differ? Scand J Soc Med 1988;16:101-4.

Heilbrun LK, Nomura A, Stemmerrnann GN. The effects of non-response in a
prospective study of cancer: 15-year follow-up. Int J Epidemiol 1991 ;20:328-3 8.

Heilbrun LK, Nomura A, Stemmerrnann GN. The effects of non-response in a
prospective study of cancer. Am J Epidemiol 1982;116:353-63.

Benfante R, Reed D, MacLean C, Kagan A. Response bias in the Honolulu Heart
Program. Am J Epidemiol 1989;130:1088-1100.

Vestbo J, Rasmussen FV. Baseline characteristics are not sufficient indicators of
non-response bias follow up studies. J Epidemiol Community Health 1992;46:617-
19.

Bostrom G, Hallqvist J, Haglund BJ, Romelsjo A, Svanstrom L, Diderichsen F.
Socioeconomic difierences in smoking in an urban Swedish population. The bias
introduced by non-participation in a mailed questionnaire. Scand J Soc Med
1993;21 :77-82.

Locker D. Effects of non-response on estimates derived from an oral health survey
of older adults. Community Dent Oral Epidemiol 1993;21:108-13.

Melton LJ, Dyck PJ, Kames JL, PC OB, Service FJ. Non-response bias in studies
of diabetic complications: the Rochester Diabetic Neuropathy Study. J Clin
Epidemiol 1993;46:341-48.

82

131.

132.

133.

134.

135.
136.
137.

138.

139.

140.

141.

142.

143.

144.

Bisgard KM, Folsom AR, Hong CP, Sellers TA. Mortality and cancer rates in
nonrespondents to a prospective study of older women: 5-year follow-up. Am J
Epidemiol 1994;139:990-1000.

Panser LA, Chute CG, Guess HA, Larsonkeller JJ, Girrnan CJ, Oesterling JE,
Lieber MM, Jacobsen SJ. The natural history of prostatism: the effects of non-
response bias. Int J Epidemiol 1994;23:1198-205.

Papageorgiou AC, Croft PR, Ferry S, Jayson MIV, Silman AJ. Estimating the
prevalence of low back pain in the general population. Spine 195;2021889-94.

Jackson R, Chambless LE, Yang K, Byrne T, Watson R, Folsom A, Shahar E,
Kalsbeek W. Differences between respondents and nonrespondents in a multicenter
community-based study vary by gender ethnicity. The Atherosclerosis Risk in
Communities (ARIC) Study Investigators. J Clin Epidemiol 1996;49: 1441-46.

Centers for Disease Control. Diskette. Atlanta, GA: Epi Info Rel 6.02, 1994.
Microsoft Corporation. CD-ROM. Redmond, WA: Microsoft Office Re] 95, 1995.
Prophone. CD-ROM. Danvers, MA: Prophone, 1992.

Centers for Disease Control. Behavioral Risk Factor Surveillance System. Atlanta,
GA: Centers for Disease Control, 1995.

SPSS Inc. Diskette. Chicago, IL: SPSS for Windows Rel 6.1.3, 1995.

Dillman DA, Sangster RL, Tamai J, Rockwood TH. “Understanding differences in
people’s answers to telephone and mail surveys.” Advances in survey research. Ed.
M.T. Braverman and J .K. Slater. San Francisco: Jossey-Bass, 1996. 45-61.

McHorney CA, Kosinski M, Ware JEJ. Comparisons of the costs and quality of
norms for the SF-36 health survey collected by mail versus telephone interview:
results from a national survey. Med Care 1994;32:551-67.

Siemiatycki J. A comparison of mail, telephone, and home interview strategies for
household health surveys. Am J Public Health 1979;69:23 8-45.

Siemiatycki J, Campbell S, Richardson L, Aubert D. Quality of response in
different population groups in mail and telephone surveys. Am J Epidemiol
1984;120:302-14.

O’Toole BI, Battistutta D, Long A, Crouch K. A comparison of costs and data
quality of three health survey methods: mail, telephone and personal home
interview. Am J Epidemiol 1986;124:317—28.

83

"11111111111'111'11111’