KEY WORDS: Attrition, longitudinal survey, registry, response rates
RUNNING HEAD: Examination of attrition, nonresponse and bias
ABSTRACT
This study examined attrition in the Trichloroethylene (TCE) Subregistry of the National Exposure Registry. The analyses focused on 3,915 persons exposed to the chemical trichloroethylene through the drinking water in their home. Baseline data were compared for subgroups of the TCE Subregistry members who were eligible to participate in the first TCE Subregistry followup. Study members were grouped according to their participation status in the first followup: remainers (n = 3,494) and losses (n = 421), and three subgroups of losses: refusals, unable to locate, and unable to contact. The comparison of demographic variables of remainers and losses revealed that remainers had a higher percent of females, currently smoked less, were older, and fewer had no education and more had education beyond high school. These differences occurred for the losses subgroups unable to locate and unable to contact, however, not for refusals. The comparison of reporting rates of remainers and losses for 23 health outcomes revealed statistically significant decreases by losses for five health conditions but the pattern of statistically significant differences for the losses subgroups was not clear-cut. Altogether, the analyses indicated that the potential for bias due to attrition was minimal.
INTRODUCTION
Longitudinal surveys offer one way to obtain information about the health of human subjects with documented exposure(s) to hazardous substances in the environment. The validity of longitudinal survey data is increasingly a salient concern, however, given the expanding application of such information into diverse contexts. In the health sciences, there is a need for valid information about the potential adverse health effects for the general population, particularly those who have experienced long-term, low-level exposures to hazardous substances in the environment. For the longitudinal studies that might provide such human information, the validity of the data might have a direct effect on the calculation of odds ratios and magnitude of risk factors (Austin et al., 1981; Greenland, 1977). In the U.S. legal system, there is an increasing use of quality survey data in all types of litigation (Crespi, 1987), including the rising volume of personal injury cases which allege harm (such as, ill health) as a result of environmental exposure (Rosenkranz, 1984).
Regardless of substantive focus, the validity of longitudinal survey data can be threatened by several factors, including the dimensions of attrition: the response rate, the specific reasons for nonresponse, and the characteristics of nonrespondents (Cooney, et al., 1988). Provided data are sufficient, it is important to examine each dimension in order to determine the potential extent and nature of nonresponse bias and to develop profiles of nonrespondents. The extent of nonresponse bias is a direct indication of the external validity of survey data (Cooney, et al., 1988; Campbell and Stanley, 1963) and is apparent when remaining members of the population are unrepresentative of the population. Nonrespondent profiles are useful to the extent that types of study losses can be anticipated and knowledge of their characteristics used in the allocation of study resources intended to reduce attrition (Liu and Anthony, 1989).
Many previous studies of attrition bias in longitudinal studies and nonresponse bias in cross-sectional studies have used the nonresponse rate as the primary indication of bias in sample estimates and multivariate patterns. Bias is, however, a more complex phenomena (Cooney, et al., 1988; Launer, et al., 1994). A study, even through it has a high response rate, might yield biased results if statistically significant differences exist between respondents and nonrespondents on variables of interest. In contrast, studies with relatively low response rates might not evidence nonresponse bias if the respondents and nonrespondents share similar attributes on the risk and outcome measures of interest (Launer, et al., 1994). Further, the effects of attrition might be associated with only some, not all, the subgroups of losses.
Most profiles of study losses are derived from comparisons between two subgroups from within the original study sample: 1) subjects who have remained in the study and 2) subjects who have been lost to followup for any reason. Findings from studies suggest that respondent loss in longitudinal and cross-sectional studies is not random (Cooney, et al., 1988; Cordray, 1983) and relates to many individual characteristics, including household size (Cobb, et al., 1957), educational attainment (Cobb, et al., 1957), age (Cobb, et al., 1957; Herzog and Rodgers, 1988; Vernon, et al., 1984), sex (Vernon, et al., 1984), smoking behavior (Criqui, et al., 1978), deviant behavior (Cordray and Polk, 1983), and physical and mental health (Cooney, et al., 1988; Cobb, et al., Vernon, et al., 1984-15), just to name a few. Unfortunately, the evidence from these studies of losses is often inconsistent. For example, some studies report higher rates of self-reported illness among losses (for example, Criqui et al (Criqui, et al., 1978), Wilhelmsen et al (Wilhelmsen, et al., 1976)) while others find no differences in self-reported health status (for example, Vernon et al. (Vernon, et al., 1984)).
Few studies when examining losses in longitudinal surveys consider the multiple aspects of attrition over time within the context of a single longitudinal study. For example, most studies of attrition do not consider how specific reasons for nonresponse relate to individual characteristics. The findings from those which consider the nuances of attrition (Cooney, et al., 1988; Vernon, et al., 1984; Goyder, 1987) merit further consideration. These more comprehensive studies of attrition suggest that losses in longitudinal studies are best perceived as a heterogeneous group (Vernon, et al., 1984); not all subgroups of losses contribute equally to the potential for bias due to attrition. For example, Cooney et al. (Cooney, et al., 1988)found that about 27.0 percent of all losses were refusals, 47.5 percent were deaths, and the remaining 25.0 percent were lost because of ill health. Study members differed from study losses due to ill health, but not the other subgroups of study losses. A more generic assessment of nonrespondents (that is, members compared with all losses) would not have identified these nuances of study member and loss similarities (and differences). The study reported here builds upon this small, yet growing literature which considers the multiple dimensions of attrition within the context of a longitudinal study.
DATA AND METHODS
Data Source
Data for this study come from the Trichloroethylene (TCE) Subregistry of the National Exposure Registry (NER). The National Exposure Registry is sponsored and maintained by the Agency for Toxic Substances and Disease Registry (ATSDR). The TCE Subregistry, a longitudinal survey, comprises 4,927 people with documented TCE exposure from 15 exposure sites located in 5 states (IL, IN, MI, AZ, and PA). This study uses baseline interview data from only 13 of these 15 sites, the coded data from 2 sites (AZ and PA) were not available at the time of this writing. At the time of the baseline interview, the TCE Subregistry members resided in 35 states, with the greatest concentration (90 percent) living in the eastern North Central region (that is, IL, IN, MI, OH, and WI). See Gist (Gist, et al., 1994) for more information about the methods and criteria used for selecting sites for inclusion in the NER.
At each site a listing is compiled of site addresses (that is, private homes) where exposure to TCE was documented. All TCE Subregistry members have documented exposure (Agency for Toxic Substances and Disease Registry, 1994) to TCE and other related contaminants found in their drinking water. Levels of exposure range from .01 to 19,308 parts per billion. The duration of exposure ranges from 1 month to 31 years, and time since exposure ranges from ongoing exposure at the time of the baseline interview to seven years prior to the baseline interview (Agency for Toxic Substances and Disease Registry, 1994). One criterion for inclusion in the NER is each respondent must have lived at an exposure site address for at least 30 consecutive days during the exposure period and during this time used the water for drinking, bathing, or cooking.
Baseline data and activities. For each of the 15 TCE sites, a complete census of homes was conducted by ATSDR to identify dwellings with validated TCE contamination of the drinking water (in all cases, the source was a private well). ATSDR attempted to identify all of the people who had ever lived at the site addresses during the exposure period, including current and previous residents. This effort included obtaining death certificate information for people who were exposed to TCE at a site address during the exposure period, but died some time prior to the date of the baseline interview (n = 240).
The baseline interviews were conducted face-to-face (except for people who had moved off-site where computer-assisted telephone interviewing, CATI, was used) and included questions about participants' duration of exposure, current and previous residences, demographic characteristics, employment status, occupation, use of tobacco products, and self-reported health outcomes. The health questions focused on 5 symptoms and 18 health problems that might have occurred from the point of birth through the date of the baseline interview. Respondents were asked to report only those symptoms or conditions that had been diagnosed by a health care provider. A knowledgeable parent or guardian was used as a proxy for subjects who were 17 years of age(n = 948, 23.5 percent). A knowledgeable proxy was used for adults who were incapacitated due to illness or other disabilities at the time of the interview (n = 94, 2.3 percent).
The participation rate for those people who were contacted and eligible to take part was very high, with little variation across the 13 exposure sites (98 percent to 100 percent).
Follow-Up data and activities. Six months after the baseline interview, a letter was mailed to confirm each registrant's listing on the TCE Subregistry asking him or her to review the spelling of his or her name, mailing address, and telephone number. One year after the baseline interview, each registrant was interviewed by telephone. The follow-up interview included a modified version of the baseline interview: questions focused on a first-time diagnosis of or continued treatment for any of the same 23 health outcomes (5 symptoms and 18 conditions), changes in employment status, occupation, and educational attainment. The follow-up interviews will continue on a biennial basis, until a decision is made to terminate the data collection effort (Agency for Toxic Substances and Disease Registry, 1994). To date, TCE Subregistry members have been involved in at least two but no more than five biennial follow-up interviews. The number of follow-up interviews depends on the date when a given exposure site was added to the TCE Subregistry. Eligibility for participation at each follow-up interview is contingent upon participation in all previous interviews.
Several methods were used to locate people for the follow-up interviews. When a valid telephone number existed from the baseline interview, that number was called for a follow-up interview. If the telephone number was invalid, multiple resources were used to obtain a valid number including directory assistance and references (three contacts requested at each interview, usually friends or relatives who did not live with participant) provided by participant (or proxy) during the baseline interview. If these sources failed, ATSDR used a private company that specialized in locating people.
Sample Population
This study included 3,915 members of the TCE Subregistry who reported their race as white, were alive at the time of the baseline survey, and had completed the baseline interview (either personally or by proxy). Nonwhite subregistry members (n = 126, 3.1 percent) were excluded from the analyses because of the great diversity of races (eight different races were reported) within the small total number of nonwhites.
Methods of Analysis
Similar to Cordray and Polk (Condray and Polk, 1983), the study reported here compares two main groups that were eligible for participation in the first followup interview of the TCE Subregistry registrants: the remainers (or participants) and losses. Baseline data on the remainers are compared with the baseline data for the losses. Next, the losses are subdivided for further comparisons with the remainers.
For the purposes of this study,
Remainers (n = 3,471) include the registrants who completed an interview (personally or by proxy) for the first followup. Remainers exclude registrants who died prior to (n = 240) or some time after (n = 23) the baseline interview.
Decedents (n = 23) are the registrants who were alive and participated in the baseline, but died some time prior to the first follow-up interview.
Losses (n= 421) include all registrants who dropped out of the subregistry after a baseline interview for any reason other than death. Within the losses category, three subgroups are identified for purposes of comparison:
(a) refusals (n = 153), or registrants who refused participation (for self or proxy) in the followup;
(b) unable to locates (UTLs) (n = 113), registrants that could not be located after extensive tracing efforts; and
(c) unable to contacts (UTCs) (n = 147), registrants that could not be contacted for interview, although ATSDR had located them.
These three subgroups account for 418 of the losses. The 3 remaining losses are excluded from comparisons of remainers with loss subgroups, but included in the comparisons of remainers with losses. These 3 subjects were lost to followup for reasons other than above; for example, language barrier and no one to translate.
Three statistical tests were used for group comparisons of the different types of variables. The Fisher's Exact test (2-tail) was used for comparing dichotomous variables (onsite, sex, race, and all health outcomes). For the categorical variables (age, educational attainment, and cigarette use), the Cochran-Mantel-Haenszel (CMH) test was used. The Student's t-test was used for continuous variables (educational attainment and age) comparisons. The analysis of education was restricted to subregistry members who were at least 18 years of age at the time of the baseline survey.
Measures: Demographic and Health
All respondent demographic and health data that are analyzed in this study are derived from the baseline interviews with participants(or their designated proxies). Measures include sex, age (date of birth), educational attainment (0-17+ years), cigarette smoking status (never smoked = smoked 100 or fewer cigarettes in a lifetime, current smoker = smoked 100 or more cigarettes and smokes now, ex-smoker = have smoked at least 100 cigarettes in a lifetime but does not smoke now), number of months subject had (for ex-smokers) or has smoked (for current smokers) cigarettes, and place of residence at the time of the baseline survey. Residency is a dichotomous variable with "onsite" defined as a registrant who lived in the same state as his or her exposure site at the time of the baseline interview. An "offsite" registrant had moved from the state of his or her exposure site some time prior to the baseline interview.
Baseline data on health were derived from questions about the health symptoms and problems experienced by registrants. For each health condition, registrants were asked two questions. The stem of the first question was, "Has a physician or other medical provider ever told you that you had or treated you for .....?". If the answer to the that question was "yes", the second question was asked, "Were you ever treated for this condition?" The specific health symptoms and conditions are listed and defined in Table 1; they cover a wide range of health outcomes. In subsequent followup interviews, the stem questions were rewritten to ask about the time since the previous interview rather than the "ever" timeframe queried for the baseline survey.
RESULTS
The overall nonresponse rates for the baseline survey (2.0 percent, all 13 sites combined) and the first followup (10.8 percent) were very low. Table 2 presents the baseline demographic characteristics of the remainers, losses, and the losses categories of refusals, unable to locates (UTLs), and unable to contacts (UTCs). Statistical test results for the remainers versus losses comparison are indicated in the column for losses. Statistical test results for the remainers versus refusals comparison are indicated in the column for refusals, and likewise for the comparison of remainers with the other subgroups of losses.
The remainers and losses, as identified at the first followup, differed on demographic variables as reported at baseline, except for the place of residence. Females were underrepresented among the losses. Losses were, on the average, about 4 years of age younger than remainers (34.5 versus 30.4 years). Losses and remainers had the same level of education, that is, the percent with at least a high school education was about 89 percent for each group. However, it is of note that for the losses, 1.0 percent had no education and 13.9 percent had a college degree or higher versus 0.3 percent reported no education and 26.8 percent with a college degree or higher for the remainers. Losses had a higher rate of current cigarette use (42.0 percent for losses versus 30.4 percent for remainers). None of the current or former smokers in the comparison groups (remainers, losses, refusals, UTLs, and UTCs) differed, however, with regard to the number of months that they have smoked cigarettes. Losses were no more or less likely to have lived on or offsite at the time of the baseline interview. These differences between the losses and remainers were characteristic of differences between the remainers and the UTCs and UTLs, but not the refusals--who represent the largest category of losses. With the exception of the percentage of female, refusals were like remainers on all measures of demographic variables. Hence, the statistically significant differences noted between the remainers and losses typify the differences between remainers and about 62 percent of the losses (UTCs and UTLs) at first followup.
Next, remainers were compared with losses and the subgroups of losses on reporting rates for symptoms and health conditions (Table 3). As before, statistical test results for the comparison of remainers with losses are indicated in the column for losses. Statistical test results for the remainers versus refusals comparison are indicated in the column for refusals, and likewise for the comparison of remainers with the other subgroups of losses. The statistical tests which compare remainers with losses are controlled for sex and age. It was not possible to control for sex in the statistical tests that compare remainers with the subgroups of losses.
Table 3 shows that the losses were neither completely different from nor similar to the remainers in terms of reported health outcomes. With regard to symptoms, the pattern shows higher reporting rates for remainers, although none reached statistical significance. However, the highest reporting rates for all groups, including remainers, were by the losses UTCs subgroup. When losses were compared with remainers for the 18 health conditions, the reporting rates for losses were statistically significantly decreased for 5 conditions (hypertension, 8.8 percent losses versus 15.1 percent remainers; ulcers, 15.9 percent versus 23.7 percent; allergy, 13.1 percent versus 18.4 percent; heart problems, 6.6 percent versus 10.4 percent; and rash, 17.6 percent versus 24.4 percent). The comparison of the remainers with each subgroup of losses revealed that the reporting rates was lower, in some cases statistically significantly lower, for each losses subgroup for these same health outcomes. There were, however, increased reporting rates (but not statistically significant) by losses for some health conditions (for example, anemia and stroke).
Overall, these comparisons indicated that losses at the first followup differed from remainers for some demographic characteristics and some health conditions. Losses subgroups showed between-subgroup homogeneity in reporting rates in terms of health outcomes but were heterogeneous in terms of demographic characteristics.
DISCUSSION
The overall nonresponse rates for the baseline survey (2.0 percent, all 13 sites combined) and the first followup (10.8 percent) were very low compared with the nonresponse rates for other longitudinal studies of health. For example, in a study of psychiatric health among the general population, Cottler et al. (Cottler, et al., 1987) reported 20 percent nonparticipation for the first followup interview. Liu and Anthony (Liu and Anthony, 1989) reported 22 percent and 30 percent nonparticipation rates in the first and second followup interviews, respectively, in a study about mental health. Eaton et al. (Easton, et al., 1992) reported a nonresponse rate of 20.1 percent at first followup for the National Institutes of Mental Health Epidemiologic Catchment Area Program survey. The high participation rate precludes participation bias being a factor in attrition and attrition bias.
Similar to Criqui et al., (Criqui, et al., 1978) the analyses of demographic characteristics revealed that losses included a higher percentage of males, a younger mean age, a lower level of educational attainment (in terms of beyond high school), and a higher percentage of current cigarette smokers compared with remainers.
The comparison of the remainers with the losses on self-reported rates of health outcomes revealed some (5 out of 25 comparisons were statistically significantly increased) differences as well. However, the nature of these differences was unlike the pattern of differences reported elsewhere (e.g., Criqui et al. (Criqui, et al., 1978), Vernon et al. (Veron, et al., 1984), Wilhelmsen et al.(Wilhelmsen, et al., 1976)), where higher rates of illness were reported among study losses. In this study, lower rates of health problems were found among losses when differences with remainers reached statistical significance.
The lower rates of illness among losses might be related, in part, to the registrants' misunderstanding of the subregistry's objectives and insufficient clarity by ATSDR in presenting the subregistry purposes. Information on refusals (maintained in ATSDR study documents) revealed that a small percentage of the refusals declined participation in the follow-up interview because: 1) the registrant had no health problems, or 2) the registrant did not have the type of health problem(s) that he or she believed were of interest to the subregistry (for example, cancer and skin rashes). To the contrary, the subregistry was designed to maintain demographic and health information on all eligible people, not merely those with no self-reported health problems.
The comparisons of remainers with the subgroups of losses revealed two related and noteworthy findings which are supported by other studies: (1) refusals were similar to remainers, and (2) losses were a heterogeneous group. Consistent with the first finding, Cooney et al. (Cooney, et al., 1988) and Goyder (Goyder, 1987) found that refusals were similar to remainers in terms of demographic characteristics and many reported health outcomes. Criqui et al. ( Criqui, et al., 1978) found that cigarette smokers, younger people, and males were overrepresented among all subgroups of losses except refusals.
With regard to the second finding--losses were a heterogenous group, the differences observed within the losses tend to support studies (for example, Cooney et al. (Cooney, el al., 1988), Goyder (Goyder, 1987), Vernon et al. (Veron, et al., 1984)) which also reveal that the reason for nonresponse is correlated with demographic and other individual characteristics. Although these studies consider variables not examined in this study (i.e., race), the pattern of their findings is consistent with those found here. For example, in a study of psychiatric health, Vernon et al. (Veron, et al., 1984) found that race is related to the reason for nonresponse. Whites were less likely to refuse participation than Hispanics and blacks. Whites were, however, more likely to be lost to followup by breaking off the interview or not being available during the study period. Cooney et al. (Cooney, et al., 1988) found the greatest contrast between people who refused and other nonrespondents. Refusals (who resembled study participants) differed from people lost for other reasons in terms of age, education, and selected cognitive abilities (for example, word fluency). Likewise, Goyder (Goyder, 1987) found that refusals tended to have different characteristics than people lost to followup for other reasons.
If the patterns described previously hold, they suggest relatively favorable prospects for future TCE Subregistry follow-up studies. One such pattern is that the comparisons of health outcomes across all groups and subgroups indicated that remainers were not unequivocally sicker (as judged by positive reporting of health outcomes) than study losses. A second pattern is that the largest group of study losses (refusals, 37.8 percent of all losses, 4.0 percent of all persons eligible for participation in the followup) were very similar to the remainers in the health and demographic comparisons. Although the UTCs and UTLs subgroups differ from remainers, it is possible that their representation among the losses will diminish over time. For example, many people (n = 42) in the UTC subgroup could not be contacted because they lived offsite and had no telephone. Although ATSDR mailed them a toll-free number which would allow them to take part in the followup interview and maintain subregistry membership, few took advantage of this option and were therefore lost to future followup studies.
Further, these findings also have at least three direct implications for future NER activities and perhaps other longitudinal studies of health where members include people with a risk factor. First, from the point of baseline, it is important to maintain information on the reasons for nonparticipation (refusal, unable to locate, language barrier, and the like), as well as documentation on the reasons why people refuse to participate. Such information is critical for the effective monitoring of ongoing data collection activities. The predominant reasons for refusal might change from one data collection period to another. At followup one, the main reason for refusal might include a misunderstanding of ATSDR's interest in all subjects regardless of their health status. At the next followup, the main reason for refusal might be the perception of excessive burden among proxies who must answer all questions for themselves, as well as their dependents who are also subregistry members. Both reasons can be dealt with (and have been accordingly) only to the extent that they are known.
Second, study members should be reminded regularly of their potential contribution to study objectives. Future baseline and follow-up communications with potential and existing TCE Subregistry participants should highlight the irrelevance of subject health with regard to subregistry eligibility.
And last, because losses tended to be younger, male, or have lower educational levels, it might be useful to target this type of participant at the baseline interview and, if possible, gather more complete locating information ( for example, more contacts) from this subgroup that is more prone to attrition. Also, this subgroup of losses might benefit from more frequent or a different kind of contact (such as, telephone calls for address updates or postcards) between surveys in order to help participant maintain interest and an up-to-date record of current addresses.
As suggested by others (e.g., Groves and Lyberg (Groves and Lyberg, 1988) , VonKorff et al. (VonKorff, et al., 1985)), the demonstration of nonresponse bias (or the lack of it) appears to be contingent upon a full assessment of the reasons for study losses. In particular, are needs to assess the relative size of each type of loss, and the way each loss category compares with the remainers in terms of demographic and other relevant characteristics. Although the impact of attrition might be a potential threat to external validity, the findings of this study support Goudy ( Goudy, 1976), who argues that researchers tend to exaggerate the role of nonresponse per se in data quality.
TABLE 1. Question Wording for Symptoms and Health Conditions
| SYMPTOMS | WORDING |
|---|---|
| Fatigue | Frequent periods of fatigue or tiredness |
| Nausea | Frequent periods of nausea |
| Numbness | Weakness, paralysis, or numbness in the arms, hands, legs, or feet |
| Anxiety | Frequent periods of anxiety, nervousness, or depression |
| Headaches | Frequent or severe headache |
| HEALTH CONDITIONS | |
| Cancer | (specify type and site) |
| Hypertension | High blood pressure |
| Anemia | Anemia or other blood disorders |
| Kidney | Kidney disease |
| Urinary | Urinary tract disorders (including prostate trouble) |
| Stroke | Effects of a stroke |
| Seizures | Seizures, tremors, spells, or epilepsy |
| Ulcers | Ulcers, gall bladder trouble, stomach or
intestinal problems |
| Liver | Liver problems |
| Asthma | Asthma, emphysema, or chronic bronchitis |
| Allergy | Other respiratory allergies or problems such as
hay fever |
| Diabetes | Diabetes |
| Arthritis | Arthritis, rheumatism, or other joint disorders |
| Heart problems | Rheumatic fever, heart disease, or other heart problems |
| Speech Impairment | Speech impairment |
| Hearing Impairment | Hearing impairment |
| Rash | Skin rashes, eczema, or other skin allergies |
| Mental retardation | Mental retardation |
TABLE 2. Summary of Demographic Characteristics and Comparison Statistics
| GROUPS | |||||
|---|---|---|---|---|---|
| Remainers | Losses | Losses Subgroups | |||
| Refusals | Unable to
Locates - UTLs |
Unable to
Contacts - UTCs |
|||
| NUMBER
(% of Baseline) |
3,471
(89.2%) |
421†
(10.8%) |
158
(4.0%) |
113
(2.9%) |
147
(3.8%) |
| SEX RATIO | |||||
| % Female | 52.9%§ | 45.8%*** | 49.2%* | 51.3% | 43.5%* |
| AGE (years)> | |||||
| 0-9
10-17 18-24 25-34 35-44 45-54 55-64 65+ |
9.6%
13.2 11.0 21.3 16.6 10.2 8.3 9.6 |
12.4%***
14.5 23.0 13.8 7.6 6.9 5.9 |
10.1%
17.1 8.2 13.3 18.4 10.8 11.4 10.8 |
11.5%***
20.4 16.8 25.7 10.6 6.2 6.2 2.6 |
15.6%***
7.5 22.4 32.0 10.9 5.4 2.7 3.4 |
| CIGARETTE SMOKING (18+ years of age) | |||||
| Current
Former Never |
30.4%
16.6 53.0 |
42.0%***
11.8 46.1 |
31.0%
13.9 55.1 |
49.6%**
8.0 42.4 |
48.3%***
12.2 39.4 |
| PLACE OF RESIDENCE | |||||
| % offsite | 15.1% | 16.6% | 14.6% | 20.4% | 16.3% |
| EDUCATION LEVEL (19+ years of age) | |||||
| None
Elementary High school College Graduate |
0.3%
9.4 63.4 23.5 3.3 |
1.0%**
9.8 74.9 11.9 2.0 |
-----
10.9 68.2 14.5 5.4 |
2.7**
6.8 82.4 8.1 ---- |
0.9**
9.2 77.9 11.9 ---- |
| EDUCATION LEVEL (years) | |||||
| Mean
(sd) |
11.8
(2.4) |
11.1
(2.4) |
11.8
(2.3) |
10.7***
(2.6) |
10.9***
(2.3) |
| AGE (years) | |||||
| Mean
(sd) |
34.5
(19.9) |
30.4***
(18.9) |
35.9
(21.3) |
26.9***
(16.8) |
27.2***
(16.1) |
† - 3 cases included in the Losses category are cases that were lost because of a
reason other than refusal, inability to locate or contact (i.e., language barrier)
§ - Percent of group or subgroup
*** - indicates different from remainers at p < .01
** - indicates different from remainers at P < .05
* - indicates different from remainers at p < .10
TABLE 3.Comparison of Baseline Health Data According to Follow-up 1 Participation
| GROUPS | |||||
|---|---|---|---|---|---|
| Remainers
(n = 3,471) |
Losses†
(n = 421) |
Losses Subgroups | |||
| Refusals
(n = 158) |
Unable to
Locates UTLs (n= 113) |
Unable to
Contacts UTCs (n = 147) |
|||
| SYMPTOMS | |||||
| Fatigue
Nausea Numbness Anxiety Headaches |
12.0%§
6.4 12.6 16.9 18.8 |
10.5%
6.2 8.6 16.4 16.9 |
9.5%
3.5 7.6 15.8 15.8 |
8.8%
4.4 7.1 13.3 14.2 |
12.9%
10.2 10.9 19.7 20.4 |
| HEALTH CONDITIONS | |||||
| Cancer (1)
Cancer (2) Hypertension Anemia Kidney Urinary Stroke Seizures Ulcers Liver Asthma Allergy Diabetes Arthritis Heart problems Speech impairment Hearing impairment Rash Mental retardation |
4.2%
11.6 15.1 9.7 5.0 16.0 1.6 3.6 23.7 2.2 15.5 18.4 4.2 19.6 10.4 3.1 10.1 24.2 0.6 |
3.3%
7.1 8.8** 10.9 4.5 13.3 2.6 4.0 15.9** 1.9 15.2 13.1** 2.1 15.4 6.6** 3.6 10.9 17.7** 0.2 |
5.1%
12.5 10.8 8.2 3.2 10.8 2.5 4.4 16.4** 1.3 12.6 11.4** 1.3 16.4 6.3** 3.8 15.2 16.4** ---- |
1.8%
---- 5.3** 8.0 4.4 15.0 3.5 1.8 18.6 1.8 13.3 9.7** 2.6 14.2 6.2 3.5 6.2 15.0** ---- |
2.7%
---- 9.5** 16.3 6.1 15.0 2.0 5.4 13.6** 2.7 19.7 17.7 2.7 15.6 7.5 3.4 10.2 21.1 0.7 |
† - 3 cases included in the Losses category are cases that were lost because of a
reason other than refusal, inability to locate or contact (i.e., language barrier)
§ - Percent of group or subgroup
*** - indicates different from remainers at p < .01
** - indicates different from remainers at P < .05
* - indicates different from remainers at p < .10
ACKNOWLEDGMENTS
This research was supported in part by the appointment of Dr. Sarah L. Allred to the Research Participation Program at the Agency for Toxic Substances and Disease Registry (ATSDR). The program is administered by the Oak Ridge Institute for Science and Education through an interagency agreement between the U.S. Department of Energy and ATSDR.
REFERENCES
Agency for Toxic Substances and Disease Registry. (1988). National Exposure Registry: Policies and procedures manual (revised). U.S. Department of Health and Human Services, Public Health Service, Atlanta.
Agency for Toxic Substances and Disease Registry. (1994). National Exposure Registry Trichloroethylene (TCE) Subregistry baseline report. (Revised).1994. PB95-154571. U.S. Department of Health and Human Services, Public Health Service, Atlanta.
AUSTIN, M., CRIQUI, M., BARRETT-CONNER, E., and HOLDBROOK, M. (1981). "The effect of response bias on the odds ratio." Am. J. Epidem. 114:137-143.
CAMPBELL, D.T. and STANLEY, J.C. (1963). Experimental and quasi-experimental designs for research. Rand McNally, Chicago, IL. p. 6
COBB, S., KING, S., and CHEN E. (1957). "Differences between respondents and nonrespondents in a morbidity survey involving clinical examination." J Chr Dis 6:95-108.
COONEY, T., SCHAIE, K., and WILLIS S. (1988). "The relationship between prior functioning on cognitive and personality dimensions and subject attrition in longitudinal research." J. Gerontol 43:12-17.
CORDRAY, S. and POLK, K. (1983). "The implications of respondent loss in panel studies of deviant behavior." J Res on Crime and Delinquency 20:214-242.
COTTLER, L., ZIPP, J., ROBINS, L., and SPITZANGEL, E. (1987). "Difficult-to-recruit respondents and their effect on prevalence estimates in an epidemiological survey." Am. J. Epidem.125:329-339.
CRESPI, I. (1987). "Surveys as legal evidence." Public Opinion Quart 51:84-91.
CRIQUI, M., BARRETT-CONNER, E., and AUSTIN, M. (1978). "Differences between respondents and non-respondents in a population-based cardiovascular disease study." Am. J. Epidem. 108:367-372.
EATON, W.W., ANTHONG, J.C., TEPPER, S., and DRYMAN, A. (1992). "Psychopathology and attrition in the epidemiologic catchment area surveys." Am. J. Epidem. 135:9, 1051-1171.
GIST, G.L., BURG, J.R., and RADTKE, T.M. (1994). "The site selection process for the National Exposure Registry." J Envir Health 56(6):7-12.
GOUDY, W.J. (1976). "Nonresponse effects on relationships between variables." Public Opinion Quart 40:360-369.
GOYDER, J. (1987) The Silent Minority. Westview Press, Boulder, CO.
GREENLAND, S. (1977). "Response and follow-up bias in cohort studies." Am. J. Epidem. 106:184-187.
GROVES, R., and LYBERG, L. (1988). "An overview of nonresponse issues in telephone surveys." In: Telephone survey methodology (Groves, R., and Beimer, P., eds.). John Wiley & Sons, New York. pp.191-211
HERZOG, A., and ROGERS, W. (1988). "Age and response rates to interview sample surveys." J Gerontol 43:200-205.
LAUNER, L., WIND, A., and DEEG. D. (1994). "Nonresponse pattern and bias in a community based cross-sectional study of cognition functioning." Am. J. Epidem. 139:803-812.
LIU, I. and ANTHONY, J. (1989). "Using the "mini-mental state" examination to predict elderly subjects' completion of a follow-up interview." Am. J. Epidem. 416-426.
ROSENKRANZ, E.J. (1984). " The pollution exclusion clause through the looking glass." The Georgetown Law Journal 74:1237-1300
VERNON, S., ROBERTS, R., and EUN, S.L. (1984). "Ethnic status and participation in longitudinal health surveys." Am. J. Epidem. 119:99-113.
VONKORFF, M., COTTLER, L., GEORGE, L., EATON, W., LEAF, P., and BURNAM, A. (1985). Nonresponse and nonresponse bias in the ECA surveys. In: Eaton W, Kessler L, editors. Epidemiological field methods in psychiatry: the NIMH Epidemiologic Catchment Area Program. Academic Press, New York, NY.
WILHELMSEN, L., LJUNBERG, S., WEDEL, H., and WËRKO, H. (1976). "A comparison between participants and non-participants in a primary preventive trial." J Chr Dis 29:331-339.
Agency for Toxic Substances and Disease Registry, 1825 Century Blvd, Atlanta, GA 30345
Contact CDC: 800-232-4636 / TTY: 888-232-6348
