Participation Bias in AIDS-Related Telephone Surveys: Results From the National AIDS Behavioral Survey (NABS) Non-Response Study.

Survey research continues to be the most important methodology for examining behavioral issues related to the AIDS epidemic. Surveys are central to obtaining information on the epidemiology of AIDS, the distribution of behaviors that spread the disease, the factors and conditions that influence high risk activities, and the effectiveness of programs targeted at AIDS prevention. The usefulness of AIDS-related surveys depends on how successfully relevant populations can be sampled, and the reliability and validity of the information obtained from self-reports. Unfortunately, survey research in general has been plagued with numerous methodological difficulties that can seriously jeopardize the quality of the data. The fact that the information being sought in AIDS-relevant surveys is often highly sensitive, involving the need to obtain histories of sexual behavior and drug use, likely accentuates methodological problems. One particularly important issue involves participation bias. Given that, in our culture, sex is viewed as a highly private activity, people often feel embarrassed or threatened when asked to talk about their sexual behavior or attitudes with a stranger (Catania, McDermott, & Pollack, 1986). This tendency to avoid discussing sex, together with the potential stigma attached to the AIDS epidemic, may increase rates of refusal and exacerbate participation bias in AIDS behavioral surveys (Gagnon, 1988; Catania, Gibson, Marin, Coates, & Greenblatt, 1990).

The use of phone surveys to collect AIDS-relevant data is becoming an increasingly popular alternative to face-to-face interviewing, especially given the new developments in sampling and data collection technologies and the typically lower costs of collecting data by phone (Miller, Turner, & Moses, 1990). Moreover, the urgent need to move rapidly in collecting data that maps the course of the epidemic, the prevalence and antecedents of behaviors involved in transmitting HIV, and the success of intervention efforts (Catania, Gibson, Chitwood, & Coates, 1990) suggests that the expedience of phone surveys may be particularly advantageous. However, despite its potential advantages, telephone surveys often have lower overall response rates in comparison to face-to-face interviews (Groves & Kahn, 1979). In a meta-analysis of the two types of surveys, face-to-face interviews had a mean response rate of 75 percent, while telephone interviews showed a mean rate of 69 percent (De Leeuw & van der Zouwen, 1987). While it has been suggested that the lower response rates may be due in part to difficulties in establishing interviewer-respondent rapport and trust in phone surveys (Groves & Kahn, 1979), phone surveys also suffer from special problems involving noncontacts (Miller et al., 1990). Unanswered phones, difficulties finding respondents at home, and the increased use of answering machines can make even the initial task of contacting the respondent considerably difficult.

Nonresponse, whether due to refusals or noncontacts, can have serious repercussions for surveys intended to derive prevalence estimates of risk behaviors. To the extent that nonrespondents are different from respondents on survey measures, statistics based on respondent data alone will be biased estimates of the population parameters (Groves & Lyberg, 1988). Although participation bias is certainly not peculiar to sex research, surveys that focus on sexual behavior may be more susceptible than surveys on most other topics. Studies have found, for example, that sex survey participants, relative to nonparticipants, tend to be older and less educated (Johnson & DeLamater, 1976), and are less comfortable disclosing sexual information (Bradburn, Sudman, Blair & Stocking, 1978; Catania et al., 1986). Moreover, there is research suggesting that sex survey participants engage in more frequent vaginal intercourse and masturbation, have more varied and unusual sexual histories, have greater levels of noncoital sexual activity, and have more exposure to erotic material (c.f. Catania, Gibson, Chitwood, & Coates, 1990). Unfortunately, most of the evidence of participation bias in sex research is based on college students or other limited samples. We know little about the impact of nonresponse in community studies of sexual behavior, especially those based on telephone surveys. Efforts to understand the nature of participation bias in surveys of this type may be complicated by the different sources of nonresponse (refusals vs. noncontact) inherent in these studies.

Given the potential utility and importance of phone surveys in examining the prevalence and determinants of AIDS-related behaviors, special effort should be directed at estimating the degree and direction of participation bias in AIDS survey phone interviews, both as a function of refusal to participate and because of problems involving noncontacts. Such work would provide an important step toward developing techniques for maximizing participation among subgroups typically underrepresented in AIDS surveys.

METHODS

Sample Design and Procedure

The National AIDS Behavioral Surveys (NABS) were designed to identify AIDS-relevant risk behaviors and their correlates in a population-based sample of adults aged 18-75 residing in the contiguous United States. The overall study design allows us to describe sexual practices; HIV antibody test seeking; and levels of AIDS-related knowledge, beliefs, and social skills over the adult life-span both for the United States as a whole and for the major high AIDS prevalence areas.

A particular emphasis was placed on oversampling young (18-49) Black and Hispanic populations within the major AIDS epicenters. The sample designed to obtain our target distribution of respondents consisted of three major components: a national sample (N=2,673), a central city random digit dial (RDD) sample (N=8,263), and an Hispanic listing sample (N=3,060). Together these three samples yielded a total N of 13,690.(1) All interviews were conducted by telephone. Interviews were conducted in both English and Spanish (55% of all Hispanics chose to be interviewed in Spanish). A minimum of 17 call attempts were made at a variety of times and days before the number was retired.(2) Details on the sample design and procedure are reported elsewhere (see Catania et al., 1992; Dolcini et al., 1993). Household refusals and noncontacts together yielded a response rate of 66.0 percent for the total sample, 69.7 percent for the national cross-section, 60.6 percent for the central city sample, and 78.9 percent for the Hispanic listed sample, after adjusting for estimated noneligibles.

NABS Non-Response Survey

Although the response rates for this study compare favorably with other telephone and face-to-face surveys in this area of research, we believe it is important to attempt to specify characteristics of nonparticipants and to estimate possible biases as a result of nonparticipation. To examine these issues, an additional survey of nonrespondents was conducted. The sampling frame of nonrespondents was comprised of two general groups: people who refused twice to participate in the main study (i.e., they still would not participate after refusal conversion efforts), and people who were extremely difficult to contact (i.e., we were unable to contact the respondent after 17 call attempts). These two groups comprised the entire sampling frame for the nonresponse survey, from which a random sample of approximately 1 in 2 was drawn.

The first group of nonrespondents (i.e., refusers) were further divided into two groups or types. The majority (74.3%) represented household refusals, whereby someone in the household refused to participate prior to enumeration. The nonresponse survey in this case consists of a household enumeration (obtaining age, gender, and ethnicity of individuals living in the household) and questions on health care and family income. The second group of refusers consists of respondent refusals, whereby the actual household member selected for the interview refused to participate (household enumeration information was already obtained). In these cases, we attempted to conduct a shorter 3 minute survey that included questions that we believe may be related to nonparticipation and sexual behavior (i.e., church attendance, perceived confidentiality, time spent at home, prior survey experience).

The second group of nonrespondents consisted of individuals who were particularly difficult to contact. Because of necessary time and cost restrictions, phone surveys generally must impose a cut-off point for the number of call attempts that can be made to any given household. The maximum of 17 call attempts that was imposed on the NABS was liberal by most standards. However, we felt it was important to attempt to describe those individuals who were not included in the study because of contact problems, in addition to those nonparticipants who refused. Most individuals who were recruited into the nonresponse survey from this difficult-to-contact group completed the entire interview, while a smaller portion agreed to complete the shorter survey.

A total of 1,545 individuals were recruited into the nonresponse study: 668 were refusers and 877 were obtained from the difficult-to-contact sample. The overall response rate for the nonresponse survey was 33.3 percent. The response rate for the refuser portion of the sample was 27.1 percent, while the response rate for the difficult-to-contact portion was 40.4 percent. These response rates, although low by typical survey standards, are comparable to other AIDS-related nonrespondent surveys. For example, in a study of nonrespondents from the 1990 National Household Survey of Drug Abuse (NHSDA) (Casper, 1992), an overall response rate of only 34 percent was obtained despite a $10 financial incentive. Similar to our nonresponse survey, the NHSDA interviewers were more successful in converting noncontact cases than refusal cases, with response rates of 45 percent and 25 percent respectively. Another AIDS-related nonresponse survey was conducted by investigators of the National Household Seroprevalence Study (NHSS) (Horvitz et al., 1989). Focusing exclusively on refusal nonrespondents, these researchers obtained a response rate of 26 percent when initial refusals were recontacted and asked to provide risk behavior information and a blood sample. Moreover, this relatively low response rate was obtained in the context of a large financial incentive ($175). A better, but still modest, response rate was obtained when initial refusers were offered a substantial incentive but were not asked to provide a blood sample.

It should be noted that the sampling frame for the present study is comprised of particularly "hard-core" nonrespondents. Unlike the studies cited above, this nonresponse survey does not include initial refusal conversion cases or respondents who participated after the typical number of call-backs employed in telephone surveys. In the present study, nonrespondents were eligible for inclusion only after they had refused participation twice or had not been contacted after seventeen call-backs. Given these stringent eligibility criteria, we believe our overall response rate of 33 percent for the nonrespondent study is admirable.

RESULTS

Demographic Comparisons

Table 1 presents demographic comparisons of main-study participants, hard-to-contact respondents obtained from the nonresponse study, and refusals (also interviewed in the nonresponse study). In comparison to the main study participants, the hard-to-contact group had significantly greater proportions of men (50 vs. 43%), Blacks (30 vs. 22%), and individuals earning over $60,000 a year (13 vs. 8%). The refusal group comprised less than half the proportion of Hispanics and a significantly greater proportion of Whites than were represented in the main study. There were also greater percentages of older people (50 and older) and smaller proportions of younger individuals (under 40 years) among refusers, relative to participants. Finally, individuals in the highest income bracket were over-represented among refusers.

Table 1. Demographic Comparisons of Participants, Refusals, and Hard-to-Contact Respondents: NABS Non-Response Study

                    Participants(a)       Hard-to-Contact
                       % N                Respondents % (N)
GENDER
 Men                43.2 (5643)           50.1 (439)(**)
 Women              56.8 (7430)           49.9 (438)(**)
RACE/ETHNICITY
 White              39.6 (5171)           34.0 (298)
 Black              22.5 (2935)           30.3 (265)(**)
 Hispanic           35.9 (4700)           34.0 (298)
 Other              2.0 (238)              1.7 (15)
AGE
 18-29              30.0 (3920)           30.5 (266)
 30-39              28.8 (3768)           32.3 (281)
 40-49              17.8 (2327)           22.6 (197)
 50-59              9.5 (1245)             7.0 (61)(**)
 60-75              13.8 (1810)            7.6 (66)(**)
INCOME
 <10,000            23.2 (2847)           21.3 (182)
 10,000-20,000      26.7 (3277)           22.4 (191)
 20,001-40,000      30.5 (3739)           28.3 (242)
 40,001-60,000      11.0 (1353)           14.6 (125)
 >60,000            8.5 (1044)            13.3 (114)(**)

                    Refusals % (N)
GENDER
 Men                43.2(289)
 Women              56.7(379)
RACE/ETHNICITY
 White              56.2 (363)(**)
 Black              24.8 (160)
 Hispanic           15.5 (100)(**)
 Other               3.5 (23)
AGE
 18-29              15.1 (87)(**)
 30-39              19.1 (110)(*)
 40-49              19.3 (111)
 50-59              17.4 (100)(**)
 60-75              29.2 (168)(**)
INCOME
 <10,000            19.0 (88)
 10,000-20,000      21.8 (101)
 20,001-40,000      31.3 (145)
 40,001-60,000      15.3 (71)
 >60,000            12.5 (58)(**)

Note. For the purpose of these analyses, participants represent only those respondents who were contacted within the main study criteria of 17 call attempts, hard-to-contact respondents are those obtained only after 18 or more call attempts (most of whom completed the entire interview), and refusals are respondents who refused twice to participate in the main study but who later agreed to respond to a shorter interview. The refuser group in these analyses includes both "household" refusals (someone in the household refused prior to selecting a respondent) and "enumerated" refusals (selected respondents refused after household enumeration was obtained).

(a) Based on total (central city + national + Hispanic list) unweighted sample: excludes participants recruited after 18 or more call attempts; N = 13,073.

(*) Significantly different from participants at p < .05.

(**) Significantly different from participants at p < .01.

These same demographic comparisons between main study participants, hard-to-contact respondents (nonresponse study), and refusals (nonresponse study) were also made within the national, central cities, and Hispanic list sub-samples. This allowed us to determine whether characteristics of participants versus nonrespondents varied by specific sample criteria. In general, the same trends are evident in these different subsamples, although differences are often nonsignificant due to smaller N sizes. For example, the greater proportion of men among hard-to-contact cases, relative to the participants, is found in all three samples. In both national and central cities samples, there is a greater proportion of Blacks among hard-to-contact respondents than among participants, while Whites are overrepresented among refusers. Also in the national and central cities samples, older respondents (especially those 60-75) are overrepresented among refusers, while younger respondents (especially those under 30 years) are underrepresented among the refusal group. Even though the age range is restricted and sample sizes are quite small, this same general pattern is found among the Hispanic list sample, with a greater proportion of the oldest group (40-49) found among refusals, relative to participants.

Given the relatively low response rate of the nonresponse study, particularly among refusals, it is difficult to know how representative our nonrespondent sample is of all refusals. Indeed, we cannot establish for certain the external validity of our refusal group, since the majority of refusals occurred before we could obtain any information at all. However, we can determine among the enumerated refusals whether there are any differences between those who agreed to participate in the nonresponse study and those who refused any level of participation. In other words, if we consider only those individuals whose demographic characteristics were known prior to refusing participation in the main study, we can compare those who also refused participation in the nonresponse study with those who agreed to respond to this shorter survey.

Table 2 compares these enumerated refusals who participated in the nonresponse study with those who did not participate. Although there are a few minor differences (Whites and older refusers were a little more likely to participate in the nonresponse survey), no gender, race, or age differences were statistically significant. Therefore, there is some preliminary evidence that our sample is representative of refusers in general.

Table 2. Social Demographic Characteristics of Non-Response Participants vs. Refusals

                  Non-Response          Non-Response
                  Participants %(N)     Refusals %(N)

GENDER
  Men             46.5 (140)            44.0 (162)
  Women           53.5 (161)            56.0 (206)

RACE/ETHNICITY
  White           54.8 (161)            50.6 (181)
  Black           24.1 (71)             26.0 (93)
  Hispanic        18.4 (54)             20.4 (73)
  Other            2.7 (8)               3.1 (11)

AGE
  18-29           13.3 (38)             13.5 (47)
  30-39           16.8 (48)             22.4 (78)
  40-49           21.0 (60)             21.3 (74)
  50-59           18.5 (53)             17.0 (59)
  60+             30.4 (87)             25.9 (90)

PERCENT (N)       45.0 (302)            55.0 (368)

Note. Based on enumerated households only.

Social-Psychological Comparisons

Table 3 presents social-psychological comparisons of participants, hard-to-contact respondents, and refusals obtained in the nonresponse survey. Findings indicate that refusers are considerably more likely than participants to believe that information obtained in surveys will not be held confidential. Specifically, almost 41 percent of refusers but only 13 percent of participants said they disagree a lot or kind of disagree with the statement that "surveys conducted by universities or colleges will keep your answers confidential or private." Interestingly, hard-to-contact respondents also differ from participants in this regard, with a significantly smaller proportion of the hard-to-contact group perceiving confidentiality as high. Ease of sexual self-disclosure was also far lower among refusers, with more than half of this group indicating that it would be either kind of hard or very hard to talk about sex in an AIDS survey. Only 13.1 percent of participants and 17.5 percent of hard-to-contact respondents fell into this category. Refusers were significantly more likely than the other two groups to attend church services one or more times per week. As expected, the hard-to-contact group spent more time away from their home than both participants and refusers. Differences in past survey participation were also evident. Participants were most likely to have been involved in at least one survey within the past six months (41.5%), followed by hard-to-contact respondents (36.2%), with refusers being least likely to have prior survey experience (31.2%). Finally, comparisons of the number of sexual partners in the past year were made between main-survey participants and hard-to-contact respondents. Findings indicate a lower proportion of respondents having zero partners and a higher proportion of respondents having two or more partners among the hard-to contact group, although this latter difference is not statistically significant.

Table 3. Social-Psychological Comparisons of Participants, Hard-to-Contact Respondents, and Refusals: NABS Nonresponse Study

                                                     Hard-to-
                                                     Contact
                                  Participants(a)   Respondents(b)
                                      % (N)             % (N)

PERCEIVED CONFIDENTIALITY(d)
  Low                             13.1 (1574)       17.5 (136)
  Medium                          35.9 (4329)       41.8 (324)(**)
  High                            51.0 (6145)       40.6 (315)(**)
SEXUAL SELF-DISCLOSURE(e)
  Low                             24.7 (3173)       24.8 (203)
  Medium                          38.3 (4953)       42.7 (353)
  High                            37.1 (4802)       32.8 (271)
CHURCH ATTENDANCE
  1 or more/wk                    32.6 (4251)       33.1 (277)
  < 1 wk                          42.5 (5534)       43.2 (361)
  Never                           24.9 (3251)       23.7 (198)
TIME AWAY FROM HOME(f)
  Low                             21.2 (2691)       12.0 (97)(*)
  Medium                          65.6 (8321)       68.2 (549)
  High                            13.2 (1669)       19.8 (159)(*)
PRIOR SURVEY EXPERIENCE(g)
  Yes                             41.5 (5390)       36.2 (302)
  No                              58.5 (7602)       63.9 (535)(*)
SEXUAL PARTNERS IN THE
 PAST YEAR
  Zero                            12.5 (1477)       9.2 (64)(**)
  One                             74.5 (8830)       74.0 (515)
  Two or more                     13.0 (1553)       16.8 (117)

                                   Refusals(c)
                                     % (N)

PERCEIVED CONFIDENTIALITY(d)
  Low                             40.6 (88)(**)
  Medium                          35.5 (77)
  High                            24.0 (52)(**)
SEXUAL SELF-DISCLOSURE(e)
  Low                             51.0 (125)(**)
  Medium                          25.3 (62)(*)
  High                            23.7 (50)
CHURCH ATTENDANCE
  1 or more/wk                    44.1 (128)(**)
  < 1 wk                          33.1 (96)
  Never                           22.8 (66)
TIME AWAY FROM HOME(f)
  Low                             28.0 (69)
  Medium                          58.9 (145)
  High                            13.0 (32)
PRIOR SURVEY EXPERIENCE(g)
  Yes                             31.2 (89)
  No                              68.8 (196)(**)
SEXUAL PARTNERS IN THE PAST
 YEAR
  Zero                              --
  One                               --
  Two or more                       --

(a) Based on unweighted sample (central city + national + Hispanic list); excludes participants recruited after 18 or more call attempts; N = 13,073.

(b) Respondents recruited after 18 or more attempts,

(c) Includes refusals who agreed to participate in non-response survey (household refusals + respondent "enumerated" refusals),

(d) Respondents were asked "How much do you agree or disagree that surveys conducted by universities or colleges will keep your answers confidential or private?" Low = disagree a lot + kind of disagree, medium = kind of agree, high = agree a lot.

(e) Respondents were asked "Do you think talking about sex in an AIDS survey would be: very easy to do, kind of easy, kind of hard, or very hard to do?" Low = very hard + kind of hard, medium = kind of easy, high = very easy.

(f) Hours in a normal week spent outside the home: low = < 1 SD below mean, medium = mean + 1 SD, high = > 1 SD above mean.

(g) Yes = participated in 1 or more survey in past 6 months (not including current survey). No = did not participate in any survey during the past six months.

(*) significantly different from participants at p < .05.

(**) significantly different from participants at p < .01.

These same analyses were conducted separately for the national, central cities, and Hispanic subsamples. Again, while small sample sizes reduced the number of significant differences between participants and nonrespondents, most of the trends found in the total sample were also found within subsamples. Some exceptions were found in the Hispanic list group. Unlike the national and central cities samples, little variation between participants and the two groups of nonrespondents was evident in the Hispanic list sample with respect to ease of sexual self disclosure, past survey participation, and number of sexual partners.

Correlates of Sexual Risk Behavior

Ultimately, it is important to determine in AIDS surveys the extent to which nonparticipation biases estimates of risk behavior. Although point estimates of bias are not possible with the present data, we can determine, in a general sense, whether exclusion of the two categories of nonrespondents would likely overestimate or underestimate sexual risk behavior in the population.

Because we have complete data on hard-to-contact cases, we are able to determine that individuals who do not participate because they were difficult to find may generally be a higher risk group than participants. This is based on our finding that respondents who could only be contacted after 18 or more call attempts were more likely to have multiple partners (although the difference is not statistically significant) and less likely to have no sexual partners, relative to study participants. Although sexual risk behavior data were not obtained for refusal nonrespondents, we can use participant data to determine whether characteristics or factors that are associated with the refuser group are also associated with risk behavior. Further, we can examine the validity of the association between hard-to-contact individuals and number of sexual partners by determining whether other characteristics or factors associated with being hard to contact are also associated with sexual risk behavior among participants.

Pearsons correlations among the demographic/social-psychological factors and the number of sexual partners were calculated within each of the three subsamples of participants. In all three subsamples, age was inversely related to number of partners, although the association is weaker within the Hispanic list sample due to the more narrow age range. Correlations are -.384, -.367, and -.124 for the national, central cities, and Hispanic list samples, respectively. Men report having significantly more partners in all three samples, with correlations ranging from .191 to .224. Income is positively related to number of sexual partners. Ease of sexual self-disclosure and church attendance are inversely related to number of sexual partners, while time away from home shows a significant positive association. Analyses also show statistically significant positive associations between number of sexual partners and being Black, believing that surveys are held confidential, and having prior survey experience, although the magnitude of these associations are relatively small (correlations range from .05 to .08).

DISCUSSION

In sum, findings indicate that individuals in higher income brackets and older people are most likely to refuse to participate in the National AIDS Behavioral Survey. Hispanics are considerably less likely to refuse participation than other ethnic groups. A number of social-psychological factors are also related to the tendency to refuse participation in this study. In general, refusers were more likely to (a) believe that information obtained in surveys would not be held confidential, (b) believe it would be hard to talk about sex in an AIDS survey, (c) attend church services one or more times per week, and (d) have no prior survey experience within the last 6 months. Correlational analyses suggest that many of these refuser attributes are negatively associated with number of sexual partners. Although the demographic evidence is quite weak (only age is related to both a greater likelihood of refusing and lower numbers of sexual partners), some social-psychological attributes of refusers, such as their higher church attendance and greater difficulty discussing sex, suggest they may generally represent a lower risk group. This conclusion is consistent with evidence from a recent telephone survey finding that nonresponders to sexual survey questions tend to be less sexually experienced than responders (Wiederman, Weis, & Allgeier, 1994).

Respondents who were contacted only after 18 or more calls show a different pattern of characteristics relative to main-survey participants. Specifically, these difficult-to-contact respondents are overrepresented among men, Blacks, and individuals in higher income brackets. They are also more likely to believe that scientific surveys are not held confidential, are less likely to have participated in a survey with the last 6 months, spend less time at home, and have more sexual partners than main study participants. The finding that hard-to-contact respondents report having more sexual partners is consistent with some of the associations between number of sexual partners and the demographic/social-psychological factors considered above. For example, among participants, being male, Black, and spending less time at home are all positively associated with number of sexual partners.

Findings suggest that AIDS-related phone surveys that employ standard call-back procedures may underrepresent higher risk respondents. That is, individuals who cannot be contacted within commonly used call-back limits may, as a group, be engaging in more high risk sexual behaviors than individuals who can be reached within fewer call attempts. Conversely, results of this study suggest that individuals who refuse participation may, in general, represent a lower risk group in terms of AIDS-related sexual behaviors than study participants. While the evidence of this type of bias among refusers is not strong and could not be examined directly in this study, there is an indication that individuals who refuse to participate in AIDS surveys differ from those who could not be contacted.

To the extent that non-respondents who could not be reached by phone are "balanced out" by non-respondents who refuse participation, perhaps our estimates of sexual risk behavior are quite accurate. Of course, given the limitations of this study, it is impossible to determine whether this is the case. At the very least, this study suggests that efforts to minimize different sources of nonparticipation could have different effects on the nature of participation bias and estimates of sexual risk behavior. In addition to the difficult task of increasing participation rates in nonrespondent surveys, future research of this type would benefit from multivariate analyses that examine the independent influence of demographic and social/attitudinal factors and how they may interact to affect nonresponse. More detailed analyses would allow for a better determination of the impact of different types of nonresponse and suggest a plan for adjusting our estimates of AIDS-related risk that accounts for nonparticipation bias. While in the present study small sample sizes in many of the nonrespondent subgroups did not allow for these more detailed analyses, this research does confirm the importance of considering nonresponse bias in telephone surveys of sexual behavior. Of particular relevance is the need to investigate how the special problems of telephone surveys, such as noncontact difficulties, combine with typical refusal problems to affect our results and conclusions.

(1) Discrepancy between total sample N and the sum of the three subsamples is due to a small overlap of national and central city components.

(2) The total sample N of 13,690 also includes respondents who were recruited into the nonresponse study after 18 or more call attempts but who completed the entire interview.

REFERENCES

Bradburn, N., Sudman, S., Blair, E., & Stocking, C. (1978). Question threat and response bias. Public Opinion Quarterly, 42, 221-234.

Casper, R. A. (1992). Follow-up of non-respondents in 1990. In C. Turner, J. Lessler, & J. Gfroerer (Eds.), Survey measurement of drug use: Methodological studies (pp. 155-173). DHHS Publication No. (ADM) 92-1929. Washington, DC: Government Printing Office.

Catania, J. A., Coates, T. J., Stall, R., Turner, H., Peterson, J., Hearst, N., Dolcini, M. M., Hudes, E., Gagnon, J., Wiley, J., & Groves, R. (1992). Prevalence of AIDS-related risk factors and condom use in the United States. Science, 258, 1101-1106.

Catania, J. A., McDermott, L., & Pollack, L. (1986). Questionnaire response bias and face-to-face interview sample bias in sexuality research. The Journal of Sex Research, 22, 52-72.

Catania, J. A., Gibson, D., Chitwood, D., & Coates, T. (1990). Methodological problems in AIDS behavioral research: Influences on measurement error and participation bias in studies of sexual behavior. Psychological Bulletin, 108, 339-362.

Catania, J. A., Gibson, D., Marin, B., Coates, T, & Greenblatt, R. (1990). Response bias in assessing sexual behaviors relevant to HIV transmission. Evaluation and Program Planning, 13. 19-29.

De Leeuw, E. D., & van der Zouwen, J. (1987). A methodological comparison of the data quality in telephone and face-to-face surveys: A meta-analysis of the research literature (Publication No. 36). Amsterdam, The Netherlands: University of Amsterdam.

Dolcini, M. M., Catania, J. A., Coates, T. J., Stall, R., Hudes, E., Gagnon, J., Pollack, L. (1993). Demographic characteristics of heterosexuals with multiple partners: The National AIDS Behavioral Surveys (NABS). Family Planning, 25, 208-214.

Gagnon, J. (1988). Sex research and sexual conduct in the era of AIDS. Journal of AIDS, 1, 593-601.

Groves, R., & Lyberg, L. (1988). An overview of nonresponse issues in telephone surveys. In R. Groves, P. Biemer, L. Lyberg, J. Massey, W. Nicholls II, & J. Waksberg (Eds.), Telephone survey methodology (pp. 191-211). New York: Wiley.

Groves, R., & Kahn, M. (1979). Surveys by telephone: A national comparison with personal interviews. San Diego, CA: Academic Press.

Horvitz, D. G., Weeks, M. F., Visscher, R. E., Folsom, R. E., Hurley, P. L., & Wright, R. A. (1989). A report of the findings of the National Household Seroprevalence Survey Feasibility Study. Paper presented at the Annual Meetings of the American Statistical Association, Anaheim, California, August.

Johnson, W. & DeLamater, J. (1976). Response effects in sex surveys. Public Opinion Quarterly, 40, 165-181.

Miller, H., Turner, C., & Moses, L. (1990). AIDS: The second decade. Washington, DC: National Academy Press.

Wiederman, M. W., Weis, D. L., & Allgeier, E. R. (1994). The effect of question preface on response rates to a telephone survey of sexual experience. Archives of Sexual Behavior, 23, 203-215.

Manuscript accepted May 21, 1998

This study was funded by NIMH Grant No. MH43892 and MIMH/NIA Grant No. MH46240.

Direct correspondence to Dr. Heather Turner, Department of Sociology, Horton Social Science Center, University of New Hampshire, Durham, NH 03824.