GREGSON, SIMON DPhil*†; ZHUWAU, TOM MSc†; NDLOVU, JOSHUA‡; NYAMUKAPA, CONSTANCE A. BSw†
RELIABLE DATA on sexual behavior are an essential prerequisite for the development and evaluation of sexually transmitted disease (STD) control and other reproductive health interventions. This is especially so in sub-Saharan Africa and other low-development settings subject to extensive HIV epidemics and/or high prevalence of STDs. In these settings, reliable data on sexual behavior are needed (1) to understand the influence of behavioral and other determinants of patterns of HIV and STD transmission;1 (2) to understand the influence of cultural, economic, psychosocial, and structural determinants of sexual behavior;2 (3) for behavior (change) surveillance designed to aid interpretation of HIV/STD surveillance data;3 and (4) for evaluation of specific HIV/STD prevention strategies. 4
Unfortunately, problems such as selective participation, poor recall, and incorrect response due to desire to conceal socially proscribed activities—social desirability bias 5 —frequently result in serious bias in sexual behavior data. 2 Extensive work has been conducted to develop methods that reduce the impact of bias. 6–9 However, more research is needed on alternative modes of survey administration within diverse populations so that a consensus can be built on the methods most likely to yield reliable and valid data in any given context. 5
Social Desirability Bias
In this article we describe and evaluate an interview strategy developed during work in rural areas of Manicaland Province, Zimbabwe, that was designed with a view to reducing social desirability bias in sexual behavior surveys in low-development environments. Social desirability bias is particularly problematic, as it can differ by location and over time. The former can lead to incorrect conclusions being drawn from studies that compare the relative levels of potentially high-risk behavior for HIV and STD infection in different populations. The latter can provide a false impression of temporal changes in behavior, especially where intensive control activities have been implemented, and can distort results from randomized controlled trials of candidate intervention strategies.
Social desirability bias is most likely to occur when data are sought on attitudes toward or experience of activities that run contrary to dominant local social norms or the respondents’ perceptions of the views of the person(s) conducting the interviews. 5 In any given interview, the extent of bias can be exacerbated by shyness or fear of breach of confidentiality. Bias may be reduced where respondents understand and value the research objectives and appreciate the detrimental consequences of providing incorrect or incomplete information.
Thus, questionnaire designs and survey and interview strategies that provide clear information on the purpose and rationale of the research and involve clearly understandable, easily verifiable procedures that reduce potential embarrassment, ensure confidentiality, match respondents with interviewers with whom they feel most comfortable (in terms of sex, 5 age, and socioeconomic background), and support interviewers in securing the trust and cooperation of respondents are the most likely to be effective.
In addition, (1) careful ordering of questions to deter interviewers from falsifying responses so as to skip sensitive questions and (2) careful wording of questions to reflect the preferred direction of bias 10 and deny easy retreat to socially acceptable answers 11 will tend to reduce bias.
In Western populations, self-administered postal, telephone, and computer-assisted surveys that provide greater privacy typically yield more complete reporting of highly sensitive behaviors. 7 However, this advantage may be offset by greater nonresponse and consequent participation bias. These methods also preclude the advantages that can flow from developing a strong rapport with respondents and limit opportunities for providing clarifications of the meaning of individual questions and for within-interview verification and correction of mistakes—benefits often associated with face-to-face interview (FTFI) methods.
In an African setting, a computer-assisted self-interview method (ACASI) 12 worked well among more-educated respondents in an urban population, 13 but doubts remain as to its practicability in the less-developed rural areas where the majority of people live. In these circumstances, in-depth, less structured interviews, in which time is taken to develop a good rapport with individual respondents, yield fuller reporting of socially proscribed behaviors than more formal, standardized approaches. 14, 15 While this method is too time-consuming to be used in a large-scale survey, these findings suggest that a more informal, less rigid, less mechanistic approach to survey interviewing could improve the reliability of reporting on sensitive topics.
Hybrid approaches, in which the most sensitive questions are self-administered in the presence of the enumerator, may provide the best balance between the need to build rapport and motivation and the need to guarantee privacy (and thus confidentiality). 11,16 Deferring questions on the most sensitive topics until late in the interview provides the enumerator with more opportunity to establish a good rapport before these questions are addressed. The interview process tends to sensitize respondents and thereby influences responses to later questions. 17 Thus, in a face-to-face interview situation, where suitably selected and well-trained enumerators are deployed, this process can be used to neutralize respondents’ initial preconceptions of the responses most likely to find favor with their enumerators. 2
In the mid-1990s, we conducted a study of the early sociodemographic impact of the HIV epidemic in rural areas of Manicaland Province in eastern Zimbabwe. 18 The local Manyika people are renowned for their secretive nature and have a strong taboo against women having extramarital sexual relationships. Using information collected from women met at beer halls, we estimated that 50 to 60 local women were actively engaged in commercial sex work at the time. However, only 2 of the women (0.48%) interviewed by female enumerators using standard population-based survey methods reported having more than 2 sex partners in the previous 4 weeks. 19
During subsequent preparations for a randomized controlled trial of a behavioral intervention, we sought to develop a survey methodology that would elicit more complete reporting of extramarital sexual behavior. In a series of formative research studies, we established that people preferred to be interviewed by “qualified,” same-sex adults who were not known locally but were familiar with the local dialect and customs. Most people preferred self-completion interview methods (the exceptions were older and less educated people). Respondents were more likely to report recent experience of unprotected sex with casual partners when a simple confidential voting interview method (see below) was used. 20
The method was improved and tested again in a pilot study for the intervention trial and was evaluated more extensively in the subsequent baseline survey. In the remainder of this article, we describe the method and the design and results of the baseline survey evaluation.
Informal Confidential Voting Interviews
The Informal Confidential Voting Interview (ICVI) approach constitutes a blend of face-to-face interview (FTFI) and self-completion methods. The first part of the interview is conducted with an informal variant of the FTFI method and is used to establish motivation, to build a strong rapport between respondent and enumerator, and to sensitize the respondent to the nonprejudicial viewpoint of the enumerator and the study. Questions gradually progress from relatively straightforward matters to more private topics. Time is allowed for some informal discussion of the subjects raised. Information on the most sensitive personal subjects is sought in the second half of the interview by means of a secret voting procedure based on Pocket Chart Voting, a participatory rapid assessment technique used in community development projects. 19,21
In the current study, the secret voting procedure was applied with use of lightweight, portable wooden “voting” boxes (Figure 1). The boxes were divided into three compartments, each with a narrow slot in the top labeled with a different color. Sets of three short strips of cardboard in the same colors were used as voting tokens. The tokens were all subdivided into six sections, each prelabeled with a number. The colors of the voting strips and the numbers marked on the sections were cross-referenced to questions in the questionnaire.
In each interview, time was set aside for the enumerator to provide a full explanation of the reasons for using the voting box and the ways in which the design and associated procedure reduce possible embarrassment and ensure confidentiality. The enumerator then wrote the respondent's survey identification number on the back of the voting strips before handing them one by one to the respondent in a predetermined sequence. The enumerator read out the corresponding questions from the questionnaires, one at a time, and the respondent wrote his or her answers in the appropriate sections on the voting strips. The boxes had lids that respondents used as screens to conceal what they were writing from the enumerators and were prelocked with keys held by the supervisors.
Respondents could seek clarification on the meanings of questions from their enumerators but were discouraged from discussing the substance of their responses. After completing each strip, a respondent would place it into the box through the slot labeled with the same color and mix it with those completed by earlier participants. Thus, the information provided was kept secret from the enumerators.
The field supervisors, who did not know the identities of individual respondents, subsequently matched the voting responses to the appropriate questionnaires (using the survey identification numbers) and questions (using the colors of the voting strips and the question numbers marked on the six sections of each strip).
The application of the ICVI method reported upon here was embedded within the baseline population-based survey for the proposed HIV prevention trial in 12 rural locations in Manicaland. The survey was carried out between July 1998 and January 2000, and the sites comprised 4 roadside trading centers, 4 commercial farming estates, and 4 subsistence farming areas. Male and female residents (present in a local household for at least 4 nights in the past month and at the same time one year ago) aged 15 to 54 years identified in a preliminary household census were considered eligible. Written informed consent was sought as a precondition for enrollment (prior ethical approval for the study was granted by the Research Council of Zimbabwe, the Institutional Review Board of the Biomedical Research & Training Institute, and the Applied and Qualitative Research Ethics Committee, Oxford, United Kingdom).
Household and individual participation rates were 98% (8233/8386) and 79% (9826/12,383), respectively. Of the nonparticipants, 20% (509) were known to be refusers and the remainder were still absent or unavailable after 3 household visits. The most common reason for declining to participate was unwillingness to provide the dried blood spot sample required for HIV and STD testing, due to a fear of Satanism that was rife in Zimbabwe at the time. Further details of the baseline survey methods are reported elsewhere. 20,22 The study population for the current analysis was restricted to sexually experienced adults in the most sexually active age range (15–49 years).
The ICVI method was evaluated with use of a short series of questions on sex partnerships, including questions on numbers of partners within various time frames (lifetime, past 12 months, past month, and current) and details of partnerships in the past month (sex acts, condom use, and characteristics of partners). Questions were constructed so as to elicit simple yes/no/don't know or numerical responses and thereby to minimize nonparticipation and response errors. When respondents had writing difficulties, simple picture codes were used to illustrate standardized symbols for yes (tick), no (cross), and so on.
Nonliterate respondents and a sample of literate respondents were interviewed by the FTFI method. Two enumerators, one male and one female, were instructed to use the FTFI method in each study site so that any difference in reporting of socially proscribed behaviors could be measured. These enumerators used the FTFI method in all interviews in the households they visited so that the selection process was random, and they were rotated as the survey team moved from site to site to minimize interviewer effects. Other cases where the FTFI method was used—for example, due to enumerator error or respondent refusal to use the ICVI method—were quantified.
In the questionnaires, the sexual behavior questions followed a lengthy section on sociodemographic characteristics and were asked in order of increasing sensitivity. The survey introduction and early part of the interview process was used to sensitize respondents in ways intended to encourage them to be more open about the later questions on sexual behavior. The questions on age at first sex (and thereby on experience of sexual debut) and time since last sexual intercourse were asked with use of the FTFI method, even in interviews where the ICVI method was used to obtain the data on partnership details.
All interviews (ICVI and FTFI) were conducted by social science graduates; most were qualified social workers skilled at establishing relationships of trust with clients. Enumerators familiar with the local language and customs were selected and matched for sex with respondents. Enumerators were trained to make culturally appropriate introductions and to explain the importance of the research objectives, the possibility that incorrect decisions might be made if unreliable data were provided, and the measures being taken to maintain confidentiality of information. Enumerators were trained to adopt an informal/flexible approach so that their rapport with respondents could develop as interviews progressed and to adopt a nonprejudicial approach. During the survey, project supervisors carried out routine quality control checks on the interview process and on all completed questionnaires.
Comparison of ICVI and FTFI Methods
We calculated the proportions of male and female respondents considered unable to be interviewed by the ICVI method because of total illiteracy. Thereafter, the evaluation of the ICVI method was conducted by reference to literate respondents only. The sociodemographic profiles of men and women interviewed by the two methods were contrasted to identify any possible participation bias. The completeness and internal consistency of the data obtained by the two interview methods were compared.
The statistical significance of observed differences in reporting between the ICVI and FTFI methods was assessed with use of standard statistical chi-square tests. Age, marital status, education, and socioeconomic environment–adjusted odds ratios and 95% confidence intervals for reporting more than one partner currently, in the past month, and in the past year by interview method were calculated by means of logistic regression analysis with STATA version 5. 23 Polygynists were excluded. For each time interval, adjusted negative binomial regression coefficients for total numbers of sex partners reported by respondents with more than one partner were computed to assess whether application of the ICVI method was associated with reporting of higher numbers of partners.
Two hundred twenty-seven of 3460 men (6.56%) and 410/4363 women (9.40%) were insufficiently literate to be interviewed by the ICVI method. These respondents were interviewed by the FTFI method, and their records were excluded from subsequent analyses of the relative effectiveness of the ICVI method. In 54 other cases, no data on literacy status were collected; 39 were assumed to be literate because they were interviewed by the voting method; the remainder did not vote and were dropped from the analysis.
Of those considered sufficiently literate for the ICVI method, 77.76% of men and 73.44% of women were interviewed in this way. ICVI and FTFI respondents were similar in age, employment, and recent health profiles. Slightly higher proportions of women interviewed by the ICVI method had a secondary education (51% versus 47%) and were currently married (35% versus 32%). A higher percentage lived in subsistence farming areas (41% versus 32%), whereas fewer lived in roadside trading centers (34% versus 42%). Among men, slightly more ICVI respondents were married (24% versus 21%) and more lived on estates (44% versus 30%) rather than at roadside trading centers (30% versus 45%).
For both men and women, similar proportions of ICVI and FTFI interviewees reported abstinence from sexual activity for more than 12 months. More of the men interviewed by the ICVI method (23% versus 20%;P = 0.096) but fewer of the women (29% versus 34%;P = 0.003) were HIV-positive. After adjustment for differences in age level, marital status, education level, and study location, the difference in HIV infection status remained nonsignificant for men (OR = 1.17;P = 0.171) but significant for women (OR = 0.84;P = 0.035).
Item Response Rates and Internal Consistency
Levels of missing data for selected questions for the two methods are compared in the top half of Table 1. Missing data are more common when the ICVI method is used, particularly on the later questions covering partnership characteristics. However, levels remain low in absolute terms.
Frequencies of internally inconsistent responses for selected questions for the two methods are compared in the second half of Table 1. In the final three comparisons shown, both questions for which responses are compared were asked by the same interview method, so that each inconsistency in reporting could have occurred due to error in either or both responses. Again, internal inconsistencies were more common in interviews where the ICVI method was used but generally remained below 5%.
Comparison of Reports of Risk Behavior for HIV/STDs
Overall, 18.9%, 10.3%, and 41.4% of men (n = 3,171) and 2.2%, 1.8%, and 5.4% of women (n = 3,876) reported more than one current sexual relationship and more than one sex partner in the past month and past year, respectively. The figures for current partnerships were higher than those for partners in the past month, probably because the former included sexual relationships that were still considered to be ongoing but that may not have been active in the past month.
Adjusted odds ratios (ORs) for reporting multiple sex partners by period before the interview for men and women interviewed by the ICVI versus FTFI method are shown in Figure 2A. Male and female respondents interviewed by the ICVI method were more likely to report multiple sex partners in all periods: current (OR = 1.33 and P = 0.028; OR = 5.24 and P < 0.001), past month (OR = 1.71 and P = 0.002; OR = 2.92 and P < 0.001), and past year (OR = 1.35 and P = 0.002; OR = 1.97 and P = 0.003).
Adjusted ORs for reporting more than one partner in the month preceding the interview by socioeconomic group are shown in Table 2. Higher numbers of young and single men reported multiple sex partners when the ICVI method was used. The difference between methods was of only borderline significance for married men as a whole but was significant for those cohabiting with their wives. The effect was highly significant among men interviewed in subsistence farming areas but not for those in roadside trading centers and estates.
Higher numbers of both young and older women reported having multiple sex partners in the preceding month when interviewed by the ICVI method; the effect was strongest among married women. Those cohabiting with their husbands were least likely to report extramarital partners (OR = 0.40; 95% CI, 0.17–0.91); those who did so had all been interviewed by the ICVI method (9/881 versus 0/380;P > 0.05). Similarly, women living in subsistence farming areas were less likely than women in estates and business centers to report multiple sex partners (OR = 0.36; 95% CI, 0.19–0.66). Again, the few who did so had all been interviewed by ICVI (13/1175 versus 0/336;P > 0.05).
In the negative binomial regression analyses for men (Figure 2B), use of the ICVI method was positively associated with number of partners in the past year (P = 0.025; n = 1188), number of partners in the past month (P = 0.105; n = 233), and number of current partners (P = 0.703; n = 471). For women, there were negative associations with number of partners in the past year (P < 0.001; n = 217), number of partners in the past month (P = 0.135; n = 77), and number of current partners (P = 0.086; n = 86).
Improvements in the reliability of sexual behavior data are needed to guide the development of more effective HIV and STD control programs, particularly in resource-poor settings. While it is unrealistic to expect to eliminate social desirability bias entirely, an improvement in reporting would facilitate identification of significant trends in behavior. A particular concern is the need for methods that are sufficiently robust to control for possible shifts in perceptions as to what constitutes socially acceptable behavior or is regarded as such by personnel engaged in HIV-related surveys. In the absence of such methods, identification of effective HIV/STD control strategies through randomized controlled trials and other scientific research designs with behavioral as well as biomedical endpoints will continue to be problematic.
Qualitative data collected in the current study populations indicate that the effects of the HIV epidemic and the associated public information campaigns have been such that even men now tend to underreport their numbers of sex partners, particularly when interviewed in health-related surveys. The results of the large-scale trial of the ICVI method described in this article show that this method is associated with significant increases in the proportions of men and women who report what constitutes socially proscribed sexual behavior in rural Zimbabwe, over periods ranging from the present to a year.
The strongest evidence of increased reporting of multiple sex partners was found in the sociodemographic subgroups for which this behavior is least socially acceptable, e.g., women (especially married women), cohabiting spouses, and people living in subsistence farming areas. Method of interview shows little or no effect among divorced and widowed men and male labor migrants who live apart from their wives, all of whom are relatively free to engage in casual relationships. Among women, there was stronger evidence of increased reporting of multiple sex partners in the shorter, more recent time periods (current and past month). This was probably because, over longer time periods, greater numbers of women will have had multiple regular partners, and these are less likely to be underreported.
The absence of a similar pattern for men could be because they are less likely to underreport casual partners. More men but fewer women in the ICVI group were HIV-positive than in the FTFI group. This suggests that the women's results are more conservative, since men and women reporting larger numbers of lifetime partners both had higher HIV prevalence (P < 0.001).
We were able to achieve a good participation rate in a rural population with a relatively high level of basic literacy. Item-nonresponse and inconsistency in reporting were kept to modest levels. The enumerators reported no major problems with interruption of the flow of interviews; indeed, the voting procedure introduced a certain novelty value and added interest to the proceedings. The method is inexpensive and practical to apply in household surveys in low-development settings and performed well in comparison with FTFIs, even when the latter were applied in favorable circumstances—i.e., with carefully chosen and trained enumerators and procedures to enhance rapport and motivation and to counter respondents’ initial perceptions about what might constitute “correct” responses.
We believe that the reporting of multiple sex partners is higher with the ICVI method principally because respondents are more certain that the personal information they provide will be kept confidential and are more relaxed about giving this information when not obliged to discuss it openly with their interviewers. However, some of the difference in reporting in comparison with FTFIs may be due to the increased misreporting associated with the method due to the inability of enumerators to check the internal consistency of responses during interviews. 6 In the current study, the extent of such errors was kept low by recruiting, training, and closely supervising skilled enumerators. Even so, in population subgroups from whom reports of multiple sex partners are uncommon, the distribution of these errors will be skewed towards greater reporting of multiple partners and will exaggerate the strength of any positive effect of the interview method on reporting.
However, questionnaire completion errors might be expected to have a less biased effect for questions about numbers of sex partners for respondents reporting more than one partner. For men, these also show a consistent pattern of increased reporting of high-risk behavior for HIV infection. For women the opposite is true, but this can be explained by a greater underlying tendency by women who do not regularly engage in commercial sex work to conceal any extramarital sexual activity. Thus, those who report extramarital relationships only in ICVIs generally report smaller numbers than those who do so even in FTFIs.
Furthermore, our results are consistent with those from studies of other methods that offer a similar degree of privacy, conducted in both Western 7 and African contexts (Wolff and Woelk, personal communications); these studies also showed reductions in social desirability bias. Nonetheless, caution must be exercised in interpreting the results. Further scientific evaluations of similar methods, including studies employing biomarkers, 24,25 are needed both to validate these results and to establish their broader relevance.
The effect of procedures integral to the method that are designed to increase motivation, develop rapport, and sensitize respondents to the nonprejudicial study objectives was not measured in this study, as the same background survey procedures were used in both ICVI and FTFI interviews. However, earlier studies have shown that these measures—which are less feasible in, for example, computer-assisted interviews—can reduce social desirability bias. 11
Other limitations of the ICVI method are that it is relatively time-consuming to conduct, it requires skilled interviewers who are thoroughly trained and supervised, it limits the complexity of questions that can be asked, and it is not suitable for non-literate respondents. The latter was not a major problem in rural Zimbabwe and would not be expected to be so in many of the populations in southern Africa where STDs are common and the incidence of HIV infection is currently highest. 26 Even in countries where overall literacy levels are very low, the ICVI method could still be applied in some subpopulations with a high HIV/STD incidence, e.g., in the typically more educated younger age groups and urban populations.
Finally, a useful distinction has been drawn between the concepts of risk screening (dichotomous indication of presence of risk behavior) and risk assessment (continuous indication of level of risk behavior). 11 The results of this study suggest that more women underreport experience of extramarital partnerships but that more of the men who report such relationships underreport their numbers of extramarital partners. If this is so, it might indicate that the ICVI method is more effective at improving risk screening for women but risk assessment for men.
1. Aral SO. Sexual behaviour in sexually transmitted disease research. Sex Transm Dis 1994; 21 (suppl): S59–S64.
2. Catania JA, Chitwood DD, Gibson DR, Coates TJ. Methodological problems in AIDS behavioural research: influences on measurement error and participation bias in studies of sexual behaviour. Psych Bull 1990; 108: 339–362.
3. Amon J, Brown T, Hogle J, et al. Behavioral Surveillance Surveys (BSS): Guidelines for Repeated Behavioral Surveys in Populations at Risk of HIV. Arlington, VA: Family Health International, 2000.
4. Hayes R, Mosha F, Nicoll A, et al. A community trial of the impact of improved sexually transmitted disease treatment on the HIV epidemic in rural Tanzania: 1. Design. AIDS 1995; 9: 919–926.
5. Schopper D, Doussantousse S, Orav J. Sexual behaviours relevant to HIV transmission in a rural African population: how much can a KAP survey tell us? Soc Sci Med 1993; 37: 401–412.
6. Dare OO, Cleland JG. Reliability and validity of survey data on sexual behaviour. Health Trans Rev 1995; 4 (suppl): 93–110.
7. Aral SO, Peterman TA. Measuring outcomes of behavioural interventions for STD/HIV prevention. Int J STD AIDS 1996; 7 (suppl 2): 30–38.
8. Caldwell JC, Orubuloye IO, Caldwell P. Methodological advances in studying the social context of AIDS in West Africa. In: Basu AM, Aaby P, eds. The Methods and Uses of Anthropological Demography. Oxford: Clarendon Press, 1998.
9. Gribble JN, Miller HG, Rogers SM, Turner CF. Interview mode and measurement of sexual behaviors: methodological issues. J Sex Res 1999; 36: 16–24.
10. Weinhardt LS, Forsyth AD, Carey MP, Jaworski JB, Durant LE. Reliability and validity of self-report measures of HIV-related behaviour: progress since 1990 and recommendations for research and practice. Arch Sex Behaviour 1998; 27: 155–180.
11. Wolff B. Draft Protocol for Monitoring Risk Behaviour for HIV Transmission During Phase III Vaccine Trials. London: London School of Hygiene and Tropical Medicine, 2000.
12. Turner CF, Ku L, Rogers SM, Londberg LD, Pleck J, Sonenstein FL. Adolescent sexual behaviour, drug use and violence: increased reporting with computer survey technology. Science 1998; 280: 867–873.
13. Van de Wijgert J, Padian N, Shiboski S, Turner C. Is audio computer-assisted self-interviewing a feasible method of surveying in Zimbabwe? Int J Epidemiol 2000; 29: 885–890.
14. Konings E, Bantebya G, Caraël M, Bagenda D, Mertens T. Validating population surveys for the measurement of HIV/STD prevention indicators. AIDS 1995; 9: 375–382.
15. Voeten HACM, Meester EI, Egesah OB, Habbema JDF. Experiences with method triangulation in a study on sexual behaviour of young adults aged 15–29 years in Nyanza province, Kenya. Presented at the IUSSP Conference on Partnership Networks and the Spread of HIV and Other Infections, Chiang Mai, Thailand, 7–10 February 2000.
16. Johnson AM, Wadsworth J, Wellings K, Field J, Bradshaw S. Sexual Attitudes and Lifestyles. Oxford: Blackwell Scientific Publications, 1994.
17. Bradburn N, Sudman S. Asking Questions: A Practical Guide to Questionnaire Design. San Francisco: Jossey–Bass, 1983.
18. Gregson S, Anderson RM, Ndlovu J, Zhuwau T, Chandiwana SK. Recent upturn in mortality in rural Zimbabwe: evidence for an early demographic impact of HIV-1 infections? AIDS 1997; 11: 1269–1280.
19. Gregson S, Zhuwau T, Anderson RM, Chandiwana SK. Is there evidence for behaviour change in response to AIDS in rural Zimbabwe? Soc Sci Med 1998; 46: 321–330.
20. Gregson S, Nyamukapa CA, Garnett GP, et al. Sexual mixing patterns and sex-differentials in teenage exposure to HIV infection in rural Zimbabwe. Lancet 2002; 359: 1896–1903.
21. Srinivasan L. Tools for Community Participation: A Manual for Training Trainers in Participatory Techniques. Washington, DC: PROWWESS/UNDP, 1990.
22. Gregson S, Terceira N, Kakowa M, et al. Study of bias in antenatal clinic HIV-1 surveillance data in a high contraceptive prevalence population in sub-Saharan Africa. AIDS 2002; 16: 643–652.
23. Hamilton LC. Statistics with STATA 5. Belmont: Duxbury Press, 1998.
24. Udry JR, Morris NM. A method for validation of reported sexual data. J Marriage Family 1967; 29: 442–446.
25. Cowan FM, Johnson AM, Ashley R, Corey L, Mindel A. Antibody to HSV-2 as serological marker of sexual lifestyle in populations. BMJ 1994; 309: 1325–1329.
26. UNAIDS Report on the global HIV/AIDS epidemic, 2002. Geneva: UNAIDS, 2002.