Social desirability in reporting paying for sex and risky behaviours: comparing two techniques for handling missing data

Sexual behaviour and personal expenses are among the most sensitive topics in surveys. Thus, sample members asked about paying for sex might misreport or refuse to answer such questions. However, surveys about sexual behaviour have been increasingly common in the last decades for the study of Sexually Transmissible Infections and for policy evaluation. The level of social desirability associated to paying for sex may be influenced by prostitution policies, through attitudes. In this work, I discuss the legislation models regulating sex work in the United Kingdom, showing a shift of policies focus from sex workers to clients and a consequent stigmatization of men who pay for sex. The shift of UK policies towards the reduction of the demand for sexual services had the explicit aim to “challenge attitudes” toward paying for sex. I review the evidence from the literature on the effect of prostitution policies on attitudes toward sex work and I discuss how attitudes may influence reporting paying for sex. The sensitive nature of questions on paying for sex may influence not only the measurement but also the sampling strategy adopted by social researchers to investigate the phenomenon. I discuss different sampling techniques adopted in the literature on sex workers’ clients in relation to the substantive findings derived by these studies. The sociology, economic and epidemiological literature have studied sex workers’ clients linking the concept of risky behaviours with paying for sex: however, both concepts are subject to social desirability. I hypothesize these two sets of behaviours to be characterized by high item missing data, and that missing data is not missing completely at random (MCAR). Also, I hypothesize that the statistically significant correlation of paying for sex and risky behaviours may not hold under different missing data handling techniques. In order to test these hypothesis, I use data from the National Survey for Sexual Attitudes and Lifestyles (Natsal-2). This is a nationally representative sample of over 12.000 adults (aged 16-44) living in Britain in 2000. I confirm that paying for sex and risky behaviours have higher missing data then other non sensitive socio-demographic items (excluding partners’ characteristics). Also, item missing data are not MCAR, neither in terms of socio-demographic characteristics nor in terms of paradata on respondents’ embarrassment, commitment and privacy of the interview setting. Moreover, the propensity to produce valid answers does not vary widely across the behaviours considered. I reject the hypothesis that multiply imputed values have higher prevalence of paying for sex and risky behaviours; finally, the correlation of paying for sex and risky behaviours is not significantly different under the two missing data handling techniques considered (listwise deletion and Multiple Imputation by Chained Equations). The finding that paradata are associated with item non response in paying for sex and risky behaviours is particularly promising. As these factors can be informative of the non response process, the inclusion of such items in data collections on sensitive topic is considered advisable, when possible under budget constraints. The evidence that, in this analysis, the estimate of the correlation of paying for sex and risky behaviours do not change under different missing data handling techniques is encouraging as it does not undermine other evidence from the literature. To assess the external validity of this finding, further research may evaluate whether estimates of the correlation of paying for sex and risky behaviour differ by missing data handling technique in different contexts, and/or under different subsample specifications. Overall, I believe this work makes an important contribution not only on the measurement and analysis of paying for sex, but also on the analysis of the correlation of different sensitive behaviours.