A potential inflating effect in estimation models: Cautionary evidence from comparing performance enhancing drug and herbal hormonal supplement use estimates

ObjectivesThis paper compares two indirect prevalence estimation methods that offer protection beyond anonymity and are suitable for self-administration, for investigating the epidemiology of transgressive or socially sensitive behaviours.DesignIn this self-report study, 513 participants (58.7% male) from sports clubs across the UK and southern Ireland were asked to complete an anonymous survey containing the recently developed Single Sample Count (SSC), along with a comparative method Unrelated Question Model (UQM), using prohibited performance-enhancing drugs/substances (PED) as sensitive and hormone-boosting herbal supplements (HS) as non-sensitive control questions.MethodThe survey comprised of sections of SSC, UQM, social projection and simple network scale up methods. Respondents were asked to indicate whether they preferred the SSC or UQM for more protection and ease of completion.ResultsA large discrepancy was observed in prevalence estimates for PED using the UQM (58.4%) and SSC (19.8%), but not for HS (54.9% and 54.0%, respectively). The SSC prevalence estimate for PED was in keeping with the results from social projection (13.8% in own sport; 26.1% in all sports) and network scale up (19.3% for known and suspected doping combined). A clear preference was logged for SSC.ConclusionSSC, but not UQM, showed good concurrent validity with social projection and personal networks for PED; and good discriminant validity with HS. The observed discrepancy could be explained by strategic responding which can inflate the proportion of ‘yes’ answers in the UQM. Adaptation of the UQM for self-administration may lead to an unwanted upward response distortion via strategic responding.

► A concerning level of performance enhancing drug and herbal supplement use was shown in a sub-elite athletic population.
► The Single Sample Count (SSC) but not the Unrelated Question Model (UQM) showed good concurrent and discriminant validity.
► For the first time, a potential noncompliance-driven overestimation is evidenced in a respondent-controlled UQM setup.
► Theoretical proof is given that adaptation of the UQM can lead to an unwanted and previously unaccounted response bias.