Abrupt peaks in perceived risk of occasional drug use after changing the question order in a repeated self-administered survey

Abstract

BackgroundQuestion-order changes in repeated surveys can distort comparisons. We want to describe the evolution of drug risk perceptions among Spanish adolescents and assessing whether the 2006 peaks in perceived risk of occasional drug use can be explained by question-order changes.MethodsThe subjects were secondary students from a biennial national survey during 2000–2012. A one-off intervention was applied in 2006, replacing the two-adjacent items on perceived risk of occasional and regular use of each drug by non-adjacent items. Annual prevalence of high-risk perception were obtained for occasional and regular use of cannabis, heroin, cocaine and ecstasy. Subsequently, the 2006 percent level change (PC) in such were estimated prevalence using segmented Poisson regression, adjusting for various student and parent covariates.ResultsThe 2006 PC in prevalence of high-risk perception of occasional drug use ranged from +63% (heroin) to +83% (ecstasy). These PCs were very high in all considered subgroups. However, the 2006 PC in prevalence of high-risk perception of regular drug use ranged from 1% (heroin) to 12% (cannabis). The evolution of preventive interventions does not suggest alternative causal hypotheses for 2006 peaks other than question-order changes.ConclusionWithin the cognitive heuristics framework, the 2006 spikes in perceived risk of occasional drug use were most likely due to a release of the anchor exerted by perceived risk of regular drug use over that of occasional use triggered by 2006 question-order changes. In repeated surveys it is inexcusable to pre-test the effect of any change in questionnaire format.

Reassessing and Extending the European Standards of Care for Newborn Health: How to Keep Reference Standards in Line with Current Evidence

“…After evaluating the results, a small trend (R 2 = 0.1324) could be identified regarding the order of the surveys and the number of respondents: namely, the further down the list of standards, the less respondents were recorded. This phenomenon is not new, as also other scholars reported on the impact of, e.g., question order on the results [16,17]. However, in our case, as the number of standards was relatively limited (n = 20), the trend was negligible as standards which were listed 9th or 11th were answered by 19 and 18 respondents, respectively.…”