The Quality of Data Obtained from Self-Completed Questionnaires in a Survey of High School Students

This study examines issues of data quality in a survey conducted in a nationwide probability sample of 8985 students in the last four years of high school education in Greece. Respondents completed an extensive questionnaire whose main topic of investigation was the use of licit and illicit substances (tobacco, alcohol, cannabis, etc.). We examined the effect of sex, age and school performance on data quality. Specifically, we related these factors to the probabilities of correctly observing filters in the questionnaire, of certain inconsistencies in responses, of reporting difficulty in understanding the questions and being able to answer honestly. It was found that all these factors have strong effects. Boys’ responses presented more problems than girls’ (median odds ratio in a series of separate logistic regressions = 1.35) and younger respondents’ more than older (median odds ratio 1.34 for ages 13–14 and 1.06 for ages 15–16, compared to ages 17–18). The strongest effect was related to school performance compared to the best students (school marks 18–20), median odds ratios were 1.31 for marks 15–17, 1.76 for 12–14 and 3.25 for 10–11. Implications for questionnaire design are discussed.