Comparison between random and convenience samples in a multicenter survey to evaluate medical students’ quality of life

Abstract

Evaluating medical students’ mental and physical health is challenged by difficulties in obtaining randomized samples and the limitations of convenience samples. We conducted a multicenter study to assess the educational environment, quality of life, and emotional competence of medical students. Between 2011 and 2012, a total of 1,350 randomly selected students from 22 schools and 1,201 volunteer students from 50 schools across Brazil completed all questionnaires (WHOQOL-BREF, VERAS-Q, IRI, RS-14, BDI, PSQI, ESS, IDATE, MBI, and DREEM). Monitoring, support for local researchers, and personalized feedback strategies were applied to ensure the participation of randomized students, achieving a response rate of 81.8%. The platform was also available to volunteers. The statistical analysis examined the effect of these two recruitment strategies using general linear models controlling for sex, age, body mass, course year, physical activity and metabolic equivalents, school type, city population, and location. A significance level of 5% and effect sizes estimated by Cohen’s eta-squared were applied to the variables of interest, both using the Bonferroni correction. The volunteer group had more women, fewer students from the final course years, and a larger number of students from private schools and larger cities. These variables largely explained the statistically significant differences and effect sizes observed between randomized and volunteer groups. In conclusion, although some valuable lessons and motivational strategies were identified, the considerable effort required to achieve high adherence through active outreach may not be justified, as results between random and volunteer samples showed minimal differences in questionnaire responses. Our findings suggest that future studies should consider lighter motivational strategies to reduce the burden on research teams. Data and R scripts to replicate the statistical analysis are available in Harvard Dataverse at https://doi.org/10.7910/DVN/YECV8E.