Enhancing representativeness in population-based surveys to improve data quality and decision-making.

Online surveys often suffer from low response rates, raising concerns about the representativeness and validity of resulting estimates. We evaluated the impact of calibration weighting on data from a cross-sectional online survey on mental health and well-being conducted among university students at the University of Torino, Italy, which had a ~ 10% response rate (eligible population ~ 79,000). Calibration was performed on data from 5,284 students using the raking method and auxiliary variables (sex, course area, and course cycle) to improve representativeness. The unweighted estimate for depressive symptoms prevalence (46.9%, 95% CI: 45.5-48.3) closely matched the weighted result (46.6%, 95% CI: 45.1-48.1). Estimates for suicidal behavior risk were also consistent (34.4%, 95% CI: 33.0-35.7 vs. 34.9%, 95% CI: 33.5-36.4), with only a slight difference observed for anxiety symptoms (72.2%, 95% CI: 70.9-73.4 vs. 69.6%, 95% CI: 68.2-71.0). Unweighted and calibrated estimates were also similar for well-being outcomes (mean overall well-being score: 31.8, 95% CI: 31.4-32.1 vs. 31.9, 95% CI: 31.5-32.3, respectively). These findings suggest that robust estimates may still emerge from large-scale online university surveys exploring mental health outcomes, despite low participation. Further research is needed to assess whether calibration methods yield consistent results across different populations, outcomes, and survey designs.