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

Area of Science:

• Psychiatry and Mental Health
• Biostatistics
• Survey Methodology

Background:

• Online surveys are prone to low response rates, potentially compromising data representativeness and validity.
• Assessing mental health and well-being in university populations is crucial for targeted interventions.
• Low response rates in online surveys can lead to biased estimates, particularly in sensitive areas like mental health.

Purpose of the Study:

• To evaluate the impact of calibration weighting on the representativeness of online survey data.
• To assess the validity of estimates for mental health and well-being outcomes from a low-response-rate online survey.
• To determine if calibration weighting can improve the accuracy of prevalence estimates for depressive symptoms, suicidal behavior risk, and anxiety symptoms.

Main Methods:

• A cross-sectional online survey on mental health and well-being was conducted among university students.
• Calibration weighting using the raking method was applied to survey data (n=5,284) with auxiliary variables (sex, course area, course cycle).
• Unweighted and weighted estimates for depressive symptoms, suicidal behavior risk, anxiety symptoms, and overall well-being were compared.

Main Results:

• Unweighted and weighted prevalence estimates for depressive symptoms were highly consistent (46.9% vs. 46.6%).
• Estimates for suicidal behavior risk also showed close agreement between unweighted and weighted data (34.4% vs. 34.9%).
• Slight differences were observed for anxiety symptoms (72.2% vs. 69.6%), while well-being scores remained similar.

Conclusions:

• Robust mental health estimates can be obtained from large-scale online university surveys despite low participation rates, particularly when using calibration weighting.
• Calibration weighting appears effective in enhancing the representativeness of online survey data for mental health outcomes.
• Further research is recommended to validate calibration methods across diverse populations, outcomes, and survey designs.