Student Finance and Wellbeing Study (SFWS) Scotland 2023-2024: technical report

6. Weighting

The 2,429 valid survey responses were weighted to control for the effects of sampling and to match population estimates more closely.

The first stage of weighting accounted for the sampling design. Participating institutions were each provided with a target number of eligible full-time and part-time students to randomly sample, apart from the two smallest participating colleges who were instructed to invite all eligible students. Selection weights were calculated to account for differential probability of selection between eligible students in different institutions and between full-time and part-time students within institutions. The selection weights were calculated using eligible student population data provided by institutions during the sampling process.

As a result of the institutions undertaking the sampling, no information about individual sampled non-responding students was available to use in weighting. This prevented the use of non-response modelling, which requires data on each individual responding and non-responding case. However, the aggregate characteristics of responding and sampled cases were compared for each participating institution and found to be on average broadly similar.

The selection weights were calibrated to estimates of the eligible Scottish student population, including both participating and non-participating institutions. Calibration adjusts the profile of the weights to match the distribution of population estimates. For college students, 2022-2023 estimates provided by the SFC were used. For university students, 2021-2022 estimates published by HESA were used as these were the most recent available. Three variables were included in the calibration matrix: institution type (universities and colleges), full-time or part-time status, and course type (postgraduate, undergraduate, and further education).

After calibration, the weights were checked for outliers and the top 15 cases trimmed. The effect of trimmed and untrimmed versions on key demographic characteristics and survey outcomes was checked. As the trimmed version had better efficiency than the untrimmed version and minimal bias, it was chosen as the final weight. The final weights have a design effect (deff)^[8] of 1.38, 72% efficiency, and effective sample size of 1760.