Using administrative register data for adjusting non-response bias in the finnish gambling harms survey

Abstract

Background: Low response rates are an increasing problem in population-based gambling surveys. Selective non-response may cause biased findings. Supporting information from administrative registers, whenever available for non-respondents can be utilized to estimate the effect of non-response to the gambling-related outcomes. The aim of this study is to evaluate the effect of non-response to the prevalences of two gambling measures: gambling participation and problem gambling. Methods: Population-based Finnish Gambling Harms mixed-mode (online and postal) Survey 2016 was conducted among 18-year-olds or older in three geographical regions in Finland (response rate 36.2%). Weighted prevalences of gambling measures were calculated exploiting the respondents’ data (n = 7,153). The study sample (N = 19,741) was individually linked to socio-demographic data from Statistics Finland to obtain information on both respondents and non-respondents. Multiple imputation was utilized to calculate the adjusted prevalences of gambling measures by register-based variables: sex, age, residential area, family structure, household equivalised disposable income, highest education degree, employment status, and native language. Crude prevalences were compared against weighted and non-response adjusted prevalences. Results: For gambling participation, there was no difference between the crude (81.9% [95% CI 81.0–82.8%]) and the weighted (83.2% [95% CI 82.3–84.0%]) prevalences (p-value 0.09), or between the crude and the non-response adjusted (82.3% [95% CI 81.6–83.0%]) prevalences (p-value 0.49). However, the non-response adjusted (2.8% [95% CI 2.4–3.3%]) prevalence of problem gambling was higher compared to the crude (1.9% [95% CI 1.6–2.3%]) prevalence (p-value 0.002), while there was no difference between the crude and the weighted (2.2% [95% CI 1.9–2.7%]) prevalences (p-value 0.26). Conclusions: Non-response had an effect of problem gambling prevalence in a Finnish Gambling Harms Survey 2016. The presence of non-response bias should be checked when analysing population surveys. Using administrative register data enables unique opportunities to increase the reliability of the results and to adjust the estimates for non-response. Trial registration: Clinical trial number: not applicable.