Analysis of selection bias in the pilot of a longitudinal study on ageing in Spain

Demonstrating that obtaining a probabilistic national sample for a longitudinal study on ageing in Spain is exposed to selection bias. Quantifying losses that will occur when administering the questionnaire at different stages.

Methods: Cross-sectional study on non-institutionalized Spanish population fifty years of age or older conducted between 2010 and 2011. Through multivariable logistic regression, characteristics of residence census tracts that were accessed (n = 5813) or not (n = 7023) to be included in the sampling frame, from which (n = 1677) were accessed or not (n = 2875) to participate in the study, and of the people who answered (n = 1398) or did not (n = 346) an in-person questionnaire following a telephone one. Additionally, the reasons for the negative responses were studied.

Results: Poor health and disability are the most frequently cited specific reasons for not being a part of the framework or not participating in the study (14.4% and 27.9%, respectively). In both cases, the most frequent negatives in the census sections is of lower socioeconomic status or those located in Catalonia, Guipúzcoa and Vizcaya. Individuals over 80 years of age respond less frequently to the in-person questionnaire. At each stage of information gathering, between 8.6% and 18.4% of people are lost.

Conclusions: Probabilistic sampling in sampling points chosen for convenience allow more resources to be devoted to increasing response rates in groups with the least participation. The concentration of questions in a single, shorter questionnaire prior to drawing blood is proposed.