[Analysis of selection bias in the pilot study of a longitudinal study on aging in Spain]

To demonstrate that selection of a probabilistic sample at a national level for a study of aging in Spain is subject to selection bias. To quantify the losses produced after each phase of the administration of a questionnaire.

We performed a cross-sectional study of the Spanish community-dwelling population aged 50 years or older between 2010 and 2011. Through multivariate logistic regressions, the characteristics of the census tract of the patients' residence were compared between those who agreed (n = 5,813) or refused (n = 7,023) to be included in the sampling frame and between those who agreed (n = 1,677) or refused (n = 2,875) to participate in the study. The individual characteristics of persons who responded (n = 1,398) or refused to respond (n = 346) to a face-to-face questionnaire administered after a telephone interview were also compared. In addition, the reasons for refusal were studied.

The most frequent specific reasons for refusing to be included in the sampling frame or to participate in the study were poor health and disability (14.4% and 27.9%, respectively). In both cases, refusal was more frequent in the census tracts of districts with a lower socioeconomic level or those located in Catalonia, Guipúzcoa or Biscay. Individuals older than 81 participated less frequently in the face-to-face questionnaire. Between 8.6% and 18.4% of participants were lost at each stage of information retrieval.

Probabilistic sampling in sampling points chosen by the researchers would allow more resources to be devoted to increasing response rates among the groups who are less likely to participate. Questions should be concentrated in only one shorter questionnaire, administered before blood extraction.