Selection bias in a survey for children of the old-old and the data adjustment with propensity score

Abstract

  Data provided by adult children as well as their older parents would be useful to gain in-depth understanding of family support in old age. In this study, 2,136 children were invited to participate in a mailed questionnaire survey through their parents (n＝823) who were 77 years of age or older and had responded to a personal interview survey. We examined the selection bias among children who responded to the mail survey (n＝685) and a possible weight to adjust the bias. Logistic regression analyses were conducted to assess the likelihood of being a respondent, using the information provided by their parent. The results showed that children living near their parents, with fewer siblings, and expected by their parents to provide care for them were more likely to respond. Then, weights were calculated using the likelihood for each child as the propensity scores. The percentage of the children who were recognized as providers of support by their parents was higher among the respondents of the mail survey than the percentage in the total sample including nonrespondents. However, the differences were reduced after adjusting the respondents’ data, which suggests the effectiveness of the weights.