Abstract
Lately, more and more empirical studies have adopted multilevel analysis because of its distinctive merits: using information about clustering (data clustered within individuals or individuals clustered within societies) , multilevel analysis produces less biased point estimation and more appropriate interval estimation than ordinary linear models. It is widely seen that in order to exploit this advantage, one needs hierarchically stratified data such as panel data or internationally comparable cross-sectional data. However, multilevel modeling can be flexibly applied to most ordinary cross-sectional data. To demonstrate, an analysis using National Family Research of Japan, which includes a wealth of information on family relationships, was conducted. The focus is on the subjective goodness of relationships with parents. There are a maximum of four relationships, and these may correlate with each other within individuals; thus, a multilevel analysis might be appropriate. Results showed that respondents who live near their parents have a more positive view of their relationships than those who live together or live far apart. At the same time, respondents who give to or receive from their parents a moderate amount of financial aid have more positive relationships. These results suggest that an excessive involvement with parents, which might be a result of family-dependent welfare provision in Japan, leads to decreased emotional welfare. In terms of the quantitative method, this study provides an example of how a multilevel analysis can be flexibly applied to widely available cross-sectional data.
