2000 Volume 15 Issue 1 Pages 165-180
Missing data occur from a refusal or nonresponse in the process of social surveys. The amount of missing data is recently increasing, and raising serious problems for sociological researchers. They mostly treat respondents that include missing data as “incomplete cases,” and exclude them all from an analysis. This procedure gives a bias to the result of any analysis. In this paper, we show that the imputation of missing data by log-linear models is useful for appropriate treatment of missing data. We took a contingency table of intergenerational mobility in education as an example and compared the results of two analyses; one considered an impact of missing data, the other didn't at all. We found that the analysis excluding missing data overlooked the large amount of downward mobility. Especially, the mobility from secondary education of fathers to compulsory education of children was distorted.