Search

This paper investigates how two different methods of multiple imputation compare to casewise deletion as a method of dealing with missing data in the presence of categorical data. We ask the question, for a typical model found in sociological research on families that comes from a widely used dataset with typical missing data patterns, could substantive conclusions differ based on choice of method used for handling the missing data? Using data from the National Survey of Families and Households (NSFH) we test how less refinement of the dependent variable may influence the results we test models with three dependent variables measuring fertility intentions, one continuous, one 3-level categorical, and one dichotomous. Results show a clear difference in the substantive conclusions that a researcher would draw due to how the missing data was handled in the context categorical data. Both the SGR and MCMC imputation models performed nearly identical and a researcher would have reached the same conclusions about what variables were significant. However, using casewise deletion compared to either of the MI models would have resulted in substantive differences. Since a limitation of this work is that we do not which model most closely approximates the "true" parameter estimated in the population, the next step for further research would be to conduct a simulation of similar comparisons where results from the different methods can be compared to known values. ..PAT.-Unpublished Manuscript [ABSTRACT FROM AUTHOR]