Imputation of Categorical Data with Small Subcategories: Assessing the Results of Imputing Race in the 2007 - 2008 National Ambulatory Medical Care Survey (NAMCS)

• Activity Number: 471
• Type: Contributed
• Date/Time: Wednesday, August 3, 2011 : 8:30 AM to 10:20 AM
• Sponsor: Section on Survey Research Methods
• Author(s): Elizabeth Goldberg*+
• Companies: Centers for Disease Control and Prevention
• Keywords: imputation ; categorical data ; IVEware

Abstract

The National Ambulatory Care Survey is a national probability sample survey of visits to nonfederal office-based physicians in the United States. Of particular interest to many researchers in health care policy is the effect of race on health care usage and outcomes, but race was missing at relatively high rates, peaking in 2008 at 33%. Evaluating the effectiveness and utility of the results of such imputations in categorical data, especially data with small sub categories, can be challenging. The evaluation of race imputation in NAMCS is a case study in how to evaluate categorical imputation. Graphical methods were used to compare imputed vs. non-imputed data. In addition, race was set to missing for a subsample of known cases in order to evaluate whether the model correctly imputed these known cases. The differences in the accuracy of the model were interesting because some very small categories were imputed with far more accuracy than some larger sub groups. This paper discusses the methodology and the findings for the research.