Leveraging Auxiliary Information on Marginal Distributions in Nonignorable Models for Item and Unit Nonresponse
When handling nonresponse, government agencies and survey organizations typically are forced to make strong, and potentially unrealistic, assumptions about the reasons why values are missing. We present a framework that enables users to reduce reliance on such assumptions by leveraging information from auxiliary data sources, such as administrative databases, on the marginal distributions of the survey variables. We use the framework to impute missing values in voter turnout in the U.S. Current Population Survey (CPS), showing how it allows us to make more reasonable assumptions about the missing values than those used in previous CPS turnout estimates.
READ FULL TEXT