This paper examines methods to correct for biases in auxiliary data (e.g. voter lists, consumer databases) using higher quality reference data (e.g. census data) in order to produce better ratio estimates (regression estimates). Survey practitioners are increasingly eager to use auxiliary data (frame information) for statistical inference. This is particularly true in light of a rapidly deteriorating survey environment driven by budget cuts in government agencies and more stringent laws (ex. TCPA) driven by privacy concerns.
Some innovative practitioners started using population frames such as voter lists and consumer databases for survey inferences, especially for political campaign work after 2012 (Nickerson & Rogers, 2014). The predictive scores that campaigns and some partisan polling firms use are in essence, regression estimators derived from both survey data and auxiliary data. Outside of political campaigns, such model based approach to obtain more accurate (in terms of MSE) estimates are already in use in the fields of small area estimation methodology.
In theory, with a good frame data and a good model, researchers should be a
|