Assessing Nonresponse Bias and Measurement Error Using Statistical Matching

The estimation of nonresponse bias and measurement error share the problem of usually not having a criterion to assess the quality of the estimate. Nonresponse bias analysis often uses responders within the survey sample who are in some way similar to nonresponders to estimate the potential bias. This depends on the variables within the survey being related to both the likelihood of responding and also being related to the measure being estimated. An alternative method uses record linkage to get data about nonresponders from another source, often administrative data. The record linkage studies are limited by what data might be available and the quality of the linkage. Measurement error studies have similar problems with estimating error. Some methods include using the internal consistency of responses, reinterviews, or record linkage to provide a measure of the quality of response. Statistical matching uses surveys from different samples which have some similarities to compare estimates. The matching can be based on demographics and other sample characteristics which are thought to be related to the survey estimates. Propensity scores are one type of variable often used in matching.