Journal of Official Statistics, Vol.3, No.4, 1987. pp. 403–418
Covariances of Measurement Errors in Survey Responses
Willard L. Rodgers and A. Regula Herzog
Abstract:Using data from administrative records, census counts, and maps to assess errors in responses to a wide range of survey items, we investigated discrepancies in survey answers from interviews with a sample of the adult population in a metropolitan area. We found evidence that commonly made assumptions about the independence of measurement errors are often incorrect. Specifically, we found: (1) substantial correlations between discrepancy scores on most of the survey items and the values from the validating sources; (2) frequent, though generally small, correlations between discrepancy scores on one variable and measured values on other variables; and (3) that discrepancy scores on different variables tend to be somewhat correlated with one another only if the variables are within a single topic area. Such violations of assumptions can have profound effects on estimates of population parameters from survey data; for example, there may be substantial biases in estimates of regression coefficients.
Keywords:Measurement error; validation; matching; estimation.
Copyright © Statistics Sweden 1996-2018. Open AccessISSN 0282-423XCreated and Maintained by OKS Group