The Accuracy of Imputations in the American Community Survey: Evidence from Comparisons to the National Longitudinal Survey of Youth

When respondents to the American Community Survey (ACS) do not respond to particular questions, the Census Bureau imputes a response. For each variable that contains imputations, an imputation variable indicates which cases have imputed values. Although the frequency of imputation is clear to data users, the accuracy of the imputation is unclear. We assess the accuracy of ACS imputations by comparing ACS data to data for the same individuals from the National Longitudinal Survey of Youth (NLSY). The NLSY79 cohort includes individuals born 1957-1964, while the NLSY97 cohort includes individuals born 1980-1984. The ACS and NLSY questionnaires overlap in a large number of topic areas, including age, race/ethnicity, place of birth, education, employment, and marital status. By linking the ACS to the NLSY79 and NLSY97 through PIK crosswalks, we compare NLSY responses to imputed ACS responses and to reported (i.e., not imputed) ACS responses. (NLSY does not impute for item non-response, so all NLSY data are reported.) For each variable, we split the matched sample into 2 subsamples: those for which the ACS variable is imputed and those for which the ACS variable is reported. Then we compute measures of similarity between ACS and NLSY responses for each subsample. The results identify specific topical areas that produce more or less agreement between NLSY responses and imputed ACS responses.