Journal of Official Statistics, Vol.18, No.4, 2002. pp. 559–576
Using Administrative Records to Improve Small Area Estimation: An Example from the U.S. Decennial Census
Elaine Zanutto and Alan Zaslavsky
Abstract:We present a small area estimation strategy that combines two related information sources: census data and administrative records. Our methodology takes advantage of administrative records to help impute small area detail while constraining aggregate-level estimates to agree with unbiased survey estimates, without requiring the administrative records to be a perfect substitute for the missing survey information. We illustrate our method with data from the 1995 U.S. Decennial Test Census, in which nonresponse follow-up was conducted in only a sample of blocks, making small area estimation necessary. To produce a microdata file that may be used for a variety of analyses, we propose to treat the unsampled portion of the population as missing data and impute to complete the database. To do so, we estimate the number of nonrespondent households of each ``type' (represented by a cross-classification of categorical variables) to be imputed in each small area. Donor households for these imputations can be chosen from the sampled nonresponse follow-up sample, the respondent households, or the administrative records households (if they are of sufficient quality). We show, through simulation, that our imputation method reduces the mean squared error for some small area (block-level) estimates compared to alternative methods.
Keywords:Imputation; missing data; nonresponse follow-up; iterative proportional fitting; loglinear models; mass imputation.
Copyright © Statistics Sweden 1996-2018. Open AccessISSN 0282-423XCreated and Maintained by OKS Group