Mary H. Mulry

Detecting and Treating Verified Influential Values in a Monthly Retail Trade Survey

Abstract In survey data, an observation is considered influential if it is reported correctly and its weighted contribution has an excessive effect on a key estimate, such as an estimate of total or change. In previous research with data from the U.S. Monthly Retail Trade Survey (MRTS), two methods, Clark Winsorization and weighted M-estimation, have shown potential to detect and adjust influential observations. This article discusses results of the application of a simulation methodology that generates realistic population time-series data. The new strategy enables evaluating Clark Winsorization and weighted M-estimation over repeated samples and producing conditional and unconditional performance measures. The analyses consider several scenarios for the occurrence of influential observations in the MRTS and assess the performance of the two methods for estimates of total retail sales and month-to-month change.

Setting M-estimation Parameters for Detection and Treatment of Influential Values

Abstract Recent research on the use of M-estimation methodology for detecting and treating verified influential values in economic surveys found that initial parameter settings affect effectiveness. In this article, we explore the basic question of how to develop initial settings for the M-estimation parameters. The economic populations that we studied are highly skewed and are consequently highly stratified. While we investigated settings for several parameters, the most challenging problem was to develop an “automatic” data-driven method for setting the initial value of the tuning constant φ, the parameter with the greatest influence on performance of the algorithm. Of all the methods that we considered, we found that methods defined in terms of the accuracy of published estimates can be implemented on a large scale and yielded the best performance. We illustrate the methodology with an empirical analysis of 36 consecutive months of data from 19 industries in the Monthly Wholesale Trade Survey.

Comparison of 2010 Census Nonresponse Follow-Up Proxy Responses with Administrative Records Using Census Coverage Measurement Results

Abstract The U.S. Census Bureau is currently conducting research on ways to use administrative records to reduce the cost and improve the quality of the 2020 Census Nonresponse Followup (NRFU) at addresses that do not self-respond electronically or by mail. Previously, when a NRFU enumerator was unable to contact residents at an address, he/she found a knowledgeable person, such as a neighbor or apartment manager, who could provide the census information for the residents. This was called a proxy response. The Census Bureau’s recent advances in merging federal and third-party databases raise the question: Are proxy responses for NRFU addresses more accurate than the administrative records available for the housing unit? Our study attempts to answer this question by comparing the quality of proxy responses and the administrative records for those housing units in the same timeframe using the results of 2010 Census Coverage Measurement (CCM) Program. The assessment of the quality of the proxy responses and the administrative records in the CCM sample of block clusters takes advantage of the extensive fieldwork, processing, and clerical matching conducted for the CCM.