|
Activity Number:
|
222
- Administrative Record Research for the 2020 Census
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Monday, July 31, 2017 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Government Statistics Section
|
|
Abstract #324457
|
View Presentation
|
|
Title:
|
Alternative Methods for Selecting Households for Administrative Record Enumeration in Place of Nonresponse Follow-Up for the 2020 Census
|
|
Author(s):
|
Richard Griffin*
|
|
Companies:
|
U.S. Census Bureau
|
|
Keywords:
|
Linear Programming ;
Likelihood Ratio ;
Receiver Operating Curve
|
|
Abstract:
|
Plans for the 2020 Census include using two models for assigning a proportion of the Nonresponse Follow-up (NRFU) workload for Administrative Record Enumeration (ADRECENU) in place of NRFU after one visit. A linear programming approach has been tested which allows for maximizing the number of housing units identified for ADRECENU subject to constraints. This paper describes three alternative methods to assign NRFU workload to ADRECENU using the predictions from these two models and compares the results from a sample of NRFU housing units using a definition of true positive and false positive rates based on both models. One alternative uses a likelihood ratio in the mode of the Fellegi-Sunter (1969) process. The second alternative uses the minimum Euclidean distance to the optimal point of a Receiver Operating Curve (ROC) to define a set of cut-offs. The third alternative is a simple approach to define the Euclidean Distance to the (1,1) point for each pair of prediction probabilities and ranking from smallest to largest distance. Alternatives are compared based on proportion designated for administrative record enumeration and true positive and false positive rates.
|
Authors who are presenting talks have a * after their name.
