|
Activity Number:
|
327
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Wednesday, August 11, 2004 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Survey Research Methods
|
| Abstract - #300642 |
|
Title:
|
Multiple System Estimation with Erroneous Enumerations
|
|
Author(s):
|
Paul P. Biemer*+ and G. Gordon Brown and Dean H. Judson
|
|
Companies:
|
RTI International and UNC-CH and RTI International and U.S. Census Bureau
|
|
Address:
|
PO Box 12194, Research Triangle Part, NC, 27709-2194,
|
|
Keywords:
|
latent class analysis ; population census ; undercount ; finite mixture models ; correlation bias
|
|
Abstract:
|
This paper presents a latent class modeling approach for multiple system estimation that accounts for varying levels of incompleteness and undetected erroneous enumerations in the population lists. Our approach assumes that one of the lists is based upon administrative records with errors that are assumed to be locally independent of the other enumeration-based lists. For k >2 lists, the resulting data take the form of an incomplete 2k contingency table which can be represented by a latent class model where the latent variable is an individual's residency true status (i.e., resident or nonresident of the population). Latent class analysis is used to estimate the expected values of the observed cells of this table and then to project these estimate onto the unobserved cells in order to estimate the total number of population members. Using artificial populations, we evaluate the improvement in mean squared error using this approach compared with other log-linear estimation approaches from the census undercount and capture-recapture literature.
|
