Speaker: Nicholas Nagle (University of Colorado at Boulder/University of Tennessee)
Title: Survey Weighting with Informative but Imprecise Benchmarks
Abstract: Survey weights are often adjusted so that the estimated totals match with known benchmark totals. This practice is limited by the requirement that benchmarks be perfectly known and the tendency for survey weight variability to increase as more benchmarks are included. We modify the Iterative Proportional Fitting adjustment method to incorporate benchmarks that are imprecisely known. This allows the use of benchmarks controls from sources of data that are not currently eligible for benchmarking to, such as auxiliary surveys or other incomplete records. This method also allows us to efficiently increase the number and types of benchmark data that are used for survey weighting. We present results from efforts to adjust public use microdata samples to generate estimates and microsimulations for small areas (i.e. tracts and block groups). (archived paper)
The attached preprint (to the right) provides background reading for the presentation. Also available at http://dx.doi.org/10.1080/00045608.2013.843439 (subscription may be required).
Location:
- Carnegie Mellon: contact William Eddy (bill@cmu.edu)
- Census Bureau headquarters: Room 1, contact Nancy Bates (nancy.a.bates@census.gov)
- Cornell University, Ithaca campus: Ives 109, contact Lars Vilhuber (lars.vilhuber@cornell.edu)
- Duke University: contact Jerry Reiter (jerry@stat.duke.edu)
- University of Michigan: Room 3443 ISR-Thompson, contact Maggie Levenstein (maggiel@umich.edu)
- University of Missouri: contact Scott Holan (holans@missouri.edu)
- University of Nebraska-Lincoln: Room TBD: contact: Allan McCutcheon (amccutcheon1@unl.edu)
- Northwestern University: contact Zach Seeskin (z-seeskin@u.northwestern.edu)
- Streaming video: [click here] (link active about 5 minutes after start of seminar)