next up previous index
Next: asa.journal.jasa.tm Up: ASA Journals (6) Previous: ASA Journals (6)

asa.journal.jasa.app


ASA Journal Session: JASA Applications & Case Studies

Session Slot: 8:30-10:20 Tuesday

Estimated Audience Size: 100

AudioVisual Request: none


Session Title: JASA Applications & Case Studies Invited Paper

Theme Session: Yes

Applied Session: No


Session Organizer: Lambert, Diane Bell Labs, Lucent Technologies


Address: Bell Labs, Lucent Technologies Room 2C-256, 600 Mountain Ave. Murray Hill, NJ 07974-2008

Phone: 908 582 6509

Fax: 908 582 3340

Email: dl@bell-labs.com


Session Timing: 110 minutes total (Sorry about format):

110 minutes total....however you want Opening Remarks by Chair - 2 minutes First Speaker - 45 minutes Discussant - 15 minutes Discussant - 15 minutes Discussant - 15 minutes Rejoinder - 10 minutes Floor Discusion - 8 minutes )


Session Chair: Lambert, Diane Bell Labs, Lucent Technologies


Address: Bell Labs, Lucent Technologies Room 2C-256, 600 Mountain Ave. Murray Hill, NJ 07974-2008

Phone: 908 582 6509

Fax: 908 582 3340

Email: dl@bell-labs.com


1. Not Asked or Not Answered: Multiple Imputation for Multiple Surveys

Gelman, Andrew,   Department of Statistics, Columbia University


Address: Department of Statistics, Columbia University, New York NY, 10027

Phone:

Fax:

Email: gelman@stat.columbia.edu

King, Gary, Department of Government, Harvard University

Liu, Chuanhai, Bell Labs, Lucent Technologies

Abstract: We present a way to analyze responses from a series of independent cross-sectional surveys when some questions are not asked in some surveys and some respondents do not answer all questions posed. Our method also applies to a single survey in which different respondents are asked different questions or different sampling methods are used in different strata or clusters. In all these cases, it is impossible to impute answers for each survey or respondent separately. Therefore, we combine information across surveys and respondents using a hierarchical regression model that has covariates at the levels of respondent and survey. This model allows us to impute the missing responses for items not asked or not answered. Information from survey weights is exploited in the hierarchical model by including the variables on which the survey weights were based. Population parameters of interest are estimated by reweighting responses - both observed and imputed. Comparing imputed to non-imputed data leads naturally to diagnostics for checking the fit of the imputation model. An application to pre-election public opinion polls, in which not all the questions of interest are asked in all the surveys, that motivated the development of our model will be considered in depth.


Discussant: Binder, David   Statistics Canada


Address: Business Survey Methods Division, 11-A R.H. Coats Building, Statistics Canada, Ottawa, Ontario, CANADA K1A 0T6

Phone: 613 951 0980

Fax: 613 951 1462

Email: binddav@statcan.ca


Discussant: Judkins, David R.   Westat, Inc.


Address: Westat, Inc., 1650 Research Blvd. Rockville, MD 20850

Phone: 301 315 5970

Fax: 301 294 2034

Email: judkind1@westat.com


Discussant: Santos, Robert L.   NORC


Address: 1155 East 60th St., Chicago, IL 60637

Phone: 773 256 6081

Fax: 312 753 7886

Email: judkind1@westat.com

List of speakers who are nonmembers: None


next up previous index
Next: asa.journal.jasa.tm Up: ASA Journals (6) Previous: ASA Journals (6)
David Scott
6/1/1998