Sponsoring Section/Society: ASA-Business and Economic Statistics
Session Slot: 2:00- 3:50 Sunday
Estimated Audience Size: 100 dws
AudioVisual Request: none dws
Session Title: Multiple Imputation of Income and Assets in Economic Surveys:
Application and Assessment
Theme Session: Yes/No
Applied Session: Yes/No
Session Organizer: Paulin, Geoffrey D. Bureau of Labor Statistics
Address: BLS-DCES 2 Mass. Ave. NE (PSB 3985) Washington DC 20212
Phone: 202-606-6870 x200
Fax: 202-606-7006
Email: paulin_g@bls.gov
Session Timing: 110 minutes total (Sorry about format):
110 minutes total Opening Remarks by Chair - 5 or 0 minutes First Speaker - 30 minutes (or 25) Second Speaker - 30 minutes Third Speaker - 30 minutes Discussant 1 - 10 minutes (or none) Discussant 2 - 10 minutes (or none) Floor Discusion - 10 minutes (or 5 or 15)
Session Chair: Rubin, Donald B. Harvard University
Address: Cambridge, MA 02138
Phone:
Fax:
Email:
1. Multiple Imputation of Income in the Consumer Expenditure Survey: Evaluation of Statistical Inferences
Raghunathan, Trivellore E., University of Michigan
Address:
Phone:
Fax:
Email: teraghu@umich.edu
Paulin, Geoffrey D., Bureau of Labor Statistics
Abstract: Non-response is a problem common to many surveys. For example, many respondents fail to report incomes for some or all working members of their families in the U.S. Consumer Expenditure Interview Survey. Because these data are so important to economic and other analyses, a complete set of data is desirable. The U.S. Bureau of Labor Statistics and U.S. Bureau of the Census have conducted investigations that use model-based, multiple imputation methods as a viable way of obtaining valid inferences when the data are subject to nonresponse. In the course of this research several approaches have been examined for both theoretical and practical reasons (including feasibility of implementation). This paper first examines four approaches: "Partial" Bayesian (with and without total expenditures as a predictive variable) and "Fully" Bayesian (with and without total expenditures as predictor). Although the "Fully" Bayesian model with total expenditures as predictors is found to be the best model of those tested from a statistical standpoint, the use of expenditures as predictors in the imputation model, at least in theory, may have implications for econometric analyses using these data. The second part of this paper examines the implications of not including expenditures as predictors in the imputation model by analyzing inferences based on certain econometric models.
2. Multiple Imputation in the Survey of Consumer Finances: Experience from the 1989-1998 Surveys
Kennickell, Arthur, Federal Reserve Board
Address: 20th and C Streets NW, Wash. DC 20551
Phone: 202-452-2247
Fax:
Email: m1abk00@FRB.GOV;
Abstract: The SCF collects information on a broad range of financial variables, many of which are highly skewed in the population. Given the sensitive nature of some of the survey questions, item nonresponse is a serious concern.Beginning in 1989, the Survey of Consumer Finances (SCF) has used multiple imputation (MI) to compensate for missing data. This development arose from a compelling need for logically consistent procedures that would not mask the level of uncertainty in imputation. Overall, the SCF experience with MI has been very successful. This paper discusses the implementation of MI for the SCF, and the reception of multiply-imputed data by users.
Discussant: Bradley, Ralph Bureau of Labor Statistics
Address: 2 Mass. Ave. NE Wash, DC 20212
Phone: 202-606-6573
Fax:
Email:
Discussant: Folsom, Jr., Ralph E. (not confirmed) Research Triangle Institute
Address: Research Triangle Institute PO Box 12194 RTP, NC 27709-2194
Phone:
Fax:
Email:
List of speakers who are nonmembers: None