The 131st Annual Meeting (November 15-19, 2003) of APHA

Abstract #63708

Ming Yin, PhD¹, Richard D. Cohn, PhD¹, Darryl C. Zeldin, MD², and Samual J. Arbes, PhD². (1) Analytical Sciences, Inc., 2605 Meridian Parkway, Suite 200, Durham, NC 27713, 9193137631, myin@asciences.com, (2) Division of Intramural Research, NIH/NIEHS, 111 T.W. Alexander Drive, Buiding 101, D236, Research Triangle Park, NC 27709

Recently, Schafer and Schenker (2000, JASA) proposed an analytic method to produce appropriate variance estimates for conditional predictive mean imputation. Their method only deals with ignorable missing. We extend their method to a more general setting including both types of missing data, where the non-ignorable missing mechanism is known up to a few parameters. This method is based on asymptotic expansions of point estimators and their associated variance estimators and produces a first order approximation to Rubin’s repeated-imputation inference with an infinite number of imputations. It can be more efficient than multiple imputation with a small number of imputations. Next, we show this newly developed analytic method can be successfully applied to data from The First National Survey of Lead and Allergens in Housing. One of the analysis goals of the Survey is to estimate the percentage of households nationally with elevated allergen levels above some threshold, e.g., cockroach allergen (Blag 1) concentration >= 0.1 U/gm. Two kinds of missing data are introduced in the survey. First, a portion of allergen measurements is completely missing. Second, some measurements are below the lower limit of detection where the limits vary from one assay to another. Complications are introduced when an important threshold is coincident with one of the smaller values of lower detection limits.

Learning Objectives:

• Participants in the session will be able to learn a new and effective technique to analyze data with both ignorable and non-ignorable missings. These kinds of missing data are common in public health related survey and laboratory studies.

Keywords: Biostatistics, Public Health

Presenting author's disclosure statement:
I do not have any significant financial interest/arrangement or affiliation with any organization/institution whose products or services are being discussed in this session.