Some Issues in Analytic Inference from Complex Surveys

Colorado State University, Fort Collins CO

Investigators

Abstract

This award is funded under the American Recovery and Reinvestment Act of 2009 (Public Law 111-5). Surveys represent a key source of data in a wide range of scientific disciplines, and are often an important input in public and corporate decision-making processes. While the primary use of survey data is to describe the population from which the sample was drawn, analysts are increasingly using those data to make inference about statistical models. This use of survey data in model estimation and analysis is referred to as analytic inference for survey data. Survey data producing organizations usually recommend that survey weights be used in all model fitting with survey data, in order to account for the sampling design and the post-sampling adjustments. This often significantly complicates the work of data analysts, in large part because the effect of survey weighting in analytic inference is still incompletely understood. This research will explore some long-standing open issues in analytic inference for surveys. First, within the survey-weighted mode of inference, the PIs will investigate improved variance estimation methods and distributional theory. Second, they will evaluate a number of alternatives to survey weighting in model fitting and inference. This research will advance the state of knowledge in how to properly conduct analytic inference for survey data by bringing together concepts, methods and approaches from survey statistics and other areas of statistics. Ultimately, it will enhance the ability of data analysts to conduct perform analytic inference for survey data. Because of the importance of survey data collection and analysis in so many disciplines, the results of the proposed research will be widely applicable.