All Times ET
Program is Subject to Change
Stratified Multiple Imputation Estimation for Complex Business Surveys (308155)*Juana Sanchez, UCLA
Keywords: Multiple Imputation, Nonresponse, Estimation, Complex business surveys
This paper addresses two different but complementary aspects of item and unit nonresponse in complex business surveys: the prevention of nonresponse before it occurs, and the effect of multiple imputation on estimates of population parameters. A methodology based on visualization of longitudinal and cross-sampling-strata patterns of unit and item nonresponse helps gain an understanding of temporary attrition of large businesses sampled with certainty and suggests modification to the survey design to prevent future nonresponse. An estimation technique that uses stratified multiple imputation, based on the information in the longitudinal and cross-strata missing data patterns, properly accounts for the existing nonresponse patterns. After a bibliographical survey of the rare use of multiple imputation estimation in the context of complex business surveys, the methodology of this paper is placed in that context and illustrated with a 2008-2013 panel of businesses participating in the Business Research and Development and Innovation Survey, linked to the Longitudinal Business Database,. The estimates of total R&D obtained with the new methodology are compared with those obtained using weighting and substantial differences are found.