The Effects of Editing and Imputation on Measured Plant-Level and Aggregate Productivity Growth in a Panel of US Manufacturing Plants

• Activity Number: 96 - Advances in the Measurement of Inputs, Output, and Productivity in the US Federal Statistical System
• Type: Topic Contributed
• Date/Time: Monday, August 8, 2022 : 8:30 AM to 10:20 AM
• Sponsor: Quality and Productivity Section
• Author(s): Thomas Kirk White* and Hang Kim and Martin Rotemberg
• Companies: U.S. Census Bureau and University of Cincinnati and New York University
• Keywords: statistical editing; imputation; productivity; enterprise data; panel data; Bayesian

Abstract

We investigate the effects of editing and imputation in the U.S. Census Bureau’s Annual Survey of Manufactures (ASM) on measures of aggregate and plant-level productivity growth. Previous research found that the Census Bureau’s imputations in the ASM, particularly regression imputes, tend to decrease measured within-industry productivity dispersion (White, Reiter and Petrin 2018; Rotemberg and White 2021). We add longitudinal ratio-of-ratios edits to the simultaneous Bayesian edit-imputation method of Kim et al. (2015), applied to the 2009-2013 and 2014-2018 panels of the ASM. We estimate plant-level and aggregate productivity growth in the manufacturing sector in the raw reported data (i.e., before editing and imputation), the Census Bureau’s edited-imputed data, and the Bayesian edited-imputed data.