Measurement errors and estimation biases with incomplete social networks: replication studies on intra-firm inventor network analysis

Abstract

The number of studies exploring the role of intra-organizational inventor networks in firm knowledge recombination and diffusion processes has surged in recent years. The typical approach of these studies–to construct intra-firm inventor networks based on archival patent grant data–suffers from a major issue: incomplete data. This incomplete data may have implications for network measures and regression estimates based on these measures. To shed light on these implications, this study explores the consequences of missing data for inventor network studies in the field of technology invention. We do so by comparing networks based on granted patent data–the incomplete data–with networks based on patent application data that also include failed patent applications–the more complete data. The findings from replications of two prior studies–one firm-level study and one inventor-level study–indicate that intra-firm network measures are systematically biased for both network-level and inventor-level measures and cause bias in regression estimates. We further find that these systematic measurement errors are also statistically significantly related to the outcomes in the studies, thereby implying omitted variable biases in the effect estimates in prior studies. These findings have implications for research on networks of scientific collaborations specifically, and networks based on incomplete archival data more generally.