Measuring labour productivity is difficult. This is in part why recent studies on the productivity impact of telework focus on occupations where productivity is easily measured, such as call centers. One would think that matched employer-employee panel survey data on telework and self-reported productivity collected for stayers who experience an exogenous change in their number of telework days might provide an alternative. Since these data allow researchers to analyze changes in employees’ self-reported productivity while controlling for firm fixed effects and ruling out reverse causality, they might identify the impact of telework on stayers’ true productivity.

We show that even with such data, the impact of telework on productivity is not identified if telework affects self-reported productivity by affecting not only true productivity but also other factors—such as satisfaction with one’s work arrangement—that likely depend on true productivity. Under these conditions, not controlling for these other factors leaves the other factors-to-self-reported productivity channel open but controlling for them leads to collider bias (Pearl and Mackenzie, 2018; Imbens, 2020), i.e. generates a spurious correlation between telework and self-reported productivity.