Extended Data Table 1 Assessing the robustness of declined disruption with increased collaboration distance in science
- From our dataset of scientific teams, we selected 7,681,669 scientists who published two or more papers. These scientists have published 13,711,470 papers between 1960 and 2020, yielding 45,078,179 paper-author records. We use this dataset to build stepwise regression models and explore the robustness of the relationship between remote collaboration (the value equals one if team members spread across cities, zero otherwise) and disruption, starting from a model without any controls, and then adding team size, time period, average career age, knowledge diversity, tie strength, the field of study, author fixed effects, and finally, an interaction term between time and remote. We note that the remote work penalty—the negative relationship between remote collaboration and disruption—is robust across all specifications. When teams move from 0 km to more than 600 km collaboration distance, for example, the predicted disruption probability, holding other variables constant, declines from 20.4% to 19.5% (p-value < 0.001 for two-side Student’s t-test), or 4.4% in relative terms.
- Note: All statistical tests are two-sided t-test and no adjustments were made for multiple comparisons. For Model 5-7, standard errors (in parentheses) are clustered at the author level. * p < 0.05; ** p-value < 0.01; *** p-value < 0.001. We used the REGHDFE package in STATA1646 to implement the fixed-effects regressions.
