This paper measures the causal effect of office buildings on collaboration among information workers.

We estimate building quality using a mover design using rich data on emails, messages, and co-authored documents between Microsoft employees. This design computes building quality as the increase in high-frequency collaboration observed after a person moves to a building relative to similar employees who did not move. Our measure of building quality provides a comparable ranking of buildings. Using it, we document vast differences in building quality, with buildings at the 75th percentile of the quality distribution increasing collaboration by 56% more than those at the 25th percentile. These differences do not reflect regional patterns and are partly explained by the building's design attributes. A 10 p.p increase in the share of dining spaces instead of offices increases building quality by 9%