NiNiProblem statement
Post-pandemic evidence on remote work and innovation is contradictory, and the contradiction is not random: studies that measure output volume tend to find no penalty, while studies that measure novelty tend to find one. That pattern suggests the disagreement is about the dependent variable, not about the world.
This study takes the position that distance does not reduce ideas; it reshapes the network that carries them. Co-located teams generate novelty through incidental strong-tie contact. Distributed teams lose that channel but gain reach into weak ties across the organisation. The net effect is an empirical question that has rarely been posed as one.
Research questions
- RQ1. How does the share of remote workdays relate to the novelty of a team's output, holding output volume constant?
- RQ2. Does weak-tie contact across organisational units mediate that relationship?
- RQ3. Does the relationship differ for teams doing exploratory versus exploitative work?
Data and setting
Two years of anonymised collaboration metadata from a single multinational firm — message graphs, meeting graphs and document co-authorship — joined to a quarterly innovation survey already run by the firm. Metadata is used at the team-week level; no message content is accessed.
Novelty is operationalised as the semantic distance between a team's deliverables and the firm's prior corpus, computed with sentence embeddings and validated against expert ratings on a held-out sample of 200 deliverables.
Identification strategy
The firm's staggered return-to-office mandate applied to business units on different dates for reasons unrelated to team performance, which supports a difference-in-differences design with team and time fixed effects. Parallel-trends assumptions are tested on the pre-mandate period and reported whether or not they hold.
Limitations
One firm, one industry, and an embedding-based novelty measure that rewards lexical difference. Expert validation bounds but does not remove that risk. Findings speak to knowledge work with digital deliverables and should not be extended to lab or field research settings.