The impact of the 'open' workspace on human collaboration

Ethan S. Bernstein; Stephen Turban · article · 2018-07-02 Reader 🖥️

Coordination Cost Multiplayer vs Single-Player AI Organizational Legibility

Removing spatial boundaries to make offices 'open' and transparent does not increase collaboration; across two corporate field studies it cut face-to-face interaction by roughly 70% and prompted a partial, compensating rise in email/IM, because heightened observability triggers a natural human response to restore privacy by withdrawing from co-present peers and re-routing interaction to electronic channels.

theoretical · empirical — Peer-reviewed field study (Philosophical Transactions B, 2018) reporting two intervention-based quasi-field experiments. Theoretical in Adler's sense: it establishes what is true about how open architecture changes human interaction, rather than prescribing what managers should do. It carries strong practical implications for office/org design, but its own contribution is descriptive-causal and evidentiary. Form is empirical/quantitative: interrupted-time-series comparisons of sensor-measured behavior, OLS with fixed effects, plus latent-space network models.

Structure

Part	Role
Framing and the divided prior theory (Introduction)	Sets up the empirical question and the tension it resolves
Study 1 — OpenCo1 'war on walls' (methods + results)	Primary evidence
Study 2 — OpenCo2 replication with dyadic detail (methods + results)	Conceptual replication, robustness, and mechanism-narrowing
Discussion — three cautionary tales + summary	Interpretation and generalization the authors claim

Key terms

open / unbounded / boundaryless office
A workspace from which spatial boundaries (walls, doors, cubicle dividers) have been removed so that a whole floor is open, transparent and everyone is mutually observable; in Study 2 concretely, cubicles replaced by undivided, unwalled tables of six-to-eight, with desks still ~2 m apart.
spatial boundaries
Physical partitions that modularize the environment and perform a functional role: clarifying who is watching and who is not, who belongs, who controls what, to whom one answers — thereby constraining interaction and mitigating overload, distraction and other symptoms of bounded rationality.
transparency
Architectural observability — the property that everyone can see everyone else all the time. The paper's ironic hinge: this physical 'transparency' is the very thing that suppresses interaction (echoing Bernstein's transparency-paradox line of work).
collaboration / collective intelligence
Operationalized as measured interaction: F2F minutes plus email/IM message volume. Collective intelligence is 'distributed intelligence arising from the social interaction of individuals' that predicts group performance more than individual ability. The paper slides between interaction (what it measures) and collaboration/collective intelligence (what it infers).
face-to-face (F2F) interaction
A machine-defined event, measured in minutes: two or more sociometric badges facing each other (uninterrupted IR line-of-sight), detecting alternating speech, within 10 m, sustained until any criterion lapses for >5 s. The primary dependent variable.
sociometric badge
A wearable sensor logging behavior at 10 ms resolution: IR for whom one faces, microphone for whether one speaks/listens (not content), accelerometer for movement/posture, Bluetooth for location. Provides objective behavioral data replacing self-report — the methodological linchpin.
propinquity
Physical proximity as a predictor of social interaction — 'one of the most robust findings in sociology' and the prior the paper challenges: in the data, openness dwarfs distance, and distance affects F2F about twice as much as email but is small overall.

Propositions

explicitcorroboratedAcross two intervention field studies at Fortune-500 corporate headquarters, transitioning from cubicles to open, boundaryless offices reduced face-to-face interaction by roughly 70% (72% at OpenCo1: 5266 down to 1492 minutes over 15 days, ~5.8 to ~1.7 h/person/day; 67-71% at OpenCo2).conf: high
supports ←Slik påvirker de ulike kontorløsningene ansattes trivsel og «emosjonelle utmattelse» · p1 — Loose, same-direction alignment at the thesis level only: both hold that office architecture is a material lever on employee outcomes and that fashionable/assumed layouts are not automatically best (Bodin Danielsson's 'don't chase trends'; Bernstein & Turban's finding that a deliberate open-office move backfired). The outcome domains genuinely differ — Bernstein#p1 measures a ~70% drop in face-to-face interaction, whereas this piece concerns wellbeing/emotional exhaustion — and the truncated Norwegian teaser names no winning layout and makes no open-plan-specific claim. The support is thematic, not a match of specific empirical findings.
explicitThe lost F2F time was accompanied by an increase in electronic interaction, but only a partial and asymmetric one: at OpenCo1 emails sent rose 56%, received (To) 20%, cc 41%, IM messages 67%, IM words 75%; at OpenCo2 email rose 22-50%. Electronic channels took up only 'some of the slack' — F2F fell by thousands of minutes while electronic rose by tens of messages, so this is substitution of channel, not a one-for-one replacement of interaction.conf: high
explicitThe proposed mechanism is a balancing feedback loop of privacy restoration: increasing observability ('transparency') triggers a natural human response to preserve privacy by socially withdrawing from co-present peers (e.g., headphones, avoidance) and re-routing interaction to email/IM rather than converse in front of a visible audience of colleagues.conf: high
extendsTBM 426: The Trouble With Mirrors · p13 — Cutler argues organizational change is gated by relational/political conditions, so improving visibility/legibility does not by itself move the system. Bernstein & Turban supply hard, cross-domain behavioral evidence and name the specific relational condition: increasing physical visibility (an open, transparent floor) moved the system the wrong way because a privacy-preservation response dominated the information/visibility gain, cutting the very F2F interaction it was meant to create.
explicitPhysical proximity (propinquity) has only a small effect on F2F interaction and a negligible effect on email, and is dwarfed by the effect of openness itself — challenging the robust sociological prior that proximity is the primary driver of workplace interaction.conf: high
explicitF2F and electronic (email) interaction networks respond differently — on some measures in opposite directions — to the same change in the built environment, and changes in one fail to predict changes in the other. An office redesign therefore alters not just the channel and volume of interaction but the interaction topology: whom a person interacts with.conf: high
implicitThe study measures interaction flows, not collaborative output: 'collaboration' and 'collective intelligence' are operationalized entirely as interaction volume (F2F minutes plus electronic message counts), and reduced F2F is treated as reduced collaboration — but decision quality, work output and performance are never directly measured.conf: high
implicitThe F2F-to-electronic substitution is assumed to be net-negative for productivity, but the only outcome evidence is one firm's executives qualitatively reporting a productivity decline; the causal chain 'less F2F -> worse collaboration -> lower performance' is otherwise imported from prior media-richness theory rather than established in these data.conf: low
implicitThe withdrawal response is framed as a 'natural human response' and thus implicitly universal, yet it is observed only in two US Fortune-500 corporate headquarters over short windows (~2 months after the move); its generality across national cultures, organization types, office designs and long time horizons is assumed, not tested.conf: medium
explicitObjective, sensor-based behavioral measurement (sociometric badges plus server logs of email/IM) is methodologically superior to the self-report and observer-interpretation methods that dominated prior open-office research, and this measurement upgrade is what makes the counterintuitive result credible.conf: high

Arguments

Removing spatial boundaries to create open, transparent offices reduces rather than increases collaborative face-to-face interaction. strength: high

Explicit premises:

Two independent field studies both show F2F interaction falling ~70% after the move to open space.
Interaction was measured objectively (sociometric badges + email/IM server logs), not by self-report.
Results are robust to sensitivity analyses (alternative distances, lag times, speaking patterns) and replicate across two firms and two open designs.
In Study 2, the openness effect dwarfs the effect of physical distance between workstations.

Implicit premises:

F2F-interaction volume is a valid proxy for collaboration and collective intelligence.
Privacy is a fundamental human drive that reasserts itself when observability increases.
Cubicles are the appropriate baseline/counterfactual for judging 'open' offices.
The ~2-month post-move observation window reflects a settled steady state rather than a transient adjustment or novelty phase.

Inferences:

Authors gesture at, but do not establish, that the net effect on organizational output is negative (one firm's exec self-report + borrowed media-richness theory).
The result is presented as a general human tendency, though only two US Fortune-500 HQs were studied.
The effect is treated as stable, though long-run equilibrium was not observed.
Systems reading the authors do not draw explicitly: the intervention targeted one interaction flow (intended more F2F) and set off an unmodeled balancing loop (privacy restoration) that overwhelmed the intended effect.

The substitution is driven by a privacy-restoration response to increased observability, not by changes in physical proximity.

Explicit premises:

The openness effect dwarfs the small distance effect on F2F and the negligible one on email.
The pattern is consistent with the fundamental human desire for privacy and with evidence that privacy can raise productivity.
F2F and email networks decouple, indicating fundamentally different mechanisms drive the two channels.

Implicit premises:

Increased observability is the operative variable, rather than confounds such as noise, changed adjacencies/neighbors, or altered tasks.
The privacy-preserving response is a stable, generalizable human tendency.

Inferences:

Alternative mechanisms (noise avoidance, changed physical adjacency of collaborators, Hawthorne/novelty effects) are not ruled out, because the mediating variable — actual privacy-seeking behavior — is inferred from volume shifts rather than measured.

Researchers must not treat electronic-interaction data as a proxy for face-to-face interaction. strength: high

Explicit premises:

F2F and email networks respond differently, and on some measures oppositely, to the same built-environment change.
Across latent-space (LPCM) and machine-learning (Random Forest) models, changes in one network fail to predict changes in the other.

Implicit premises:

The two networks capture genuinely different social processes rather than noisy versions of the same process.

Inferences:

Prior studies (e.g., Enron email analyses) that inferred F2F patterns from electronic data may be systematically misleading.

Narratives

OpenCo1's 'war on walls': a Fortune-500 multinational completely transformed one wall-bounded floor of its global headquarters into an open, transparent, boundaryless space, hoping to spark more F2F collaboration. Among 52 tracked employees, F2F time fell 72% (from ~5.8 to ~1.7 h/person/day) while email and IM rose, and executives reported that productivity declined.
→ Primary evidence and the counterintuitive punch — the firm engineered openness specifically to increase F2F interaction and got 72% less of it.
OpenCo2's replication: during a multi-year HQ redesign, 100 employees (1830 dyads) moved from cubicles to undivided tables of six-to-eight. F2F fell 67-71% and email rose 22-50%; physical distance mattered far less than openness; and network models showed F2F and email networks moving independently, even in opposite directions.
→ Conceptual replication that adds robustness, rules proximity out as the driver, and establishes the network-decoupling finding.
The everyday move: in the open floor, 'rather than have an F2F interaction in front of a large audience of peers, an employee might look around, see that a particular person is at his or her desk, and send an email' — while others don large headphones to isolate themselves and appear busy because everyone can see them.
→ Makes the privacy-restoration mechanism concrete and relatable; shows the balancing loop as an individual micro-decision repeated across the floor.

Questions addressed

✅ Does removing spatial boundaries to create open offices increase face-to-face interaction?
No. In both field studies F2F interaction fell by about 70% after the transition to open space.
✅ What happens to electronic interaction (email, IM) when offices go open?
It rises (email up 20-56%, IM up 67-75% in Study 1; email up 22-50% in Study 2), taking up some but not all of the lost F2F interaction.
✅ Does physical distance between workstations explain the change in interaction?
No. Distance has only a small effect on F2F and a negligible effect on email; the openness effect dominates it.
✅ Are F2F and electronic interaction networks good proxies for one another?
No. They respond differently (sometimes oppositely) to built-environment change; changes in one do not predict changes in the other, so an office redesign also changes whom people interact with.
⭕ Exactly which factors beyond proximity and visibility govern the response, and how does open space affect collective intelligence?
Acknowledged as open: 'until we understand those factors...' and 'we still have much to learn about how collective intelligence works.' The privacy mechanism is proposed but not measured.
⭕ Did reduced F2F interaction actually harm collaborative output or performance?
Not established. Interaction was measured; output was not, apart from one firm's executives qualitatively reporting a productivity decline. The paper treats reduced F2F as reduced collaboration without validating that proxy.

Gaps

incompleteInteraction is measured; outcomes are not. The study equates a ~70% drop in F2F-interaction minutes with reduced collaboration and productivity, but the only outcome evidence is one firm's executives qualitatively reporting a productivity decline (confidential, self-reported, uncontrolled). In systems terms it measures a flow (interaction) and infers a stock (collaborative capability/output) without measuring the stock — and never tests whether some of the electronic substitution (async, focused, documented) improves rather than harms certain outcomes.
Read next: Read studies that link open-office redesign to measured output/decision quality/performance, not just interaction; the media-richness empirical literature on when leaner channels help or hurt; and deep-work/async-collaboration research on whether shifting interaction off the floor can raise productivity for some task types.
incompleteThe mechanism — a balancing feedback loop of privacy restoration under observation — is hypothesized but not instrumented. No measure of headphone use, avoidance, or perceived observability was collected, so the mediating variable is inferred from volume shifts. Competing explanations (noise avoidance, changed physical adjacency of one's actual collaborators, task changes, or a Hawthorne/observer effect from being badged) are not ruled out.
Read next: Read Bernstein's own transparency-paradox work (2012) and privacy/boundary-management theory (Altman) for the mechanism; look for studies that directly measure the mediator (privacy-seeking behavior, perceived observability) rather than inferring it, and that separate 'loss of enclosure' from 'change of neighbors'.
incompleteGeneralizability is thin: n=2 organizations, both US Fortune-500 headquarters, specific open designs (undivided tables), short observation windows (3 and 8 weeks) beginning only ~2 months after the move, so a transient adjustment/novelty phase cannot be distinguished from long-run equilibrium. The framing universalizes ('a natural human response') without testing national/cultural context, org type, which open designs or mitigations (quiet zones, activity-based working, norms) change the result, or steady-state behavior a year on.
Read next: For a leader housing teams (esp. in a Norwegian flat, high-trust, quiet-office context), read cross-cultural office-design studies, longitudinal (>1 year) open-office evaluations, and activity-based-working assessments to see whether the effect holds where privacy norms, trust levels and design specifics differ.
incompleteThe counterfactual is narrow: the studies compare open space only against cubicles. They cannot tell a decision-maker what to do instead, only that a cubicles-to-fully-open move backfired — private offices, hybrid/activity-based layouts, and remote/distributed work are untested comparators. Nor can they separate 'losing enclosure' from 'changing who sits nearby'. The implicit assumption that architecture is the right leverage point (rather than norms, incentives, or meeting design) is left unexamined.
Read next: Read comparative office-typology studies (private vs cubicle vs open vs ABW vs remote) and, in Meadows' leverage-point terms, work on whether interaction is more effectively shaped by rules/norms/incentives than by the physical structure the paper treats as the intervention.

The impact of the 'open' workspace on human collaboration

Structure

Key terms

Propositions

Arguments

Narratives

Questions addressed

Gaps

Related documents