Distance is a durable penalty that hybrid work has not repealed

The fourth of ten units turns from the signal to the distance it crosses. Cooperation over distance costs more than the channel that carries it. Olson and Olson (2000) reviewed over ten years of field and laboratory studies of collocated and noncollocated group work and concluded remote work succeeds only where common ground is high, the work is loosely coupled and the group is ready for both. None of those four conditions is bandwidth. Distance remains a durable penalty, because it works through what people share and how their tasks depend on one another. The productivity gains since 2020 are bounded to work the distance literature never claimed to describe. The fifth unit turns to the communities that sustain such work.

Bandwidth is not what distance removes, and mutual knowledge is what fails

Kiesler and Cummings (2002) reviewed the research on proximity and distance and located the strongest effect in informal, spontaneous encounter, the conversations by the coffee machine and in the lunch room. Those encounters build the shared context colleagues draw on later, and a scheduled call supplies none of it.

Nardi and Whittaker (2002) asked what a face-to-face meeting supplies that a channel does not, and their answer was relational. From studies of workplace networks they argued people must meet in person to establish the relationships business dealings later rest on. The workers they interviewed described the mechanism: shared bodily activity, eating and drinking together, and simply showing up. A medium can therefore be rich on the dimensions a system measures and poor on the one that decides whether the pair help each other next month.

Cramton (2001) supplies the mechanism. From thirteen geographically dispersed teams she built a model of the mutual knowledge problem: members assume a shared context that is in fact local, and the gaps between them are systematic rather than random. She named five recurring failures, including lost context, unevenly distributed information and the misreading of silence. Dispersed partners therefore misattribute the failures they meet, reading a colleague's quiet as agreement when it is a channel artefact, and the damage compounds because neither side can see the misreading.

The cost of distance is measurable, and it tracks how many people are involved

Herbsleb and Mockus (2003) used change-management records and survey data from a distributed software organisation and found distributed work items took about two and a half times as long as comparable collocated items. Distributed items involved more people, and that number tracked the calendar time to completion. Herbsleb, Mockus, Finholt and Grinter (2000) found cross-site delay rose where a site's colleagues were perceived not to help under heavy load, while workers believed they helped remote colleagues as much as local ones. Espinosa, Slaughter, Kraut and Herbsleb (2007) added the moderator: distance harmed coordination, but shared team knowledge and presence awareness mitigated it. Bjorn and Ngwenyama (2009) show the repair is deliberate, since teams rebuilt shared meaning by working through their breakdowns.

The strongest case for remote work rests on countable, loosely coupled tasks

Bloom, Liang, Roberts and Ying (2015) randomised volunteers at Ctrip, a Chinese travel agency of about 16,000 employees, to home or office work for nine months. Home working raised performance by 13 per cent and halved attrition, about nine points from more minutes per shift and four from more calls per minute. The task was answering calls, which is individual, loosely coupled work. Choudhury, Foroughi and Larson (2021) found a work-from-anywhere transition for patent examiners raised output 4.4 per cent without changing rework incidence. That is again individual, countable work, close to the conditions Olson and Olson (2000) predicted would survive distance. The trials assume a reliable connection, so the cost a metred connection adds is not in them.

The post-2020 evidence restores the penalty when the unit is collaboration

Yang and colleagues (2022) used telemetry from 61,182 employees of one firm over the first six months of 2020 and found firm-wide remote work made collaboration networks more static and siloed, with fewer bridges, and moved communication from synchronous to asynchronous. The network result is no longer one firm's. Zuzul and colleagues (2025) analysed metadata from more than 360 billion emails across 4,361 organisations and found siloing rose in 2020 with less stable group membership. Gibbs, Mengel and Siemroth (2023) examined data on more than 10,000 skilled professionals at an Indian technology company: hours rose, output fell slightly and productivity fell by 8 to 19 per cent. Emanuel and Harrington (2023) separated who chooses remote work from what it does to them in a Fortune 500 firm's call centres. Remote workers answered 12 per cent fewer calls per hour; when the centres closed, previously on-site staff fell 4 per cent relative to those already remote, a third of the initial gap, and an 8 per cent gap persisted. Most of the difference was negative selection, with a negative treatment effect inside the rest.

Why hybrid is contested nonetheless

Hybrid is the settlement the evidence seems to be reaching, and Barrero, Bloom and Davis (2021) predicted it early. They surveyed more than 30,000 Americans and estimated that 20 per cent of full workdays would be supplied from home after the pandemic, against 5 per cent before. That forecast rests on improved experience and new investment rather than measured output. Bloom, Han and Liang (2024) ran a six-month controlled trial of hybrid working for 1,612 employees of a Chinese technology company. Resignation rates fell by a third, concentrated among non-managers, women and long commuters, while performance was unchanged on null equivalence tests. That is a strong result for retention and a null one for output, and it says nothing about the collaboration network.

Contested question: does measured productivity answer the distance claim?

Three positions divide the distance literature. Olson and Olson (2000), Cramton (2001) and the software-engineering studies (Herbsleb et al., 2000; Herbsleb and Mockus, 2003; Espinosa et al., 2007) hold that distance imposes a durable coordination penalty better media do not remove, because it lies in common ground and coupling rather than the channel. The productivity studies (Bloom et al., 2015; Choudhury et al., 2021; Bloom, Han and Liang, 2024) hold that the penalty is conditional, and under loose coupling and high common ground can be small, absent or reversed. Emanuel and Harrington (2023) and Gibbs and colleagues (2023) hold a third position cutting across both: the headline productivity numbers are confounded, because selection decides who works remotely and output is not coordination.

What would settle the disagreement is a design measuring both at once. Hold the work and the people constant, and vary whether they are near or far. Record coordination outcomes such as cross-site delay and the bridges in the collaboration network alongside individual output and selection. Emanuel and Harrington's (2023) natural experiment and Bloom, Han and Liang's (2024) trial are the nearest templates, but neither measures coordination. Until one study carries both halves, the two bodies of evidence will keep appearing to disagree.

Set the two literatures side by side. The penalty rests on a decade of field and laboratory studies (Olson and Olson, 2000; Kiesler and Cummings, 2002), a model from thirteen dispersed teams (Cramton, 2001) and change-record studies of software teams (Herbsleb and Mockus, 2003). The productivity case rests on two randomised trials and a natural experiment, each on countable output (Bloom et al., 2015; Choudhury et al., 2021; Bloom, Han and Liang, 2024). While the dispute is open, a designer should name the outcome the arrangement must serve and measure it. The strongest objection remains that the coordination measures are correlational, since nothing assigns distance.

Where the argument is heading

The direction is toward designs that carry both halves. The classical literature (Olson and Olson, 2000; Cramton, 2001) established the coordination penalty. The post-2020 literature (Yang et al., 2022; Zuzul et al., 2025; Gibbs et al., 2023; Emanuel and Harrington, 2023; Bloom, Han and Liang, 2024) has begun to supply samples and randomisation. The open question is whether the network and delay measures of the first can be built into the randomised designs of the second.

The sources this unit rests on

Each note states what the document did and found. Identifiers were resolved on 2 November 2026.

• Barrero, J. M., Bloom, N. and Davis, S. J. (2021). Why working from home will stick. doi:10.3386/w28731 Surveyed more than 30,000 Americans and projected that 20 per cent of full workdays would be supplied from home after the pandemic, against 5 per cent before. It frames the hybrid question, but is a forecast, not measured output.
• Bjorn, P. and Ngwenyama, O. K. (2009). Virtual team collaboration: building shared meaning, resolving breakdowns and creating translucence. doi:10.1111/j.1365-2575.2007.00281.x Studied communication breakdowns in two globally distributed virtual teams and traced them to differences in lifeworld, organisational and work-process structures. It shows the distance penalty is met with deliberate repair, not removed by the medium.
• Bloom, N., Han, R. and Liang, J. (2024). Hybrid working from home improves retention without damaging performance. doi:10.1038/s41586-024-07500-2 Ran a six-month randomised trial of hybrid working for 1,612 employees of a Chinese technology company; quit rates fell by a third, the gain concentrating among non-managers, women and long commuters, and performance was unchanged. It is the strongest randomised evidence that hybrid is net-positive, and it measures output rather than coordination.
• Bloom, N., Liang, J., Roberts, J. and Ying, Z. J. (2015). Does working from home work? Evidence from a Chinese experiment. doi:10.1093/qje/qju032 Randomised call-centre employees at Ctrip to home or office for nine months and found home working raised performance 13 per cent, about nine points from minutes worked and four from calls per minute, while halving attrition. It is the experimental case for remote productivity, on individual, loosely coupled work.
• Choudhury, P., Foroughi, C. and Larson, B. Z. (2021). Work-from-anywhere: the productivity effects of geographic flexibility. doi:10.1002/smj.3251 Exploited the staggered transition of United States patent examiners to a work-from-anywhere programme and found a 4.4 per cent increase in output with no change in rework incidence, extending the productivity gain to geographic flexibility on countable individual output.
• Cramton, C. D. (2001). The mutual knowledge problem and its consequences for dispersed collaboration. doi:10.1287/orsc.12.3.346.10098 Built a model of the mutual knowledge problem from thirteen geographically dispersed teams, naming five recurring failures such as the uneven distribution of information and difficulty reading salience. Being systematic and misattributed, they damage coordination invisibly.
• Emanuel, N. and Harrington, E. (2023). Working remotely? Selection, treatment and the market for remote work. https://www.newyorkfed.org/research/staff_reports/sr1061 Separated selection from treatment in a Fortune 500 firm's call centres: remote workers answered 12 per cent fewer calls per hour, a 4 per cent treatment effect accounted for a third of the gap when the centres closed, and an 8 per cent gap persisted as negative selection, with lower call quality and promotion rates. Raw productivity comparisons therefore confound who chooses the arrangement with what it does.
• Espinosa, J. A., Slaughter, S. A., Kraut, R. E. and Herbsleb, J. D. (2007). Team knowledge and coordination in geographically distributed software development. doi:10.2753/mis0742-1222240104 Field study of software teams identifying technical, temporal and process coordination needs; distance harmed coordination but shared team knowledge and presence awareness mitigated it, and shared task knowledge mattered more when members were collocated. It specifies what buys the penalty down and what cannot.
• Gibbs, M., Mengel, F. and Siemroth, C. (2023). Work from home and productivity: evidence from personnel and analytics data on information technology professionals. doi:10.1086/721803 Used data on more than 10,000 skilled professionals at an Indian technology company; hours rose, output fell slightly and productivity fell 8 to 19 per cent, with coordination time up while uninterrupted work shrank. It is the clearest recent evidence that more hours do not mean more output.
• Herbsleb, J. D. and Mockus, A. (2003). An empirical study of speed and communication in globally distributed software development. doi:10.1109/tse.2003.1205177 Modelled change-management and survey data and found distributed work items took about two and a half times as long as collocated ones, replicated in a second organisation. The delay tracked the number of people involved, locating the cost in coordination rather than the medium.
• Herbsleb, J. D., Mockus, A., Finholt, T. A. and Grinter, R. E. (2000). Distance, dependencies and delay in a global collaboration. doi:10.1145/358916.359003 Measured delay and site interdependence in a multi-site organisation and found cross-site delay rose with the degree to which remote colleagues were perceived not to help under heavy load, while workers believed they helped them as much as local ones. A mutual misperception sits inside the delay.
• Kiesler, S. B. and Cummings, J. N. (2002). What do we know about proximity and distance in work groups? doi:10.7551/mitpress/2464.003.0007 Reviewed the proximity and distance literature and located the strongest effect of distance in informal, spontaneous encounter and in weakened identification with the larger group. It is why richer media do not close the gap.
• Olson, G. M. and Olson, J. S. (2000). Distance matters. doi:10.1207/s15327051hci1523_4 Reviewed over ten years of field and laboratory studies of collocated and remote group work and reduced success to four conditions: common ground, coupling, collaboration readiness and technology readiness. It is the anchor claim this unit tests.
• Yang, L., Holtz, D., Jaffe, S., Suri, S., Sinha, S., Weston, J., Joyce, C., Shah, N. P., Sherman, K., Hecht, B. J. and Teevan, J. B. (2022). The effects of remote work on collaboration among information workers. doi:10.1038/s41562-021-01196-4 Used telemetry from 61,182 employees of one firm over the first six months of 2020 and found firm-wide remote work made collaboration networks more static and siloed, with fewer bridges, and shifted communication from synchronous to asynchronous. It is the largest coordination-side evidence since 2020, and it measures the network rather than output.
• Nardi, B. A. and Whittaker, S. (2002). The place of face-to-face communication in distributed work. doi:10.7551/mitpress/2464.003.0008 Argues from studies of workplace networks that face-to-face meetings establish and sustain the relationships distributed work depends on, through shared physical activity rather than information transfer. It explains why a richer channel does not remove distance.
• Zuzul, T., Cox Pahnke, E., Larson, J. and colleagues (2025). Dynamic silos: increased modularity and decreased stability in intraorganizational communication networks during the COVID-19 pandemic. doi:10.1287/mnsc.2022.02797 Analysed more than 360 billion emails across 4,361 organisations and found siloing rose in 2020 with less stable group membership, extending the collaboration-network evidence beyond one firm.