Both Sides of the Remote Work Argument Are Right

Here is the state of the evidence, which almost nobody arguing about this has read.

Atkin, Schoar and Shinde ran a randomized controlled trial in the data entry sector in India, assigning workers to home or office rather than letting them choose. The people sent home were 18 percent less productive. Two thirds of that gap showed up on the first day.

Bloom, Han and Liang ran a randomized controlled trial at Trip.com across 1,612 engineers, marketers and finance staff, assigning two days a week at home. Performance grades did not move across the next two years of reviews. Quit rates fell by a third.

Same method. Same question. Opposite answers. If you have a firm opinion about remote work, one of those studies is your favorite and you have probably never mentioned the other one out loud.

Two rigorous trials disagreeing is not a reason to ignore the evidence. It is the evidence telling you that you are holding the wrong variable.

The scoreboard, honestly kept

For the office camp: Atkin and colleagues found the 18 percent gap under random assignment. Yang and colleagues, studying a very large firm, found that going fully remote made collaboration networks more static and more siloed, and pushed communication toward asynchronous channels. Emanuel, Harrington and Pallais found that sitting near teammates raised the coding feedback junior engineers received by 18.3 percent.

For the remote camp: the Trip.com trial above. Bloom and colleagues also found a 13 percent performance gain among call center staff at Ctrip in an earlier experiment. Choudhury, Foroughi and Larson found that letting patent examiners at the USPTO work from anywhere raised output.

Both lists are real. Both sets of authors are careful. And the moment you put them side by side, the pattern stops being about geography.

The most cited anti-remote number is a fraction, and fractions have a bottom

The study people quote most is Gibbs, Mengel and Siemroth, who tracked more than ten thousand professionals at a large IT services firm. The headline is that productivity fell somewhere between 8 and 19 percent. It gets posted roughly every week by someone announcing that the experiment is over.

Read the components. Total hours worked rose by about 30 percent. Average output did not significantly change. Productivity is output divided by hours, so a flat numerator over a larger denominator produces a smaller number, and the sentence "productivity fell" is arithmetically correct.

It is also read, almost universally, as "less got done." Less did not get done. The same got done, and it consumed considerably more of people's lives. Whether that is a catastrophe or a rounding error depends entirely on whose ledger you are holding. From the firm's side, output held. From the person's side, the commute came back as work and then brought friends.

That is a real problem and it is worth taking seriously. It is a sustainability problem, not an output problem, and the two have completely different fixes. One is answered by a building. The other is not.

What actually moved, in every study that found a loss

Go back through the negative findings and read the mechanism rather than the headline. Gibbs and colleagues reported that time in coordination and meetings rose while uninterrupted work hours shrank considerably, and that people networked with fewer individuals and business units. Yang and colleagues reported networks becoming static and siloed. Emanuel and colleagues reported junior engineers receiving less feedback from senior ones.

Not one of those is a sentence about location. Every one is a sentence about a condition.

Here is the uncomfortable part. In most offices, none of those four conditions was built. They were being held up by proximity accidents: the hallway correction, the overheard context, the senior engineer glancing at a screen on the way past. Nobody designed any of it. Nobody could have named it as a system. It simply happened, for free, all day, because people were in the same room.

Remote work did not break those conditions. It removed the thing that was accidentally holding them up, and then everybody looked at the wreckage and blamed the laptop.

Your office may have been load bearing. That is not a compliment to the office.

The finding that should humble everybody

Buried in the Atkin study is a result that deserves more attention than the headline. The researchers measured what workers preferred, then randomly assigned them anyway.

Workers who preferred working from home were 12 percent faster and more accurate at baseline. Genuinely stronger performers. And they were substantially less productive at home than at the office: 27 percent less, against 13 percent for the people who preferred the office.

The people most confident that they work better from home were the people whose work suffered most from it. If that stings, it should sting evenly. The executive who is certain the office is the answer, and the engineer who is certain it is not, are both reasoning from a sample of one, and the one study that checked found self-knowledge pointing the wrong way.

So should your team come back?

If the four conditions are not deliberately built and maintained in your organization, then yes. Come back. The office will help, and the research showing it helps is real.

But be clear about what that decision is admitting. It says the team functions when people can see each other and degrades when they cannot, which means the thing holding it together is adjacency rather than anything anyone built. The office is a reasonable answer to that problem. It is not a solution to it, and it is an expensive place to keep one.

The inverse is the part leaders tend to skip. If a team holds its shared picture, protects its deep work, decides at the level where the knowledge sits and turns experience into changed behavior, then that team is going to be effective distributed, hybrid, or crammed into an open plan floor with a ping pong table nobody uses. Location stops being the interesting variable because the conditions are carrying the load.

Which makes this a leadership question wearing a real estate costume. A return to office mandate is a condition intervention that nobody designed as one, aimed at a constraint nobody diagnosed, and measured by badge swipes.

What to do before you pick a side

The useful move is not to decide about location. It is to find out which condition is actually binding, because the answer changes what you should do and location is downstream of it.

• Count the uninterrupted hours a maker on your team got last week. Not scheduled. Actual.
• Ask two people who depend on each other to describe the current priority separately, then compare the answers.
• Take one decision that went up the chain last month and name what the approver knew that the requester did not.
• Name one thing the team learned this quarter that changed a rule, a routine or a decision. If nothing comes, that is the finding.

Whichever of those four is worst is the thing costing you performance. It was probably costing you in 2019 too. The building was just covering for it.