The Mandate Nobody Diagnosed

A return to office mandate is an intervention. It has a dose, measured in days. It has a target population, which is usually everybody. It has a cost, which is other people's time and money. And like any intervention it is supposed to be aimed at something.

Read a few of the announcements and see how many name what they are aimed at in terms precise enough to fail. Collaboration. Culture. Innovation. Serendipity, which is a word that means the good thing that happens by accident, deployed here as the goal of a policy. Almost none of them state a measurable condition, a baseline for it, or what result would count as the mandate not having worked.

That is the actual problem, and it is upstream of whether the office is good or bad. An intervention with no stated endpoint cannot be evaluated, and an intervention that cannot be evaluated will be judged on whether people showed up.

What happened to the firms that did it

Ding and Ma searched public announcements from S&P 500 companies and identified 137 firms that had publicly required employees back for at least several days a week. They then compared what happened afterwards against firms that had not.

On the financial side, no significant improvement in profitability. No significant improvement in market valuation. On the employee side, using Glassdoor ratings, declines in overall satisfaction, in work-life balance, in views of senior management and in perceptions of corporate culture.

A companion study looked at what happened to the people. Firms that announced mandates saw abnormal turnover rise by 13 to 14 percent on average, and the increase was more pronounced among senior employees, more skilled employees and women. Those firms then took significantly longer to fill the resulting vacancies and saw their hire rates fall. A larger share of departing employees moved to lower-ranked positions at their next employer, which is what it looks like when people pay for flexibility out of their own career progression.

The attrition was not random. It was concentrated among the people with the most options, which is the definition of the group you least wanted to test this on.

Two caveats, stated plainly because this series is not in the business of stacking the deck. Both of those are working papers rather than peer-reviewed journal articles, and both rely on difference-in-differences comparisons using publicly posted reviews rather than on random assignment. They are strong observational evidence. They are not proof, and they should not be quoted as though they were.

The obvious objection is selection. Firms that announce mandates may already be struggling, and a struggling firm would show flat profits and unhappy employees regardless. Ding and Ma looked directly at what predicted a mandate and reported results consistent with managers using them to reassert control over employees and to blame employees for poor firm performance. That is a claim about motive drawn from observational data, which is the part of their work a careful reader should hold most loosely.

But notice where the objection leaves the argument even if you grant it entirely. If mandates follow poor performance rather than causing it, then the mandate is downstream of the problem rather than a fix for it, and the organization has responded to a performance shortfall by changing where people sit. That is not a rescue. It is a different version of the same finding.

The case for the office, at full strength

Now the other side, and it deserves better than it usually gets from people quoting the studies above.

Atkin, Schoar and Shinde randomly assigned data entry workers to home or office, which removes the selection problem entirely, and the home group was 18 percent less productive. Emanuel, Harrington and Pallais found that sitting near teammates raised the feedback junior engineers received on their code by 18.3 percent. Emanuel and Harrington found that remote work degraded call quality particularly for inexperienced workers and reduced promotion rates.

Those are real effects, two of them from designs stronger than anything in the studies of mandates, and anyone who waves them away is arguing from preference rather than evidence. The office does things. It does them for novices more than for experts, for neighbors more than for the organization, and it does them without anyone having to design them, which is genuinely valuable and genuinely hard to replace.

So the honest position is not that mandates are stupid. It is that a mandate is a very blunt instrument aimed at a real target, fired at everybody, with the result measured by attendance.

The dose exists, and somebody actually measured it

Here is the finding that should reframe the whole conversation, and it comes from the one design capable of answering the question. Choudhury, Khanna, Makridis and Schirmann randomized how many days individual employees worked from the office. Not whether. How many.

Workers assigned around two days a week in the office reported greater work-life balance, more job satisfaction and less isolation from colleagues than workers assigned either more or fewer days. Their performance ratings were no different from anyone else's.

Bloom, Han and Liang's randomized trial of 1,612 people arrives from the other direction and lands in the same place. Two days a week at home, performance grades unchanged over the following two years of reviews, quit rates down by a third.

Two experiments, different countries, different industries, both pointing at an interior optimum. Almost every mandate in the wild is a corner solution.

When the best two experiments both find the answer in the middle, a policy of five days or zero is not a reading of the evidence. It is a preference with a memo attached.

This argument is older than you think

It is tempting to treat all of this as a pandemic artifact that will resolve once everyone calms down. Two numbers say otherwise.

Gajendran and Harrison pooled 46 studies covering nearly 13,000 employees and found telecommuting had small but mainly beneficial effects on outcomes including perceived autonomy and lower work-family conflict. That meta-analysis was published in 2007.

And Barrero, Bloom and Davis report that full days worked at home accounted for 28 percent of paid workdays among Americans aged 20 to 64 as of mid-2023, roughly four times the 2019 rate and about ten times the rate in the mid-1990s. The level has been broadly stable rather than reverting.

A quarter of the working week is not a phase to wait out. It is an operating model that arrived without being designed, which is a sentence that should sound familiar to anybody who has read the rest of this series.

What a designed version would look like

None of this argues for doing nothing. It argues for doing the thing an intervention requires before it is deployed.

• State the constraint before the policy. Which condition is failing, and how do you know? Attention, alignment, authority and adaptability degrade differently and are visible in different places. A policy that cannot name its target cannot be evaluated by anything except compliance.
• Pick the dose, not the corner. The experiments point to the middle. Two days is not a compromise between two camps, it is where the measured optimum keeps landing, and it is available to anyone willing to read the studies rather than the comment section.
• Target the population the research targets. The benefits of proximity concentrate on people who are learning and on people who need to reach across a boundary. A uniform mandate spends the same budget on an expert executing codified work, where the evidence says the return is close to zero.
• Measure the condition rather than the badge. Badge swipes measure presence the way a smoke detector measures cooking. Count uninterrupted hours, cross-team ties, decisions resolved without escalation and feedback reaching junior people. Those are the things the research actually measured.
• Say in advance what would count as it not working. If nothing could, it was never an intervention. It was a position.

The uncomfortable symmetry

The remote camp should take the same medicine. A blanket commitment to fully distributed work, adopted because it is popular internally and cheap on real estate, is also an intervention with no stated endpoint. It is aimed at conditions nobody measured and it has its own concentrated cost, which the previous piece in this series documents falling on the least senior people in the building that no longer exists.

Both mandates are the same error wearing different politics. Somebody picked a location policy, applied it to everyone, and called the resulting compliance a result.