I had lunch with a man from my sailing club in March and it went well. Not politely well. We were the last two in the place, someone was stacking chairs around us, and at the door we both said we should do that again properly.

We did not. I saw him at the club a dozen times afterwards, it was friendly every time, and neither of us ever mentioned lunch again. In November he took a job in Adelaide and that was the end of it.

Nothing went wrong. That is the part I had trouble with for a while. The catch-up worked and produced nothing, and my first assumption was that I had misread the whole thing.

Men see friends in person about as often as women do

The shortfall in male friendship is not where most people put it.

In Pew Research Center's survey of 6,204 American adults, fielded in September 2024, men reported seeing their close friends in person slightly more often than women did: 31 per cent at least weekly against 28 per cent. Where men fell behind was everything in between. Thirty-two per cent spoke to a close friend by phone or video at least weekly, against 38 per cent of women, and the gaps on texting and social media ran the same way.

Men have the catch-up. What is missing is whatever turns one catch-up into a series, and that is a narrower problem than "I have no friends".

The gap widens at exactly the wrong moment

After a conversation, the other person likes you more than you think they do.

Boothby, Cooney, Sandstrom and Clark called this the liking gap and measured it repeatedly in a 2018 paper in Psychological Science. Their samples were undergraduate strangers paired in a lab, first-year students meeting dorm-mates, and members of the public at a workshop - young, mostly, and none of them a man in his forties trying to convert a lunch. In their first study, strangers talked in pairs for five minutes and then rated each other on a seven-point scale. Actual liking averaged 5.82. What each person believed the other thought of them averaged 5.17, a reliable shortfall (t(34) = -5.83, p < .001).

The finding I keep coming back to is their fourth study, which ran inside a real workshop with a hundred participants and took the measurement twice, before the conversation and after it. Beforehand, actual liking was 3.59 and believed liking 3.22, a gap of 0.37. Afterwards, actual liking had risen to 4.30 and believed liking to 3.61, a gap of 0.68.

Both numbers went up. The distance between them almost doubled.

So the point at which you are least able to read how a conversation went is the point immediately after it, which is also the point at which you decide whether to arrange the next one. That is a badly designed sequence, and I had been running it for years without noticing.

A longer conversation does not close it

The obvious defence is that a really good long lunch must be different from a five-minute lab conversation.

It is not. The same paper's third study covered conversations from two minutes to forty-five minutes and found the gap at every length, with no reliable relationship between duration and size (interaction p = .37). "We really hit it off, it will sort itself out" has nothing standing behind it. The study did not test a two-hour lunch. Nothing in its data suggests length helps, but past forty-five minutes I am extrapolating and should say so.

There is a detail in the first study that made me wince. Among people high in shyness the gap was 1.14 points. Among the least shy it was 0.25 and did not reach significance. The man most likely to need a second arrangement, because he has fewest of them, is also the man least able to tell that it would be welcome.

You are most wrong about the men you know least

The error is not evenly spread across your contacts.

Liu, Rim, Min and Min ran thirteen preregistered studies published in the Journal of Personality and Social Psychology in 2023 on how much people appreciate being contacted out of the blue. Across 1,602 participants, senders consistently underestimated it, and the underestimate was substantially stronger for weak ties than for close ones (partial eta squared .046 against .010). When they had people send real messages, senders predicted an appreciation rating of 5.50 and recipients reported 5.87.

Weak ties are precisely the category the sailing club man was in. So is nearly everyone a man in his forties might plausibly convert into a friend: the bloke from the old job, the other father at the fixture, the neighbour two doors down. The people you are most wrong about are the people the whole thing depends on.

A related study by Sandstrom, Boothby and Cooney puts a number on how wrong. In their 2022 paper in the Journal of Experimental Social Psychology, 286 people were asked to talk to strangers over the course of a week. On the first day, only 40 per cent expected the first person they approached to be willing to talk to them. Ninety-two per cent were. Across the week, 1,164 of the 1,336 conversations that happened were with the first person approached.

What happens when you just tell people the research?

Almost nothing, which is the honest and slightly deflating answer.

Both groups tested it. Liu's team ran a supplemental study in which they gave participants the finding directly and then measured whether they actually got in touch with anyone. Interest ticked up without reaching significance, and there was no significant increase in the behaviour itself. Sandstrom's team found that a week of practice with strangers moved expected rejections from 0.89 to 0.23 and kept them there at follow-up, while the number of conversations people actually had did not differ from the control group.

Attitudes move. Diaries do not.

I can confirm this from the inside. I have known about the liking gap for over a year and I still walked out of a good lunch, felt the usual small uncertainty about whether he had enjoyed it as much as I had, and did nothing about it for eight months. Understanding the bias does not disarm it. It just gives you a name for what you are doing while you do it.

Which means the fix has to be structural. Something that produces the second meeting without either man having to make a judgement, in the ten seconds after a good conversation, about whether the other one wants it.

Set the next one before you stand up

The window closes when you get up from the table, so the arrangement has to be made while you are still sitting at it.

That is the whole mechanic. Before the goodbye, while the conversation is still running, put a specific next date in the air and get the phone out:

Same time in a month? I'll put it in now before we both forget.

Three things are doing work there. The date makes it a decision instead of a project. Getting the phone out immediately converts a pleasantry into a diary entry, which is the actual difference between the two outcomes. And "before we both forget" is an honest and low-stakes reason that asks nobody to say anything about how much they enjoyed the afternoon.

Timing is the whole reason it works. Do it while the estimate is still cheap to ignore, because the moment you are on your feet, the thing you consult is your own reading of how it went, and the research above says that reading is wrong in a predictable direction.

The standing slot

Better again: make it something that recurs, so nobody has to ask twice.

First Thursday of the month. A fortnightly walk. The same pub the week after payday. Once an arrangement has its own schedule, the decision has been taken permanently and the awkward ask never has to be made again. I have written up the mechanics of that first ask, including the wording and what to leave out of it, because that part stops most men before they get anywhere near a second meeting.

Two things I did not expect from a standing slot. A cancellation stops meaning anything, because there is another one in a fortnight, so neither man has to interpret it. And the arrangement survives long stretches where neither of you would have initiated, which turns out to be most of the year.

A recurring diary entry also has no feelings about you. It does not have to be talked into existence each time by whichever man currently feels most confident about the friendship, and waiting for that man is what kills these things.

This week

Take the last catch-up you had that went well and produced nothing. There will be one, and it is more likely to be recent than you think.

Send a specific date. Not "we should do that again", which is the sentence that killed the last attempt, but an actual day and place with the phone in your hand. The evidence says the message will land better than you expect it to, and that it will land better still if you barely know the man.

Then, when you are there, book the next one before you stand up.

If the wider question of how a full life ends up with an empty middle tier is the one you are actually sitting with, I have written about that here.