“What do you think of this?”
That is the last message in one version of an experiment funded by the University of California, Berkeley. It goes to a friend you don’t know all that well, a couple of texts into an exchange over the last few hours. Then nothing comes back.
If you have been on that end of it, you probably built a theory about the person on the other side. Most theories land somewhere between forgot and couldn’t be bothered, and neither one has much feeling in it. A paper published on 31 August in Personality and Social Psychology Bulletin argues that the theory usually runs too cold. Sophia Li, Coral Zheng and Juliana Schroeder ran three experiments on both sides of that silence, and in two of them the people on the silent side, recalling a real lapse in one study and imagining one in the other, put their own bad feeling higher than the waiting side predicted.
195 people, and their own message threads
The first study drew on 195 adults recruited on Prolific, each asked to go back through their messages and find the last real instance. It is a between-subjects design: 101 people were asked to recall a time they had stopped replying to someone, and 94 different people a time someone had stopped replying to them. Participants were told to check their phones rather than reach for whatever came to mind first, and to pick someone who was neither a stranger nor a close friend or family member. The median gap since the waiting party’s last message was ten days.
What people actually chose was looser than the instruction. Asked to describe the relationship, 46.2 per cent called the other person a friend, 34.4 per cent an acquaintance, and 8.7 per cent immediate family or a very close friend, a category the instruction had ruled out. Another 4.6 per cent picked a stranger, which it had ruled out too.
Asked how badly they felt afterwards, on a scale running from 1 (“not bad at all”) to 5 (“extremely bad”), the people who had gone quiet averaged 2.27. The people who had been left waiting put the other party, in their own recalled case, at 1.56. That difference sits at d = 0.77.
Two more things sat inside it. The waiting group specifically underestimated guilt (2.35 reported against 1.85 predicted) and sadness (1.84 against 1.46). They also slightly underestimated relief, 1.89 reported against 1.56 predicted, which keeps this from being a tidy story about remorse. Of the eight emotions measured, those are the three the waiting side misread.
The people who had gone quiet read the other side almost exactly right: 2.11 predicted against 2.11 reported. The one thing they got wrong was anger, which they slightly overestimated, at 1.87 predicted against a reported 1.52. That is the fourth misread emotion in the study, and the only one running the other way.
The accidental silences
The gap sits in the average, but it is not spread evenly across the silences it attaches to.
People who said their non-response had been closer to accidental than deliberate felt worse about it, not better; the correlation between intent and bad feeling ran negative, at r = −.29. The people waiting did not track it. Their predictions were flat against their own reading of how deliberate the silence had been, r = −.002, and the interaction between perspective and intentionality was statistically significant.
Split the sample at the midpoint of that intentionality scale, which ran from “completely accidental” to “completely on purpose”, and the size of the misreading changes character. Among the eighty-four participants whose situation fell below the midpoint, the ones who had gone quiet averaged 2.62 while the waiting group predicted 1.50, at d = 1.17. Among everyone else, at or above the midpoint, the gap was still there, at d = 0.52. The paper is explicit that the misreading is not confined to the accidental cases.
One wrinkle is worth naming, because it went against the researchers’ own written prediction. They had registered in advance the expectation that people who were left waiting would overestimate how far the silence was a response to something they themselves had done. In these recalled cases, on the paper’s own reading, the direction reversed: the people who had gone quiet described more intent behind it (3.75) than they were credited with (3.19).
“Sorry, but I can’t talk right now”
The second study put 601 people into a written scenario instead of their own memories, and added the comparison that makes the finding practically useful. Each participant was assigned one side of the exchange and then saw two versions of it: one in which the friend simply did not reply, and one in which a single line came back, “Sorry, but I can’t talk right now”, and nothing more.
The perspective effect reappeared, accounting for about fifteen per cent of the variance on the paper’s own eta-squared. And the one-line reply did shrink the misreading. Picturing no reply at all, people put their own bad feeling at 3.23 while the other side predicted 1.94. Picturing the brush-off, they put it at 2.33 against a prediction of 1.82. Smaller gap, still a gap, still statistically significant at d = 0.49. The short reply narrows the distance between what someone feels and what you think they feel. It does not close it.
It moves something else too. When nobody replied in the scenario, the waiting participants were asked to pick one of three reasons the researchers had supplied, and they split almost evenly: no bandwidth, the conversation wasn’t interesting, I was annoying them. When the one-line reply arrived, 57.7 per cent landed on “they didn’t have the bandwidth” and 19.7 per cent on “I was annoying them”. Seven words of acknowledgement moved the waiting participant’s explanation away from themselves.
The live study’s headline test
For the third study the researchers built the situation rather than describing it. The survey was opened to six hundred people. Fifty never matched or hit technical trouble, and another ninety-two were lost when they or their partner timed out mid-study, leaving 458 people in 229 complete pairs, each of which had had an eight-minute get-to-know-you conversation using the kind of questions designed to build closeness quickly. Then one partner was offered a choice: sit through a second five-minute conversation, in which they would be interviewed about their own personality and asked to write a short summary of it, or skip it and go straight to the end of the study for the same payment. The researchers say they positioned skipping as the more appealing option, since only pairs where someone left would enter the main analysis. In 175 pairs someone took it, and those pairs, 350 people, are the sample for the primary analysis.
One thing to hold before the numbers: the scale flipped between studies. The first two ran from “not bad at all” to “extremely bad”, the third from “extremely good” to “extremely bad”, so nothing below can be lined up against the numbers above, and three further measures in this study sit on scales of their own.
The headline comparison came back flat. People left sitting in the chat window predicted their partner felt 3.35 bad about it; the partners reported 3.26. Paired, the difference did not reach significance, at p = .357, and the interval around it still allows a small gap in either direction.
The gap returns twice elsewhere in the same study. One is a preregistered secondary analysis of the 54 pairs in which the potential leaver stayed and did the personality interview. Asked how they would have felt had they walked, they put it at 4.22 against their partner’s prediction of 3.24, at d = 0.93. The other is a subgroup of the pairs where someone did leave, split on a moderator the authors say plainly they had not registered in advance: how far each person considered the behaviour to be ghosting at all, on a scale from “not at all” to “extremely”. Among the 121 participants who rated it above the midpoint, 46 who left and 75 different people who were left, the underestimation returned, 3.93 against 3.51, at d = 0.43. Those are not paired figures, and the paper prints no comparison for the people below the midpoint. A more inclusive cut that also keeps the participants sitting exactly on the midpoint puts the same comparison at 3.72 against 3.43, at d = 0.30. What it does print is that the leavers’ bad feelings climbed with how strongly they read their own silence as ghosting, r = .42, while the waiting side’s predictions stayed nearly flat against their own reading of it, r = .13.
What survived the live test, and survived it large, is a different misreading. On a scale from “extremely positive” to “extremely negative”, people left waiting put their partner’s feelings towards them at 3.41; the partners reported 2.29, at d = 0.91. And on a scale from “not at all” to “very much”, the waiting side put their partner’s interest in a friendship at 1.53, against a reported 2.58. The person who walked out of the chat window was, on average, warmer about the person left in it than that person guessed.
The word each side used
The live study turned up something the first two could not, because it had both halves of the same interaction on record.
They did not agree on what had happened. Asked how much the behaviour counted as ghosting, the people who left averaged 2.56; the people they left averaged 3.06. The leavers had been reminded repeatedly that they had chosen to skip the conversation and leave someone waiting alone without explanation. They still scored it lower.
The authors are direct about the limits. A short online study, they write, cannot fully capture the stakes of a long relationship coming apart. Their participants had known each other for eight minutes and, the paper thinks, most likely expected never to encounter each other again. And the waiting partner had been told in advance that the other might not rejoin, which may have preemptively reduced the uncertainty that typically comes with being ghosted. Together, they write, those limits made the study a relatively conservative test of their own hypothesis. Separately, in the paper’s own limitations section, assigning ghosting to people in an experiment poses problems both ethically and for internal validity, and the authors suggest embedding short surveys in messaging platforms as one way to catch the real thing.
What this piece rests on, then, is five-point scales filled in by people recalling a text thread, people imagining one, and people who had met eight minutes earlier, with the materials, data and analysis scripts posted openly. That is enough for a claim about averages, with the paper’s own scope attached: in recalled experiences, in imagined scenarios, and in the live conversation where the non-response was recognised as ghosting, the state behind a non-response is more uncomfortable, and less indifferent, than the person waiting assumes, and in the scenario study a short acknowledgement narrowed the distance between the two readings. Why the person you are thinking of has not replied sits outside it. The samples were US adults describing acquaintances and friends they were not close to, plus strangers paired for eight minutes, not your friend. And if a repeated silence from someone close to you is doing real damage, a therapist or counsellor is the right address for it.
The usable part
The residue is small enough to fit in a sentence. If you are the one who has gone quiet, a one-line acknowledgement was the thing that closed part of the gap in the scenario study, though that study never varied what went into the line: the apology, the reason and the implied later all arrived together, and the authors say they cannot tell which part did the work. If you are the one waiting, the flat, indifferent version of the other person you have assembled is probably a worse likeness than the data supports; misreading someone as indifferent could, the authors argue, reduce the chance of anyone picking the thread back up. Silicon Canals has written before about the person left waiting on a message, and this is the same silence from the other chair.
What would settle it is a comparison the paper does not report: both members of a pair rating the same silence as ghosting, and their feelings weighed against each other inside that pair. The authors’ own proposal for better evidence points away from the lab altogether, towards short surveys dropped into the messaging apps where the silence actually happens.
