Why People Skip the Room: The Forecasting Error Paper No. 12, The Gathering Architecture Papers Author: David Adler. Publisher: GatheringPoint.news. Free and ungated. Web version: https://gatheringarchitecture.com/papers/why-people-skip-the-room This file is the complete plain-text copy of the paper, provided so text-only readers and AI agents can read the full work. ======================================================================== Gathering Architecture Why People Skip the Room: The Forecasting Error Why people skip a gathering they want to attend, what may be a forecasting mistake, what it costs organizers, sponsors and guests, and three tools that correct it. A Gathering Architecture report · Working draft · Nov 18, 2026 In this report 01 The question behind this report 02 What is at stake, and what nobody has counted 03 Why we do not go: the forecasting error 04 Building the muscle: what you do as an individual 05 Three tools for one mistake: what to change in your event 06 What we do not know yet 01 The question behind this report The usual explanation for an empty seat is that the person would rather stay home. This report looks at a different explanation: that they expected the room to be worse than it will be. We found research suggesting that people predict conversations with others will be more awkward and less rewarding than they turn out to be, and that the people who actually have those conversations find the opposite of what they feared. That is our starting point, not a finding about events. What follows tells the story with evidence where evidence exists, and says so plainly where it does not. Inference If the forecasting error holds for gatherings, three tools work on it at three different moments. Marketing gets people to the door, facilitation corrects the guess inside the room, and practice makes the correction last. All three are our inference from the research, not tested results. A note on how to read this report. Cited findings carry a source in the final section. Our own reasoning is labeled as inference. Where the evidence is thin, comes from industry surveys, or rests on a small sample, we say so in the text. This report has a companion on how digital contact builds the wish to meet, The Path From Digital to Face-to-Face: Read it here . 02 What is at stake, and what nobody has counted Nobody has measured how many attendees stay home because of the forecasting error, and we found no study that comes close, so this report does not offer a number. The nearest figures are adjacent, not direct. PheedLoop, an event software company, found in its data on more than 1,070 events that the median no-show rate is about 20 percent, roughly 1 in 5 expected attendees, and it notes the figure does not separate people who cancelled from people who simply did not appear. It also gives no reasons, so some share of that fifth stays home for good reasons, such as illness and travel, and we cannot say how large the forecast share is. In the Sandstrom, Boothby and Cooney study, only 4 in 10 people expected the first stranger they approached to talk to them, and 9 in 10 did. A poll for the food-delivery app Yelp Eat24 found 55 percent of 2,000 adults make excuses to skip social gatherings at least some of the time, but it is a self-reported poll with an income skew and its reasons were mixed, so we treat it as color and not evidence. Inference The loss is shared. The organizer pays for the seat that stays empty, the sponsor or exhibitor for the audience that never arrives, and the guest for the introduction that never happens. Of the three, the guest's loss may be the largest and is the one no ledger records: the friendship, the job, the partner or the change of direction that starts in a room and does not start for the person who sends the polite note. We label this inference because the research shows people enjoying conversations more than they predicted, not that any conversation changed a life. [Attendees lost to the forecasting error: no measurement exists yet; this is the study someone should run.] For your event You can start measuring your own version this month. Ask no-shows one question in the follow-up email, "What made you decide not to come?", with a choice that reads "I expected it to be awkward or not worth it," and compare that share with the people who attended and said afterward how it went versus what they expected. 03 Why we do not go: the forecasting error Your browser cannot play this video. Download it here. Watch: The Neuroscience of the First Hello (5 min 49 sec). An illustrated explainer on the forecasting error: why we expect a conversation with a stranger to go worse than it does, the commuter experiments, the liking gap, and why practice shrinks the mistake. It is a companion to this section, not a substitute for it. The studies it draws on are cited in the Sources below. In 2014, Nicholas Epley and Juliana Schroeder published nine experiments, involving more than 800 people, most of them commuters, that ask a plain question: why do people sit silently next to strangers? Commuters were told to talk to a stranger, to sit in solitude, or to commute as usual. Those told to connect reported the most positive experience, with no cost to their productivity. Separate groups asked to predict how each condition would feel expected the opposite, believing that connecting would be less pleasant and less productive than solitude. The authors trace part of the error to a belief about other people. Participants assumed strangers were less interested in talking than they actually were, and because they never tested the assumption, they never learned it was wrong. A 2021 follow-up by Schroeder, Lyons and Epley found the same shape: people expect a conversation to be less pleasant than it turns out to be, because they expect the other person to be uninterested. The limits are real. The participants were strangers on commutes, not colleagues at a conference, and Epley himself notes that talking to strangers may not bring the lasting benefits of time with friends. Inference Deciding whether to attend a gathering is also a forecast, and it may carry the same bias. Someone weighing a trip pictures a room that costs more and gives less than the room will. If that holds for conferences, part of the barrier to attendance is a mistaken prediction, which is something an organizer can address. This has not been tested on event attendance, and we flag it as a research question. For your event Your prospect is deciding with a biased forecast, so correct it. Describe the people and the company in the room rather than the program, and put a host in the room whose job is to start conversations: to make introductions, to signal that everyone came to meet, and to take the first step off the attendee's hands. The evidence for a host is indirect. In Epley and Schroeder's experiments, the researchers' instruction to connect acted as permission, and in a lab waiting room, people who were spoken to were as positive as people who did the speaking. Nobody has tested a facilitator at a conference, so treat the host as a promising move to measure, not a proven one. 04 Building the muscle: what you do as an individual A host can lower the cost of the first step, but a host cannot be everywhere, and permission borrowed from someone else is not the same as a skill you own. The research suggests the skill can be trained. In a 2022 study in the Journal of Experimental Social Psychology, Gillian Sandstrom, Erica Boothby and Gus Cooney gave 286 students at two universities, one in the United States and one in the United Kingdom, a week-long scavenger hunt played on a phone app. Each task asked the player to find a stranger who fit a description, such as someone wearing interesting shoes. Half the players had to talk to the stranger, while a control group only had to observe. Compared with the controls, the talkers grew less pessimistic about rejection, more optimistic about how conversations would go and more confident, and they reported more conversations with strangers in the week that followed. The beliefs shifted gradually over the week, which points to repetition, not a single good experience, as the active ingredient. The same researchers' earlier work explains what is being overcome. In a 2021 analysis across seven studies and 2,304 people, Sandstrom and Boothby sorted the fears into three kinds: that the conversation will not be enjoyable, that the other person will not like you, and that you lack the skill to carry it. Across those studies the fears turned out to be overblown. A related result from Boothby and colleagues in 2018 is called the liking gap. After conversations lasting anywhere from two minutes to 45 minutes, people consistently underestimated how much their partners liked them and enjoyed the conversation, and the people who had more negative thoughts about the interaction showed the larger gap. Sandstrom and her coauthors also note that earlier attempts to talk people out of these fears by telling them the fears were exaggerated had failed, which is why their intervention asked people to do the thing and see. Inference Read together, these studies suggest that the insecurity behind not starting a conversation is a set of three predictions (it will be dull, they will not like me, I will not know what to say) and that each prediction gets corrected by evidence the person gathers personally, in small repeated doses. That makes starting conversations resemble a muscle in one respect: it responds to reps. It also suggests the scavenger hunt worked partly because it supplied a task and a reason to approach, which is a version of the permission a host provides. Neither point has been tested at conferences, and the participants were students playing a game, not professionals on a show floor. For you, as an attendee or a member Treat your next event as practice, and begin with a small, specific number, such as three conversations, rather than a vague aim to network. Give yourself a reason to speak, such as a question about the session you just left or the person's badge, because the research suggests a task makes approaching easier. Keep the first conversations short, and afterward write down two things: how you predicted it would feel and how it did. The gap between those two notes is the evidence your next forecast will draw on. Everything in this paragraph is our inference from the studies above, offered as a way to test the research on yourself. For your event Do not make the host carry the whole job. A host can supply the first reps and then step back, which turns your event into a place where attendees build a skill they take home. Build a light task into the program, such as a find-someone-who prompt on the name badge or a question printed at each table, so that every attendee has a reason to speak that does not depend on a person with a title. Test: compare how many new conversations attendees report, and how comfortable they say they felt, between sessions that offered a prompt and those that did not. 05 Three tools for one mistake: what to change in your event This section speaks to you directly. Everything in it is inference, drawn from the research above and not tested on events. Each move comes with a cheap test, so that you can find out whether it works at your event before committing to it. Marketing gets people to the door: sell the meal, not the menu. Lead with who will be in the room and what time with them is like, and treat the program as supporting evidence. The attendee's picture of the room is the weak point, so make it easier to picture, and introduce registrants to one another before they arrive. Earlier research found that telling people their fear is exaggerated did not change it, so expect marketing to help and not to finish the job. Test: run two versions of a registration page, one led by the program and one led by the people, and compare conversion. Facilitation corrects the guess inside the room: put a host in the room. Hire or designate someone whose job is the first step: introducing people, signaling that everyone came to meet, and getting the first laugh going. The classic version is the tumler, the resort social director who got guests mingling. Add a question printed at each table or a find-someone-who prompt on the badge, so that every guest has a reason to speak. This rests on the permission logic in the forecasting section and has not been tested at conferences. Test: compare the number of new conversations attendees report at tables with a host and tables without one. Practice makes it last: build reps into the program. Give attendees small, structured chances to start conversations, so that the skill goes home with them. Test: compare how many conversations attendees report, and how comfortable they felt, between sessions that offered a prompt and those that did not. The forecast conundrum. A conundrum, as we use the word, is a structural bind in which every party acts sensibly and the result is still worse for all of them. Attendees underestimate how good the room will be, and event marketing usually describes the room in terms of its program, which is information. Information is what a menu carries, so the pitch can confirm the attendee's low forecast instead of correcting it. This is our inference, offered as a hypothesis to test. 06 What we do not know yet The central hypothesis of this report, that the decision to skip a gathering carries the same forecasting error as the decision to skip a conversation with a stranger, has not been tested on event attendance by anyone we found. The supporting research is adjacent: Epley and Schroeder studied strangers on commutes, and Sandstrom and colleagues studied students playing a scavenger hunt. The no-show figures come from one software company's data and carry no reasons. Nobody has tested a host or a table prompt at a conference. Three studies would move this from argument to evidence. The first would ask people deciding whether to attend an event to predict how it will feel, and then ask them afterward how it did feel. The second would ask no-shows why they stayed home, with the forecast as one choice among several. The third would compare new conversations between tables with a host and tables without one. Any of these could be run by a single organizer, and we would welcome the results. Sources and notes Pages marked (opened) were read in full for this report. The rest were read as search excerpts, so confirm any figure drawn from them against the original before publication. Forecasting error: Epley and Schroeder, 2014, Journal of Experimental Psychology: General (opened) ; British Psychological Society digest ; ScienceDaily ; Schroeder, Lyons and Epley, 2021 Practice and the fear of talking to strangers: Sandstrom, Boothby and Cooney, 2022, scavenger hunt study (abstract via Sussex) ; SPSP summary by the authors ; PsyPost on the study design ; Sandstrom and Boothby, 2021, Self and Identity ; Boothby et al., 2018, the liking gap ; APS summary of the liking gap What is at stake: PheedLoop Event Data Lab Report #05, April 2026 , a median no-show rate across 1,070+ events with no reasons given; HuffPost on the Yelp Eat24 poll , self-reported and income-skewed; New York Sun on Sandstrom, Boothby and Cooney , the 40 percent versus 92 percent figures A Gathering Architecture report. Working draft: sources marked as read from excerpts should be confirmed against the originals before publication. Our own reasoning is labeled as inference.