When Should a Conversation End?

Why people keep talking long after you want them to stop.

Posted Dec 03, 2018

Estrada Anton/Shutterstock
Source: Estrada Anton/Shutterstock

People have remarkably consistent expectations about how long things should go on for.

For example, right now on Spotify's chart for the top global tracks, the lengths of the top five songs are 3:27, 3:32, 2:16, 3:12, and 3:30 minutes long. How long should a pop song be? About three minutes and 30 seconds, give or take. Here's another example: How long should a good fiction book be? The top five most-read fiction books on Amazon this week are 400, 417, 464, 322, and 815 pages. (This last one is a Harry Potter book, but the next two non-HP books are 417 and 379.) Answer: about 400 pages, unless it's part of a well-established series. What do you think the result would be if I looked at the same trends for movies — just south of two hours, right? We have these lengths pretty much down to a science.

So here's an example that's at least as prevalent in our social lives, but much more difficult to answer: How long should a conversation be? 

Emerging research by Harvard dream-team Adam Mastroianni, sage young gun, and Dan Gilbert, spry old guard, explores this issue. Their experimental paradigm: lock two people in a room and see how long they converse before asking to be let out. What they found: People have no idea how long the other person wants to talk for. 

Of course, how long two people decide to talk for depends a lot on how long they expect to talk for going in. The technical term psychologists use to describe this expectation is a prior. As in, your belief about how long conversations go in general prior to knowing anything specific about this particular conversation. Previous research by cognitive scientists Tom Griffiths and Josh Tenenbaum showed that people's prior expectations are rather well-defined across a range of issues. People expect movies to be mostly just about 110 minutes, with a few going longer than two hours, but almost none going shorter than 90 minutes. They expect poems usually to be a few lines long, but allow that some can go up into the thousands. They expect that the majority of cake recipes call for precisely 60 minutes in the oven. And they even expect the reigns of most Egyptian pharaohs to have lasted on the order of only a few years.  

Furthermore, Griffiths and Tenenbaum showed that these prior expectations have a significant influence on people's predictions about how long something will last for, even when you give someone information about how long it's gone on so far. For example, they asked, "If your friend read you her favorite line of poetry and told you it was line 5 of a poem, what would you predict for the total length of the poem?" The typical respondent might say 10 or 20 lines. They even asked such ponderous questions as, "If you opened a book about the history of ancient Egypt to a page listing the reigns of the pharaohs, and noticed that at 4000 BC a particular pharaoh had been ruling for 11 years, what would you predict for the total duration of his reign?" 

Mastroianni and Gilbert played out this kind of scenario in the lab. Instead of asking them how long they think it will go, their participants were allowed to decide for themselves when to stop. They paired people up and asked them to talk. When they mutually agreed to terminate the conversation, the experimenters asked them a couple follow-up questions. Then they were on their way. In a poster at a recent meeting of Society for Judgment and Decision Making, they reported that only 15 percent of people left the conversation when they actually wanted to. 

The average conversation lasted just under 30 minutes. If you had asked me beforehand, I would have predicted that this was longer than anyone wanted to talk. As it turned out, about half of the participants wanted to leave earlier, and half wanted the conversation to keep going when their interlocutor suggested they stop. If you wanted to stop early, you wanted the conversation only to go about half as long as it did. If you wanted to keep going, you wanted the conversation to keep going for about twice as long as it did. One partner thought they were reading Animal Farm. The other was ready to plunge into the whole Lord of the Ring trilogy. 

Part of the reason for the mismatch was that no one else knew what the other person wanted. When people tried to guess what their partner had wanted, they really had no idea. We may have a pretty strong understanding of how long we want something to go on for, but it seems like we're not as successful in extending that understanding to include other people. Further compounding the issue, people had no idea that they had no idea about their partner's expectations. My guess would be that this is because people were basing their estimates of how long the conversation should go more on their prior expectations than any of their partner's conversational cues. 

Mastroianni and Gilbert's findings have profound and immediately implementable societal implications: We need to be clear with other people when we're done talking to them. We don't know how long they want to talk for. They don't know how long we want to talk for. And the two parties likely have different expectations anyway. Let's just accept that and be more clear with our interlocutors about when we're ready to move on to the next activity. 


Griffiths, T. L., & Tenenbaum, J. B. (2006). Optimal predictions in everyday cognition. Psychological science, 17(9), 767-773.

More Posts