Wanting to talk, and what to trade it off with
Before you read any further, please ask yourself how good you are, relative other people, in answering the following question: “Dendrophobia is the fear of what? A: Spiders, or B: Trees.”
Next, please ask yourself how good you are, relative other people, in answering a variant of this question: “Dendrophobia is the atypical fear of what? A: Spiders, or B: Trees.”
It is plausible that adding the word “atypical” did not only help you to find your answer, but that it also increased your confidence about your answering skill, relative to that of other people. Correct? (But note: they get the exact same question.) Now, suppose that you are in a quiz game with a team member, and in your internal team discussion you have to be really quick, such that only one of you can talk. If the current challenge is the second version of the dendrophobia question, would you be willing to speak up and explain your thoughts, more so than in case of the first version? Most people would do so, although it is unclear whether talking is really more appropriate in the second version: if matters are clear already, do you really need air time?
More generally, what is the economic cost of talking? The benefit that talkers create is, at least theoretically, quite strong: the desire to talk increases the supply of information, and hearing multiple opinions makes it possible for the listener to benchmark the statements against each other. (This benchmarking is especially helpful in cases where every speaker introduces his or her own biased view, as demonstrated by Battaglini, 2002.) Experiments that allow for a competition of talkers confirm that having such a competition can indeed be informative (Wilson and Vespa, 2016) but they also show that problems of coordination between the candidate talkers arise: they need to figure out who should talk how much (Grabova, 2023).
Vespa and Weizsäcker (2023) investigate a novel experiment that highlights one aspect of this discussion: the informational trade-off between talking and listening. If a person talks, she cannot listen to others at the same time, and vice versa. It is a tradeoff that arises in most business meetings that we attend, and in most private conversations, too. Only one of us can show his or her skills in speculating and formulating, about dendrophobia or anything else.
To judge whether we should talk or listen, we may follow an important principle from the field of linguistics: relevance. Conversational utterances are always relevant to the talker, in some way or other, and the listener’s task is to find out about the way in which they are relevant. Dan Sperber and Deirdre Wilson developed a whole subfield of linguistics, relevance theory, based on the premise that for both the talker and the listener, cognition and communication is relevance oriented. The theory’s key element is nicely worded as the “implicit guarantee of relevance”: it is known that the talker would say something only if he has something relevant to say (Sperber and Wilson, 1995). The listener can therefore examine the statement, and its context, to find the relevant parts and put them together.
Applying this simple logic in an exact way is difficult, however. The logic implies that the listener makes her belief update not only based on the statement, but also based on the fact that the talker decides to make a statement. This fact indicates a sense of urgency, which carries information: the talker is sufficiently convinced of what he has to say. While this is informative, it also leaves room for interpretation by the listener, because she cannot know why exactly the talker is so confident. What’s worse, the interlocutors have to account for all of these inferences from the perspectives of both roles. If person X decides that he wants to talk, then he considers that person Y takes this sense of urgency into account when she makes her inferences after hearing the statement. He also knows that Y considers her own talking decision with the analogous consideration in mind: she would believe that he would believe that … and so on. It is quite impossible to judge how well real people understand all this. A vast source of possible misunderstandings.
In an experiment, however, one can shut out a large part of this difficulty and avoid that the decision to talk is informative per se. The trick is to make the statement so informative – using numeric values that can be sent from the talker to the listener – that if a statement is made, then it carries all information that a talker has. The basic principle of an implicit guarantee of relevance is not affected by this experimental trick (the talker still talks if and only if he finds it relevant) and the experiment is a suitable representation of the talking-listening tradeoff.
One disadvantage of such a laboratory experiment arises, though: it is artificial. The information structure is abstract and boring and, clearly, real-world conversations are not made by sending numerical values to and fro. One should not, therefore, extrapolate too much from the laboratory to the outside world. Or, one may include a less abstract (and admittedly less controlled) parallel variant of the experiment, with more fun and realism: the dendrophobia question and other questions like it.
A key advantage of such an experiment, one that applies even to the abstract version of the game, is that one can compare different experimental conditions: in one condition, everyone is well informed, i.e., has a relatively precise signal structure (the insertion of words like “atypical”, or other help), whereas in another condition the signals are less precise.
The result of this variation is an instance of the hard-easy effect on people’s relative confidence (Moore and Healy, 2008). If a task is hard, we all tend to think that we are relatively bad at it. Think of calculating square roots of large numbers in your head: no-one would believe that he or she is in the top group of square-root takers. Now think of an easy task, like driving a car without accidents. Here, most people believe that they are in the top group, i.e., in the upper deciles of the distribution. This pattern is clearly irrational because the upper deciles cannot magically grow only because driving a car is simpler than solving a math problem. What we tend to forget in our self-centered assessment is that driving a car is simpler than a math problem for everyone in the distribution.
In the trade-off between talking and listening, this exact pattern appears, too. Inserting a word like “atypical” helps everyone in the experiment, but the willingness to talk and explain one’s thoughts, versus to listen, increases by about a third, from 38 to 51 percent. In the numerical variant of the experiment (more boring, but better controlled communication tasks), the effect size is smaller but still significant, with about a one-tenth increase in the frequency of talking. In all parts of the experiment, participants are talking more if it is easy to talk. In situations where the topic is difficult, there is more silence.
Author: Georg Weizsäcker
Battaglini, Marco (2002), Multiple Referrals and Multidimensional Cheap Talk. Econometrica 70(4), 1379-1401.
Grabova, Iuliia (2023), Confidence and Communication: Too much air time for some? Working paper, Humboldt-Universität zu Berlin.
Moore, Don A. and Paul J. Healy (2008), The trouble with overconfidence. Psychological Review 115, 502–517.
Sperber, Dan, and Deirdre Wilson (1995), Relevance: Communication and Cognition (second edition). Blackwell Publishing.
Vespa, Emanuel, and Georg Weizsäcker (2023), Do we talk too much? Working paper, Humboldt-Universität zu Berlin.
Vespa, Emanuel, and Alistair Wilson (2016), Communication with multiple senders: An experiment. Quantitative Economics 7 (1), 1-36.