What Is Signaling?

Noah Smith complains about people like me:

There’s a fad in the economics world that annoys me. The fad is to describe every human action as “signaling.” This has to stop, people. … It’s become fashionable in the economics world to label any and every human social interaction as a form of signaling. The most enthusiastic promoter of this way of thinking is GMU economist Robin Hanson. Fashion isn’t self-expression — it’s signaling. Leisure isn’t about fun — it’s about signaling. And so on.

The problem is, this notion of “signaling” isn’t really what Spence had in mind. Spence’s signaling model was about proving yourself by doing something difficult — something so difficult that someone who didn’t have what it takes wouldn’t even bother. But most of what Hanson is talking about is just communication, not Spence-style signaling. Even if hipsters wax their moustaches in order to prove their hip-ness, that doesn’t mean there are a whole bunch of wannabe hipsters out there who just didn’t have what it takes to wax their moustaches. Communication, like signaling, is costly. But it’s not a matter of jumping through hoops to prove yourself. (morefollowup)

Let’s distinguish three different kinds of messages I might send with my waxed moustache:

1) “I have thick shiney hair.” This message is verifiable. Soon enough, others can just directly check if it is true. So I don’t need to pay costs to send this message, though I may pay costs to create the nice hair.

2) “Hipster is one of my interest areas.” If you and I are going to talk anyway, but must pick a conversation topic, we may share a sufficient common interest in finding talk topics of mutual interest. In such a context, it can be enough for me to just tell you about my interests. You can just accept my claims for the purpose of picking a talk topic. Technically, this is a “cheap talk” message.

3) “I am especially devoted to the hipster ethos” or “I especially embody hipster ideals.” That is, I am especially willing to identify myself as a hipster, and my personal features are an especially high quality match to ideal hipster features, including having a creative and contrarian yet attractive and coherent personal style that fits with current hipster fashions. These messages are hard to verify, and the interests of observers and I conflict. While observers want to accurately rank me relative to others, I may want them to estimate me as having maximal devotion and quality. Since verification and cheap talk won’t work here, I have to show, not just say, my messages.

To show my hipster devotion, I can choose an appearance that is sufficiently off-putting to ordinary people at work, home, church, etc.. By paying the cost of putting off possible associates, I show my devotion to hipsterism. To show my hipster features, I can pay to track hipster fashions and to continually search in the space of possible styles for a combination that simultaneously reflects current fashions while being creative, coherent, and showing off my best personal features. Not being a hipster, I don’t know how exactly that works for them. But I do know, for example, that since lipstick and tight clothes make some bodies look better while making other bodies look worse, they are costly signals of the quality of lips and body shape. There must be similar factors for showing off hipster qualities.

More generally I call a message “signaling” if it has these features:

  1. It is not sent mainly via the literal meanings of words said.
  2. It is not easily or soon verifiable.
  3. It is mainly about the senders’ personal features, perhaps via association with groups.
  4. It is about sender “quality” dimensions where more is better, so senders want others to believe quality is as high as possible, while others want to assess more accurately. Such qualities are not just unitary, but can include degrees of loyalty to particular allies.

Cheap talk cannot send a message like this; one cannot just say such a thing, one must show it. And since it cannot be verified, one must show it indirectly, via how such features make one more willing or able to do something. And since willingness and ability track costs, these are “costly” signals.

When weighted by how much the messages matter to us, and by how much effort we put into adjusting them, I’d say that most of our communication is “signaling” of this sort. Most of the private value, if not most of the bits.

GD Star Rating
Tagged as:
Trackback URL:
  • Daniel Carrier

    > More generally I call a message “signaling” if it has these features:

    You’re listing features of signaling, but you’re making it sound like they define signaling, which they do not. Signaling is when you pay a cost to show something because the benefit given that it’s true makes it worth it, and the benefit given that it is false would not be worth it.

    Talking isn’t signaling because it’s cheap and easy to lie. There is a cost to lying, but since the cost is only paid if you lie, it’s not signalling.

    Things that are easily and quickly verifiable cannot be signaled since it would never be worthwhile to pay the cost. Although you have to be careful with that one. If someone simply can verify something, you may need to signal it before they actually do. It’s generally possible to get around this with cheaper methods though, so I’m not sure how much this happens.

    Senders must want people to believe it’s as high as possible, or they wouldn’t benefit from the cost of signaling.

    • Faze

      So signaling is any costly gesture whose expense can be taken by another as a measure of commitment.

    • Joe In Morgantown


      Signaling does not have to expensive if you do have the quality being signaled, it only has to be expensive if you do not have the quality.

    • UWIR

      There is a massive presumption of truth in speech; barring a highly improbable claim, a perceived ulterior motive, or particularly low credibility, people tend to assume that anything they hear is true. Speech that depends on that presumption (which is most speech) is not signalling. However, it is quite possible for speech to be signalling. However, for speech to be signalling, there usually has to be some other agent whose actions are being relied upon. For instance, claiming that your product is low fat can viewed as signalling in the sense that a product that isn’t low fact has a greater chance of resulting in penalties from regulatory agencies.

      • Daniel Carrier

        I’d say most speech is signalling. Small talk isn’t about the explicit information transfer. It’s about showing you care enough to spend five minutes just to show that you care.

  • I think Noah forgets to factor in the opportunity cost — not just the cost. When you wax your mustache, you are giving up some particular forms of employment … But then not for all. In the labs I work in, no one would bat an eye … It becomes a more nuanced problem.

  • pgbh

    Noah: “People say college signals quality X. But there are better ways to show off X, for example Y and Z. So it doesn’t make sense that people would choose college to display X.”

    He could be correct, but I doubt it. Another possibility is that a college degree signals additional qualities, which are not correlated with Y and Z. Then a college degree could be the best way to show off all these qualities together.

    Noah rejects the idea that college could be the best way to signal either intelligence or conscientiousness. He doesn’t consider that it could be the best way to signal both together.

    Or more likely, college shows off both, combined with additional qualities we also value but don’t like to discuss too openly.

  • One problem with your definition is that it is all contemporaneous, whereas in practice you’ve gone beyond this with regard to the object of signaling.

    This is relevant to the Noah debate. Brian Caplan regards the signals education provides to employers as contemporaneously useful to them. (And in this he might actually weaken his case.) Noah’s Japanese example (where college is a moratorium) does seem to refute Caplan’s explanation.

    The main reason I find to think college education has a large signaling component is its inefficiencies. There isn’t much concern about the effective transmission of knowledge. But what is being signaled? It probably has to do with personal associations, such as with the “intellectually powerful” professariat (per Robin) more than useful personal traits (per Brian).

  • DanielHaggard

    Nice clarification… I really like that you included 1) – as I think this is an important feature of human signalling that distinguishes it from signalling in other animal domains.

    Semantic values of utterances tend to get less important as signalling behaviours become more prevalent. This is why – say – in a scientific context you should want to do your best to avoid signalling behaviours so that the focus can remain on the semantic values of the theories under consideration.

    I do think you should include other features of signalling that hail from the standard signalling story.

    5) There needs to be an asymmetry of information between parties.
    6) Transference of that information is mutually beneficial.
    7) Cheating by the sender of information should have a short run benefit for the individual sender, and short run cost for the receiver – but a long run cost for both the sender and receiver as the system of communication breaks down.

    Also – in the domain of human signalling I’d add another feature that I think is important. Because talk is cheap – we actually go to great lengths to increase the costs of speaking. We punish people, for instance, who say the wrong things on twitter etc.

    By increasing the costs in this way – we actually enable certain forms of speech to become costly enough to serve as honest signals.

    • On #5, there can still be a signaling message even when the audience already knows the message. Many messages are sent to a large audience, where some of that audience already knows. On #6, many listeners choose wide listening strategies, which let in info that hurts them sometimes. If they know that a particular info would hurt them they might try to not listen to that, but it can be hard to target your listening that specifically. On #7, with large groups of senders and receivers each one need not internalize the consequences of hurting the communication system. So individuals might well send signaling messages that hurt the system.

      • DanielHaggard

        Thanks for the reply – I should have probably also included #8 That sender and receiver should have competing interests; but I’m going to guess that you would imagine scenarios in human signalling where the interests of sender and receiver don’t compete in any significant way.

        I have to think about your reasons for leaving the others out. In the evolutionary biology case costly signalling was proposed to explain how do individuals with competing interests exchange information given that there is an incentive to cheat/lie and ruin the communication system? How does such a system maintain equilibrium? That’s why the signals must be costly, to make the short run benefits of cheating not worth it – which protects the system as a whole.

        So if you’re going to leave #7 out, for instance, then I lose my understanding of why we evolved a COSTLY signalling system at all. Because it’s that cost structure described in #7 that makes the costliness necessary.

  • andrew

    Why isn’t all communication signalling? Why isn’t all behavior signalling? I am thinking Robin Trivers and self deception. We are just better off if everyone is unaware that we are one big complex signal. Aren’t all prices just signals?

    Don’t let Noah frame the debate he is too earnest. He would rather be liked than correct like when we attacked you for comparing rape and cuckolding. You are being aware and honest and that signal is not welcome in public discourse.

  • Pingback: “Signaling” isn’t about signaling – Mortgage Rates()

  • Pingback: Measuring Students – mindlevelup()

  • Pingback: 24 – Signaling | The Bayesian Conspiracy()

  • Pingback: Some Noise on Signaling – Rational Reflection()