A Cat in a Bag: People Under Incomplete Information

Man is a mystery.

— F. M. Dostoevsky

The seller of a used car knows what happened to the gearbox last winter. The buyer sees a clean body and hears a confident engine for fifteen minutes. In 1970 George Akerlof showed that this kind of information asymmetry can destroy a market. If the buyer cannot tell a good car from a lemon, he offers an average price. Owners of good cars do not want to sell at the average price and leave. The share of bad cars rises, the buyer lowers his price further. One side’s lack of information degrades the composition of the offers themselves.

Potential partners are not cars. But the informational problem is recognisable: each knows more about himself than the other does, and the most important qualities are hard to verify in advance.

“I am reliable”, “I don’t play games”, “family comes first for me”, “I know how to talk about feelings” — these are cheap signals. They can be said by a person who matches the description and by a person who knows the description goes down well. Lying is not required. People often sincerely overestimate their own reliability. They describe an intention, while their partner needs a forecast of behaviour under fatigue, fear and conflict. A profile makes the problem worse: it selects the traits a person controls. The photo, the wording and the list of values are the carefully prepared side of the data. What is harder to control continuously shows up later.

There is no magic question a bad partner will fail to answer correctly. As soon as the question becomes known, the right answer becomes cheap.

Good options sometimes look worse

Akerlof’s model has an unpleasant corollary: high quality can disappear from the supply precisely because it cannot be proved in advance. The buyer pays for the average, the owner of high quality refuses an average valuation and leaves; after he leaves, distrust becomes even more reasonable. What matters for dating is not a literal market but the same selection in favour of properties that are easy to display quickly.

Dating has no single price, but the selection can work in a similar way. Someone who promises brightly, quickly and without qualifications is more visible than someone whose best properties show up in a sequence. Reliability fits badly into a profile: anyone can write “I keep my word”, and a person who really has the quality cannot attach ten years of future behaviour to the phrase. In a market of quick impressions, a deep but slowly observable quality loses, for a while, to a confident presentation.

The channel rewards what is easy to show and undervalues what is expensive to verify. Charisma can belong to a reliable person and modesty to an unreliable one; early selection still favours whoever produces the better early impression. Reputation, mutual acquaintances and repeated situations make it possible to account for qualities that do not fit into a promise.

This changes the receiver’s task. You do not have to guess a “true essence” from someone’s tone of voice. You have to create conditions in which the types start behaving differently. Cheap promises leave everyone looking the same; small commitments, time and watching actions gradually separate the sample.

The expensive signal

Michael Spence called a signal informative when its cost differs between types of participant. In relationships that signal is consistency: arriving on time many times over, keeping small promises, staying respectful after a refusal, admitting a mistake without an immediate reward. A single episode is easy to act out; a sequence is usually more expensive to maintain for someone whose ordinary behaviour contradicts it. That does not make a series an infallible test. A large gift is expensive in money terms but weakly connected to reliability: it costs a rich man little, and it can quickly put the recipient under an obligation. Cost conveys quality only when it arises from the very property being tested.

Experiments using descriptions of acts of friendship and romance show that costly actions are read as stronger signals of commitment. But what was measured was how convincing the signal was to an observer, not the sender’s actual honesty. An expensive gesture is harder to imitate, yet it can still be a strategic investment. For predicting reliability, what matters more is the link between the cost and the property being tested, and repetition across independent situations.

Informativeness comes from the link between cost and the quality you need to learn about, not from the absolute price. Accepting a boundary calmly is usually harder for someone inclined to control. Keeping a consistent story going for a long time is harder for someone who lies. Treating people you need nothing from with respect is harder for someone whose politeness is a tool of the deal.

Reputation as collective memory

Tight communities carried information about a person between relatives, neighbours and colleagues. Bad behaviour in one relationship worsened future opportunities, good behaviour accumulated a reputation. Along with the memory came control, gossip and little freedom to leave. The modern city reduced village surveillance, but it also deprived the stranger of a ready-made history. After vanishing from one chat you can open another; a platform’s shared memory is weak, and public ratings of people would quickly turn into harassment.

As a result, personal observation over time takes on a larger role. It is slower than recommendations and better for autonomy.

Five repetitions of one nice story are still one story. Different circumstances work better: interest, a small disappointment, a change of plan, one’s own mistake, money, friends, being served by a stranger. On a first date almost everyone shows behaviour in maximum-motivation mode; independent samples reveal what survives a change of mode. Natural situations are valuable because they are harder to stage for effect, and they require no hidden exams.

The Bayesian logic here is simple. The first impression sets a prior probability. Every new action updates it. A strong promise without behaviour changes little. A repeated sequence changes more. One bad episode may be noise; the same pattern across different circumstances is already a signal.

Time does not guarantee truth

The advice to “just wait” also has limits. Some people can keep up an image for a long time, especially when the stakes are high. Others show their worst qualities only after dependence has appeared — shared housing, a child or financial entanglement. You pay for that information with growing investments and with the cost of a mistake. Reversible stages let you accumulate information before the large bets. Small commitments first, larger ones later. This has nothing to do with corporate vetting: the ability to keep promises only becomes visible after the promises themselves. Shared housing, a joint account and a child provide a lot of data, but they are far too expensive as diagnostic tests.

There is almost no unfakeable sign. Several independent observations and small commitments give you more than one perfect question. It makes sense to raise the stake as the data accumulate. The bag is opened gradually — before housing, money and the whole of a future life are tied to it.

Main sources

Akerlof, G. A. (1970). The market for “lemons”: Quality uncertainty and the market mechanism. Quarterly Journal of Economics, 84(3), 488–500.

Spence, M. (1973). Job market signaling. Quarterly Journal of Economics, 87(3), 355–374; Zahavi, A. (1975). Mate selection — a selection for a handicap. Journal of Theoretical Biology, 53(1), 205–214.

Yamaguchi, M., Smith, A., & Ohtsubo, Y. (2015). Commitment signals in friendship and romantic relationships. Evolution and Human Behavior, 36(6), 467–474.