Malleable Lies

Lying Behavior When the Stakes Are Large

In our “Malleable Lies” paper, we study whether we can predict behavior of contestants in the British TV game show Golden Balls, by carefully analyzing what they say.


In Golden Balls, the two finalists independently decide to either split or steal the jackpot. If both split, each gets half the jackpot. If one splits and the other steals, the one who steals gets the entire jackpot while the other gets nothing. If both steal, both get nothing.


Before the two finalists make their decision, they get a few minutes to talk to each other. Most contestants try to convince their counterpart that they intend to split the money with him/her.

This leads to very tense situations. The Golden Balls episode in which two contestants, Sarah and Stephen, faced a jackpot of £100,000 has become famous. It has been discussed on many blogs, for example on . One of their screenshots summarizes what happens quite succinctly:

In a previous paper, we studied the determinants of peoples cooperative choices in this game. Amonst others, we investigated the role of age, gender, the stakes, and reciprocity in cooperative choices. You can read a summary of that paper here .


In the present paper, we study what people say and whether what they say can predict what they do.

A large literature in both psychology and economics shows that people do not like to lie. There are various theoretical accounts for this: people may want to maintain a positive self-image , they may feel guilt if they lie, or they may be afraid to harm their reputation.


Much research studies lying in this binary way: lie or no lie, promise or no promise. In our past paper, we also took this approach. In line with lying aversion, we found that promises are strong predictors of behavior: people who promise to split are 31 percentage points more likely to split than those who do not.


However, in real-life free-form communication people have way more options at their disposal than “lying or not lying”, “promising or not promising”. Rather, they can deceive by omitting, obfuscating, or stretching the truth.

This leads to our malleability hypotheses:

  • Lies are less costly if they are malleable to ex-post interpretation as truth. Therefore:
    • Contestants who plan to steal will prefer making malleable statements over unmalleable statements.
    • People who plan to split have no reason to make such malleable statements.
  • As a result, stealers will reveal themselves by their use of malleable statements.


What are malleable statements? We say that statements can be made more malleable by making them more implicit or by making your own behavior conditional on that of the other person. This leads to the following 2-by-2 typology (with examples):


The prediction is that cooperation is:

  • highest for those that make explicit unconditional statements
  • intermediate for those that make statements that are either explicit or unconditional
  • low for those that make implicit conditional statements or no qualifying statements

This is exactly what we find!


Results are robust to how we deal with the fact that people often make multiple types of statements (e.g., only considering strongest statement, percentage of statements from a category). Differences are statistically and economically significant.

There is much evidence that people are conditionally cooperative: preferring to cooperate if their opponent does so as well. If they are, and if they realize that the opponent’s statements have predictive value, we would expect contestants to react to their opponents statements.

We find that they do not!


Decisions to either split or steal are not affected by the statements made by the opponent. This suggest that either (1) contestants are not conditionally cooperative, and/or (2) they are unable to extract the predictive power of their opponent’s words. Given the large literature on conditional cooperation, and the large literature on people’s inability to clearly distinguish between truth and lies, the second explanation seems more likely.


Overall, our paper shows that statements that carry an element of conditionality or implicitness are associated with a lower likelihood of cooperation. By doing so, it confirms that malleability is a good criterion for judging the credibility of cheap talk.

We applied our 2-by-2 malleability typology to the specific case of a 2 person Prisoner Dilemma setting in a Game Show, but it can straightforwardly be applied to classify statements in other settings, such as politics, business, or personal relations.

The British TV game show Golden Balls thus allowed Uyanga Turmunkh, Martijn van den Assem,and myself to examine lying behavior in a controlled environment with large sums of money at stake. The paper was published in Management Science in 2019. It is available on the Management Science website or in working paper format on SSRN .


Dennie van Dolder
Senior Lecturer in Economics