Friday, September 7, 2018

Turning a Weakness within Game Theory into a Strength: A Review of 'Testing' Game Theory by Daniel Hausman (2005)


         In Hausman’s article on testing game theory, the author identifies the primary difficulty in empirically testing the axioms of the analytic technique. In order to conduct an experiment on game theory a scientist must impose experimental conditions prior to testing. The scientist determines who the players are, what strategies are available to the players, and the payoffs for each strategy. For example, an economist may present a game theory test in which the players are two business professionals choosing a strategy based on monetary payoffs (make money or lose money). On the other hand, an evolutionary biologist may present a game theory scenario in which the players are two animals choosing a strategy based on evolutionary payoffs (survival or death). After imposing these conditions, the emergent test of game theory is what Hausman describes as a “Human Premeditated Game Theory” (HP game theory). Since every discipline produces a different game theory test, Hausman argues that the economist and evolutionary biologist are conducting their own tests of HP game theory and not tests of game theory itself. 
         Since pre-determined conditions are necessary to apply a test of game theory and since those same conditions inherently turn a test of game theory into a test of HP game theory, there is no way to test game theory itself. Thus, Hausman concludes, attempting to test the axioms of game theory is a futile effort. Instead, Hausman emphasizes the value of using game theory to test claims about people’s preferences.
         Hausman explains the varying levels of control a scientist has in determining conditions for a test of game theory. At the lowest level of difficulty is determining how many players there are and what strategies are available to those players. With slightly greater difficulty, scientists can control the beliefs players have concerning permissible strategies, the physical outcomes of strategy combinations, and the knowledge available to other players (including their beliefs about the beliefs of each other). The greatest difficulties for the scientist are determining the payoffs and the preferences of the participants. Determining the payoffs in a game theory experiment are critical because they are used "to make any substantial predictions concerning laboratory behavior". Furthermore, the payoffs are built on the assumption that the participant’s preference is to choose the strategy that offers the most favorable payoff, or the dominant strategy. 
      Yet when applied in experimental tests of game theory, there are times when individual preferences lead participants to choose the non-dominant strategy. Rather than consider this anomaly as disconfirming the effectiveness of game theory in predicting behavior, Hausman suggests that scientists should use game theoretic anomalies to study the factors influencing preferences. In an economist’s HP game theory, the economist assumes that the participant will prefer to maximize the monetary payoff by choosing the dominant strategy. However, other unknown factors (what Hausman calls preferences) may influence the participants to choose the non-dominant strategy with the lesser monetary payoff. For example, participants may care about the monetary payoffs other participants receive, what other participants trust them to do, whether the outcome is fair, etc. Hausman concludes that the true value in applying game theory is this opportunity to understand how those personal preferences are affecting decision-making.


     The seemingly fatal flaw of applying game theory, in my opinion, is the assumption that the participant will choose to maximize personal payoff with the dominant strategy. As analysts we are taught to identify our assumptions and reconsider how those assumptions are affecting our analysis. Here, Hausman identifies the assumption that is misleading scientists and leaving them surprised when participants choose the non-dominant strategy.
   Although this assumption is a weakness, I agree with Hausman that we can turn this weakness into a strength by using game theory to understand how personal preferences affect a participant’s decision regarding the payoffs. An opportunity to expand on empirical tests of game theory is to conduct a survey of the participants alongside the experiment. These surveys should attempt to reveal the preferences, or personality characteristics, of the target population. How important are the values of fairness, generosity, or selfishness to these participants? With a large enough sample, this survey/game theory experiment can reveal two things: the target population’s beliefs regarding these factors and what factors tend carry the most influence in decision-making for very specific, HP game theory situations.
Although this combined experiment could provide a more wholistic understanding of the target population’s preferences and the decision-making process for participants, challenges remain because a scientist is limiting participants to two strategies and payoffs to only four mutual outcomes. Thus, it will be difficult to generalize results from a wide-scale study of this kind because real-world “HP game theory” situations will typically present decision-makers with more possible strategies and far more potential outcomes.

Source: Hausman, D. M., (2005). ‘Testing’ game theory. Journal of Economic Methodology. 12(2), 211-223.


  1. Tom,

    It seems that with HP game theory, a different definition of "rational" actor would be needed. If individuals are not making the most personally beneficial choice based on predetermined outcomes, can they be considered rational within the classic game theory model? Would consistency in choice based on their preferences be a better metric?

    1. William,
      You are correct in that Hausman is in essence proposing a new definition for what is "rational". In the classic model these anomalies would suggest that those individuals are in fact "irrational". However, as Hausman digs further he finds that their decisions were justifiable and rational. The differences in anomalies lies in what preferences were more influential to the individual: ideas of fairness and being amicable or ideas of being self-serving. Thus, consistency in choice based on preferences would in fact be a better metric in defining a "rational" decision. However, the gray area that emerges with this metric for a rational decision is that it strays from the idea of an objectively rational decision (which game theory attempted to establish); as now being rational is considered 'to each their own'.

  2. It seems like Hausman argues that we need to know more about how people makes decisions because even in a situation such as a prisoners dilemma, they will make a decision consistent with past decision-making despite it being in their best or worst interest. What are your thoughts on this Tom?

    1. Harry,
      You identified an important argument that Hausman wanted to discuss: although we expect decisions to be based solely on potential outcomes, in fact personal preferences are just as influential (if not more) in the decision-making process. Thus, Hausman suggests that future research should attempt to understand decisions at the individual-level: how do the outcomes and personal preferences interact with each other to produce the decision? This is a challenging research question and future game theory research can build on this opportunity to better understand the decision-making process by examining the role of personal preferences closer.

  3. I think Hausman argues that we can benefit from understanding why someone picks the non- dominant strategy. However, what classifies the strategy as dominant or non- dominant. Its not always a black and white situation. Yes one strategy may maximize profits, however, the other strategy may still increase profits, draw in investors and drive up sales. Even though the other strategy is maximizing profits more, the other payouts might add up to more in the long run or the other person values the other payouts more. So is one strategy actually dominant? Or how do we standardize the definition of "dominant strategy"?

  4. Alyssa,
    In the classical model that Hausman is discussing, the "dominant strategy" is that which is rational, or maximizes the personal payoff. But as you mention, the other non-dominant strategy may be equally if not more beneficial in more ways like building trust and cooperation with the other players. If the other players can trust you to make the mutually beneficial decision with them, this can increase profits in the long run for all participants. That is exactly why Hausman challenges the current definition a of rational decision and the idea of a dominant strategy. It is difficult to define a dominant strategy due to the influence of personal preferences and the presence of anomalies. Thus, Hausman argues that the important thing to learn is not what constitutes a dominant strategy but rather understanding the why behind the decision of a traditionally dominant or non-dominant strategy.

  5. Your article has me wondering if game theory can be modified to allow for certain participants to act in their supposed "irrational" self interest (ie, for respect, community harmony, or benevolence) rather than proper self interest. Can game theory be applied when two players have different payoff desires?