If human action could not be predicted, the results of experiments into the Allais Paradox, preference reversal, conjunction fallacy, and on and on, should be a random walk. Since they’re not, human action can be to some measure predicted. If an Austrian believes all such research should be discounted by 100%, I’m taking issue with whatever prior gives that result.
And no, using priors in the way I described is not circular reasoning. Recall Bayes’ Rule for updating on evidence: P(H|E) = P(H)P(E|H)/P(E)
To be as clear as possible, let’s use the following example. We want to test the hypothesis “People act rationally and self-interested.” As a definition for rationality, let’s say people’s STABLE preferences disallow intransitivity.
Say I start with a prior for the hypothesis, P(H)=0.9. The likelihood that we see experimental evidence of intransitive preferences given this hypothesis must be fairly low, but there could always be experimental error, so P(E|H)=0.05. This estimate is where my priors come in as I described above. If I think it’s equally likely for an experiment to show evidence of intransitivity as transitivity, even given my hypothesis, P(E|H)=0.5.
I discount by my estimate that there will be experimental evidence of intransitive preferences regardless. P(E) = P(H1)P(E|H1) + P(H2)P(E|H2). Given P(Intransitivity results | People are rational) = 0.05, P(Intransitivity | People are irrational) = 0.95, we have, for the case of the believer P(E) = 0.05 0.9+0.1 0.95 = 0.14 and, for the case of the skeptic, 0.5 0.9+0.1 0.5 = 0.5. So, for the believer, evidence of intransitivity gives 0.90.05/0.14 = 0.32, and for the case of the skeptic, 0.9 0.5/0.5 = 0.9, ie, no updating.
Mises argues that we can never have any evidence of intransitive preferences because preferences are not stable. Thus, the preference reversal evident in choosing Gamble 1 in Part 1 and Gamble 2 in Part 2 of the Allais Paradox can never be evidence of intransitive preferences. But, I argue that if we show, in study after study, across the majority of people, that the preference for Gamble 1A and Gamble 2B is stable over time—seconds, weeks, months, years, lifetimes even!--that we should discount the skeptic argument P(Intransitivity results | People are rational) from 0.5 to something lower, akin to P(E|H) = 0.05.
But that’s not where it begins. I’m saying that experimental evidence of such preference stability should change your probability estimate of P(Preferences are stable) from 0.5 (This variable is mystical, completely unknown, sublime and unknowable even to a superintelligent AI with the capability of doing a nanosweep of your entire noggin) to something much higher, like 0.9 (I am pretty damn sure this preference is stable because the evidence says so and evolutionary psychology suggests it’s universal). Even if you want to leave it highly unknown, P=0.51, this will change your update according to the evidence. So it’s not circular reasoning. It’s using priors/updates on one hypothesis (preferences are stable) to update on another hypothesis (people are sometimes irrational).
If you’re arguing that we should remain radically uncertain even in the face of such evidence, I want to know the priors you assign. Saying “it’s unknown” isn’t enough. How unknown is it? I have trouble believing it’s really a 50⁄50 split. Are we really equally likely to see most people choose Gamble1A and Gamble 2B in every experimental study with highly statistically significant results across times and cultures as we are to see a random walk? If so, how come we never see random walks?
Human action can indeed be to some measure predicted.
For instance, if I conducted an experiment with 100 people wherein I presented each person with the opportunity to place their bare hand on a red hot burner on a stove, including leaving it there for one minute, I predict 100% would say no. I could even model that experiment.
However, this kind and degree of predictability is meaningless in the context of economic modeling.
What if the person who is being presented with the choice in the Allais Paradox just lost their mother to death, as well as losing their job in the same week? How does this affect the model? What does that research show?
How does one account for decisions made without adequate consideration, or when the decision maker doesn’t understand the problem? What about the follow on effects of choices made in the past which encumber via contract, or cause emotional or financial pain, such that the decision is not rational or the risk assessment is distorted? Or the reverse when the rewards have been great in the past?
How many life choices exist in such pristine, simple and clear conditions as the Allais Paradox?
Are not our choices, responsibilities, assets, liabilities, obligations, future earnings, job markets, work relationships, preferences, skills, talents, capital, regulatory environments, choices of other people, comparative advantages, currency fluctuation, taxation, inflation, religious beliefs, IQ, education, weather, genetics, resource allocation, scarcity, social stability, time constraints, competing demands, influence of peers, influence of media, family relationships, beliefs about the future, and more, all knit into each decision made?
Are you really claiming that the minor complexities presented in such a simple model as the Allais Paradox rise to the level of mathematically illuminating, for the purpose of useful economic modeling, the myriad decisions inherent in daily life? After all, everything in life depends, at some significant level, on exchange of productivity, which is generated as a result of the decisions of life.
My point is that the models relating to human action which are herein employed as proofs, are not sufficiently complex to be useful or meaningful in economic modeling.
You criticize the Austrian school on the basis of presuppositions which are designed to note the limits of our ability to construct theories or predict future events. At the same time, all that is offered to suggest we are not limited are simplistic and wholly inadequate models which do nothing to solve the problem. As long as Mises claims we can’t know or test these things and no one else shows that we can, I have to agree with Mises.
Besides, if these things were knowable, Mises would never have accepted stopping at this level. He would have anticipated and likely discovered and modeled the information so as to press another layer deeper, in hopes of gaining a greater mastery of the subject.
If human action could not be predicted, the results of experiments into the Allais Paradox, preference reversal, conjunction fallacy, and on and on, should be a random walk. Since they’re not, human action can be to some measure predicted. If an Austrian believes all such research should be discounted by 100%, I’m taking issue with whatever prior gives that result.
And no, using priors in the way I described is not circular reasoning. Recall Bayes’ Rule for updating on evidence: P(H|E) = P(H)P(E|H)/P(E)
To be as clear as possible, let’s use the following example. We want to test the hypothesis “People act rationally and self-interested.” As a definition for rationality, let’s say people’s STABLE preferences disallow intransitivity.
Say I start with a prior for the hypothesis, P(H)=0.9. The likelihood that we see experimental evidence of intransitive preferences given this hypothesis must be fairly low, but there could always be experimental error, so P(E|H)=0.05. This estimate is where my priors come in as I described above. If I think it’s equally likely for an experiment to show evidence of intransitivity as transitivity, even given my hypothesis, P(E|H)=0.5.
I discount by my estimate that there will be experimental evidence of intransitive preferences regardless. P(E) = P(H1)P(E|H1) + P(H2)P(E|H2). Given P(Intransitivity results | People are rational) = 0.05, P(Intransitivity | People are irrational) = 0.95, we have, for the case of the believer P(E) = 0.05 0.9+0.1 0.95 = 0.14 and, for the case of the skeptic, 0.5 0.9+0.1 0.5 = 0.5. So, for the believer, evidence of intransitivity gives 0.90.05/0.14 = 0.32, and for the case of the skeptic, 0.9 0.5/0.5 = 0.9, ie, no updating.
Mises argues that we can never have any evidence of intransitive preferences because preferences are not stable. Thus, the preference reversal evident in choosing Gamble 1 in Part 1 and Gamble 2 in Part 2 of the Allais Paradox can never be evidence of intransitive preferences. But, I argue that if we show, in study after study, across the majority of people, that the preference for Gamble 1A and Gamble 2B is stable over time—seconds, weeks, months, years, lifetimes even!--that we should discount the skeptic argument P(Intransitivity results | People are rational) from 0.5 to something lower, akin to P(E|H) = 0.05.
But that’s not where it begins. I’m saying that experimental evidence of such preference stability should change your probability estimate of P(Preferences are stable) from 0.5 (This variable is mystical, completely unknown, sublime and unknowable even to a superintelligent AI with the capability of doing a nanosweep of your entire noggin) to something much higher, like 0.9 (I am pretty damn sure this preference is stable because the evidence says so and evolutionary psychology suggests it’s universal). Even if you want to leave it highly unknown, P=0.51, this will change your update according to the evidence. So it’s not circular reasoning. It’s using priors/updates on one hypothesis (preferences are stable) to update on another hypothesis (people are sometimes irrational).
If you’re arguing that we should remain radically uncertain even in the face of such evidence, I want to know the priors you assign. Saying “it’s unknown” isn’t enough. How unknown is it? I have trouble believing it’s really a 50⁄50 split. Are we really equally likely to see most people choose Gamble1A and Gamble 2B in every experimental study with highly statistically significant results across times and cultures as we are to see a random walk? If so, how come we never see random walks?
Human action can indeed be to some measure predicted.
For instance, if I conducted an experiment with 100 people wherein I presented each person with the opportunity to place their bare hand on a red hot burner on a stove, including leaving it there for one minute, I predict 100% would say no. I could even model that experiment.
However, this kind and degree of predictability is meaningless in the context of economic modeling.
What if the person who is being presented with the choice in the Allais Paradox just lost their mother to death, as well as losing their job in the same week? How does this affect the model? What does that research show?
How does one account for decisions made without adequate consideration, or when the decision maker doesn’t understand the problem? What about the follow on effects of choices made in the past which encumber via contract, or cause emotional or financial pain, such that the decision is not rational or the risk assessment is distorted? Or the reverse when the rewards have been great in the past?
How many life choices exist in such pristine, simple and clear conditions as the Allais Paradox?
Are not our choices, responsibilities, assets, liabilities, obligations, future earnings, job markets, work relationships, preferences, skills, talents, capital, regulatory environments, choices of other people, comparative advantages, currency fluctuation, taxation, inflation, religious beliefs, IQ, education, weather, genetics, resource allocation, scarcity, social stability, time constraints, competing demands, influence of peers, influence of media, family relationships, beliefs about the future, and more, all knit into each decision made?
Are you really claiming that the minor complexities presented in such a simple model as the Allais Paradox rise to the level of mathematically illuminating, for the purpose of useful economic modeling, the myriad decisions inherent in daily life? After all, everything in life depends, at some significant level, on exchange of productivity, which is generated as a result of the decisions of life.
My point is that the models relating to human action which are herein employed as proofs, are not sufficiently complex to be useful or meaningful in economic modeling.
You criticize the Austrian school on the basis of presuppositions which are designed to note the limits of our ability to construct theories or predict future events. At the same time, all that is offered to suggest we are not limited are simplistic and wholly inadequate models which do nothing to solve the problem. As long as Mises claims we can’t know or test these things and no one else shows that we can, I have to agree with Mises.
Besides, if these things were knowable, Mises would never have accepted stopping at this level. He would have anticipated and likely discovered and modeled the information so as to press another layer deeper, in hopes of gaining a greater mastery of the subject.