The Death of Behavioral Economics
thebehavioralscientist.com
thebehavioralscientist.com
- Priming is social psychology, not behavioral economics. Most behavioralists I know are openly scornful of priming studies, which as the author correctly notes often do not replicate. (Leading example of the kind of thing behavioralists think is BS: “prime people with words like ‘Florida’ or ‘retired’ and they will walk more slowly.”). That kind of study rarely if ever published in Econ journals.
- For his other claim that “Most behavioral interventions don’t replicate,” I would say that most behavioral economists don’t study or design interventions! Much of their work is experimental in nature and focuses on developing more realistic decision-theoretic mathematical models of choice which can be applied in economic modeling outside the lab.
What the author actually has in mind is not contemporary behavioral economics, but rather a subset of social psychology.
Those results used to receive reasonably wide media coverage and were taught to MBAs (generally not by the economists!), but starting around 2010 a bunch of very serious critiques of their statistical methods were raised which undermined confidence in the entire enterprise.
It was never supposed to become the replacing paradigm. It doesn't have its own theory of how individual behaviour (microeconomics) composes into system behaviour (macroeconomics). But it showed that just summing up individual behaviour of independently and identically distributed individuals doesn't give you the macro picture that should tell us something about the real world either. It at best can describe a simulated game world, which only occassionally predicts something about our world.
Behavioral economics is de facto applied psychology and as such you'd obviously expect it to have study replication problems. Of course it will never be fully exact science as it tries to explain an evolving dynamic system of nonuniform units. Does it render it useless as the author points out? - unlikely.
The field's most prominent findings—loss aversion, priming, ego depletion—fail to replicate. That shoud be publicized as widely as possible, lest the millions who have read and adopted behavioural economics concepts continue to use them to guide treatment, government policy, business practice, etc.
Perhaps yes, at least as a starting point until better evidence produces a replicable, actionable model.
[1] https://www.dukechronicle.com/article/2021/08/duke-universit...
August 10, 2020
I would imagine there are many such cases in social sciences, that are not really clear cut.
Also in general, people can not reliably the fallacies they are prone to, so it is a constant uphill struggle. Maybe immediately crying "fraud" is not always appropriate.
This line is a damning criticism in hindsight.
So, so far, one lousy, or lazy and rushed up paper publication.
Discrete models of intelligence (ie., AI) fail to capture continuous dynamical phenomena (eg., growth, body adaption, etc.) The equivocation is, "a discrete simulation of (spatio-temporally continuous) X possess the same properties as X".
fMRI "investigations" into the brain look at region blood perfusion (etc.) as a guide to "activation". Equivocations here abound: "activity in blood" is "activity in dynamical behaviour response", etc.
Social psychology and behavioural economics: humans have no hysteresis (ie., individuals have no personal histories) so their aggregate behaviour is context-insensitive and can be studied "as a pattern". The equivocation: "situational behaviour" is "context-free behaviour".
And so on.
It is obvious, from the outside, that these maximally-reductionist projects are doomed to fail. The target of study (humans) cannot be reduced to one small category of explanation which can then be generalised free from extremely simplifying assumptions.
What we have in each case is largely hyper-specialised academics, routinely with no great deal of self-awareness, taking some paradigm principle in their field and forcing all human behaviour into it.
What is disheartening to me is how mindnumbing an effect this has on the public. The current wave of AI hysteria has left many unable even to parse ordinary questions such as, "What do you think of what I'm wearing?"
..which according to an NLP-AI defender is an invitation to have 1trillion historical documents summaried into a response.
Rather than the /obvious/ invitation to a particular human to exercise, based on their personal history of aesthetic experiences, a judgment of taste. For them to tell me their opinion.
The whole world of interior human life is routinely obliterated in this hysteria and we are left thinking that humans merely "generate text" on the occasion that otehr humans "generate text".
This, and all of these fields, are mind-numbing reductions of human complexity into toy models which barely function on their own critera -- let alone as models of any natural phenomenon. This is all pseudoscience.
Humans may all have histories, they also are synchronized: they grow up together raised by the same LED screen whispering in their ears.
I m sure you re right it'd be hard to explain and simulate the human economic interactions with less energy than required to just run it with real humans, but you can probably make a lot of easy short sub models now that we are all that synchronized.
And look at me writing to you in good enough English from Hong Kong while I was raised in France. I really feel we can be reduced more to patterns simply because we have to adapt a lot more to a common denominator. I had to internalize a lot of american concepts just to be understood when I express myself in English and therefore must surely be more sensitive to American stimuli than my personal history from the countryside of Normandy would suggest.
> to explain and simulate the human economic interactions with less energy than required to just run it with real human
I think we have become mystified by the type of "engineerable reductionism" that very modern science has enabled. This religion of "engineering approximations" has taken the popular imagination captive, academics included, and has led to people being extreme reductionists.
As an anti-dote, I think we should consider problems which are physically impossible to compute; interactions which require infinite energy to measure; simulations which take the age of the universe to run.
We find problems of this kind everywhere and they out-number the solvable ones infinity-to-few. We cannot compute the shape of a table from the positions of its atoms (10^30 states) let alone human behaviour from god-knows-what (the state of some neurones, ad prices, or blood flow).
In the face of what cannot be known either in principal, or ever in practice, we should resist engineering solutions.
I dont think there can ever be a reductive science of human behaviour. We are in exactly the position almost all humans in history have been to matters such as chemistry: unable to make any pronouncements whatsoever.
I think in the case of "the hard problems" in irreduibly complex cases, we should wholesale reject the reductive scientific method. It can only lead to pseduoscience.
We should instead be extremely careful and locally theoretical empricists, as the anchinets. And say only, "over here, this happens", and "over there, that happens". And so on.
Any more is charlatanism, delusion -- and worse -- an invitation to the public to a superstitious pseudo-scientific religion.
I agree with this sentiment, for example, in the case of wholly electronic ballots.
But the behavioral economics of academic grants will continue to guffaw at the idea of denying monetary incentives laid on the altar of the technology godz.
OK