Rationality repost

I was going to comment on this until I realised that ‘Jim’ from the comments section of the link posted by John above, summed up what I was getting at better than I could. Here it is…..

‘The “problem” with game theory is not really rationality, but mathematical modeling. That is, what allows game theory to be consistent with so many possibilities is that it usually involves mathematical modeling without empirical discipline. Relaxing the rationality assumption would make things worse, not better; the models would be consistent with more things. Much of behavioral economics suffers from the same problem as traditional game theory in that it is consistent with many things. One has to distinguish between experiments and models. Experiments are valuable in identifying how people behave, but they are not a substitute for models, because without a model, it is difficult to identify general principles. Behavioral economics uses both tools. Hopefully, someday, experiments–or some other source of evidence–will sufficiently discipline models so that the new models themselves will be more predictive than traditional game theory, but that day has not yet come.’ (italics added).

I might add that whereas it may (or may not) be true that the Historical School were devoid of ‘theoretical insights’ this is better than theories that are devoid of empirical predictions and facts.

Quine perhaps put it best (from my point of view)

‘As an empiricist I continue to think of the conceptual scheme of science as a tool, ultimately, for predicting future experience in the light of past experience . Physical objects are conceptually imported into the situation as convenient intermediaries — not by definition in terms of experience, but simply as irreducible posits comparable, epistemologically, to the gods of Homer’. (italics added)

The various theoretical assumptions of economics, it seems to me, have no ‘better’ ontological status than the Gods of Homer (actually they have less because at least some people claim to have seen Zeus and Aphrodite, but no one has ever claimed to have seen the various postulates of economics).

BUT.

In physics these postulates actually lead us to make real world predictions that enable us to like, you know……do stuff. Trivial stuff, admittedly, like putting a man on the moon, but still.

What, precisely, do the ‘laws’ of economics (and game theory) actually predict , apart from trivialities, phenomena that could easily be explained some other way, and things that didn’t happen?

Re brendan’s post – good points, but I think that this is a widely-used but misleadingly idealized picture of how clean physics is.

There is a relatively narrow area even within physics where theory makes clear, verifiable predictions and that is the area that many focus on when picking on poor old game theory. Sure, QED can predict the Lamb shift to a gazillion decimal places, but take a look at areas like solid state physics or theoretical chemistry to name but two and you see a different picture. To give just a couple of examples:

– remember the cold fusion fiasco? there were several papers by very smart people giving theoretical explanations of how this could happen.

– Roald Hoffmann, a Nobel prize-winning theoretical chemist, gave a theoretical explanation in the late ’80s sometime for a phenomenon called “bond-length isomerism” that later turned out to be a misinterpretation of X-Ray crystallographic data.

I don’t want to pick on Roald Hoffmann here – his on-base percentage is high – but complex systems where unambiguous prediction is difficult start a long way before you get to the even more messy world of economics.

Just using rationality in a value-neutral, purely descriptive way would greatly reduce the harm the word does. “Rational” is normally a buzz-word — science is rational, scientists are rational, some historical actors are rational (or at least, much more so than others), and it is assumed that if everyone were more rational, the world would be a better place.

It isn’t hard to find cases when rationality is harmful, either because it’s predatory or because the assignment of costs and benefits leaves out important considerations. Rationality is a way of choosing actions, but also a way of framing justifications or proposals, and if rationality is too high on your queue of justifications, you will probably end up justifying harmful behavior in cases in which one aspect of a situation has been adequately formulated, whereas spokespersons concerned with other aspects of the situation are still speaking “pre-rationally”. External costs is an example; the human cost of economic progress is another.

It has been noted that certain sorts of sociopaths are fully rational. This is especially so if you lump traditional moral taboos in the “emotional” category, so that the taboo on torturing children is on the same formal footing as the taboo on eating shrimp.

Your post seems mostly directed at “rational” as a predictive explanatory category, as economists extend their sway outside pure market forms, for example, the ludicrous attempts to portray the family as a system of exchange where mother, father, and child are all behaving rationally in order to get what they want. This involves hypothecating imaginary “child-services” and assigning them imaginary numerical values.

An odd feature of this is that evpsych people explain that even our most irrational gut feelings are rational from the point of view of genes. This is another case of expanding the concept of rationality beyond its possible meaningful uses.

With rationality, we often emough we find ourselves in angels- dancing- on- the- head- of- a- pin territory, where all the implications of a concept are made explicit, no matter how ridiculous they might be.

Many attempts to resolve policy questions are grounded on hypothetical, schematic rational choices, often using on unquantifiable values, which are then generalized to reality. The “ticking bomb in a schoolhouse” justification for torture is one such. (Taking the rationality calculation seriously, I concluded that torture would be justified if the ticking bomb were threatening a class of innocent pre-schoolers, but not if it were merely threatening three or four annoying, defiant preteens. But obviously this was not a rational calculation at all; a bomb in a preschool was chosen by those who wanted to justify torture precisely because it had the maximum irrational emotional power.)

Note: I think (I think ) what I am getting at is very close to what has been stated by the so-called Post-Autistic Economics school . But I am willing to stand corrected on this matter, especially by someone who actually knows what they are talking about.

In terms of the thread of this post: ‘rationality’ like ‘irrationality’ is not some metaphysical distinction that descended from on high, but is a real world classification used by specific people in specific situations. Therefore the real issue is: can it be used consistently by people across situations? And for this to be the case there need to be public, agreed upon criteria for its application. These may of course be culturally specific (suicide bombing strikes ‘us’ as being irrational, but not presumably the bombers themselves. Ipso facto, the invasion of Iraq might well strike ‘them’ as being irrational but not ‘us’), but that in itself calls into question the universality of the use of the word.

In other words, we need public definitions: an act shall be defined as ‘rational’ if it meets criteria x, y, and z. And ‘we’ (i.e. the people using the word) have to agree on this.

This difference in viewpoints is key. As John points out, there are certain actions that may be ‘irrational’ from ‘my’ point of view, but not irrational from the point of view of the gene. And, presumably, vice versa.

Like all these categorisations, I think they are harmless as long as universalist or even metaphysical claims are not also smuggled in. Unfortunately they normally are.

I got interested in the concept of rationality by way of psychiatry, where its roughly synonymous with sanity. That of course gets us nowhere, since we don’t know what sanity is. In the event, the central idea seems to be the kind of thinking that views a propostion in context. Rather like Witgenstein coming to the conclusion that propositions were uttered by human beings.
The notion of a rational economic man -a disembodied calculator of advantage and gain – has struck me as about as reliable as the labor theory of value.

The one that is most obviously reasonable, though not invariably compelling, is transitivity. If I prefer A to B, and B to C, I should prefer A to C. Others imply preferring more of a good to less (of course, the term ‘good’ is part of a circular definition). Broadly speaking, game theory requires rationality in the form of consistency, but this is often tangled up with assumptions about self-interest.

Interesting, because in my (not-statistically-valid) empirical experienc of life most human beings have a wide variety of situations where their preferences are NOT transitive. This is very easy to demonstrate with pairwise selection of flavours of ice cream. That would seem to be a bit of a problem for these mathematical models.Cranky

Given any data on any observed set of problems involving the selection of one or more choices from a set of alternatives, the observed choices can be represented as the maximisation of an appropriately specified function.

Yeah, I suppose that’s true as far as it goes, but this not very helpful because for many people this ‘appropriately specified function’ is defined as ‘I won’t be bothered with maximizing any freaken functions’.

When I am buying a car I will try to maximize the function, but if I’m buying a DVD player, then I’ll just walk into the centre commercial downstairs and buy whichever one is on sale today.

I would argue that for any particular slice of the population (defined by income and demographics) there is indeed a strong pattern of choices they make – and this empirical pattern can be termed ‘rational behavior’ for this particular segment.

The more precisely and narrow this slice is defined, the more clear you’ll be able to see what the empirical ‘rational behavior’ is. If you could collect information about people who bought brand-new Chrysler minivans in 2000, you would’ve probably found out that their socio-economic situations were very similar.

Brendan: “To summarise: your point only works if we were all agreed about ends and disagreed merely on means.”

Ah, no. It works provided what you’re interested in is means. That’s still an interesting question even if we don’t agree on ends – because sometimes we actually have to make decisions. Just because game theory isn’t a magical theory of everything, doesn’t mean it’s worthless.

Brendan, I really don’t see it that way. Game theoretic models ask the question: if you have a particular set of strategies at your disposal in a particular situation, what ought (“normative alert”) you do? If, faced with a sufficiently similar situation in the real world, people behave differently, there are several possibilities: they don’t see the choice space as you see it, they don’t have the same objective you do, or they are taking an action which, had they but known how to behave, would have done them better by their own lights. The first two are standard problems with all models, namely, that they are only approximations to reality. In this, economics is no different than physics, except that their models cover an unchanging underlying reality which allows more precision than the messiness we poor social scientists are left to observe.

In the third objection, though, game theory actually has something to teach us. My pithy annalysis would come from an actual game (although in this case a game against nature) blackjack. A “game-theoretic” analysis of blackjack tells you how to play the game. That can be useful even if few players play that way. The goal is certainly not a predictive model of how actual players play blackjack. Such a model might be interesting, but for very different reasons.

I’d say, think twice about using Bayes’ theorem in any case where you don’t have a well defined chance set up . You don’t always have objectively based priors, and when you don’t some of us are rather skeptical of using Bayes’ theorem.

Mathematical Claim, then “If this claim is accepted, it’s evident that the definition of rational choices as those that maximise an objective function is empty, since all choices satisfy this criterion.”

Well. No.

1. Even if the definition given above is circular – in the sense that all behavior is rational by definition – it is not necessarily empty. As social scientists we still might want to know what exactly the function being maximized consists of. If I see you doing something, which in layman’s terms could be called ‘irrational’, I still might want to understand the reason behind the madness. I wouldn’t call you irrational or nothin’ but, I would still be interested in the peculiar kind of a utility function, the preferences, the motives, that generate a particular type of behavior.

2. Maximization itself implies an existence of a structure. Generally speaking, well ordered preferences (transitive and complete) over a given choice set. Hence to maximize something, one has to assume consistency and comparability of alternatives in the first place – the (5) above. So the standard definition of rational preferences in economics is actually quite parsimonious, at least relative to how other disciplines define rationality. In absolute terms it is not, because:

3. The “rational” in economics refers not to people but to their preferences. Preferences are rational, people … well, who knows. Which is not a trivial distinction. I might have transitive and complete preferences but that does not automatically imply I “act” on them. I think Conchis above is making the same point. And,
4. Your Mathematical Claim smacks of the whole ‘Revealed Preference Approach’, without the revealing part. Not that there’s anything wrong with that. What’s wrong with that is to equivocate transitivity with consistency within that approach or in the context of the above definition. Consistency can only be determined if you have more than one observation since it necessarily involves comparing stuff. If you want to judge only a single observed choice, welp, then it’s gonna be consistent by definition since all you can go by is that choice itself. So in a way, there’s an intertemporal aspect here – if you see me doing one thing and then another how can you know whether I’m an irrational type of guy or whether my preferences have changed in the meantime, while still being complete and transitive. That’s the basic shortcoming of both the axiomatic approach to “rationality” and the mathematical claim above. Both are static. That’s what motivated Samuelson, though it became pretty obvious that even if you paribus all the ceteris you still can’t squeeze much out if it.

I guess these last two points agree with the spirit of the post, although for completely different reasons. And what Game Theory has to do with any of it beats the hell outta me.

Quick question: is there any difference between “game theory” and “rational choice theory,” or are they just different names for the same ideas? Or, rather, are they theories that overlap in important ways, but cannot accurately be called the same thing?

“Game theory is reasonably well-defined and…can be regarded as a subset of rational choice theory”

This is the essence of my disagreement. GT is a tool, a method. It is not a theory in the same sense that RTC is a theory – a theory of human behavior. GT is a theory in a mathematical sense in the same way as say, Functional Analysis is a theory. So to the fact that ‘rationality’ gets used in confusing and ambigous ways, you gotta add the confusing and ambigous ways in which the word ‘theory’ gets used. This is why I’m objecting to conflating the two. You wanna argue about the validity/usefulness of the assumption of rationality in economics (or Social Science in general) that’s one thing. You wanna argue about the validity/usefulness of game theoretic methods in economics (and beyond) that’s another. There’s hardly any value added to the discussion by slapping these two topics together.

I don’t think there’s anything intrinsic about game theory that implies rationality. Note that your discussion about rationality above has nothing to do with game theory per se, it’s just sort of appended ad hoc to a sentence that mentions GT. Usually the strong rationality (I’d rather call it ‘sophistication’, as in the case with SPNE) that you’re objecting to is added in, but that’s because of the standard problem that … well, when you get rid of rationality you don’t have much to replace it with. This is true of most economic models, whatever the tools you’re using.

Another counter example. Games which have both ‘rational’ and ‘crazy’ or ‘random acting’ agents. The rationality assumption lies in the equilibrium concept used (or not), not in the game itself.

Also, the basic problem in modeling human behavior is that you want your model to generate non-trivial sharp predictions which have a hope of being tested against real life data (contrary to what some posters assert above). That’s not easy. Rationality by itself in fact doesn’t usually deliver, partly because in economics at least it’s a much weaker concept then most people think. So how do you get these non-trivial sharp predictions? You gotta throw in something else. In non-game-theoretic models the assumption of price taking behavior (which in fact in many cases is not strictly speaking “rational”) usually gets you there. In game theoretic models you need some kind of sophistication on the part of the agents – have them think about how the other player is going to act, whether certain strategies are credible or not etc. Or you assume that players who play strategies which result in lower payoffs get eliminated over time – evolutionary approach. Either way, you got to have something there to make it work which goes beyond rationality.

The length of my posts, I think, reflects the fact that I’m a wordy kind of guy, but also that there’s too many issues here being discussed at once, btw.

Reading all of these comments just seems to make JQ’s point over and over again. The very investment some commenters have in the word and their legitimacy in using it in the way(s) they feel useful seems to me to confirm his observation that it is simultaneously empty and dangerous.

Yea, well, the trick is to specify the function appropriately.