Thanks to Sarah Weiler for prompting me to write this post
Rational choice models are a way of thinking about the social world that models individuals as rational agents and seeks to explain social phenomena as the result of the interaction between these different rational agents. Games like the prisoner's dilemma are the most well known example of this type of thinking about the social world. It would be crazy to think that this type of modelling is the only thing you need to think well about the social world. My claim is that this type of thinking about the social world is often a very good framework to try to understand the social world.
What are rational choice models?
Part of the confusion that I think often permeates discussions of rational choice models is that the word ‘rational’ here means something different to what it means in ordinary language or philosophy. Rational here is a set of assumptions about individuals preferences. Rational choice theories assume that individuals' preferences follow two axioms: firstly that they are complete and secondly that they aren’t transitive. Completeness here means that individuals have preferences over all states of the world - they either prefer one to the other or are indifferent between them but they are not unrankable. An example where this wouldn’t be the case is where there are two states of the world that individuals are choosing between that are just incomparable. For instance, it’s not obvious that one’s individual happiness and the ability to fulfil one’s moral responsibilities are even the sorts of things that can be compared. Transitive means that individual preferences can’t make them go round in a circle. For instance, if I prefer A to B and B to C I’m not allowed to then prefer C to A.
A second point of confusion is what the word ‘preference’ means here. Again, it means something different to why the word is used in ordinary language. Preference here is purely about observable behaviour rather than being located in the mind. When individuals make choices about states of the world we can infer from this a ‘preference relation’ - the thing that has to obey the axioms of completeness and transitivity. For instance if I’m choosing between two political parties you can infer a preference relation from my choice which may be very different from the preference relation that would correspond to what was good for my utility in the philosophical sense but may still have predictive power.
What unifies the axioms of completeness and transitivity is that they preclude behaviour that is erratic. For instance, if when choosing between apples, oranges and pears my choices varied wildly while everything else stayed the same - for instance if crunch of an apple was fundamentally incomparable sweetness of an orange - then the axioms of rationality would be violated and I couldn’t use rational choice theory to model this part of the social world. But I don’t think that this is what we see. In general it seems like individuals don’t display this pretty weird erratic behaviour and so I don’t think objections to these axioms are particularly strong reasons to reject rational choice modelling.
Rational choice models integrate other approaches
The choice between rational choice modelling and other ways of thinking about the social world - for instance ways of thinking that centre norms or psychology - are not sharply at odds with one another. The bare bones of rational choice models that I described about - the preference relation - is essentially useless by itself. To forge it into something that we can use to understand social phenomena we need to explain what individuals have preferences over and why. We can infer this from data - individuals make choices and, unless their behaviour is pretty strange in a way that violates our key axioms - we’ll know what their preference relation is (although this becomes much harder if we have very little data and a large number of possible choices.) In this way rational choice theory in the way it’s practised in economics abstracts away from human preferences in the same way that biology abstracts away from chemistry. Biology takes certain chemical principles as fixed and then tries to understand how life works given those chemical principles. Similarly, rational choice theories shift the onus from human preferences onto decisions given those preferences.
Sometimes it seems clear most of what we don’t understand about social phenomena comes from not understanding how individuals' preferences interact rather than why they have the preferences they do. This is clearest in economics, the field where rational choice models were first developed and are today most embraced. When we have a recession the thing that’s changed isn’t people’s preferences over money, leisure and how much they like work. Why we have the preferences we do for these things are interesting questions in themselves, but the 2008 recession didn’t happen because suddenly everyone decided that they didn’t like money.
Sometimes the interactions of individuals given their preferences is clearly not the variable that explains an important social phenomena. Why is it plausible that India and Pakistan will engage in a full scale land war while this is not the case for the UK and France. The material situations of the two states are very similar. All four are second tier powers and all are nuclear armed. The variable that explains much more of their animosity seems to be the history between the two countries and the preferences this has led their populations and political leaders to have.
Note in neither of these examples is does a rational choice approach preclude a sociological view nor vice-versa. It’s that the social phenomena that needs explaining looks more conducive to examination by one mode of thinking rather than another. This can be extended to the purported challenge to rational choice models made by psychology.
Classical game theory makes much stronger assumptions than rational choice does as a framework. It assumes that individuals are good at finding what the best strategy for them to take is, or at least that the game has occurred often enough that the best strategy has been learnt or selected for, and also that when there’s an element of chance involved individuals will make decisions according to expected utility theory.
Expected utility is the idea that when faced with a choice that leads to different states of the world, probabilistically individuals choose the option that leads to the greatest sum of utility in the different worlds, weighted by the probability that those worlds come about. Crucially, this is not the same as making decisions according to expected value. For instance, it may be that a risky investment make me more money on average than a safe one, but expected utility theory still predicts I buy the safer investment because I’m hurt by losing a large amount of money so much more than I’m hurt by losing a smaller amount, while when I win lots of money by life improves by only a bit more than when I win a smaller amount.
Work psychology has shown pretty conclusively that these are often, although by no means always, bad assumptions to be making. For instance, rather than making decisions according to expected value theory, people often make decisions under uncertainty according to prospect theory. In this model individuals are loss averse according to a fixed reference point. There are a number of other ways that psychologists have found in which individuals deviate from the more stringent assumptions of classical game theory but these can still be integrated into a rational choice model and can still be modelled using mathematics as is the case under classical game theory.
Another way in which a sociological and psychological approach challenges the classic assumptions of game theory in a way that can still be integrated into a rational choice framework is by giving people preferences over strategies. In the classic prisoners dilemma game, in which there are two prisoners who are offered deals to give information on the other in exchange for a reduced sentence, classic game theory says that both prisoners take the deal leading to a worse outcome for both. However, this assumes that neither of the players have preferences over the strategies that they use. For instance, one or other could think that giving up information is bad in and of itself, leading to their cooperation despite their still having the preference not to go to prison for more years nor caring at all about their accomplice. Like the psychological challenge before it, this sociological challenge can still be incorporated into a mathematical rational choice theory, while still challenging the standard game theory assumptions.
Why I like rational choice theory
So far I’ve tried to show that often when people are arguing against rational choice theory they’re thinking it makes much stronger claims than it is in fact making. In this section I’ll argue why I think that trying to explain social phenomena by taking preferences as given is often a good way to approach the study of the social world.
The first attractive property of rational choice models where the focus is on choices given preferences, is they’re often extremely good at doing science with. The property of being mathematical forces one to make extremely explicit the assumptions that one is making and allows one to derive logically the conclusions that follow from those assumptions. The scientific advantage of this is that rational choice models lead to extremely precise, extremely falsifiable predictions. In contrast, it’s often unclear exactly what sociological models are predicting and exactly which assumptions lead to the prediction the theory is making. This makes them both harder to falsify, harder to derive predictions from and harder outsiders to understand and check. It’s also often very easy to apply rational choice models to situations outside of the evidence that informed the development of the model because the model mechanically outputs predictions based on the parameters chosen, while it’s often hard to do the same with sociological models. This is not to say that I never think that sociological models are valuable - I think understanding the way that kinship works is one of the key variables one needs to understand to understand the grand story of human society - but just that all else being equal I prefer a rational choice model.
Alternatively, one could try to think about the same process described by a rational choice model qualitatively. This is much a stronger challenge to rational choice models than sociological models that limit themselves to trying to understand preferences, and on this question I come down very hard in favour of rational choice models. A good example of this is poverty traps. A qualitative description of poverty traps might say something like poor people are poor because they don’t have enough money to eat so they aren’t productive so they don’t have enough money to eat ect. However, if you try to model this, you find that this qualitative description of the mechanism doesn’t in fact get you a poverty trap!1 If you earn some money when you work you spend some of it on food. If this makes you a little more productive then eventually you iterate your way out of poverty. This fact is very easy to see with mathematics but very hard to see verbally.
I’m currently working on models of nuclear war and I’m using a rational choice model called the bargaining model of war to try to understand when a nuclear war would happen. In this case I think rational choice models are really useful for the sorts of policy related questions I’m working on. I’ve assumed I’m in a situation where we might get into a nuclear war between two adversaries. This seems like the sort of thing that might happen - two nuclear armed states are competing over territory, or power and maybe harbour some animosity towards one another and I want to see when that situation will spiral out of control into a nuclear war and try to recommend policy that can reduce the likelihood that happens given we’re in that situation. The only assumptions I need are that both states don’t want to be nuked but both want something that they can’t both have and neither are absolutely unwilling to use nuclear weapons.
One could imagine other ways of approaching the problem of nuclear war which try to reduce the likelihood of nuclear war by changing preferences, for instance by trying to stop the narrative of competition with China developing in the US, or, if I had influence over the Chinese government, reduce the importance of Taiwan. The reason I’m not working on these approaches is partly a matter of taste and partly that I see value in understanding how we get to nuclear war in the very general case of two adversaries who want a good that neither can have all of.
Thanks to my development economics professor Dr.Moav for this example
You're wrong about this point:
"But I don’t think that this is what we see. In general it seems like individuals don’t display this pretty weird erratic behaviour and so I don’t think objections to these axioms are particularly strong reasons to reject rational choice modelling. "
In reality, people DO prefer A to B, B to C, and C to A. It happens all the time. In fact, this was demonstrated all the way back in 1954: https://www.jstor.org/stable/1909827 and then Kahneman and Tversky made a whole career out of exploiting this type of thing.
There are further issues when trying to extrapolate individual choice to group choice, since transitivity isn't even almost allowed in group decisions. Marquis de Condorcet proved that in the 1700s.
As George Box famously said, all models are wrong, some are useful.
You have to ask yourself if a rational choice model is useful in a given situation. This will generally not be the case, but maybe there are edge cases in which it is.