What makes a useful theory?

If conventional economic theory is so wrong (as we are repeatedly told) why does it survive so well?

This post by UnlearningEcon prompted me to think again about why economics, despite widely accepted empirical data from behavioural econ, is broadly taught in the same way as before, and why its basic assumptions still underpin much modern research.

Some have a sociological explanation for this. In this view, economists are invested in the old approaches, have spent decades honing specific mathematical skills, and effectively collude to make sure new ideas do not displace the old. The top journals only accept papers that cite the same old work, perpetuating the models. Science, as they say, advances one funeral at a time. No doubt there's something to this, but I don't think economists are quite so closed minded.

There is a clue in the above article:
"...Euclidean geometry, despite being incorrect, is more effective than non-Euclidean geometry in some engineering and architecture."
In practice, it's hard to see why an engineer or architect would even think of using non-Euclidean geometry - unless they are calculating a space probe's route to Neptune (or in a few specific scenarios relating to aeroplane flight paths). In earth-bound contexts, Euclidean geometry is so close to being right that there's no point bringing in non-Euclidean approaches.

UnlearningEcon complains about the use of "as if" theories: when a theory has been empirically disproved by data, why do we keep pretending it's true? Why model people as if they are rational utilitarians when they're clearly not?

A theory is useful because it enables us to predict - and therefore control - the world. Two properties make a theory good at this job:

  1. It should accurately reflect the world
  2. It should be practically generalisable
The most solid, widely-used theories do well on both criteria: the laws of thermodynamics or statistics are highly accurate (no better model has been found) and they can be expressed in relatively simple, generic forms that can apply in many contexts.

Some theories, like Newtonian mechanics, are known to be slightly inaccurate in some contexts (Einstein's relativity superseded them) but are still so close to being right - typically within a millionth of a percent - that it hardly matters. And because Newton's laws are so much easier to generalise and use than Einstein's, it's a more useful framework.

Now, economics. We do know that the rules of economics - utility maximisation, the irrelevance axiom, etc - are wrong in lots of real-world contexts. They aren't so wrong as to be useless, but they are far less accurate than Newton's laws. UnlearningEcon mentions the priority heuristic, which does make better predictions of people's behaviour in lotteries than its competitors, expected utility or expected value theory.

But the priority heuristic is hard to generalise. Knowing how people prioritise gains and probabilities in a formally-specified gamble does not tell us how they prioritise money versus pleasure, or more leisure versus a bigger TV. It doesn't say much about how producers will choose what products to offer, or about the larger picture of a market or a whole society.

Standard utility and price theory, on the other hand, is very general. Theories relying on those underpinnings can be applied to any market, and to a lesser extent to non-market contexts (there are lots of theories about the economics of crime or marriage, for instance).

If you're looking for a way to predict what happens in a life situation, of course you'd rather have an accurate, empirically-supported theory to tell you, based on data, what to expect. But if you don't have one, because the data hasn't been collected yet in your specific situation, a second best approach is to rely on a generalising a theory from other scenarios.

Classical economics lets you do that. You can borrow rules that work reasonably well, apply them in your context and have a ready-made set of tools and predictions that have a chance of being roughly right. Maybe nobody has gathered enough data to make a theory specific to the crab fishing market off the Isle of Mull, but the theory tested in the trading pits of the Chicago commodity markets reads across close enough.

More realistic economic theories have two ways of emerging, then.

First, we can keep collecting more situation-specific data and create theories for each context. Then we can independently develop a Mull Fishing theory, a theory of social media clickthrough rates, a theory of how people use shared resources. It will be a lot of work, but the predictions will be well calibrated to their environments and probably quite accurate.

Second, we can work on generalising the psychological and behavioural theories that have been developed in narrower domains. I think this is the direction that holds out hope for better economics.

I anticipate that behind the priority heuristic lies a deeper explanation, that can tell us why people make choices based on minimum gain, and why probability is more important than maximum profit. The Vlaev et al paper linked from UE's article reviews examples of work that points in that direction, but it is not a unified theory in itself.

Some behavioural economists disagree that a unified theory will be possible, while others see it as the next important step in the field's development. I'm with the second group, but regardless: we don't quite have one yet.

In the meantime, it's inevitable that economics practitioners and teachers will rely on the one general theory they have: neoclassical utility maximisation. And, in 80% of the practical decisions they must make, they will be right to do so.

Comments

Popular posts from this blog

What people want, cognitive goods, models of persuasion and why we avoid important information: the cognitive economics session at the AEA conference

Introducing System 3: How we use our imagination to make choices

Discussion 2 of 3: No spooky action at a distance - a theory of reward