Lithium in the Water Supply and Our Preferences

Big Think on one of my favourite topics - changes in behaviour caused by mineral levels in the body:

Communities with higher than average amounts of lithium in their drinking water had significantly lower suicide rates than communities with lower levels. Regions of Texas with lower lithium concentrations had an average suicide rate of 14.2 per 100,000 people, whereas those areas with naturally higher lithium levels had a dramatically lower suicide rate of 8.7 per 100,000.

Interesting stuff.  But what does it have to do with economics?

Strictly speaking, the fact that lithium concentration might lead to dramatic differences in suicide rates does not violate the standard rational utility-maximizing assumptions of neoclassical economics, but it still makes me a little uneasy about those assumptions.  Note I use the term 'might' here - there may be other reasons causing the differences in suicide rates; I can think of many reasons why ceteris paribus (all else being equal) does not hold here.   But for the sake of argument, let's assume that the suicide rate is a function of lithium concentration levels in the water supply.

Suppose the decision to commit suicide is done on a rational basis - that a person does a cost/benefit analysis of staying alive, and if the costs outweigh the benefits, then they make the terrible decision.  If lithium lowers depression and lowers the costs of staying alive, then this is fully rational behaviour.  So what's the problem?

The difficulty I have with it is this - that if our preferences are so easily influenced by factors we cannot easily detect (and lithium levels are by far not the only case - consider Toxoplasma infections), then it makes predicting human behaviour that much more difficult.   If we observe today that an agent "chooses A out of A and B", we can use that to predict future behaviour, particularly when combined with a few preference axioms such as transitivity.  However, if their preferences are greatly influenced by transient microscopic factors we cannot easily observe, then what hope do we have?  As someone who both researches human behaviour in the academic world and someone who uses it daily in the private sector (most of my day job can be classified as 'marketing') I find this deeply unsettling.

I would love to hear your thoughts on this, as this is again another blog post where I've provided more questions than answers.

Comments

[Leigh Caldwell]

Stable preferences are definitely one of the more dangerous simplifications of the rational model of decision-making.

These kind of experiments, and many other behavioural results (not all involving chemical modification!), have shown that preferences can easily be influenced by context and by various factors. However, as far as I know, there has been no serious attempt to incorporate this into economic models.

Within psychology (in the judgment & decision-making subfield) researchers do typically try to construct low-level decision-making models which explain some of these results - however they do not really aggregate these models to look at their economic consequences. So we don't really know whether (for example) a partial equilibrium will easily become unstable because of changing preferences, or whether social welfare really is maximised under a system of complete markets where agents do not know their own future preferences.

As you point out, this is a really important area to understand for people in marketing, so there is every reason why we would want to construct better models to explore it. Some researchers are working on this area under the banner of "cognitive economics" (Marco Novarese for example, and me) but it isn't yet a widespread field.

I do think that better mathematical modelling of real ("irrational") behaviour is the natural next step for behavioural economists, but lots of the old guard of that field (Thaler and De Bondt at least) don't agree with me. I think their attitude is that it's unrealistic to think we'll be able to model phenomena like this, and we should instead learn to better accept the limits and caveats of economic modelling.

[Dan]

Somehow I don't think that a 2 page paper by a psychiatrist bothered to control for loads of other (economic and not) factors that could cause higher suicide rate. So my prior is that this is just spurious correlation.

[Chris S]

Following up on the comments and the second article, it seems there is another potential causal route:

1) Rain water dilutes lithium levels (and maybe all ground-sourced minerals)
2) Higher rain levels correlate with lower lithium levels
3) Higher rain levels correlate with lower sunshine levels
4) Lower sunshine levels correlate with higher suicide levels

If this causal route holds, then the original correlation is not spurious - but it isn't causal either.

However ... we can replace "affected by lithium" to "affected by sunshine". Now, sunshine levels are certainly more easily detected, but - how many national marketing plans factor in regional weather?

Nevertheless, Mike's original concern is valid. How can you trust your predictive methods if the range of outcomes is swamped by factors you can't even detect? And - even if you can detect them, aren't you in the same boat if your model simplifies those factors away?

This is one of the reasons that I find complexity theory so refreshing, as it takes on some of these assumptions head-on, and recognizes that much methodology that we trust may be subject to buried counter-correlations. That said - complexity theory is very hard to *apply* if your goal is to improve your predictive ability.

[Wonks Anonymous]

If lithium concentration is also stable, then behaviour should be stable as well and you can continue making accurate predictions.

[Lord]

Such effects affect each of us continually. Every time we decide on what to have for lunch they impinge on us as the desire for something salty, sweet, or sour. It is not so much a matter of individual effects though, but of systemic ones that can shift public preferences significantly over a short period of time. It would seem the entire fashion industry is predicated on the very fact that they do. There are probably aspects of fashion that can be studied, creation, propagation, penetration, duration, but predictability may be beyond what is possible.

[Gabriel Kreindler]

Short interpretation of your problem: no perfect prediction of preferences is possible because there is some error term / stochasticity. Lithium levels might change all of a sudden and affect preferences, but let's say we weren't expectin lithium levels to have an effect. So a priori this is like a shock. If we had better instruments we could have extracted some more factors out of the error term.

If there are many microscopic factors that change often (transient) and if you assume there is no overall bias, then the law of large numbers suggests that there won't be any large bias. This should be even more true in the aggregate. As in the standard motivation for accepting normally distributed errors in linear regressions - imagine they are in fact averages of very many iid shocks (perhaps with a different distribution).

Even if there are few microscopic factors, the variations in time are still stochastic and perhaps it makes sense (given we don't know anything else about them) to assume their expected effect on behaviour is nill. Until we get a result of the kind "Any change in mineral levels in the body, *in any direction* will always make people more depressed/ whatever", it seems the only natural thing to do.

So I don't see any real problems for neoclassical econ here.

[David]

If you start reading up on behavioural econ (Dan Ariely's books and Nudge are a good start), you start to realize that the rational economic agent assumption has some flaws to it. The lithium thing, assuming it holds ceteris paribus, is just one example (I blog about a couple other examples here: http://dmkarp.blogspot.com/2009/10/scented-nonsense.html)

I don't think it's unreasonable to think that environmental factors influence behaviour. I think it's pretty obvious it does. It just means that while incentives matter and our economic predictions of how people act will hold true most of the time, we can never completely predict how every person will react in every situation. People aren't robots making perfectly consistent decisions using some kind of well-defined algorithm, after all.

[Gabe]

I agree with Dan. I suspect ceteris paribus fails and the correlation is spurious.

[gold Guild Wars 2]

Suppose the decision to commit suicide is done on a rational basis - that a person does a cost/benefit analysis of staying alive, and if the costs outweigh the benefits, then they make the terrible decision. If lithium lowers depression and lowers the costs of staying alive, then this is fully rational behaviour.