There is an unfortunate divide in the rationalist tribe between Bayesians and believers in scientism. Bayesians are those who rationally incorporate all sources of information when choosing what credence to have in different propositions. You have prior credences that are set by common sense, theoretical arguments, empirical information and so on. You then update from those priors with new information, whether that is from personal observation, social science, theory or whatever. Believers in scientism in contrast form their beliefs by putting lots of weight on published scientific evidence over other types of evidence.
I think scientism is the wrong approach and can be costly. It is also among the leading reasons not to defer to the scientific establishment on some key questions; it is a case where the requirements of epistemic modesty are not as they might appear. A lot of experts with comprehensive knowledge of the scientific literature have the wrong epistemology and oftentimes, for that reason, we shouldn’t defer to them. I will illustrate this with examples.
Masks
The Bayesianism vs scientism cleavage has played out recently in the disagreement between experts about the efficacy of face masks. Some experts have come out strongly in favour of masks, whereas some have cast doubt on their efficacy. For example, in August, a Swedish epidemiologist said:
“Mr Ludvigsson noted that in a meta-analysis by the WHO of 29 studies that showed face masks were effective, only three concerned their use outside hospitals and of those that did not none involved Covid-19.”
This is a paradigmatic scientistic attitude. Mr Ludvigsson casts doubt on whether masks work outside hospitals and for covid because there haven’t been any studies testing them outside hospitals and for covid. The Bayesian approach, in contrast, would update on the information from all of the studies, even though the context is not exactly the same in the studies as what is being proposed now.
They would also take into account the basic science of viral transmission, and use common sense. We know that covid spreads via droplets that are released from the mouth or nose when an infected person coughs, sneezes, or speaks. People can catch covid when those infectious droplets get into their mouth, nose or eyes. This is where common sense kicks in: if you put a mask in the way of the droplets from my mouth or nose, then that makes it less likely that the droplets will get from my mouth or nose into yours. We even have video footage of masks doing the blocking.
Even if masks have only been tested for influenza and not covid, we still understand the mechanism by which they work – droplet blocking. If they work for influenza, which spreads in the same way as covid, then they will very likely also work for covid.
The scientistic view is that to know whether we should recommend that people wear masks in enclosed spaces like shops, buses and the tube, we would need to have a high-quality study (preferably an RCT) of that exact thing. However, that exact thing can never be done twice. Any intervention that we impose will be different in some way from the version of the intervention that has been tested.
To illustrate, consider this dialogue:
Scientism: Ok, we have the evidence: masks work on buses.
Bayesian: Great! One thing – all the masks you tested were blue. Do yellow masks work?
Scientism: Well, we didn’t study that, but of course yellow masks work.
Bayesian: How do you know if you didn’t study it?
Scientism: There is nothing about yellow masks that would make them not work if blue masks work – colour is irrelevant.
Bayesian: Do you have a scientific study showing that colour is irrelevant?
Colour is indeed irrelevant. The argument for this is that the colour of the masks is not going to have an effect on the mechanism by which the mask works, which is blocking droplets. Knowledge of the mechanism of transmission tells us the results will generalise to yellow masks. But this argument is not available to scientism.
This is why the experts have disagreed about masks. World-leading experts have got this one wrong, and it is down to their mistaken scientistic epistemology. Unusually, someone who just relied on sturdy common sense would be more likely to be right than an expert who knew the whole scientific literature really well but assimilated it in a scientistic way.
The minimum wage
I am going to preface this by saying that I am in favour of more redistribution – I am opposed to the minimum wage because it is bad for low-skilled people. Economics 101 states that if you introduce a minimum wage, employers will demand less labour – they will lay people off or demand fewer hours from them. If you think the demand curve for labour slopes downwards, then a minimum wage will cause firms to economise on labour.
In spite of this, as this IGM poll shows, there is active disagreement among leading economists about whether minimum wages will make it harder for the lower paid to find work:
Some of the experts’ answers are influenced by how interpret the term ‘noticeably’, but others justify their arguments with scientistic claims, such as
- “I’m not aware of any strong evidence demonstrating this result.” (David Autor)
- “The empirical evidence now pretty decisively shows no employment effect, even a few years later. See Dube, Lester and Reich in the REStat” (Michael Greenstone)
- “Yes, I know the Econ 101 answer but the evidence suggests the effect on employment is between small and 0.” (Richard Thaler)
These statements suggest that one’s view of the minimum wage should be determined by what the median study finds about the effects of the minimum wage. But really when we are assessing this claim we need to consider economic theory, common sense and all the other evidence that demand curves for labour slope downward. Here is Bryan Caplan on the strength of the case from economic theory and common sense:
“In the absence of any specific empirical evidence, I am 99%+ sure that a randomly selected demand curve will have a negative slope. I hew to this prior even in cases – like demand for illegal drugs or illegal immigration – where a downward-sloping demand curve is ideologically inconvenient for me. What makes me so sure? Every purchase I’ve ever made or considered – and every conversation I’ve had with other people about every purchase they’ve ever made or considered.”
He goes on to argue that empirical evidence from other parts of economics should also update us towards minimum wages having a disincentive effect. This includes
- The literature on the effect of low-skilled immigration on native wages. The strong consensus is that high levels of low skill immigration have little effect on native wages. This implies a near horizontal demand curve for low-skilled labour.
- The literature on the effect of European labor market regulation, which shows that regulations that increase the cost of hiring people causes high unemployment in Europe.
- The literature on Keynesian macroeconomics. Keynesian macroeconomics suggests that nominal wage rigidity causes short-term unemployment. This makes it highly likely that government-imposed wage rigidity causes unemployment.
- The literature on the effects of price controls, on rent, energy and agriculture, which bolster the textbook story of price floors.
On this last point, Caplan makes a version of the point I made above about masks:
“If you object, “Evidence on rent control is only relevant for housing markets, not labor markets,” I’ll retort, “In that case, evidence on the minimum wage in New Jersey and Pennsylvania in the 1990s is only relevant for those two states during that decade.” My point: If you can’t generalize empirical results from one market to another, you can’t generalize empirical results from one state to another, or one era to another. And if that’s what you think, empirical work is a waste of time.”
The empirical work on the employment effects of the minimum wage is mixed. The majority of studies find a negative effect on employment, but some prominent studies do not. But this is pretty much what we would expect to find if the minimum wage did indeed have a negative effect on employment. Empirical social science research is not very good. Most published research is false. John Ioannidis has found that nearly 80% of the reported effects in the empirical economics literature he studied are exaggerated, typically by a factor of two, and with one third inflated by a factor of four or more.
Minimum wage studies are especially likely to find false negatives. Studies of the minimum wage usually test the effects of minimum wage levels that are low ($4 – $11) and so they are antecedently likely to find effects that are dotted either side of zero. Moreover, when assessing the effect of the minimum wage, you are comparing a counterfactual to the real world that is affected by innumerable other forces, including every other labour market policy, the plans of literally every business in the area, all other economic forces that could affect a region, etc. Then there is the problem of actually measuring changes in demand for labour properly. So, we have very noisy data measuring a small treatment effect – of course we will find some surprising results, which are very likely false negatives. The rational thing to do is not to update very much.
In advance of examining any evidence about the employment effects of the minimum wage, your prior that it reduces employment should be something like >99%. Upon observing any study showing that the minimum wage does not have an effect on employment, you shouldn’t update much because empirical research is not very good. Caplan establishes this formally here.