I guess the follow up question to this is: what superior alternative do you propose we use for evaluating competing government policies?
One problem is of uncertainty (this was pointed out by Keynes) - people simply do not trade higher risk for higher profits, even if the mean value increases in the mathematical sense. Rather, they set a horizon of uncertainty.
The other big problem is that utility is ill-defined for an individual. We are competitive beings and value many things for reasons that are related to how other people value them. This cannot be modeled within the theory of rational choice. (It's also somewhat related to the fact that supply-demand is not that great model.)
In other words, we should augment the models to take the above issues (at least, IMHO those are the biggest problems but certainly not the only ones) into consideration.
I think at this point even the most conventional academic economist would start giving you sideways glances if you asserted that people are always rational utility maximisers. It seems that most econometric modelling takes a top-down approach; implicitly making homogenising assumptions about the models 'atoms' (e.g. individuals, firms, states, countries etc.).
I often wonder if the future of economic modelling is to be found in 'agent-based' models, where you start at the lowest level and aggregate up. This approach makes it possible to account for the effects of non-homogeneous actors and to also include 'non-rational' behaviours observed in experimental/behavioural economics.
I feel that the other issue is how beginner-level economics is taught at university. I would agree that there's no practical way to teach economics to beginners without making a bunch of simplifying assumptions and presenting a bunch of simplified models for study (e.g. a perfectly competitive market). However, I think the problem is that insufficient effort (if any) is made to emphasise that these models do not really reflect reality, and are just 'toy models' to help you develop your analytical thinking.
At best, in a real world context, understanding these models gives you an analytical framework for reasoning about whether some policy proposal is pointing in the right or wrong direction, though even that is a pretty fraught assertion.
EDIT: I figured I'd address some of the comments below regarding randomised-controlled trials, since its related to agent-based modelling / simulation. Comments below have pretty much highlighted the crux of the problem with economics and public policy (albeit without comprehending the implications): barring the rare 'natural experiment', its virtually impossible to collect 'clean' data that allows comparison of different policies. It's true that RCTs are being run in a 'public-administration' context, and the idea is gaining popularity. Most notably, the UK government has run a bunch of RCTs on various things.
But the crucial thing to note is that these RCTs are generally only run for 'low-stakes' issues. For instance, one of their RCTs involved varying the wording of their tax authority's "you haven't submitted a tax return" letter to determine what gets the highest compliance rate.
How comfortable, from an ethical standpoint, would you feel running state-by-state RCTs to determine the best government healthcare policy? Particularly given that the outcome in some states will be higher levels of preventable death and permanent disability? I've chosen healthcare here because the connection between different policies and higher or lower rates of mortality is fairly direct. But for any other government policy (of any significance), you're generally looking at the same sort of stakes. It's just that the ultimate ill-effects of a bad policy are not immediately apparent, take a long time to unfold, and often only eventuate at the margins.
Let's say you're the King of the United States. One of your noblemen comes in to your palace and says 'Your majesty, I suggest we eliminate corporate taxation and replace the lost revenue with a tax on consumption.' What happens next?
Oh and use "muddling through" more openly, because that's what life is about no matter how much data you collect. So don't pretend that there even IS an "optimal" answer, and/or that you know it and/or even that anyone can find it. Let's admit that we are all just trying things and that failure IS an option. So create systems that a) actually encourage and allow experimenting, and b) let those fail safely.
However, if you do happen to have political opinions (that you think are well-reasoned), then you are engaging in this form of 'misguided analysis' that economists also engage in (albeit unknowingly). As for the 'economists' who claim to 'know the optimal answer', I guess I've just been fortunate to have never encountered one. More precisely, I've never run in to one in the various federal economic public policy organisations where I've spent the majority of my career. Where do they hang out?
And good luck designing experiments around, say, national healthcare policy that 'fail safely'. I think you'll find that an alarming number of them end up 'failing deadly'.
I don't agree with the idea that people, and society,
are completely impermeable to analysis.
This is just like medicine: You can theorize all you want, but in the end only a clinical trial shows you if your plans actually work. The subject is too complex. All you can do with your models is use them as input into the trials. If you skip trials you will get disaster and failure on an unsafe scale.
And good luck designing experiments around, say,
national healthcare policy
Though we would probably just have one about who offended the gods.
Single payer is a great example, Californian or NY voters will likely vote in such a system while Texans would likely opt into a free market approach. Wait a few years and see the outcomes of each policy. Now you can make an evidence based decision rather than a political one.
I wonder what would be the best way to implement something like this? Just spitballing, but what about Federal grants to partially fund the 'treatment' States (with non-participating States acting as controls), with the policy issue and various 'treatments' negotiated between the Fed and State level? And then after that, States sign on using a mechanism similar to California proposition voting during State elections maybe?
Or maybe even push it down to a local government level? I seem to remember somewhere in the US conducting trials of basic income around the 50s or 60s. Maybe there's a model there...
The thing is, political views don't diverge quite as much as it is being made to seem. Gerrymandering scrambles the signal[0] quite a bit and the Electoral College then magnifies[1] the apparent differences.
[0] http://theweek.com/speedreads/541955/howgerrymandering-works...