Which is more convincing? A well-structured argument or a data based inference?

[Squatch347]

I recognize the subject line somewhat involves two related concepts, but as I was perusing some discussions today it occurred to me that some people find a structured argument or causal description more compelling than a data set or regression model.


Specifically, the discussion was between Austrians and Keynesian economists. Austrians have long been distrustful of demand side econometrics (essentially economics based statistics arguments) preferring the study of institutions, laws, interactions, and their study of interactions.


For a bit more clarification allow me to pose an example I run into quite often.



Minimum wage reduces employment opportunities for low skilled workers.

For: Increasing a wage is essentially increasing a price on a good or service. Since demand curves are obviously downward sloping, this necessitates that a lower quantity of labor is consumed.

Against: Using data in a state that raised minimum wage shows no statistical difference in unemployment rate changes between that state and one that didn’t raise minimum wage.


I’ll spare you the back and forths.


Both of these seem like relatively coherent arguments. Neither is clearly conclusive in the classical sense and when I step back I would seem to think, at first, blush they have relatively the same “strength.”

So which do you find more convincing? Or is one type of argument inherently more convincing?

The latter wouldn’t seem to be the case in my opinion given that if we take either form to its extreme:

If a then b.
A.
Therefore B

Or

The set {2,4,6,8} contains only even numbers.


Both are absolutes.


That would seem to indicate that any preference would be subjective (relative strength of arguments aside). So given that, I’m curious to hear all of your takes on this.

[CliveStaples]

For: Increasing a wage is essentially increasing a price on a good or service. Since demand curves are obviously downward sloping, this necessitates that a lower quantity of labor is consumed.

Against: Using data in a state that raised minimum wage shows no statistical difference in unemployment rate changes between that state and one that didn’t raise minimum wage.

The conclusion of deductive arguments are only as reliable as the premises of the argument. And since it's difficult to know a priori how people will act, or even the principles that will guide their actions (if indeed there are any principles guiding their actions), it makes sense to gather empirical data to check the accuracy of one's assumptions about the motivations and decision-making principles at work in a given population.

However, any data-based inference will be drawn on the basis of some sort of deductive, inductive, of abductive argument, so I don't see any difference in kind between the two arguments, only that one happens to cite empirical evidence while the other refers only to the given premise that has implicitly either a priori or empirical support.

[Sigfried]

I suppose it depends a bit on what conclusion I'm after but generally I'm a practice over theory person when the two come head to head.

The statistics will be more convincing to me if the argument is, "should we do this?" Often its one thing to imagine what should happen, but another to see what does happen. What does happen is often a complex web of actions and reactions that can be hard to predict and the only way to find out is to do.

That said I still find the theory arguments compelling. I don't doubt the effect of demand and supply, I just think they are often caught up in very complicated situations where simple models will tell you part of the picture, but often not the final outcome.

[Squatch347]

Clive,

That they are similar in kind is an interesting take I hadn’t thought of. You are right that the examples I used were, of courses appeals to a separate argument or inference (microeconomic law) rather than some kind of purely deductive reasoning. That you are right about deductive reasoning being only as strong as the premises goes without saying as well.

Here is where I might make a small difference though. The appeal to microeconomic law is an appeal to relationships that are true given certain assumptions (rational utility maximizing), most of which are general principles applied across a spectrum of individuals (though not necessarily, the OP is a bit broader than just social sciences) and so at least provides some clear expectations of when the argument might not hold. I’m not sure the same clarity comes with the data based inference.

Perhaps this is just my personal bias about how people use statistics, but I think the assumption cloud in the example rebuttal (and indeed in most data based inferences) is usually less clear. There are assumptions about statistical power, population dynamics, demand characteristics, etc. that aren’t as clearly defined (neither are clearly defined for sure).

Returning to my last point in the OP, I suppose I am resorting to the inference being “further from ideal” than the deduction. The pro position seems very much of the deductive form, and the question is about whether that specific premise holds true, while the con would seem to have questions of a larger nature (is that set related to the argument, are even numbers the right category, etc). Perhaps my bias stems more from my take that it is easier to get closer to the ideal from with a deductive argument than a data inference.

---

Sig,

I think your section about “should we do this” arguments is very interesting. I have a similar view the opposite way (well to some extent, some Austrians refuse to use data at all, which seems silly).

Interestingly, it is the complexity of most of these subjects that makes me a bit wary of the data based argument (or at least of the data driven hypothesis argument). It is very easy to gather up a bunch of data, find a correlation and declare a relationship that might or might not hold in the future. To quote brokers, past performance is not indicative of future results. You are dead on when you note that these are complex subjects, with lots of correlations and the number of possible assumed conditions grows very rapidly and it is often impossible to fully understand the relationships (if the system is emergent) and given that, virtually impossible to identify the conditional assumptions.

I suppose, to echo what I wrote to Clive, in my experience the deductive types of reasoning tend to have those conditions more clearly stated than the data based inferences. And when I’ve seen deductive arguments put forward, they tend to be about larger group sets where individual differences net out.

That said, my early experience with deductive arguments was in the social sciences (anthropology and archeology) where they were being applied to individual finds or cases. Those seemed much less tenable to me and far more speculative. Arguments like, “we found a fire pit,” “fire pits are associated with sleeping areas,” “this was a sleeping area.” I realize it is a fallacious argument, but it is a broad example of something I remember being quite common. It also highlights what I mentioned above, the assumptions being made in it seem pretty clear to me, which gives me fewer misgivings than an argument where they aren’t as open. “In 45 of 53 sites firepits were found in sleeping areas indicating the culture valued the protection of fire in family life.”

Now that I’ve contradicted myself several times, I think you make a very valuable point when you state that it depends on what kind of argument is being made. You mention the “what should we do” cases, but aside from those, what kind of criteria drive a decision between deductive/inferential arguments being more persuasive?