Need a little help with a priori vs empiricism argument....

Economics Economics Questions
1

I’m having a conversation on another forum, and am seeking some advice. Here’s the conversation so far.

ME: My implication is that economists come to very different conclusions from the same empirical data, which implies some kind of fundamental disagreement, or conflicting principles

Him: Wouldn’t that depend on the data specifically? Economists use a method called statistical regression to analyze date e.g the relationship between age and income. There are plenty of things an economist must be very critical of when both creating a sample and running the data. Due to these complications, economists can argue that a failure to account for a correlated variable (omitted variable bias) or a sampling the population in a manner that introduces a bias, can under-represent the relationship, over-represent the relationship, or indicate the data cant be trusted at all. There is a great deal of argument in this area because of the complications of making observations in this most dynamic system we call an economy.

Me: The principle of Human Action, aka the science of praxeology is the basis for all economic activity. I feel this is largely ignored or not understood by most mainstream economists, and is seldom, if ever, even mentioned by any mainstream economists.

Him: Economics is all about human behavior. In fact, all of economics is based on some fundamental assumptions about human behavior. Praxeology is not ignored, but its addressed in a different manner. For instance, expectations, the idea that what people believe is going to happen has an impact of what actually happens, is an example of taking into consideration human characteristics in a complex system. However, the study of human behaviors is something that can be observed like anything else, and we can make very strong statements about peoples behaviors based on observations. What economic purpose would it serve to make any model about our economy that can not be empirically be shown to be close to reality?

Me: empirical data cannot be properly understood without reasoning from the right first principles.

Him: I’m struggling to grasp this statement here. Typically, data is collected, and then relationships between data sets are calculated using statistical analysis. This is called econometrics. The use of advanced statistical analysis to calculate relationships between variables. Economists (who are in today more closely associated with mathematicians than other fields of study), are taught to be very astute when drawing conclusions from data. They are taught to be skeptical of their data, do be critical of all types of errors that can occur, because their colleagues will be critical of them.

Just looking for a little knowledge and advice. :slight_smile:

2

Seems like you’ve got a decent debate partner there. That sounds like just the kind of person I would want to have a disagreement with. I have a feeling this is what you’re looking for…

Praxeological Economics and Mathematical Economics

You might also check some of the first links here.

For more detail, see Hoppe’s Economic Science and the Austrian Method…in particular the section “Praxeology and Economic Science”. And for more, see this list.

3

I am beginning to think that this answer might be best:

We don’t reject the scientific method, mainstream economists do. Statistical analysis without controlling of variables is useless in science. Since you can’t control all variables in economics, mainstream economics is in fact unscientific and no better than a salesman trying to sell you snake oil.

Now, if you reason from first principles with arguments like “humans act, therefore, blah blah,” then you will always have a true argument.

4

Well that doesn’t make it unscientific at all. Does it also make it unscientific because we can never know what distribution a variable comes from? That’s why we have the central limit theorem. Does it make it unscientific that we can’t run infinitely many trials? That’s why we have inference based on variances. The fact that we don’t have the luxury of randomised controlled trials is the entire reason econometrics exists; for overcoming obstacles similar to those we used to have in the past, but don’t have anymore thanks to people not quitting before they start. It’s absurd to claim that without RCTs something is not science, and doing so is not only setting yourself too an easy task, allowing yourself to fall back on making “self-evident” assumptions, it’s just plain lazy.

If you reason from first principles you’ll arrive at true conclusions given that your assumptions are actually true and the logical steps you filter them through are valid (this is also known as mathematics, by the way). How do you know your assumptions or logical steps actually hold in reality? You test your them with, you guessed it, statistics.

5

Econometrics qua econometrics is flawed because it does not measure what it purports to measure. The human brain has around 100 billion neurons (about as many as the stars in the Milky Way galaxy, just to give you an idea of the scale) that interact in gazillions of combinations with one another. There are 7 billion such brains on the planet that wake up every morning and spend 16-20 hours going about whatever it is they choose to go about. The idea that you can blindly apply statistical methods that break down on much less complex phenomena, such as weather, to the human economy is downright silly.

The correct approach to empirical questions is as a matter of economic history. Mises explains all of this in Theory and History, I recommend you take a look.

Clayton -

6

The problem with economic models based on statistical analysis can be summed up in one word: heteroskedasticity.

7

Well firstly, heteroscedasticity doesn’t result in biased estimators. Heteroscedasticity is the last Gauss-Markov assumption, required for minimum variance unbiased estimators. Second, people didn’t just go “Oh no, heteroscedasticity! Looks like the jig is up”. They created generalised least squares and heteroscedasticity-robust variances.

8

@Marginal: Almost the entire body of probability theory and statistics is not applicable to non-stationary processes, of which the human economy is a gilt example. And as soon as you start applying the tools of probability theory which are applicable to non-stationary processes - such as Bayesian inference, Markov chains, and so on - you’re right back in the realm of “verbal, deductive logic” so spurned by Keynesian and other mainstream economists.

To make matters worse, it is impossible to derive causal relations from statistical methods. In other words, you can’t do deduction with inductive methods. More powerful methods (such as Solomonoff induction) must be employed but these methods are so abstract that they have as yet little application to real-world problems.

Clayton -

9

Every philosophical discipline (including natural philosophy, aka science) makes assumptions. The question is the relative quality of this set of assumptions versus that set of assumptions.

Let us ask the question, “Does raising the minimum wage rate improve or disimprove social well-being?” In answering this question, assumptions will have to be made. I can assume things like “humans prefer more to fewer goods and prefer goods sooner to later” and attempt to answer the question with deductions based on these assumptions. This is the praxeological method. Or, I can assume the axioms of statistical methods which are, to be sure, much less open to question than the statement “humans prefer more to fewer goods” if for no other reason than that there is infinitely less to argue about in the axioms of statistical methods.

However, it is not sufficient to take the axioms of statistical methods as my starting point, I must assume something else, namely, that those methods can be meaningfully applied to human behavior in order to answer questions like the above. And this assumption is much, much weaker - refutable, in fact - than the assumption that humans prefer more to fewer goods. Hence, you have reduced the uncertainty in one portion of your axioms at the expense of increasing it many times more in your other assumptions or, even worse, at the expense of starting with false assumptions.

Mathematics is not a panacea. There is no magical “truthiness” in maths. Mathematics is a ritualized (formalized) subset of natural language, nothing more. It is useful for reducing uncertainty in the investigation of problems which can be adapted to the limitations of mathematical methods but blind application of mathematics to problems which are not properly reduced to mathematics is just sophistry, numerology, astrology, quackiness. This is precisely what a large portion of modern economic theory is.

Clayton -

10

Sorry, I just need to clarify: Are you using the phrase “non-stationary” in the context of a non-stationary time series? That’s what I guessed, given the topic of statistics, but the second half of your paragraph isn’t consistent with that. So… I’m not really sure. If you are, then thankfully not every time series we have to deal with is non-stationary, plus cross-sectional data doesn’t face that problem.

As for “back in the realm of ‘verbal, deductive logic’”, I have no problem at all with verbal logic. It would be insane to require that the only method of solving a problem be to use esoteric symbols. On the contrary, the vast majority of mathematical proofs I’ve done have been entirely verbal. Your assertion that this is “spurned by Keynesian and other mainstream economists” is an absurd mischaracterisation.

You can derive causality to within a specified level of significance. Obviously you can’t get to 100% confidence without running infinitely many trials. But statistics isn’t supposed to be a replacement for models based on deductive logic. It’s there to test that assumptions/predictions in the model hold up in reality. It’s there to disconfirm models which have made errors in assumptions or logical steps. A filter to get rid of bad models and retain and improve upon good ones. I’m trying to find a quote I heard a long time ago, but I’m not having any luck. It was something along the lines of: [of physics] no model is ever proven true, the best it can hope for is to survive.

11

You say it’s “the praxeological method”, but that’s nothing unique to Austrian economics. Everybody does that. We all make assumptions, which we think are reasonable, that consumers maximise their utility, you get more utility from more goods, for most goods there’s diminishing marginal utility, we prefer stuff now to stuff later, firms maximise profit, etc. Those are assumptions you need to build models, and they’re reasonable assumptions to make (of course they’re not always assumed though). The problem, when answering questions on the minimum wage, is that these harmless assumptions aren’t enough. The answer to whether an increase in the minimum wage will cause unemployment depends crucially on the market structure of what you’re applying it to. Is the market something approximating perfect competition (in which case a higher minimum wage will cause unemployment), or does the market display a degree of monopsony power over labour (in which case a minimum wage will increase employment)? You can’t go around assuming things like this or determining them a priori. This is something that needs to be measured empirically before the question can be answered.

12

Could you describe in more detail the scenario in which these circumstances would obtain? What do you mean by the market displaying a degree of monopsony power over labor?

13

Monopsony power over labour is where firms don’t take a prevailing market wage as given (as in perfect competition) but can affect the market wage through the quantity of labour they choose to employ.

This could happen due to insufficient competition among firms in the labour market, search frictions, imperfect labour mobility.

The point being, you can’t just assume whatever you like (in this case, perfect competition), you’ve got to measure it.

14

I don’t think anyone has ever seriously argued that perfect competition exists or can exist or even what it would actually look like in the real world. There are problems in the real world and “friction” but these problems exist for Marxist theorists as well, such as yourself it seems.

15

As has been alluded to I think above, pure empiricism cannot distinguish between multiple possible causes and effect. Case closed, that’s it.

16

…

You seem to have missed the point. And actually, several industries come extremely close to resembling perfect competition (where statistically the null hypothesis of perfect competition can’t be rejected at the 95% confidence level). Completely beside the point though. The point is: In reality, the labour market for a specific industry will lie somewhere between perfect competition and total monopsony. Where it lies on this spectrum is important for determining the effect a minimum wage will have (create unemployment or not). You can’t just assume where the industry in reality is, you must measure it. That’s where statistics comes in.

Marxist?! I don’t think so.

17

I’m speaking of a stationary process as described here, in which mean and variance do not change over time. If 10% of people have become doctors during a specific period in the past, that says absolutely nothing about the future distribution of occupations because there is no stable average and no reason we should expect there to be. The probability of rolling a 5 on a die is stationary because that’s what a die is… an object that produces 5’s according to a stationary distribution.

Perfect confidence in a 1.0 correlation between variable X and variable Y still would not establish causation. You just can’t get from correlation to causation and statistics by definition is the study of correlations between random variables.

OK, I agree with everything in this paragraph but the OP was regarding the use/validity of econometrics, in particular and some other peripheral issues. Econometrics is flawed for the reasons Mises explains here. Econometrics wants to make statements like, “If we raise taxes, there will be X% chance of 4% growth in GDP over the next 10 years.” Such statements are nonsensical when analyzed against the categories of class vs. case probability.

Clayton -

18

Well it does. Am I missing something here… Do you have a heterodox school of physics based entirely upon “self-evident” truths and deductive logic rather than empiricism?

19

Yes, but the issue is the quality of the assumptions - are they true? This question can only be answered by going over the methodological issues. But the mainstream schools of economics don’t often talk about methodology, they prefer to do “real work”. But what’s the use of doing “real work” with flawed methods and getting nonsense results?

But the degree of “monopsony power over labor” is not something that is measurable any more than the “deviation of the Fed target interest rate from the natural rate of interest” is measurable. There is no counterfactual price of labor or interest rate against which we can measure the actual price of labor or interest rate to determine whether or how far the market has deviated from “perfect competition.” So, measurement is useless in this regard.

Clayton -

20

If statistical analysis in the social sciences is invalid, then we are unable to know very much about the world around us. All we can talk about is tendecies which may be at work. For example, even if the ABCT is correct, it may have little economic impact (as Andrew Young and a few others have found). Other factors may also be operating in the background.

Suppose we want to know whether a gender-based wage gap exists. We could try to reason out the answer (as Austrians like Block do). But this would not tell us whether our reasons were correct, as opposed to a vast number of other arguments. We would need to perform some kind of quantitative analalysis to tease out possible causes.