Hello everyone. About a month ago I posted about my desire to make a video refuting Unlearning Economics' recent video critiquing Marxist economics. I've done a lot of work on the video since then and I have some thoughts on the popular criticisms of the empirical literature on the labor theory of value that UE peddles. I wondered if anyone on this forum might have more to add or any corrections, any feedback would be appreciated as I'm looking to not make sophomoric mistakes (my educational background is in psychology and it has not been especially helpful in my study of political economy).

UE repeats the popular argument that's been floating around for some time that the observed correlation between labor values and the money value of sectoral output is spurious. This goes back (as far as I can tell) to Andrew Kliman's 2002 paper claiming that the correlation is an artifact of "industry size." As far as rebuttals to the empirical evidence for the LTV goes, pretty much everyone seems to be drawing from Kliman. Kliman's critique is picked up by Bichler & Nitzan and from them it gets picked up by UE.

The section in my video talking about the empirical evidence goes through these points:

1) UE cites an old spreadsheet by Bichler & Nitzan which claims to show that the correlation between aggregate price ($) and aggregate value (SNAL, or socially necessary abstract labor) can emerge even if there is no correlation on a unit level if both are multiplied by a third variable, in this case industry size. Bichler & Nitzan define industry size simply as the # of units of output.

As Cockshott, Cottrell, and Valle Baeza point out, there's a simple algebraic error here, because their example implicitly requires adding the $/unit and SNAL/unit of incommensurable commodities (i.e., $/pencil + $/barrel of oil) since these are different sectors. Calculating a correlation coefficient requires deriving the sample means of both variables, but deriving a sample mean requires adding all items in a sample together, which we cannot do here for either $ or SNAL since the dimensions are not homogenous.

There is a coherent argument to be made here (that the correlation between prices and labor values is spurious and driven by industry size), but this spreadsheet is incoherent and demonstrates mathematical ineptitude. Any unit level correlation would have to be within sectors to ensure that both vectors remain homogenous.

CasP theorist Blair Fix responds by insisting Bichler & Nitzan didn't make a dimensional error and that if we accept they did, it follows that the price system itself is dimensionally invalid. This is pure cringe to be blunt.

2) The idea that "industry size" drives the correlation might seem superficially plausible, but on closer examination it falls apart. For one, what precisely even is industry size? The energy input of an industry? The literal physical space it takes up? Bichler & Nitzan argue that it's just the # of units of output produced.

But there's a way to put this to the test: if size is a third variable driving everything, then logically, any input that should scale up with industry size would correlate just as well as labor. This is not seen, as for example pointed out by Zachariah (2006). Not even energy input correlates nearly as well as labor with the money value of sectoral output.

UE alludes to this response in his video (he doesn't actually detail this response though, he just mentions that Cockshott has responded to the charge of spurious correlation) and then retreats to another argument...

3) UE says that Cockshott "misses the key take away," which is not that industry size specifically is driving the correlation, but that any number of variables could possibly be driving it. UE just suggests that "historical factors" are these variables and doesn't elaborate. But this is a totally vacuous argument. One could always suppose that a correlation is spurious no matter how convincingly researchers demonstrate that it is causal. It's not a testable refutation. If you want to argue that a correlation is spurious, you have to specify a measurable variable you think is driving the correlation so that tests can be conducted to see if it's really driving the correlation.

I think the problem is that UE as well as Bichler & Nitzan seem to think that correlating prices and labor values on a sector level leaves the door open to the correlation being spurious, whereas a unit level correlation would rule out spuriousness. If Marxists demonstrate that the correlation holds up on a unit level, then this would be proof of the LTV, but correlations on the sector level could easily be spurious and thus are worthless. Bichler & Nitzan said this an exchange with Cockshott in 2010:

"…Cockshott doesn’t correlate average unit price with average unit labour time (as revealed by average unit dollar cost with some transformation). Instead, he correlates total sales with total labour values (as revealed by total dollar cost with some transformation). This move from individual commodities to sectoral aggregates introduces the further hazard of spurious correlation.”

There's two problems with their argument here: 1) capitalist economies just don't keep data at such a granular level, calculating unit prices and unit values just isn't econometrically feasible, so unit level correlations are going to remain elusive. But 2) this doesn't really matter regardless, because finding the correlation holds up on a unit level would not actually definitively remove the threat of spuriousness. Again, one could always suppose that a correlation is spurious. Cockshott relying on sectoral aggregates doesn't introduce this hazard; in principle, this hazard is always there.

However, Marxists have provided strong theoretical reason to believe that a causal connection exists between prices and labor values, have provided further evidence that the correlation isn't spurious by testing other inputs, and other predictions of the LTV like labor-saving technical innovation leading to falling profit rates have been empirically confirmed (which of course UE also disputes, though that's beyond the topic of this post). As such, the burden of proof is on the LTV's critics to demonstrate that the correlation is spurious. Until then, these correlations stand as robust evidence for the LTV.

What is meant by "industry size"? IIRC Bichler and Nitzan do not specify but thankfully Kliman does. And as I suspected, industry size is defined in terms of value. Should we therefore be surprised that the correlation disappears when we divide both prices and values by value? The last paragraph of section 4 even brings this up.

What Kliman, Bichler and Nitzan are doing is trying to claim "correlation != causation". But that is only a valid argument if a theory stems from correlation alone. The LTV is a causal theory. To actually deboonk it you have to put forward a competing testable causal value theory that has greater explanatory power. Reactionary economists never do this.

The intent is to show that capital somehow brings something to the table, that value comes from something other than living labour. Hence production functions and other epicycles. Somehow these people are not concerned that theory that maintains the status quo might be "spurious".

There is quite a bit I could write about this topic, but I the lack time. If you want more detailed info you can DM me.

0. The results derived from the statistical LTV do *not* depend on the results of correlation studies.

The statistical LTV (developed in Laws of Chaos and How Labor Powers the Global Economy) is built around baskets B of commodity products: Any basket has an unpredictable monetary price M(B) and a labour content L(B).

The statistical LTV predicts that M(B)/L(B) is well-approximated by a defined constant (average wage/wage share) and that this approximation improves with the size of a random basket B.

This is not premised on a significant correlation between M(B_industry) and L(B_industry), but the latter is rather a consequence of the theory.

1. 'Industry size' is *ill defined* and cannot even be considered as a factor to be "adjusted for".

2. Costs of production are mediating variables between labour content and prices. Adjusting for mediating variables *destroys* causal association.

See any basic book about causal inference.

3. Kliman believes that 'real' correlations remain invariant after prices and labour content are divided by costs. This is *provably wrong* using the properties of the covariance operation.

See https://www.youtube.com/watch?v=9-96kJSFFVI&ab_channel=PaulCockshott

The arguments raised by Kliman et al. have no grounding in causal inference or statistical methodology. They would not stand a rigorous review.

That said, point 0 still holds.

Looking at the video "Critique of Klimans theory of price" around 16:44, does Kliman seriously claim profit rates should be the same despite different organic compositions of capital? This seems like an extremely elementary error. Even with a naïve (monetary) look at this, only in terms of money, one trivially realizes that firms with higher overhead have lower ROI. This should be obvious to anyone with even rudimentary knowledge of accounting.

Aside: it irks me that a lot of these arguments involve specific examples rather than symbolic manipulation.

I also see the same argument I just made, at 28:00 in "Kliman the empirics". Kliman's argument could be made even more silly by "correcting" for the prices in each industry. That is, transforming:

log P ∝ log W

into

log P - log P ∝ log W - log P

or

0 ∝ log W - log P

which is clearly silly. But Kliman would have you believe this is perfectly fine.

I knew about Kliman's proposal that industry size be defined by costs of production and how controlling for this just removing the mediating variables. UE does not even mention Kliman however and his critique of the correlational studies boils down to citing the spreadsheet of Bichler & Nitzan and then addressing the response of Cockshott et al. by just supposing that even if it's not "industry size," the correlations are probably spurious due to some vague "historical factors."

Nor does UE even mention Farjoun & Machover and their probabilistic framework. His critique of the LTV boils down to declaring that labor values are unmeasurable (including an easily falsified statement that socially necessary labor time has never been measured without using wages) and criticizing the correlational studies on vacuous grounds.

If UE had even passing knowledge of Laws of Chaos he probably wouldn't have totally misunderstood what the transformation problem is; he thinks it concerns the price theory of the "simple" LTV rather than prices of production. He does not seem aware at all that the transformation problem is premised on profit-rate equalization and concerns a different price theory.