He writes: "Actually, my concern is that it the literature on this
issue
seems to be limited in its statistical sources. For example,
following that
literature (given your own purpose, I see), you use the NIPA data
for
calculating both prices and values. But, do we know how are actually
"cooked" those data?"
Well, most of the studies use I-O data, not national income data.
Alejandro Valle and I are exceptions. I don't think the limited
number of sources is a big problem, given that the studies do use
different data -- different years, different countries, and
different sets of accounts (to some extent), and somewhat different
categories -- but the results keep coming out the same. My results,
for instance, are very similar to everyone else's (actually,
"better"), even though values are computed on the basis of the TSS
interpretation of Marx's value theory, not the Bortkiewicz
interpretation.
How the data are cooked by, for instance, the BEA, is known, more or
less -- if you have the patience to slog through it all. I think
the data are serviceable, though you have to be careful. (For
instance, I first used the "capital consumption" series that is part
of what they call "property-type income" to compute depreciation.
Then I noticed big changes in the numbers in 1981 and 1986. I
called the BEA, and learned that these series are based on tax
returns and are affected by changes in tax laws -- which is the
reason for the changes in 1981 and 1986. So I redid everything
using the "depreciation" series.)
How particular researchers cook the data after that is usually not
very clear, because no journal will publish 20 pages on that.
Ochoa's dissertation is an exception; he tells you everything in
minute detail.
Ale: "What strikes me about LTRP is the dynamics, I think, it would
predict:
Given a wage level, to have higher profits, capitalists should hire
more
people or go to intensive labor branches. Don't you think that this
wouldn't be the world we are seeing daily? So maybe there is
something
"wrong" in the statistics which are being used and, more
importantly, those
statistics are taken uncritically to conduct the tests. (Maybe you
feel
that I'm going out of your original point... which is true!)"
Yeah, the most obviously weak point in this stuff is that there
isn't and never has been any theory backing it up. We're supposed
to believe it because these are the "facts." Besides being crude
empiricism, however, the problem is, as you say, that these supposed
"facts" contradict other well known facts -- "all experience based
on immediate appearances," as Marx put it.
I think the "statistics" are ok, if by this you mean the *data*.
But if you mean the *measures* used to evaluate them, I agree --
they (aggregate correlations, measures of deviation) are fatally
flawed.
The data are my friends. They told me exactly what I had suspected
ahead of time. There was simply no reliable evidence that values
have any influence on prices, once spurious correlation is removed,
and any variable that has the same p.d.f. as the values is as
"close" to prices as the values are.
The problem is rather the rampant misinterpretation of the data by
means of meaningless and misleading measures of value-price
relations. Looked at in one way, in other words, the data falsely
SEEM to support the LTRP. But looked at properly, the *exact same*
data disconfirm the theory.
Ale: "Other point concerning the usual procedures is the constant
rate of
exploitation assumption you refer to in ope-l 436. ...
'[Stiebeling] cites data from the US 1880 Census of Production to
show that industries with high organinc composition tend to have
high rates of exploitation, and viceversa.'"
This is very interesting. I found the opposite pattern. Sectors
with a high ratio of nonlabor costs to living labor tended to be
relatively high wage sectors (this undoubtedly has to do with
monopoly). I mention this in a footnote. It is this phenomenon
that makes prices and values "close" when you assume equal rates of
exploitation ("adjust" the labor-times) instead of counting each
hour of labor as an hour of labor. Sectors that use less living
labor, and thus have low unadjusted values, typically have high
wages, which means that adjustment raises their values. The
labor-intensive, high-value sectors, typically have low wages, so
adjustment reduces their values. So the values all become closer to
one another (in my data set, their variance is reduced by a factor
of 20!).
It is this, not any influence of values on prices, that gives rise
to "small" price-value deviations. When prices and values are
uncorrelated, as they are in my data set and, I suspect, most
others' as well, the average price-value deviation is merely
measuring the dispersion of the data. So, by shrinking the variance
of the values, the adjustment procedure yields markedly smaller
average deviations -- and markedly smaller coefficients of variation
as well. This is a mere statistical artifact, which in no way means
that the adjusted values are better predictors of prices than the
unadjusted ones. They aren't -- there is no reliable evidence that
the values are correlated with prices.
To understand the effect of variance-shrinking, consider a very
simple example. Imagine that all prices = 1.2. So they are
unaffected by values. Whether an industry's value is high or low,
its price is 1.2. Now, the mean absolute deviation of price from
value is the average of abs(1.2 - Vj). If values are spread widely
apart, you'll have large deviations. But if they are all clustered
closely together (1.2, 1.21, 1.19, etc.), then the deviations will
be small. Big deal. In either case, values have no influence on
prices. So the mean absolute deviation and similar measures are
pretty meaningless.
Ciao
Andrew