Date: Mon, 09 Mar 1998 16:17:14 +0000 (GMT)
From: Andrew Brown <a.brown-AT-mdx.ac.uk>
Subject: Re: BHA: Realist Methods


I have firstly an elaboration of my objection to simple 
correllation and regression in response to Doug / Howard:

>The crucial difference between our use of 
>correlations and positivism's is that for us, the statistical 
>relations are just evidence for an explanation and not the 
>explanation itself.

> even mere correlation is often a good, first
> piece of evidence as to whether, for example, banks are redlining,
> whether people who have lost welfare coverage are doing worse, or
> whether international inequality is a consequence of capitalist
> penetration.
> 

Everyone on the list would agree with these statements. This maybe
why the point I suggested concerning simple correlation and regression
appears both peculiar and incorrect. However, unless I am missing 
something, then Bhaskar is committed to the view that empirical 
regularities are vary rare - requiring all the unlikely conditions of 
a closure. This would also be something on which everyone on the list 
would agree. But the original two statements strongly suggest that 
empirical regularities in the form of clear simple correlations - the 
architypal 'constant conjunction' ? - are quite frequent. Hence they 
contradict the claim that the social world is open - so if one is a 
Bhaskarian then a there appears to be a problem? I think this point 
is a particular case of the relatively familiar suggestion that 
fundamental RB concepts (open / closed systems, actualism etc.) are 
not precisely defined leading to some problems. Or am I simply 
barking up the wrong tree? Given the absoultely fundamental nature of 
the concepts at stake it seems at least to be a point worth clearing 
up. For example how are we to define open / closed systems if we 
allow empirical regularities to occur in both?

On minor econometric detail: 

> Actually, there is no real
> ontological difference between what is presupposed by, say, chi-square on
> the one hand and zero-order regression or correlation on the other.  Each
> is measuring covariation between two variables -- event regularities. Thus,
> if regression is to be ruled out on that basis, so, too, should chi-square.

As I understand it a multiple regression co-effecient does not 
measure the relation between two 'actual' variables. Rather the 
co-efficient is equivalent to a simple regression of the dependent 
variable against a 'non-actual variable'. This non-actual variable is 
the variation in the relevant independent variable 
*once all other independent variables have been controlled for*. But 
maybe I have misunderstood your meaning Doug?

Hans E raises perhaps the most fundamental 'caveat' - the 
stochastic nature of independent variables. I don't know how 
damaging the problem is - on Hans E's authority it seems to be more 
a fundamental objection rather than a mere caveat. Though this would 
be an objection with which positivists are equally at home.

Thanks,
andy.



Andrew Brown,
School of Economics,
Middlesex University,
Queensway,
Enfield.
EN3 4SF

tel 0181 362 5512


     --- from list bhaskar-AT-lists.village.virginia.edu ---