Date: Thu, 24 Sep 1998 18:09:08 +0100
From: Colin Wight <Colin.Wight-AT-aber.ac.uk>
Subject: Re: BHA: Regression paper


Hi Andy, Doug, Wendy, Tony etc.

Andy, I'm not sure why you think Doug's account 

smacks of empirical realism
>(Hans E made this point in discussions some months back). You are 
>introducing 'given' empirical data (of course, as Marshall pointed 
>out months ago data is always theory laden) as variables. Yet real 
>mechanisms are not given in this way according to CR; rather we have 
>to retroduce their existence from the flux of empirically perceived 
>actual events that they produce. 

Isn't Doug suggesting only that the variables used in the regression
analysis are the empirical indicators of strong tendencies at play and
aren't themselves the mechanisms. This doesn't imply that these variables
ontologically exhaust the matter. Nothing in CR denies the validity of
empirical testing of propositions. 

Now to Tony, if I may quote his own text at him, on p. 77 he argues that,
'Of course, even if regularity stochasticism [by which he means a
committment to Humean constant conjuctions, I think] is not a valid
doctrine some constant conjunction of events of the sought-after kind may
come about, at least over some limited region of time-space." 

Given this possibility, isn't regression at least implicitly accepted as a
valid tool? If we have a constant conjuctions in an open systems isn't this
an indicator of a potentially strong set of tendencies in a given causal
complex? A set of tendencies that require explanation. Note that this
differs from the way Tony puts it in his book, because what we are
reasearching now is not the "certain basic conditions" that make such a
constant conjunction possible. But the causal tendencies inherent in the
mechanisms themselves.

Now I'm not saying that regression analysis is perhaps a "good" tool to use
and it maybe that it is altogther too crude, and even in instances where
constant conjunctions do occur in open systems it will still require
supplementation. However, to say that it is not the only tool is not to say
that it is of no use. 

Wendy, I'm interested in hearing more about path analysis.

Cheers,




 alsoexhaust the matter. It is these considerations that 
>prompted the account I posted months ago of the circumstances under 
>which regression is possible given CR.
>
>On the other hand I find your example of bank redlining fairly 
>persuasive. I tend towards the conclusion that CR has certain 
>problems since it finds difficulty in accounting for your example. 
>
>On the point (2) raised by Tony, I agree with you: IF we have entered 
>all relevant mechansims as independent variables (this requires our 
>own construction based on retroduced mechansims as I suggested in 
>earlier posts - NOT the employment of 'given' data) then the error 
>term will be random due to the interaction of upteen minor, 
>non-systematic factors and we will have a CR rationale for multiple 
>regression.
>
>Wendy, I take it your point refers to multicollinearity? If so then I 
>fully agree with you that this is an important problem but it is one 
>that econometricians would readily concede. 'Path analysis' sounds 
>interesting - though I'm not sure I fully understand it. 
>However, would it not have the same problem - 'empirical 
>realism' - as I have suggested Doug's procedure has?
>
>thanks,
>andy.
>
>
>     --- from list bhaskar-AT-lists.village.virginia.edu ---
>
>
------------------------------------------------------------------------

Dr. Colin Wight
Department of International Politics
University of Wales
Aberystwyth
telephone: +44 (0)1970-621769
fax      : +44 (0)1970-622709
---------------------------------------------------------------------------


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