File spoon-archives/bhaskar.archive/bhaskar_1998/bhaskar.9803, message 7

Date: Thu, 05 Mar 1998 16:48:49 +0000 (GMT)
Subject: BHA: econometrics (was Need Help Fast)

Hi again Marshall,
Hope the proposal is successful.

As regards multiple regression I find myself in agreement with your 
assessment. The CR social ontology is essentially one of interacting 
relatively enduring social structures. Therefore in principle  
multiple regression can serve the same purpose as experiment in 
natural science. It can disentanlge interacting causal mechanisms to 
reveal non-actual structural mechanisms at work. A stochastic error 
term is perfectly acceptable in picking up the noise left once all 
relevant structures have been included. Thera are many caveats to 
this concerning structural stability, functional form, 
quantifiability, high degree of inter / intra relationality [more on 
this below], etc. ensuring that multiple regression is nowhere near 
as useful as natural experiment where control over variables is 
possible. But still, the perhaps surprising conclusion must be that a 
clear rationale for multiple regression is provided by CR.

The *actual* existence of regularities must still be ruled out 
according to CR. It would be a volte-face should CR 
allow actual constant conjunctions in the social realm hence simple 
correlations and simple regressions must be ruled out by the CR 
ontology as must actual probability distributions (hence the clear 
distribution of income you refer to appears to be problematic 
for CR). The point is that multiple regression, as opposed to simple 
regression, affords a method (however imperfect) for accessing 
non-actual mechanisms.   

Thus I tend to agree with you that Nick's view that CR would resist 
multiple regression is incorrect. And my previously expressed 
view that no hypothesis testing is allowed for by CR is incorrect. I 
agree with you that hypothesis testing using multiple regression has 
a CR rationale even given Sayer's discussion referred to by Rakesh. 

This view can be related to the many debates on this 
list concerning the transformation of social structures. You and I 
appear to agree with Colin's interpretation of the TMSA which is, I 
believe, that social structures are for the most part reproduced 
rather than transformed through the medium of social agency 
(apoligies to Colin if this is not his view). In econometric terms 
this suggests that 'structural stability', a key requirment for 
multiple regression, is inbuilt into the CR ontology.

I should add that while I agree with your own and, indirectly, 
Colin's view as an interpretation of CR, I have reservations about CR 
itself despite its many appealing aspects. Capitalist social 
structures may be best characterised as being in ongoing structural 
transformation rather than reproduction. And it would be very helpful 
if our concepts of structures could be integrated in a meaningful, 
ordered way so as to see society as a single totality rather than as 
the CR 'plurality-within-(partial)totality'. (CR suggests structures 
are intra-related but I am not aware of any means afforded by CR to 
order such complexity so as to comprehend it. Without this then 
CR may have to *hope* social reality is simple enough to 
comprehend [so possibly to measure via multiple 

Given such a transformational social ontology and integrated 
social structures then multiple regression, if useful at all, 
would be interpreted as measuring aspects of ongoing social 
transformation / development. The usefulness of basic descriptive 
stats would be enhanced also.


Andrew Brown,
School of Economics,
Middlesex University,

tel 0181 362 5512

     --- from list ---


Driftline Main Page


Display software: ArchTracker © Malgosia Askanas, 2000-2005