Date: Mon, 9 Mar 1998 15:41:58 -0500
Subject: BHA: Realist Methods


Thanks Andy and Howard for your responses to my post.

Andy,

First, let me say that I am no methods expert but as I have just finished
running scores of statistical tests -- including multiple regressions -- on
our new welfare legislation, I hope I am not an inconsistent realist.  I
still don't think I am.

I also apologize, Andy, if my reply was somewhat confusing because I was
also replying not to what you had said but what others had said.  I think
that applies to the point about chi-square. I, too, think the issues raised
are important for us to collectively clarify.

For my part, I do not think that Bhaskar is committed to the view that
empirical regularities are very rare -- not empirical regularities per se.
If Bhaskar is committed to that claim, then, it seems to me, he is
obviously wrong.  There are many empirical regularities.  The sun rising
every 24 hours, all sorts of tendencies associated with demographics, etc.

What Bhaskar is committed to, I think, is the view that in an open system,
there are no _invariant_ regularities -- nor even statistical regularities
with invariant probabilities associated with them.  Yet, it is such
invariances that are needed for regularities to support explanations
conforming to the covering law model.  With regard to regularities of this
form, I think Bhaskar is totally correct:  They are rare if they exist at
all and do not in any case constitute an explanation.

Invariant regularities can be contrived in closed systems.  We construct a
micro-world in which only certain mechanisms operate without interference
so that what they produce is invariantly regular. But in open systems,
regularities occur too.  They are just not invariant regularities.  But it
is on that quality of invariance that positivism rests.  It is crucial.

I am intrigued by the account of multiple regression that you and others
have advanced. I agree that in multiple regression, we are examining the
contribution of each independent variable on the dependent variable,
controlling for all other independent variables in the equation.

Although I have never thought about this in these terms before, my initial
response is to disagree that the effects of those controlled variables are
non-actual.  In fact, what the regression equation is controlling, it seems
to me, is precisely the actual effects of the controlled variables.  If
there were no actual effects of these variables, the regression equation
would not pick them up.  Insofar as the regression equation does pick them
up, they are not only actual but empirical as well.

Think of this in terms of vector forces.  The regularity or non-regularity
that results at the zero-order level is due not just to the independent
variable but as well to the vector sum of all those other forces operating
simultaneously.  Since they are operating, they are actual.  What is _not_
actual, perhaps, is the regularity that would have resulted were those
actual effects not operative.  The real, non-actual effect is what the
regression equation displays. So it isn't the effects of the controlled
variables that is non-actual but the regularity that would have resulted
from the isolated effect of the independent variable of interest.

In the kind of analysis I've just been running, I want to see if there is a
relationship (a regularity) between the cut-off of welfare and quality of
life.  Any bivariate statistical test, including zero-order regression,
tells me that.  Then, I want to see if this is a real effect or a spurious
correlation.  To test this, I use multiple regression or two-way ANOVAs,
etc.  But the bivariate relationship is still important to establish first.

No bivariate relationship established, however, will be invariant.  It can
be counteracted or mitigated just as we know.  And, I say again, the
statistical relationship is not the explanation but only evidence for an
explanation.

I am not familiar with the criticism of regression that Hans raises, but it
does not seem to me a critique distinctive to a realist point of view.  As
you suggest, it is a criticism that even positivists would acknowledge.  My
own realist response to those criticisms is to acknowledge their merit and
the fallibility of all knowledge and techniques and to ask what alternative
technique works better.  I think Hans mentioned something, but I don't
think we in sociology are very familiar with it.

Thanks again, Andy, for the comments.

doug











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