From: "Marshall M. A. Feldman" <Marsh-AT-uri.edu>
Subject: RE: BHA: RE: cr and social science
Date: Wed, 6 Feb 2002 21:11:17 -0500


Andrew,

I'm not sure I understand your question, but on the surface here are some
answers. If we're talking about traditional experimental design, the
researcher creates a random assignment: individuals are assigned at random
(i.e. with equal probability) to different experimental groups. So, the
probability distribution is the one the researcher used for the assignment.
In survey research, the same answer applies, only this time it's related to
the sampling procedure. If, for example, we are studying businesses, we
might select firms from a list of all businesses at random, where the
randomness is generated by some random number generator. The sampling
procedure would not favor some firms over others, so the sample would be
called "random." Again, the probability distribution is one the researcher
used for selecting the sample.

I think both approaches offer good justification for assuming randomness and
the specific probability distribution. I don't see why we have to assume
closure of the experimental situation itself.

Of course, there's a huge literature on both subjects, and we can dream up
many complications and ways to address them. However, I don't think the
issue is that we're assuming randomness here; the researcher is actually
doing things at random according to a probability distribution of the
researcher's choosing. The issue might have to do with identifying the
population or considering/ignoring certain characteristics of the subjects.

BTW, I don't think this is unique to social science or statistical research.
When physicists study cloud chambers, don't they also assume "randomness"
(stochastic features, actually -- in other words, that probability rather
than certainty is at work) when they (1) consider sub-nuclear particles as
having a complex, probabilistic relation between their location, mass, and
energy ala quantum mechanics, (2) the particles collide and follow paths
with certain probabilities, and (3) observation is also subject to "noise"
so that there is probabilistic error in measurement (stochastic measurement
error). Even when I was a first-year college student doing experiments in
classical mechanics, we used to make multiple observations and use averages
of our measurements as the best measure. The underlying assumption was that
there was some random measurement error. (Remember, "random" means
probabilistic; it does not mean with no pattern). Also note that in the
quantum mechanics example all three sources of randomness are independent of
experimental closure.

	Marsh Feldman

-----Original Message-----
From: owner-bhaskar-AT-lists.village.virginia.edu
[mailto:owner-bhaskar-AT-lists.village.virginia.edu]On Behalf Of Andrew
Mearman
Sent: Wednesday, February 06, 2002 8:38 PM
To: bhaskar-AT-lists.village.virginia.edu
Subject: RE: BHA: RE: cr and social science



Doesn't randomisation imply closure, simply by
*assuming* that the sample *is* random?  What is the
basis for assuming such a probability distribution?

Andrew


--- Marshall Feldman <marsh-AT-uri.edu> wrote:
> Hi all,
>
> Tone makes an excellent point. I don't think the
> prevalence of randomization
> in biomedical research is a problem for CR because
> these studies are not
> designed to uncover causal mechanisms. Instead,
> they're designed to
> determine the effects of a drug (rather than how the
> drug causes those
> affects). This seems legitimate but superficial.
> Without knowing the causal
> mechanisms, we can only hope our luck holds out
> using the drug in the future
> (the problem of induction).
>
> The whole issue of randomization raises some other
> interesting questions.
> One is whether or not randomization is a viable way
> to achieve closure.
> Although it does not close the experiment by
> eliminating confounding causes,
> it may make these causes cancel out so that we can
> say the experimental
> manipulation is "quasi-closed" (i.e., the observed
> effects are only those of
> the manipulated causes). This doesn't bring us
> closer to uncovering
> mechanisms or necessity, and perhaps the most
> damning aspect of the
> literature on experimental design for social and
> behavioral science is its
> over-emphasis on effects and quasi-closure coupled
> with its almost complete
> blindness to questions of how we study mechanisms
> and the structure by which
> things necessarily achieve their causal powers.
>
> The issue also brings up the question of random
> sampling, which aims to
> obtain "representative" samples. It too ignores
> mechanisms. This question,
> however, raises questions not only with
> experimentation in the social
> sciences, but also statistical analysis.
>
> 	Marsh
>
> 	Dr. Marshall Feldman, Associate Professor
> 	Department of Community Planning and Landscape
> Architecture
> 	94 West Alumni Avenue, Suite 1
> 	204 Rodman Hall
> 	The University of Rhode Island
> 	Kingston, Rhode Island 02881-0815
>
> 	Tel.	401.874.5953
> 	Fax	401.874.5511
> 	Email:   marsh-AT-uri.edu
> 	http://www.uri.edu/cels/cpl/marsh.html
>
>
> > -----Original Message-----
> > From: owner-bhaskar-AT-lists.village.virginia.edu
> >
> [mailto:owner-bhaskar-AT-lists.village.virginia.edu]On
> Behalf Of Tone
> > Skinningsrud
> > Sent: Tuesday, February 05, 2002 8:00 AM
> > To: bhaskar-AT-lists.village.virginia.edu
> > Subject: Re: BHA: RE: cr and social science
> >
> >
> > Dear Fredrik, and others,
> >
> > To adress your last question first:
> > randomisation (i.e. randomised experiments), as
> far as I know, is an
> > experimental procedure used specifically in the
> social (and psychological)
> > sciences. The rationale for this procedure,
> according to a
> > leading theorist
> > in experimental and quasi-experimental
> methodology, Donald T. Campbell, is
> > to enable the experimenter to assess the effects
> of different treatments
> > completely independent of the prior status of the
> object (persons etc)
> > under study.
> >
> > This rationale seems to be consistent with a
> Humean, i.e. contingent,
> > conception of causality which seeks conjunctions
> of events - correlations
> > between impact and result, regardeless of the
> internal structure of the
> > object being studied. This rationale is not
> consistent with a
> > conception of
> > causality as the necessary relationship between
> events, which
> > would make us
> > interesed in studying how objects with certain
> properties react
> > to specific
> > (adequate) treatments. Thus, Campbell's idea of
> the randomised experiment
> > is quite different from Bhaskar's definition of
> the experiment, which you
> > quoted in your article in 'Hften fr kritiska
> studier' - which emphasises
> > the purpose of releasing and studying a specific
> type of mechanism in
> > isolation.
> >
> > I see the rationale of the randomised experiment
> as contrary to basic
> > assumptions in critical realism, specifically the
> conception of causality.
> > One does not get closer to isolating the operation
> of one mechanism by
> > disregarding (or randomising)the properties of the
> objects under study.
> > And, by the way, Piaget did not use randomised
> experiments. He just
> > experimented with his own children.
> >
> > I find the theoretical reasons for rejecting
> randomised experiments
> > convincing, however, it is a dilemma that a large
> part of medical
> > (pharmaceutical and clinical) research is based on
> this methodology, as
> > well as certain psychological disciplines. May be
> each experiment would
> > have to be scrutinised to see whether it is
> designed to reveal mechanisms
> > rather than correlations.
> >
> > Tone
> >
> >
> >
> > At 22:16 04.02.02 +0100, you wrote:
> > >Hi
> > >
> > >My interpretation has always been that
> experiments are imposible in the
> > >social sciences only beacause of practical and
> ethical reasons,
> > and that we
> > >therfore have to wait for sittuation where there
> are reasons to
> > belive one
> > >or a few mechanisms are dominant, such as crises.
> But my
> > interpretation may
> > >be wrong for subjective reasons.
> > >
> > >Then there is the question whether there are any
> example of closed
> > >experiments at all? (Isnt it therfore we use
> randomisation ;-)
> > >
> > >Best,
> > >
> > >Fredrik
> > >
> > >
> > >     --- from list
> bhaskar-AT-lists.village.virginia.edu ---
> > >
> > >
> >
> >
> >      --- from list
> bhaskar-AT-lists.village.virginia.edu ---
> >
>
>
>
>      --- from list
> bhaskar-AT-lists.village.virginia.edu ---


__________________________________________________
Do You Yahoo!?
Send FREE Valentine eCards with Yahoo! Greetings!
http://greetings.yahoo.com


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



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