I hesitate to chime in here, because I agree with most of what has been
said. Also, as my name has been invoked as an authority of sorts, I am afraid
of muddying the water. But if I can step back and speak without trying to be
"authoritative," it seems to me that all of the distinctions we make-- between
experiments, quasi-experiments, nonexperimental work, etc.-- is part of an
attempt to bring some order to a wide range of activities. The categories are
somewhat fuzzy, in my opinion. Certainly randomization and manipulation are the
gold standard. But it is hard to argue that (e.g.) psychophysical research
using one or two subjects (in a within design) isn't true experimentation.
One not-so-minor point: The purpose of randomization is not to produce
groups that are approximately equal--matching can do a much better job. 1)
Randomization is the only method that guarantees that there is no confounding
except by chance, and 2) the probability that differences between groups occured
by chance can be assessed by inferential statistics (which assume the random
model).
best
don
Donald McBurney
University of Pittsburgh
jim clark wrote:
> Hi
>
> On Wed, 26 Feb 2003, Hatcher, Joe wrote:
>
> > I would argue that random assignment is only a means to
> > an end, the end being having at least two groups that are
> > assumed to be roughly equal on all variables. Seen that way,
> > a within-subjects design is simply another means of achieving
> > the same end. I understand the limitations, but would argue
> > that with appropriate counter-balancing these can be
> > overcome. This may be an argument of convenience for me or a
> > result of dissonance-reduction on my part; we have a very
> > small subject pool and strongly encourage within-subject
> > designs where possible
>
> I would agree with Joe here. Moreover, within-subject designs
> are certainly the dominant design in many areas of psychology
> that have a (perhaps undeserved according to Mike?) reputation as
> being experimental (e.g., cognitive, perception, ...).
> Within-subject designs also generally increase the power of the
> design/analysis, as the error term is usually smaller when
> individual differences in subjects are reduced.
>
> > > From: Mike Scoles
> > > I'm going to stick with Sir Fisher and reserve the term
> > "experiment" for > situations where there is random
> > assignment to conditions. I do not know > of > any
> > within-subjects designs that would not be better as mixed
> > designs. > Within-subject designs are too easily compromised
> > by history, maturation, > instrumentation, attrition, and
> > (sometimes) test sensitization and > regression issues.
> > Let's see, the only one of the "Big 7" that I left out > was
> > subject selection--the major problem with quasi-experiments.
> > Of > course, > good quasi-experiments can provide information
> > as useful as a marginal > experiment.
>
> If you think of it as random assignment of conditions to subjects
> (rather than the reverse) then within-subject designs can be
> accommodated. That is, it is possible make the conditions
> orthogonal to many (all?) potentially confounded variables in
> within-subject designs.
>
> I would also extend Mike's last statement even further.
> Non-experimental designs can also provide information as useful
> as (or more useful than) marginal or poorly designed experiments.
> Many students appear to believe (wrongly, I think) that an
> experiment necessarily means causal inferences are warranted,
> whereas in fact only well-designed experiments permit strong
> causal inferences.
>
> Best wishes
> Jim
>
> ============================================================================
> James M. Clark (204) 786-9757
> Department of Psychology (204) 774-4134 Fax
> University of Winnipeg 4L05D
> Winnipeg, Manitoba R3B 2E9 [EMAIL PROTECTED]
> CANADA http://www.uwinnipeg.ca/~clark
> ============================================================================
>
> ---
> You are currently subscribed to tips as: [EMAIL PROTECTED]
> To unsubscribe send a blank email to [EMAIL PROTECTED]
---
You are currently subscribed to tips as: [EMAIL PROTECTED]
To unsubscribe send a blank email to [EMAIL PROTECTED]
