> On Wed, 11 Apr 2001, Larry Z. Daily wrote:
>
> > Good morning all,
> >
> > I have a colleague who is interested in doing a longitudinal study who asked
> > me about the pros and cons of such a design. We're meeting soon to go over
> > them. I don't really do longitudinal research, but I've been looking into it
> > for her. My question: other than the Campbell & Stanley problems with any
> > one group, pre-test post-test design (e.g., history, maturation) are there
> > any other problems with longitudinal research I should be aware of?
> >
With longitudinal designs you have potential "time of testing" confounds. If
you were tracking political involvement in 10, 20 and 30 year olds, started in
1960, and retested in 1970 and 1980, and found political involvement low in 10
year olds, high in 20 year olds, and low again in 30 year olds, "age" is confouned
with time of testing. Perhaps 20 year olds are naturally politically active. Or,
perhaps there was something different about 1970 (like Vietnam).
You also have to worry about selective drop out. If the less politically
active people "dropped out" of the study between 1960 and 1980, the average
involvment score would rise accordingly. One way to deal with the drop out problem
is to analyze the characteristics of those who dropped out, and see if they differ
from the rest.
The most obvious drawback of the design is the time involved (unless the time
span is a few months).
One strategy to deal with history confounds is to get the measurements more
frequently (e.g. every week, for instance) rather than every month or year. This
permits you to determine (more easily) if a particular outside event affected the
DV.
If you were to use a cross-section, time of testing is held constant, but have
now counfounded age with birth cohort (the generation). My understanding of the
difference is that cohort effects last for a long time, while time of testing are
more transitory (though, logically, cohort effects would have to start as
time-of-testing effects).
Even though cohort effects are held constant in longitudinal designs (you'd use
only one generation), you might hesitate generalizing the results from this one
generation to others - i.e. the baby boomer generation may show a different age
pattern than another generation.
If time and effort permit, you can consider a cohort-sequential design that
permits you to estimate the magnitude of age, cohort, and time-of-testing
confounds. Start with a cross-section, and follow each cohort of the cross section
longitudinally.
--
---------------------------------------------------------------
John W. Kulig [EMAIL PROTECTED]
Department of Psychology http://oz.plymouth.edu/~kulig
Plymouth State College tel: (603) 535-2468
Plymouth NH USA 03264 fax: (603) 535-2412
---------------------------------------------------------------
"What a man often sees he does not wonder at, although he knows
not why it happens; if something occurs which he has not seen before,
he thinks it is a marvel" - Cicero.
