On Mon, 3 Jul 2000 Robert N�meth <[EMAIL PROTECTED]> wrote:
> Could somebody please advise me in the following problem:
>
> I have a summary score consisting of 5 different items, which are
> definitely not independent from each other.
By "summary score", do you mean you are using as a dependent variable
the sums of rating scores for the 5 different items? If so, how are
the rating scores defined? (And notice that by summing the scores you
are implicitly treating them as though they were, at least to a decent
approximation, of interval-scale quality; insisting on using only the
ordinal quality of the resulting score is then rather beside the point,
not much different from locking the barn door after the horses have been
stolen, to use an epithet from my youth.) ... If not, what DO you mean?
> Each item describes the severity of a given symptom (medical) on a
> scale of absent, mild, moderate or severe.
My initial inclination would be to score each item 0, 1, 2, or 3 (or, if
you prefer, 1, 2, 3, 4) respectively, sum the 5 item scores (thus
producing a response variable whose potential range of values is from 0
to 15 (or from 5 to 20), and apply a two-way ANOVA to see if anything
interesting emerges. (I might consider a 3-way ANOVA on the item scores,
using items as a 5-level factor.) If nothing emerges, I doubt whether
methods that use only the ordinality of the item scores would show
anything at all.
If ANOVA produces interesting results, and if the results lend
themselves nicely to interesting interpretation(s), one may then worry
about whether the results are possibly attributable to having treated
the data as though they were of (approximately) interval quality.
One way of pursuing such worries is to apply a dual scaling analysis
(otherwise called correspondence analysis) to see whether the "best"
scaling of the item scores displays approximately equal successive
intervals. (If so, stop worrying. If not, substitute the scale values
arising from the scaling analysis and repeat the ANOVA(s) on the
redefined response variable(s).)
For advice on dual scaling, consult S. Nishisato's books on the
topic, or consult Professor Nishisato himself:
Shizuhiko Nishisato <[EMAIL PROTECTED]>,
to whom I've copied this message.
> These assessments will be done two times (baseline and end value) for
> each individual, who are assigned to two different treatment groups
> within each center. (Repeated assessments in a parallel-group
> multicenter design using correlated measures.)
I take it you meant, "... assigned randomly to one of two different
treatment groups ...".
An alternative form of analysis for such a design would be an analysis
of covariance, using the "end value" as the response variable and the
"baseline" as the covariate. This is just a different way of looking at
the problem, not necessarily a better way; given the coarseness of your
item scores, I wouldn't expect much in the way of improved sensitivity
to possibly interesting effects, but you never can tell. Do model
interaction between the covariate the the treatment groups, at least for
the initial analysis; if your ANCOVA routine doesn't permit that, use
either a general linear model (GLM) routine, or a multiple-regression
routine, using a dichotomous variable to distinguish between your
treatments.
> Is there any generalisation of the Cochran-Mantel-Haenszel methods
> for this situation?
Sorry, I'm not familiar with these methods.
> One further question is whether somebody could point me to any
> reference for a CMH method for symmetry tests (McNemar or Bowker) or
> in general for agreement statistics?
> Many thanks in advance
>
> Robert N�meth
>
> _____________________________________________
> Robert N�meth
> Focus Clinical Drug Development GmbH
>
> http://www.focus-cdd.de
> Email: [EMAIL PROTECTED]
> Tel: +49 (0) 2131 155 315 Fax: +49 (0) 2131 155 378
> _____________________________________________
> --------------------------------------------------
> Focus Clinical Drug Development GmbH, Neuss
------------------------------------------------------------------------
Donald F. Burrill [EMAIL PROTECTED]
348 Hyde Hall, Plymouth State College, [EMAIL PROTECTED]
MSC #29, Plymouth, NH 03264 603-535-2597
184 Nashua Road, Bedford, NH 03110 603-471-7128
===========================================================================
This list is open to everyone. Occasionally, less thoughtful
people send inappropriate messages. Please DO NOT COMPLAIN TO
THE POSTMASTER about these messages because the postmaster has no
way of controlling them, and excessive complaints will result in
termination of the list.
For information about this list, including information about the
problem of inappropriate messages and information about how to
unsubscribe, please see the web page at
http://jse.stat.ncsu.edu/
===========================================================================
