In your other post, you state that these are Likert-type items that
cover Satisfaction in several particular ways. Likert items are
routinely handled as continuous scales. The other modern way to
deal with them might be as the logistic items for Rasch scaling, ets.
It takes a weird sample or a bad job of question design to make
"ordinal" a choice that requires simple rank methods.
It seems to me that you have a set of items that make up a scale,
and also a general question that would summarize it. In this sort of
case, it is usually a foregone conclusion that there were be pretty
good positive correlations among the items and with the summary;
so that is not an interesting issue. "How big?" is the question.
This framework seems to invite the use of the RELIABILITY procedure,
to see how high and how consistently the items do intercorrelate.
Cronbach's alpha gives a measure of how well, if you don't just use
their average correlation. I would also compute the summary of the
items as a simple average of them, and get the correlation of that
with the Overall Satisfaction question.
Using a preliminary set of questions is known to improve the quality
of a summary score; and, sometimes, to focus it on particular content.
("How sick is this patient?" will be affected by which symptoms and issues
you first bring to mind.)
The sort of followup question that comes to mind is the sort that looks
at some *other* sort of criterion. Then, one may ask whether Likert
average and the Overall Satisfaction question perform the same: Do
they correlate the same with this other criterion? -- That is, assuming
that the simple correlations are about the same: Does either of them
add "significantly" to the association that you see for the other one?
--
Rich Ulrich
> Date: Mon, 17 Jun 2013 09:12:35 -0700
> From:
[hidden email]> Subject: Re: Ordinal regression on a data set
> To:
[hidden email]>
> @ Kendall
> First I want to check if there exists any relationship between the IVs and
> the DV. (which I have done using the Kendalls Tau). Again, I would like to
> use the IVs (questions 1-9) to predict the DV (Overall satisfaction last
> question). Can I take the IVs as factors or covariates and should they be
> treated as nominal or ordinals? Have tried to vary all of them but the
> values I get are out of the question for sure. Please advice.
>
>
Locked
