Katherine,
Yours is hard. It needs something like a tobit regression, except
that's for categorical/continuous regression, where you first have to
pass a threshold to do something, and then you can count how many
times you do it. E.g. for smoking cigarettes, you first model the
probability that they smoked ANY cigarettes, and then model how many
they smoked, given that they smoked any.
Jamie, yours is easy. Well, the answer is easy. Doing it isn't. You
can use a multilevel multinomial logistic regression. For this sort
of data, categorical with three outcomes, and two groups, you can
usually use multinomial regression. When you have people measured on
multiple occasions, you can do a multilevel model (you can always do a
multilevel model instead of a repeated measures t-test or anova, for
example). You just need to combine them both. However, multinomial
regression is quite hard, multilevel models are hard, and when you
have multilevel multinomial models, it's more like hardsquared than
hard+hard. (You come across terms like adaptive quadrature and markov
chain monte-carlo when you try to do this sort of thing). (You might
be able to use a generalized estimating equation approach, or a
sandwich estimator instead).
If possible, for both of you, I'd suggest you find someone who knows
about this stuff, and talk nicely to them. Specifically, Jamie, see
if you can track down Tim Croudace. That might be hard, but he can
probably point you in the right direction.
Jeremy
2009/4/9 Katherine Sang <[log in to unmask]>:
> Dear Jamie and all,
>
> I am facing a similar dilemma - I have surveyed respondents before and after
> an intervention (matched pairs). Respondents are asked to indicate if they
> use a particular item, with three possible responses (no, yes for less than
> an hour, yes for more than an hour). The data is non-parametric... I was
> thinking a Wilcoxon Matched Pairs test...can anyone tell me if I am right or
> barking up the wrong tree??
>
> Thanks,
>
> Kate
>
>> Date: Thu, 9 Apr 2009 10:45:06 +0100
>> From: [log in to unmask]
>> Subject: Repeated measures categorical data analysis
>> To: [log in to unmask]
>>
>> Sorry if this is a simple question but I have designed a study in
>> which I gathered some incidental, opportunistic data, without really
>> thinking properly as to how I could analyse it. Consequently, I am now
>> stuck!
>>
>> 32 participants (16 in two groups) answered all the same 22 questions.
>> Subsequent to answering each question, participants indicated whether
>> their answer was based on a guess, a feeling or a memory. Each
>> participant provided 22 of these categorical ratings, thereby
>> presenting each participant with a percentage with which they chose
>> each of the three ratings. I thought initially, I could analyse the
>> data by using a mixed analysis of variance with 1 between and 1
>> within. However, of course, the observations are not independent of
>> one another. Moreover, they average out to 33.33% across the
>> within-subject. It doesn't seem like a chi-square test is a good idea
>> either though, as each of the 22 for each subject would be more
>> related to one another than any of the other observations. Does anyone
>> know what analysis I can use - or is it only possible to explore this
>> data?
>>
>> Thanks for any help - it is much appreciated!
>> --
>>
>> Jamie Brown
>> PhD Candidate
>> University of Cambridge
>> Dept. of Experimental Psychology,
>> Downing Street
>> Cambridge, CB2 3EB
>>
>> Phone: +44 1223 765 206
>> Email: [log in to unmask]
>> Website: http://www.psychol.cam.ac.uk/lara/ and http://www.implab.org/wiki
>
> ________________________________
> Windows Live just got better. Find out more!
--
Jeremy Miles
Psychology Research Methods Wiki: www.researchmethodsinpsychology.com
|