2

We ran a study in which participants were presented with four items (let's call them service requests) and to each of these requests, two user profiles of people who could satisfy those requests were matched. Our participants had to rate each of those matched profiles on four Likert scales, each concerned with different aspects of the matched profiles.

We want to know if factors such as word length of the request, profile bio word length and number of past services listed in the profile are related to how participants rate those profiles. We have multiple observations per participant and the observations themselves are paired in terms of post.

I am assuming we need some kind of mixed effects regression analysis but I'm not sure how to structure the data and/or what model to apply. I'm using R for analysis and, as well as recommendations, a good example to copy as well as pointers to clear tutorials would be helpful.

Robert Long
  • 53,316
  • 10
  • 84
  • 148
Victoria
  • 21
  • 2
  • You presumably want a mixed-effects version of ordinal logistic regression. I have an example here: [Is there a two-way Friedman's test?](http://stats.stackexchange.com/a/161872/7290) – gung - Reinstate Monica Sep 14 '15 at 23:05
  • Thanks, but data are Likert scale, so not logistic. There is debate about treating Likert scales as parametric (e.g., http://www.theanalysisfactor.com/can-likert-scale-data-ever-be-continuous/). But seems fine by this: http://pure.au.dk/portal/files/70360382/SimulationPosterJK.pdf. We're using lmer from R's lme4 package with participant as a group (of 8 observations) and nested request within participant (within participant group, 4 requests, each with 2 grouped observations). I think this is OK. See: http://jaredknowles.com/journal/2013/11/25/getting-started-with-mixed-effect-models-in-r. – Victoria Sep 16 '15 at 03:34
  • Certainly not *binary* logistic regression, but you do need *ordinal* logistic regression. Try reading the linked thread. – gung - Reinstate Monica Sep 16 '15 at 04:00

1 Answers1

1

A generalised linear mixed model for an ordinal outcome is what you need. This can be fitted with the ordinal package in R using the clmm which fits a cumulative link mixed model.

Robert Long
  • 53,316
  • 10
  • 84
  • 148