Help with regression and related nominal values

#1
Hey all. I'd like some input on a regression analysis. I have data that measures the reactions of individuals to a number of objects and to a number of questions. It looks something like this (Q=question):

Code:
      Q1   Q2   Q3...
Obj1  45   23   52
Obj2  65   67   02
...
I have responses like this for many subjects so it actually looks like this:

Code:
      Q1   Q2   Q3...
Obj1  45   23   52
Obj2  65   67   02
Obj1  87   03   12
Obj1  65   32   73
Obj2  12   54   94
...
I want to run a multivariate regression using Q1 as a dependent variable and the other questions as IVs in order to rank the predictive quality of those independent variables. How do I do this without incorporating the "object"? The response values are related to and vary by those specific objects, but I can't include them in the regression as they are nominal values.

Any help would be appreciated. Just FYI, I'm using R for analysis.
 

Junes

New Member
#2
Hi and welcome.

Some context would be good I think. What are the "objects"? Are they all instances of the same class? How many of them are there?

I think you need a multilevel/mixed model approach. I found this nice starter in R, though it's low on theory.
 
#3
Hi, and thanks for looking at this. The objects are just images of different situations. They span the gamut from desirable to undesirable so the reaction value will vary depending on what question is being asked (e.g., "How much do you like this?" versus "How much would you like to avoid this?"). There are just over 100 of these situations.
 
#6
I did look into this, Junes, and I think this is helpful. The GLM method in R does allow for variable grouping so that I can include that as a modifier to the other independent variables. Thanks for your help.

Hi and welcome.

Some context would be good I think. What are the "objects"? Are they all instances of the same class? How many of them are there?

I think you need a multilevel/mixed model approach. I found this nice starter in R, though it's low on theory.