Can I use a regression here ?

#1
I am doing my thesis. My dependent variable is 'love style' , this however has 6 levels. I have 5 independent variables or 'predictors'. Can I use a regression to predict which IV best predicts the DV ? and if so which one and will I have to deal with each of the love styles as separate analyses ??:confused:
 
#3
My understanding is that yes, you can include "love style" with 6 levels as a predictor, however, you will have to treat each level of "love style" as it's own IV. For example;

If your original table headers look like:

IV1 - IV2 - IV3 - Love Style

You will have to break the love style field out into the 6 options;

IV1 - IV2 - IV3 - LS1 - LS2 - LS3 - LS4 - LS5 - LS6

Now, in each of the LS1 columns you will have a 1 or a 0 depending on which it was. Say "Love Style" was a 1 (or maybe it said thump, I don't know. Anyway, there are 6 options) You would have LS1's value be 1, while all the rest (LS2 - LS6 were 0's). Now say in the next observation Love Style was a 4 (or whatever it might say, gallop for all I know), you would now have 0's in all LS's except for LS4 which would hold a 1 to indicate this is the value of "Love Style". Does this make sense?

You will also have to check for multicollinearity between all IV's and the LS's. Now, when you do the regression analysis if you see a pattern throughout the LS's, then you should be able to say that; "Yes, Love Style IS a predictor", but as far as how much the P value shows this is something you won't be able to get. However, this should be irrelevant as you are only trying to prove or disprove the null hypothesis anyway.

Someone may have a different take on it, but this is how I have been conducting my regressions in this situation.

If you don't change this categorical predictor (Love Style), typically the program running your regression will view it as a numerical indicator of the output variable.
 
Last edited:

Dragan

Super Moderator
#4
I am doing my thesis. My dependent variable is 'love style' , this however has 6 levels. I have 5 independent variables or 'predictors'. Can I use a regression to predict which IV best predicts the DV ? and if so which one and will I have to deal with each of the love styles as separate analyses ??:confused:


Given what you've provided an available option to consider would be (polytomous) logistic regression.

If the the IV's are categorical then another option to consider would be Log-linear analysis.
 
#5
Thanks all, I have it sorted. Apparently, I use Stepwise regression because although I am predicting, I have no specific predictions, its like more an exploratory model. Just to clarify though - my DV is love style (6 of them) my IVs are self esteem (continuous), attachment style (3 categories) and personality (5 categories) and then gender , age, relationship status. Understanding at last ! I think ....