If I am using factor scores (Anderson-Rubin scores to be precise) as independent variables in multinomial logistic regression, do I still need to do a Box-Tidwell test (IV x LnIV) to meet the assumption of linearity?
If I still have to do the Box-Tidwell test, how do I cope with the issue of log transforming the factor score (as these range between -2 to +2). If I add 2 or 3 in the factor score IVs, will that influence my regression results.
Thanks.
Thank you so much Jos, that was very helpful and simple as well. Made my life easy.
Highly appreciated.
Instead of using factor scores you could also use factor based scores, i.e. the mean od the variables loading high on the factor in question. In this way the scale will be similar to the scales of the relevant variables.
You could also transform the factor scores to Likert type scores, e.g. 7-point scores (or any other scale):
min = lowest factor score
max = highest factor score
Likert7 = 1 + (((factor_score - min)/(max-min))*6)
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS