Regression Using Likert Scales (Total Scores)

#1
I conducted a survey that consisted of 5 different likert scales. I totaled the scores for each scale. I would like to know if the total scores of 4 of the scales serve as predictors of the scores on the 5th scale.

I am struggling to determine which regression method, if any, is most appropriate for my data.

Does anyone have any suggestions? I am using SPSS.
_________________________________
- I considered MULTIPLE REGRESSION, which would require me to treat my total score for the outcome variable as a continuous variable, which I know is debatable. However, my stats book says the outcome variable needs to be unbounded, so I am assuming I would violate that assumption since my variable would be bounded (scale ranges from 30-150, but data scores only range from 38-145)

- I am also considering ORDINAL LOGISTIC REGRESSION, perhaps by binning the total scores of my outcome variable.

- Another possibility I am exploring is CATEGORICAL REGRESSION (CATREG in SPSS), which uses the Optimal Scaling approach.

Thoughts? Guidance? Need more info?

Thanks!
Scott
 
#2
By the book, you shouldn't be using OLS regression with ordinal IVs, but the reality is that this is frequently done anyways.

One thing to look out for with survey data, however, is multicollinearity between the IVs. In short, if the IVs highly correlate with one another, the regression would be unable to determine the amount of variance that is explained between each of the IVs on the DV. If you are only looking for a share of impact (i.e. the standardized beta scores), and are not interested in getting a predictive equation, I'd recommend using the Shapley Values Regression. You can run it in R using the "relaimpo" package. The basic function would be calc.relimp(model, dataframe).
 
#3
Ok, thanks. I am indeed mainly interested in the share of impact, so I will take a look at the Shapely Values Regression. I've never used R before, but should be able to figure it out! Please let me know if you know of any good guides or sources that can assist me. Thanks again!
 
#4
R has a bit of a learning curve, but it's easier to use advanced techniques like Shapley Values Regression than with SPSS.

First save your .sav data file as a .csv.

Then install R, and install the relaimpo package.
Type the following in the R console:

mydata <- read.csv("C:\\where my file is located")
library(relaimpo)
calc.relimp(DV~IV1+IV2+IV3+IV4, mydata, type = c("lmg"), rela = TRUE )

The output will be a proportional share of impact of your IVs on the DV.