# More than one outcome (dependent) variables in ordinal logistic regression

I want to run ordinal logistic regression (OLR) in SPSS. My data include 6 predictor variable (two continuous and 4 categorical ) but my outcome variables are also 6 (categorical-likert scale). e.g my dependent variable is Business development and 6 likert scale questions were asked for this, increase in profit, sales, size, asset, marketing and labour.

1-If I composite these 6 variable into one by first sum all the 6 variables and then recode them into again five categories like before (In SPSS in Transform, Compute variable and then recode into different variables). I think in this way I lost my original data and may categorize them wrong.

2-If I think of using Categorical Principle component analysis for reducing the dependent variables, it will give results (object scores) in continuous form on which I have to run linear regression. I do not want to run linear regression as my original data is categorical.

3-So option left is to run OLR without combining the dependent variable on each DV. It means that there is no model for Business development and 6 models for profit, sales....

My question is that

• Is it preferable to run OLR with each DV and then summarize the results
• OR Is there any other method for reducing DV and running OLR.

• OR Is there any better method than ordinal logistic regression for this data.

A more flexible approach is fitting an item response theory model which treats business development as a latent trait which is input to the six likert scale manifest variables according to a proportional odds model with a set of intercepts and slopes. The R package mirt is capable of fitting these models. This is a more general approach to combining these measures. The "ability" is a generalized concept of a sum-score which accounts for the fact that some questions may have very low or very high prevalence, and that differences may be of varying weights in that regard. This is a rough, high level explanation, but you should be able to conduct some self-directed research to see if this is a useful approach to combining multiple ordinal measures.