# What Regression Analysis Should I Employ

#### CA92

##### New Member
Hi, I would really appreciate some help with this.
I am looking to see if employees' organisational history (such as voluntary/involuntary turnover) and their contract type (part or full-time) impacts on their organisational commitment and empowerment.
So I have 2 Dependent Variables (Commitment Questionnaire and Empowerment Questionnaire, both Likert-scales)
And I have 3 Independent Variables (Occupational history (3 levels), contract type (2 levels) and I want to use gender as well, to see this impacts on the difference).

Please anyone help me, either by telling me which Regression to use
Much Appreciated :tup:

#### noetsi

##### No cake for spunky
If you have a likert scale DV than either ordered or multinomial logistic regression is where you want to go. You mentioned that you had unequal error variance on another thread. I think you will find when you look up logistic regression that it does not assume equal error variance (and it is not suprising you have it with a non-interval dependent variable). I believe unequal error variance is inherent or at least common in categorical DV.

#### CA92

##### New Member
Thanks for getting back to me so quickly, just my project seemed to take a serious turn this morning, and wanted to figure out what analysis I should be using so I don't spend the whole weekend worrying.
No problem if you can't, but do you know how I could best decide if I was an ordered, or a multinominal logistic regression? If not, no problem, just stats isn't my strongest suit and I find myself getting more confused the more I google!
Thanks again

#### noetsi

##### No cake for spunky
In general ordered logistic regression makes sense when there is some logical order to your categories so that say level 3 is higher on some dimension than level 1 and 5 is higher on it than 3. Ordered logistic regression makes an assumption about how you could dichotomize the more than two levels of the DV into two levels (which is important in terms of calculation and perhaps theoretically although I don't know the theory behind this). Essentially it assumes that the results don't depend on how you would dichotomize the dv. There is a test of this (sas calls it the Score Test for Preportional Odds Assumption although each software probably has a different name).

The nulls is that in fact the results did not depend on how you dichotomized the DV. You want the null to be true, if you reject it you can not use the ordered logit model as I understand it - you would then try the multinomial model. Paul Allison warns that with many IV and a large sample size you may reject the null too often, but unless you have a lot of expertise in this method I am not sure you have much choice but to not use ordered logistic regression in this case. You can read Allison's comments on the method in chapter six of "Logistic Regression Using SAS" (2nd ed)

Note that I am anything but an expert in ordered regression which I rarely use (I usually have dichotomous data or make it dichotomous which is simpler but not reccomended). You should read some online or book material to follow through on my comments.

#### CA92

##### New Member
Thanks again so much! Currently trying to sort out my undergraduate thesis, and don't remember covering regression in much detail at all. Thanks again for your help