# Which Analysis to use? Likert Scale

I have 3 independent variables (1-5 Likert Scale) questions and I want to check how well these three can predict/explain my DV (1-5 Likert scale)

The three independent variables are: 1. Quality of information 2. Accessibility of staff 3. Quality of technical advice

My DV is: Overall evaluation of service center

All variables are ordinal (1 = low... 5 = high)

Which analysis would be appropriate to run here? I would prefer an easy approach and I think Ordinal Logistic Regression is way too complicated. Can I use a Linear Regression?

Basically, I want to be able to say that (for example) "quality of technical advice" is better at predicting "overall evaluation" than "Accessibility of staff"

Also, I have a 0 value on all variables ("No opinion", so in fact all are measured on 0-5 Likert scale)). How should I treat this variable? Can I replace the 0s with the mean of the observations?

Many thanks!

Fredrik

Likert scales are usually "disagree" to "agree" scales and not "low" to "high". See wikipedia http://en.wikipedia.org/wiki/Likert_scale -

In your case of a low to high scale, no opinion seems a missing value and not a value below "low, so i would argue you should not use a 0 there.

Methodologically I think there is no problem in using a Linear regression. In true Likert scales (disagree to agree) the item can be seen as an interval variable with interval characteristics and quasi-normal distribution (this info also in the wikipedia page), so no problems in using a linear regression. The problem is that your scale may not be a true Likert and therefore you will be in shakier grounds.

• Many thanks Jacques! Actually, the scales are (1-5) as below: Not important - very important, Not Satisfied - very satisfied, & Performs much worse - Performs much better. Does that change your answer? Commented Aug 25, 2014 at 6:28
• Your answers seems Likert like. Given that, the "no comment' "no opinion" answers are valid answers and correspond to the middle of the scale. Ordered Logistical Regression as suggested by @Walter is a more correct method but I think OLS is also applicable. Be aware that you may face some problems such a OLS result that is above your maximum value (2?) This is discussed in this CV question stats.stackexchange.com/questions/92902/… Commented Aug 25, 2014 at 12:55
• Thanks Jacques! I know that the interpretation of the OLS variables will be tricky, but at this point I'm mostly looking for significance and an idea which IV is the stronger predictor. Commented Aug 26, 2014 at 7:17

From the reading that I've done, it seems like an Ordered Logit (Ordered Logistical Regression) would be most appropriate here.

• Thanks Walter. I have also noticed that OLR would be appropriate, but personally I find it hard to interpret the results and to present the results in a straightforwards way. Commented Aug 25, 2014 at 6:27