Regression with Likert Scales

Question

My study

I have seven independent variables (likert scales 1-5) who are skewed to the right. I have one independent variable (likert item 1-5) which is skewed to the left. With SPSS (24) I want to measure what influence the independent variable have on the dependent variable.

I have read multiple articles that discuss the analysis of Likert based data, but can't find a test that fits with my data. I have tried Ordinal Regression, but seeing as I get very high numbers I believe the results are biased. (https://i.sstatic.net/MFkxa.jpg)

Questions

1) Could the problem be that my independent variable are scales and thus have a wide range of answers, because they consist of combined items? And if so, do I have to transform them?

2) Or do I have to tranform the dependent variable, seeing as it is left skewed?

3) Are there other ways to measure regression for Likert scale variable?

Thank you very much in advance. If you need additional information, don't hesitate to ask me.

Here are two graphs, visualizing my independent and dependent variable: https://i.sstatic.net/8qqHF.jpg

Answer 1

I see that your dependent variable ("High Urgency WhatsApp" or "High_Urcengy_WhatsAp") has 5 levels, 1-5. You have posted results marked "OLS"; the standard meaning is Ordinary Least Squares, which is linear regression, but your output indicates that you have used ordinal regression with a logit link. I think you need to clearly distinguish between OLS regression and ordinal regression. Perhaps you will want to try each and compare results--that would be reasonable. People endlessly debate the wisdom of treating a 5-level rating as an interval-level variable for the purpose of linear regression. How you handle this may depend on the preferences and standards of your audience for this research. You can also read up on the arguments in many threads on this site if you search for "Likert" and sort by number of votes.

You will not find a transformation for this dependent variable that produces a normal distribution, if that is what you are after. And for "Mean Score ISE" the skew can probably only be removed by creating 3 categories and thus sacrificing detail. But more important (if you are using OLS) is the degree to which the regression residuals are normal. If you are using ordinal regression, I see little justification for any transformation at all.

You also refer to "bias", "the problem", and "high numbers", but it is not clear from your post what you mean by each of these.

Kudos for recognizing the difference between Likert items and Likert scales.