I have data from a 5 point Likert scale item which approximately 1/4 of the responses only account for 1 through 4 of the 1 through 5 available.

Do I have to standardize the min/max response categories that is reported in SPSS descriptives for output or will creating z-scores be sufficient?

The variables have differing degrees of distributions due to this and the transform functions (e.g., SQRT, LG10) have not solved my problem. My sample (N=162) is the first part of a test/retest for instrument reliability. I have run all the processes thinking that what looked normal and based on criterion (2 or 3 times the SD is sufficient to estimate skewness to identify 'normal distribution' regardless of visual).

I'm at the point of frustration and exhaustion...any expert advice?

  • $\begingroup$ I'm sorry, I can't follow your question. It sounds like you're trying to assess normality. Why? You make several assertions in your third paragraph but I really can't tell what you're getting at. What parametric tests are you referring to? $\endgroup$ – Glen_b Feb 27 '14 at 2:22
  • $\begingroup$ You are applying those terms somewhat vaguely. Anyway, Likert scale is more often considered an ordinal scale, rather than interval. $\endgroup$ – Germaniawerks Feb 27 '14 at 8:59

A test of significance on "normality" is often very uninformative. With large samples these tests almost always reject the hypothesis that sample data are from a normally distributed population, but often the degree of non-normality is only small and has no impact on the reliability of the statistcal analyses. Graphic methods such as q-q- plots are often more useful.

Moreover, if you perform analyses such as regression, ANOVA, t-test, then the normality of the variables is uninteresting. Instead, you would have to inspect the distribution of the residuals (prediction errors).

Even more: if you want to perform analyses such as regression, ANOVA, t-test, then with such a large sample size as yours, even a normal distribution of the residuals doesn't matter much.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.