0
$\begingroup$

I am working on analyzing data from a questionnaire, and I am unsure whether I have chosen the correct analysis for my research. I have chosen regression correlation, but I might be wrong.

Any input is much appreciated, as I am currently stuck.

I have 31 variables. One asks for self-reported success on a task. The other 30 are statements about specific factors connected to the environment where the task was executed.

I am looking for correlation; is high/low success rate (agree/disagree) correlated with high/low values(agree/disagree) in the factors.

Data is from 4 point-Likert scales (strongly disagree, disagree, agree, strongly agree) and divided into two groups (agree and disagree), and coded 1 and 2.

Is it correct to use correlation matrix (jamovi) and Spearman's rho for this analysis? Spearman (non-parametric) chosen as the variables violate normality.

Improve this question
$\endgroup$
5
  • 1
    $\begingroup$ Likert data are ordinal categorical. Perhaps treated as if numerical interval. For example comparing two Likert variables with two-sample Welch t test instead of Wilcoxon rank sum. Unless you fully understand and agree that this requires the gap btw 1 & 2 to be similar in impact to gap btw 3 & 4 on a 5-point Likert scale you may risk making 'false discoveries' treating Likert as if numerical. Somehow it seems to me the perils of pretending Likert data are num. are much greater when using them in a regression. // On this matter you'll need advice from someone more comfortable with the pretense. $\endgroup$
    – BruceET
    Apr 1, 2021 at 15:06
  • 1
    $\begingroup$ one independent variable and 30 dependent variables: could be, but I would guess that you have the terms the wrong way round. dependent should refer to an outcome or response that you want to explain or even to predict and independent anything else that can be used to help do that. One of several reasons why these terms are widely disapproved and (very slowly) fading away is that it's all too easy for people to get them the wrong way round. $\endgroup$
    – Nick Cox
    Apr 1, 2021 at 15:06
  • 1
    $\begingroup$ I regard the comment of @BruceET as signalling caution like my answer. $\endgroup$
    – Nick Cox
    Apr 1, 2021 at 15:08
  • $\begingroup$ " is high/low values in independent correlated with the same values in the independent" Correlation is not a matter of whether two variables have the same values. For example, correlation between height and weight could be of interest, but it is impossible for a person's height to equal their weight if only because the units of measurement differ. It is hard in practice to get a strong positive correlation between Likert scales unless many people score the same on both scales, but that's because the variables have the same ranges, not because correlation is measuring agreement. $\endgroup$
    – Nick Cox
    Apr 1, 2021 at 17:49
  • $\begingroup$ Thank you for this reply! I have edited the question a bit. Hopefully, it is more clear now. $\endgroup$
    – Jasmine
    Apr 1, 2021 at 17:50

1 Answer 1

Reset to default
1
$\begingroup$

A variable with four distinct possible values can't possibly be normally distributed even if the four distinct values are regarded as (equivalent to) interval scale measurements, which is what any normality test is doing on your behalf. It has a (very) discrete distribution.

What does homogeneity mean precisely in your case?

The bigger deal is whether

  1. Pearson correlation makes sense for variables of this kind, on which different researchers will give views all of the way from possibly helpful in practice to utterly wrong in principle, given the measurement scale. (Why is normality thought to be needed any way? For inference about Pearson correlation, which is possible without invoking normality, say by bootstrapping or permutation tests.)

  2. Spearman correlation ditto, but here opinion would generally be more benign, although still with a big red flag about the difficulties caused by many ties.

jamovi is a program name. I've never used it. It may or may not allow all the analyses mentioned here.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

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