I use a likert type questionnaire (7 point scale) in order to measure four variables: usefulness, ease of use, ease of learning and satisfaction. Hence in the questionnaire I have four different sub scales and each one has likert items which measure one variable.

I want to find the correlation between the variables. In order to do that, I sum the responses of the likert items for each sub scale and then i correlate the sums using Pearson's coefficients. Is it a good tactic?

  • $\begingroup$ (Sub)scale sum of "Likert" rating items is known as Likert or summative construct. Using it means that you treat each individual item as interval level variable. For, if you want to treat them as ordinal summation makes no sense. $\endgroup$ – ttnphns Jun 27 '15 at 12:33
  • $\begingroup$ Check stats.stackexchange.com/questions/187820/… and I would suggest reading about IRT models. $\endgroup$ – Tim Apr 1 '16 at 8:25

Yes. Sum of the responses of the Likert Item for each sub scale can be used to find the correlation among these sub scales.

You can also find the mean of the responses for each sub scale and them can use them to find the correlation among these variables.

  • $\begingroup$ How would finding sums and means enable finding of correlations? $\endgroup$ – Nick Cox Jan 16 '16 at 12:28
  • $\begingroup$ suppose there are 5 questions (items) to measure usefulness and you have labelled them as item1, item2, item3, item4 and item5. The summing of the responses of these items means that you just add them like item1+item2+item3+item4+item5. If you are using ms-excel then sum the items for first respondent and then copy the formula for other respondents. If you are using SPSS then use the option compute from transform menu. $\endgroup$ – itfeature.com Jan 23 '16 at 12:13
  • 1
    $\begingroup$ That's one way of collapsing several items into one. It says nothing whatsoever about the correlations between subscales: you might as well claim that the sum of 1,2,3,4,5 tells me about the differences between them. $\endgroup$ – Nick Cox Jan 23 '16 at 16:41
  • $\begingroup$ I just gave you an example of a sub-scale not for all. Your all sub-scales (usefulness, ease of use, ease of learning and satisfaction) needs to be same computation. After computing scores (summing the questions of each sub-scale), you can perform correlation between them, regression analysis and some comparisons such as ANOVA. $\endgroup$ – itfeature.com Jan 25 '16 at 3:28
  • $\begingroup$ I suggest that you edit your answer to make clearer what you are proposing. $\endgroup$ – Nick Cox Jan 25 '16 at 8:25

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.