I have measurement items like (just as an example):

"Having taken medicine A ....

  • I felt that I am feeling better
  • I felt that it helped me
  • I felt that it is necessary to take it again
  • I felt that I might need it

Each of the four items were measured on a 7 point likert scale with "1 = Not at all" ... "4 = Neutral" ... to "7 = Very much". How can I merge them into one construct (named for instance) "effect level"? And how can I check for reliability?

I am using SPSS if that is necessary.

  • $\begingroup$ When you're talking about reliability, I assume you're referring to internal consistency/reliability rather than, e.g., test-retest reliability? $\endgroup$ – Ian_Fin Oct 5 '16 at 8:15

The quick-and-dirty solution would be to calculate the sum or average of the items and use this value as the aggregate. You could then calculate Cronbach's alpha as a measure of internal consistency (available in SPSS from Analyze > Scale > Reliability). The problem with this solution is that it treats all items as equal and assumes that all of their variance is variance you are interested in. However, in reality, the items are not equal and some are more related to your latent variable of interest than others. There is also likely some variance in each item that you are not interested in - variance unique to that item or due to measurement error. For this reason, a better solution would be to use a latent variable model such as factor analysis or its generalization, structural equation modeling. However, this is more complicated and will require some learning.

| cite | improve this answer | |

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.