Hi everyone,
What does it mean when a subscale does not correlate with the total score on the full scale? How does one interpret such a finding?
Thank you.
Hello Shaun,
I suspect there could be several explanations. One obvious one is that the overall scale has two or more separate domains (factors / subscales) within it, and therefore any single one of those domains, by mere dint of comprising a separate factor from the other(s), might not correlate very highly with (m)any other domain(s). So, when you obtain a total that comprises scores from separate domains (a practice that I regard as pretty suspect in many cases), who knows what the total means or how it might be "obliged" to correlate with its separate components.
Of course, it also depends on what you mean by "not correlate". For example, is .75 an indication of lack of correlation in the kind of situation you are considering? In my experience, different researchers have different views about that, sometimes depending on which barrow they want to push.
Hi Shaun Hiu ;
your factors (subscales) may be orthogonal. In this situtation, subscale scores will not correlate with the total score. To investigate this, you can use bi-factor modelling. But first of all, you have to conduct exploratory (or you also know the construct of the scale ==> confirmatory) factor analysis. Then, you can get an idea about the factor structure.
Bests
You will need to provide a lot more information to make better understand the problem. I would report the sample size, total number of indicators of the overall scale and the subscale, the specific items of the scale and subscale, and univariate and bivariate summaries.
Many people make the mistake of creating scales from formative when they are using a modeling approach that assumes they are reflective indicators (see the classic measurement article by Bollen and Lennox (1991) for a detailed discussion of this issue). If the indicators and analyses are properly specified, you may have evidence that the underlying theory is incorrect and the measure may need revising. I'm also curious about the observed correlation. I'm doubtful that they are completely uncorrelated, as most social science constructs have some degree of inter-correlation.
Article Conventional Wisdom on Measurement: A Structural Equation Perspective
That said, I would first start with ensuring that you have a conceptually sound instrument. I would look at the conceptual framing of all the indicators and provide univariate and bivariate summaries.
- Badges
- Science topic
- Similar topics
- Psychology
- Psychometrics