# Validation of a scale for a different population (CFA)

I'm currently adapting a coping questionnaire which has been widely used in sports to be used with a different population (musicians), so I need to address its reliability and validity with this new population. I'm a novice in statistics therefore any advice will be very helpful.

The questionnaire has 39 items on a 5 point Likert scale (1=not at all; 5= very much). In sports population, EFA and CFA have shown a 10 first order factors and 3 second order factors. Since I'm unsure if in the new population (musicians) those 10 factors will be the same, I'm wondering how I should proceed to test validity and reliability. My questions are:

1.Should I calculate the Cronbach's $\alpha$ for every 10 first order factor from the original questionnaire in order to test internal consistency (and perhaps delete some items) and then proceed with a CFA? Or should I run an EFA with all 39 items?

1. Once those X first order factors have been established, will I just need to run a CFA to confirm (or not) that they load in a 3 (perhaps just 2) second order factors?

2. I'm collecting my responses to the questionnaire right at this time, and I'm expecting to get between 100-120 observations. Is this sample enough to perform those tests or would I need a larger one?

• if you are wanting to test if the structure from sports is the same in the musicians id use a CFA. Test model fit by looking at coefficient estimates and model fit statistics. But 120 observations will not be enough to fit 10 factors and and 39 items with ordinal data. Commented Mar 10, 2013 at 21:17

A confirmatory factor analysis would be used to test whether the questionnaire had the same structure among musicians as among athletes. EFA would be used to see what the factor structure is among musicians. Either might be right, but I think the latter is more likely what you want.

But before doing any FA, I'd look at item statistics. Perhaps some items that worked well with athletes do not work with musicians. Are there any items that get very skewed responses? They may not be contributing much. I would also look at all the correlations (39x38/2 of them). I might even make a huge scatterplot matrix to look at all the scatterplots.

Then I'd do EFA.