Hello, I found a survey about attitude toward language in an article which I find fit to use it in my thesis.
The survey has 30 questions, which were put into 5 factors through factor analysis with varimax solution and were calculated the internal consistency estimate of reliability for each factor.
Items with a minimum eigenvalue of 1.0 and that loaded on a factor at a level of .45 and above were kept; items that loaded on multiple factors at .32 or higher or with little theoretical connections to the factor were deleted
The problem is if I want to use this survey, is there anything I can do to validate or make sure it is reliable. I can only think of giving it a pilot test and calculate cronbach's alpha for each item.
For an N, you would typically need 5x the the number of items for this set of scales, which is 150, and thus large for a pre-test. The other alternative is simply to go ahead and use the items in your survey, and hope that you get a meaningful set of scales.
If you do go ahead, the appropriate next step would be to use Confirmatory Factor Analysis (rather than Exploratory Factor Analysis), to see if your results replicate the original findings.
I suggest that you do pilot study (though 150 is large, you could still include this data in your actual study if everything works out) and then perform CFA as suggested by David L Morgan to see if you are able to replicate the results. Then you could increase the number of participants!
Khoa Nguyen, I am a bit suspicious about the scale you are using if the number of factors in it was determined by the number of eigenvalues > 1 (known as the Kaiser criterion) and if the factors were rotated by using varimax.
Neither of these procedures is recommended by methodologists whom I respect and, in my experience, neither of these procedures has been beneficial. Instead, I (and others) recommend parallel analysis for identifying the likely number of factors, and an oblique rotation (e.g., oblimin or promax) for seeing which items belong on particular factors.
If you like, I could post a number of articles that support the above and/or provide additional advice that might be helpful.
I would be inclined to:
a. mistrust the results that you refer to as having been previously obtained with the scale,
and
b. conduct fresh exploratory factor analyses with your own data.
Please also be aware that a coefficient alpha is best calculated for items within a particular factor. Alpha is not calculated for individual items. (I think you might simply have confused the words "factor" and "item" at the end of your question.)
- Badges
- Science topic
- Similar topics
- Social Science