The theoretical factor structure of a questionnaire is linked to its Likert scale response? If I change the response scale this may affect the structure? For example, instead of using a scale from 9 answers (Original questionnaire), I'm using the same scale but with 5 answers. I can compare the theoretical factor structure of original questionnaire (9 answers) with the questionnaire of 5 answers?
The underlying question is whether modifying the response scale will have an effect on the correlation matrix, which is the basis for the factor structure. In general, modifying the variance is less important here than modifying the covariance (i.e., the pairwise pattern of responses).
My personal feeling is that going from 9 to 7 is not very likely to change how the items correlate with each other, but going down to 5 might.
For a general discussion on how many points to use on a Likert-type item, here is a book chapter by two of the top experts in the field:
http://www.sagepub.com/kumar4e/study/Chapter%209/Questionnaires.pdf
Zed,
in general, if you reduce the number of categories the variance of the items will also be reduced. Therefore one would expect reliability and validity to be somewhat smaller. I would not expect the differences between 9 and 5 categories to be very large.
9 categories may often be too large, respondents generally use between 5 to 7 categories at the utmost and ignore (do not use) more categories. Therefore the factorial structure of the data using 9 categories may be similar to the data using 5 categories.
As use suggest, you could compare the factorial structure of both versions of the questionnaire an then report the differences.
An interesting article is (you can find it via Google):
Preston, C. C. & Colman, A. M. (2000). Optimal number of response categories in rating scales: Reliability, validity, discriminating power, and respondent preferences. Acta Psychologica, 104, 1-15. doi:10.1016/S0001-6918(99)00050-5
Kind regards,
Karin
If you performed factor analysis or item response theory (IRT) on the data using a certain number of LIkert scale responses, you will need to redo the analysis with the new response patterns. It is not always true that people will response the same (relatively) with fewer or more categories. Karin is correct that the variance will necessarily be reduced, but the pattern of responses may be different. There is no general rule, and thus you should recalibrate your tool accordingly.
The underlying question is whether modifying the response scale will have an effect on the correlation matrix, which is the basis for the factor structure. In general, modifying the variance is less important here than modifying the covariance (i.e., the pairwise pattern of responses).
My personal feeling is that going from 9 to 7 is not very likely to change how the items correlate with each other, but going down to 5 might.
For a general discussion on how many points to use on a Likert-type item, here is a book chapter by two of the top experts in the field:
http://www.sagepub.com/kumar4e/study/Chapter%209/Questionnaires.pdf
- Badges
- Science method
- Similar topics
- Mathematical Sciences
- Statistics
- Applied Statistics
- Survey Design