We have a questionnaire, which have many questions on 6-point likert scale. So these variables are ordinal, not normally distributed. In performing factor analysis, there are two major methods in extracting factors, one is maximum likelihood and the other is principle axis factors. The maximum likelihood is said to perform better than PAF, but in our case the item variables are ordinal (1-6 likert scale), will this be okay to use maximum likelihood, or better to use PAF?
$\begingroup$
$\endgroup$
1
-
$\begingroup$ A description of different factor extraction methods. All of them are linear FA that is for interval scale data, and ML method assumes multivariate normality. But we often treat likert-type rating scale as interval. If you need to treat it as ordinal, here is a number of alternatives to classic FA. $\endgroup$– ttnphnsCommented Jun 29, 2016 at 22:54
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
Although late , i hope this will help : https://www.tandfonline.com/doi/abs/10.1080/02664763.2011.610445?journalCode=cjas20 The FACTOR stand alone factor analysis program ( Urbano Lorenzo Seva) offers many other useful alternative for ordered scales, Available at: https://scholarworks.umass.edu/pare/vol19/iss1/5
https://scholarworks.umass.edu/cgi/viewcontent.cgi?article=1317&context=pare
In general for ordinal scales the exploratory factor analysis is performed with PCA then you switch to PAF method because it outperforms the PCA . The Maximum Likelihood Factor Analysis under confirmatory method outperforms both PCA and / PAF . I would pay attention to the multivariate normality assumptions ( mardia statistic being one helpful index) given the scales in use have ordinal nature . MLFA requires a big sample size analysis however , a sample size in excess of 400 and better above 600 subjects are good to begin with .
Warmest Jordanian Regards. Mohammad Alkhateeb @ www.hodhodata.com
