Determining number of factors during EFA using Maximum Likelihood factor analysis in SPSS
HTML-код
- Опубликовано: 21 авг 2024
- In this video, I provide a demonstration of an approach (i.e., Maximum likelihood factor analysis) to aid you in making a decision regarding the number of factors that best account for the intercorrelations among your measured variables during factor analysis. In the video, I demonstrate a process of generating measures of model fit for one-, two-, three-, and four-factor models, including the chi-square goodness of fit test and the RMSEA (root mean square error of approximation).
Some of the discussion points I raise are addressed in:
Fabrigar, L. R., & Wegener, D. T. (2012). Exploratory factor analysis. Oxford University Press.
Feel free to download a copy of the dataset here: drive.google.c...
During the presentation, I use an Excel file to make certain calculations. You can download a copy of the final file here: drive.google.c...
In the video, I reference another presentation on Parallel analysis as a basis for determination of factors. Here is the link: • Parallel analysis for ...
Thank You very much for your interpretation Dr. Crownson. This is exactly what I was looking for, so much details
You are very welcome!
Thank you Dr Mike!
My pleasure!
Thanks for sharing this. Can you please make an EFA demonstration video with STATA or R?
Thank you so much Dr for this great video .
Using ML estimator method assumes that the observed variables are normally distributed whereas our variables are 5 points likert scale which are not . What about this assumption ?
Thank you again .
Hi Mostafa, thank you for your question. You are correct that the ML estimator does assume multivariate normality. This is particularly important when it comes to testing factor coefficients for significance and the chi-square goodness of fit test. In the context of EFA, testing factor coefficients is rarely done - and generally does not seem to be included as an option in most statistical packages. (Usually testing of factor loadings occurs in the context of confirmatory factor analysis; and non-normality in that context tends to lead to increased type 1 error rate when testing whether coefficients are different from 0). With respect to the chi-square goodness of fit test, non-normality can have an impact by inflating the chi-square test value (where you have an increased type 1 error probability; but in that case, your at greater risk of rejecting a reasonable model). From what I've read (e.g., Fabrigar & Wegner, 2012; and others), ML EFA is 'reasonably robust' to multivariate non-normality. However, there's not tons of guidance on how bad is 'bad enough'. I tend to adopt the principal laid out by Fabrigar & Wegener (2012) and Pituch and Stevens (2016) that it's a good idea to make determinations on the number of factors using multiple methods. When the suggested number of factors differs, you might consider the question 'why'. Another thing to keep in mind is the interpretability of models with different numbers of factors. I hope this helps!
@@mikecrowson2462 thank you so much Dr Mike , i appreciate your reply .
Thank you Dr Mike. A very interesting presentation. I have RMSEA of .078 and it decreases to .028, is it good or i have to keep the first number of factors? Thanks again.
Dr. Crowson, I have a question. Let's say I have 20 items for EFA. How can I determine the minimum sample size for EFA? Please let me know.