[Lindsey Chesus]

Hello,

I ran a factor analysis, and the pattern matrix is coming up as:

Pattern Matrix a

a Rotation failed to converge in 25 iterations. (Convergence = .000).

What does this mean and how can I fix it? I ran another one for a different set of data and had no issue. 

Disclaimer: I am a statistics novice. Please be gentle. 

[Nicolas Stefaniak]

in order to better understand your problem, it would be helpful to state which function (and the package you’re using).

I guess that the problem is due to the method that you’re using for extract factors. You’re probably using the maximum likewood.

If you’re using the function fa() from the psych package, you can use either minres or eighted least squares (WLS).

The argument in the function is fm (see psych documentation).

Finally, this problem can occur when your correlation (or covariance) matrix does not suit well for factor analysis. Probably that your KMO (keyser olkin meyer) is weak. Check this point before running factor analysis.

Best.

I’m not native speaker and I make probably mistakes in English, so please be also gentle  😉

[Lindsey Chesus]

Hello,

Thank you for the reply. Your English is excellent, so no worries 🙂

If I am correct as to what you are asking, I am using SPSS 22. I followed Andy Field’s instructions from his video on youtube as to running it. I am using the eigenvalue extraction choice, and the value is set to one. The method is principal axis factoring. If I left something out, please let me know.  

[Nicolas Stefaniak]

I do not use SPSS, but try anything else principal axis.

Note that in “discovering statistics using R”, Andy Field consider factor analysis and principal component as the same analysis. I do not know what he is saying in his video, but it is not exactly the same (even if I share his view, I know that some reviewer could disagree with his view and making factor analysis should be a better alternative.

I’m pretty sure that SPSS provides the KMO. Can you provide the value of the KMO?

[Lindsey Chesus]

I am looking at perceived outcomes and actual outcomes for a program, and in the perceived outcome analysis I was able to see the pattern matrix. Will changing the output for the actual outcome influence the comparison? The KMO is .521

[Nicolas Stefaniak]

This KMO should be considered as miserable, this is why your factor analysis does not converge. This means that your data do not suit well for factor analysis. So my advice would be to think about another analysis (perhaps cluster analysis – try reading Revelle’s advices on the question in the psych help package in R. He developed a function called iclust.

The other (but probably worse) solutions are :

-Try using the other methods in SPSS, 

-Try if it is possible to increase the number of iteration (perhaps 100).

[Lindsey Chesus]

Oh boy.

I re-ran it in SPSS using 100 iterations, and it worked (said it converged in 53). 

Is this bad?

[Lindsey Chesus]

I went and looked at my variables, and there were some in there that were not supposed to be included. After removing the annoying variables, my KMO was .786, and it only took 11 iterations. Thank you for your help! 

[Author]

Posts