# Unexpected eigenvalues in parallel analysis for factor analysis in SPSS

Would greatly appreciate if someone could clarify which eigenvalues I am supposed to compare when using parallel analysis to determine factor retention.

I am running Principal Axis Factoring in SPSS 24. For Parallel Analysis (PA) to determine the number of factors, I am using the syntax provided by Brian O'Connor to run parallel analysis in SPSS. I used the 'rawpar' programme as my data is not normally distributed.

I am unclear which eigenvalues I am supposed to compare:

• The 'Initial eigenvalues' from the PAF output and the Prcntyle values from the PA output?
• The 'Raw data' values from PA and the Prcntyle values from PA

The initial eigenvalues produced by SPSS are different from the Raw data values produced by PA and I'm confused by that as I thought they should be the same, but maybe I misunderstood.

Here's the PAF output:

And the PA output:

• This seems SPSS-specific & so may be hard to answer for people who don't use SPSS. Can you paste in your output? People may be able to figure it out from looking at it. – gung - Reinstate Monica Jul 7 '17 at 12:15
• I included a link to a screenshot of the SPSS above, it is hard to understand when I paste it in as the headings for the column jump all over the place? I have tagged SPSS now too. Thanks for getting back to me. – sjb Jul 7 '17 at 12:24
• I expect the answer is like this. On your SPSS factor analysis output pic, you display the results of PAF factoring extracting 10 factors. It looks like a full-blown (iterative) PAF. The results of "PA" (Parallel analysis) pic display eigenvalues of the reduced correlation matrix without iterations. I.e. it is same as you set in PAF number of iteration 1 or 0 (check it). So, iterative PAF has not been done. – ttnphns Jul 7 '17 at 12:36
• In iterative PAF you have to assume the number of factors to extract. You extracted 10. In the parallel analysis procedure (after Connors?) which syntax you used the author does not initially extract any specific number of factors. Under "PAF" method he actually sets squared mult. corr. (smc) on the diagonal, extracts eigenvalues, displays them for you in the "Raw data" column, and does not proceed to iterate. Find lines compute smc = 1 - (1 &/ diag(inv(cr)) ). call setdiag(cr,smc). compute realeval = eval(cr). – ttnphns Jul 7 '17 at 12:55
• ...To asses how much this is valid or are there better ways, please read the source article or contact with the author. There is not just single approach to doing parallel analysis. – ttnphns Jul 7 '17 at 12:56