Number of factors using parallel analysis

[PleaseAdvise]

Hi,

this is my first post in this forum and I hope that someone can help me I am writing on a paper for my university and I am not really sure whether my interpretation of parallel analysis is correct.

After running this syntax (rawpar) in SPSS, I got the following result:

Actual Eigenvalue / Average Eigenvalue / 95th Percentile Eigenvalue

1) 3,33 / 0,70 / 0,84
2) 1,66 / 0,56 / 0,67
3) 0,71 / 0,45 / 0,53
4) 0,55 / 0,35 / 0,43
5) 0,40 / 0,27 / 0,35
6) 0,21 / 0,19 / 0,25

According to this paper: "Factors or components are retained as long as the ith eigenvalue from the actual data is greater than the ith eigenvalue from the random data."

-> Retain 5 factors because 3,33 > 0,84; 1,66 > 0,67 ....

However, eigenvalues 3,4,5 are smaller than 1. Does it mean that only 2 factors should be retained?

Please advise how many factors should be retained Thank you very much!

[Saegetus]

Hi,

this is my first post in this forum and I hope that someone can help me I am writing on a paper for my university and I am not really sure whether my interpretation of parallel analysis is correct.

After running this syntax (rawpar) in SPSS, I got the following result:

Actual Eigenvalue / Average Eigenvalue / 95th Percentile Eigenvalue

1) 3,33 / 0,70 / 0,84
2) 1,66 / 0,56 / 0,67
3) 0,71 / 0,45 / 0,53
4) 0,55 / 0,35 / 0,43
5) 0,40 / 0,27 / 0,35
6) 0,21 / 0,19 / 0,25

According to this paper: "Factors or components are retained as long as the ith eigenvalue from the actual data is greater than the ith eigenvalue from the random data."

-> Retain 5 factors because 3,33 > 0,84; 1,66 > 0,67 ....

However, eigenvalues 3,4,5 are smaller than 1. Does it mean that only 2 factors should be retained?

Please advise how many factors should be retained Thank you very much!

Well, it's not too much to advice here: Once you've decided to use the paralel analysis in order to determine the optimal number of factors you should consider all five.

The eigenvalue default rule of greater than one is just an alternative method of selecting significant factors wich has been considered as being too restrictive by some analysis experts.

My advice will be to check for the last two factors (if from previous knowledge or conceptual perspective are useful for your research objective).

Trivial, non-significant factors might emerge into consideration group when using paralel analysis.

Hope it helps!

[Ian Straus]

.....
However, eigenvalues 3,4,5 are smaller than 1. Does it mean that only 2 factors should be retained? .....

Yes, that is what it means.

[PleaseAdvise]

Thank you for your answers! I finally retained 3 factors and saved them as Anderson Rubin scores (standardized).

My next issue: I want to do a cluster analysis.

1) Should I use the Anderson Rubin scores to do the cluster analysis or the original variables?
2) How do I determine the number of clusters? I just tried 2,3,4, ... clusters but that seems not really scientific

Please advise Thank you!

[Ian Straus]

[QUOTE=PleaseAdvise;18996]
2) How do I determine the number of clusters? I just tried 2,3,4, ... clusters but that seems not really scientific

QUOTE]

Clustering is not all that scientific. It's more of an art form.

You might save cluster membership and then compare the clusters. Are they significantly different from one another in ways that matter to your project? At what point does the additional cluster stop telling a story that you can make sense of? Do one or two variables seem to dominate the cluster formation?

[Sherman Whipple]

As I recall Anderson Rubin estimates are considered orthogonal so there should not be a problem staying with them for your cluster analysis.