Exploratory Factor Analysis problems

#1
Independent Variable: 49 state survey of state court administration: how much power does the state court administrator have over 31 separate management functions: Total control (coded as 3), Shared control with local government (2), or none (1). So a state in which all Human Resources functions are dictated by the state HQ would have TOTAL control. A state in which the Procurement was split with some local government control would he SHARED. Etc.

Dependent Variable: 24 state survey of the case clearance rate for the state. So, in state X there was 1,387,788 cases filed in 2013, 1,335,007 disposed for a case clearance rate of 96.2%.

Theory from Literature: States with higher amounts of centralization (IV) should be more efficient (higher DVs).

The Problem: Using the 49 states, I perform an initial rotation with Principal Component Analysis that determined there were 9 factors to be extracted. Kaiser-Meyer-Olkin Measure of Sampling Adequacy = .622. Second rotation: Extraction = Principal Axis Factoring; Rotation Methods= Promax with Kaiser Normalization. Resulting scree plot indicates two “elbows”: Factor 3 and Factor 5.

But, I get worried about the notion that I’m trying to use factor scores based on 49 states and apply them to by 24 state data set.

Using just the survey results from the 24 states with caseload data as well I run the EFA again. Initial rotation with Principal Component Analysis results in no KMO score; “This matrix is not positive definite.” There is a scree plot generated and a component matrix made up of 8 items, but it is clear there’s an error here.

Question: Can I use factor scores based on 49 states and apply them to my 24 state data set? If not, how can I execute an exploratory factor analysis otherwise?
 
Last edited:

noetsi

Fortran must die
#2
Why are you applying the factor scores to only 24 states? I don't see that in your discussion. You can always see if a set of factors from one sample is replicated in another sample, but you can't assume because you did find it in one you will find it in another. This is true of any statistic, and would have been true even if you were doing 49 states. And the 24 states data of course could be substantively different from the 49 because the states vary.

You might want to look at this in terms of your error.

http://www-01.ibm.com/support/docview.wss?uid=swg21477275

http://www2.gsu.edu/~mkteer/npdmatri.html
 
#3
> Why are you applying the factor scores to only 24 states?

Because I don't have dependent variable data for the other states.

> And the 24 states data of course could be substantively different from the 49 because the states vary.

So I can't use it, which is what I thought.

Ok, thanks.