Item Exploratory Factor Analysis - results changing when code is rerun

[Sam Honer]

I have a sample of N = 401 individuals who completed a self-report measure on mobile-use dependency. In an effort to validate the survey I plan on first running an exploratory factor analysis (EFA) followed by a confirmatory factor analysis (CFA). The questionnaire consisting of 18 binary items so I am using item factor analysis.

I have randomly split the sample into two to carry out EFA on one sample and CFA on the other using the following code:

set seed 12345
generate random = uniform()
sort random
generate group= .
replace grp = 1 in 1/200
replace grp = 2 in 201/401

Then to run the item exploratory factor analysis (iEFA), I run the following code:

preserve
drop if grp == 2
tetrachoric i1-i18, posdef
matrix mysigma = r(Rho)    // saving the tetrachoric correlation matrix

factormat mysigma, n(401) mineigen(1) blanks(0.2) // factor analysis
screeplot
rotate, promax(3) oblique kaiser blanks(0.3)  
restore

Every time I rerun the the above chunks of code, I obtain different loadings and sometimes a different number of factors.
What am I missing here?

[Bruce Weaver]

Are you re-opening the data file when you rerun your code (to get back to the same initial state)?

Also, why are you specifying n(401) on your factormat command when you are using only 200 observations?

[Sam Honer]

Hi Bruce, yes every time I reopen the file or rerun the code within the same session this happens.

*The n = 401 is a mistake that should say 200 - thanks for pointing out.

[Bruce Weaver]

When I modify your code to use the data from this UCLA webpage, I get the same result every time. What do you get when you run my modified code?

cls
forvalues i = 1/2 {

use "https://stats.idre.ucla.edu/stat/stata/faq/tetra", clear
set seed 12345
generate random = uniform()
sort random
generate grp= .
replace grp = 1 in 1/100
replace grp = 2 in 101/200

preserve
drop if grp == 2
tetrachoric female-sci, posdef
matrix mysigma = r(Rho)    // saving the tetrachoric correlation matrix

factormat mysigma, n(100) mineigen(1) blanks(0.2) // factor analysis
screeplot
rotate, promax(3) oblique kaiser blanks(0.3)  
restore

}

[Joseph Coveney]

Every time I rerun the the above chunks of code, I obtain different loadings and sometimes a different number of factors.
What am I missing here?

You've sorted it.

When you -restore- the dataset, it's not the same dataset as what you had at first: it's sorted differently, and so when you go to sort it again, you start from a different sort order.