Uniqueness calculation in EFA

General help and assistance with jamovi. Bug reports can be made at our issues page: https://github.com/jamovi/jamovi/issues . (If you're unsure feel free to discuss it here)

by David » Fri Nov 23, 2018 11:37 am

Hi, in a five factor EFA solution the uniqueness calculation seems a bit off, not sure why. Hand calc gives me a uniqueness of 0.66 for a promax rotated solution but the jamovi displayed value is 0.74. Hand calc of the unrotated solution is correct, so I'm not sure why rotated factor loadings don't add up to displayed uniqueness. Using jamovi 0.9.5.5.
David
ps omv attached
Attachments
bfi_sample.omv
(143.01 KiB) Downloaded 55 times
David
 
Posts: 28
Joined: Wed Jun 13, 2018 10:18 am

by Ravi » Fri Nov 23, 2018 12:04 pm

Hi David,
So by default we use "Minimal Residuals" as the factor extraction method (just like the R package "psych" that we use in the background). However, "factanal" uses maximum likelihood as the extraction method. If you change the extraction method to maximum likelihood, the uniqueness should be the same again.

Cheers,
Ravi
User avatar
Ravi
 
Posts: 140
Joined: Sat Jan 28, 2017 11:18 am

by David » Fri Nov 23, 2018 12:09 pm

Hi Ravi, maximum likelihood also gives 0.66 uniqueness for a five factor promax rotated solution (hand calc) - but the displayes uniqueness is 0.74 (jamovi 0.9.5.8)
David
 
Posts: 28
Joined: Wed Jun 13, 2018 10:18 am

by David » Fri Nov 23, 2018 12:10 pm

jamovi 0.9.5.8 ... not sure why the shades appeared!
David
 
Posts: 28
Joined: Wed Jun 13, 2018 10:18 am

by David » Fri Nov 23, 2018 12:17 pm

jmv syntax shows nFactors = 2 but five are extracted using parallel method...could this be something to do with it? See attached.
Attachments
Screenshot 2018-11-23 at 12.14.54.png
Screenshot 2018-11-23 at 12.14.54.png (164.48 KiB) Viewed 1064 times
David
 
Posts: 28
Joined: Wed Jun 13, 2018 10:18 am

by David » Fri Nov 23, 2018 12:22 pm

..answering my own question, NFactors = 2 or NFactors = 5 in jmv syntax doesn't change things...
David
 
Posts: 28
Joined: Wed Jun 13, 2018 10:18 am

by Ravi » Fri Nov 23, 2018 12:32 pm

But it does match the R result you calculated using 'factanal' now right?
User avatar
Ravi
 
Posts: 140
Joined: Sat Jan 28, 2017 11:18 am

by David » Fri Nov 23, 2018 12:46 pm

Using factanal handcalc uniqueness = 0.66 but displayed = 0.74, just like in psych. Probably something happening that I don't understand...I had thought that rotated factor loading should still sum to same communality / uniqueness as unrotated?
David
 
Posts: 28
Joined: Wed Jun 13, 2018 10:18 am

by David » Fri Nov 23, 2018 1:06 pm

Further digging shows that with orthogonal rotation (varimax) factor loadings combine correctly to displayed uniqueness, but not for oblique rotation (promax). Extraction method doesn't seem to make a difference.
David
 
Posts: 28
Joined: Wed Jun 13, 2018 10:18 am

by David » Fri Nov 23, 2018 1:30 pm

aah, pattern vs structure matrix!! Sorry I'm a bit rusty on fa
David
 
Posts: 28
Joined: Wed Jun 13, 2018 10:18 am

Next

Return to Help

cron