• determinant of matrix in factor analysis

    From alizaesther8@gmail.com@21:1/5 to All on Wed May 20 11:11:01 2020
    I'm trying to do a factor analysis on a questionnaire with 63 questions. Andy Field's SPSS book says the determinant of the correlation matrix shouldn't be less than .00001. Mine is 5.125E-16, and that's after removing the 8 questions with the highest
    intercorrelations (R>.7). Any suggestions? Thank you in advance.

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From alizaesther8@gmail.com@21:1/5 to All on Wed May 20 11:13:51 2020
    p.s. N = 335

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From Rich Ulrich@21:1/5 to All on Wed May 20 17:34:55 2020
    On Wed, 20 May 2020 11:11:01 -0700 (PDT), alizaesther8@gmail.com
    wrote:

    I'm trying to do a factor analysis on a questionnaire with 63 questions. Andy Field's SPSS book says the determinant of the correlation matrix shouldn't be less than .00001. Mine is 5.125E-16, and that's after removing the 8 questions with the highest
    intercorrelations (R>.7). Any suggestions? Thank you in advance.

    I never noticed that criterion anywhere, but my (many)
    FAs were almost always performed on smaller sets of items,
    any my items usually were Likert-type scale items. What do
    you have as your questionnaire items, more specifically?

    You do not have 10 or 20 times the number of cases as
    your number of items. That's a frequent recommentation.
    But that advice is based on items with not-large correlations,
    where that large of an N is needed to get a robust structure
    to emerge.

    If you do have Likert-type items, correlations of
    more than 0.70 would be extremely large, since that
    is about the maximum they should yield from their
    inherent reliability.

    What does your attempt at FA give you? Reasonable
    factors? Are the estimated communalities reasonable
    estimates for the item reliabilities?

    - In case of small N (which yours /might/ be), I have
    done FA with fixed estimates of reliability for the diagonal.

    Hope this help.

    --
    Rich Ulrich

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)