I am comparing two different methods for factor analysis in Stata: principal factors (as specified by the pf option) and maximum likelihood (as specified by the ml option). Here are the two commands I am comparing:

sysuse auto
factor price mpg headroom trunk weight length turn displacement gear_ratio, mineigen(1) pf
factor price mpg headroom trunk weight length turn displacement gear_ratio, mineigen(1) ml

I specify mineigen(1) to only keep factors with eigenvalues greater than unity. The factor loading matrix for pf retains one factor, as would be expected, given there is only one factor with eigenvalues greater than unity (eigenvalue = 5.8). Why does the factor loading matrix for ml retain five factors, when there are only two factors with eigenvalues greater than unity?

I'm not sure if the answer might have something to do with the solution for ml being a Heywood case; the help-file for Heywood case doesn't mention anything about the factor loading matrix.

Here is the output:

. factor price mpg headroom trunk weight length turn displacement gear_ratio, mineigen(1) pf
(obs=74)

Factor analysis/correlation                      Number of obs    =         74
    Method: principal factors                    Retained factors =          1
    Rotation: (unrotated)                        Number of params =          9

    --------------------------------------------------------------------------
         Factor  |   Eigenvalue   Difference        Proportion   Cumulative
    -------------+------------------------------------------------------------
        Factor1  |      5.80299      5.27017            0.8901       0.8901
        Factor2  |      0.53283      0.23069            0.0817       0.9719
        Factor3  |      0.30214      0.11137            0.0463       1.0182
        Factor4  |      0.19077      0.19287            0.0293       1.0475
        Factor5  |     -0.00210      0.02635           -0.0003       1.0472
        Factor6  |     -0.02845      0.01025           -0.0044       1.0428
        Factor7  |     -0.03869      0.04112           -0.0059       1.0369
        Factor8  |     -0.07981      0.08074           -0.0122       1.0246
        Factor9  |     -0.16055            .           -0.0246       1.0000
    --------------------------------------------------------------------------
    LR test: independent vs. saturated:  chi2(36) =  673.52 Prob>chi2 = 0.0000

Factor loadings (pattern matrix) and unique variances

    ---------------------------------------
        Variable |  Factor1 |   Uniqueness 
    -------------+----------+--------------
           price |   0.4846 |      0.7652  
             mpg |  -0.8148 |      0.3362  
        headroom |   0.5524 |      0.6948  
           trunk |   0.7398 |      0.4527  
          weight |   0.9756 |      0.0483  
          length |   0.9546 |      0.0887  
            turn |   0.8605 |      0.2596  
    displacement |   0.9138 |      0.1649  
      gear_ratio |  -0.7831 |      0.3867  
    ---------------------------------------

. 
end of do-file

. do "C:\Users\NAE215\AppData\Local\Temp\1\STD4dc4_000000.tmp"

. factor price mpg headroom trunk weight length turn displacement gear_ratio, mineigen(1) ml
(obs=74)
number of factors adjusted to 5

<log omitted>

Factor analysis/correlation                      Number of obs    =         74
    Method: maximum likelihood                   Retained factors =          5
    Rotation: (unrotated)                        Number of params =         35
                                                 Schwarz's BIC    =    151.498
    Log likelihood = -.4278555                   (Akaike's) AIC   =    70.8557

    
Warning: Solution is a Heywood case; that is, invalid or boundary values of uniqueness.

    --------------------------------------------------------------------------
         Factor  |   Eigenvalue   Difference        Proportion   Cumulative
    -------------+------------------------------------------------------------
        Factor1  |      3.87317      1.84053            0.4924       0.4924
        Factor2  |      2.03264      1.18001            0.2584       0.7508
        Factor3  |      0.85263     -0.07200            0.1084       0.8592
        Factor4  |      0.92463      0.74196            0.1176       0.9768
        Factor5  |      0.18267            .            0.0232       1.0000
    --------------------------------------------------------------------------
    LR test: independent vs. saturated:  chi2(36) =  673.52 Prob>chi2 = 0.0000
    LR test:   5 factors vs. saturated:  chi2(1)  =    0.77 Prob>chi2 = 0.3793
    (tests formally not valid because a Heywood case was encountered)

Factor loadings (pattern matrix) and unique variances

    -------------------------------------------------------------------------------
        Variable |  Factor1   Factor2   Factor3   Factor4   Factor5 |   Uniqueness 
    -------------+--------------------------------------------------+--------------
           price |   0.3182    0.7650    0.5600   -0.0000   -0.0000 |      0.0000  
             mpg |  -0.5844   -0.4288    0.0811   -0.3382    0.1998 |      0.3137  
        headroom |   0.6619   -0.0440   -0.1115    0.1121    0.0975 |      0.5254  
           trunk |   1.0000   -0.0063    0.0018   -0.0000    0.0000 |      0.0000  
          weight |   0.6758    0.5263   -0.1410    0.4482   -0.1185 |      0.0316  
          length |   0.7292    0.3721   -0.1515    0.4769   -0.1867 |      0.0445  
            turn |   0.6037    0.3428   -0.2584    0.4783   -0.1577 |      0.1976  
    displacement |   0.6128    0.5804   -0.2571    0.3750    0.2438 |      0.0215  
      gear_ratio |  -0.5137   -0.6260    0.5868    0.0001    0.0000 |      0.0000  
    -------------------------------------------------------------------------------
0

There are 0 best solutions below