table/collect PCA proportion of variance

[Ariel Karlinsky]

Hi,

I'm trying to create a table (Stata 17) of several PCA models that shows the share of variance explained by each principal component ("Proportion" column in the results), but I'm unsuccessful and searching for help on this did not help either.

. sysuse auto
(1978 automobile data)

. pca mpg price weight

Principal components/correlation                 Number of obs    =         74
                                                 Number of comp.  =          3
                                                 Trace            =          3
    Rotation: (unrotated = principal)            Rho              =     1.0000

    --------------------------------------------------------------------------
       Component |   Eigenvalue   Difference         Proportion   Cumulative
    -------------+------------------------------------------------------------
           Comp1 |       2.2225      1.63291             0.7408       0.7408
           Comp2 |      .589588      .401676             0.1965       0.9374
           Comp3 |      .187912            .             0.0626       1.0000
    --------------------------------------------------------------------------

Principal components (eigenvectors) 

    ----------------------------------------------------------
        Variable |    Comp1     Comp2     Comp3 | Unexplained 
    -------------+------------------------------+-------------
             mpg |  -0.6021    0.4307    0.6723 |           0 
           price |   0.5037    0.8582   -0.0987 |           0 
          weight |   0.6195   -0.2792    0.7337 |           0 
    ----------------------------------------------------------

Any advice would be more than welcome.

[Hemanshu Kumar]

Ariel Karlinsky I do not know if it possible to manipulate matrices once collected, and I also don't know how to pass a matrix to -collect- unless it is a stored result. Thus here is my code, which feels unnecessarily complicated. Perhaps someone else has a much easier way of doing it!

collect clear

capture program drop make_prop
program define make_prop, rclass
    args eigenval trace
    mat wanted = `eigenval' / `trace'
    return mat Prop = wanted
end

sysuse auto, clear
pca mpg price weight

make_prop e(Ev) e(trace)

collect get r(Prop)
collect label levels rowname r1 "Proportion", modify
collect style cell rowname, nformat(%5.4f)
collect layout (colname) (rowname)
noi collect preview

which produces:

------------------
      | Proportion
------+-----------
Comp1 |     0.7408
Comp2 |     0.1965
Comp3 |     0.0626
------------------

[Ariel Karlinsky]

Thanks! for my purposes this is probably good enough!