Always interesting in these cases to
contrast a first principles solution
and recourse to a canned program.
Let's invent some data:
clear
set obs 100
forval i = 1/5 {
gen i`i' = uniform() > 0.5
}
A first principles solution is
(1) set up some matrices
mat chi2 = J(5,5,0)
mat pval = J(5,5,0)
(2) loop over varlists and populate the matrices with
results:
local i = 0
qui foreach x of var i* {
local ++i
local j = 0
foreach y of var i* {
local ++j
tab `x' `y', chi
mat chi2[`i',`j'] = r(chi2)
mat pval[`i',`j'] = r(p)
}
}
(3) decorate the matrices
unab x: i*
matrix rownames chi2 = `x'
matrix colnames chi2 = `x'
matrix colnames pval = `x'
matrix rownames pval = `x'
(4) show results
mat li chi2
symmetric chi2[5,5]
i1 i2 i3 i4 i5
i1 100
i2 .01463915 100
i3 1.1576392 .52040936 100
i4 .36363636 1.4492754 1.999184 100
i5 .55512978 .42413558 .2398025 1.003613 100
mat li pval
pval[5,5]
i1 i2 i3 i4 i5
i1 1.524e-23 .90369711 .28195568 .5464936 .45622921
i2 .90369711 1.524e-23 .47066711 .22864426 .51488065
i3 .28195568 .47066711 1.524e-23 .15738391 .6243488
i4 .5464936 .22864426 .15738391 1.524e-23 .31643784
i5 .45622921 .51488065 .6243488 .31643784 1.524e-23
=============================
-makematrix- is two steps:
(1) spend 30 minutes reading the help file and
figuring out syntax
(2)
makematrix , from(r(chi2)) cols(i*) : tab i*, chi
i1 i2 i3 i4 i5
i1 100
i2 .01463915 100
i3 1.1576392 .52040936 100
i4 .36363636 1.4492754 1.999184 100
i5 .55512978 .42413558 .2398025 1.003613 100
makematrix , from(r(p)) cols(i*) : tab i*, chi
i1 i2 i3 i4 i5
i1 1.524e-23 .90369711 .28195568 .5464936 .45622921
i2 .90369711 1.524e-23 .47066711 .22864426 .51488065
i3 .28195568 .47066711 1.524e-23 .15738391 .6243488
i4 .5464936 .22864426 .15738391 1.524e-23 .31643784
i5 .45622921 .51488065 .6243488 .31643784 1.524e-23
I am not clear why Stata does not regard the last matrix
as symmetric.
Nick
[email protected]
Michael Blasnik
> you may want to check out -makematrix- available from SSC.
Daniel Egan
> > I have a large list of indicator variables for which I
> would like the
> > equivalent of a correlation matrix, but with pearsons chi2
> and/or the
> > significance of pearsons chi2 as the cell contents instead of the
> > correlation coefficient. Similarly, I would like only the lower half
> > matrix.
> >
> > Is there a packaged way of doing this?
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
