Dear Statalisters,
I'm using Stata 8.
I have a question about the sample code in -heckprob- ([G-M]
pp.80-1). Here is an excerpt:
----------
clear
set seed 12309
set obs 5000
gen c1 = invnorm(uniform())
gen c2 = invnorm(uniform())
sum c1 c2
matrix P = (1,.5\.5,1)
matrix A = cholesky(P)
local fac1 = A[2,1]
local fac2 = A[2,2]
gen u1 = c1
gen u2 = `fac1'*c1 + `fac2'*c2
summarize u1
replace u1 = u1/r(sd)
summarize u2
replace u2 = u2/r(sd)
drop c1 c2
gen x1 = uniform()-.5
gen x2 = uniform()+1/3
gen y1s = 0.5 + 4*x1 + u1
gen y2s = 3 - 3*x2 + .5*x1 + u2
gen y1 = (y1s>0)
gen y2 = (y2s>0)
heckprob y1 x1, sel(y2 = x1 x2) nolog
----------
What I don't get the idea of is the matrix-processing part
of the above code.
It looks like a symmetric positive-definite matrix is first
geneated, then is decomposed into the product of lower triangular
matrix and its transpose, whose [2,1] and [2,2] elements are
used to construct correlated two (bivariate normal) error
terms.
But I don't see why Cholesky decomposition is required here
to create these error terms.
I'm afraid no persuasive explanations can be found in the
manual, though it might be just a standard routine for
programmers.
Any suggestions welcome.
Thanks in advance.
K.I.
*
* 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/
