I am probably missing something here, but why not use
predict pred if e(sample)
and then calculate its variance directly?
Nick
[email protected]
Renzo Comolli
>
> I would like to have measures of goodness of fit for
> intreg depvar1 depvar2 indepvars, het(subset_of_indepvars) robust
> constraints(1)
>
> The measure I would really like is variance of predicted y,
> but I would
> settle for other measures too.
>
> I can't use -fitstat- because it was written for the Stata 7
> version of
> -intreg-.
> I hacked the code of -fitstat- enough to make it work for
> Stata 8 version of
> -intereg-, but it does not work if I specify the option -,
> het()- and I
> don't understand the code enough to hack it further.
> (I wrote to Mr. Long months ago, so that is not an option)
>
> I post here my hacked version of the code to compute variance
> of predicted
> y, but fails (as said) if the -, het()- option is specified.
>
> capture program drop vystar
> program define vystar, rclass
> version 6.0
>
> tempname n_obs n_parm b v_ystar v v_y v_x v_e factor bnocon
>
> mat `b' = get(_b)
> local depv `e(depvar)'
>
> *-> get information from last model estimated
> mat `v_e' = J(1,1,1)
> sca `n_obs' = e(N)
>
> *-> compute numbers of variables and parameters
> sca `n_parm' = colsof(`b')
> local n_rhs = `n_parm' - 2 /* for var(e) */
>
> local varlist : colnames(`b')
> parse "`varlist'", parse (" ")
>
> * 2/11/00
> * to deal with rhsnam gt 80 characters, can't use string functions
> local rhsnam ""
> local i 1
> while "``i''"!= "" {
> if "``i''"!="_cons" & "``i''"!="_se" { local
> rhsnam "`rhsnam'
> ``i''" }
> local i = `i' + 1
> }
>
> *-> compute Var(y*) using cov_x
>
> scalar `factor' = 1/(`n_obs'-1)
> quietly mat accum `v_x' = `depv' `rhsnam' if
> e(sample), deviations
> noconstant
> scalar `v_y' = `factor' * `v_x'[1,1] /* sum y/(n-1) */
> mat `v_x' = `factor' * `v_x'[2...,2...] /* cov
> matrix of rhs vars
> */
> mat `v_x' = `v_x'[2...,2...]
> mat `bnocon' = `b'[1,1..`n_rhs'] /* trim off _con */
> mat `v' = `bnocon' * `v_x' * `bnocon''
> mat `v_e'[1,1] = e(sigma)*e(sigma)
> mat `v' = `v' + `v_e'
> scalar `v_ystar' = `v'[1,1]
> display "Variance of Fitted y* :" " " `v_ystar'
> return scalar v_ystar = `v_ystar'
>
> end
*
* 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/
