On Thu, 11 Oct 2007, Marcos Larios wrote: > Hi all, I'd been usng Gretl for over 3 years now and I need it > to do two tests: Bartlett's Test and Levene's test for > Homogeanity of Variance, to use them in the residuals of some > models...
> The thing is that when I do the "genr sample = > (POPM<=maxband&POPM>=lowband)" code in Gretl's console it does > it allright but when I do it as a function "? > bartlett(RESID01,POPM,5)" it does nothing, it only shows the > printf output, but it doesn't generates the sample series and > much less changes the sample size of the workfile. The design of user-defined functions in gretl is intended to ensure that such functions don't have any surprising side-effects -- that is, they are "safe to use" for people who don't know exactly what goes on inside them. The implications of this design decision are spelled out in Chapter 10 of the Gretl User's Guide. Two points are particularly relevant here: 1) Any variables defined inside a function are thrown away when the function exits, unless they are made available as return values, and those return values are assigned by the caller of the function. 2) You can change the sample all you like inside a function, but when the function exits the sample reverts to what it was before the function was called. For the purposes of a function that implements Bartlett's test, I don't think these things should be problematic. Within the function, you can reset the sample repeatedly and calculate variances. All you need return is a test statistic and/or p-value -- or just print the test result. In case it's helpful, I'm putting below a script that solves a simpler but related problem, namely printing an F-test for the hypothesis that the variance is the same, for some given variable, in the first and second halves of the sample. A couple of comments follow. <gretlscript> function vfun (series y) set messages on set echo on scalar n2 = $nobs/2 smpl ; -n2 n1 = $nobs - 1 v1 = var(y) smpl --full smpl +n2 ; n2 = $nobs - 1 v2 = var(y) printf "v1 = %g, v2 = %g\n", v1, v2 Ft = (v1 > v2)? v1/v2 : v2/v1 pval = pvalue(F, n1, n2, Ft) printf "F(%d,%d) = %g, p-value %g\n", n1, n2, Ft, pval return scalar pval end function open data9-7 scalar pv = vfun(QNC) </gretlscript> Comments: (1) You should use set echo on set messages on inside your function when debugging. Then remove these lines when you're sure things are working right. (2) Unfortunately there's an annoying bug in smpl, whereby a "-" indicating retard-the-sample-end-or-start gets confused with the "-" that precedes an option flag. The form I used above, "-n2", happens to be a lucky string that doesn't trigger the problem. (I think the solution to this is to insist that the options to smpl be given in their full form, with two dashes. I'll fix that for gretl 1.6.6.) Allin Cottrell
