Re: [Gretl-users] Regressograms

On Sat, 14 Jan 2012, AGUSTÍN ALONSO RODRÍGUEZ wrote: Regarding nonparametric methods, current gretl offers loess regression and the Nadaraya-Watson estimator -- see the functions loess() and nadarwat(). The regressogram is not a built-in method but it can be scripted in gretl. An example follows. The data, attached, are from the example at http://fedc.wiwi.hu-berlin.de/xplore/ebooks/html/anr/anrhtmlnode20.html <hansl> function matrix regressogram (series y, series x, matrix cuts) scalar nbins = cols(cuts) - 1 matrix R = zeros(2*nbins, 2) scalar k = 1 loop i=1..nbins -q smpl (x >= cuts[i]) && (x < cuts[i+1]) --restrict R[k,1] = mean(y) R[k+1,1] = R[k,1] R[k,2] = cuts[i] R[k+1,2] = cuts[i+1] - 0.001 k += 2 smpl --full endloop colnames(R, "y x") outfile &quot;(a)dotdir/regresso.plt&quot; --write print "plot '-' using 2:1 w l notitle, \ '-' using 3:2 w p notitle" printf "%10.5g\n", R print "e" print y x --byobs print "e" outfile --close gnuplot --input=&quot;(a)dotdir/regresso.plt&quot; --output=display return R end function open mcycle.gdt matrix cuts = seq(0,19) * 3 matrix R = regressogram(y, x, cuts) print R </hansl> Allin Cottrell