On Wed, 30 Jan 2008, nadaud(a)centre-cired.fr wrote: [...] Let me recap: you're adding two column vectors; the second one is obtained by dividing a scalar by all the elemnts of a column vector. \footnote{You could have used the syntax a./b, but this is immaterial.} There has been some work recently in the area of casting a 1x1 matrix to a scalar, and a few bugs were fixed. Try using 1.7.1, or perhaps the snapshot version. Alternatively, try what happens after substituting the line matrix erlog = vecun + (coeffaids[4,1]).*(1/what[,1]) with scalar tmp = coeffaids[4,1] matrix erlog = vecun + tmp.*(1/what[,1]) or, better still, scalar tmp = coeffaids[4,1] matrix erlog = vecun + tmp./what[,1] What worries me more than this is that, unless I'm mistaken, a practically identical scripts triggers a different behaviour. Maybe you're using two different PCs with different versions of gretl? Riccardo (Jack) Lucchetti Dipartimento di Economia Università Politecnica delle Marche r.lucchetti(a)univpm.it http://www.econ.univpm.it/lucchetti

Hello listers, greetings from Paris ! Just returned and glad to read you ! Well, I tried both of your advises but the problem persists. Jack as, you wrote, there is probably something lurking around, and this is why I decided to write to the list because it seemed serious enough to report it. Initially, I throught I made a mistake in copying the code of the main script, but after struggling most of yesterday (30th january) it seemed that the same code did run differently in two executions !!? I will try to make the computations using series instead and tell you. cheers sincerely Franck PS: I have good hope to provide around march / april some functions and scripts to compute spatial tests in gretl.

Hi again ! Please send a testcase, that would be very helpful. Also, please be explicit on the version of gretl you're running. Ok Jack, I send it as soon as possible (sure tomorrow). Finally, I installe 1.7.1 today, rebooted and started again. The problem persist but I could bypass it using genr instead of matrix functions. Here is some sample code : ============== in my script file ======== ###calcul à partir de séries genr elr_log=1+(1/yhat_s01)*$coeff[4] genr elr_car=1+(1/yhat_s02)*$coeff[11] matrix melrlog=meanc(elr_log) matrix melrcar=meanc(elr_car) matrix merl=melrlog~melrcar merl =============================== it works ok, and here is the return code: =========================== ###calcul à partir de séries ? genr elr_log=1+(1/yhat_s01)*$coeff[4] Matrice elr_log remplacée ? genr elr_car=1+(1/yhat_s02)*$coeff[11] Matrice elr_car remplacée ? matrix melrlog=meanc(elr_log) Matrice melrlog générée ? matrix melrcar=meanc(elr_car) Matrice melrcar générée ? matrix merl=melrlog~melrcar Matrice merl générée ? merl merl (1 x 2) 0.16027 0.14265 ============================= So using genr and the $coeff from system estimation with fitted saved as yhat_s01 and yhat_s02, I can compute my elasticities and show them in a row vector (merl) however, there is still something a bit puzzling: using genr I expected two series of elasticites (elr_log ; elr_car) but gretl returns two matrices. Well its ok, I just need the results to compute standard errors of mean elasticities via the delta method (the objective of the whole thing !). So, I will submit a sample of the data and the full script as soon as possible so that all this can be investigated. Woosh ! that is really great news ! And I must admit i did not notice because I am fully mobilized on the system estimation and elasticities computations. I saw that my doc is out of date... I think that this will greatly ease the development of these scripts ! thanx a lot to mention it ! cheers sincerely Franck

Dear all listers As I wrote yesterday, here is my script and part of my data (a sample of 500 households). The following script does system estimation by ITSUR on two variables home energy and motorfuel. then I compute elasticities for each household. Initially i intended to do it with matrix commands only. But as I encountered the problem i resorted to saving the fitted values and using genr commands. However, after these command the script computes those same elasticities but with matrix commands. So let ei = income elasticity for equation i I have eih = 1 + beta[i] / what[i,h] essentially i generate the vector of ones vecun = ones(500,1) then compute b = beta[i] * (1/what[i,h]) to finally add them : ei = vecun + b but here gretl do not want to add those two vectors. you can see it by yourself executing the script. Notes : 1) i am under 1.7.1 just installed yesterday. The problem persists 2) in system estimation, the matrix accessors are a bit different. There is an extra line & column in $vcv and $coeff but I do not remember what it is exactly althrough it seems to be related to ML ? greetings and see ya ! sincerly Franck

On Fri, 1 Feb 2008, Riccardo (Jack) Lucchetti wrote: It's related to the extra coefficient needed to test the restriction; I forgot to trim that off before storing $coeff and $vcv, for the case of an LR test. That's now fixed in CVS. Allin.