Hello everyone! i'm a medical student from Italy, so not an expert in statistics and I apologize in advance for my linguistic or technical errors. I would like to ask if there is a way, using Gretl, to perform a series of unique logistic regressions with a single command, keeping the same nominal variable and analyzing its relationship with a series of measurement variables extracted from an excel database. let me explain: I have a single nominal variable , A, and a series of measurement variables B C D (100 items). usually I will open Gretl, load the database, click on "logit", "binary", select the nominal variable and the first regressor, get the data I need (O.R., p value, IC 95%), copy them, and paste them into Word or Excel. then I repeat all again using the second regressor, then the third, and so on...a huge waste of time and energies. Is there a way to avoid this exhausting task, automating the process and performing the calculation operation once, and then getting a table with the results in the column? Do you think it's possible? could you help me with this task? thank you in advance Sincerely Davide _______________________________________________ Gretl-users mailing list Gretl-users(a)lists.wfu.edu http://lists.wfu.edu/mailman/listinfo/gretl-users

Am 21.11.18 um 20:12 schrieb Davide Bertani: > > I would like to ask if there is a way, using Gretl, to perform a > series of unique logistic regressions with a single command, keeping > the same nominal variable and analyzing its relationship with a series > of measurement variables extracted from an excel database. let me > explain: I have a single nominal variable , A, and a series of > measurement variables B C D (100 items). usually I will open Gretl, > load the database, click on "logit", "binary", select the nominal > variable and the first regressor, get the data I need (O.R., p value, > IC 95%), copy them, and paste them into Word or Excel. then I repeat > all again using the second regressor, then the third, and so on...a > huge waste of time and energies. Hi Davide, typically it is unusual to estimate three separate equations if you think all your explanatory variables B, C, D could have an influence. You would put them all in at once. (Unless you have a good reason not to, but you haven't mentioned something like that AFAICS.) In case this caused confusion, the term "binary" in this context doesn't mean bivariate, but it means that the dependent variable has only two possible values, 0 and 1. cheers, sven _______________________________________________ Gretl-users mailing list Gretl-users(a)lists.wfu.edu http://lists.wfu.edu/mailman/listinfo/gretl-users

Am 22.11.18 um 11:46 schrieb Davide Bertani: > Thank you guys for the answers > @sven I understood the use of the "binar" term: as a matter of fact, > my nominal variable is actually binary, spacing from 0 to 1. Problem > is, my thesis tutor wants UNIVARIATE as well as multivariate analysis > of these data, so I'm forced to estimate all 100 separates (the > univariate analysis, as far as I understand) and then I will estimate > the 100 variables at once (multivariate analysis if Im not wrong). And > Im looking for help for accelerating the First task:) Ah OK, I think I understand. You don't just have B,C,D as regressors but actually have 100 (potential) regressors. (First I thought you had 100 observations per variable, not 100 variables.) Then you need to do some programming with gretl's scripting language hansl indeed. You could try the following: <hansl> list reg = B C D # <and so forth> # an alternative way is to use a series range as in 'list reg = B..ZZ', see the manual ch. 14. loop foreach r reg logit A const $r # use the $coeff (and other) accessor if you want to store the result endloop </hansl> A working example with a built-in dataset would be as follows: <hansl> open credscore list reg = Age..Selfempl loop foreach r reg --quiet logit Acc const $r endloop </hansl> HTH, sven _______________________________________________ Gretl-users mailing list Gretl-users(a)lists.wfu.edu http://lists.wfu.edu/mailman/listinfo/gretl-users

Here I am again guys! First thing first, thank you very much Sven for your advice. With your help and with my friend's, I managed to adapt the code you sent me to my needs, and make it work on Gretl. So thank you very very much for your help! I am now facing a second, bigger challenge: with Sven's script, I got the standard errors, p values, O.R. and IC 95% that I needed. Now I have to tabulate them, using Excel or Word. Problem is that the Gretl results file, which I attached to the mail, shows (along with the useful data obtained) many other things that I do not need, lines and lines of text that I should manually delete hundreds of times (the akaike criterion, the schwarz criterion, hannan- quinn etc.). What I need is a series of rows displaying ONLY the ES, P-values and so on, instead of having to search through the list of results that the Gretl logit analysis gave me. DId I manage to explain my problem? does anyone have any suggestions about it? I add to the mail three things: the script I used, the word file of the results I got, and a facsimile of the column I would like to create. Thank you all! Cordially, Davide Il giorno ven 23 nov 2018 alle ore 21:10 Davide Bertani < dbertani92(a)gmail.com&gt; ha scritto: > thank you very much Sven!! I'll try to adopt the solution you sent me > (maybe with some help from a friend of mine, big computer brain), I'll let > you know if it works! best regards, > Davide > > Il Gio 22 Nov 2018, 10:59 Sven Schreiber <svetosch(a)gmx.net&gt; ha scritto: > >> Am 22.11.18 um 11:46 schrieb Davide Bertani: >> > Thank you guys for the answers >> > @sven I understood the use of the "binar" term: as a matter of fact, >> > my nominal variable is actually binary, spacing from 0 to 1. Problem >> > is, my thesis tutor wants UNIVARIATE as well as multivariate analysis >> > of these data, so I'm forced to estimate all 100 separates (the >> > univariate analysis, as far as I understand) and then I will estimate >> > the 100 variables at once (multivariate analysis if Im not wrong). And >> > Im looking for help for accelerating the First task:) >> >> >> Ah OK, I think I understand. You don't just have B,C,D as regressors but >> actually have 100 (potential) regressors. (First I thought you had 100 >> observations per variable, not 100 variables.) >> >> Then you need to do some programming with gretl's scripting language >> hansl indeed. You could try the following: >> >> <hansl> >> >> list reg = B C D # <and so forth> >> >> # an alternative way is to use a series range as in 'list reg = B..ZZ', >> see the manual ch. 14. >> >> loop foreach r reg >> >> logit A const $r >> >> # use the $coeff (and other) accessor if you want to store the result >> >> endloop >> >> </hansl> >> >> A working example with a built-in dataset would be as follows: >> >> <hansl> >> >> open credscore >> list reg = Age..Selfempl >> loop foreach r reg --quiet >> logit Acc const $r >> endloop >> </hansl> >> >> >> HTH, >> >> sven >> >> _______________________________________________ >> Gretl-users mailing list >> Gretl-users(a)lists.wfu.edu >> http://lists.wfu.edu/mailman/listinfo/gretl-users > >

Am 28.11.2018 um 15:40 schrieb Davide Bertani: I am now facing a second, bigger challenge: with Sven's script, I got the standard errors, p values, O.R. and IC 95% that I needed. Now I have to tabulate them, using Excel or Word. Problem is that the Gretl results file, which I attached to the mail, shows (along with the useful data obtained) many other things that I do not need, lines and lines of text that I should manually delete hundreds of times (the akaike criterion, the schwarz criterion, hannan- quinn etc.). What I need is a series of rows displaying ONLY the ES, P-values and so on, instead of having to search through the list of results that the Gretl logit analysis gave me. DId I manage to explain my problem? does anyone have any suggestions about it? You probably need the --quiet switch (which it seems isn't documented for logit, but works). Plus the $coeff and other $-accessors as I briefly mentioned before. Check this out: <hansl> open credscore list reg = Age..Selfempl loop foreach r reg --quiet logit Acc const $r --quiet printf "First coefficient estimate: %g\n", $coeff[1] printf "1st coeff. std err: %g\n", $stderr[1] matrix tstats = $coeff ./ $stderr printf "1st coeff. t-stat: %g\n", tstats[1] endloop </hansl> Instead of printing it out inside the loop, you could also append to a result matrix one-by-one, and print (or store) the final matrix after the loop. hth, sven _______________________________________________ Gretl-users mailing list Gretl-users(a)lists.wfu.edu http://lists.wfu.edu/mailman/listinfo/gretl-users

On Sun, 2 Dec 2018, Davide Bertani wrote: > Dear Sven; > Thank you again, very very much, for your help. > I tried to do as you suggested, and everything went well until I tried to > add the p value (written as "$pvalue") among the accessors. At that point, > in the results column, under "p value" gretl did not return the required p > value but instead the word "nan" (not a number). The $pvalue accessor is not available for regression coefficients. You need to decide between one-tailed or two-tailed. For a two tailed test this would do the job (since the pvalue() function gives you the right-tail probability): pval = 2 * pvalue(z, abs($coeff[i] / $stderr[i])) That's for coefficient i. Note that coefficient 1 is the constant, so for a single independent variable you probably want i = 2. Allin Cottrell _______________________________________________ Gretl-users mailing list Gretl-users(a)lists.wfu.edu http://lists.wfu.edu/mailman/listinfo/gretl-users