Lassologit error

[Said Kaawach]

Hello everyone,

I am having the following error when running lassologit

lassologit loan_close `all' `fe', postlogit
predict double prob_loan_close if e(sample), pr postlogit  lic(ebic)

Error:

. predict double prob_loan_close if e(sample), pr postlogit  lic(ebic)
Use lambda=118.077676822632 (selected by EBIC).
    st_matrixcolstripe():  3300  argument out of range
          ereturn_beta():     -  function returned error
           PostResults():     -  function returned error
                 <istmt>:     -  function returned error

Has anyone encountered this error before? If so, what should I do? I tried installing, updating stata as suggested before on statalist. Also, I have installed the lassopack and still got this error
Thank you for your time.

[William Lisowski]

No experience here, but in looking into lassologit I notice some differences between your syntax and that described in the output of the help lassologit command.

    Prediction

    After selecting a model, we can use predict to obtain predicted
    probabilities or linear predictions.

    First, we select a model using lic() in combination with
    postresults as above:
        . lassologit v58 v1-v57
        . lassologit, lic(ebic) postresults

    Then, we use predict:
        . predict double phat, pr
        . predict double xbhat, xb

    pr saves the predicted probability of success, and xb saves the
    linear predicted values.

    Note that the use of postresults is required.  Without
    postresults, the results of the estimation with the selected
    penalty level are not stored.

I also note that Stata 16 has built in lasso commands which may accomplish the same thing as the lassopack routines.

[Said Kaawach]

No experience here, but in looking into lassologit I notice some differences between your syntax and that described in the output of the help lassologit command.

Prediction

After selecting a model, we can use predict to obtain predicted
probabilities or linear predictions.

First, we select a model using lic() in combination with
postresults as above:
. lassologit v58 v1-v57
. lassologit, lic(ebic) postresults

Then, we use predict:
. predict double phat, pr
. predict double xbhat, xb

pr saves the predicted probability of success, and xb saves the
linear predicted values.

Note that the use of postresults is required. Without
postresults, the results of the estimation with the selected
penalty level are not stored.

I also note that Stata 16 has built in lasso commands which may accomplish the same thing as the lassopack routines.

Thank you Professor William. I believe that both are correct.

I used the command that you provided me with and I got the same exact error.

. lassologit, lic(ebic) postresults
Use lambda=118.077676822632 (selected by EBIC).
    st_matrixcolstripe():  3300  argument out of range
          ereturn_beta():     -  function returned error
           PostResults():     -  function returned error
                 <istmt>:     -  function returned error

Thank you again for your fast response

[Said Kaawach]

Update: I found missing values (they equaled to ".") in one of my variables. I removed the missing values and the code ran normally.

[Achim Ahrens]

This has been fixed with Said's help. Many thanks to him for pointing this out.

If you encounter this error, please update lassologit using

net install lassopack, from("https://raw.githubusercontent.com/statalasso/lassopack/master/") replace

to version 1.4.2 (or above).

[Federico Antonioli]

Dears,

a silly question probably.

I am running one logit lasso for each year of a panel.

I am not interested in the post-estimation logit, but rather on the variables the lasso selects for each year (i.e., whether the post logit converges or not is not an issue now).

From the help file I do not see any macro containing the selected variables (i.e., the macro e(selected) available for rlassologit and lassologit for instance).

A second option is to use the Stata18 lasso command (selected variables are stored into the macro e(allvars_sel)), but there the issue is that when the post estimation does not converge the loop I wrote stops working (see below the code I used, maybe there's something missing or incorrect).

forvalues i=2004(1)2020 {
lasso logit dep_var var1 var2 varN if year==`i', selection(cv, folds(5)) nolog 
estimate store logit`i'
g sel_probit`i'=e(allvars_sel)
di e(allvars_sel)
if _rc == 0 {
continue
}
}

Many thanks,

Federico

[An Dao]

Dears,

I would like to apply lasso for my model. Unforturnately, my dependent variable has zero inflated distribution. I wonder whether anyone of you experienced such issue and share code with me? Thank you in advance. An Dao