Varying outcomes of observations with lasso and elasticnet

Lukas Seubert

Join Date: Jul 2022
Posts: 2

Varying outcomes of observations with lasso and elasticnet

13 Jul 2022, 08:51

Dear Stata community members,

I have a question regarding the elasticnet and lasso commands on Stata 17. I ran these codes for elastic net (1), ridge (2) and lasso(3):
1

Code:

elasticnet linear x1 $vldummy $vlcontinuous, alpha(0(0.05)1)

Code:

elasticnet linear x1 $vldummy $vlcontinuous, alpha(0)

Code:

lasso linear x1 $vldummy $vlcontinuous

with x1 being the dependent variable. I am trying to compare the estimates of the 3 methods for variable selection on my dataset.

I ran this command several times and always saved the estimates using:

Code:

estimates store elasticnet

Code:

estimates store elasticnet1

etc.

Then using

Code:

lassogof

I got the following table for all estimates:

Code:

-------------------------------------------------
       Name |         MSE    R-squared        Obs
------------+------------------------------------
 elasticnet |    2.961392       0.2433      5,113
elasticnet1 |     2.93792       0.2308      4,827
elasticnetT |    2.955981       0.2527      3,608
elasticnetT1|    2.950256       0.2541      3,605
elasticnetT2|    2.962259       0.2430      5,113
elasticnetT3|    2.961392       0.2433      5,113
      lasso |    2.961809       0.2432      5,113
     lasso1 |    2.994677       0.2343      5,120
     lassoT |    2.961809       0.2432      5,113
    lassoT1 |    2.961809       0.2432      5,113
    lassoT2 |    2.961809       0.2432      5,113
    lassoT3 |    2.961809       0.2432      5,113
      ridge |    2.874946       0.2577      3,416
     ridge1 |    2.938475       0.2413      3,416
     ridgeT |    2.874946       0.2577      3,416
    ridgeT1 |    2.881256       0.2561      3,416
    ridgeT2 |    2.872305       0.2584      3,416
    ridgeT3 |     2.87792       0.2570      3,416
-------------------------------------------------

My question is: why do the results vary sometimes and are the same other times (e.g. elasticnet and elasticnetT3). In large datasets such as mine it would be very unlikely that the random sample is the same twice. More importantly, why does the number of Obs vary so drastically (What does the Obs number tell me?) for elasticnet? The entire dataset has 10721 observations.

I could not find and info on this via

Code:

help elasticnet

Code:

help lasso

Any help or hints would be greatly appreciated!

Best regards,

Lukas

Tags: data, elasticnet, lasso, lassopack, regression

Announcement

Varying outcomes of observations with lasso and elasticnet