# Lasso on squared parameter

Assume a linear regression problem where I want to force sparsity of some parameters. However, due to some physics, I know that one of my parameters is always positive. For instance, I have that

$$y=\sum \beta_ix_i+\epsilon$$ where $\beta_5\geq0$

Is it safe to find the parameter estimates through maximizing the penalized likelihood below while just adding the constraint $\beta_5\geq0$

$$l_p=l(\boldsymbol\beta)+\lambda \sum |\beta_i|$$

By safe I mean, can we still interpret the sparsity results the same way we do in the lasso and if yes why, is there another way to do it using an $l_1$ norm, or does this minimization retain the lasso properties at the MLE.

• What are the "lasso properties at the MLE"? – AdamO Apr 9 '18 at 2:23
• @AdamO consistency in estimation and selection, under some regularity conditions first introduced by Tibshirani and then expanded by Hui Zou in 2006. – Wis Apr 9 '18 at 2:25
• Wis, There was a related question recently, but I cannot find it anymore. Did you delete it? – Richard Hardy Apr 9 '18 at 5:24
• @RichardHardy yes I did, it was put on hold as it was unclear. – Wis Apr 9 '18 at 5:25

## 1 Answer

The MLE estimate that you get for your parameters will not include the prior you have of B5 being positive. However unless your model has a hard time fitting and you end up having negative estimates for B5, it's unlikely that not introducing that prior will change your interpretation in any meaningful way - it might just mean that your confidence intervals would have been slightly different if the prior had been there.

• Could you please clarify this answer so that I can accept it. – Wis Apr 19 '18 at 3:37
• What do you mean by "clarify" ? – f.g. Apr 22 '18 at 13:54