Subscribe Bookmark RSS Feed

Adding Constraints to Coefficients in Regression Models

prossman

Community Trekker

Joined:

Jul 30, 2014

I am working on a market mix model and the impact of promotions on sales. I would expect the coefficients to be positive or 0, but some of the models are producing negative coefficients.

Is there a way to add a constraint to the coefficients of some variables to force them to be positive?


Thanks!

Paul

1 ACCEPTED SOLUTION

Accepted Solutions
kevin_c_anderso

Community Trekker

Joined:

Jun 5, 2014

Solution

Paul, you can use the Parameter Bounds pulldown in the Nonlinear platform, specify the bounds you wish, and then iteratively fit a linear model.

The Nonlinear platform will make you specify a model in the Formula Editor and supply some Initial Values for the parameters.  You won't enjoy some of the model diagnostics you might be using in another platform (R^2, for instance, is not provided for nonlinear models), but at least you can constrain the parameters appropriately and (if your model converges) get a hopefully useful answer.

2 REPLIES
kevin_c_anderso

Community Trekker

Joined:

Jun 5, 2014

Solution

Paul, you can use the Parameter Bounds pulldown in the Nonlinear platform, specify the bounds you wish, and then iteratively fit a linear model.

The Nonlinear platform will make you specify a model in the Formula Editor and supply some Initial Values for the parameters.  You won't enjoy some of the model diagnostics you might be using in another platform (R^2, for instance, is not provided for nonlinear models), but at least you can constrain the parameters appropriately and (if your model converges) get a hopefully useful answer.

prossman

Community Trekker

Joined:

Jul 30, 2014

Thanks. This worked perfectly.