causal forest - Understanding the output, what are pv_values? and a question on variable interactions.

[yjacoby]

Hi,

I am new to using the grf package and am trying to understand the causal forest algorithm.

Looking at the causal forest object, what are the "_pv_values"?

A second question is, other than variable importance, can the causal forest also highlight the most 'important' variable interactions?

Thank you in advance

[erikcs]

Hi @yjacoby, all the attributes with an underscore are internal forest data structures not meant to be accessed by the user (_pv_values are stored prediction values), you can see available functions here

[yjacoby]

Thank you so much for your quick response @erikcs.

In regards to my second question, while predictions from a causal forest give us the treatment effect for each individual with variance estimates. I was wondering if there is a way to gain standard errors and variance estimates on the coefficients of interactions? For example if there are two covariates the coefficient, x_1* x_2 * Treatment? Of course in my case there are more than two covariates so the process of discovering the important interactions is a little more involved.

Apologies if my question is unclear.

[yjacoby]

I have found that this topic was discussed in the thread : #238

Thank you for all of your help.