Testing assymetry between 2 models with F-test


I would like to know whether it is acceptable to use a general linear F-test, supposed to compare two models (a full model and a restricted one), when the full model is built on the restricted one from dummy variables.

I am working on 2 models:

1) Y = a + b.X + e
2) Y = a1.D + a2.(1-D) + b1.X.D + b2.X.(1-D) + v, with D = 1 if X>0 and 0 otherwise

The model 2 is an existing model that is presented through scientific papers, so the problem shouldn't come from a bad model specification.
There, I considered 2 as the full model and 1 as the restricted model and computed the following F-stat :

Fstat = [(RSS_1 - RSS_2) / (k2-k1)] / [RSS_2 / (n-k2)]

With n the number of observations, k1 the amount of parameters of model 1 (k1=2) and k2 the number of parameters of model 2 (I'm however not so sure whether I should take 3 or 4 because of the double constant...)

I'm doing this for three different sets of data, but for all three of them I am getting a pvalue of 100%, which upsets me.

Am I doing something wrong or are these results somehow possible?
Thanks for your help.
I actually had a problem in the computation of RSS_2 so I solved the problem of pval=1, however would someone still know if I should consider 3 or 4 parameters to model 2?

I also got advised by a university friend to rather use a Davidson-MacKinnon test for non nested models. Would it be better than this F-test?
Last edited: