# How does hypothesis testing with multiple regression relate to a reduced best subsets model?

Let's say we have the model:

$$Y_i=\beta_0+\beta_1 a_1+\beta_2 b_2+\beta_3 c_3+\beta_4 d_4+\beta_5 e_5+\beta_6 f_6+\beta_7 g_7$$

The goal is descriptive - we want to find out which variables $$x$$ have a relationship with $$Y$$. For each variable $$(a,b,...,f,g)$$, we are testing the null hypothesis that $$\beta=0$$.

Now suppose we use the best subsets technique to reduce the model and find that an optimal reduced model is, say:

$$Y_i=\beta_0+\beta_2 b_2+\beta_5 e_5+\beta_6 f_6$$

We'll still have a model output with coefficients and p-values to conduct hypothesis testing for these $$\beta$$ values. But what about the discarded ones from the full model that are not in the reduced model? I'm pretty sure you can't just conclude they're insignificant just because they're not in the reduced model.

Or should you conduct the hypothesis testing on the full model, and then for the sake of forming a simple model you conduct best subsets?