I developed a conditional logit model in Stata. The model is good and the variables are highly significant. Then I do mfx predicted (PU0) to determine the marginal effects of variables $\frac{dy}{dx}$. The problem is when I get the table of the marginal effects of variables in Stata: the significance of the variables becomes very small and insignificant. Why and therefore I can comment or not the marginal effects?
Example coefficients of smoker 0.6666 and the significance P value 0.000 but when I put mfx compute to have $\frac{dy}{dx}$ smoker marginal effect is 1.2 and significance is $p= 0.521$.