When using margins to evaluate the effect of a continuous variable in a non-linear model (e.g., logit, nbreg), it is common to look at the marginal effect over a range of values of such variable. On occasion, I find that the confidence interval (95) of the predicted marginal effect of a very low value of the variable in question overlaps with the confidence interval (95) of the predicted marginal effect of a very high value of the same variable. This is the case even though the coefficient of the variable in the model is highly significant (p-value .01) and has a confidence interval with a small range that is different than 0. So, if a plot the results with marginsplot, for example, I can see that the lower end of the CI for a high value of the variable overlaps with the higher end of the CI for a low value of the variable. Can I still interpret the statistically significant coefficient in the model as implying that low and high values of the continuous variable are different?
0 Response to Confidence intrevals and marginal effects significance for continuos variables in non-linear models
Post a Comment