I have a model with 13 independent variables (all of them are significant and 2 of them are categorical variables) and 678 observations. I was wondering if there is any way to check the coefficients stability. My approach is to start from the model with one variable and then find that coefficients. Then I’ll add another variable by subset selection method and find the new variables and coefficients. I’ll continue until that all variables are in the final model. Hence, I have 15 models (2 categorical variable have 3 levels each) which have different variables (1 to 15 variables including the categorical variables levels). I was wondering if I can show all the coefficients in one graph. For instance, for X1 which is in all models I’ll have 15 coefficients, for X2 which I’ll have 14 coefficients etc.

Any feedback about how can I have all the coefficients in one graph and check their variation for all models or any other approach that help me to check the Regression coefficient stability would be grateful. Thanks.

  • $\begingroup$ What do you mean by "coefficient stability"? $\endgroup$
    – 3x89g2
    Sep 3, 2016 at 7:44
  • $\begingroup$ It means the coefficients have the same sign and almost are smooth when new variables are added to the model (not having a radical change). If the correlational structure is not robust (not stable) by adding any new variable the variable coefficient and its sign will change. $\endgroup$
    – Amir
    Sep 5, 2016 at 2:52


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.