Version: Stata 16
Dear Statalisters,
I have difficulties with testing my interaction term in a fixed effects negative binomial regression model. The interaction term is omitted due to collinearity. I understand why it is omitted, however I was wondering wether there is an alternative for my moderator “experience”.
Here are some further information regarding variables and dataset:
I’m testing whether developers exiting a project (IV: “exit”) have a negative effect on project code quality (patch acceptance rate) (DV: “patch_ac”). I have an unbalanced panel dataset comprising 1,106 projects (12,187 observations). I xtset the data using “project” as panel variable and “release_index” as time variable. My IV “exit” is coded as dummy variable (1 = developer exit takes place; 0 = no developer exit). As my IV “patch_ac” is a count variable I’m using a negative binomial regression (fixed effects). All in all there are 204 instances where a developer exits a project.
Furthermore, I’d like to test whether “exiting developer experience” is moderating the effect between “exit” and “patch_ac”. The higher experienced an exiting developer is the more the negative relationship between “exit” and “patch_ac” is strengthened. “experience” is measured as number of releases a developer was active in a project. Whenever a developer exits his project (exit=1) my moderator “experience” has a value ranging from 1 to 36, however if no developer exit occurs (exit=0) the variable “experience” is also 0 as there is no loss in experience in that project. When I now want to test the interaction term "exit x experience” it is omitted due to collinearity as my moderator “experience” and my interaction term “exit x experience” are identical.
Do you have any idea for how to logically adjust my moderator “experience” that I can use it for my regression?
I am new to these topics in statistics, so I am sorry for this basic question.
Thank you very much in advance.
Maja
Dear Statalisters,
I have difficulties with testing my interaction term in a fixed effects negative binomial regression model. The interaction term is omitted due to collinearity. I understand why it is omitted, however I was wondering wether there is an alternative for my moderator “experience”.
Here are some further information regarding variables and dataset:
I’m testing whether developers exiting a project (IV: “exit”) have a negative effect on project code quality (patch acceptance rate) (DV: “patch_ac”). I have an unbalanced panel dataset comprising 1,106 projects (12,187 observations). I xtset the data using “project” as panel variable and “release_index” as time variable. My IV “exit” is coded as dummy variable (1 = developer exit takes place; 0 = no developer exit). As my IV “patch_ac” is a count variable I’m using a negative binomial regression (fixed effects). All in all there are 204 instances where a developer exits a project.
Furthermore, I’d like to test whether “exiting developer experience” is moderating the effect between “exit” and “patch_ac”. The higher experienced an exiting developer is the more the negative relationship between “exit” and “patch_ac” is strengthened. “experience” is measured as number of releases a developer was active in a project. Whenever a developer exits his project (exit=1) my moderator “experience” has a value ranging from 1 to 36, however if no developer exit occurs (exit=0) the variable “experience” is also 0 as there is no loss in experience in that project. When I now want to test the interaction term "exit x experience” it is omitted due to collinearity as my moderator “experience” and my interaction term “exit x experience” are identical.
Code:
xtnbreg patch_ac i.exit##c.experience i.release_index, fe
I am new to these topics in statistics, so I am sorry for this basic question.
Thank you very much in advance.
Maja
