Timeline for Non-linear omitted term in Poisson vs OLS regression
Current License: CC BY-SA 4.0
6 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Oct 16 at 13:03 | history | edited | lippi | CC BY-SA 4.0 |
added 1 character in body
|
| Oct 16 at 12:57 | comment | added | lippi |
Thank you very much. I have now clarified that the difference is observable even if an individual z term is added to the model. In fact, the estimate of this term's coefficient will also be biased. As for the range, the bias is observable for both positive and negative x values. In my simulations x follows a standard normal distribution.
|
|
| Oct 16 at 12:54 | history | edited | lippi | CC BY-SA 4.0 |
added 42 characters in body
|
| Oct 16 at 12:48 | comment | added | EdM |
What's the range of your x values? Your f(x) can be close to linear for x<0. Also, your models both omit an individual term for z. That's often not a good idea and might be contributing to a different c coefficient for the x:z interaction that you would see if you included an individual z term.
|
|
| Oct 16 at 10:22 | history | edited | lippi | CC BY-SA 4.0 |
added 97 characters in body
|
| Oct 16 at 10:06 | history | asked | lippi | CC BY-SA 4.0 |