Hi all,
I have a panel of households over 8 years. The dependent variable is a binary response, 0 or 1. The main explanatory variables are dummies for quintiles of wealth. I would like to run a logit regression with household fixed effects, controlling for year fixed effects and time variant controls such as age, etc.
When I run the simple xtlogit model without year dummies and controls, the regression works fine.
xtset hh_id year
xtlogit y i.wealth, fe
However, when I include year dummies, the maximum likelihood will drop to 0 in the 2nd iteration and it says "not concave". This can go on for 300 iterations and the estimation never converges. The same problem exist when controlling for age.
Do you have any idea why this happens and how I can solve this?
Just one description of the panel structure, all the households have similar pattern of response. They start with a 0 response (in an arbitrary year) followed by some more 0 responses (or none), and all end with a positive response of 1. However, the start and end years and age values can vary for different households.
Thanks for your help in advance,
Fatima
I have a panel of households over 8 years. The dependent variable is a binary response, 0 or 1. The main explanatory variables are dummies for quintiles of wealth. I would like to run a logit regression with household fixed effects, controlling for year fixed effects and time variant controls such as age, etc.
When I run the simple xtlogit model without year dummies and controls, the regression works fine.
xtset hh_id year
xtlogit y i.wealth, fe
However, when I include year dummies, the maximum likelihood will drop to 0 in the 2nd iteration and it says "not concave". This can go on for 300 iterations and the estimation never converges. The same problem exist when controlling for age.
Do you have any idea why this happens and how I can solve this?
Just one description of the panel structure, all the households have similar pattern of response. They start with a 0 response (in an arbitrary year) followed by some more 0 responses (or none), and all end with a positive response of 1. However, the start and end years and age values can vary for different households.
Thanks for your help in advance,
Fatima
