Dear Statalists,
I have read the discussion of negative binomial and poisson in statalist. There is still a confusion about negative binomial regression with fixed effects, and I see that some studies follow Blundell et al.'s (2002) suggestion of using pre-sampled means as an initial condition to represent unobserved heterogeneity. But I am hesitant about how to implement this in stata.
--------------------------------
Here is an example from the study by Li et al. pp. 5071-5073.
Even after controlling for known firm characteristics, some firms are likely to have permanently higher numbers of alliances than others because of omitted firm-specific effects. One interpretation of the firm fixed effect in Equation (2) is that it reflects the “entry level of alliance formation” of each firm, that is, the number of past alliances at the beginning of the panel data sample. We control for firm fixed effects following the “presample mean scaling” method of Blundell et al. (1999), Aghion et al. (2013), and Bloom et al. (2013) specific to regressions involving count data.
--------------------------------
A specific context: y is the number of patents and ypsm is the pre-sample mean of y. Based on my knowledge, I would add ypsm to the model as a normal variable, i.e.
, and I wonder if this is correct?
Perhaps this is a rather newbie question, but ey I want to make sure it is as correct as possible as it is part of my research.
Thanks a lot for your help.
Dezhi
I have read the discussion of negative binomial and poisson in statalist. There is still a confusion about negative binomial regression with fixed effects, and I see that some studies follow Blundell et al.'s (2002) suggestion of using pre-sampled means as an initial condition to represent unobserved heterogeneity. But I am hesitant about how to implement this in stata.
--------------------------------
Here is an example from the study by Li et al. pp. 5071-5073.
Even after controlling for known firm characteristics, some firms are likely to have permanently higher numbers of alliances than others because of omitted firm-specific effects. One interpretation of the firm fixed effect in Equation (2) is that it reflects the “entry level of alliance formation” of each firm, that is, the number of past alliances at the beginning of the panel data sample. We control for firm fixed effects following the “presample mean scaling” method of Blundell et al. (1999), Aghion et al. (2013), and Bloom et al. (2013) specific to regressions involving count data.
--------------------------------
A specific context: y is the number of patents and ypsm is the pre-sample mean of y. Based on my knowledge, I would add ypsm to the model as a normal variable, i.e.
Code:
xtnbreg y ypsm x1 x2 i.year
Perhaps this is a rather newbie question, but ey I want to make sure it is as correct as possible as it is part of my research.
Thanks a lot for your help.
Dezhi
