[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
Re: st: FW: residuals after conditional fixed effects negative binomial regressi
I am not very familiar with the specifics of -xtnbreg-, but I could
give you an answer if the question is "why do I not get residuals like
the ones you get with `normal' OLS regression?".
If you regress income on education, you will get parameters with which
you can predict someones income if you know someones education. The
difference between these predictions and observed income is the
residual. This is possible because the thing you model and predict has
the same unit as what you observe.
With negative binomial (and poisson) regression you observe a count
(number of events) and you model and predict a rate (number of events
per unit of time or space). Similarly, with logistic regression you
observe whether or not some event occured and you model and predict
the probability of the event occuring. Here you cannot get residuals
because what you observe and what you predict do not have the same unit.
The type of residuals used most often for these types of models are
deviance residuals and pearson residuals and I have no idea why STATA
will not give them after -xtnbreg-. It also does not give them after
-nbreg- or -poisson- (the count models for non-panel data), but they
can be obtained after estimating these models with -glm-.
I hope this partial answer helps,
--- "Jose A. Aleman" <jaaleman@p...wrote:
> why is it that I cannot compute residuals after estimating a model
> with xtnbreg?
* For searches and help try: