F****n
发帖数: 3271 | 1 Suppose I have a linear model of this form:
y = [1/(1+b)]x1 + [c/(1-b)]x2 +e
1. I can identify b,c by estimating 1/(1+b) and c/(1-b) using OLS, right?
2. If so, is it possible to calculate standard errors for b and c, as well
as their p value?
Thanks! | c*******2
发帖数: 8 | 2 You can try delta method to find the limiting distribution and thus obtain
the C.I., P-values etc;
1. Basically, you can reparameterize your model as:
Y=A*x1 + B*X2+e
Then your original parameter b and c has a one-to-one function that maps to
the new parameter A and B. You can get estimation of A and B based on MLE (
or OLS), then also MLE of b and c
2. Using Delta method to drive limiting distribution of b and c, then
conduct test or construct intervals based on the distribution. | F****n
发帖数: 3271 | 3 Thanks! Is it possible to directly obtain SEs from MLE by treating b and c
as parameters and modify the calculation of log likelihood?
to
【在 c*******2 的大作中提到】
: You can try delta method to find the limiting distribution and thus obtain
: the C.I., P-values etc;
: 1. Basically, you can reparameterize your model as:
: Y=A*x1 + B*X2+e
: Then your original parameter b and c has a one-to-one function that maps to
: the new parameter A and B. You can get estimation of A and B based on MLE (
: or OLS), then also MLE of b and c
: 2. Using Delta method to drive limiting distribution of b and c, then
: conduct test or construct intervals based on the distribution.
| G**7
发帖数: 391 | 4 If you are using any software, they all produce the estimated beta and
estimated standard error. |
|
