g******2
发帖数: 234 | 1 Suppose y_i is the response variable constrained in the interval [a,b], and
I'm interested in modeling y_i against some predictors x_1i, x_2i, etc. If
the true underline model is y_i=min(max(a,x_i*beta+e_i),b), where e_i is
Gaussian error, is there any way to estimate beta without bias?
Any suggestions or references will be very helpful.
Thanks! | b*****h
发帖数: 95 | 2 Is it about censored regression? Tobit?
Or using conditional MLE since the error is Gaussian. | k*******a
发帖数: 772 | 3 可以用MLE吧， 假设phi(.)是n(X_i*beta, sigma^2)的CDF f(.)是pdf
如果 y_i=a, likelihood = phi(a)
如果 y_i=b, likelihood = 1 - phi(b)
如果 a < y_i < b, likelihood = f(y_i)
这个和interval censored survival 很类似...
然后optimize这个likelihood | g******2
发帖数: 234 | 4 Thanks for the reply!
Is there any R package dealing with this kind of regression? All I found on
web are for survival.
Thanks! | k*******a
发帖数: 772 | 5 write down the likelihood function
optimize using a optimization procedure in R such as optim or some thing
else
on
【在 g******2 的大作中提到】
: Thanks for the reply!
: Is there any R package dealing with this kind of regression? All I found on
: web are for survival.
: Thanks!
|
|
