e****t
发帖数: 766 | 1 when fitting logistic regression,
it indicates that "Model Convergence Status
Complete separation of data points detected. "
So i use exact logistic regression,
but in the output, it indicates
"# indicates that the conditional distribution is degenerate."
what does this mean? What else I can do?
i am thinking to use poisson regression...
any suggestions ?
Thanks in advance!! |
s*r
发帖数: 2757 | 2 http://support.sas.com/rnd/app/papers/exactlogistic2009.pdf
If the permutation distribution is degenerate (has only one value),then the
procedure does not produce any statistics and does not output
the distribution.
ucla has recommended ways to handle this
http://www.ats.ucla.edu/stat/mult_pkg/faq/general/complete_sepa |
z******n
发帖数: 397 | 3 呃，我觉得你说的不大对。
你给的页面里面提到：
What are the techniques for dealing with complete separation or quasi-
complete separation?
... ...
Exact method is a good strategy when the data set is small and the model is
not very large. Below is a sample code in SAS.
proc logistic data = t2 descending;
model y = x1 x2;
exact x1 / estimate=both;
run;
这表明exact logistic regression可以用来解决data complete separation的问题。
但complete separation并不是degenerate
按我的理解，degenerate distribution在exact test里面是指所关心的参数a的充分统
计量T的条件分布是退化的（关于冗余参数b的充分统计量S取条件）。为了做exact
test，需要用permutation或MCMC计算Pr(T|S=s)。在很多情况下，S=s是一个极强的约
束，使得整个空间中，仅有原始观测样本y能够满足S=s，即条件分布Pr(T|S=s)只有一
个取值，称为退化的。
一个例子是参数b对应的变量中包含至少一个连续变量，此时很难找到一个异于原始y的
y.new，使得S=s严格成立。此时exact test失效
the
【在 s*r 的大作中提到】
: http://support.sas.com/rnd/app/papers/exactlogistic2009.pdf
: If the permutation distribution is degenerate (has only one value),then the
: procedure does not produce any statistics and does not output
: the distribution.
: ucla has recommended ways to handle this
: http://www.ats.ucla.edu/stat/mult_pkg/faq/general/complete_sepa
|
s*r
发帖数: 2757 | 4 the ucla page only covers complete separation
degeneration is only a pathological case under exact method
is
【在 z******n 的大作中提到】
: 呃，我觉得你说的不大对。
: 你给的页面里面提到：
: What are the techniques for dealing with complete separation or quasi-
: complete separation?
: ... ...
: Exact method is a good strategy when the data set is small and the model is
: not very large. Below is a sample code in SAS.
: proc logistic data = t2 descending;
: model y = x1 x2;
: exact x1 / estimate=both;
|
z******n
发帖数: 397 | 5 complete separation可以用SVM了:)
【在 s*r 的大作中提到】
: the ucla page only covers complete separation
: degeneration is only a pathological case under exact method
:
: is
|
e****t
发帖数: 766 | 6 Thank you so much for you guys discussion!!!!
very helpful..
I will go back and try firth logistic regression to see if it works.
proc logistic data = t2 descending;
model y = x1 x2 /firth;
run; |
