a*********n
发帖数: 1331 | 1 sampling不可能，因为data是secondary data，不是自己收集的。
记得有统计方法解决selection bias，叫heckman two steps，但是有人提出这方法有
问题 （具体不记得了，哪位能给个reference吗？）
好像记得有人提出用simultaneous equation modeling。但这方法没什么人用啊。难道
也是问题一大堆？（reference? )
现在大家都用什么方法？ |
a*********n
发帖数: 1331 | |
s*******1
发帖数: 146 | |
s*******i
发帖数: 741 | 4 heckman test to get inverse miller ratio
【在 a*********n 的大作中提到】
: sampling不可能，因为data是secondary data，不是自己收集的。
: 记得有统计方法解决selection bias，叫heckman two steps，但是有人提出这方法有
: 问题 （具体不记得了，哪位能给个reference吗？）
: 好像记得有人提出用simultaneous equation modeling。但这方法没什么人用啊。难道
: 也是问题一大堆？（reference? )
: 现在大家都用什么方法？
|
J**Y
发帖数: 34 | 5 Heckman's 2-step approach relies on joint normality assumption on the error
terms. Estimation results could be sensitive to this assumption. |
g*k
发帖数: 1797 | 6 simultaneous equation method actually is the same as the instrumental
variable analysis. Basically you need to find a reasonable instrumental
variable, it has to be related to your independent variable but not
associated with your outcome. Then you will need to run simultaneous two
equations for the analysis to control for selection bias.
Biostatisticians like to use the propensity score to control for selection
bias, but some health services researches don't believe it. |
a*********n
发帖数: 1331 | 7 Thanks! Has Heckman or someone else solved this problem? What method?
error
【在 J**Y 的大作中提到】
: Heckman's 2-step approach relies on joint normality assumption on the error
: terms. Estimation results could be sensitive to this assumption.
|
a*********n
发帖数: 1331 | 8 very helpful!
But I have about 50 independent variables. It is not possible to find such a
variable that is related to all the independent variable but not related to
the outcome. Any solution?
【在 g*k 的大作中提到】
: simultaneous equation method actually is the same as the instrumental
: variable analysis. Basically you need to find a reasonable instrumental
: variable, it has to be related to your independent variable but not
: associated with your outcome. Then you will need to run simultaneous two
: equations for the analysis to control for selection bias.
: Biostatisticians like to use the propensity score to control for selection
: bias, but some health services researches don't believe it.
|
v*******y
发帖数: 5530 | 9 你总不会有50个variables of interest把
a
to
【在 a*********n 的大作中提到】
: very helpful!
: But I have about 50 independent variables. It is not possible to find such a
: variable that is related to all the independent variable but not related to
: the outcome. Any solution?
|
g*k
发帖数: 1797 | 10 In a project you usually should focus on only 1-3 independent variables, and
other variables that you want to control for should be called "covariates",
or called "confounders".
Do you know any biostatistician or eplidemiologiest, or health services
researcher? An experienced health services researcher should be the best
person as a co-investigator in your project. |
u*****n
发帖数: 3277 | 11 factor in this variable: consider it as a new variable because of its
variance.
【在 a*********n 的大作中提到】
: sampling不可能，因为data是secondary data，不是自己收集的。
: 记得有统计方法解决selection bias，叫heckman two steps，但是有人提出这方法有
: 问题 （具体不记得了，哪位能给个reference吗？）
: 好像记得有人提出用simultaneous equation modeling。但这方法没什么人用啊。难道
: 也是问题一大堆？（reference? )
: 现在大家都用什么方法？
|
a*********n
发帖数: 1331 | 12 unfortunately, I do have 50+ independent variables to test. I can only use
about 10 of them as control variables.
【在 v*******y 的大作中提到】
: 你总不会有50个variables of interest把
:
: a
: to
|
