h**l 发帖数: 4883 | 1 I am trying to compare the academic performance of 2 groups of students. the
dependent variable is the “number of passes on assessments”. The students
take a varied number of assessments ranging anywhere from 5 to 12. I think
it's a binomial distribution. But would Poisson regression work? What are
the essential differences between Poisson and Binomial? Are they equivalent
in this case? |
s*********e 发帖数: 1051 | 2 do you mean "negative binomial" ^_^?
standard Poisson definitely won't work, since your outcomes range from 5 to
12. however, truncated poisson might. |
h**l 发帖数: 4883 | 3 Thanks. So basically students can fail maximally 5 tests and then they'll be
kicked out.
So the study will have 12 tests, most of the students will fail less than 5
tests so the odds ratio would be # of failed tests/ 12. But for those who're
kicked out, the odds ratio would be 5/# tests they take. In this case, it's
negative binomial?
well, so I think I should use proc GENMOD? how do you perform logistic
regression of negative binomial variables?
Thanks a lot.
to
【在 s*********e 的大作中提到】 : do you mean "negative binomial" ^_^? : standard Poisson definitely won't work, since your outcomes range from 5 to : 12. however, truncated poisson might.
|
s*********e 发帖数: 1051 | 4 what exactly is the objective of your study?
with binomial, you are trying to predict the probability to pass all 12 exams.
with poission / nb, you are trying to predict the number of exams passed. |
h**l 发帖数: 4883 | 5 The objective of the study is to compare the academic performance of 2
groups of students. The tricky part is that there are students who failed 5
tests and dropped out immediately after they failed 5. We don't want to
exclude these students who didn't have chance to take all 12 exams.
exams.
【在 s*********e 的大作中提到】 : what exactly is the objective of your study? : with binomial, you are trying to predict the probability to pass all 12 exams. : with poission / nb, you are trying to predict the number of exams passed.
|
b***t 发帖数: 348 | |
h**l 发帖数: 4883 | 7 well, each test can only have two outcomes: pass and fail... it's just that
the students can have different number of tests depending on if they pass or
not, and the number has a minimum of 5 and maximum of 12.
【在 b***t 的大作中提到】 : multinomial
|
x*******i 发帖数: 1791 | 8 multimomial. 我也认为是。
the number of passed in some fix number of total test.
不是poisson。 poisson没有上线。 |
x*******i 发帖数: 1791 | 9 这个不难,用classical方法,做一个link function 就可以了。 |
h**l 发帖数: 4883 | 10 no...
Multinomial has k outcomes. When k=2, multinomial=binomial.
So binomial is like flipping a coin
and multinomial is like throwing a die.
【在 x*******i 的大作中提到】 : multimomial. 我也认为是。 : the number of passed in some fix number of total test. : 不是poisson。 poisson没有上线。
|
|
|
n*******m 发帖数: 101 | 11 how about poisson for rate? |
n*******m 发帖数: 101 | 12 BTW, if a student quits whenever he fails 5 times, it's no longer a binomial
distr or multinomial. Use the negative binomial as the link will work. |
h**l 发帖数: 4883 | 13 Does negative binomial have infinite tests until the students fail 5 test?
binomial
【在 n*******m 的大作中提到】 : BTW, if a student quits whenever he fails 5 times, it's no longer a binomial : distr or multinomial. Use the negative binomial as the link will work.
|
W**********E 发帖数: 242 | 14 我想你是不是可以用POISSON模型,但用不同的OFFSET(parameter is not just lamda
but lamd/offset)。 Y/OUTCOME就是错误的次数,OFFSET 就是总共经历过的测试。打
个比方,Y=5, OFFSET=5=>5次测试内发生5次错误的概率; Y=5, OFFSET=10=>10次测试
内发生5次错误的概率;Y=1, OFFSET=12=>12次测试内发生1次错误的概率。 |
n*******m 发帖数: 101 | 15 樓上正解。log(mu_i/t_i)=X*beta->mu_i=t_i*exp(X*beta),把t_i帶入就行了。 |
x*******i 发帖数: 1791 | 16 没明白为什么是5-12之间的数? 1-4为什么不行?说中文吧。说实话就没明白你想干啥。
multinomial 和binomial一回事。 |