如何比较不同logistic回归模型中所得OR值？

是 否 +2 论坛币

k人 参与回答

经管之家送您一份

应届毕业生专属福利!

求职就业群

赵安豆老师微信：zhaoandou666

经管之家联合CDA

送您一个全额奖学金名额~ !

立即领取

感谢您参与论坛问题回答

经管之家送您两个论坛币！

+2 论坛币

向大家请教一个问题，如何比较不同logistic回归模型中所得OR值是否存在显著差异？非常感谢，^_^

问题的背景是：假设有一组有冠心病病史的病人，得心梗的患者相对于没有得心梗的患者服用某种药物的OR值经多因素调整之后为OR1。另外一群没有冠心病病史的病人中，得心梗的患者相对于没有得心梗的患者服用某种药物的OR值经相同的多因素调整之后为OR2。现在想比较在不同的这两组人群中的，多因素调整之后的OR1和OR2是否有区别？假设两组人群的人数分别为n1和n2，对应的多因素调整的OR1和OR2标准差分别为SE1和SE2，不知道大家是否知道相关方法？非常感谢？^_^

扫码加我 拉你入群

请注明：姓名-公司-职位

以便审核进群资格，未注明则拒绝

分享0 收藏0 回帖

关键词：Logistic回归模型 logistic回归 logistic ogistic logisti 冠心病 标准差 模型 如何

ｍａｒｋ一下，
等我忙完手头的事　想一想

So here you are making inference on comparing the means of two populations, using two samples.
In this case, it's reasonable to assume the two samples are independent, right?

Usually the estimator for the mean follows normal distribution, and we do can do a z-test, if the variance is known, or a t-test, if the variance is unknown.
H0: mu1-mu2=0
z = (xbar1-xbar2-0) /square root of (Var1 + Var2)

But in this case, the estimator for the mean(Odds Ratio) follows lognormal distribution, right?
We do something similar to the above, but slightly different. Hypothesis test for lognormal distribution is not common. I can refer you to a 2003 paper which is not long, only 19 pages.
It answers your question on page 9-10.

K. Krishnamoorthy and  Thomas Mathew (2003), "Inferences on the means of lognormal distributions using generalized p-values and generalized confidence intervals", Journal of Statistical Planning and Inference 115 (2003) 103 – 121

Hope this helps.

I met with my stat professor.

1. Can take the shortcut by taking log of the lognormal distributed random variables (Odds Ratio in this case) and doing a z-test or t-test?
If the two population variances are unknown and we only have estimates of the two variances, which is a typical case, then we cannot take the shortcut by taking log and doing a t-test.
If the two population variances are known, then it's OK to do so with slight modification.

2. In case of logistic model, there is a built in command in SAS to do hypothesis test for comparing the means of two populations. But he cannot recall the command. So check SAS user guide if you don't want to go through the paper I referred.

楼上高手，学习！
请否告知是哪个学校的？

misslost 发表于 2011-9-22 08:25
I met with my stat professor.

1. Can take the shortcut by taking log of the lognormal distribute ...

非常感谢你的回答，但是感觉似乎并没有直接解决我的问题，不知能否说得更为详细，^_^

moonstone 发表于 2011-9-25 11:31
非常感谢你的回答，但是感觉似乎并没有直接解决我的问题，不知能否说得更为详细，^_^

你没有population variance，不能做z test， 我解释了也没用。——或者等我忙完手头的活再说。

到目前为止，我只能告诉你，答案就在我提的2003paper 9-10页。
LZ只能您自己花时间去看了。

还有个答案在SAS user guide 里有，可是在哪里，我就不清楚。
也许你可以找个精通SAS logistic model的人问问。

stata18 发表于 2011-9-23 08:57
楼上高手，学习！
请否告知是哪个学校的？

多谢，不必了吧

感谢回复！

OR值是优势比logistic回归中，OR值=1，表示该因素对疾病的发生不起作用，OR值大于1，表示该因素是一个危险因素，OR值小于1，表示该因素是一个保护因素