怎么对系数做T检验啊

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

数据分了2组，健康组和非健康组，做完了LOGIT regression以后，怎么对这两组数据中的每个变量的系数做T检验呢？

万分感谢啊，拜托大家了！

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1 回帖

关键词：t检验 regression regressio regress logit 健康

I think you can compute Wald test, but a better option may be to run the whole data together by including interaction terms using the dummy of the group in the model?

jjjj6666 发表于 2014-3-18 08:31
I think you can compute Wald test, but a better option may be to run the whole data together by incl ...

但是现在需要对系数做T检验，导师没让我做别的检验。我知道怎么对样本做T检验，但是对系数的T检验真是一点都不会啊，您能告诉我从STATA的那里可以做嘛？我用的是LOGIT回归，在那个菜单选项里有关于这个检验的吗？多谢！

you can use this

(b1-b2)/sqrt(se(b1)^2+se(b2)^2)

where b1 is the coefficient from the 1st logit, b2 is the corresponding coefficient from the 2nd logit.  For logit, this will be z-statistics, not t, I think(?)

I still think running the model with interaction term is a better approach.

jjjj6666 发表于 2014-3-19 06:18
you can use this

(b1-b2)/sqrt(se(b1)^2+se(b2)^2)

真的很感谢您的回复，我不怎么熟悉STATA的COMMENTS，您知道怎么用菜单方式实现这个命令吗？多谢！

I think you need to use command window since you need to compute yourself, for example, if g is your group variable with 0 for one of the subgroup and 1 for the others,

* run 1st logit
logit y x if g==0
* record b and SE2
mat B1=e(b)
mat V1=e(V)

* run 2nd logit
logit y x if g==1
* record b and SE2
mat B2=e(b)
mat V2=e(V)

to test the difference of coefficient for x, we compute the p-value

local p=2*(1-normprob(abs(B1[1,1]-B2[1,1])/sqrt(V1[1,1]+V2[1,1])))
di `p'

or do another way, which is better by taking into consideration the estimates are from sub-samples :

logit y x if g==0
est store a
logit y x if g==1
est store b
suest
test [a_y]x = [b_y]x

jjjj6666 发表于 2014-3-20 10:49
or do another way, which is better by taking into consideration the estimates are from sub-samples : ...

我昨天自己又查了些资料，觉得你说的interaction好像也不错的，这个用菜单的形式可以实现吗？

if using menu, under data->create or change data->other variable creation commands->interaction expansion

then depend on if the other variables are continuous, if continuous, select  I.varname1*varname3, select the variable for 健康组/非健康组 in the categorical variable, and the x (continuous) variable in continuous variable selection, click ok.  It will generate new variables named start with _I

Then run the logit reg by include all the new variables named as _I in addition to the variables you already used with robust SE

It will be easy to use command window as (if you do logit Y on x by group):

xi: logit Y x i.group*x, robust

and then test the interaction term

test _IgroXx_1

楼主你好，我想请问下你是怎么解决的，我也想检验两个回归系数是否有差异，我的是面板数据，好像用suest不行，请问怎样做t检验