变量很显著但估计系数过大怎么办？

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

小妹求教：回归结果所关心的估计系数很显著，但系数值过分偏大，是什么原因引起的呢，如何改进？多谢！

直觉上讲，我所关心的变量对因变量的影响程度大致也就是10%~20%，但回归结果却达到0.78+，不知道是哪错了，望大侠指点指点，感激不尽！！


另外，在半对数模型中，当解释变量表示的是某两个值的比重时，其回归后的系数又如何解释呢？

多谢多谢！！

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2 回帖

关键词：怎么办 半对数模型 是什么原因 回归结果 感激不尽 怎么办 系数

sysuse auto
gen foreign1=foreign*1000

regress price foreign
regress price foreign1


hope you can see the difference.

大白菜2012 发表于 2012-10-23 13:08
sysuse auto
gen foreign1=foreign*1000

thank you very much for your kindly reply!

But how this changes related with my question? I am look forward to hope you can give me a detailed explanation!

thanks!

zhienboai 发表于 2012-10-23 14:43
thank you very much for your kindly reply!

But how this changes related with my question? I am  ...

my sample answered your first question.

for example, when you define a time measure, say, year, you can make it as 1999, 2000, 2001, and 2002. But you have the other option, to make it as 1,2,3,and 4.

the coefficient of year in model will be changed but not for the coefficients of other variables.

zhienboai 发表于 2012-10-23 14:43
thank you very much for your kindly reply!

But how this changes related with my question? I am  ...

when you have perfect prediction issue in your model, you will see this phenomenon too: very large coefficent but it is statistically significant.

http://www.ats.ucla.edu/stat/mult_pkg/faq/general/complete_separation_logit_models.htm

非常感谢！