多元回归中type I error 与显著性水平

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

本人是个统计菜鸟，投了一篇论文中用了多元线性回归，两年的数据，每年数据中自变量有5个，因变量有8个，分别作了逐步进入回归分析，两年的数据共作了16次。审稿意见如下：
If I understand the statistical analysis correctly, then separate stepwise regression analyses were conducted for each behavioral outcome variable (active time, retirement-sunset interval, etc.) for each year. If 16 separate analyses were conducted, then if alpha = 0.05, then the chance of type I error is very high, and some of the observed "significant" relationships are likely to be spurious. This problem could be addressed by reducing the alpha level, but this is likely to result in an unacceptably low statistical power, given the small sample sizes.  Alternatively, a smaller set of outcome variables specifically related to the study hypothesis could be chosen (it is unclear how playing time, for example, is related to the energy maximization hypothesis), or a multivariate approach considering several outcome variables simultaneously could be employed.

哪位大神能否指点一下，我不明白他说的解决方案具体是啥。比如减少因变量，感觉即便减少了也不会对剩下的变量结果有影响啊

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：Error 多元回归 type err relationship error

应该把年份作为自变量，做8次就可以了。

如果把多组实验分别做一般的两两比较，会使犯一类错误的概率增大的。如果因变量有8个，也只能做8次多元线性回归。因为年份可以作为因素，放到模型中。所有，论文评阅专家提意见是有道理的。

gxnnhgm66 发表于 2015-7-13 23:35
应该把年份作为自变量，做8次就可以了。

不是很明白，把年份加到自变量里面？但年份不是连续变量啊