- UID
- 117239
- 热情
- 15
- 人气
- 6
- 主题
- 1
- 帖子
- 939
- 精华
- 0
- 积分
- 486
- 分享
- 0
- 记录
- 0
- 相册
- 0
- 好友
- 0
- 日志
- 0
- 在线时间
- 504 小时
- 注册时间
- 2007-5-8
- 阅读权限
- 20
- 最后登录
- 2014-11-14
升级 95.33% - UID
- 117239
- 热情
- 15
- 人气
- 6
- 主题
- 1
- 帖子
- 939
- 精华
- 0
- 积分
- 486
- 阅读权限
- 20
- 注册时间
- 2007-5-8
|
去STATS 20X的FORUM看了下~~~~
Stats! Feb 5 2008, 11:02 AM Post #1
For question three i got weak evidence over necessity for a quadratic term, however it is made so difficult by the fact there are nearly ten variables. Should I attempt to fit a quadratic?
mike Feb 5 2008, 11:23 AM Post #2
No.
0000 Feb 5 2008, 03:55 PM Post #3
would it be any harm to put a quadrtic term on the model if it improves it? I have tried to put a quadrtic term to the model and made residual plot of the fitted values look better than it did.
(after the model building process, the p-value of no quadratic term needed is significant at the 5% level, so I thoght it was a good idea to fit a quadrtic term on the model.)
mike Yesterday, 09:44 AM Post #4
The residual plot shows no real need for a quadratic term.
guest Yesterday, 07:23 PM Post #5
but its way better if you log(height) ?
QUOTE(mike @ Feb 7 2008, 09:44 AM)
The residual plot shows no real need for a quadratic term.
mike Today, 10:07 AM Post #6
residual plot from the initial model shows constant scatter, so why would you log the response????????
Remember, a good model is a simple model.
这么看的话。。就是两个都不需要做咯?? |
|