k*****u
发帖数: 1688 | 1 如图所示。有少量的点，residual随着predict增大而增大，是不是我漏掉了某些重要
的变量？ |
A*******s
发帖数: 3942 | 2 y>=0 ?
【在 k*****u 的大作中提到】
: 如图所示。有少量的点，residual随着predict增大而增大，是不是我漏掉了某些重要
: 的变量？
|
h***i
发帖数: 3844 | 3 看着有点像，response都《=0，然后用linear regression fit的？不知是否。
【在 k*****u 的大作中提到】
: 如图所示。有少量的点，residual随着predict增大而增大，是不是我漏掉了某些重要
: 的变量？
|
t****a
发帖数: 1212 | |
p***r
发帖数: 920 | 5 how about add another factor? |
k*****u
发帖数: 1688 | 6 谢谢楼上的几位兄弟。我再继续搞
看看能不能搞明白 |
h****i
发帖数: 79 | 7 It seems to me there is another totally independent linear factor. |
G**7
发帖数: 391 | 8 Read Keutner's book on regression. This case was discussed there. |
G**7
发帖数: 391 | 9 You sent me a message abot the book. Th book is: applied linear statistical
model. One of the athors is keutner. |
f***a
发帖数: 329 | 10 One possibility: irrelevant predictors could cause such problem.
naive example:
y = random error + irrelevant terms + relevant terms
=> residual = y - y_hat = random error + irrelevant terms
if your fitted irrelevant terms are linear, then linear residuals will be
produced for observations whose irrelevant attributes contribute (non-zero). |
