包含R的线性回归模型中的误差项

时间:2009-11-26 04:07:54

标签: r regression linear

我想知道是否有办法为线性回归模型包含错误术语,如:

r = lm(y ~ x1+x2)

1 个答案:

答案 0 :(得分:4)

代码r = lm(y ~ x1+x2)意味着我们将y建模为x1和x2的线性函数。由于模型不完美,因此会有一个剩余项(即模型无法适应的剩余项)。

在数学方面,正如Rob Hyndman在评论中所指出的那样y = a + b1*x1 + b2*x2 + e,其中ab1b2是常数,e是你的残差(假设是正态分布的。)

要查看具体示例,请考虑R附带的虹膜数据。

model1 <- lm(Sepal.Length ~ Sepal.Width + Petal.Length + Petal.Width, data=iris)

现在我们可以从模型中提取常量(相当于ab1b2,在这种情况下也是b3

> coefficients(model1)
(Intercept)  Sepal.Width Petal.Length  Petal.Width 
1.8559975    0.6508372    0.7091320   -0.5564827

已计算模型中使用的每行数据的残差。

> residuals(model1)
           1             2             3             4             5       
0.0845842387  0.2100028184 -0.0492514176 -0.2259940935 -0.0804994772
# etc. There are 150 residuals and 150 rows in the iris dataset.

(编辑:剪切摘要信息不相关。)


编辑:

您在评论中提到的Error值,在aov的帮助页面上有解释。

If the formula contains a single ‘Error’ term, this is used to
specify error strata, and appropriate models are fitted within
each error stratum.

比较以下内容(改编自?aov页面。)

> utils::data(npk, package="MASS")
> aov(yield ~  N*P*K, npk)
Call:
   aov(formula = yield ~ N * P * K, data = npk)

Terms:
                       N        P        K      N:P      N:K      P:K    N:P:K Residuals
Sum of Squares  189.2817   8.4017  95.2017  21.2817  33.1350   0.4817  37.0017  491.5800
Deg. of Freedom        1        1        1        1        1        1        1        16

Residual standard error: 5.542901 
Estimated effects may be unbalanced

> aov(yield ~  N*P*K + Error(block), npk)
Call:
aov(formula = yield ~ N * P * K + Error(block), data = npk)

Grand Mean: 54.875 

Stratum 1: block

Terms:
                    N:P:K Residuals
Sum of Squares   37.00167 306.29333
Deg. of Freedom         1         4

Residual standard error: 8.750619 
Estimated effects are balanced

Stratum 2: Within

Terms:
                        N         P         K       N:P       N:K       P:K Residuals
Sum of Squares  189.28167   8.40167  95.20167  21.28167  33.13500   0.48167 185.28667
Deg. of Freedom         1         1         1         1         1         1        12

Residual standard error: 3.929447 
1 out of 7 effects not estimable
Estimated effects may be unbalanced