3
简单线性回归中回归系数的导数方差
在简单的线性回归,我们有y=β0+β1x+uy=β0+β1x+uy = \beta_0 + \beta_1 x + u,其中u∼iidN(0,σ2)u∼iidN(0,σ2)u \sim iid\;\mathcal N(0,\sigma^2)。我导出的估计: β1^=∑i(xi−x¯)(yi−y¯)∑i(xi−x¯)2 ,β1^=∑i(xi−x¯)(yi−y¯)∑i(xi−x¯)2 , \hat{\beta_1} = \frac{\sum_i (x_i - \bar{x})(y_i - \bar{y})}{\sum_i (x_i - \bar{x})^2}\ , 其中x¯x¯\bar{x}和y¯y¯\bar{y}是的样本均值xxx和yyy。 现在,我想找到的方差β 1。我衍生像下面这样: 无功(^ β 1)= σ 2(1 - 1β^1β^1\hat\beta_1Var(β1^)=σ2(1−1n)∑i(xi−x¯)2 .Var(β1^)=σ2(1−1n)∑i(xi−x¯)2 . \text{Var}(\hat{\beta_1}) = \frac{\sigma^2(1 - \frac{1}{n})}{\sum_i (x_i - \bar{x})^2}\ . 推导如下: Var(β1^)=Var(∑i(xi−x¯)(yi−y¯)∑i(xi−x¯)2)=1(∑i(xi−x¯)2)2Var(∑i(xi−x¯)(β0+β1xi+ui−1n∑j(β0+β1xj+uj)))=1(∑i(xi−x¯)2)2Var(β1∑i(xi−x¯)2+∑i(xi−x¯)(ui−∑jujn))=1(∑i(xi−x¯)2)2Var(∑i(xi−x¯)(ui−∑jujn))=1(∑i(xi−x¯)2)2×E⎡⎣⎢⎢⎢⎢⎢⎢⎛⎝⎜⎜⎜⎜⎜∑i(xi−x¯)(ui−∑jujn)−E[∑i(xi−x¯)(ui−∑jujn)]=0⎞⎠⎟⎟⎟⎟⎟2⎤⎦⎥⎥⎥⎥⎥⎥=1(∑i(xi−x¯)2)2E⎡⎣(∑i(xi−x¯)(ui−∑jujn))2⎤⎦=1(∑i(xi−x¯)2)2E[∑i(xi−x¯)2(ui−∑jujn)2] , …