Random Effect Model

最新推荐文章于 2022-02-12 14:11:39 发布

数据科学小白

最新推荐文章于 2022-02-12 14:11:39 发布

阅读量1.3k

点赞数

分类专栏：统计推断

本文链接：https://blog.csdn.net/weixin_39174856/article/details/108431809

版权

统计推断专栏收录该内容

23 篇文章 7 订阅

订阅专栏

Since $\sigma^2_u,\sigma_\alpha$ are unknown, we could use Feasible GLS.
The FGLS is equivalent to a quasi-difference transformation:
$Y_{it}-\hat \lambda \overline Y_i=(X_{it}-\hat\lambda \overline X_i)'\beta+(\varepsilon_{it}-\hat \lambda \overline \epsilon_i)$
where
$\hat\lambda=1-[\hat \sigma^2_u/(\hat\sigma^2_u+T\hat \sigma^2_\alpha)]^{1/2}$
then run OLS.
Then we get the Random Effect Estimator(RE).
The intuition: let $\lambda=1-[\sigma^2_u/(\sigma^2_u+T\sigma^2_\alpha)]^{1/2}$
$Cov(\varepsilon_{it}-\lambda\overline \varepsilon_i,\varepsilon_{is}-\lambda\overline \varepsilon_i|X_i)=0$