First of all sorry for the bad title. I picture this is a standard problem, i just don't know how to deal with it

I have the following model

\begin{equation} Y_{ij} = \mu + E_i + \varepsilon_{ij} \end{equation} Where $1 \leq i \leq N$ and $1 \leq j \leq M$. And $E_i \sim \mathcal{N}(0,\sigma^2_E$) and $\varepsilon_{ij}\sim \mathcal{N}(0,\sigma^2_\varepsilon)$ are error terms all independent among each other. Note that $E_i$ appear in many equations and $\varepsilon_{ij}$ in only one.

I would like to know how is the standard way to estimate $\mu$ and the variance of the estimation, after observing all $Y_{ij}$.

My first approach is to estimate by the mean \begin{equation} \hat{\mu} =\frac{1}{MN} \sum_{i=1}^{N}\sum_{j=1}^M Y_{ij} \end{equation} I am able to calculate the variance of $\hat{\mu}$, that yields \begin{equation} \text{Var}(\hat{\mu}) = \frac{\sigma^2_{\varepsilon}}{MN} + \frac{\sigma^2_E}{N} \end{equation} I can manage to estimate $\sigma^2_\varepsilon$ with \begin{equation} \hat{\sigma^2_\varepsilon} = \frac{1}{N(M-1)}\sum_{i=1}^M\sum_{j=1}^N(Y_{ij} - \bar{Y_i})^2 \end{equation} But I am not sure how to estimate $\sigma^2_E$.


These are called random effects or variance components models. The Wikipedia page has the basics and you can search for additional details with the correct terms.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.