%\VignetteIndexEntry{Pooled location and scale estimators}
%\VignetteKeyword{rQCC}
%\VignetteKeyword{pooled estimator}
%\VignetteKeyword{unbalanced}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\documentclass{article}[12pt]
\usepackage{Sweave}

\usepackage[breaklinks]{hyperref}

\usepackage{amsmath,xcolor}
\addtolength{\textwidth}{0.5in}
\addtolength{\oddsidemargin}{-0.25in}
\setlength{\evensidemargin}{\oddsidemargin}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\title{Pooled location and scale estimators}
%-------------------------------------------------------------------
\author{Chanseok Park\footnote{Applied Statistics Laboratory,
Department of Industrial Engineering, Pusan National University, Busan 46241, Korea.
His work was supported by the National Research Foundation of Korea(NRF) grant funded
by the Korea government(MSIT) (No.\ 2022R1A2C1091319).}
~{and}
Min Wang\footnote{Department of Management Science and Statistics,
The University of Texas at San Antonio, San Antonio, TX 78249, USA.}
}
\date{December 2022}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{document}
\maketitle
\begin{abstract}
In this note, we provide a brief summary of pooled location and scale estimators, 
which are provided in the \texttt{pooledEstimator} function of the robust quality control chart (rQCC) R package.
The \texttt{rcc} function is based on the pooled location and scale estimators provided here.
\end{abstract}

%====================================================================
\section{Location Estimation}
The pooled location estimator has three types, denoted by A, B, C. 
Type A estimator is obtained by the unweighted average of the unbiased estimators.
Type B estimator is obtained by the weighted average with the weight proportional to the sample size.
Type C estimator is obtained by the optimally weighted average in a sense of the BLUE.

Suppose that there are $m$ samples and $\hat{\mu}_i$ is an unbiased estimator of $\mu$ 
from the $i$th sample of size $n_i$. 
The \texttt{pooledEstimator} calculates the pooled estimator for the sample mean (default), 
the sample median, and the Hodges-Lehmann \cite{Hodges/Lehmann:1963} estimator 
based on one of the following three pooling methods.
\begin{itemize}
\item Type A: \\
\[
\bar{\hat{\mu}}_A =  \sum_{i=1}^m w_i \hat{\mu}_i,
\]
where $w_i = 1/m$.
\item Type B: \\
\[
\bar{\hat{\mu}}_B =  \sum_{i=1}^m w_i \hat{\mu}_i,
\]
where $w_i = n_i/\sum_{j=1}^{m} n_j$.
\item Type C: \\
\[
\bar{\hat{\mu}}_C =  \sum_{i=1}^m w_i \hat{\mu}_i,
\]
where $w_i$ is given by the BLUE in \cite{Park/Ouyang/Wang:2022}.
\end{itemize}
It should be noted that $\bar{\hat{\mu}}_A$, $\bar{\hat{\mu}}_B$, and $\bar{\hat{\mu}}_C$ are 
all unbiased for $\mu$.

%====================================================================
\section{Scale Estimation}
The \texttt{pooledEstimator} also calculates the pooled sclae estimator for the standard deviation (sd),
range, median absolute deviation, and Shamos estimators
based on one of the following three pooling methods.

Suppose that there are $m$ samples and $\hat{\sigma}_i$ is an estimator of $\sigma$ from the $i$th sample of size $n_i$.
Let $C_i$ is an unbiasing factor so that $E(\hat{\sigma}_i)/C_i = \sigma$.
Type A estimator is obtained by the unweighted average of the unbiased estimators.
Type B estimator is obtained by the weighted average with the weight proportional to the unbiasing factors.
Type C estimator is obtained by the optimally weighted average in a sense of the BLUE.
\begin{itemize}
\item Type A: \\
\[
\bar{\hat{\sigma}}_A =  \sum_{i=1}^m w_i \cdot \frac{\hat{\sigma}_i}{C_i},
\]
where $w_i = 1/m$.
\item Type B: \\
\[
\bar{\hat{\sigma}}_B =  \sum_{i=1}^m w_i \cdot \frac{\hat{\sigma}_i}{C_i}
\]
where $w_i = C_i/\sum_{j=1}^{m} C_j$.
\item Type C: \\
\[
\bar{\hat{\sigma}}_C =  \sum_{i=1}^m w_i \frac{\hat{\sigma}_i}{C_i},
\]
where $w_i$ is given by the BLUE in \cite{Park/Ouyang/Wang:2022}.
\end{itemize}
It should be noted that $\bar{\hat{\sigma}}_A$, $\bar{\hat{\sigma}}_B$, and $\bar{\hat{\sigma}}_C$ 
are all unbiased for $\sigma$.



\bigskip
%====================================================================
\section{R Usages}
%====================================================================
%%-----------------------------------------------
\noindent\fbox{\parbox{\textwidth}{%
\begin{color}{red}
\texttt{> \# Type A with mean (default)} \\
\texttt{> x1 = c(1,2)} \\
\texttt{> x2 = c(2,5,9)} \\
\texttt{> data = list(x1, x2)} \\
\texttt{> pooledEstimator(data)}
\end{color}
}}
\smallskip

\bigskip\noindent\fbox{\parbox{\textwidth}{%
\begin{color}{red}
\texttt{> \# Type C with HL1 (Hodges-Lehmann)} \\
\texttt{> pooledEstimator(data, estimator="HL1", poolType="C")}
\end{color}
}}
\smallskip


\bigskip\noindent\fbox{\parbox{\textwidth}{%
\begin{color}{red}
\texttt{> \# Type A with standard deviation (sd) } \\
\texttt{> pooledEstimator(data, estimator="sd")}
\end{color}
}}
\smallskip

\bigskip\noindent\fbox{\parbox{\textwidth}{%
\begin{color}{red}
\texttt{> \# Type C with Shamos} \\
\texttt{> pooledEstimator(data, estimator="shamos", poolType="C")}
\end{color}
}}
\smallskip




%%===============================================
\bigskip
\bibliographystyle{unsrt}
\bibliography{rQCC}


%%===============================================
\end{document}
%%===============================================