\documentclass{article}

\usepackage[utf8]{inputenc}
\usepackage[french]{babel}
\usepackage[T1]{fontenc}




\usepackage{graphicx}


\usepackage{amsmath}
\usepackage{amsthm}
\usepackage{amsfonts}
\usepackage{xcolor}

\setlength{\headheight}{-50pt}
\setlength{\textheight}{708pt}

\usepackage{pdfpages}
%\VignetteIndexEntry{Estimation functions}

\usepackage{Sweave}
\begin{document}
\SweaveOpts{concordance=TRUE}
\input{estima-concordance}

\thispagestyle{empty}

The goal of the estima function is to estimate the coefficients of the two centered autologistic regression:


\begin{align*} logit(p_{i,t}) &= X^{T}_{i,t}\beta + \beta_{past} \sum_{j \in N_{i}^{past}} \!\!\!Z_{j,t-1} +  \rho_{1}\sum_{j \in N_{i}} Z^{**}_{j,t} + \rho_{2}Z_{i,t-1} \\
\Leftrightarrow \qquad p_{i,t} &=  \frac{exp(X^{T}_{i,t}\beta + \beta_{past} \sum_{j \in N_{i}^{past}} \!\!\!Z_{j,t-1}+ \rho_{1}\sum_{j \in N_{i}} Z^{**}_{j,t} + \rho_{2}Z_{i,t-1})}{1+exp(X^{T}_{i,t}\beta + \beta_{past} \sum_{j \in N_{i}^{past}} \!\!\!Z_{j,t-1}+ \rho_{1}\sum_{j \in N_{i}} Z^{**}_{j,t}+ \rho_{2}Z_{i,t-1})}
\end{align*}

where $Z_{i,t}$ is a binary variable of parameter $p_{i,t}$, $N_{i}$ is the neighborhood of the site $i$ for the instantaneous spatial dependence, $N_{i}^{past}$ is the neighborhood of the site $i$ for the spatio-temporal dependence (spread of the illness)  and $Z_{i,t-1}^{**}$ is given by:



$$ Z^{**}_{i,t} = Z_{i,t} - \frac{exp(X^{T}_{i,t}\beta  + \beta_{past} \sum_{j \in N_{i}^{past}} \!\!\!Z_{j,t-1}+ \rho_{2}Z_{i,t-1})}{1 + exp(X^{T}_{i,t}\beta  + \beta_{past} \sum_{j \in N_{i}^{past}} \!\!\!Z_{j,t-1}+ \rho_{2}Z_{i,t-1})}. $$

\bigskip
Estimation uses the pseudo-likelihood:

$$
\mathcal{L}(\beta,\beta_{past},\rho_1,\rho_2) = \prod_{t = 1}^{T} \prod_{1 \leq i \leq n} (p_{i,t})^{z_{i,t}}(1-p_{i,t})^{1-z_{i,t}}.
$$For more detail see Gegout-Petit, Guérin-Dubrana, Li, 2019.

\bigskip

The parameters of spatio-temporal dependence
$\rho_1$, $\rho_2$, $\beta_{past}$ can be interpreted as practical biological processes:

\begin{itemize}
\item Instantaneous spatial dependence $\rho_{1}$. It quantifies the spatial autocorrelation between neighbours for the occurence of the event at each time $t$,
\item Temporal dependence $\rho_{2}$. It quantifies the temporal dependence on the previous year's status,
\item Coefficient $\beta_{past}$: it quantifies the spread of the illness coming from the  previous year's status of the neighbours
\end{itemize}


The function "estima"  estimates the parameters with different possibilities for $\beta_{past}$ and $\sum_{j \in N_{i}^{past}} \!\!\!Z_{j,t-1}$:

\begin{itemize}
\item[if "covpast = FALSE :] estimates the parameter
   $\beta = \begin{pmatrix}
       \beta_{0} \\
       \beta_{1} \\
       \beta_{2} \\
       \beta_{3} \\
    \end{pmatrix}$
  and
   $X^{T}_{i,t} = \begin{pmatrix}
       1 \\
       x_{i,t}^{1} \\
       x_{i,t}^{2} \\
       x_{i,t}^{3} \\
    \end{pmatrix}$
    where $x_{i,t}^{j} \forall j \in (1,2,3)$ is a spatio-temporal covariate. There can be 0, 1, 2 or 3 covariates. In this case, there is no regression on   $\sum_{j \in N_{i}^{past}} \!\!\!Z_{j,t-1}$  ($\beta_{past}=0$).


\item[if "covpast = TRUE" :] the function estimates the parameters
  $ \beta = \begin{pmatrix}
       \beta_{0} \\
       \beta_{1} \\
       \beta_{2} \\
       \beta_{3} \\
    \end{pmatrix}$
    and
    $\beta_{past}$.

\end{itemize}




\end{document}