583 
583 
regression procedures. Further discussion about the underlying theory can be

584 
584 
found in the \scAD\ user's manual, but it is not necessary to understand the

585 
585 
theory to make use of the procedure.

586 

\begin{figure}[h]


586 
\begin{figure}[htbp]

587 
587 
\centering\hskip1pt\beginpicture

588 
588 
\setcoordinatesystem units <.18in,.04in>

589 
589 
\setplotarea x from 0 to 16.5, y from 0 to 50

...  ...  
675 
675 
analysis. An alternative approach, which avoids these difficulties, is to employ

676 
676 
\ADM's robust regression procedure, which removes the undue influence of

677 
677 
outlying points without the need to expressly remove them from the data.

678 

\begin{figure}[h]


678 
\begin{figure}[htbp]

679 
679 
\centering\hskip1pt\beginpicture

680 
680 
\setcoordinatesystem units <.18in,.04in>

681 
681 
\setplotarea x from 0 to 16.5, y from 0 to 50

...  ...  
783 
783 
template from a nonlinear regression model to a robust nonlinear regression

784 
784 
model.

785 
785 

786 

\begin{figure}[h]


786 
\begin{figure}[htbp]

787 
787 
\centering\hskip1pt\beginpicture

788 
788 
\setcoordinatesystem units <.18in,.04in>

789 
789 
\setplotarea x from 0 to 16.5, y from 0 to 50

...  ...  
826 
826 
better than leastsquare estimates in the presence of moderate or large

827 
827 
outliers. You can lose only a little and you stand to gain a lot by using these

828 
828 
estimators.

829 

\begin{figure}[h]


829 
\begin{figure}[htbp]

830 
830 
\centering\hskip1pt\beginpicture

831 
831 
\setcoordinatesystem units <.18in,.04in>

832 
832 
\setplotarea x from 0 to 16.5, y from 0 to 50

...  ...  
889 
889 
concentrations of the reactants are known only approximately.

890 
890 
See~Table~\ref{tab:runs} for what they are.

891 
891 
%xx \htmlbegintex

892 

\begin{table}[h]


892 
\begin{table}[htbp]

893 
893 
\begin{center}

894 
894 
\begin{tabular}{@{\extracolsep{1em}} l c l}

895 
895 
\bf Run 1&$s_1(0)=\theta_5=1\pm0.05$&$s_2(0)=\theta_6=1\pm0.05$\\

...  ...  
2135 
2135 
produced this way. Also, a plot of $y_i$ versus $x_i$ gives the user an

2136 
2136 
indication of what the probability distribution of the parameter looks like.

2137 
2137 
(See Figure~\ref{fig:05}.)

2138 

\begin{figure}[h]


2138 
\begin{figure}[htbp]

2139 
2139 
\centering\hskip1pt\beginpicture

2140 
2140 
\setplotsymbol ({\eightrm .})

2141 
2141 
%\setcoordinatesystem units <.7in,5in> Have to adjust so labels don't overlap.

...  ...  
2578 
2578 
the estimates produced by the \textsc{mcmc} method, for different sample sizes

2579 
2579 
(25,000 and 2,500,000 samples) for the \texttt{catage} example.

2580 
2580 
%xx \htmlbeginignore

2581 

\begin{figure}[h]


2581 
\begin{figure}[htbp]

2582 
2582 
%xx \htmlendignore

2583 
2583 
\centering\hskip1pt\beginpicture

2584 
2584 
\setplotsymbol ({\eightrm .})

...  ...  
2911 
2911 
standard deviations fixed. The number \texttt{N} should be between~1 and~9. The

2912 
2912 
smaller the number, the more the correlation is reduced. For this example (see

2913 
2913 
Figure~\ref{fig:07}), a value of~3 seemed to perform well.

2914 

\begin{figure}[h]


2914 
\begin{figure}[htbp]

2915 
2915 
\includegraphics[height=3.5in, width=\textwidth]{mcrb350K.png}

2916 
2916 
\emptycaption

2917 
2917 
\label{fig:07}

...  ...  
3787 
3787 
The plot of \texttt{qa} and \texttt{qb} demonstrates the extra information about

3788 
3788 
the probability distribution of the current state contained in in the current

3789 
3789 
value of~\texttt{y(t)}. (See Figure~\ref{fig:08}.)

3790 

\begin{figure}[h]


3790 
\begin{figure}[htbp]

3791 
3791 
\centering\hskip1pt\beginpicture

3792 
3792 
\setplotsymbol ({\eightrm .})

3793 
3793 
\setcoordinatesystem units <3.2in,2.5in>
