[Lme4-commits] r1796 - www/JSS

[noreply at r-forge.r-project.org]

Author: bbolker
Date: 2013-02-14 01:19:41 +0100 (Thu, 14 Feb 2013)
New Revision: 1796

Added:
   www/JSS/figs/
Modified:
   www/JSS/glmer.Rnw
Log:
added print() around xyplot for Sweave
various tweaks
added figures directory



Modified: www/JSS/glmer.Rnw
===================================================================
--- www/JSS/glmer.Rnw	2013-02-13 23:17:49 UTC (rev 1795)
+++ www/JSS/glmer.Rnw	2013-02-14 00:19:41 UTC (rev 1796)
@@ -62,15 +62,16 @@
 \begin{document}
 \section{Introduction}
 \label{sec:intro}
+\bmb{equations are all defined in part~I \ldots}
 
-
 \section{Generalized Linear Mixed Models}
 \label{sec:GLMMdef}
 
 The generalized linear mixed models (GLMMs) that can be fit by the
 \pkg{lme4} package preserve the multivariate Gaussian unconditional
 distribution of the random effects, $\mc B$
-(eqn.~\ref{eq:LMMuncondB}).  Because most families used for the conditional
+(eqn.~\ref{eq:LMMuncondB}).  
+Because most families used for the conditional
 distribution, $\mc Y|\mc B=\bm b$, do not incorporate a separate scale
 factor, $\sigma$, we remove it from the definition of $\bm\Sigma$ and
 from the distribution of the spherical random effects, $\mc U$.  That
@@ -144,7 +145,8 @@
 to $\bm u$ so the unscaled conditional density is indeed well-defined
 as a density, up to a scale factor.
 
-To evaluate the integrand in (\ref{eq:GLMMlike}) we use the value of
+To evaluate the integrand in (\ref{eq:GLMMlike}) 
+we use the value of
 the \code{dev.resids} function in the GLM family.  This vector,
 $\bm d(\yobs,\bm u)$, with elements, $d_i(\yobs,\bm u), i=1,\dots,n$,
 provides the deviance of a generalized linear model as
@@ -370,15 +372,15 @@
 }
 zm <- zeta(m1, -3.750440, 3.750440)
 dmat <- exp(-0.5*zm$sqrtmat^2)/sqrt(2*pi)
-xyplot(as.vector(dmat) ~ rep.int(zm$zvals, ncol(dmat))|gl(ncol(dmat), nrow(dmat)),
+print(xyplot(as.vector(dmat) ~ rep.int(zm$zvals, ncol(dmat))|gl(ncol(dmat), nrow(dmat)),
        type=c("g","l"), aspect=0.6, layout=c(5,3),
        xlab="z", ylab="density",
        panel=function(...){
            panel.lines(zm$zvals, dnorm(zm$zvals), lty=2)
            panel.xyplot(...)}
-       )
+       ))
 @ 
-  \caption{Comparison of univariate integrands (solid line) and standard normal density function (dashed line)}
+  \caption{Comparison of univariate integrands (solid line) and standard normal density function (dashed line) \bmb{is something wrong?  Do these agree TOO well?}}
   \label{fig:densities}
 \end{figure}
 
@@ -430,9 +432,9 @@
 \begin{figure}[tbp]
   \centering
 <<tfunc,fig=TRUE,echo=FALSE>>=
-xyplot(as.vector(dmat/dnorm(zm$zvals)) ~ rep.int(zm$zvals, ncol(dmat))|gl(ncol(dmat), nrow(dmat)),
+print(xyplot(as.vector(dmat/dnorm(zm$zvals)) ~ rep.int(zm$zvals, ncol(dmat))|gl(ncol(dmat), nrow(dmat)),
        type=c("g","l"), aspect=0.6, layout=c(5,3),
-       xlab="z", ylab="t(z)")
+       xlab="z", ylab="t(z)"))
 @ 
   \caption{The function $t(z)$, which is the ratio of the normalized
     unscaled conditional density to the standard normal density, for