version 0.5-1

ChangeLog

@@ -1,3 +1,28 @@
+2009-08-08  Douglas Bates <bates@stat.wisc.edu>
+
+	* [r401] DESCRIPTION, inst/doc/Usinglmer.Rnw, inst/doc/Usinglmer.pdf:
+	  New release on R-forge
+	* [r400] inst/doc/Usinglme.Rnw, inst/doc/Usinglme.bib,
+	  inst/doc/Usinglmer.Rnw, inst/doc/Usinglmer.bib: Update to AFS version
+
+2009-08-08  Martin Maechler  <maechler@stat.math.ethz.ch>mmaechler
+
+	* DESCRIPTION, man/Animal.Rd, man/AvgDailyGain.Rd, man/BIB.Rd,
+	  man/Bond.Rd, man/Cultivation.Rd, man/Demand.Rd, man/Genetics.Rd,
+	  man/HR.Rd, man/IncBlk.Rd, man/Mississippi.Rd, man/Multilocation.Rd,
+	  man/PBIB.Rd, man/SIMS.Rd, man/Semi2.Rd, man/Semiconductor.Rd,
+	  man/TeachingI.Rd, man/TeachingII.Rd, man/WWheat.Rd, man/WaferTypes.Rd,
+	  man/Weights.Rd: update to CRAN`s version 0.4-4 (from 2008-06 !)
+
+2009-07-28  Martin Maechler  <maechler@stat.math.ethz.ch>
+
+	* .: move mlmRev, MEMSS (and SASmixed) from
+	  "svn.r-project.org/R-packages/trunk/"
+
+2006-10-23  Douglas Bates <bates@stat.wisc.edu>
+
+	* [r384] Remove poorly defined example
+
 2006-09-09  Douglas Bates  <bates@R-project.org>
 
 	* man/BIB.Rd: Update this and all other examples for lmer back in

DESCRIPTION

@@ -1,14 +1,16 @@
 Package: SASmixed
-Version: 0.4-4
-Date: 2008-06-25
+Version: 0.5-1
+Date: 2009-10-21
 Title: Data sets from "SAS System for Mixed Models"
 Maintainer: Douglas Bates <bates@stat.wisc.edu>
 Author: Original by Littel, Milliken, Stroup, and Wolfinger
-  modifications by Douglas Bates <bates@stat.wisc.edu>
-Description: Data sets and sample lmer analyses corresponding
-  to the examples in Littel, Milliken, Stroup and Wolfinger
-  (1996), "SAS System for Mixed Models", SAS Institute.
-Suggests: lattice, lme4 (>= 0.999375-19)
+        modifications by Douglas Bates <bates@stat.wisc.edu>
+Description: Data sets and sample lmer analyses corresponding to the
+        examples in Littel, Milliken, Stroup and Wolfinger (1996), "SAS
+        System for Mixed Models", SAS Institute.
+Suggests: lattice, lme4
 LazyData: yes
-License: GPL version 2 or later
-Packaged: Wed Jun 25 12:42:15 2008; bates
+License: GPL (>= 2)
+Packaged: 2009-10-26 19:14:08 UTC; bates
+Repository: CRAN
+Date/Publication: 2009-10-26 20:51:35

INDEX

@@ -0,0 +1,25 @@
+Animal                  Animal breeding experiment
+AvgDailyGain            Average daily weight gain of steers on
+                        different diets
+BIB                     Data from a balanced incomplete block design
+Bond                    Strengths of metal bonds
+Cultivation             Bacterial innoculation applied to grass
+                        cultivars
+Demand                  Per-capita demand deposits by state and year
+Genetics                Heritability data
+HR                      Heart rates of patients on different drug
+                        treatments
+IncBlk                  An unbalanced incomplete block experiment
+Mississippi             Nitrogen concentrations in the Mississippi
+                        River
+Multilocation           A multilocation trial
+PBIB                    A partially balanced incomplete block
+                        experiment
+Semi2                   Oxide layer thicknesses on semiconductors
+Semiconductor           Semiconductor split-plot experiment
+SIMS                    Second International Mathematics Study data
+TeachingI               Teaching Methods I
+TeachingII              Teaching Methods II
+WaferTypes              Data on different types of silicon wafers
+Weights                 Data from a weight-lifting program
+WWheat                  Winter wheat

inst/doc/Usinglmer.Rnw

@@ -45,7 +45,7 @@ A file on the SAS Institute web site (\textsf{http://www.sas.com})
 contains all the data sets in the book and all the SAS programs used
 in \citet{litt:mill:stro:wolf:1996}.  We have converted the data
 sets from the tabular representation used for SAS to the
-\code{groupedData} objects used by \code{lmer}.  To help users familiar
+\code{data.frame} objects used by \code{lmer}.  To help users familiar
 with \code{SAS PROC MIXED} get up to speed with \code{lmer} more quickly,
 we provide transcripts of some \code{lmer} analyses paralleling the
 \code{SAS PROC MIXED} analyses in \citet{litt:mill:stro:wolf:1996}.
@@ -79,29 +79,32 @@ The basic specification of the model requires a linear model
 expression for the fixed effects and a linear model expression for the 
 random effects.  In \code{SAS PROC MIXED} the fixed-effects part is
 specified in the \code{model} statement and the random-effects
-part in the \code{random} statement.  In \code{lmer} the
-arguments are called \code{fixed} and \code{random}.
+part in the \code{random} statement.  In \code{lmer} the fixed effects
+and the random effects are both specified as terms in the
+\code{formula} argument to \code{lmer}.
 
 Both \code{SAS PROC MIXED} and \code{lmer} allow a mixed-effects model
 to be fit by maximum likelihood (\code{method = ml} in SAS) or by
 maximum residual likelihood, sometimes also called restricted maximum
 likelihood or \textsf{REML}.  This is the default criterion in
 \code{SAS PROC MIXED} and in \code{lmer}.  To get \textsf{ML}
-estimates in \code{lmer}, set the optional argument
-\code{method="REML"}.
+estimates use the optional argument \code{REML=FALSE} in the call to
+\code{lmer}.
 
 \section{Important differences}
 \label{sec:differences}
 
 The output from \code{PROC MIXED} typically includes values of the
 Akaike Information Criterion (\textsf{AIC}) and Schwartz's Bayesian
 Criterion (\textsf{SBC}).  These are used to compare different models
-fit to the same data.  The output of the \code{summary} function applied
-to the object created by \code{lmer} also produces values of \textsf{AIC}
-and \textsf{BIC} but the definitions used in \code{PROC MIXED} and in
-\code{lmer} are different.  In \code{lmer} the definitions are such that
-``smaller is better''.  In \code{PROC MIXED} the definitions are such
-that ``bigger is better''.
+fit to the same data.  The output of the \code{summary} function
+applied to the object created by \code{lmer} also produces values of
+\textsf{AIC} and \textsf{BIC} but the definitions used in older
+versions of \code{PROC MIXED} are different from those used in more
+recent versions of \code{PROC MIXED} and in \code{lmer}.  In
+\code{lmer} the definitions are such that ``smaller is better''.  In
+some older versions of \code{PROC MIXED} the definitions are such that
+``bigger is better''.
 
 When models are fit by \textsf{REML}, the values of \textsf{AIC},
 \textsf{SBC} (or \textsf{BIC}) and the log-likelihood can only be
@@ -111,15 +114,15 @@ compared between any models fit to the same data.  That is, these
 quality-of-fit criteria can be used to evaluate different
 fixed-effects specifications or different random-effects
 specifications or different specifications of both fixed effects and
-random effects.  The greater flexibility of model comparisons when
-using maximum likelihood is the reason that this is the default
-criterion in \code{lmer}.
+random effects.  
 
 We encourage developing and testing the model using likelihood ratio
 tests or the \textsf{AIC} and \textsf{BIC} criteria.  Once a form
 for both the random effects and the fixed effects has been determined,
 the model can be refit with \code{REML = TRUE} if the restricted
-estimates of the variance components are desired.
+estimates of the variance components are desired.  Note that the
+\code{update} function provides a convenient way of refitting a model
+with changes to one or more arguments.
 
 \section{Data manipulation}
 \label{sec:data}
@@ -152,13 +155,15 @@ data animal;
 would require that the \code{trait} and \code{animal} variables be
 specified in a class statement in any model that is fit.
 
-In \s{} these data could be read from a file, say \texttt{animal.dat},
+In \R{} these data could be read from a file, say \texttt{animal.dat},
 and converted to factors by
 \begin{Schunk}
 \begin{Sinput}
-animal <- read.table("animal.dat", header = TRUE)
-animal$trait <- as.factor(animal$trait)
-animal$animal <- as.factor(animal$animal)
+animal <- within(read.table("animal.dat", header = TRUE),
+             {
+                 trait <- factor(trait)
+                 animal <- factor(animal)
+             })
 \end{Sinput}
 \end{Schunk}
 In general it is a good idea to check the types of variables in a data 
@@ -170,14 +175,6 @@ sapply(Animal, data.class)
 str(Animal)
 @
 
-To make specification of models in \code{lmer} easier and to make graphic
-presentations more informative, we recommend converting from a
-\code{data.frame} object to a \code{groupedData} object.  This class of
-objects contains a formula specifying the response, the primary
-covariate (if there is one) and the grouping factor or factors.  The
-data sets from \citet{litt:mill:stro:wolf:1996} have been
-converted to \code{groupedData} objects in this directory.
-
 \subsection{Unique levels of factors}
 \label{sec:nested}
 
@@ -197,14 +194,14 @@ In \code{SAS} this nesting of factors is denoted by \code{wafer(et)}.  In
 within ET''.  If both levels of nested factors are to be associated
 with random effects then this is all you need to know.  You would use
 an expression with a \code{"/"} in the grouping factor part of the
-formula for the \code{groupedData} object.  Then the random effects
-could be specified as
+formula in the call to \code{lmer} object.  The random effects term
+would be either
 \begin{Example}
-  random = list( ET = ~ 1, Wafer = ~ 1 )
+  (1 | ET/Wafer)
 \end{Example}
 or, equivalently
 \begin{Example}
-  random = ~ 1 | ET/Wafer
+  (1 | ET:Wafer) + (1 | ET)
 \end{Example}
 
 In this case, however, there would not usually be any random effects
@@ -213,33 +210,31 @@ only random effects are at the \code{Wafer} level.  It is necessary to
 create a factor that will have unique levels for each \code{Wafer}
 within each level of \code{ET}.  One way to do this is to assign
 <<semiconductorGrp>>=
-Semiconductor$Grp <- with(Semiconductor, ET:Wafer)
+Semiconductor <- within(Semiconductor, Grp <- factor(ET:Wafer))
 @ 
-%$
-after which we could specify a random effects term of \code{(1 | Grp)}.
+after which we could specify a random effects term of \code{(1 |
+  Grp)}.  Alternatively, we can use the explicit term
+\begin{Example}
+  (1 | ET:Wafer)  
+\end{Example}
 
 \subsection{General approach}
 \label{sec:generalApproach}
 
-As a general approach to importing data into \s{} for mixed-effects
+As a general approach to importing data into \R{} for mixed-effects
 analysis you should:
 \begin{itemize}
 \item Create a \code{data.frame} with one row per observation and one
   column per variable.
-\item Use \code{ordered} or \code{as.ordered} to explicitly convert any
+\item Use \code{factor} or \code{as.factor} to explicitly convert any
   ordered factors to class \code{ordered}.
 \item Use \code{ordered} or \code{as.ordered} to explicitly convert any
   ordered factors to class \code{ordered}.
-\item If necessary, use \code{getGroups} to create a factor with unique
+\item If necessary, use interaction terms to create a factor with unique
   levels from inner nested factors.
-\item Specify the formula for the response, the primary covariate and
-  the grouping structure to create a \code{groupedData} object from the
-  data frame.  Labels and units for the response and the primary
-  covariate can also be specified at this time as can \code{outer} and
-  \code{inner} factor expressions.
-\item Plot the data.  Plot it several ways.  The use of trellis
-  graphics is closely integrated with the \code{nlme} library.  The
-  trellis plots can provide invaluable insight into the structure of
+\item Plot the data.  Plot it several ways.  The use of lattice
+  graphics is closely integrated with the \code{lme4} library.
+  Lattice plots can provide invaluable insight into the structure of
   the data.  Use them.
 \end{itemize}
 
@@ -399,7 +394,7 @@ anova(fm1Mult)
 (fm5Mult <- lmer(Adj ~ 1 + (1|Grp), Multilocation))
 anova(fm2Mult)
 (fm2MultR <- lmer(Adj ~ Trt + (Trt - 1|Location) + (1|Block), Multilocation,
-                  control = list(msV = 1, niterEM = 200)))
+                  verbose = TRUE))
 @