1
\documentclass{article}
6
% During writing: a draft:
9
\draftfalse \ifMM\drafttrue\fi
11
\ifdraft %% generate tableofcontents down to the \paragraph
12
\setcounter{tocdepth}{5}
15
% a) a statistician's needs
16
% b) statistical analysis packages supported by ESS
23
% 3) ange-ftp/EFS/tramp
33
%4. ESS as an open-source project
43
% b) what's next for ESS
44
% ESS internet resources
51
\addtolength{\topmargin}{-1cm}
52
\addtolength{\textheight}{+1cm}
54
\renewcommand{\baselinestretch}{1.5}
56
\addtolength{\oddsidemargin}{-0.5in}
57
\addtolength{\textheight}{0.2in}
58
\addtolength{\textwidth}{1in}
59
\ifMM\addtolength{\textheight}{2cm}\fi
62
\usepackage[authoryear,round]{natbib}
63
%or (if you have an unshiny latex installation)
64
%\newcommand{\citep}[1]{{\{\sf#1\}}}
71
%\DeclareGraphicsExtensions{.jpg,.pdf,.png,.mps,.ps}
72
\DeclareGraphicsRule{.ps}{eps}{.ps.bb}{}
73
\DeclareGraphicsRule{.ps.Z}{eps}{.ps.bb}{'/bin/zcat -c #1}
74
%\graphicspath{{jcgs/}}
77
\ifx\pdfoutput\undefined
78
%% Stuff wout hyperref
79
\def\url#1{\stexttt{#1}} % To help fit in lines ?AJR: stextsf?
81
%% Stuff with hyperref
83
%%\hypersetup{backref,colorlinks=true,pagebackref=true,hyperindex=true}
84
\hypersetup{backref,colorlinks=false,pagebackref=true,hyperindex=true}
86
%%---End of package requiring ---------- Own Definitions -------------
88
\newcommand*{\regstrd}{$^{\mbox{\scriptsize{\textregistered}}}$}
89
\newcommand*{\tm}{$^{\mbox{\scriptsize\sc tm}}$}
90
\newcommand*{\SAS}{\textsc{SAS}}
91
\newcommand*{\Splus}{\textsc{S-Plus}}
92
\newcommand*{\XLispStat}{\textsc{XLispStat}}
93
\newcommand*{\Stata}{\textsc{Stata}}
94
\newcommand*{\Rgui}{\textsc{Rgui}}
95
\newcommand*{\Perl}{\textsc{Perl}}
96
\newcommand*{\Fortran}{\textsc{Fortran}}
97
\newcommand*{\Scmt}[1]{\hbox{\qquad {\footnotesize \#\#} \textsl{#1}}}
98
\newtheorem{defn}{Definition}[section]
99
\newtheorem{ex}{Example}[section]
101
\newcommand{\stexttt}[1]{{\small\texttt{#1}}}
102
\newcommand{\ssf}[1]{{\small\sf{#1}}}
103
\newcommand{\elcode}[1]{\\{\stexttt{\hspace*{2em} #1}}\\}
104
\newcommand{\file}[1]{`\stexttt{#1}'}
105
\newcommand{\US}{{\char'137}} % \tt _
106
\newcommand{\marpar}[1]{\marginpar{\raggedright#1}}
107
\newenvironment{Salltt}{\small\begin{alltt}}{\end{alltt}}
109
\newcommand{\emptyfig}{
110
\hspace*{42pt}\rule{324pt}{.25pt}\\
111
\hspace*{42pt}\rule{.25pt}{10pc}
115
\newcommand{\ESSfig}[1]{\centering{#1}}
117
%% Use \begin{Comment} .. \end{Comment} for internal comments
119
\newenvironment{Comment}{\begin{quote}\small\itshape }{\end{quote}}
121
\else %% this requires
122
\usepackage{verbatim}
123
\let\Comment=\comment
124
\let\endComment=\endcomment
128
%%--------------------------------------------------------------- Start Text
130
\title{Emacs Speaks Statistics (ESS): A multi-platform, multi-package
131
intelligent environment for statistical analysis}
133
%%For blinded submission:
136
%%For regular review:
137
\author{A.J. Rossini \and Richard M. Heiberger \and Rodney A. Sparapani
138
\and Martin M{\"a}chler \and Kurt Hornik \footnote{%
140
A.J. Rossini is Research Assistant Professor in the Department of
141
Biostatistics, University of Washington and Joint Assistant Member at
142
the Fred Hutchinson Cancer Research Center, Seattle, WA, USA
143
\url{mailto: rossini@u.washington.edu};
145
Richard M. Heiberger is Professor in the Department of Statistics at
146
Temple University, Philadelphia, PA, USA \url{mailto: rmh@temple.edu};
148
Rodney A. Sparapani is Senior Biostatistician in the Center for Patient
149
Care and Outcomes Research at the Medical College of Wisconsin,
150
Milwaukee, WI, USA \url{mailto: rsparapa@mcw.edu};
152
Martin M{\"a}chler is Senior Scientist and Lecturer in the Seminar for
153
Statistics, ETH Zurich, Zurich, Switzerland
154
\url{mailto: maechler@stat.math.ethz.ch};
156
Kurt Hornik is Professor in the Institut f{\"u}r Statistik,
157
Wirtschaftsuniversit{\"a}t Wien and the Institut f{\"u}r
158
Wahrscheinlichkeitstheorie und Statistik, Technische Universit{\"a}t
159
Wien, Vienna, Austria \url{mailto: Kurt.Hornik@r-project.org}}}
162
\date{$ $Date: 2003/10/22 17:34:04 $ $\tiny Revision: 1.255$ $}
166
%% To cite everything
177
\ifdraft{}%% large line skip -- not for draft
179
\renewcommand{\baselinestretch}{1.5}
180
%%- \baselineskip=2pc
184
Computer programming is an important component of statistics
185
research and data analysis. This skill is necessary for using
186
sophisticated statistical packages as well as for writing custom
187
software for data analysis. Emacs Speaks Statistics (ESS) provides
188
an intelligent and consistent interface between the user and
189
software. ESS interfaces with SAS, S-PLUS, R, and other statistics
190
packages under the Unix, Microsoft Windows, and Apple Mac operating
191
systems. ESS extends the Emacs text editor and uses its many
192
features to streamline the creation and use of statistical software.
193
ESS understands the syntax for each data analysis language it works
194
with and provides consistent display and editing features across
195
packages. ESS assists in the interactive or batch execution by the
196
statistics packages of statements written in their languages. Some
197
statistics packages can be run as a subprocess of Emacs, allowing
198
the user to work directly from the editor and thereby retain a
199
consistent and constant look-and-feel. We discuss how ESS works and
200
how it increases statistical programming efficiency.
203
\noindent Keywords: Data Analysis, Programming,
204
S, \SAS, \Splus, R, \XLispStat, \Stata, BUGS, Open Source Software,
205
Cross-platform User Interface.
207
\section{Introduction}
208
\label{sec:introduction}
210
Most statistical research activities, particularly data analysis and
211
communication, involve some form of computing. The computer user
212
interface is thus placed in the central role of facilitating
213
statistical tasks. While presentation of character and graphical
214
information is the most visual component of a user interface,
215
perhaps a more critical component is how the computer interprets user
216
input. A familiar, well-understood set of input behaviors can provide
217
large gains in efficiency. This paper introduces Emacs Speaks
218
Statistics (ESS) \citep{ESS}, a software package which provides a
219
common interface to a variety of statistical packages on the most
220
common computing platforms.
222
ESS is an interface to statistical packages that provides tools which
223
facilitate both statistical software development and data analysis.
224
ESS provides assistance with both writing and evaluation of analysis
225
code for both interactive and batch statistical packages. ESS
226
currently supports the S family of languages (including S
227
\citep{BecRCW88,ChaJH92,ChaJ98}, \Splus\regstrd\ \citep{Splus}, and R
228
\citep{ihak:gent:1996,R}; \SAS\regstrd\ \citep{SAS:8}; \Stata\
229
\citep{Stata:7.0}; \XLispStat\ \citep{Tier90} and its extensions Arc
230
\citep{Cook:Weisberg:1999} and ViSta \citep{youn:fald:mcfa:1992}; BUGS
231
\citep{BUGS}; and Omegahat \citep{DTLang:2000}. ESS can be extended
232
to accommodate most statistical packages which provide either an
233
interactive command-line or process batch files for instructions.
235
We start by describing the Emacs text editor, the underlying platform
236
on which ESS is built. Next, we discuss how ESS enhances a
237
statistician's daily activities by presenting its features and showing
238
how it facilitates statistical computing. We conclude with a short history
239
of the development of ESS.
240
% and conclude with future extensions and related work.
245
Emacs is a mature, powerful, and extensible text editing system which
246
is freely available, under the GNU General Public License (GPL), for a
247
large number of platforms, including most Unix\regstrd
248
distributions, Microsoft Windows\regstrd\ and Apple Mac\tm\ OS. There
249
are two open-source implementations of Emacs: GNU Emacs
250
\citep{GNU-Emacs} and XEmacs \citep{XEmacs}. Emacs shares many
251
features with word processors, and some characteristics with operating
252
systems, including many facilities which go beyond ordinary text
253
editing. More important to our goals, Emacs can control and interact
256
\paragraph{Keyboard and Mouse Input.}
257
When Emacs was originally written, character-based terminals were the
258
most advanced method of computer access. Common Emacs commands were
259
mapped to key sequences, creating keyboard shortcuts for convenience.
260
Over the last decade, Emacs has been extended to use graphical
261
windowing systems, such as X11\tm, Microsoft Windows, and Apple Mac
262
OS, which allow additional forms of input, for example using a mouse,
263
and which encourage multiple applications to share a single display.
264
Presently, Emacs is more often used with a GUI, with commands bound to
265
mouse actions, but having commands also associated with key sequences
266
is an important ergonomic and time-saving feature. Emacs menus and
267
toolbars on the display screen allow mouse access to frequently used
268
actions and provide a graphical alternative when the user does not
269
know or can not recall a key sequence; these are also subject to
272
\paragraph{Buffers give Emacs control.}
273
Emacs buffers are the interface between the user and computer. They
274
can be considered to be a collection of scratch pads that both the
275
user and computer can read, write, and respond to. The user can
276
simultaneously edit many files and control numerous programs by
277
opening multiple buffers. With disk files, the working copy of the
278
opened file is placed in an Emacs buffer where it can be viewed and
279
edited either by the user or automatically by Emacs or another program
280
under the control of Emacs. Emacs can save a backup of the contents
281
to disk at specified intervals. Emacs presents buffer contents in
282
ways which optimize reading and navigation activities. One example of
283
program control is the embedding of the interactive operating system
284
command line interpreter, called a shell, within Emacs. Variations on
285
this theme are used to control programs such as statistical packages
286
which take input from and provide output to the command-line. The
287
resulting buffers provide a copy of the entire transcript of the
288
interaction, which can be edited and searched while the program
291
\paragraph{Major and Minor Modes.}
292
Emacs capabilities are extended by loading
293
%% AJR: If we remove ``or byte-compiled'' we need to remove ``text'',
294
%% since it is wrong.
297
files containing commands and functions written in Emacs Lisp (elisp)
298
\citep{RChassell1999}, which is a dialect of Lisp
299
\citep{PGraham:1996}. Emacs commands can be called interactively by
300
pressing a key sequence mapped to the command or by name.
301
%% AJR: this is actually true for M-x long-command-not-bound-to-keys,
302
%% but I'm not telling anyone this!
303
% Rodney: I don't understand what the problem is with telling them.
304
The most important extensions to Emacs take the form of modes, which
305
provide specific enhancements to the editing behavior.
307
Major modes provide a customized environment consisting of mapped key
308
sequences and associated commands for performing tasks such as file
309
editing, reading mail, or browsing disk directories. Only one major
310
mode can be active for a given buffer at any time. Major modes also
311
can be written to intelligently control other programs such as
312
statistics packages. Major modes for file editing are often
313
determined by the file type or extension, i.e. the characters at the
314
end of the file name that follow a period like \stexttt{txt},
315
\stexttt{s}, or \stexttt{sas}. Examples of this kind of major mode are
316
\stexttt{ESS[S]} and \stexttt{ESS[SAS]}. Major modes understand a
317
file's syntax and grammar and therefore provide intelligent
318
actions such as automatic indentation; navigation in units of
319
characters, words, lines, sentences, paragraphs, function definitions,
320
and pages; syntax-based fontification and colorization; and
321
reformatting based on programmed conventions.
323
Minor modes provide complementary services that that are applicable
324
across major modes. Many minor modes can be active at once. For
325
example, \stexttt{font-lock-mode} allows Emacs to highlight, with
326
fonts or colors, the syntax of a programming language whose
327
characteristics are described within a major mode like
328
\stexttt{ESS[S]}. The \stexttt{overwrite-mode} determines whether
329
typed characters replace the existing text or are inserted at the
330
cursor. Minor modes can emulate the key sequences used by another
331
editor such as \stexttt{vi}. In addition, they can be used to perform
332
version control operations and many other operations which are nearly
333
identical across file types.
335
\paragraph{Network Support.}
336
Emacs allows transparent access to remote files over a network. This
337
means that the user views, edits, and saves files on a remote machine
338
exactly as if they were on the local machine. Mechanisms for both
339
open (\stexttt{ange-ftp} and \stexttt{EFS} use ftp) and secure
340
(\stexttt{tramp} uses scp or ssh) access are available. Emacs can
341
also monitor and control remote processes running in a shell buffer.
343
\paragraph{Editing Extensions.}
344
Most programming and documentation tasks consist of editing text.
345
These tasks can be enhanced by contextual highlighting and recognition
346
of special reserved words appropriate to the programming language in
347
use. In addition, Emacs also supports folding, outlining, tags, and
348
bookmarks, all of which assist with maneuvering around a file. Emacs
349
shares many features with word processing programs and cooperates with
350
markup-language document preparation systems such as \LaTeX,
351
\textsc{html}, or \textsc{xml}.
353
Tracking changes to a text file made by multiple users, potentially in
354
different locations, is the job of source-code control programs.
355
Emacs interacts with standard source-code control programs such as CVS, RCS,
356
and SCCS through minor modes such as \stexttt{vc-mode}. These
357
source-code control systems facilitate documenting and tracking edits and
358
changes to a file. More importantly, they allow for branching and
359
merging of versions so that material present in an older version of
360
the file can be recovered and inserted into a newer version in a
363
Comparison of files, two or three drafts of a paper for example, is
364
simplified by \stexttt{ediff}.
365
An example is shown in Figure \ref{f.ediff}.
366
The lines that are similar are highlighted in the two
367
buffers, one for each file, and the specific words that mismatch are
368
highlighted in a contrasting color. \stexttt{ediff} has many tools
369
for working with the differences in files and in entire directories.
370
When combined with the patch utility or a source-code control system,
371
it provides the user with the ability to insert, delete or modify only
372
the differing portions of text files.
375
\ESSfig{\includegraphics[angle=270,width=\textwidth]{ediff-sas}}
376
%\url{http://software.biostat.washington.edu/ess/doc/figures/ediff-sas.gif}
377
\caption{Ediff of two versions of a file.}
381
% Rodney: You can't just mention complex functionality like etags and
382
% speedbar. We would need to introduce these packages. Since they
383
% aren't critical to ESS at this time, let's ignore them.
384
Emacs has many other important features. Emacs provides file-manager
385
capabilities, such as \stexttt{dired} (discussed in Major and Minor
386
Modes above) and \stexttt{speedbar}, both of which interface to the
387
computer's directory structure. Emacs stores the complete history of
388
commands issued in an editing session, allowing a flexible and fairly
389
complete undo capability. More importantly, for modes which control
390
processes, the process input history is stored for recall as well as
391
for later editing for printing or re-use. Emacs also includes web
392
browsers, mail/newsgroup readers, and spell checking.
394
In addition to being an extremely powerful editor, Emacs also
395
includes capabilities usually found in an operating system. Thus, it
396
provides a strong foundation for constructing an integrated
397
development environment focused on the needs of statisticians. Emacs'
398
power, flexibility, portability, and extensibility make it a solid
399
platform on which to construct a statistical analysis user interface.
401
\section{ESS extends Emacs}
402
\label{sec:ess-extends-emacs}
404
Statistical programming is the writing of computer programs for data
405
analysis and processing. These programs might be written in a
406
computer language that requires a compiler, such as \Fortran\ or C.
407
But, more likely, they are written in a statistical analysis language
408
that only requires an interpreter such as R, \SAS, \Stata, or \XLispStat.
409
General purpose languages such as \Fortran, C/C++, Java, and PERL
410
have integrated development environments which facilitate writing and
413
ESS extends Emacs to provide an integrated development environment for
414
statistical languages. It offers a single
415
interface for a variety of statistical computing tasks including
416
interactive data analysis and statistical programming.
417
ESS is able to provide a functional and extensible interface
418
which is uniform and consistent across multiple statistical packages.
419
This is done by adding shortcuts and features for accelerated editing
420
of files as well as by interacting with the particular statistical
421
packages to provide, for example, control of input/output, assistance
422
with evaluation, and specialized parsing of help files.
424
ESS supports the S family (S, \Splus, and R) interactively. \SAS\ and
425
BUGS are also well supported for batch processing. \Stata\ and
426
\XLispStat\ (including ARC and ViSta) are supported through
427
highlighting and process-interfacing.
429
\subsection{Features and capabilities}
430
\label{sec:ESS:features}
432
\paragraph{Syntactic highlighting and indentation of source code.}
433
The programmers task is eased when language constructs (such as
434
reserved words, function calls, strings, and comments) are visually
435
identifiable and when lines of code are automatically indented to a
436
depth appropriate to their context (e.g., if--then clauses, loops).
437
ESS provides both of these to the programmer by including a
438
description of the syntax of each supported statistical language in
439
the form used by \stexttt{font-lock-mode}.
441
Figure \ref{f.font} shows an example of font-locking a complicated S
442
statement. The top panel shows an \stexttt{if} statement with a long
443
expression in the condition and a multi-line consequence. The keyword
444
\stexttt{if} is shown in purple, the string \stexttt{"deltat"} in
445
RosyBrown. The comments are in red. Everything else is in the
446
standard font. The consequence is indented and the continuations of
447
the consequence are further indented. The matching parentheses are
448
shown in green. The cursor is indicated by a solid box. In the
449
bottom panel, we replaced the matching parenthesis with an unbalanced
450
bracket. Emacs immediately marks that with the paren-mismatch font,
451
bright purple in this example. On a black and white terminal we would
452
use bold, underline, italic, and reverse-video, rather than colors, to
453
distinguish the fonts.
455
% Figure \ref{f.font} shows a black-and-white example of font-locking a
456
% complicated S statement. The top panel shows an \stexttt{if}
457
% statement with a long expression in the condition and a multi-line
458
% consequence. The keyword \stexttt{if} is shown in an underlined font,
459
% the string \stexttt{"deltat"} in an italic underlined font. The
460
% comments are in an italic font. Everything else is in the standard
461
% font. The consequence is indented and the continuations of the
462
% consequence are further indented. The matching parentheses are marked
463
% by a bold foreground and a shaded background. The cursor is indicated
464
% by a solid box. In the bottom panel we replaced the matching
465
% parenthesis with an unbalanced bracket. Emacs immediately marks that
466
% with the paren-mismatch font, bright purple on
469
The font selection and the indentation depth are automatically
470
supplied by Emacs as the lines are typed. The user has several
471
options for mapping of colors or fonts to each of the syntactic types.
473
% black-and-white font-mapping for display here. On a color terminal
475
purple for the keywords, red for comments, green for matching parens,
476
and inverse-video purple for mismatched parens. Emacs makes default
477
choices of colors and ESS provides several other optional schemes.
479
\begin{figure}[tbp]%h
481
\includegraphics[angle=270,width=\textwidth]{font-cor-s}
482
\includegraphics[angle=270,width=\textwidth]{font-incor-s}%
484
%\url{http://software.biostat.washington.edu/ess/doc/figures/font-cor-s.jpg}
485
%\url{http://software.biostat.washington.edu/ess/doc/figures/font-incor-s.jpg}
486
\caption{We illustrate here with fonts and colors appropriate for a
487
color display. On a black and white terminal we would use bold,
488
underline, italic, and reverse-video. On a color terminal we
489
would use a selection of colors.}
493
Since S syntax is similar to that of C, ESS uses the Emacs tools for
494
reformatting S code to match particular styles. Common C format styles,
495
as well as locally customized styles, are defined by specifying the indentation
496
level for nested statements, location of open-braces (at the end or at the
497
beginning of a line), indentation offsets for if-then-else constructs,
498
and similar characteristics.
500
Syntax highlighting can be used to help enforce coding
501
standards. Figure \ref{f.hilock} illustrates a standard for
502
\SAS\ programming that says all \stexttt{PROC} statements must use the
503
\stexttt{DATA=datasetname} option.
506
\ESSfig{\includegraphics[angle=270,width=\textwidth]{hilock-sas}}
507
%\url{http://software.biostat.washington.edu/ess/doc/figures/hilock-sas.gif}
508
\caption{Enforce coding standards. The standard here is
509
that all \stexttt{PROC} statements must use the
510
\stexttt{DATA=datasetname} option. Lines that satisfy the
511
standard turn green, lines that don't turn red.
512
Ambiguous ones turn yellow.}
516
\paragraph{Process interaction.}
517
Emacs has historically referred to processes under its control as
518
``inferior'', accounting for the name inferior ESS (\stexttt{iESS}) to
519
denote the mode for interfacing with the statistical package. Figure
520
\ref{f.ess-demo} shows the S language program \stexttt{ess-demo.s} in
521
the top buffer in \stexttt{ESS[S]} mode and the executing R process in
522
the bottom buffer \stexttt{*R*}. The \stexttt{iESS} major mode of the
523
\stexttt{*R*} buffer is crafted for command-line editing. This mode
524
remembers and uses the command history, allowing for the recall and
525
searching of previously entered commands. Filename completion for
526
local directories is also available.
529
\ESSfig{\includegraphics[angle=270,width=\textwidth]{ess-demo}}
530
%\url{http://software.biostat.washington.edu/ess/doc/figures/ess-demo.jpg}
531
\caption{Line-by-line execution of a command file. The cursor is
532
placed on a line in the \stexttt{ESS[S]} buffer and then with a single
534
the line is sent to the \stexttt{*R*} buffer for
535
execution. The output of the package goes directly to the
536
editable \stexttt{*R*} buffer.}
540
\paragraph{Source-level Debugging.}
541
ESS facilitates the editing of source code files, sets of commands
542
written for a statistical analysis package, and allows the user to
543
load and error-check small sections of source code into the package.
544
This is done through several mechanisms. First, the presence of
545
unbalanced parentheses or mismatched/unterminated quotes is
546
immediately evident with syntactic highlighting of the source code.
547
Second, functions are provided for simple and consistent execution of
548
user-specified or natural units of the code (function definitions in S
549
or \XLispStat, \stexttt{PROC \dots\ RUN;} sections in \SAS). An
550
error-free evaluation lets the user execute the next section of code;
551
if errors arise, the user edits the current unit and re-evaluates.
552
Once the code is verified, an entire buffer, or file, of code can be
553
sent to the package as a unit. This file can also be used as a batch
554
file for routine analysis at a later time. Finally, output from the
555
statistics package is normally captured directly by Emacs and placed
556
into a buffer from where it can be edited and searched. Particular
557
forms of output such as requests for help pages and log-file output
558
can be diverted into special buffers with modes crafted to facilitate
559
reading. These modes include tools for automatically placing the
560
cursor on the first \stexttt{ERROR}, for example in \SAS\ and S.
562
\paragraph{Interactive transcripts.}
563
A transcript records all commands entered by the analyst and the
564
corresponding text-based responses such as tables and comments
565
generated by the statistics package during an interactive statistical
566
analysis session. Once a transcript file is generated, for example by
567
saving an \stexttt{iESS} buffer, \stexttt{transcript-mode} assists
568
with reuse of part or all of the entered commands. ESS understands
569
the transcript's syntax, especially the potential prompt patterns used
570
during the interactive analysis. ESS provides tools to facilitate
571
editing and re-evaluating the commands directly from the saved
572
transcript. This is useful both for demonstration of techniques and
573
for reconstruction and auditing of data analyses. Special ESS
574
functions can ``clean'' S language transcripts by isolating all input
575
lines and placing them in a new S language source file. Transcript
576
cleaning facilitates the use of an exploratory interactive analysis
577
session to construct functions and batch files for routine analysis of
580
\paragraph{Remote access to statistics packages.}
581
ESS provides transparent facilities for editing files and running
582
programs which might reside on numerous remote machines during the
583
same session. The remote machine could be a different platform than
586
\paragraph{Manipulating and Editing Objects (S family).}
587
For languages in the S family, ESS provides object-name completion of
588
both user- and system-defined functions and data. ESS can dump and
589
save objects (user- and system-generated) into formatted text files,
590
and reload them (possibly after editing).
592
\paragraph{Help File Editing (R).}
593
ESS provides an R documentation mode (\stexttt{Rd-mode}) which assists
594
in writing help files for R functions, objects, and other topics worth
595
documenting. \stexttt{Rd-mode} provides the ability to view and
596
execute code embedded in the help file in the same manner as ESS
597
handles code from any S language source file. It provides syntax
598
highlighting and the ability to submit code directly to a running ESS
599
process, either R or \Splus, for evaluation and debugging. This
600
latter feature is useful for ensuring that code developed using R runs
603
\paragraph{Cooperation across Multiple Tools.}
604
Statistical packages are intended for either general statistical
605
analyses or for specialized forms of statistical analyses. The
606
specialized statistical packages can be far more efficient for their
607
intended activities, but this is balanced by their inability to
608
perform a wide range of general statistical functions. Tightly
609
coupled inter-operability between general and specialized packages
610
rarely exists, but such a facility is often desired. For example, a
611
general purpose package such as R does not perform Bayesian analyses
612
as easily as BUGS does. On the other hand, BUGS lacks breadth in the
613
range of analyses and results it can generate. For this reason, BUGS
614
is often distributed with R packages, like the diagnostic packages
615
CODA and BOA, which assist with importing and analyzing the results in
616
R. Another point of contention is the difference in the interfaces
617
between general packages and specialized packages. ESS helps by
618
providing a single point of contact to both tools, though the typical
619
interfaces (interactive for R, batch for BUGS) are different.
621
%\item[Rodney:] I can't speak for everyone, but the BUGS users I know are
622
% .. lots deleted by AJR
623
%will make sense. Besides, the most pressing need for me is to get
624
%ESS-elsewhere to work with ESS[BUGS] rather than creating inferior-BUGS.
626
%\item[Rich:] How does making ``ESS-elsewhere work with ESS[BUGS]'' differ
627
%from ``creating inferior-BUGS''? My question is predicated on the assumption
628
%that ESS-elsewhere is a (generalized) minor-mode that makes the location
629
%of the program irrelevant.
630
%See my ESS-elsewhere quibble below.
632
%\item[Rodney:] It's the whole batch BUGS vs. interactive BUGS thing. Batch
633
%BUGS with ESS-elsewhere; interactive BUGS with inferior-BUGS which does not
634
%and will never exist. If you want to call ``inferior-ESS'' ``ESS-elsewhere''
635
%why do you need two different names? ``inferior-ESS'' is a terrible name
636
%so the change would be fine with me, but you can't have it both ways? Is
637
%this why you keep saying that ESS-elsewhere works for SAS? See
638
%response to quibble below.
641
%\stexttt{iESS[SAS]} was designed to mimic as well as possible \stexttt{iESS[S]}.
642
%I need to read doc/README.SAS to make sure all of its options are represented
643
%here. --- Not yet done.
647
\paragraph{Simplifying Keymap Differences.}
648
Simple conflicts between interfaces are exemplified by different key
649
sequences for editing tasks such as cut, copy, paste, beginning of
650
line, end of line, etc. These may be the most aggravating because our
651
fingers are typing ``instinctively'', but differences in interfaces
652
circumvent this learned behavior. ESS solves this problem by
653
providing a uniform interface to keyboard actions across the variety
654
of statistical packages that might be used. That is, the same key
655
sequences are used for cursor movement, evaluation, and basic tasks
656
such as loading files for editing.
658
\paragraph{Concurrent Use of Multiple Machines and Operating Systems.}
659
It can be useful to have multiple statistical processes running
660
simultaneously, either on a single machine or a variety of machines.
661
This capability assists with large-scale numerical simulations as well
662
as code design and testing across multiple versions of statistical
665
\subsection{Interactive Processing.}
666
\label{sec:interactive}
668
The increased popularity of exploratory data analysis as well as the
669
advent of simple GUIs has made interactive data analysis an important
670
component to statistical practices.
671
ESS uses three different approaches for communicating with statistical
674
\paragraph{Inter-Process Communication.}
675
Packages that use the command-line interface are run as an inferior
676
process in an Emacs buffer, with the standard input and output of the
677
package redirected to the buffer. Packages that do not use the
678
command-line interface must be run as an independent process, possibly
679
with limited cooperation.
681
ESS can use the Windows DDE (Dynamic Data Exchange) protocol to
682
provide one-way communication directly to packages which function as a
683
DDE server. ESS can control the actions of the package, but it can not capture
684
the results directly. Transcripts must be physically copied to an
685
Emacs buffer to get the transcript editing features.
687
Statistical packages that use neither the standard input/standard
688
output protocol nor DDE can not be directly controlled by Emacs. But, ESS
689
can still provide an editing environment for these statistical languages. The
690
user must either manually cut and paste the edited code into the
691
package or save the edited files and run them in a batch environment.
692
The Microsoft Windows versions of \SAS, \Stata, and \XLispStat\ are
695
One useful extension in ESS is relaxation of the requirement that the
696
statistics program be available on the local machine. ESS provides
697
both transparent editing of files and execution of statistics packages
698
on a remote machine with \stexttt{iESS[S]} or \stexttt{iESS[SAS]} (see
699
below). All the editing and interaction features described for the
700
local machine work equally well on the remote machine. The
701
interaction, including all the unique features of working with ESS,
702
appears to the user as if the program were running on the local
703
%rmh: The interaction ... appears ....
704
machine. If the X11 Windowing system is running on the local machine,
705
it is even possible to bring up visual displays and graphics from
706
remote Unix systems onto a local Microsoft Windows or Apple Mac
709
\paragraph{Interactive S family}
710
ESS for S family statistical languages, \stexttt{iESS[S]},
711
replaces the \Splus\ Commands window or the R GUI window. In addition
712
to running the S family language process, \stexttt{iESS[S]} mode provides the
713
same editing features, including syntactic highlighting and
714
string-search, as the editing mode \stexttt{ESS[S]}. It also provides
715
an interactive history mechanism; transcript recording and editing;
716
and the ability to re-submit the contents of a multi-line command to
717
the executing process with a single keystroke. \stexttt{iESS[S]} is
718
used with S, \Splus, and R on Unix and with Sqpe and R on Windows.
720
The \Splus\ GUI on Windows can be a DDE server. There are two
721
advantages to using even this limited communication with the \Splus\
722
GUI through ESS. First, through \stexttt{ESS[S]} mode the user gets
723
the full editing capabilities of Emacs. Second, S language commands
724
% rmh: S, not \Splus, in both places. The {\it language} is S.
725
% The previous session could be R or ATT S. It is not restricted to \Splus.
726
are sent from the editing mode \stexttt{ESS[S]} buffer and from
727
transcript buffers from previous S sessions directly to the GUI
728
Commands window with the same Emacs key sequences as are used with ESS
729
on Unix. Hence the user can work in a powerful editing environment
730
and is protected from the delay and ergonomic challenges of using the
731
mouse for copy and paste operations across windows.
733
\paragraph{Interactive \SAS.}
734
\stexttt{iESS[SAS]} is a mode that allows text-based \stexttt{PROC} by
735
\stexttt{PROC} interaction with an inferior buffer running an
736
interactive \SAS\ session on either the local or a remote computer.
737
\stexttt{iESS[SAS]} mode works by redirecting standard input and
738
output from \SAS\ to ESS. Currently, the \stexttt{iESS[SAS]} mode can
739
run on any computer, but the \SAS\ process it is controlling must be
740
running on a Unix machine. This process is very efficient for dial-up
741
network connections to a remote computer with \SAS\ installed. The
742
resulting interface is similar to the SAS character terminal
743
interface, but with Emacs key sequences.
745
%Rodney: What is this paragraph about? I'm going to comment it out
746
%because I don't recognize what it is supposed to be. Maybe somebody
749
% rmh: Round umpteen. Yes, this is redundant with the third paragraph
750
% of interprocess communication. I still want it here. Try this rephrasing.
752
\stexttt{ESS[SAS]} mode can be used in conjunction with the \SAS\
753
Display Manager to allow simultaneous access to Emacs for editing
754
\SAS\ language code and to the \SAS\ mouse-based interfaces to the
755
graphical routines and help system.
757
%%%%% AJR: WHY IS THIS SUBSTANTIALLY DIFFERENT THAN BATCH? I KNOW ITS
758
%%%%% SLIGHTLY DIFFERENT, BUT SUBSTANTIALLY?
759
%%%%% rmh: this is a form of interaction, not batch. I restored the
760
%%%%% first paragraph with some expansion.
761
%\paragraph{\SAS---Interactive cooperation with the \SAS\ Display Manager.}
762
%ESS users who write data analysis code in \stexttt{ESS[SAS]} mode in Emacs
763
%often need to work with the \SAS\ Display Manager's
764
%mouse-based interface to the graphical
765
%routines, the help system, and other non-text-based features.
766
%%The authors of ESS prefer the Emacs environment for
767
%%the text-file interaction with \SAS, that is with editing and
768
%%managing input command files and output listing and log files,
769
%%even on computer systems which run
770
%%the \SAS\ Display Manager environment.
771
%In this situation, the user
772
%designs the command file in \stexttt{ESS[SAS]} mode and highlights
773
%regions to be forwarded to \SAS\ for processing.
775
%% Rodney: I don't see this as a feature of either ESS or emacs. And,
776
%% what the authors prefer is certainly not germane.
779
%% rmh: This is my preferred mode for interacting with SAS. I tried another
780
%% rephrasing. It {\it is} a feature of ESS. ESS is able to provide ESS[SAS]
781
%% mode for the text processing (.sas .lst. log files) and simultaneously
782
%% let the user have interactive graphical access.
783
%%This can be done by either:
785
%%\item copying and pasting the marked regions to the \SAS\ Editor window
786
%% and then pressing the \stexttt{RUN} button. Highlighted sections of
787
%% the \SAS\ Listing window are brought back to Emacs to be read in the
788
%% \stexttt{ESSlst} mode editing environment.
789
%%\item submitting the marked region for Batch File Processing (see the
790
%% next section) but using the mapped key sequences to append to the log
791
%% and listing files instead of replacing them.
794
\subsection{Batch File Processing.}
795
\label{sec:batch-file}
797
Batch file processing with statistical analysis packages is a better
798
choice than interactive processing when the execution times are longer
799
than the user is willing to wait as well as for regularly updated
800
statistical reports and figures. ESS provides a means to shorten the
801
debugging cycle for writing code intended for batch evaluation by
802
containing the whole process, both writing and evaluation, within
805
\paragraph{Batch \SAS.}
806
\label{sec:sas-batch}
808
\SAS\ is a popular choice for processing and analyzing large amounts
809
of data. However, interactive \SAS\ is rarely used in these situations
810
due to the length of time involved. Instead, a file containing \SAS\
811
commands is created and \SAS\ executes these commands in the background,
812
or batch, while the user moves on to other activities.
814
ESS facilitates \SAS\ batch with \stexttt{ESS[SAS]}, the mode for files
815
with the \stexttt{sas} extension. ESS defines \SAS\ syntax so that
816
\stexttt{font-lock-mode} can highlight statements, procedures,
817
functions, macros, datasets, comments and character string literals in
818
\SAS\ programs. Optionally, the same language features are
819
highlighted in the \SAS\ log with the addition of log notes, warnings
822
For files with the \stexttt{sas} extension, ESS binds the function
823
keys in \stexttt{ESS[SAS]} mode to match the definitions used by \SAS\
824
Display Manager. These definitions are optionally available in all
825
modes. They are particularly useful when viewing \SAS\ log and
826
listing files (with extensions of \stexttt{log} and \stexttt{lst}
829
Only one function key press is needed to submit a \SAS\ batch process.
830
Other function keys open the \SAS\ program, the \SAS\ log and the
831
\SAS\ listing buffers. When accessed in this manner, the \SAS\ log
832
and \SAS\ listing buffers are automatically updated since they may
833
have been appended or over-written by the \SAS\ batch process. In
834
addition, the \SAS\ log is searched for error messages and the error
835
messages, if any, are sequentially displayed with consecutive key
838
Another function key opens a \SAS\ permanent dataset for editing or
839
viewing. An option is provided so that the tab and return keys
840
operate in typewriter fashion like they do in \SAS\ Display Manager.
841
This option also defines a key to move the cursor to a previous
842
tab-stop and delete any characters between its present position and
843
the tab-stop. This is a \SAS\ Display Manager feature that is not
844
typically available in Emacs.
846
The \SAS\ batch process runs on the computer where the \SAS\ program
847
resides. This is important because any \SAS\ permanent datasets
848
referenced in a \SAS\ program only exist on the computer running \SAS.
849
If the \SAS\ program resides on a remote computer, then the
850
log and listing are also accessed remotely. The net result is that
851
running \SAS\ batch on remote computers is nearly transparent to the
857
%The \SAS\ batch process can run on the same computer on which the
858
%emacs session is running or it can run on a remote computer. For
859
%remote jobs, files are transparently saved (with ftp or scp or kermit)
860
%and the batch process is transparently submitted through a telnet or
864
%Rich: I have some terminology quibbles here with how the term
865
%``ESS-elsewhere'' is used with SAS BATCH. I think of S-elsewhere or
866
%ESS-elsewhere as a trick to make a \stexttt{telnet-mode} buffer think
867
%it is \stexttt{iESS-mode} buffer.
869
%I don't think of file saving, editing, and retrieval as an example
870
%of ESS-elsewhere. Neither is submission of the remote job;
871
%that is just an ordinary shell command in an ordinary shell buffer.
872
%M-x SAS probably is an example of ESS-elsewhere, but I
873
%designed it before I thought of the ESS-elsewhere concept.
875
%The initial idea behind S-elsewhere was to run an interactive S or S-Plus
876
%session on a remote computer in telnet (or equivalent) buffer. The trick
877
%was to make the \stexttt{telnet-mode} buffer accept C-c C-n and
878
%related commands from the \stexttt{S-mode} buffers. Hence I had to
879
%make the \stexttt{telnet-mode} buffer think it is \stexttt{iESS-mode}
880
%buffer. The ``elsewhere'' part of the name is entirely related to a
881
%different start up procedure. Once the connection is made, there is
882
%{\em no} difference visible to the user. The buffer shows itself to
883
%be an ordinary \stexttt{iESS} buffer. Tony generalized S-elsewhere
884
%to ESS-elsewhere to allow other languages than S to be used
887
%I have been using ESS for SAS remote BATCH for years, ever since you
888
%and I started working on this together. We initially defined
889
%ess-sas-submit-method to encapsulate the location of the sas process.
890
%Except for a few lines of elisp to get the connection started, the user
891
%behavior has been identical whether the SAS process is on the same or
892
%different machine. Since we are not interactively talking to the SAS
893
%process in an inferior-ESS buffer I don't see this as ESS-elsewhere.
895
%Rodney: Remote submissions of SAS batch jobs never worked. I only got
896
%it to work a couple of weeks ago. The problem was with the cd command.
897
%You need to ignore the beginning and end of the expanded buffer name
898
%that are the ange-ftp/EFS/tramp stubs which tell Emacs what the remote
899
%username and hostname are. I find the batch usage of ESS-elsewhere
900
%entirely consistent with the interactive behavior. OTOH, I don't find
901
%the terminology particularly illuminating. I think ESS-remote or
902
%ESS-net would be more meaningful.
905
\paragraph{Batch BUGS.}
906
BUGS software performs Markov Chain Monte Carlo integration. There is
907
an interactive capability, but it is not often used since the analyses
908
can be very time-consuming. Most BUGS programs are executed as batch
909
processes. ESS facilitates BUGS batch with \stexttt{ESS[BUGS]}, the
910
mode for files with the \stexttt{bug} extension. ESS provides 4
911
features. First, BUGS syntax is described to allow for proper
912
fontification of statements, distributions, functions, commands and
913
comments in BUGS model files, command files and log files. Second,
914
ESS creates templates for the command file from the model file so that
915
a BUGS batch process can be defined by a single file. Third, ESS
916
provides a BUGS batch script that allows ESS to set BUGS batch
917
parameters. Finally, key sequences are defined to create a command
918
file and submit a BUGS batch process.
920
\paragraph{Batch S family.}
921
ESS provides 2 facilities for batch processing of S family language files.
922
The first is to execute the contents of a file using buffer-evaluation. This
923
differs from interactive processing only by the number of commands
924
being evaluated; errors can be found by examining the resulting
925
transcript. The second is the load-source mechanism, which provides a
926
means of jumping to errors in the source file, but doesn't display the
927
evaluated commands in the transcript. These mechanisms provide
928
different tools for debugging the source files.
930
\section{History of ESS}
931
\label{sec:ESS:history}
933
ESS is built on Emacs, the editing system for which Richard Stallman
934
won a MacArthur Foundation Fellowship in 1990. Emacs has a long
935
history of being a programmer's editor. Many statisticians got their
936
first taste of the power of Emacs with \Fortran\ mode which was
937
introduced in 1986. As statisticians' preferences changed from
938
generalist compiled languages such as \Fortran\ to specialist
939
statistical analysis packages like S and \SAS, Emacs modes soon
942
The ESS environment is built on the open-source projects of
943
many contributors, dating back over 10 years.
944
Doug Bates and Ed Kademan wrote S-mode in 1989 to edit S and \Splus\
945
files in GNU Emacs. Frank Ritter and Mike Meyer added features,
946
creating version 2. Meyer and David Smith made further contributions,
947
creating version 3. For version 4, David Smith provided significant
948
enhancements to allow for powerful process interaction.
950
John Sall wrote GNU Emacs macros for \SAS\ source code around 1990.
951
Tom Cook added functions to submit jobs, review listing and log files,
952
and produce basic views of a dataset, thus creating a SAS-mode which was
955
In 1994, A.J. Rossini extended S-mode to support XEmacs. Together
956
with extensions written by Martin M{\"a}chler, this became version
957
4.7 and supported S, \Splus, and R.
958
In 1995, Rossini extended SAS-mode to work with XEmacs.
960
In 1997, Rossini merged S-mode and SAS-mode into a single Emacs
961
package for statistical programming; the product of this marriage was
962
called ESS version 5. Richard M. Heiberger designed the inferior mode
963
for interactive \SAS\ and SAS-mode was further integrated into ESS.
964
Thomas Lumley's Stata mode, written around 1996, was also folded into
965
ESS. More changes were made to support additional statistical
966
languages, particularly \XLispStat.
968
ESS initially worked only with Unix statistics packages that used
969
standard-input and standard-output for both the command-line interface
970
and batch processing. ESS could not communicate with statistical
971
packages that did not use this protocol. This changed in 1998 when
972
Brian Ripley demonstrated use of the Windows Dynamic Data Exchange
973
(DDE) protocol with ESS. Heiberger then used DDE to provide
974
interactive interfaces for Windows versions of \Splus. In 1999,
975
Rodney A. Sparapani and Heiberger implemented \SAS\ batch for ESS, which
976
relies on files rather than standard-input/standard-output, on Unix,
977
Windows and Mac. In 2001, Sparapani added BUGS batch file processing
978
to ESS for Unix and Windows.
980
This history is summarized in Table \ref{tab:timeline}.
985
\begin{tabular}{c ll c ll}
989
& \multicolumn{2}{c}{S-mode} && \multicolumn{2}{c}{SAS-mode} \\
990
\cline{2-3} \cline{5-6}
991
1989 & v.1 & (GNU Emacs, Unix, S/S+) && \\
992
1990 & & && & (GNU Emacs, Unix, SAS editing) \\
993
1991 & v.2 & (GNU Emacs, Unix, S/S+) && \\
994
1993 & v.3 & (GNU Emacs, Unix, S/S+) && \\
995
1994 & v.4 & (GNU Emacs/XEmacs, Unix, S/S+) && v.1 & (GNU Emacs, Unix, SAS batch) \\
996
1995 & v.4.7 & (GNU Emacs/XEmacs, Unix, S/S+/R) && v.2 & (GNU Emacs/XEmacs, Unix, SAS batch) \\
997
\cline{2-6}\\[-3.5ex]
999
& \multicolumn{5}{c}{Emacs Speaks Statistics (ESS)} \\
1001
&\multicolumn{2}{c}{Emacs, Operating Systems} &&\multicolumn{2}{c}{Additional Functionality}\\
1002
\cline{2-3} \cline{5-6}
1003
1997 & v.5.0 & (GNU Emacs/XEmacs, Unix) &&& Stata, XLispStat, SAS interactive \\
1004
1998 & v.5.1.1 & (GNU Emacs/XEmacs, Unix/Windows) &&& S+elsewhere; Windows: S+/R\\
1005
1999 & v.5.1.10 & (GNU Emacs/XEmacs, Unix/Windows/Mac) &&& SAS batch; Omegahat \\
1006
2001 & v.5.1.19 & (GNU Emacs/XEmacs, Unix/Windows/Mac) &&& Unix/DOS: BUGS batch; Mac: R \\
1007
%% 2002 & v.5.1.20 & (Emacs/XEmacs, Unix/Windows/Mac) &&& Unix/DOS: BUGS batch, Mac: R \\
1008
%% 2002 & v.5.2 & (Emacs/XEmacs, Unix/Windows/Mac) &&& ? \\
1012
\caption{History and Ancestors of ESS}
1013
\label{tab:timeline}
1017
%\section{Future Work}
1019
%%%There are two active areas of extensions for user environments. One
1020
%%%is to enhance the capabilities of the IDE for statistical practice;
1021
%%%this includes implementing such common IDE features as object
1022
%%%browsers, tool-tips, and interfacing cleaning. The other is to target
1023
%%%appropriate potentially useful programming methodologies for transfer
1024
%%%to statistical practice.
1026
%%%Literate Programming methodologies \citep{Knuth:1992,NRamsey:1994} are
1027
%%%a natural fit for statistical practice. We refer to the application
1028
%%%to statistical analysis as Literate Statistical Practice
1029
%%%\citep{rossini:dsc:2001}. The tools used are Noweb
1030
%%%\citep{NRamsey:1994} and either \LaTeX, \textsc{html}, or \textsc{xml}
1031
%%%for documenting and explaining the analysis. This approach to
1032
%%%programming encourages the use of a literary documentation style to
1033
%%%explain the programming code for the data analysis. The program can
1034
%%%then be extracted from the documentation text for realizing the
1035
%%%statistical analysis.
1037
%%future enhancement perhaps
1038
%%ESS provides the same ESS-elsewhere support for BUGS batch
1039
%%that it does for \SAS\ batch (see above).
1041
%Important extensions which should be implemented in future
1042
%versions include class browsers, analysis templates, tool-tips, and
1043
%similar features. Class browsers can be thought of as a tree or
1044
%outline for presenting datasets, variables and functions in the
1045
%context of what they represent; this allows for rapid and appropriate
1046
%inspection. Analysis templates would allow statistics centers and
1047
%groups to provide standardized templates for initiating an analysis.
1049
%Additional statistical packages can easily be added to ESS.
1051
%%While most IDE features have been developed for object-oriented
1052
%%languages, the above also can apply to non-object oriented
1055
%ESS is one of the first Rapid Application Development (RAD)
1056
%environments intended for statisticians. It provides
1059
\section{Conclusion}
1062
ESS provides an enhanced, powerful interface for efficient interactive
1063
data analysis and statistical programming. It allows the user
1064
complete control over the communications among the files in which the
1065
analysis is specified, the statistical process doing the computation,
1066
and the output. Because all are within the same programming
1067
environment, and therefore are accessed with the same editing and
1068
searching concepts and the same key sequences, user efficiency is
1069
increased. It is completely customizable to satisfy individual
1070
desires for interface styles and can be extended to support other
1071
statistical languages and data analysis packages.
1074
%Rich: see my discussion of ESS-elsewhere terminology.
1076
%What is the behavior for remote SAS that is new for 2002?
1078
%Rodney: SAS batch now works and Kermit was added as a method of transfer.
1080
%ESS-elsewhere provides interactive and batch processing
1081
%with \SAS\ running on a remote machine that is accessed over a
1082
%network. This provides a powerful development environment for \SAS.
1085
%%includes support for syntax highlighting and template-based source file
1086
%%generation that provides the capability of specifying all the necessary
1087
%%parameters for a BUGS batch run in a single file.
1089
\bibliographystyle{plainnat}
1090
\pdfbookmark[1]{References}{section.7}
1091
%\addcontentsline {toc}{section}{\numberline {}References}
1092
\bibliography{ess-intro-graphs}
1097
\section{Appendix: ESS Resources on the Internet}
1098
%\addcontentsline {toc}{section}{\numberline {}ESS Resources on the Internet}
1101
\paragraph{Latest Version.}
1103
ESS is constantly in flux. New versions of statistical
1104
packages, Emacs and operating systems require new releases of ESS to
1105
support them. The latest stable version of ESS can be found on the web at
1106
\url{http://software.biostat.washington.edu/statsoft/ess/}. To get help
1107
with problems, send e-mail to \url{mailto: ess-help@stat.math.ethz.ch}.
1108
The latest development, hence unstable, version can be obtained by
1109
anonymous CVS. First type:
1112
:pserver:anoncvs@software.biostat.washington.edu:/var/anoncvs login}
1114
You will be prompted for a password which is ``\stexttt{anoncvs}''.
1118
:pserver:anoncvs@software.biostat.washington.edu:/var/anoncvs co
1121
\paragraph{Additional documentation.}
1123
An expanded version of the present paper is in \citep{RMHHS:2001}. A
1124
general introduction and usage instructions can be found in
1125
\citep{heiberger:dsc:2001}; in addition, one which is more focused on
1126
\SAS\ can be found in \citep{heiberger:philasugi:2001}. The
1127
documentation that comes with ESS provides details of its
1128
implementation as well as examples of its use.
1133
%%% Local Variables: