Statistical Software for Today and Tomorrow

Edward J. Wegman Jeffrey L. Solka

Center for Computational Statistics Code B10

George Mason University NSWCDD
Fairfax, VA 22030 Dahlgren, VA 22448
January 13, 2005

Abstract. This article is a relatively short exposition on the current

state of the art with regards to statistical software along with some
prognostications as to the future trends in the area. The article is
somewhat biased in its discussions of the state of the art with a
necessitated focus on those packages that seem most promising with
regards to their current and future role in the academic research
communities.

Keywords and Phrases. statistical software

AMS Subject Classification. Primary: 62-04; Secondary: 68-04

1 Introduction
This article is an assessment of the state of the art of the statistical software
arena as of 2004. Much like a snapshot is only part of a movie, this article is
a snapshot of a currently evolving system which is being driven in part by the
needs of the user community and capitalistic forces. We do however make some
predictions as to what the next few frames of the “statistical software” movie
might be.
We have attempted to touch upon a few commercial packages, a few general
public license packages, a few analysis packages with statistics add-ons, and a
few general purpose languages with statistics libraries. For the most part, our
discussions are cursory in nature although we have attempted to be unbiased in
our reviews.
r r
We begin our discussions with SAS° . SAS° , in our opinion, represents the
Microsoft of the statistical software companies and hence our discussions in this

1
section are a bit more lengthy. We next proceed to BMDP. BMDP is a package
which had its birth in the bio-medical arena. Next, we proceed to SPSS. SPSS
r
is another commercial package like SAS° and in many ways it represents one
r r
of the main competitors to SAS .We then proceed to discuss S-PLUS°
°
. S-
r
°
PLUS has quickly evolved from its birth as the S language into a bona fide
commercial package with a loyal user base.
We change gears a bit in the next section and discuss a non-commercial
implementation of the S language, R. R, like LATEX, is an example of a general
public license success story. The R language benefits from the “many eyes” that
examine the code and its large user base provides a user with an alternative to
the standard “help desk” to turn to for assistance. R is currently one of our
languages of choice for the development of research level software.
The next several sections focus on commercially available mathematical anal-
ysis software that have statistical add-on packages. The first of these, MATLAB,
is also one of our favorites. MATLAB is a language with strong visualization
and matrix capabilities. A discussion of Mathematica follows. Mathematica is
touted as the de facto standard for symbolic computation with a strong graphics
capability. These capabilities make it an asset for calculations associated with
mathematical statistics.
Our next section briefly examines several general purpose software devel-
opment languages that have various statistics libraries. The first one, JAVA,
provides the user with the capability to offer students and clients access to sta-
tistical software on-line. This along with its ability to seamlessly execute on
numerous platforms running numerous operating systems makes JAVA a strong
contender as a general purpose implementation language.
We next discuss C++. C++, like JAVA, possesses the strength of being
an object oriented language. This allows one to extend pre-existing standard
template libraries in a rather straight-forward manner. C++ has been around
long enough to have a large numerical analysis user base associated with it.
The final general purpose language that we discuss is PERL, the Practical
Extraction and Reporting Language. PERL is a highly flexible language that
allows a user to craft very dense code to perform many purposes. Free form
text processing is one of the strengths of PERL and this has led to it being
embraced by many diverse communities including the bioinformatics and the
computer security communities.
In the next section of the paper we attempt to predict what will be the next
major driving forces that will influence the statistical software development
process. This is a very difficult task given the fluid nature of the commercial
and research sectors. We do believe that some of the predictions will come to
fruition though.
In the final section of the paper we review some of the ground that has been
covered within. We have chosen to make our synopsis somewhat succinct given
the rather lengthy discussions that have proceeded this point in the paper. We
close this particular section with a some very brief philosophical musings.

2
2 Current Statistical Packages

r
SAS °
r
SAS° had its birth as a statistical analysis system during the late 1960’s.
r
°
SAS grew out of a project in the Department of Experimental Statistics at
North Carolina State University. This project led to the formation of the SAS
r
Institute in 1976. Since its infancy, SAS° has evolved into an integrated system
for data analysis and exploitation.
r r
SAS° is really much more than a simple software system. The SAS° website
claims that their software resides at 40,000 sites worldwide including 90 percent
of those companies on the Fortune 500 list. We believe that this expansion has
come about in part because the SAS management has aligned themselves with
the recent “statistical-like” advances within the computer science community
such as data mining. In fact if one visits the SAS website at www.sas.com one
is hard pressed to ascertain that the site details a statistical software product.
An initial perusal of the site leads one to believe that the product that is be-
ing offered by the company is a data mining product. This clever integration
of mathematical/statistical methodologies, database technology, and business
r
applications has helped propel SAS° to the top of the commercial statistical
software arena.
r
SAS° has been utilized in a myriad of different business areas including the
automotive industry, the telecommunications industry, the life sciences area,
the home land security area, the economic forecasting area, and the waste and
fraud detection area. SAS also continues to remain one of the languages that
are typically taught at academic institutions world-wide.
r
The architecture for the SAS approach is called the SAS° Intelligence Plat-
form. This Platform is really a closely integrated set of hardware/software
components that allow a user to fully utilize the business intelligence (BI) that
can be extracted from their client base. The SAS Intelligence Platform consists
r r
of the following components: the SAS° Enterprise ETL Server, the SAS° Intel-
ligence Storage, the SAS Enterprise BI Server, and the SAS Analytic Technolo-
r
gies. Our real interest within this chapter is the SAS° Analytic Technologies,
but we believe that it is beneficial to at least touch upon the nature of the other
components.
r
The SAS° Enterprise ETL Servers are really just a means to provide a uni-
fied database environment to handle metadata and to provide multi-threaded
platform independent access to the business information within a GUI-based
environment. The SAS Intelligence Storage System serves as a repository for
the information that is ultimately analyzed using analytic application in order
r
to provide BI insights. The SAS° BI Server provides the user with universal
format type access to their collected data structures along with novice user facili-
tated interfaces to the analytic software that is used to obtain the BI insights. In
this manner, a user who is relatively new to the data analysis area can still obtain
quality insights into their data and obtain the BI that is needed to allow them

3
 r
to compete in today’s highly competitive markets. The SAS° Analytic Tech-
nologies contain the advanced mathematical and statistical algorithms along
with visualization frameworks which allow even a mathematically/statistically
unsophisticated user to interpret the insights obtained by these algorithms.
The analytic technologies are really the areas that we are most interested
in within this current tome. The SAS management has been very astute in
marketing their analytic capabilities in such a manner that the buyer can easily
ascertain their relevance to their particular problem of interest. The SAS ana-
lytic technologies include data and text mining, forecasting and econometrics,
operations research, quality improvement, and statistical analysis. We would
like to take a moment to briefly touch upon each of these analytic capabilities,
ending with the statistical analysis one.
The data and text mining capability is focused on the use of text mining
technologies to foster business applications, although one can easily formulate
alternative application in the areas of homeland security. We believe that the
r
creators of SAS° have demonstrated great foresight in extending their analy-
sis capabilities in order to provide text mining capabilities. It is the authors’
contention that text mining will continue to play a more prominent role in the
formulation of effective national scientific strategies. In fact, some of the giants
in the IT arena, Microsoft and Apple, have recently been gearing up to com-
pete with the stalwart text search company Google. There are certainly fruitful
areas for interaction between standard statistical and computational statistical
procedures and text mining.
The econometric, time series analysis, and forecasting area is an area often
r
associated with statistics. SAS° has done a good job of closely coupling time
series analysis to business applications. They provide the capability to deal
with the computational complexities associated with making large quantities of
accurate forecasts utilizing large volumes of data. They also provide novice users
with an interactive interface that facilitates their ability to make meaningful
forecasts.
Operations research and quality improvement are two areas that one might
not immediately identify as statistical applications. However, these two areas
are very relevant to the current culture within the business community. The
astute reader will, after a moments reflection, realize that optimization, the
enabling technology in operations research, and quality control/quality inspec-
r
tion have received previous attention from the statistics community. SAS° has
employed very clever business practices in bringing these to the forefront as
analytic product areas.
Finally, we turn our discussions to the statistical analysis analytic capa-
bility. Many of the previous discussions with regard to the capabilities of their
computer/database support architectures lend support to the statistical analysis
analytic capability. Their statistical analysis capability can be broken down into
the sub-areas of statistical analysis, exploratory data analysis, matrix program-
ming language, matrix programming interface, guided data analysis, market
research, and other statistical components. Let us take a moment to consider
each of these in turn.

4
 The SAS/STAT component contains those capabilities that one normally
associates with a data analysis package. These capabilities include functionality
to support analysis of variance (ANOVA), regression, categorical data analysis,
multivariate analysis, survival analysis, psychometric analysis, cluster analysis,
r
and nonparametric analysis. One of the strengths of the SAS° SAS/STAT
package is the fact that the package is constantly being upgraded with each
release in order to reflect the latest and greatest algorithmic developments in
the statistical arena.
The exploratory data analysis area is supported by the SAS/INSIGHT pack-
age. This package is designed to analyze data via linked visualization strategies.
These strategies can be linked across multiple windows and they allow an adroit
user to uncover trends, spot outliers and readily discern subtle patterns within
their datasets that would not be apparent using other analytical methods. The
hallmark of this portion of the SAS software is the interactive nature of the data
analysis procedure.
r
SAS° provides the user with a convenient matrix programming language
via their SAS/IML software system. This system allows statisticians, mathe-
maticians, and computer scientists to develop their own matrix-based analysis
systems. Formulation of many statistical algorithms are facilitated by the use
of a matrix based language.
A matrix programming interface to this language is provided within the
SAS/IML Workshop GUI. This GUI gives the user an integrated editor/debugger
for the development of SAS/IML based programs. The SAS/IML Workshop
GUI also provides the user with the ability to call SAS procedures and external
C/Fortran/Java functions.
The SAS/LAB system provides a user with a prompt-driven guided data
analysis methodology. The system uses graphical methods to represent basic
methods of statistical analysis. In this way, the user is guided through the
analysis procedure.
The last SAS analytical package that we wish to discuss is the SAS Market
Research package. This package provides the user with a point and click in-
terface that allows them to conduct many common market research functions.
Some of these functions include discrete choice analysis, cojoint analysis, and
multi-dimensional preference scaling.
Many of the methodologies that we previously discussed serve to support
the SAS BI model. These technologies provide a user with on-line analytical
processing (OLAP) capabilities in order to obtained summarized data views.
This user-based analysis is facilitated by providing the user with easy access to
his data, with an interactive data exploration environment, and with a means
to facilitate web-based reporting in order to achieve flexible content delivery.
r
The user is referred to [1] for a more in-depth look at the impact of SAS° in
the BI arena.
The BI process is of course facilitated through the use of data exploration
and analysis via visualization frameworks. The user is free to choose from a
plethora of different visualization methods including surface, prism, block, and
chloropleth maps. As is the case with virtually all modern data analysis packages

5
the user has the capability to output the graphics to any of a number of different
graphics devices.
r
In summary, SAS° has become one of the “biggest players” in the statistical
software arena. It has done a great job at marketing itself as a data mining
company. It also has aggressively expanded its client base to include numerous
companies, government agencies, and academic institutions. It will continue to
be an active statistical software company for the foreseeable future and, based
on prior company maneuvers, one can expect that SAS will aggressively position
itself to meet the future needs of the business communities.

BMDP
BMDP originated during the 1960s as a bio-medical analysis package. In
many ways BMDP has remained true to its roots and this is evidenced by
its long list of clients which includes such biomedical giants as Astra Zeneca,
Bristol-Myers Squibb, Eli Lilly, the FDA, Glaxo Wellcome, and Merck. BMDP
is currently produced by Statistical Solutions. In addition BMDP has chosen to
align itself with a number of other currently popular statistical products includ-
ing StatXact 5.0, LogXact 5.0, SOLAS, EquivTest, SigmaPlot, Meta Analysis,
and SYSTAT. These alliances have extended the base capabilities of BMDP on
many different fronts.
The current release of BMDP is Professional Version 2.0. BMDP is avail-
able for a variety of platforms/operating systems including Microsoft Windows
XP, MS DOS, HP 9000, Sun Solaris OS, IBM AIX, Dec Alpha, Dec Vax/VMS
and more. The base distribution of BDMP includes routines for data manipu-
lation, data description, group comparisons, various plots including histograms,
frequency tables, correspondence analysis, regression, maximum likelihood esti-
mation, non-linear regression, analysis of variance, multivariate analysis, non-
parametric analysis, cluster analysis, missing value analysis, survival analysis,
and time series analysis.

SPSS
The SPSS software system was developed in the late 1960s by SPSS Chair-
man of the Board Norman H. Nie, C. Hadlai (Tex) Hul, and Dale Brent. The
three of them were Stanford University graduate students at the time. Nie and
his colleagues founded SPSS in 1968. SPSS incorporated in 1975. It established
its headquarters in Chicago, Illinois where it currently resides. SPSS became
publicly traded in August of 1993.
SPSS was originally a software package designed for mainframe use. SPSS
introduced SPSS/PC+ for computers running MS-DOS in 1984. This was fol-
lowed by releases of a UNIX software version in 1988 and a Macintosh version
in 1990.
SPSS 13.0 for Windows was current as of 2004. The current release of the
r
base distribution for MAC° OS X is version 11.0. There is also a server version
of the software and and instructional versions for undergraduate and graduate
academic institutions.

6
 SPSS supports numerous add-on modules including one for regression, ad-
vanced models, classification trees, table creation, exact tests, categorical analy-
sis, trend analysis, conjoint analysis, missing value analysis, map-based analysis,
and complex samples analysis. In addition SPSS supports numerous stand-
alone products including AmosTM (a structural equation modeling package),
SPSS Text Analysis for SurveysTM (a survey analysis package that utilize nat-
ural language processing methodology (NLP)), SPSS Data EntryTM (a web-
r
based data entry package), AnswerTree° (a market segment targeting package),
°r TM
SmartViewer Web Server (a report generation and dissemination package),
r r
SamplePower° (sample size calculation package), DecisionTime° and What
r
if? (a scenario analysis package for the non-specialist), SmartViewer°
TM
for
Windows (a graph/report sharing utility), SPSS WebApp Framework (web-
based analytics package), and the Dimensions Development Library (a data
capture library).
As can be ascertained from the above discussions, the SPSS Corporation is
targeting many of the same application areas as SAS. However, it still seems
that the client bases for SAS is much more extensive than that of SPSS. The
user is referred to [2] for a recent in-depth review of SPSS as of release 12.

r
S-PLUS°
r
S-PLUS° is an extension of the statistical analysis language S. S was orig-
inally developed by Trevor Hastie, Richard A. Becker, Alan Wilks, John M.
Chambers, and William S. Cleveland while they were working for AT&T. Becker
Chamber, and Wilks provided the original description of the S language in 1988
r
[3]. A more recent tutorial on the use of S and S-PLUS° is contained in Krause
and Olsen, 1997 [4].
r
S-PLUS° is manufactured and supported by the Insightful Corporation.
R. Douglas Martin has been one of the visionaries who heavily influenced the
r
development of S-PLUS° . He has continued his association with the Insightful
Corporation but he has most recently been associated with the University of
Washington and with FinAnalytica Inc. The history of the development of S-
r
PLUS° has been characterized by contributions from such statistical luminaries
as Rob Tibshirani, Jerome Friedman, Bill Venables and Brian Ripley.Much of
the original S distribution is still available from http://lib.stat.cmu.edu.
r r
S-PLUS° is available for both Windows and UNIX platforms. S-PLUS°
provides the user with a fully extensible environment in that there is ready
r
support for the user to develop their own functions using the S-PLUS° language.
r r
In addition S-PLUS provides the user with the capability to call an S-PLUS°
°

functions within C, C++, or JAVA and the capability to call a C, C++, or

r
JAVA function from within the S-PLUS° environment.
r
Insightful also markets several add-ons to S-PLUS° and a few closely related
r
°
products. Insightful claims that S-PLUS contains over 4,200 data analysis
functions which implement modern and robust statistical procedures. One of
r
the strong points of S-PLUS° is its graphical capabilities. The graphics within
r
°
S-PLUS are “head and shoulders” above many of the the other statistical

7
 r
packages. The S-PLUS° based distribution comes well equipped to perform
many of the standard statistical analysis including the ability to generate ran-
dom data from many of the standard distributions, the ability to perform nu-
merous hypothesis tests including Student’s t-test and the Wilcoxon test. The
package also allows the user to perform linear, nonlinear, and projection pursuit
regression. In addition there are capabilities for multivariate analysis, discrimi-
nant analysis, and clustering. Standard time series analysis techniques are also
r
supported within the S-PLUS° environment.
r
For those customers looking to use S-PLUS° to analyze large datasets and
to conduct the analysis using moderately skilled technicians, Insightful offers In-
sightful Miner 3. This is a highly scalable workbench that allows non-technical
users to deploy predictive intelligence systems throughout the business enter-
prise. Insightful Miner is used by such clients as the Bank of America and
Pfizer.
r
The specialized S-PLUS° add-ons include S+ArrayAnalyzerTM (a microar-
ray analysis module), S+NuOPTTM (a portfolio and general optimization pack-
age), EnvironmentalStats for S-PLUS (a set of functions for performing statis-
tical and graphical analysis on environmental data), S+FinMetricsTM (an eco-
nomics analysis package), S+SeqTrialTM (a package for designing, monitoring,
r
and analyzing clinical trial using group sequential methods), S-PLUS° for Ar-
r
°
cView GIS (a package for the integration of S-PLUS and ArcView GIS 3.2)
and S+SpatialStatsTM (a spatial statistics analysis package).
r
In closing we would like to point out that we expect that S-PLUS° will
continue to receive a strong support from the academic and commercial sectors.
There are of course subpopulations of users in particular discipline areas that are
adamant about the benefits of S-PLUS. The user is referred to [5] for in-depth
r
discussions with regards to S-PLUS° .

R
R originated at the University of Auckland, New Zealand in the early 1990’s.
Ross Ihaka and Robert Gentleman were looking for a statistical environment to
use in their teaching lab. At the time, the labs were populated with Macin-
tosh computers and there was no suitable statistical software available for the
Macintosh environment. The two of them had some previous familiarity with S
and so they decided to implement a new language that would be based on an
S-like syntax. Initial versions of R were provided to statlib at Carnegie Mellon
University and the preliminary user feedback indicated a positive reception for
the language. This positive response encouraged them to release R under the
Open source software initiative. The first version of R was released to the public
in June of 1995. About the same time Martin Mächler came on board the R
development team.
The Open source software license agreement allows users to modify the R
software as long as their changes also fall within the Open source software license
agreement. The Open source software paradigm had previously worked well for
such software as LINUX and APACHE and this also proved to be the case for

8
R. A software system that exists under the Open source paradigm benefits from
having “many pairs of eyes” to examine the software to help insure quality of
the software. Support for R is provided via an on-line mailing list. This list
allows users to pose questions to the R “collective consciousness.” Often the
users receive responses within a matter of minutes.
R developed a huge following and it soon became too difficult for the three
of them, Ross, Robert, and Martin to keep up with the submitted software
changes. This led them to select a Core group, in 1997, of around 10 members.
This group was responsible for changes to the source code. John Chambers
joined the Core team in 2000.
The R development paradigm has led to the rapid development of the R
software system. R has been endorsed by many academic institutions as an
alternative instructional language in several of their undergraduate courses. In
addition they have also utilized it in their graduate and research programs, in
part because of the ease of disseminating their developed prototype software to
the research community at large.
As of this writing, R possesses an extensive statistical capability. R is orga-
nized into various packages. The base package contains support for 2-D and 3-D
plots, including many of the more esoteric statistical plots such as star plots.
The base package also includes sample, evaluation, cumulative distribution,and
inverse cumulative distribution functions for many of the standard statistical
distributions. There is also support for one and two sample hypothesis testing
and categorical data analysis. Besides the base package there are numerous other
packages including boot (bootstrap R and S-plus functions [6]), class (functions
for classification), cluster (functions for clustering [7]), datasets (R datasets), ex-
actRankTests (exact distribution for permutation and rank tests), foreign (read
data from Mini-Tab, S, SAS, SPSS and others), graphics (T graphics pack-
age), grDevices (R graphics devices and support for colors and fonts), grid (grid
graphics package), ISwR (package to accompany Introductory Statistics with R
[8]), KernSmooth (functions functions for kernel smoothing for Wand and Jones
[9]), lattice (Lattice Graphics), MASS ( main package of Venables and Ripley’s
Modern Applied Statistics with S [10]), methods (formal methods and classes),
mgcv (generalized additive models with generalized cross validation smoothness
estimation and generalized additive mixture models by restricted maximum
likelihood and penalized quassi-likelihood), nlme (linear and nonlinear mixed
effects models), nnet(feed-forward neural networks and multinomial log-linear
models), rpart (recursive partitioning), spatial (functions for kriging and point
pattern analysis), splines (regression spline functions and classes), stats(the R
stats package), stats4 (statistical functions using S4 classes), survival (survival
analysis, including penalized likelihood), tcltl (TCL/Tk interface), tools (tools
for package development), and utils (the R utils package).
There are also several major projects that are “R spin-offs.” “Bioconduc-
tor” is an R package for gene expression analysis. “Omega” is an R package
that is focused on providing seamless interface between R and a number of
other languages including PERL, PYTHON, MATLAB, and numerous others.
“gRaphical models” is an R package for graph theoretic modeling. “R GUISs”

9
is a project for the development of a GUI-based R package. R, for the most
part, is a command line based language. There is a small contingent that is
interested in a fully GUI-based R language. “R spatial projects” is an ongoing
effort to extend the spatial statistics capabilities that are resident in the R base
distribution. The user is referred to [11] fro a more in-depth discussion of R and
its related projects.

3 Current Analysis Packages with Statistical Li-

braries

MATLAB
MATLAB had its birth as a set of FORTRAN subroutines for solving linear
(LINPACK) and eigenvalue (EISPACK) problems. These routines were devel-
oped primarily by Cleve Moler in the 1970’s. He, Moler, later went on to teach
mathematics courses. During these courses he wanted his students to have ready
access to LINPACK and EISPACK without needing to know FORTRAN. Hence
he developed MATLAB as an interactive system to provide his students with
access to LINPACK and EISPACK. Originally MATLAB gained much popular-
ity primarily through word of mouth since it was not being officially distributed
at the time. MATLAB was rewritten in C during the 1980’s. This rewrite
provided the user with more functionality including routines for data plotting.
The parent MATLAB company, the Mathworks, Inc., was created in 1984. The
Mathworks, which is located in Natick Massachutes, is now responsible for the
sale, development, and support of MATLAB.
MATLAB has grown to have a very large user-based in government, academia,
and the private sector. The base distribution of MATLAB allows a user to read
and write data in ASCII, binary, and MATLAB proprietary format. The data
is presented to the user as an array, the MATLAB generic term for a matrix. A
scaler is simply a 1 x 1 array and a n by m matrix is a n by m array. The base
distribution comes with all of the expected mathematical functions including
trigonometric, inverse trigonometric, hyperbolic, inverse hyperbolic, exponen-
tial ,and logarithmic. In addition, MATLAB provides the user with access to
cell arrays, a data structure that can have different types of data types in each
i, j entry of the array and structures which allow a user to create a simple data
object in a manner that is analogous to a C or C++ data structure. MATLAB
also provides the user with various numerical methods including optimization
and quadrature functions.
One of the real strengths of MATLAB is its data plotting capabilities. MAT-
LAB supports standard two and three dimensional scatter plots along with sur-
face plots. In addition MATLAB provides the user with a graphics property
editor. The graphics property editor allows the user to customize virtually ev-
ery aspect of the appearance of a graph and it also serves as a framework for
GUI development.

10
 MATLAB, like R, is fully user extensible. The user is provided with an
integrated text editor that serves as a debugger for code development. One can
write either scripts or functions using the MATLAB language. Functions are
automatically compiled prior to execution.
MATLAB supports many additional toolboxes including a statistics toolbox.
The statistics toolbox supports 20 different probability distributions. There are
five associated functions with each distribution including: a probability density
function, a cumulative distribution function, an inverse cumulative distribution
function, a random number generator, and the mean and variance as a function
of the parameters. MATLAB also provides functions for computing confidence
intervals and parameter estimates for many of these 20 functions. The toolbox
also provides support for descriptive statistics, linear models, nonlinear models,
hypothesis tests, multivariate statistics, statistical plots, statistical process con-
trol, design of experiments, and hidden Markov models. The toolbox has been
through several iterations in its development cycle and it is rapidly evolving into
a bona fide package in its own right.
Wendy and Angel Martinez have worked diligently over the past several years
to extend the MATLAB statistical capabilities into the computational statis-
tics realm. Their first book [12] helped bring MATLAB into the computational
statistics arena and the accompanying software package was written to be mostly
independent of the pre-existing MATLAB statistics package. This package pro-
vided capabilities for kernel-based probability density estimation, mixture-based
density estimation, projection pursuit, along with many other more recent com-
putationally intensive methods. Their second book [13] and associated package
focuses on exploratory data analysis. This book provides MATLAB-based code
for many of the standard statistical exploratory data analysis methods and
graphics procedures. In addition, there is a greatly expanded dimensionality re-
duction section that includes MATLAB codes and discussions on the nonlinear
isometric embedding (ISOMAP) dimensionality reduction procedure.

Mathematica
The driving force behind the development of MATHEMATICA has been
Stephen Wolfram. His preeminent role in symbolic computing was foreshad-
owed by his receiving the MacArthur Prize in 1981. During the early 1980’s he
developed a language denoted SMP (Symbolic Manipulation Processor). SMP
was written in C. In 1985 Wolfram extended SMP and renamed it Mathematica.
Mathematica is most well known as a symbolic computation package but over
recent years it has developed advanced numerical algorithms and sophisticated
GUIs to facilitate the user interaction with the compute engine. Mathematica
originally came bundled with the NeXT operating system. This certainly pro-
vided the NeXT machines with a high quality application to bundle with there
initial product line.
Like MATLAB, Mathematica provides the user with an application pro-
gramming interface (API) which enables it to call Fortran or C functions, and
to also be callable from C and Fortran external programs. Mathematica can

11
function in its own right as a programming language. In addition, Mathematica
can function as a notebook document system which allows the user to keep track
of the symbolic computations and Mathematica commands that led to a partic-
ular figure in a paper. It also provides a system for complex analysis including
image and signal processing, a repository of reams of information on numerical
methods and mathematics, and a bona fide system for numerical and symbolic
computation. The Mathematica graphics techniques have grown over recent
years to rival MATLAB and even surpass MATLAB in the area of graphical
rendering of surfaces from their intrinsic equations.
Mathematica has recently provided users with statistical analysis tools. Math-
ematica provides tools for analysis of variance, classical hypothesis testing, con-
fidence interval estimation, data smoothing, univariate and multivariate de-
scriptive statistics, nonlinear and linear regression, and a large suite of powerful
optimization techniques. Mathematica may fall a little short in its ability to
perform some of the advanced statistical graphics procedures such as multi-
pane graph linking but it makes up for this weakness with its keen capability to
aid a user in arduous symbolic calculations that occur in mathematical statis-
tics or in advanced computational statistics methodologies. The user is referred
to [14] for an in-depth discussion on the uses of Mathematica in mathematical
statistics.

4 Some General Languages with Statistical Li-

braries
In this next section we will briefly highlight the statistics libraries available with
some of the more common software development languages.

JAVA
It is difficult to assess the state of the art with regards to JAVA statistical
libraries in that there may be many custom user developed packages that we
are unaware of. Given this caveat, we would like to mention three packages.
The first package had its inception under the name WebStats. It was developed
as an instructional statistical framework by Webster West of the University of
South Carolina. WebStats has recently been named StatCrunch. StatCrunch
4.0 provides the user capability to perform interactive exploratory data analysis,
logistic regression, nonparametric procedures, regression and regression diagnos-
tics, and others. The user is referred to a recent review of StatCrunch 3.0 that
appeared in the Journal of Statistical Software [15].
Another source of JAVA-based statistics functions is the Apache Software
Foundation Jakarta math project. The math project seeks to provide common
mathematical functionality to the JAVA user community. The initial code base
will consist of the following statistical algorithms: simple univariate statistics
(mean, standard deviation and confidence intervals), frequency distributions,
t-test, chi-square test, random number generation from Gaussian, exponential,

12
and Poisson distributions, random sampling/resampling, along with bivariate
regression and correlation. The user is referred to the project’s website for
additional information [16].
The final source for JAVA-based statistical analysis that we will discuss is
the Visual Numerics JSMLTM package. The JSMLTM package provides the user
with an integrated set of statistical, visualization, data mining, neural network,
and numerical packages. One would expect that the JSMLTM package is of
very high quality given Visual Numeric’s previous releases, PV-WAVETM and
IMSLTM . The reader is referred to [17] for additional discussions on JMSLTM .

C++
We will only briefly mention two of the many C++ statistical libraries. The
first one is the GNU Object-Oriented Statistics Environment (Goose). Goose is
a C++ library that is dedicated to statistical computation. The Goose project
has several goals including the creation of a useful and complete system that
takes advantage of C++’s features to improve the clarity of statistical code
and that is also easier for programmers to use. The other two main goals of
Goose are to provide its functionality to Guile and Guppi two other ongoing
GNU projects. Goose has been developed primarily under Linux but it has
been designed to be compatible with other Unix and Win32 systems. Goose
currently supports a standard set of statistical functions including cumulative
and inverse cumulative distribution functions for many of the most common
statistical distributions. In addition Goose possesses a high-quality Mersenne
twistor-based random number generator. Goose also provides a RealSet class
which is an optimized C++ container to hold statistical data. In addition
Goose provides support for t-tests, F-tests, Kruskall-Wallis tests, Spearman
tests, McNemar’s test, and Cochran’s Q test along with an implementation of
simple linear regression models. The reader is referred to [18] for more in-depth
discussions on the Goose project.
A statistics library product that may be of more interest to Microsoft Win-
dows developers is the Probability and Statistics for .NET v3.3 by WebCab
Components. This package actually consist of five packages: statistics, discrete
probability, standard probability distributions, hypothesis testing and corre-
lation and regression. Some of the capabilities offered by these components
include standard measures of centrality and dispersion, discrete and continu-
ous probability distribution functions including random sampling, cumulative
distribution function, and inverse distribution function. Modules to perform
various hypothesis tests and to compute confidence intervals are also provided.
A strength of these modules are there support for various interfaces includ-
ing C# and C++.NET. WebCab Component’s website provides the user with
extensive documentation on their package, see [19].

PERL
The Practical Extraction and Report language, PERL, has been embraced by
a large group of users from various communities. Proponents of PERL include

13
individuals in the system administration, computer security, bioinformatics, lan-
guage processing, and financial communities. One can often write PERL code
that it much shorter than equivalent code in other languages. PERL is partic-
ularly well known for its parsing and string matching capabilities. All of these
reasons led us to include a brief mention of it within the current chapter. The
comprehensive PERL archive network, CPAN) serves as a common repository
for much of the user community developed PERL code. A simple search at this
site reveals that there have been many different PERL statistics packages devel-
oped. These packages include some for simple descriptive statistics calculation,
linear regression analysis, chi square testing, rank correlation computation, and
t-test and dependent t-test. One can peruse the additional PERL functionality
at the CPAN site [20].

5 The Future of Statistical Computing

We feel that there are numerous challenges and opportunities that will help
shape the future of statistical computing. The first challenge is provided by the
non-expert user. There are more and more individuals who wish to apply sta-
tistical techniques to analyze their financial, biological, chemical, astronomical
or other type of dataset. These users usually have only a modicum of statistical
training and it is easy for them to inadvertently perform erroneous analysis on
their data. We believe that there remains much work to be done in providing
the user with computer software based aids to oversee their application of sta-
tistical techniques. We believe that these aids will ultimately take the form of
some sort of “expert system” or other artificially intelligent package and that
these aids will help the common user avoid pitfalls that are often encountered
by the novice user.
Novice users are not the only challenge to the future of statistical computing.
The statistical computing community has previously encountered the challenge
of “data in the wild” as part of their support of data mining efforts. By “data in
the wild”, we mean data that was collected under uncontrolled circumstances.
This sort of situation greatly complicates the nature of the analysis of this type
of data. We believe that this challenge will continue to exist and that it may
even grow in coming years. In fact the evolution of this “data in the wild” can
help drive the software development process on several fronts.
The first front is characterized by the need for the analysis of text data. Text
data analysis can be necessitated by the needs of computer security, bioinfor-
matics, automated scientific discovery, and intelligence research. Statistics has a
role to play in the analysis of features that might semantically capture the mean-
ing of a collection of text documents or as a means of analyzing simpler “word
count histogram” descriptions of text documents. We have recently completed
the first phase of a project to automate the discovery of serendipitous relation-
ships between disparate discipline areas [21], [22]. It is our contention that the
analysis of such data will play a more prominent role in today and tomorrow’s
document rich environment. We think that our position is well-founded based

14
on the recent interest in text mining of several computer giants, Microsoft and
r
Apple, and major statistical software companies, SAS° and SPSS.
The second front is characterized by streaming data. Streaming data can
be thought of as a dataset that manifests itself over a period of time in a serial
manner. Many sorts of data fall into the streaming category including video,
computer security, and voice. Computer security data can take the form of
session data, where a human operator interacts with the computer via an on-
going session, internet messenger chat data, simple packet exchange headers,
and many other types. The streaming nature of the data would suggest the
consideration of recursive algorithms. In fact one of the largest challenges of
the streaming data type is that visualization of said data requires graphics tech-
niques that evolve with the data. As the data changes its nature, dimensional-
ity, type (continuous vs. discrete), the visualization software must adapt/evolve
appropriately. See for example [23], [24]. for some recent discussions on these
matters. Software developers and various government agencies are only now
gearing up for concerted efforts in this area.
The third front is based on the need for statistical software in graphi-
cal/network based models. Network based models have recently received at-
tention for their use in social network analysis (SNA). SNA and in particular
dynamic SNA is an area which is of great interests not only to the sociology
community but also to those individuals prosecuting the wars on drugs and ter-
rorism. The application of statistics to SNA in general and dynamic SNA in
particular is still in its early stages. It will be interesting to see which of the ma-
jor statistical software companies come forward with a toolbox/package aimed
at graph theoretic modeling and analysis. Right now, R and Mathematica have
some preliminary offerings in this area.

6 Concluding Remarks
We have provided an “evaluation” of the current state of statistical software.
Our treatment has been tightly focused based on our own experiences and prej-
udices in this area. We have, however, attempted to be “even handed” in our
appraisal of the discussed software.
We have also attempted a limited prognostication with regard to the future
of statistical software. We have attempted to identify those issues and those
applications that are likely to shape the future of the area. We realize of course
that the future is usually full of surprises. These surprises are what makes for
an interesting journey as we go through life.

Acknowledgments
The work of the first author, Edward J. Wegman, was support by the Defense
Advanced Research Projects Agency under Agreement 8905-48174 with The
Johns Hopkins University. The work of the second author, Jeffrey L. Solka,

15
was supported by the Office of Naval Research (ONR) In-house Laboratory
Independent Research (ILIR) Program

References
[1] On-line News Published in DMRreview.com (2004).
New SAS 9 Software Revolutionizes the BI Industry
www.dmreview.com/article sub.cfm?articleId=1001037.

[2] Hilbe, J. (2003). A Review of Concurrent SPSS Products SPSS 12,

SigmaPlot 8.02, SigmaStat 3.0, Part 1. The American Statistician, Vol.
57, No. 4, 310–315.
[3] Becker, R. A., Chambers, J. M., and Wilks, A. R. (1988). The New S
Language. Wadsworth and Brooks/Cole, Pacific Grove, CA.
[4] Krause, Andreas and Olson, Melvin (1997). The Basics of S and S-Plus
(Statistics and Computing) Springer-Verlag.
[5] The Insightful Corporation (2004). The Insightful Webpage.
www.insightful.com
[6] Davisin, A. C., Hinkley, D. V., and Gill, R. (1997). Bootstrap Methods and
Their Application. Cambridge University Press.
[7] Kaufman, Leonard, and Rousseeuw, Peter J.(1990). Finding Groups in
Data:An Introduction to Cluster Analysis. Wiley-Interscience.
[8] Dalgaard, Peter(2002). Introductory Statistics with R. Springer-Verlag.
[9] Wand, M. P., Jones, M. C. (1994). Kernel Smoothing. Chapman and
Hall/CRC.
[10] Venables, W. N., Ripley, Brian D. (2002). Modern Applied Statistics with
S. Springer-Verlag.
[11] The R software compendium (2004). The Comprehensive R Archive Net-
work. cran-r-project.org
[12] Martinez, Wendy L., Martinez, Angel R. (2001). Computational Statistics
Handbook with Matlab. CRC Press.
[13] Martinez, Wendy L., Martinez, Angel R.. (2004). Exploratory Data Anal-
ysis With Matlab (Computer Science and Data Analysis). Chapman and
Hall/CRC Press.
[14] Rose, Colin, Smith, Murray D. (2002). Mathematical Statistics with MATH-
EMATICA. Springer-Verlag.

16
[15] West, R. Webster, Wu, Y., and Heydt, D. (2004). An Introduction to
StatCrunch 3.0. Journal of Statistical Software, Volume 9, Issue 6.

[16] The Apache Foundation (2004). The Jakarta Proposal for math Package.
jakarta.apache.org/commons/math/proposal.html

[17] Visual Numerics Incorporated (2004). JMSLTM Numerical Library for

JAVA Applications. www.vni.com/products/imsl/jmsl/jmsl.html
[18] GNU Project of the Free Software Foundation (2004).
JMSLTM Goose: The GNU Object-Oriented Statistics Environment.
www.gnu.org/software/goose/goose.html
[19] WebCab Components Website (2004). Probability and Statistics for .NET
v 3.3. www.webcabcomponents.com/dotNET/dotnet/pss/index.shtml
[20] CPAN Website (2004). Categories >> Data Type Utilities >> Statistics.
cpan.uwinnipeg.ca/chapter/Data Type Utilities/Statistics

[21] Solka, J. L., Wegman, Edward J., and Bryant Avory C. (2005). Identifying
Cross Corpora Document Associations Via Minimal Spanning Trees. Inter-
face 2004:Computational Biology and Bioinformatics 36th Symposium on
the Interface to appear.
[22] Solka, J. L., Wegman, Edward J., and Bryant, Avory, C. (2005). Text Data
Mining With Minimal Spanning Trees. in Handbook of Statistics, C. R.
Rao, Edward J. Wegman, Jeffrey L. Solka (Eds). to appear.

[23] Wegman, Edward J.(2003). Evolutionary Graphics and Recursive Algo-

rithms for Streaming Data. Keynote Address, Conference on Applied Statis-
tics in Ireland Mullingar, Ireland.
[24] Wegman, Edward J.(2003). Evolutionary Graphics for Streaming Data.
Joint Statistical Meeting 2003 San Francisco, CA.

Further Reading
Cti Statistics (2004). Alphabetical list of reviews.
www.stats.gla.ac.uk/cti/activities/reviews/alphabet.html
Jan de Leeuw (2004). Journal of Statistical Software.
www.jstatsoft.org
Scientific Computing World (2004). Software Reviews from Scientific Comput-
ing World. www.scientific-computing.com/reviews.html

Related Entries: ESS2, LATEXtemplate, Wiley

17