16 July 1998
The graphical interface
SHARON KARDIA
Sharon Kardia is in the Department of Epidemiology at the University of Michigan, Ann Arbor, MI, USA
This is a comparative review of Windows-based graphics packages that complements reviews of statistical packages for Windows and UNIX operating systems, which will appear on Nature's website in the coming weeks.
For scientists, graphics are an essential tool for representing and communicating complex ideas. Through a combination of lines, points, symbols, coordinate systems, words, shades and colour, graphics can reveal the story within data like no other medium. Graphics can encourage our eyes to compare, contrast and condense information coming from single or multiple sources. As Edward R. Tufte said in his book The Visual Display of Quantitative Information1: "At their best, graphics are instruments for reasoning about quantitative information." Good graphics bring to light data patterns that clearly demonstrate, or reveal the limitations of, our summary statistics.
During the last decade, the almost ubiquitous use of Windows-based computer systems by scientists has encouraged the development of many easy-to-use and stylish Windows-based graphics packages. Many of these packages provide a very high level of quality graphical data display, as well as a flexible editing environment so that scientists can create, present or publish graphics that range from the stylishly simple to the creatively complex. In general, the Windows-based graphical software packages fall into four categories:
1) large, multiple-volume software packages with combined emphasis on statistics and graphics,
2) stand-alone graphics software that is part of a larger family of statistical software,
3) small, single- or double-volume software packages with combined statistical and graphical features, and
4) small software packages with graphics only.
Of the 16 graphical software packages reviewed here, four of them were large conglomerates of statistics and graphics (S-Plus, SPSS, STATA and Statistica), two were stand-alone packages belonging to larger families of statistical software (SigmaPlot and Sygraph), nine were small statistical and graphical packages (Excel, JMP, Minitab, Origin, Prism, ProStat, Psi-Plot, Statview and Unistat) and only one (Tecplot) was a stand-alone graphics package. Although the size of a software package is often related to its level of capability, this association does not necessarily hold true for graphics software. For the 16 software packages examined for this review, there was no straightfoward relationship between graphical capabilities and size of the software package. Some of the best graphics were contained in the smaller packages.
| Complete list of software packages reviewed in this article | |||
|
Excel
JMP Minitab Origin |
Prism
ProStat Psi-Plot S-Plus |
SPSS
Stata SigmaPlot Statistica |
StatView
Sygraph Tecplot Unistat |
To arrive at an overall rating for each package, I evaluated the graphical capabilities associated with the visual display of univariate, bivariate and multivariate statistical information. Specifically, this assessment involved a particular focus on
1) graphical displays that accompany standard description of univariate distributions, regression analysis, survival analysis, time-series, quality-control and factorial analyses,
2) graphical displays of statistical diagnostics underlying these analyses (for example, leverage or influence statistics),
3) plotting statistical distribution functions with data,
4) graphics associated with multivariate statistics and visualizations,
5) three-dimensional graphical capabilities,
6) the creative potential or imaginative qualities of the graphical features and
7) the customized editing features that allow the user to sculpt graphical displays into exact, refined displays of information.
In the table (42k), these graphical topic areas were assessed in each package using a five-category scale: excellent, very good, good, fair and poor. Because the display of most quantitative graphical information is currently linked with statistical analyses, one criterion used to judge these software packages was how well graphics was integrated with statistical procedures and diagnostics. Hence, in the review below, some packages were not rated for a particular topic area (for example, survival analysis) because they did not have an option for immediately displaying the results or diagnostics associated with the statistical procedure. Note this "not applicable" rating does not mean the software did not have that particular statistical procedure, only that it did not have immediate graphical display of that procedure's results. Hence, in the review below, some packages were not rated for a particular topic area (for example, survival analysis) because they did not have an option for immediately displaying the results or diagnostics associated with the statistical procedure or because that analysis was not available in the package. Note this "not applicable" rating means that either the software did not have that particular statistical procedure or that it did not have immediate graphical display of that procedure's results. [statistical capabilities for many of these packages will be evaluated in the complementary statistics review, which will be release in the coming weeks]
To briefly summarize, I found that Statistica provided the most impressive array of graphical options. It was obvious that this package was the best and in some ways may serve as a prototype for graphical software of the information age. However, there were still limitations to this software. The software designers must keep consulting with researchers to expand their graphical capabilities to include intelligible and detailed displays of the megabytes and gigabytes of data easily generated in many fields of science (for example, genetics). Other software that had very good graphical capabilities included JMP, Sigma Plot, S-Plus, Sygraph, TecPlot, and Unistat. Not surprisingly, each of these software packages fills a different user niche, as I have tried to indicate in the "Likely Users" sections of the reviews of these packages.
e-mail: slrk@supera.hg.med.umich.edu
References
2. Cleveland, W. S. The elements of graphing data.
(Wadsworth, Monterey, CA, USA, 1985).
