Scientists Have Many Choices For Math, Stat, And Graphing Software

SIDEBAR: Selected Suppliers of Math, Stat, And Graphing Software No scientific task is more universal than the job of analyzing data. Programs for this purpose were among the first applications written for mainframe computers, and the programs made the jump to microcomputers soon after the beginning of the personal-computer revolution in the early 1980s. Despite recent mergers among statistical software companies, scientists still have an embarrassment of riches when it comes time to choose software for analyzing and graphing data. There are literally dozens of programs out there, ranging from large, highly flexible packages to those focused on a narrow subfield. Choosing the right program may be getting progressively more difficult, however, since software vendors seem to release new versions of their programs annually, with continually evolving sets of features. "The market is extremely competitive," explains Santanu Bhattacharya, director of software marketing for Microcal Software Inc. of Northampton, Mass. Microcal sells Origin, a technical graphing and data-analysis package. "If you're not bringing out a version every year, you're lagging behind," Bhattacharya says. "You're adding more things, but then you don't want to make it so cluttered that it would intimidate a user who is just starting out. The effort is also spent in trying to make the software easier and easier." NUMERICAL SYSTEM: MATLAB from The Math Works produces sophisticated graphs easily "One of the nice things about [almost] all the packages now, which wasn't true two years ago, is that they're available on both Macintosh and Windows platforms," notes Robert J. Cruise, a professor of biostatistics at Loma Linda University in California. And hardware compatibility doesn't end there. The latest version of Mathematica by Wolfram Research Inc. of Champaign, Ill., for example, is available for no fewer than 20 different platforms. Another noteworthy development is a sort of convergent evolution between statistics software and graphing software. Statistics packages are continually improving their graphing capabilities, and graphing packages are continually adding more statistical functions. "There is a convergence there, and the convergence is a very natural one, because part of doing statistical analysis is doing things like exploratory graphing, understanding what your data looks like," explains John Foster, senior product manager for Jandel Scientific Software of San Rafael, Calif. "And after you've done your analysis, you want to do presentation graphing." Jandel, which was acquired this month by Chicago-based SPSS Inc., makes SigmaPlot, a highly regarded graphing program, and SigmaStat, a general-purpose statistics program. Although Jandel encourages both programs to be used as a package, SigmaStat by itself provides a wide selection of graph types, and SigmaPlot has a number of data-analysis tools, including the ability to generate summary statistics such as standard deviations and confidence intervals, a sophisticated mathematical transform language, and flexible curve-fitting routines. Microcal's Origin is another high-end graphing program that includes several tools for statistical analysis. "We feel we can capture the low end of the [statistics] market by providing statistical utilities in the product," Bhattacharya points out. "One example of that is a graduate student in biology who needs to do statistics but really does not need to become an expert in statistics." But it's not just grad students who use Origin. Bruce Bean, a professor of neurobiology at Harvard Medical School, notes that "in our work we don't use very heavy statistics. We use Origin primarily for plotting data, for analyzing data by reading peaks, and for curve fitting-for example, to determine exponential decays or more complex kinetic models. We're using it for fairly simple things, and it does those very conveniently." BUILT IN: 3-D graphing is part of the basic Axum program, not an add-on. Another popular graphing program is Axum from Seattle-based Math Soft. Axum excels at handling large, complex data sets, producing more than 80 publication-quality graph types. Unlike some graphing programs, Axum offers 3-D graphing as part of the basic program, not an optional add-on module. Additionally, Axum features a unique utility that allows the user to display a 3-D plot in several different orientations on the same page with a click of the mouse. The primary advantage that graphing programs give the scientist is complete control over all details of the final graph. Jandel's Foster explains, "While many of your statistics programs can offer you a bar chart, what SigmaPlot offers is the complete ability to edit that bar chart, or the ability to export the chart to a JPEG format to use on the Internet." SigmaPlot comes with an optional add-on called the Visual Display Graph Library. This software is based on The Visual Display of Quantitative Information (Cheshire, Conn., Graphics Press, 1983), a highly regarded book by Edward R. Tufte, a professor of political science, statistics, and computer science at Yale University. "It's a snoozer of a title," Foster says of Tufte's book, "but it was a worldwide bestseller. What he's trying to do is to show how to effectively present your data so it can be clearly and correctly interpreted. We have taken his techniques [and] his style guidelines, and used the unique flexibility of SigmaPlot to make these types of graphs." Purchasers of the library also receive a hardbound copy of Tufte's book. While some scientists will be well served by graphing programs with a few statistical functions, others require far more statistical power. At the other end of the scale are programs like SPSS, SYSTAT, and S-Plus; these general-purpose statistics programs allow scientists to perform just about every statistical test under the sun. S-Plus from MathSoft contains more than 1,650 statistical functions, and it's based on the S language, developed at AT&T's Bell Laboratories. MathSoft claims that S is "the only modern, object-oriented language created specifically for data exploration and analysis." This language allows the sophisticated user to interface easily with programs written in C and FORTRAN, and to pass data to the operating system and other application programs. Abacus Concepts Inc. GraphPad Software Inc. Jandel Scientific Software Math Soft The Math Works Inc. Microcal Software Inc. SPSS Inc. Stat-Ease Inc. Statistical Graphics Corp. StatSoft Inc. Wolfram Research Inc. SPSS Inc. distributes both SPSS and SYSTAT. Although both programs are full-featured, the company is not competing against itself, explains Joel York, SPSS's director of marketing for scientific products. "SYSTAT and SPSS are both comprehensive statistical offerings. They have lots and lots of statistics. Finding the right features, making sure the product has your features, isn't that hard with either SYSTAT or SPSS. But they do come in different flavors. SYSTAT has the very sophisticated graphics, whereas SPSS has the very sophisticated tabular interactivity." One of the oldest statistical programs available, SPSS was created as the "Statistical Package for the Social Sciences," and York acknowledges that the program targets social scientists and people using business applications as its primary market. He notes, however, that any scientist making heavy use of data in tabular form-survey results, for example-will find SPSS especially useful. But York thinks that most life scientists would prefer SYSTAT. "SYSTAT has some of the best statistical, analytical graphics on the market. In real time, I can change from a linear scale to a log scale and see the graph change instantly in front of me, as opposed to transforming the variable and replotting the graph. It's got these nice 3-D graphics that you can rotate and look at from different angles to explore your data visually." Biostatistician Cruise, who has had extensive experience with most of the statistical software now on the market, notes that the full-featured statistical packages may be overkill for many scientists. "Are you just going to do the basic analyses that would be done in 70 percent to 80 percent of all research?" he asks. "If so, then you don't necessarily have to get the expensive packages. You can get a smaller package and still be okay." TIME-SAVING TEMPLATES: With StatView from Abacus Concepts, the suer does not have to write complex macros Cruise is a fan of StatView from Abacus Concepts Inc. of Berkeley, Calif. "I think people buy it because of its ease of use in terms of its templates. You can sit down in front of a StatView menu and pretty much figure out what you need to do, so there's not a lot of real learning time." StatView's templates can be a time-saver for scientists who repeatedly perform similar analyses. A StatView template retains every detail of the analysis, from the formulation of the null hypothesis to the size of the tick marks on each graph. And making a template is as simple as saving a file; the program doesn't require the user to write complex macros. Once the template is saved, anyone can use it for a new analysis simply by specifying a data set and variables. Statistica from StatSoft Inc. of Tulsa, Okla., is another package that "aims to bridge the gap between something that's completely comprehensive but impossible to figure out and something that's easy to figure out but doesn't have any sort of powerful statistics," in the words of Dean Westervelt, a StatSoft technician. Statistica gives the scientist a very wide array of parametric and nonparametric statistical tests, as well as reliability and survival analysis, time-series analysis and forecasting, and cluster analysis. But the program also provides hundreds of different graph types. "Our graphics engine is pretty comprehensive, and it's also easy to customize," notes Westervelt. "Statistica is truly interactive in that you can graph off of a data set, and if the graph isn't what you wanted, you can go and change the data and it will be instantly reflected in the graph. Or you can double-click anywhere on the graph and customize its features. We offer not just your standard graphs, but also a wide variety of other graphs that might not be readily available in other packages, like contour plots or 3-D functions." Some statistical packages are specialized for particular areas of science. "It seems as if the big statistics programs are written for statisticians-people who do large studies all the time," observes Harvey J. Motulsky, president of San Diego-based GraphPad Software Inc., which distributes Prism. "But most statistics programs that I've seen are not that great for working scientists who don't do statistics much of the time. Scientists have more need for graphics and curve fitting than they do for basic statistics. And when they do statistics, they don't need so much depth. What they need is an easy way to get at it and a lot of explanation, because a lot of scientists don't have a background in statistics." TAILORED FOR SCIENTISTS: Prism from GraphPad combines scientific graphing, curve fitting, and statistics. To fulfill these needs, Motulsky, a pharmacologist, wrote the first precursor to Prism in 1980. Motulsky learned his programming skills at the side of the most famous name in software: He went to high school with Bill Gates, who later would found Microsoft Corp. Describing Prism, Motulsky says, "I used to be a basic scientist, so I have a pretty good sense of what basic biologists need most of the time. We limit the choice to what I think a scientist is going to need. For many biologists, fitting curves is the most frequent statistical thing they do. Prism is designed to make curve-fitting really, really simple. We have built into Prism 15 standard equations that probably work 95 percent of the time. So you have binding curves, exponential curves for kinetics, all the standard curves built in. If you want to be more fancy, you can customize the equations." "Prism was the first package that seemed to be directed toward biological scientists," notes Andrew A. Young, vice president of physiology at Amylin Pharmaceuticals Inc. in San Diego. "The number of different graphic shapes that are available in Prism is somewhat restricted, but nonetheless they cover the range of the most commonly used and most commonly published graphic types. Anyone who's actually writing manuscripts to publish in scientific journals is well served by Prism." Young also likes Prism for another reason: "One of the biggest problems in keeping yourself organized is keeping the stuff together-keeping the graphs together, keeping the data together, keeping the analyses of the data together. And Prism is particularly useful for doing this. All of those things are kept together in one big file." While Prism is tailored for life scientists, some statistical packages target more narrow specialties. Design-Ease and Design-Expert, programs offered by Minneapolis-based Stat-Ease Inc., are "geared specifically to people who are trying to improve products or processes to get better yields, to get better quality, or to get better performance," according to Mark J. Anderson, vice president and general manager of Stat-Ease. Pharmaceutical companies frequently make use of these programs to optimize their formulations. ARRAY OF TOOLS: Origin from Microcal Software provides a variety of data-analysis features. With the programs' interactive series of menus, "we lead people through kind of a wizard-like process of setting up experimental design," explains Anderson. "They end up with a recipe sheet that tells them exactly what experiments to perform. And then when they perform the experiments and enter the data, it leads them through a distinct path that makes it simple to get the proper graphs and proper statistics needed to determine whether anything they did made a significant difference." Wizards, expert systems, and extensive help screens are increasingly becoming part of many stat packages. STATGRAPHICS Plus from the Statistical Graphics Corp. of Princeton, N.J., for example, allows the scientist to call on its StatAdvisor for help in interpreting the results of any statistical table or graph. Jandel's Foster points out that "underneath SigmaStat is an artificial intelligence engine. [It] allows you to use the program in two ways. You can either know the particular statistical test you want to run, or you might have the Advisor [a wizard feature within SigmaStat] advise you on a particular test. And all the time you're either using that test or being advised on that test, SigmaStat is looking at things like messy data, and checking for normality and equal variance-basically the fundamental underlying assumptions used in these statistical tests." Prism takes a different approach, providing written help files that scientists can access wherever they need assistance, instead of having the program itself insist on aiding the user. Explains Motulsky, "You can click on 'help' or 'interpret' at various places, and it's not just how-to-use-the-program help, it's really how-to-think-about-the-problem help, explaining how to choose the right test or, once you've chosen it, how to interpret the results." INTERACTIVE: Statistica from StatSoft Inc. offers a comprehensive graphics engine that's easy to customize. For power users-scientists who have the need for mathematical tools well beyond mere statistics-there are at least two particularly impressive software packages: Mathematica from Wolfram Research and MATLAB from The Math Works Inc. of Natick, Mass. Although these two programs overlap in significant respects, some researchers use both. One of them is George Oster, a professor of molecular and cell biology at the University of California, Berkeley. His group is developing mathematical models of molecular motors such as the protein kinesin, which "walks" vesicles up and down microtubules. "For us, Mathematica and MATLAB are complementary tools," comments Oster. "They each fill a different niche. MATLAB is a completely numerical system. It doesn't do anything analytically. We use MATLAB to get numerical estimates on things that cannot be computed analytically. And we use Mathematica to compute analytical answers that you couldn't do by hand. "Both of them do terrific graphics," Oster continues. "One of their great advantages over standard programming languages is the ease by which you can produce very sophisticated graphical presentations without any programming." And both programs contain extensive statistical functions. "We could use a specialized statistics program," notes Oster. "There are a lot of them out there that are quite good. But my philosophy is that it takes a lot of time to really get proficient at a complicated package. We're better off dealing with one or two packages that do everything. Even though they might not be the best at anything, you don't have to waste a lot of time learning more packages." And another advantage of Mathematica and MATLAB, says Oster, is that each has developed into a lingua franca. "You can count on almost any other scientist anywhere in the world having available Mathematica or MATLAB. So you can do a graph or do a statistical analysis in either of these programs, and then you can send the [file] to anyone and they'll be able to look at it." While some researchers might fear that a program as complex as Mathematica might be difficult to learn, it isn't, according to Paul R. Wellin, academic and business liaison at Wolfram Research. "I've used it to teach undergraduate students, and I've even given some workshops for high-school students, and they were able to pick it up fairly quickly. The learning curve is kind of interesting. There's an initial gradual curve, and then it steepens after you get past a certain basic amount of knowledge. But to do some basic data analysis-do some fits, some interpolations, and plot certain things-this is something you can pick up in a very, very short amount of time." With all the statistical, graphing, and mathematical software out there, how is a scientist to choose? Bruce Bean ended up with Origin after trying many other programs. "About four years ago, I got pretty much every [graphing] program that's available for Windows," he recalls. "I just liked Origin a lot in the first half-hour or so [that] I used it. I found it extremely easy to use, intuitive, very powerful, yet very simple to learn without spending a lot of time with the manual." Most of the software manufacturers mentioned in this article provide free demo versions of their programs, giving scientists the opportunity to compare features along with more qualitative aspects of the programs, such as how intuitive their interfaces are. With so many packages out there, there's no reason to settle for software that doesn't perform the exact tests or print the exact graphs that are required. And, concludes York, "In the end the best way [to choose software] is a recommendation from one of your colleagues." Robert Finn, a freelance science writer based in Long Beach, Calif., is online at finn@nasw.org.

Scientists Have Many Choices For Math, Stat, And Graphing Software

Interested in reading more?

Become a Member of