Number Crunching

Click to view the PDF file: Suppliers of Statistical Analysis SoftwareCourtesy of Stata Corp.New users of Stata Corp.'s Stata software can obtain training via NetCourses."There are three kinds of lies: lies, damned lies, and statistics," or so said Benjamin Disraeli. Fortunately, the field of statistics has come a long way in the last 150 years. With careful scrutiny, lies (and even damned lies) can often be recognized as the inappropriate use of statistics. Researchers must use statistical tests to evaluate how well experimental data supports a given hypothesis. In addition, modern experiments and studies frequently produce so much data that the only way to see the forest for the trees is with summary statistics. Indeed, most scientific journals will reject manuscripts for misuse or lack of statistical analyses.The range of approaches to scientific questions is as varied as the data and methods of analysis. Data can be analyzed across published studies (meta-analyses); extracted from large, diverse, databases (data mining); collected from surveys; and generated from explicitly designed experiments. Many statistical software products are available from outside the field of scientific experimentation, from subjects like finance, medicine, meteorology, political science, sociology, and even sports. If the design is appropriate, software from these distant fields may be applicable to scientific situations.Here, The Scientist focuses on software available for the researcher designing experiments and analyzing, interpreting, and presenting data. As the number of statistical software packages available is probably in the thousands, an exhaustive review is not feasible. Instead, this article hones in on some of the major commercial players selling a broad array of statistical tools likely to be useful to laboratory scientists. Comparisons among packages are made only in terms of tools available and general information for the potential user, with an emphasis on unique features. More rigorous performance comparisons, of interest to statisticians, can be found in software guides published in various journals.1 Two universal points of concern with any software product are input data format and output requirements. Fortunately, these factors are becoming less critical as the newer software packages accept a wider range of input formats and include graphical output options, often at publication quality. Input limitations can often be circumvented with imaginative manipulations of data files or via transfer programs. There is an obvious, but not strict, trade-off between ease-of-use and flexibility. The low-priced, easy-to-use applications usually come with standard procedures. Some of these applications are limited in terms of input format, size of data matrix, and platform support. The larger and more versatile products often have international support in different languages, are extremely flexible, and run on several platforms. Because these software products can often be purchased with or without a number of options (platform, number of users, and modules) it is difficult to compare pricing directly. Nevertheless, it is generally true that the larger and more flexible products are also more expensive.Both classes of software have their respective roles. Statistical novices will appreciate the user-friendliness of the "low-end" products, whereas harnessing the power of a "high-end" package may require taking some coursework, often available through the support company. Many labs may wish to use different products in different situations--for example, a quick-and-dirty computation of error bars for a transfection assay versus large-scale experiments guided by a statistician.Take It EasySome products are designed for users who don't use statistical programs frequently or who don't have a firm grasp of statistical concepts. For example, GraphPad InStat and GraphPad Prism from GraphPad Software Inc. of San Diego are designed to "make statistics simple to the experimentalist," according to company president Harvey Motulsky. GraphPad Prism emphasizes scientific graphing and curve-fitting but also performs basic statistics. GraphPad InStat performs basic statistical analyses and helps the user choose a test to make sense of data. Both programs offer step-by-step guidance through analyses, and are thus much easier to master than larger programs. However, GraphPad InStat does not perform as many tests as the larger packages and does not offer as many ways to manipulate and subdivide data. Input file size is limited to 1000 rows (Windows version) and 26 columns, but this is unlikely to constrain even a medium-scale experiment. Both products run on Macintosh and Windows operating systems.Researchers looking for a simple analysis package with a Microsoft Office interface might try ESBStats from ESB Consultancy of Kalgoorlie-Boulder, Western Australia. Statistix for Windows from Sigma Research Associates of San Diego is also very easy to learn, with excellent data handling, data management, graphing tools, and Microsoft Office compatibility.Those looking for a product to fit a very tight budget might consider Modstat from Modern Microcomputers of Bellingham, Wash., developed for teaching and occasional users. Although data matrix size and format is limited, more than 300 statistical tests and routines are featured in the package. Modstat also includes suggestions for types of analyses and validity of tests, without the need to refer to a manual.Comprehensive Statistical ToolsSeveral large packages are geared toward the statistical expert. These are often more flexible, and some can handle larger and more complex data than their smaller brethren can. Familiarity (and effective instruction) can make even the most cumbersome software user-friendly, but the general trend is toward ease-of-use across the spectrum. For teaching advanced courses in which statistical analysis is a major component, these products are valuable, but they are equally applicable to large experiments in academia and industry.One of the most widely used packages in academia is SAS from SAS Institute Inc. of Cary, N.C., which offers much more than its acronym (Statistical Analysis System) implies. SAS is now a family of products for data mining, visualization, geographic information systems, and applications development, among other resources, which are organized into a large array of modular packages that can run on most operating systems. SAS/STAT and STAT/QC are the products of greatest interest to the experimental scientist. The former is one of the company's flagship products and includes most standard parametric and nonparametric approaches. The software employs an open format, allowing for a modeling approach to analysis of variance (ANOVA) and other analyses. The programming-like framework to preparing run files, especially on the earlier versions, enables almost any possible input data format, and the versatility of the software makes it a good choice for teaching statistical analysis. STAT/QC offers many approaches to product design, usually in an R&D setting, but is also applicable in some experimental settings. For a versatile graphics interface SAS/GRAPH is one option that is compatible with SAS/STAT. But the SAS Institute now also supports the simple but user-friendly StatView software, as well as JMP, a useful package for data and trend exploration with a very effective link between data, statistical analyses, and graphics.Courtesy of SPSS Inc.Screenshot of SigmaPlot 2001SPSS Inc. of Chicago is a relatively old and established company, founded in 1968. Like its peers, SPSS has evolved from a single statistical package to a large family of software tools, some acquired from external sources and developed further. Most of the latest versions have only Windows support, but the array of analyses and visualization tools is staggering. The basic statistical package is SPSS Base, which offers several add-on modules. Additional tools for data entry, database management, education, and data mining are also available. Other statistical packages that are now supported by SPSS include SYSTAT and SigmaStat as well as the graphical applications SigmaPlot and Graph Express. Additional SigmaPlot modules are available for enzyme kinetic analysis and electrophysiology, as well as SigmaGel for analysis of gel images. All the tools are comprehensive in scope and easy to use. The efficient umbrella control of SPSS ensures compatibility with almost any data management system.One of the favorites for teaching is MINITAB from Minitab Inc. of State College, Pa., which began exclusively as an educational tool. Although MINITAB has expanded to the scope of its competitors, it has retained its pedagogic framework, and many students trained on the product retain it for later use. The majority of Minitab's customers, however, are commercial or quality control organizations. MINITAB Release 13 includes basic and advanced statistics, regression and ANOVA, statistical process control (SPC), design of experiments (DOE), reliability analysis, power and sample size, time series and forecasting, and measurement systems analysis (Gage R&R).STATA from Stata Corp. of College Station, Texas, has the ability to add custom commands. The program combines a wide array of analyses, can handle various data frameworks, including panel data and survey data, and has good graphing capabilities and easy-to-use functions. STATA also includes tools for data management and programming, enhancing the versatility of the software. NetCourses are available for STATA training, and user meetings are held regularly.Courtesy of Insightful Corp.Data analysis performed using Insightful Corp.'s S-PLUS.Another product with options for user-defined functions is S-PLUS from Insightful Corp. of Seattle. The package is based on the statistical analysis language S, originally developed for interactive exploration, analysis, and visualization of data. The software has the expected array of standard analyses but also excellent graphics functions, vector and matrix computations (and other mathematical capabilities), as well as tools for data exploration and data mining. The programming capabilities may appeal to industry, but S-PLUS also succeeds in an academic research setting, especially in groups with well-trained statisticians.Courtesy of StatSoft Inc.Screenshot illustrating customizability of StatSoft's STATISTICAThe difference between simple and relatively inflexible packages and complex and versatile ones is not necessarily clear-cut. For example, STATISTICA from StatSoft Inc. of Tulsa, Ok., is a very flexible and widely- used application that could appeal to both experts and beginners. The entire approach to the software is scalable, and the quality of output graphics is excellent. For some of the traditional analyses, such as ANOVA, the software includes a design wizard that can assist with appropriate model choice, in association with experimental design. STATISTICA is a family of products that includes the stand-alone BASE version, an advanced application for linear and nonlinear modeling, and packages for quality control analysis, multivariate analysis, experimental design, neural networks, power analysis, and data mining. A QUICK version is also available for Macintosh for more streamlined needs.Unfortunately, some studies cannot use standard statistical approaches, so new programs must be written to accommodate them. If it does becomes necessary to code new software for a specific application, a selection of over 2,000 statistical and mathematical routines available from Numerical Algorithms Group (NAG) of Downers Grove, Ill., will prove helpful. NAG also offers a stand-alone product for professional statisticians and scientists called Genstat.Mplus from Muthén & Muthén of Los Angeles can handle a powerful array of models and modeling approaches. This is particularly useful for the design of complex statistical models, such as latent variable mixture modeling. Mplus' approach is also valuable when the researcher is looking for complex patterns in data, searching for suites of indicators for such characteristics as disease susceptibility, or analyzing the data from randomized trials.Several products are available specifically to aid experimental design, including the Gendex DOE Toolkit from Design Computing of Clayton, Victoria, Australia. For research synthesis across published studies (meta-analyses), Comprehensive Meta-Analysis from Biostat Inc. of Englewood, N.J., includes a good database framework for archiving abstracts and other peripheral information. An alternative is Epi Meta (www.cdc.gov/epo/dpram/epimeta/epimeta.htm), distributed freely by the Centers for Disease Control.Rows or Columns?The format of input data involves not only the simple column-row orientation of a two-dimensional matrix, but also includes software-specific code and analysis-dependent instructions. Moreover, applications vary in their abilities to handle missing data. Whereas some of the more versatile applications will accept a user-defined format or can import from other applications, the lower-end packages may be more restrictive. Format transfer software can help users who are unable to translate data with simple cutting and pasting via a text editor. Several programs are available, including DBMS/Copy from Conceptual Software of Houston, Texas, which can translate and modify data among more than 90 different database management, spreadsheet, and analysis applications. This is ideal for a large research lab full of researchers with different software preferences. Stat/Transfer from Circle Systems Inc. of Seattle, for Windows and Unix, handles transfers between most of the popular statistics and database applications at a very reasonable price.For many spreadsheet users and analysts, a spreadsheet (or a database queried from a spreadsheet) is the preferred interface for modeling and statistical analysis. For these users, there are several low cost add-ins for Microsoft Excel, including modeling, multivariate methods, and nonparametric statistics from NAG. Other add-ins are available, including XLSTAT from AddinSoft of Paris, and Analyse-it from Analyse-it Software of Leeds, United Kingdom.Today's laboratory scientists depend on statistical software. But effectiveness of the software is only as good as the experimental design and quality of the data. An understanding of the theoretical framework of statistical analysis is therefore more than a little useful. To this end, reference books are a valuable guide. The most dog-eared books on this author's shelf are R. R. Sokal and F. J. Rohlf's Biometry;2 G. E. P. Box, W. G. Hunter, and J. S. Hunter's Statistics for Experimenters;3 B. F. J. Manly's The Design and Analysis of Research Studies;4 and B. S. Weir's Genetic Data Analysis II.5 Many others are available. A very useful, albeit simple, text for the lab is Basic Statistics for Laboratories by W. D. Kelley, T. A. Ratliff, Jr., and C. Nenadic.6 The next best thing to a good statistics book is a friendly and competent statistician. When initiating experiments it is best to consult statistical advice at the design stage rather than at the end of the data collection stage, when it is perhaps too late to alter design criteria. The combination of a sound experiment with appropriate statistical analysis can go a long way towards combating the oft-cited "lies" of statistics. Thus, current software solutions play a major role--revelation of the truth.Paul Wolf (wolf@biology.usu.edu) is an associate professor of biology at Utah State University.References1. A. Koch, U. Haag, "The statistical software guide '96/97," Computational Statistics and Data Analysis, 24:107-31, 1997.2. R.R. Sokal, F.J. Rohlf, Biometry, 3rd Ed., New York, W.H. Freeman and Co., 1995.3. G.E.P. Box et al., Statistics for Experimenters, New York, John Wiley and Sons, 1978.4. B.F.J. Manly, The Design and Analysis of Research Studies, Cambridge, U.K., Cambridge University Press, 1992.5. B.S. Weir, Genetic Data Analysis II, Sunderland, Mass., Sinauer Associates Inc., 1996.6. W.D. Kelley, et al., Basic Statistics for Laboratories, New York, Van Norstrand Reinhold, 1991.

Number Crunching

Interested in reading more?

Become a Member of