|
 |
|
|
|
|
 |
|
|
|
SIGMASTAT RESOURCES |
|
|
|
|
|
|
|
SYSTAT : SigmaPLOT : SIGMASCAN : TableCurve2D : TableCurve3D : PeakFIT : AutoSignal |
SigmaStat® 3.1 - Advisory Statistics for Scientists
By John A. Wass, Ph.D.
Introduction
SigmaStat is an old friend that this reviewer had originally used over 10 years ago when first learning about statistical software. At that time, SigmaStat was making the transition from DOS to Windows-based and even the DOS version seemed a significant improvement over the other programs in use at the time. The major reason for this was the corporate (and this software has passed through several corporate hands since) philosophy of making it simple with useful tools and many explanations, both in the manual and within the program. The ideal program for the statistical novice! In the ensuing years the program has been eclipsed by more "glitzy" software that uses some of SigmaStat's excellent tools and adds many statistical and graphics routines, and the seeming marketing emphasis on its sister program, SigmaPlot. Regardless, this beginners statistical program is still simple to use and includes the original tools (they pioneered the use of Wizards) that do much to recommend it to those of modest statistical needs. The companion program SigmaPlot, which will be reviewed in an upcoming issue, may be used to extend the graphics capabilities, and the two integrate so that all of SigmaStat's functions are available through with its graphics companion, assuming that both programs were purchased.
Installation and Manual
The software is made for Windows 98, 2000, NT 4.0 or XP, and this review was done on Windows Me without any problems. The company recommends at least a Pentium with 200 MHz clock, 64MB of RAM, 48MB hard disc space and a CD-ROM drive. These minimal requirements will run the program with very little speed decrement.
The single 848-page manual is still a great addition with clear exposition, much repetition, and mercifully large type. It explains both the menu driven sequences as well as the statistical theory in easy to read bites that are heavily cross-referenced. In addition, the index is actually of some value as it cleared up several problems with features that were difficult to find with the on-line help. Searching the help menu's that come with most programs is always an adventure, and unfortunately no different here, as certain subjects are well referenced but many important ones are not. For example, finding out how to get regression graphics with confidence and prediction intervals was easy with the manual, but took forever with the search function. Unlike internet searches, the user is limited to single words (you can type in more but this may be more of a hindrance than help). This is just a problem with the search function and not with the information provided, which is as substantial as that in the manual. It was rather surprising to see several indications of sloppy editing however, including reproductions of screens that were from SigmaPlot rather than SigmaStat, page references in the chapter-page format when the manual no longer uses that format, and contradictory figure descriptions.
The manual now contains one of the best reference sections that a novice statistical user could hope for. It includes the excellent introductory tests by Glantz (1) and Zar (2), as well as the more advanced 'Messy Data' text by Milliken and Johnson (3).
Notebook and Manager
 |
| Figure 1 - To View Larger Image |
number of multiple worksheets or other windows quickly. If working with a single spreadsheet, but having several analysis and graphics windows open simultaneously, the use of the tile option under the Windows menu item will display all of them at once. This is useful for examining data trends and aberrancies. Notes on each project may be instantly displayed and hidden using the 'Show summary information bar' and the navigator window hidden or recalled at the press of an icon.
There have been several improvements to the spreadsheet to make it more like EXCEL in terms of functions. These include multiple undo, more flexible column headers and row titles, freeze panes, print preview, and cell height resizing. Importing data from EXCEL is automated and the column headers come in with the data. Plain text and comma-delimited files are also imported as are SPSS and SYSTAT files and several other formats.
Statistics
This is the strongest feature and that which will appeal to the novice or non-statistician. At the core for any beginner is the Advisor Wizard that will guide them, step-by-step, through an analyses by asking leading questions. It will query the user from the most general (description versus comparison) to the more specific (data on numbers or ranks, two or more than two groups, etc) and finally recommend a test. The list of tests and routines are displayed in Table 1.
Table 1. Statistical Tests
| Descriptive Statistics t-test Rank Sum test One-, Two-, and Three-way ANOVA ANOVA on Ranks Paired t-test Signed Rank test One- and Two-way Repeated Measures ANOVA Repeated Measures ANOVA on Ranks Chi-Square Linear Regression Multiple Logistic Regression Multiple Linear Regression Polynomial Regression Stepwise Regression Best Subsets Regression Nonlinear Regression |
Pearson Product Moment Correlation Spearman Rank Order Correlation Survival - Single Group Survival - LogRank Survival - Gehan-Breslow Normality test - Kolmogorov-Smirnov Power and Sample size for t-test Paired t-test Proportions ANOVA Chi-square Correlation Fishers Exact test McNemar's test z-test |
In addition, there are a number of transformations available from a pull-down menu that will allow quick transforms of data columns as well as user defined functions to expand the type of transform. Many already defined functions are built-in and readily available through the user-defined box. From a perusal of Table 1, it is apparent that the developers have a limited, but very useful menu. They did do a good job of selecting all of the most frequently used data exploration tests and descriptive statistics, plus many of the more advanced tests that would be useful to a variety of disciplines. There are also embedded helps such as dialog boxes when the user attempts questionable actions, such as a paired t-test on unbalanced data, or a dash in the output column for the log of negative numbers. Now the minor gripes.
It would be very helpful to add the coefficient of variation to any descriptive statistics routines. This is easily calculated but is heavily used in any cursory data inspection and should automatically appear in any summary statistics. Also, SigmaStat is one of the few programs that will operate on data as it is usually collected, i.e., in columns with column identifiers. Many statistical programs require stacked data, especially for the t-tests and ANOVA's. This requires transposition of the data and the addition of an identifier column. Thus SigmaStat holds the advantage here for already formatted data. However, data needs to be sub-setted, stacked, transposed, joined, and sorted for other routines. It would be very helpful if SigmaStat could do those chores. Next would come the addition of more important choices, e.g., having a normality test is necessary but, by restricting ourselves to the KS test, we bias small data sets to normality as well as exclude some of the more useful aspects of the Shapiro-Wilks test as well as several others. The section of the Transforms pull-down menu labeled Missing Values is misleading as it merely offers substitute symbols for the blanks and does not actually impute any values. Quick Transforms merely add, subtract, and divide one column by another. It would be nice if the user had a dialog box that would allow these functions using a pre-defined constant rather than resorting to a user-defined programming box. Something like this is done with the random number generating function, where the user merely fills in the boxes and specifies the range and how many numbers are to be generated, a very valuable device.
The user-defined box will take some practice as the language is neither straightforward nor covered in the manual. Other areas, such as the very useful best subsets regression could use more amateur coaching as choosing a set based upon correlation alone may be misleading. Lastly, a few other tests that expand the general linear models (say to mixed models) would be useful
Graphics
The software does an adequate job of the quick and dirty graphics that will assist the researcher in visualizing the data and spotting trends and aberrancies. However, to quickly generate publication quality graphics as well as access advanced editing features, the user needs to purchase the sister program SigmaPlot. Still, there is a fine choice of simple graphics from scatter, line and bar graphs to pie charts, histograms, 3D scatter plots, box plots, residual plots and probability plots. The editing features include line/symbol color and type, axis scale (seven choices), and fill color.Most are easy to create and may be viewed with the data by tiling.
Summary
SigmaStat has matured slightly from the earlier versions but for what it does, it does so well that the user is always looking for upgrades. This general statistical software does not come with the usual modules that do experimental design, quality assurance, or the variety of specialized functions that are in heavy use with the larger datasets, but is superb for the novice statistician with modest needs and purse. It is therefore useful for the scientists, engineers, and students who have a need for first-pass statistics and graphics to better understand what their data is telling them and perhaps suggest further experimentation. As such, it is hoped that the company will more heavily market this fine addition to the statistical support available to the non-statistician.
References
1) Glantz SA. Primer of Biostatistics (3 ed). New York: McGraw-Hill, 1992.
2) Zar JH. Biostatistical Analysis. (2 ed). Englewood Cliffs NJ: Prentice-Hall, 1984.
3) Millikan GA, DE Johnson. Analysis of Messy Data. Volume 1: Designed Experiments. New York: van Nostrand Reinhold, 1984.
 |
 |