SigmaStat 3.0 was briefly reviewed in the November 2003 issue of The American Statistician (Hilbe 2003). The review dealt with packages which, at the time, were sold under the banner of SPSS. SigmaStat has since been licensed for sale by SYSTAT, and has had a new release. Hence my interest in taking another look at the product.

First of all, SigmaStat has not been designed with the professional statistician in mind. Rather, the product is meant for researchers who have some statistical experience, but who perhaps need more guidance in which test to use for a particular set of data. Moreover, SigmaStat output contains—unless the user specifically blocks it—a summary interpretation of the results, expressed in a manner that may be appropriate for inclusion in the statistical methods section of a clinical journal article. For instance, a short paragraph appears at the end of statistical output specifying the p value, together with how it is to be interpreted with respect to the particular test. I showhowthis is done in some of the examples used in this review. This feature of the software may appear extremely valuable to the novice researcher; but as most TAS readers know, pat interpretations are susceptible to error. If one checks the assumptions upon which the test is based, and can be assured that the interpretation given is appropriate, then thewording given in the output can more safely be included in a manuscript. Unfortunately, however, many users of the software may be apt to overlook these assumptions and simply paste the interpretation into their manuscript without serious thought. It is also possible that an entirely wrong test has been given to the data at hand. Aside from the above caveat, the ancillary interpretation is a nice feature of the package.

SigmaStat comes with an 848-page reference manual. For each SigmaStat capability, the manual presents a near textbook discussion of the theory underlying each procedure or test, together with a thoroughly worked out example. In fact, for someone having to learn basic statistics on their own, SigmaStat may be the perfect package.

The statistical capabilities of SigmaStat 3.1 include the following: basic descriptive statistics; t-test; one-, two-, and threeway ANOVA; one- and two-way repeated measures ANOVA; contingency table analysis; nonparametric analysis; Kruskal- Wallis ANOVA on ranks; Friedman repeated measures ANOVA on ranks; multivariate linear regression; polynomial regression and logistic regression; nonlinear regression; sign rank tests; Pearson and Spearman correlation analysis; logrank survival analysis; three types of Kaplan-Meier survival analysis; Gehan- Breslow survival analysis; power and sample size computation. Several other capabilities exist as well.

SigmaStat comes with a host of appropriate graphs, and allows the user to engage in basic data transformation and coding. It appears to be a nice all-around basic package for biostatistical analysis.

I believe that the reader may perhaps obtain a better idea of SigmaStat’s feel by actually viewing results of statistical procedures. I have decided on three commonly used procedures: Fishers exact test on a 2 × 2 table, a two-way ANOVA, and a logistic regression. Comparisons will be made to the output of other selected packages.

A potential problem area rests with getting data into SigmaStat. The software will directly load files formatted in SigmaStat, SigmaPlot, Lotus, Excel, and several other packages, including plain text files (comma delimited and free format). There is no problem with this. However, if you wish to paste data into the editor from that of another package, the variable name is placed on the first line. One must cut the data, column by column, and paste it in starting with the first row. This is rather time-consuming, and should be corrected in a future release. But for those who simply import data directly—for example, from Excel—there is no problem.

Finally, I have heard rumors that SigmaStat is sometimes sloppy with numeric details, but I did not identify any problem areas. Now, I may have missed them, or they might not be there any longer. I, however, did not identify any serious problems.

Fisher’s exact test is more accurate than asymptotic methods when determining the p value of a K × R table, assuming the independence of the rows and columns. SigmaStat limits the use of Fisher’s exact test to 2 × 2 tables, and furthermore restricts cell count size to be no more than five. I tried to calculate the test using the example for Fishers Exact Test as found on the onscreen help. Because one of its cells was seven, the test could not be performed. I changed the table to read as

4 1
2 3

and the test worked as described. I might add that cell counts of less than five for such tables are inappropriate for use with asymptotic tests; for example, the standard Pearson ÷2 test. So SigmaStat’s inclusion of the exact test for tables having one or more cells less than five is very useful for users who have such a design. Available asymptotic tests can be given for tables having all cell values greater than five, as well as for those tables having dependence between rows and columns; for example, repeated treatment tests.

SigmaStat’s output for the test is shown in Figure 1(a). Notice the statement of interpretation at the bottom of the output. For comparison sake, Figures 1(b) and 1(c) show the output for the same test using StatXact and Stata.

Figure 1. (a) SigmaStat’s output for Fisher’s Exact Test.

Figure 1. (b) StatXact output.

Figure 1. (c) Stata’s output..

SigmaStat provides the correct answer, and even includes expected cell values, which are required by the calculational algorithm. Neither StatXact nor Stata provide these values, but Stata adds a variety of related statistics. SigmaStat provides only a two-tailed test, but a one-sided p value can easily be calculated by dividing the two-tailed value in half.

2. TWO-WAY ANOVA

As an example I shall use a constructed clinical trial that comes with the package. The data consist of 12 observations over three variables. The dependent variable is drug response with continuous values ranging from 1.5 to 6.6. Factors include gender (Male; Female), and drug (A = treatment; B = Placebo). With the data already in the editor, one selects “Statistics” from the top selection bar, and subsequently selects “ManyWay Comparison” and “2Way ANOVA.” It is a simple task to run the model. Figure 2(a) shows the main ANOVA output without diagnostics. SigmaStat has a wide variety of ANOVA diagnostic tools that assist the user with the interpretation of the model. Figure 2(b) displays Stata output for the same model.

Figure 2. (a) SigmaStat output for two-way ANOVA.

Figure 2. (b) Stata output for the same model.

Figure 3. SigmaStat output for logistic regression.

Figure 4. Stata’s logistic capability. Top displays the coefficients and bottom shows the odds ratios.