Test Platform

Pentium 166MHz, running Windows 95 with 16 Mb RAM

Overview

GraphPad InStat 3, a unique statistics program from GraphPad Software, Inc., directs biomedical researchers through the appropriate analysis of their data. It is the perfect solution for descriptive statistics, as well as for analysis of contingency tables, linear regression and correlation, parametric and nonparametric tests for comparing control and experimental groups, and multiple regression. InStat provides appropriate, data-dependent choices for statistical tests and posttests. InStat queries users about the structure of their data, displays the proper input screens for entry and editing, and then implements the statistical tests relevant to the data set. Test results are easy to understand and include the question that is posed by the analysis, as well as unambiguous answers to the question. InStat fills a need in the statistical software market, providing biomedical researchers with easy-to-use software that effectively implements most of the desired statistical tests.

Program Features

InStat has an effective approach to data analysis. It requires that one know something about the structure and characteristics of the data to be analyzed, then directs one through the relevant statistical tests. Figure 1 displays the first screen encountered. Users must specify the analysis goals by clicking on radio buttons, which determine the data entry format choices as well as the tests that InStat will perform. For example, if one chooses to compare means, one can enter raw data, or the means, error estimates, and Ns for the data to be analyzed (figure 2). InStat will perform tests that are specific for the analysis goal and data format selected (figure 3).

InStat is ideal for researchers analyzing small sets of data. Raw data can be entered on worksheets (figure 4), imported as text files, or copied/pasted from other programs. If one chooses to analyze means and standard deviations, InStat changes the format of the data entry screen to match this choice. One can also choose to transform data, or combine variables from the program's Data menu. Some of the functions available for transforming data include log, ln, exp, sqrt, square, LOGIT, and trigonometric functions (figure 5). In addition, data can be selected and excluded from the analysis from the Data menu. Although InStat offers some data transformation features, it may be more efficient to preprocess data in another program such as Excel, which allows automation of some of the manipulations through the use of macros.

InStat's toolbar (figure 6) and navigation (figure 7) buttons, as well as useful links to context-sensitive help topics, facilitate navigation through data entry and analysis. One accesses appropriate InStat screens in a stepwise manner by clicking on the right or left arrows on the navigation bar. View any of these screens in any order by clicking on the appropriate button (figure 7).

Figure 6

Figure 7

In addition, each screen includes two buttons that link to relevant information in the online help file. For example, from the worksheet, one can display help topics that cover arranging and importing data, while the column statistics page includes links to explain results and on entering means and standard deviations.

Data Analysis Features

InStat is a guide to correct data analysis. Clearly, no software program can automatically make all of the right decisions without user intervention, but InStat does an excellent job of presenting only the relevant choices. As long as one tells InStat something about the structure and format of the data, and some of the assumptions one wishes to make, it will select the right tests for analyzing the data set. InStat implements the right statistical test for paired or unpaired data, from Gaussian (parametric) or non-Gaussian (nonparametric) distributions, using the right posttests to compare pairs of group means when ANOVA is used.

For example, when two groups are compared, InStat automatically directs one to utilize paired or unpaired t-tests with one- or two-tailed P values. One is asked whether data values are sampled from a Gaussian distribution and whether it is assumed that the populations have the same or different standard deviations. InStat will select the appropriate analysis method based on those choices. Wilcoxon (paired) and Mann-Whitney (unpaired) nonparametric tests are automatically selected if one expects non-Gaussian distributions. Parametric tests are used if one assumes Gaussian distributions, and a Welch correction to the t-test implemented for populations with different standard deviations.

InStat changes the choices available based on the structure of the data. If three or more columns of data are being analyzed, InStat correctly directs one to use ANOVA or repeated measures ANOVA for paired or unpaired data respectively (figure 8). For nonparametric methods, InStat will use the Kruskal-Wallis test for unpaired, and Friedman test for paired, data. Posttests can be used to compare pairs of group means. The Dunn posttest is used for nonparametric methods, while one can choose from Bonferroni, Tukey, Student-Newman-Keuls, and Dunnet tests for parametric ANOVA methods (figure 9).

In addition to comparing means, InStat also includes tools for regression analysis and correlation as well as analysis of contingency tables. Linear and constrained linear regression can be implemented, as well as Pearson (parametric) and Spearman (nonparametric) correlations. InStat will even automatically estimate unknowns from standard curves. Analysis of contingency tables can be performed using Fisher's exact test, the chi-square test, or the chi-square test with Yate's correction.

Although InStat does an excellent job of conducting these basic analysis tasks, it does not handle advanced statistical tests. For example, it lacks features to analyze survival curves, and to perform two-way ANOVA, factor or cluster analysis, polynomial regression, or nonlinear regression. Many research labs will find that InStat effectively covers the majority of needed analysis tasks. Perhaps the only addition that this reviewer would like to see is the inclusion of survival analysis, although this feature (along with polynomial regression, nonlinear curve fitting and two-way ANOVA) is included in GraphPad Prism, the powerful data analysis and technical graphics package from GraphPad Software. (See HMS Beagle's review of Prism).

Interpretation of Analysis

InStat presents the results of statistical tests in an easy-to-read format. All relevant information, which is dependent on the structure of one's data and the statistical tests and posttests performed, is clearly displayed in a standardized output format. For example, figure 10 shows the results of a one-way ANOVA performed on data comparing three treatments with controls and includes a Dunnet multiple comparison posttest, as well as a summary table of each comparison. In addition, the results file includes tests of the assumptions that are not shown in figure 10. Bartlett's method is used to determine whether the standard deviations are equal, and the Kolmogorov and Smirnov method tests the assumption of Gaussian distribution.

Graphical Output

InStat's graphical output is rudimentary but effective. InStat automatically generates a graph each time a new data set is created (figure 11). It is not possible to customize these graphs, and they are not of publication quality, but they do effectively illustrate salient points of the analysis. Prism offers complete flexibility for publication-quality plot creation.

Documentation and Internet Support

GraphPad Software continues its tradition of producing excellent documentation for its software products, and for data analysis topics in general. The InStat Guide to Choosing and Interpreting Statistical Tests not only helps one learn the software, but also provides a review of statistics so the user can pick the right test and interpret the results. The manual may be viewed free online, or downloaded as a PDF file. A separate online tutorial clearly demonstrates how to work in InStat. In addition, GraphPad Software provides a Data Analysis Resource Center that includes guides to nonlinear curve fitting and analysis of radioligand binding data, as well as short articles covering topics such as detection of outliers, the use of residual plots, and analysis of binding data with ligand depletion.

GraphPad InStat fills a necessary need in the statistical software market. Rather than attempt to include every possible analysis method at the expense of performance and clarity, GraphPad has included those methods most used by biomedical researchers. InStat effectively implements appropriate statistical analysis methods from an easy-to-use, intuitive interface. Despite simple graphical output and the lack of survival analysis (features that are included in Prism), InStat is an excellent software tool for the research lab. InStat is highly recommended, and should be the first statistical software program to use when analyzing data from the biomedical research laboratory.

GraphPad InStat 3.0 will work with Windows 3.1, Windows 95, or Windows NT. It requires a 486 or better processor, 4 Mb of RAM, and 3 Mb of free hard drive space. InStat 2.0 is available for the Macintosh, and requires Mac OS System 6 or later, at least 700 Kb RAM, and a hard drive. Visit the GraphPad Web site for more information on the differences between the Windows and Macintosh versions of InStat.

For a limited time, InStat is available for $99 from GraphPad Software, Inc., 10855 Sorrento Valley Road, #203, San Diego, California 92121. Contact GraphPad Software for upgrade information by phone at (800) 388-4723 (in the United States) or (619) 457-3909 (outside the United States), by fax at (619) 457-8141, or by email at sales@graphpad.com or orders@graphpad.com. A demo version of InStat is available. Ordering options include a discounted ($79) Internet delivery option. The GraphPad Software Web site includes material on their other data analysis software tools, Prism and StatMate.