Repeated measures ANOVA

•Two-way ANOVA with repeated measures in both factors. Prism 5 could only compute two-way ANOVA with repeated measures in one of the factors. Prism 6 can handle repeated measures in both factors.

•More subjects. Prism 6 now allows up to 256 subcolumns, so can perform repeated measures two-way ANVOA with up to 256 subjects per treatment.

•The Greenhouse-Geisser correction allows repeated measures one-way ANOVA to work, even when the assumption of sphericity is violated because the repeated measurements are made in too short a time interval, so that random factors that cause a particular value to be high (or low) don't wash away or dissipate before the next measurement. In this case, multiple comparisons are also computed so as to not assume sphericity.  

Survival analylsis

•Prism now computes the hazard ratio (HR) by two methods: The logrank method and the Mantel-Haenszel method.

•If people in group A die at twice the rate of people in group B (HR=2.0), then people in group B die at half the rate of people in group A (HR=0.5). Prism now reports the hazard ratio, and its confidence interval, both ways so you can report the one that makes most sense in your clinical context.

•If people in group A have a median survival time three times people in group B, then people in group B have one third the median survival of people in group A. Prism now reports the ratio of median survival times, and its confidence interval, both ways so you can choose which to report.

Analyses of column data (t tests etc.)

•Mann-Whitney tests reports 95% CI of difference between medians. The Mann-Whitney test is often described as comparing the median of the two groups. That is not precisely correct, as it is possible for two groups to have the same median, but for the Mann-Whitney test to find that the difference between the distribution of ranks among the two groups is statistically significant. However, if you assume that the two distributions have roughly the same shape, then it is fair to think of the Mann-Whitney test as comparing two medians. Accordingly, Prism 6 can now report the confidence interval for the difference of two medians.

•The Wilcoxon test computes the 95% CI of the median. The Wilcoxon matched-pairs test (nonparametric test of two paired or matched groups) reports the confidence interval of the median of the paired differences, and the Wilcoxon signed rank sum test (nonparametric test to compare a median with a hypothetical median) now reports the confidence interval of the difference between the actual sample median and the hypothetical median.

•Ratio t test. The paired t test works by analyzing the difference between each pair of values, testing the null hypothesis that the average difference is zero. With some kinds of data, the difference between before and after is not a consistent measure of effect. The differences might be larger when the "before" values are larger, and smaller when the "before" values are smaller. The ratio (after/before) may be a much more consistent way to quantify the effect of the treatment. Actually, it turns out that analyzing the logarithm of ratios works much better.

•Kolmogorov-Smirnov test.  Like the Mann-Whitney (MW) test, the Kolmogorov-Smirnov (KS) test, is a nonparametric method to compare two groups. The KS test works by comparing the two cumulative frequency distributions, and so has more power to detect subtle differences in the two distributions. In contrast, the  MW test is better at detecting changes in the median. The use of the KS test has become standard in some scientific fields. Don't confuse this test with the the other version of the KS test used for normality testing.

•Faster nonparametric tests. To avoid slow calculations, Prism reports an approximate P value for nonparametric tests with large data sets. Prism 5 also reported approximate P values when several values were identical and so tied for the same rank. Prism 6  performs the exact calculations much faster (hundreds of times faster!), and so does exact calculations even for fairly large data sets and even when several values are identical and so tie for the same rank.

•Identify outliers. One of our free QuickCalc web calculators identifies outliers from a stack of values using Grubbs' method. Because this is one of our most popular calculators, we created a new analysis in Prism 6 to identify outliers in a column of data.

•Perform many t tests simultaneously. Computing multiple t tests can be useful, so long as you correct for multiple comparisons. Prism 6 has a new  analysis to perform one t test per row of a data table (with replicates placed side-by-side). It corrects for multiple comparisons either by using either a more stringent definition of statistical significance or controlling the False Discovery Rate (FDR). You can also choose not to correct for multiple comparisons.

•More graphing choices. Prism 6 offers options to automatically graph ranks of nonparametric tests, and differences for paired t test and repeated measures ANOVA.

•Browne and Forsythe test. To test equality of variances in one-way ANOVA, Prism 6 computes both the Browne and Forsythe test as well as the Bartlett test (which previous versions computed).

•Method of Pratt. There are two ways to compute the Wilcoxon matched-pairs test when some of the pairs are identical, so the difference is zero. Prism 6 offers both choices. The new choice is the method of Pratt for tied rows.

Simulations and Monte-Carlo analyses

Simulations can be useful even when you just simulate one data set, and view the graph and analysis.Prism 5 let you simulate XY data. Prism 6 now let's you also simulate column data and 2x2 contingency tables. For XY and Column data, Prism 6 can include not only Gaussian random scatter, but also scatter computed from a Poisson or binomial distribution.

Simulations are even more useful when you repeat the simulation many times, and tabulate the results. Prism 6 makes this possible with a new Monte Carlo simulation analysis that makes it easy to tabulate the results of many simulations. First simulate a data table and run an analysis to analyze that table. Then run the new Monte Carlo analysis. Specify how many simulations you want to run, and which analysis parameters to tabulate. You can also define a “hit”, as perhaps a P value less than 0.05, or a confidence interval including a theoretical value. The results of the Monte-Carlo analysis include a table with all the tabulated data (that you can analyze further), and a table of the number of simulations that were hits.