From the Welcome or New table dialog, choose the contingency tab.
If you are not ready to enter your own data, choose one of the sample data sets.
Most contingency tables have two rows (two groups) and two columns (two possible outcomes), but Prism lets you enter tables with any number of rows and columns.
You must enter data in the form of a contingency table. Prism cannot cross-tabulate raw data to create a contingency table.
For calculation of P values, the order of rows and columns does not matter. But it does matter for calculations of relative risk, odds ratio, etc. Use the sample data to see how the data should be organized.
Be sure to enter data as a contingency table. The categories defining the rows and columns must be mutually exclusive, with each subject (or experimental unit) contributing to one cell only. In each cell, enter the number of subjects actually observed. Your results will be completely meaningless if you enter averages, percentages or rates. You must enter the actual number of subjects, objects, events. For this reason, Prism won't let you enter a decimal point when entering values into a contingency table.
If your experimental design matched patients and controls, you should not analyze your data with contingency tables. Instead you should use McNemar's test.
If you want to compare an observe distribution of values with a distribution expected by theory, do not use a contingency table. Prism offers another analysis for that purpose.
From the data table, click on the toolbar, and choose Chi-square (and Fisher's exact) test.
Your choice of effect sizes will depend on experimental design. Calculate an Odds ratio from retrospective case-control data, sensitivity (etc.) from a study of a diagnostic test, and relative risk and difference between proportions from prospective and experimental studies. All of these effect sizes apply only to 2x2 tables, so the choices will be gray if your table is larger.
If your table has two rows and two columns, we suggest you always choose Fisher's exact test to calculate the P value.
If your table has two columns and three or more rows, you can choose the chi-square test or the chi-square test for trend. This calculation tests whether there is a linear trend between row number and the fraction of subjects in the left column. It only makes sense when the rows are arranged in a natural order (such as by age, dose, or time), and are equally spaced. The test is also called the Cochran-Armitage test for trend. It is explained clearly, with equations and an example, on pages 261-265 Altman (2). You can find these pages at Google Books.
With contingency tables with more than two rows or columns, Prism always calculates the chi-square test. You have no choice. Extensions to Fisher's exact test have been developed for larger tables, but Prism doesn't offer them.
•CI of the relative risk. The method of Katz is an approximation, and we suggest not using it except to maintain compatibility with analyses done with earlier versions of Prism. There are many ways to compute the CI of a relative risk (1), and many of these seem good. Prism now offers the Koopman asymptotic score method, which we recommend.
•CI for the difference between proportions. The asymptotic method used by earlier versions of Prism is an approximation, and we suggest not using it except for compatibility. Instead choose the Newcombe/Wilson method(2). We offer that method with and without the continuity correction, and recommend the variation with that correction.
•CI of the odds ratio. The method of Woolf used by Prism 6 and earlier is an approixmation, and we suggest you not use it except to maintain compatibility. There are many good methods to compute the CI of an odds ratio (2). Prism now offers the Baptista-Pike method, which we recommend.
•CI of the sensitivity, specificity, etc. The so called "exact method" of Clopper and Pearson produces wide confidence intervals, and we suggest you don't use it except for compatibility. Instead choose the hybrid Wilson/Brown method (3).
You can also choose how you want P values formatted.
1.Fagerland MW, Lydersen S, Laake P. Recommended confidence intervals for two independent binomial proportions. Stat Methods Med Res. SAGE Publications; 2011 Oct 13.
2.Newcombe, R. G. R. (1998). Interval estimation for the difference between independent proportions: comparison of eleven methods. Statistics in Medicine, 17(8), 873–890.
3.Brown, L., Cai, T., & DasGupta, A. (2001). Interval Estimation for a Binomial Proportion. Statist. Sci, 16(2), 101–133.