A small-sample Bayes credible band around the Kaplan-Meier survival curve. The program makes a multi-colored graph. Also, the program prepares a table with tab separators between the numerical fields, for exporting to a spread-sheet. The Microsoft Internet Explorer 8 browser, and earlier, cannot work this page. Google Chrome and Mozilla Firefox and Safari can. Please click on randomBeta.htm. This is a one-sample inference.
A small-sample Bayes credible band around the difference of two Kaplan-Meier survival curves. The program makes a multi-colored graph. Also, the program prepares a table with tab separators between the numerical fields, for exporting to a spread-sheet. The Microsoft Internet Explorer 8 browser, and earlier, cannot work this page. Google Chrome and Mozilla Firefox and Safari can. Please click on randomBeta2.htm. This is a two-sample inference.
Bayes tests for r by c contingency tables and for other categories problems: please click on bayesCategories.htm.
Frequentist non-parametric tests for symmetry around zero: please click on symmetryAroundZero.htm.
A frequentist non-parametric test for the two-sample problem, using permutations. Ties are handled correctly. Please click on twoSample.htm. This test can be slow.
A frequentist non-parametric test for correlation, using permutations. Ties are handled correctly. Please click on correlationPermutation.htm. This test can be slow.
Bayes and frequentist tests for the population means of non-negative populations: please click on martMean.htm.
Frequentist and Bayes conditional tests for counts models having one degree of freedom: please click on wdf.htm.
Leave-out-one cross-validations for: central tendency, least-squares line, multinomial, contingency with structural zeroes, k-sample, blocks-and-treatments inference with missing and supernumerary measurements, and interval estimation. Please click on crossValidate.htm.
A program to work Fisher’s combination of independent p-values, corrected for file-drawer effect. Perhaps meta-analysts may find it useful. Please click on FishCombFileDraw.htm.
An animation in pure JavaScript using a crude form of Daniel Asimov’s Grand Tour algorithm to show data in higher dimensional spaces. The Microsoft Internet Explorer cannot run this. Netscape and Firefox and Opera and Safari can. Please click on grandTour.htm
Every statistics teacher warns us against constructing the formula(s) of a hypothesis test by looking at the sample, and all students and all statisticians disobey the teacher. The students and the statisticians are unknowingly trying to do multiple inference, so I use a Bonferroni correction with infinitely many weights to help them. Please click on jsBonfer.htm
Smirnov’s exact (nonasymptotic) one-sample one-sided test of fit using the sample distribution function. (Kolmogorov’s test is two-sided.) Please click on Smirnov1.htm
Holm’s method to correct p-values for multiple testing. This is usually more powerful than the Bonferroni correction, and never less powerful, and the Holm may be used wherever the Bonferroni may be used. Please click on Holmed.htm.
A toy program to build a best non-randomized frequentist hypothesis test if both the parameter space and the sample space have only finitely many points. It is to amuse students of mathematical statistics. Researchers in science will find it useless. Please click on AllSubSets.htm.
A calculator for a two-sided confidence interval for the chance of heads when throwing a possibly unbalanced coin. This is applied to McNemar’s test. Please click on BinomialConfidence.htm.
A calculator for simultaneous inference of two-sided confidence intervals for the population medians of populations from which independent subsamples were drawn independently at random with replacement. This can be used for a k-sample test and to make confidence intervals for difference of two population medians. Please click on MedianConfidence.htm.
A frequentist Monte Carlo test for blocks and treatments, where by null hypothesis the distributions in the same block are symmetric around the same center. It uses Tukey’s simultaneous test for pairs of treatments, but it is nonparametric. It can handle heteroskedasticity (inequality of variances) correctly, because it reverses signs instead of shuffling. It can handle unsymmetry correctly, provided that distributions in the same block are exchangeable and continuous, by the use of ranks. (Ranks will then cause symmetry.) Please click on BlockTreatTukey.htm.
A frequentist Monte Carlo exact test for independence of rows from columns in a complete contingency table. (Structural zeroes are forbidden.) This kind of test is recommended for big sparse tables. Please click on Mcirc.htm.
A frequentist Monte Carlo exact test for quasi-independence of rows from columns in a contingency table having structural zeroes. Please click on McmcBesag.htm.
A frequentist deterministic exact test for quasi-independence in a contingency table possibly having structural zeroes. This is a toy program for tiny samples. The program uses complete enumeration. Please click on Determ2.htm.
A frequentist Monte Carlo test for blocks and treatments. Yes, it can do missing values and supernumerary values, and related tests. It expects the distributions in the same block to be exchangeable, hence homoskedastic. Please click on http://www.toad.net/~jkaplan2/BlockTreat.htm. BlockTreat is not an applet. It is a stand-alone console program.
at dot
smtw2gh toadmail com