Features and Specifications

LogXact® Features

• Powerful Monte Carlo procedures enable fast exact inference for much larger data sets
• Exploration Mode tells you how long a problem will take before you commit computing resources
• Exploration Mode also allows use of customized exact algorithms to optimize computer performance for difficult problems
• See LogXact® overview page for more details

Thorough Documentation

LogXact is shipped with a detailed 350+ page User Manual that thoroughly documents all the binary logistic regression and Poisson regression methods as well as the user interface. The manual is packed with interesting examples showing every step of how to run each procedure. There are also examples illustrating how LogXact can handle large datasets. The manual is well indexed and has an extensive table of contents. We have maintained the high standard for user documentation of our previous StatXact® and LogXact User Manuals.

Validation of LogXact® Results

The statistical results computed by LogXact engines have been subjected to rigorous and extensive quality-assurance testing for purposes of validation. A database of a large number of datasets has been compiled at Cytel Inc. These datasets have been collected from standard textbooks, published articles, LogXact users and beta testers, as well as constructed by us in order to test across the range of possible input values.

Method 1: Comparison of exact p-values, asymptotic p-values, permutation distributions and exact parameter estimates computed by LogXact 5 engines and corresponding values computed by LogXact 2 and 4. Apart from rigorous pre-release testing, the LogXact 2 and 4 engines have been extensively used at the sites of over a thousand users for several years and are therefore mature enough to provide a "gold standard."

Method 2: Cross-validation (within LogXact 5) of exact p-values with Monte Carlo p-value confidence intervals using a very large Monte Carlo sample size (typically in excess of a million tables). The Monte Carlo intervals should include the exact p-value up to the confidence level of the settings.

Method 3: Comparison of exact p-values and exact parameter estimates with results from StatXact. Certain inference in logistic regression models can be recast as various tests in StatXact. The algorithms used in StatXact are quite different than those used for LogXact.

Method 4: Comparison of asymptotic values with the results from SAS, SPSS and S-PLUS.

StatXact statistical procedures that validate the logistic regression procedure of LogXact include:

• Wilcoxon-Mann-Whitney test
• Generalized WilcoxonöGehan test
• Normal Scores test
• Savage Scores test
• Logrank test
• Conf. Interval for the Odds Ratio of Two Binomials
• Homogeneity of Odds-ratios for Strat. 2 X 2 Tables
• Conf. Interval for the Common Odds-ratio for Strat. 2 X 2 Tables
• Wilcoxon-Mann-Whitney test for Strat. 2 X C Tables
• Normal Scores test for Strat. 2 X C Tables
• Savage test for Strat. 2 X C Tables
• Cochran-Armitage Trend test for Strat. 2 X C Tables
• Permutation test with General Scores for Strat. 2 X C Tables

  For additional details and on-site inspection of our testing documentation contact us at support@cytel.com or617.661.2011.

  The datasets used under Methods 1 & 2 are available in the LogXact 5 folder "Data". The datasets used under Method 3 correspond to the datasets used under each of the statistical procedures in StatXact 5 and as described in the StatXact 5 manual. The datasets are available in the StatXact 5 folder "Data". The datasets used under method 4 include the datasets used for methods 1 & 2.