Data
Give the decisions of your own analysis. The page runs no significance tests on your data. A group is one item or treatment.
Examples
Simulated experiment
Simulates the means of an experiment and decides each pair with a t test, |t| > 1.96. "Methods and sources" below the results gives the model.
One row for each pair of groups, with the columns group1, group2, and significant. The decision is true when the two groups differ significantly. Every pair needs a row.
or .
Columns group and mean. The means set the order of the groups and of the letters. They do not change a decision.
or .
Method
CLD-C gives the fewest distinct letters, then the fewest assignments among those displays. CLD-sigma gives the fewest assignments of a letter to a group, so the fewest letters in the table, then the fewest distinct letters among those displays. The R, Python, and npm packages stop after the first minimization; the second is experimental and exists in this page only.
Study
Simulation study
This view follows the design of the experiment of Section 8 of Ennis, Fayle, and Ennis (2012) in this browser. Each instance is one simulated experiment. The page runs CLD-C, CLD-sigma, and both two-stage solves on it, each solve with the time limit of the study settings. The 2012 study ran an exact backtracking search (FIND-AM) on 1,000 instances for each number of treatments, with a limit of 30 seconds.
Set the sizes and the time limit, then press Run the study. The tables appear here as the instances finish.