MerQuR+Statistics Glossary

Statistics Glossary → Page's test (ordered alternatives)

Page's test (ordered alternatives)concept

A nonparametric test of whether three or more conditions measured on the same subjects show an increasing (or decreasing) trend in a pre-specified order.

The Friedman test asks whether the conditions differ 'in some way'; Page's test predicts that the differences follow a given order. Within each subject the conditions are ranked, and the rank sum Rⱼ of each condition is computed. The statistic is L = Σ j·Rⱼ, where j is the condition's position in the predicted order. If the predicted trend is real, large rank sums pair with large j values and L grows; in large samples L is tested with a normal approximation.

When the ordered alternative is true, Page's test is more powerful than Friedman's because it concentrates the information in a single direction. On the other hand, if the predicted order is wrong (e.g. the trend is U-shaped) it can miss real differences. The order must be fixed before seeing the data. Its counterpart for independent groups is the Jonckheere-Terpstra test.

An everyday example

A coffee shop has the same 12 customers taste three coffees with little, medium and a lot of sugar, expecting liking to rise with sugar. Each customer ranks the three; the sweetest is usually ranked first and the least sweet usually last. While the Friedman test only says 'there is a difference', Page's test shows more powerfully that this increasing order is consistent.

In MerQur

Available as the 'Page' checkbox in the Friedman test; the order of the selected columns is the predicted increase, and the test uses a one-sided normal approximation. For a decreasing trend the columns should be selected in reverse order.

Analyses where this concept appears

Friedman

Related concepts

Jonckheere-Terpstra test (ordered alternative)Kendall's W (coefficient of concordance)