MerQuR+Statistics Glossary

Statistics Glossary → Cut point (dichotomization)

Cut point (dichotomization)concept

A threshold value that splits a continuous or many-valued variable into two groups, those below and those above it.

Some tests work only with two-category data: the binomial test counts successes and failures, and the runs test looks at the order of two kinds of values in a sequence. A numerical variable enters these tests through a cut point, for example below and above the median. Which group values exactly equal to the cut point go to must be decided in advance, because it can change the result.

Dichotomizing loses information: values just below and far below the cut point are treated the same. In analyses such as regression, splitting a continuous variable in two therefore usually costs power and can mislead. If the cut point is chosen after looking at the data to give the most significant result, the p values become invalid; it should be set in advance and with a rationale (e.g. a clinical threshold or the median).

An everyday example

A factory records the number of defective parts made each day. The quality manager labels each day 'good' or 'bad', saying 'more than 5 defective parts makes a bad day'. This makes it possible to compare the share of bad days with a target or to test whether bad days come in runs.

In MerQur

'Cut point' in the Binomial Test (in a numeric column, x ≤ c counts as a success); in the Runs Test, 'Cut point' (Median, Mean, Mode, Custom value) and the 'Values equal to the cut' option.

Analyses where this concept appears

Binomial TestRuns Test

Related concepts

Mood's median test