A term used in connection with the merging of neighbouring categories. For example, if the random variable X takes values 0, 1, and 2 with high probability and values 3, 4, and 5 with low probability, then it may be sensible to collapse the latter categories into a single ‘3 or more’ category.

As a more extreme example, suppose that, in an I×J×K×L contingency table, for variables A, B, C, D, it turns out that D is of no interest. All the categories of D may then be collapsed together to give an I×J×K table—the original table has been collapsed over D.