Are Productive People to Be Found? Robust Analysis of Sparse Two-Way Tables

Abstract

Data from individual measurements of production in a manufacturing environment may be thought of as a two-way table indexed by worker and task. If there are many tasks, such a table will normally be sparse. Adjustment for task effects is important if workers are to be compared to one another, since any two workers may have no tasks in common. Three techniques are presented for the analysis of such sparse tables: table-stripe regression, rectangle ratios, and network scores. These techniques are illustrated on data from a large industrial concern, where 6,846 person—task—weeks of data on 479 workers and 678 tasks were collected as part of an ongoing group incentive plan.