Performance data need a context to meaningfully interpret the data. On
e method of providing context for an individual unit's performance is
to compare it with other similar units. This study compares three meth
ods for selecting similar units: cluster groupings, index groups, and
benchmark groups. Each of the three methods is evaluated on a number o
f criteria, primarily the minimization of within-group variance. Bench
mark groups are the best at reducing the variation within the selected
groups, and they resist attempts to ''label '' the groupings. Cluster
groups are a close second to benchmarks in the minimization of variab
ility within groups and are considerably easier to compute and adminis
ter However clustering allows labeling that could stigmatize the group
s and threshold effects that might influence judgments about performan
ce. Index groups, while simple, do not perform well on any of the othe
r criteria.