/grouping

The domain of grouping: putting observations into kinds that can be named, counted and acted on, whether the kinds are found or given in advance. Use these principles to do the work or to review it; in a review, name each breach once, with its fix.

Principles

1. The purpose fixes the relation. State the question the groups answer and the relation that puts two observations together. Group by it, not by surface likeness such as shared form.
2. Found kinds come from the observations. Unless the kinds are given, build them from the observations. Familiar categories pull what you see toward them: use them at most to gather observations, never to form the groups. Hold each kind as provisional until it earns its place.
3. Dense inside, distinct between. Members cohere; any two groups differ in a way you can state. Groups at one level are the same kind of thing.
4. Name each group by what its members share, specifically enough that a reader can tell what belongs; a higher group's name sums up those below it. A label that only says a group exists, or borrows a familiar category, is not a name. Define each group in a sentence or two: what it includes and what it excludes.
5. Account for every observation. Each is in a group or stays alone. Challenge a catch-all group: it may hold kinds not yet found. Keep an exception as a variation or a new kind.
6. Groups that will be counted or decide something are checked by someone who did not build them. That reader assigns a sample by the definitions alone; report agreement per group, against a bar fixed in advance by the cost of a wrong conclusion. Other groups must let such a reader see that each follows from its members. A disagreement is a finding about a definition: sharpen it and check again. Agreement shows that others can reproduce the groups, not that they are right.
7. Check against observations that did not make the groups. Place fresh observations against the final groups; when there are none, reread the whole set.
8. Compare each observation with the groups so far. Group like with like, and only like. Split a group that holds two kinds; merge two groups that are one. Stop when further comparison no longer changes them.
9. Fix the unit before grouping: what one observation is, at a comparable scope and a size the purpose sets, and whether it may sit in several groups.
10. Count only settled kinds, and say what one count is. Size alone does not make a group important; importance follows the purpose.

Definitions

• Given groups: fixed in advance and applied by their definitions.
• Agreement: how often independent readers place the same observations in the same groups, beyond chance.