DTM.3:4.4 - Choose an aggregation that preserves the needed effect

Choose among individual states, a few relevant classes, a distribution over state or age since adoption, and a justified common state. Use the least detail that preserves the comparison.

When event contribution is q(x), the relevant population quantity is the appropriately weighted average of q(x), not automatically q of the average x. Weights come from the modeled events: an individual making many demonstrations can contribute more exposure than an individual making none. Population shares alone need not be the weights.

A time-since-adoption description is useful when newcomers and experienced users produce materially different events. An average state is more defensible when the effect is approximately linear over the relevant range or when variation is small enough for the decision. Test that approximation, rather than choosing a fixed number of classes.

Return to MMP.18 if shared quantities, units, conservation, state boundaries or incompatible assumptions prevent composition.