Before the percentage, find the denominator

Precision is not the same as comparability

The generation queue dataset is a useful reminder that a large, carefully assembled record still describes a particular population. A researcher must retain that population when carrying the finding into a new argument. Lawrence Berkeley National Laboratory ↗

A compact protocol

Write the unit of observation in a sentence. State the geography and window of time. Explain what qualifies an observation to enter the count, and what constitutes an exit or completion. Then ask whether the comparison uses those same definitions. If it does not, report the mismatch before calculating a ratio.

For deployment research, separate announced projects, financed projects, connected sites and operating capacity. These categories can overlap without being interchangeable. A percentage that divides one by another can be meaningful, but only if the relationship and timing are established.

An example without invented data

Imagine a record of announced sites and a second record of grid connections. A connection may serve more than one phase of a campus; an announcement may contain several planned buildings. Dividing one count by the other does not yet produce a completion rate. The first research task is entity matching. This example describes a measurement problem, not the observed performance of a real project group.