Separate Within-Run Repeatability from Day-to-Day Measurement Variation

A station can repeat a reading closely for several minutes and still shift from day to day. A short burst of measurements characterizes only the conditions within that burst. To study longer-term behavior, organize the data so that within-run variation and between-day changes remain distinguishable.

NIST’s two-level nested design guidance separates short-term repetitions from measurements across days or another appropriate time interval. Adapt the time levels to the station and the source of variation being investigated, rather than copying an arbitrary calendar schedule.

Define the two time levels

Choose a sufficiently stable reference or check item for the intended measurement. Identify what will remain unchanged within a short run and what will naturally vary between runs. Examples include reconnection, startup, environmental conditions, or an ordinary operator handoff.

Record the reference identity, instrument, fixture, recipe, date, run, and repetition number. Preserve individual readings instead of storing only a daily average. The average cannot reconstruct the within-run spread after the raw data are discarded.

Keep the hierarchy intact

Plot the run averages over time alongside the within-run spread. A station with stable averages but noisy repetitions presents a different problem from one with tight repetitions and moving averages. Both patterns can disappear when every reading is pooled into one unlabeled column.

Investigate the observed pattern

1. Review data completeness and configuration changes.
2. Compare within-run behavior across sessions.
3. Examine shifts in session averages.
4. Relate candidate causes to recorded station events.
5. Use a suitable variance analysis for the actual study structure.

Do not label every between-day difference as instrument drift. The reference, connection, environment, or procedure may have changed. Check those possibilities with independent evidence before adjusting the instrument or widening product limits.

If several operators or instruments are involved, plan their assignment deliberately. A design in which one operator always uses one instrument cannot separate those effects. Similarly, measuring every reference only once per day gives no direct within-day repetition for that reference; preserve the planned replication instead of replacing it with unrelated specimens. More readings within the same arrangement do not remove that confounding.

Report the time horizon and conditions represented by the study. The conclusion should distinguish short-term repeatability from variation across ordinary operation, providing a realistic basis for station comparison and measurement planning.