Statistical Methods
Check for a change in spread before pooling a research series
Keep the original order
Start with a sequence plot whose horizontal axis is time or acquisition order. Annotate predetermined changes in instrument, software, operator or collection schedule. Before calculating a pooled spread, check whether location and variability appear stable over the interval in which the pooled estimate would be used.
Compare comparable windows
Choose windows using recorded process events where possible. Report each window’s size and center alongside its spread. A window selected after searching for the most extreme change needs an exploratory label. A sharp change in observed dispersion can also arise from a new rounding or reporting rule.
Hypothetical check
The first four synthetic readings are 9, 11, 9 and 11. The next four are 6, 14, 6 and 14. Each group has mean 10. Their ranges are 2 and 8 units, a fourfold difference. A report giving only the two means would conceal the change; a single pooled dispersion would describe neither period particularly well.
Look back at the operating log and the reporting resolution. Retain all original observations and document any exclusion criterion before re-estimation. For a very short sequence, a descriptive plot can support further investigation but cannot reliably determine the timing or cause of a variance change.
Method checked 9 November 2026. NIST/SEMATECH Engineering Statistics Handbook, 1.3.3.25. Run-Sequence Plot.