Subgroup Analysis

Rational subgrouping is the foundation of meaningful SPC. The choice of subgroup size and composition determines whether control charts detect real process changes or just reflect noise. A poorly chosen subgroup strategy can make a capable process look unstable or an unstable process look capable.

The classic example: subgrouping consecutive parts from one machine captures within-machine variation but may miss between-machine differences. Subgrouping across machines captures total variation but masks machine-specific shifts. The "right" subgroup strategy depends on what sources of variation you need to detect versus those you want to include in baseline variability.

Subgroup size also affects statistical performance. Larger subgroups (n = 5–10) provide tighter control limits on X-bar charts (better shift detection) but may mask within-subgroup non-normality. Individual charts (n = 1) preserve distributional information but have wider limits and lower sensitivity to small shifts.

EntropyStat enhances subgroup analysis by applying the EGDF separately to each subgroup or to within-subgroup residuals, without requiring the normality assumption that traditional X-bar/R charts rely on. For individual measurements (subgroup size = 1), this is especially valuable because the Central Limit Theorem does not apply and normality cannot be invoked through averaging.

The ELDF provides a data-driven approach to subgroup validation. If you believe your subgroups are rational (homogeneous), the ELDF can test this by checking whether measurements within each subgroup form a single cluster. If the ELDF detects multiple clusters within a subgroup, it suggests the subgroup definition is mixing distinct process conditions — the subgroup is not truly rational.

This automated subgroup validation replaces the manual, judgment-based process of determining rational subgroups. Engineers typically rely on process knowledge to define subgroups, then hope the choice was correct. EntropyStat's homogeneity testing provides statistical confirmation or rejection of the subgrouping strategy, strengthening the foundation on which all subsequent SPC analysis is built.