Analyze

Multi-Vari Chart

Stages single measurements across up to four factors and splits the spread into variance components — shows where the variation actually comes from.

Multi-Vari Chart

The multi-vari chart shows every single measurement, staged across up to four factors. It answers the question a single summary statistic hides: does the variation arise between shifts, between mould cavities, or within one part?

In Minitab it lives under Stat → Quality Tools → Multi-Vari Chart. Qprovement adds the variance-component table underneath — it puts numbers on what the picture shows.

Factor order: Position 1 sits on the x axis and is also the outermost nesting level, position 2 becomes the series (colour and symbol), position 3 the panel column, position 4 the panel row. Layout and model follow the same order, so the picture and the numbers cannot drift apart.

Nested or crossed: Nested: the levels of the second factor exist only inside one level of the first (part 1 from cavity 1 has nothing to do with part 1 from cavity 2). Crossed: every level occurs with every other (machine 1 runs on both the early and the late shift).

Data & operation

Measurement column — one numeric column holding the single measurements.
Factors — two to four columns of levels. Rows with an empty level or a non-numeric measurement are skipped and reported.
Order — drag by the grip. The role label on the left moves with it.
Estimator — ANOVA (Henderson I) decomposes the sequential sums of squares via expected mean squares; REML estimates the components directly from the likelihood. On a balanced design the two agree for practical purposes.

Negative variance estimates are clamped to 0 after the solve and marked in the table. Such a case usually means there is too little data for that level.

Pitfalls

Unbalanced design. With unequally occupied cells the components depend on the estimator. The module says so; REML is the more dependable choice then.
Wrong model form. A nested design computed as crossed produces interaction terms that cannot exist. When in doubt: do the second factor's levels also exist outside the first?
Size limits. Beyond 30 levels in position 1, 12 series in position 2 or 48 panels the chart is not drawn — it would be unreadable anyway. The message names the offending column; the variance-components table stays valid. Move a factor with fewer levels to the front.
Share ≠ importance. A large variance share says where the variation arises — not whether it hurts. Only the comparison with the tolerance turns it into a statement.
Order is model. Reordering the factors changes the computed terms in the nested case, not just the picture.

Examples

This module ships with the following example datasets — load any of them in the app with a single click.

Multi-vari: injection moulding, nested — Project template

2 shifts × 2 mould cavities × 3 parts, 2 wall-thickness measurements each. Balanced and nested: most of the variation sits between the cavities, not between the shifts.

Multi-vari: filling line, crossed — Project template

2 machines × 2 shifts × 2 batches, unequally occupied with one missing combination. Crossed with REML — also shows how the module flags an unbalanced design.

Available in the following cycles

DMAIC: Analyze
DMADV: Analyze
8D: D4 — Root Cause Analysis