Multi-vari chart¶
A chart that preserves lower-level numerical observations or spread within declared process and time strata so variation within and among groups can be compared.
Core Idea¶
A multi-vari chart displays one measured process response across a declared multilevel sampling plan, preserving lower-level observations or spread so variation within groups can be compared with shifts among groups.[ref-af7bbaf16fb1][ref-085e9d940180] Its “multi” refers to variation levels rather than several response variables. Different marks can do the job: de Mast shows ranges and summaries for tea boxes, while Steiner and MacKay show individual molded-part points; neither display alone proves a physical cause.[ref-af7bbaf16fb1][ref-085e9d940180]
Scope of Application¶
Use it to compare one measured response within and among declared process or time groups:
- Tea-machine comparisons — box-weight ranges within samples contrast with recurring machine summaries.[^ref-af7bbaf16fb1]
- Tea-day comparisons — box-weight ranges within samples contrast with the marked day means.[^ref-af7bbaf16fb1]
- Molded-part hour comparisons — crossbar points within sampled hours contrast with other hours.[^ref-085e9d940180]
- Molded-part day comparisons — crossbar points within sampled hours contrast across day groups.[^ref-085e9d940180]
Clarity¶
The chart separates lower-level spread from higher-stratum shifts, a distinction a pooled histogram or lone factor mean may hide. Tea-box squares join sample minima and maxima, machine means connect within a day, and crosses mark day means; the molded-part chart instead retains individual points, without a source-defined physical unit or complete legend for every overlay.[ref-af7bbaf16fb1][ref-085e9d940180] Read each mark under its own convention, and treat an unusual stratum as a clue rather than verified bag count or molding cause.
Manages Complexity¶
Keeping a sampling address with each value or summary lets a reader see whether a large overall range concentrates inside cells or among machines, sampled hours or days. The tea chart compresses five-box sets into ranges and group summaries, while the crossbar chart retains individual parts in hour/day groups; each reduces clutter differently and leaves causal or control-limit decisions for separate evidence.[ref-af7bbaf16fb1][ref-085e9d940180]
Abstract Reasoning¶
Use three moves: identify the one response, trace each observation through its actual sampling cells and recurring group labels, and compare lower-level spread with higher-stratum shifts. Then test what pooling would erase: a histogram may preserve overall range but lose the machine-day or sampled-hour location of an unusual pattern, and that location still does not prove its physical cause.[ref-af7bbaf16fb1][ref-085e9d940180]
Knowledge Transfer¶
The same chart grammar travels literally from tea packing to molded-part production when one measured response, sampling addresses, lower-level marks and within/among-stratum comparisons are retained.[ref-af7bbaf16fb1][ref-085e9d940180] Scientific Diagram is the broader kind: observations are mapped through visual conventions to a bounded comparison, while this chart adds the one-response process-strata grammar; Representation is inherited through that parent. A factor-level-means-only display also called “multi-vari” needs a separate check for retained lower-level values or spread, and a grouped plot outside process measurement is only analogous until all four roles map.[^ref-3ba64ae2e825]
Relationships to Other Abstractions¶
Current abstraction Multi-vari chart Domain-specific
Parents (1) — more general patterns this builds on
-
Multi-vari chart is a kind of Scientific Diagram Domain-specific
Every admitted multi-vari chart is a scientific diagram of measured process variation, specialized to one response and comparison within and among multilevel sampling strata.
Hierarchy path (1) — routes to 1 parentless root
- Multi-vari chart → Scientific Diagram → Representation → Abstraction
Neighborhood in Abstraction Space¶
Multi-vari chart sits in a sparse region of the domain-specific corpus (93rd percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.
Family — Unclustered & Miscellaneous (2721 abstractions)
Nearest neighbors
- Ridgeline Plot — 0.80
- Multilevel regression with poststratification — 0.79
- Van der Waerden Test — 0.78
- Balanced repeated replication — 0.77
- Quadrant Count Ratio — 0.77
Computed from structural-signature embeddings · 2026-10-17
References¶
[^ref-af7bbaf16fb1]: Jeroen de Mast. (2002). “The multi-vari chart: a systematic approach,” chapter 4 of Quality improvement from the viewpoint of statistical method, Universiteit van Amsterdam thesis, original chapter PDF. Relevant one-based PDF pages 2–4 (printed pp. 71–73), Figure 4.1, and PDF pp. 12–13 (printed pp. 81–82), Figure 4.7. Figures were visually checked. The box-count explanation is the author's likely interpretation; the later caffeine ANOVA is separate from chart-only inspection. [^ref-085e9d940180]: Stefan H. Steiner and R. Jock MacKay. (2006). “Reducing Process Variation With Statistical Engineering: A Case Study,” Quality Progress, June 2006, pp. 33–39, author-hosted copy of original article. Relevant one-based author-hosted PDF pp. 2–5, especially Figure 3 on p. 3. The author-hosted PDF pages are not equated with journal pages. Figure 3 does not fully define every overlaid glyph or a physical response unit. [^ref-3ba64ae2e825]: Minitab. (undated). “Multi-vari chart basics”, official software documentation, “What is a multi-vari chart?” Its factor-level-mean use of the name is adjacent nomenclature, not an original worked case for the bounded process-strata identity used here.