Decomposition Plot¶
Diagnostic artifact — instantiates Trend Detection and Removal
Displays observed, trend, seasonal or cyclical, and residual components for review.
A Decomposition Plot is the artifact you look at to see a series taken apart: a stacked panel of aligned charts showing, from top to bottom, the observed series, the extracted trend, the recurring seasonal (or cyclical) component, and the leftover residual. Its whole value is simultaneity and alignment — every panel shares one time axis, so the eye can trace a single date down through the stack and see how much of a bump belonged to the trend, how much to the season, and how much was genuinely anomalous. It does not decide which component matters or test whether the residual is clean; it renders the split visible and keeps every removed piece on the page, so nothing is silently discarded. It is the review surface, not the analysis.
Example¶
An e-commerce team looks at three years of weekly sales and cannot tell a story from the raw line: it climbs, but it also saws up and down, and a few weeks spike hard. They generate a Decomposition Plot — a classic STL (Seasonal-Trend decomposition using Loess) breakdown.[n1] The top panel is the familiar noisy observed line. Below it, the trend panel shows a smooth, steady climb — the business genuinely growing. The seasonal panel reveals a clean repeating shape: a Q4 holiday surge and a January trough, the same every year. The residual panel is mostly flat noise except for two tall spikes, which the team immediately recognizes as a viral promotion and a site outage. In one glance the messy line has become four legible stories, and — crucially — the trend and season are still shown, not thrown away, so no one mistakes ordinary seasonality for a growth surge.
How it works¶
What distinguishes it from a numerical decomposition is that its output is a coordinated visual, built for review:
- Run a decomposition — additive or multiplicative, splitting the series into trend, seasonal/cyclical, and residual parts.
- Stack the panels on a shared axis — observed, trend, seasonal, residual, vertically aligned so any date lines up across all four.
- Preserve scale honestly — annotate each panel's magnitude so a small-looking residual is not mistaken for a large one, or vice versa.
- Keep everything on the page — the removed trend and season remain visible components, not deletions.
Tuning parameters¶
- Additive vs. multiplicative — whether components add or multiply; multiplicative suits a seasonal swing that grows with the level.
- Trend smoothness — how flexible the extracted trend is, which trades a smooth trend against a busier residual (and vice versa).
- Seasonal-window rigidity — a fixed repeating season versus one allowed to evolve across years.
- Panel scaling — shared versus independent vertical scales, the single choice most able to flatter or exaggerate a component.
- Residual overlay — whether to mark thresholds or highlight outliers on the residual panel to aid review.
When it helps, and when it misleads¶
Its strength is communicative: it makes a decomposition inspectable by a human, surfacing at a glance whether a movement was trend, season, or anomaly, and enforcing the archetype's rule that a removed trend stays documented rather than vanishing.
Its failure mode is that a plot persuades before it proves. Independent per-panel scales can make a trivial residual look alarming or a real one negligible; an over-smoothed trend pushes structure down into the residual where it reads as anomalies; and the plot shows a decomposition, not the decomposition — change the smoothing and the panels rearrange. The classic misuse is treating the residual panel's every wiggle as an event worth explaining, when some of it is just the leftover the chosen trend happened to produce. The discipline is to fix scales honestly, state the decomposition settings on the artifact, and confirm anything read off the residual against a separate test rather than the picture alone.
How it implements the components¶
Decomposition Plot fills the display-and-reporting slice of the archetype's machinery:
trend_residual_visual_overlay— its primary form: the aligned, stacked visual of observed, trend, seasonal, and residual on one axis.component_separation_frame— the panels enact the separation frame, partitioning observed movement into trend, cycle, and residual for the viewer.trend_reinsertion_or_reporting_rule— by keeping the extracted trend and season on the page beside the residual, it is the reporting artifact that documents what was removed rather than erasing it.
It performs no test on the residual (stationarity_diagnostic) and does not itself estimate and strip a recurring cycle (seasonal_or_periodic_cycle_control) — producing a cleaned, adjusted series is Seasonal Adjustment Procedure, whereas this artifact only displays the split; nor does it locate breaks (change_point_marker, Change-Point Detection Test).
Related¶
- Instantiates: Trend Detection and Removal — the review surface that keeps every separated component visible.
- Consumes: the components produced by a separation procedure such as Seasonal Adjustment Procedure or Rolling-Window Trend Estimate.
- Sibling mechanisms: Seasonal Adjustment Procedure · Change-Point Detection Test · Residual Stationarity Check · Differencing Transform · Rolling-Window Trend Estimate
Editorial Notes¶
Form Classification¶
Form family: Interface, Display & Cue
Rationale: Decomposition Plot operates as a user-facing prompt, display, template, or perceptual cue that shapes attention and action at the point of use because it displays observed, trend, seasonal or cyclical, and residual components for review.
Independent corroboration: The frozen evidence defines Decomposition Plot as 'Displays observed, trend, seasonal or cyclical, and residual components for review', so its operative form is Interface, Display & Cue.
Nearest alternative: Analysis, Modeling & Optimization — The coordinated visual panels are the operative review surface, while decomposition computation supplies their series.
Review outcome: Independent reviewer agreement; medium confidence.
Origin Attribution¶
Primary origin: Statistics & Experimental Design
Origin pattern: Single lineage
Present-day reach: Multi-domain
Rationale: Time-series statistics cohered aligned decomposition plots displaying observed, trend, seasonal or cyclical, and residual components on a shared axis.
Related originating lineages:
- Data Science & Analytics — Analytic visualization made STL and related decomposition panels routine diagnostic artifacts.
Review resolution: Time-series statistics cohered aligned decomposition plots displaying observed, trend, seasonal or cyclical, and residual components on a shared axis.
Review outcome: Reconciled after independent review; high confidence.
Notes¶
[n1] STL — Seasonal-Trend decomposition using Loess — splits a series into trend, seasonal, and remainder components with locally-fitted smoothers, allowing the seasonal shape to evolve slowly. The stacked four-panel plot (observed, trend, seasonal, remainder) is its canonical output and the archetypal decomposition artifact. ↩