How does it differ from an ordinary time plot?

It groups matching seasonal positions instead of showing the entire sequence in uninterrupted chronological order.

Does it discover the cycle length automatically?

No; the display uses a specified grouping.

Can a box plot replace it?

A seasonal box plot summarizes distributions, but does not retain the same within-season chronological detail.

AI Glossary / Statistics and probability / Advanced

Seasonal subseries plot

Definition

A seasonal subseries plot is a time-series display that groups observations by their position within a recurring cycle, while retaining chronological order within each group. For quarterly data, it shows separate sequences for first, second, third and fourth quarters, commonly with a mean reference line for each sequence.

What it means

The display helps distinguish differences between seasonal positions from changes within the same position over successive cycles. A high third-quarter average and rising third-quarter values are different features; this plot can reveal both.

“Seasonal” need not mean weather seasons. The chosen cycle might be a year, week or day, depending on the measurements and question. The cycle must be specified appropriately; reorganizing data into groups does not itself establish that a meaningful recurring pattern exists.

How it works

Choose the cycle length and correctly identify each observation's position within it. Group matching positions, preserving their order across cycles. Plot the response against time inside each group, using comparable vertical scales. Add group means when useful.

Inspect the relative group levels, within-group movement and unusual observations. Compare the display with the original chronological plot, because rearrangement can make some features easier to see while obscuring others.

Trend may need attention before interpreting seasonal differences. Missing periods also need explicit handling: calendar labels should not silently shift because an observation is absent.

Three practical examples

1. Service demand: group daily request counts by weekday and inspect whether Mondays differ from other days or change across successive weeks.

2. AI operations: group hourly system-load measurements by hour of day to examine recurring demand and changes within particular hours. Use consistent time-zone and daylight-saving conventions.

3. Quarterly planning: compare quarterly sales across years while checking whether an apparent seasonal peak remains similar as the business grows.

Think of it this way

Imagine sorting several calendars into matching positions: all first quarters together, then all second quarters, and so on. Inside each pile, the years remain in order. You can inspect both differences between piles and changes through each pile.

Worked example

Consider fictional quarterly request counts, measured in hundreds, for three years. Year one records 10,20,30,20 from Q1 through Q4. Year two records 14,24,34,24. Year three records 18,28,38,28.

A seasonal subseries plot uses four groups. Q1 contains 10,14,18 in year order. Q2 contains 20,24,28. Q3 contains 30,34,38. Q4 contains 20,24,28. Each group therefore has three points, with year increasing within the group.

The mean reference lines are (10 + 14 + 18)/3 = 14 for Q1, 24 for Q2, 34 for Q3 and 24 for Q4. The Q3 mean exceeds the Q1 mean by twenty hundreds of requests, or two thousand requests.

However, the mean lines do not tell the whole story. Every subseries rises by four hundreds from one year to the next. A chart showing only the four means would conceal that common increase.

The annual means are twenty, twenty-four and twenty-eight hundreds. For a simple descriptive comparison, subtract each year's annual mean from that year's four values. Year one becomes -10,0,10,0. Years two and three produce exactly the same centered values.

In this example, the recurring within-year pattern is stable after removing each year's overall level. The raw subseries reveal increasing levels; the centered values reveal a repeated shape. These are compatible observations, not contradictory conclusions.

This centering is a descriptive operation, not a complete trend model or a demonstration of the mechanism causing demand. It also uses all four quarters of each year. An operational forecast made earlier in the year could not treat that completed-year mean as already observed.

Now suppose year three's Q1 measurement is missing. The remaining observed Q1 mean would be (10 + 14)/2 = 12, rather than fourteen. That lower value would partly reflect the missing high-year observation, not necessarily a change in the underlying seasonal pattern. Keep the missing position visible and report the available count.

The example has three cycles. Its neat structure was chosen to explain the display; it does not establish that real future demand will follow the same pattern. Nor does the graph supply a significance test, confidence interval or validated forecast by itself.

Illustrative case study

Fictional Willow Services initially budgets from the four seasonal means. The subseries plot then shows that every quarter has increased across the three observed years.

The team separates its questions: which quarter tends to be high, and how is the overall level changing? It also checks calendar alignment and the missing Q1 observation before updating its analysis. For forecasting, it evaluates candidate models on later observations that were not used to fit them. This invented case illustrates how a graph can improve the questions asked without supplying a finished prediction model.

Common misconceptions and mistakes

“The four means describe every relevant feature.” Within-season changes and unusual observations may be hidden by averages.

“Any visible separation is statistically significant.” A descriptive plot does not automatically provide inferential calibration.

“Missing data can simply be removed before assigning positions.” That can shift subsequent observations into the wrong seasonal groups.

“A recurring pattern must have a seasonal cause.” Calendar grouping alone does not establish causation.

Business use

Use the plot to inspect recurring workload, demand or performance before choosing a forecasting or monitoring approach. Document the measurement interval, cycle definition, units and any transformation. Compare like periods, especially when calendar changes, reporting practices or operating hours alter exposure.

For AI-generated charts, verify the group assignments and chronological ordering against the source dates. Keep axes comparable and distinguish observations from imputed values. If a pattern changes over time, investigate rather than assuming a single historical seasonal average will remain suitable. The display supports judgment; it does not replace validation of the decision.

Check your understanding

Why is Q3's mean thirty-four?

Answer: Its values thirty, thirty-four and thirty-eight average to thirty-four.

What do the rising subseries show?

Answer: Each corresponding quarter increases by four hundreds per year.

Why can the missing Q1 value distort comparison?

Answer: Its group then represents a different set of years.

Related terms

Time series
Autocorrelation
Missing data
Seasonality
seasonal plot
trend
detrending
seasonal decomposition

Learn next

Plot the raw and year-centered worked values side by side. Explain which features are preserved and which are removed, then repeat after marking the missing Q1 observation explicitly.

Sources

NIST/SEMATECH, Seasonal Subseries Plot, grouping and interpretation: https://www.itl.nist.gov/div898/handbook/pmc/section4/pmc4431.htm

Rob J. Hyndman and George Athanasopoulos, Forecasting: Principles and Practice, second edition, section 2.5: https://otexts.com/fpp2/seasonal-subseries-plots.html

NIST Dataplot, Seasonal Subseries Plot, cycle and starting-position specification: https://www.itl.nist.gov/div898/software/dataplot/refman1/auxillar/seassubs.htm

Previous: Seasonal index Next: Seasonal variation