AI Glossary / Statistics and probability / Intermediate

Digidot chart

Definition

A digidot chart combines a stem-and-leaf display with a time-order plot of the same numerical observations. In the arrangement described by NIST, the stem-and-leaf display sits along the vertical scale with leaves extending left, while observations are plotted in time order to the right. It lets a reader inspect the distribution of values and their sequence together. It is an exploratory display, not a statistical test or a control chart by definition.

What it means

Two datasets can contain exactly the same values but tell different stories when placed in different orders. A distribution display shows where observations concentrate and how widely they vary. A time plot shows when high and low values occur, potentially revealing shifts, trends or alternating patterns. Bringing these views together helps prevent a pooled distribution from concealing changes over time. The name refers to the statistical chart in this entry, not to unrelated products or digital marking technologies with similar names.

How it works

Collect a sequence of numerical observations with an order or timestamp. Choose a stem-and-leaf scale that represents the desired significant digits and state its key. Arrange the stems along a vertical measurement scale with leaves to the left, and use the corresponding scale for the time plot to the right. Plot every observation at its sequence position or actual timestamp. Keep the chronological values in their original order even if leaves are sorted within stems. Label units, rounding and the time axis. Read the marginal distribution and sequence jointly; if the data are extensive or have many digits, separate aligned plots may be clearer.

Three practical examples

• Process review: display measured component dimensions and their production order to investigate whether a pooled spread hides a shift.
• Service monitoring: inspect response times from a short sequence of comparable requests alongside their value distribution.
• AI testing: review latency measurements across successive evaluation runs while checking whether changes coincide with recorded test conditions.

Think of it this way

Think of reading both a basket's contents and the order in which the items were collected. The inventory tells you what is present, while the collection record tells you when it appeared. Neither view substitutes for the other, and combining them can reveal a pattern hidden by the inventory alone.

Worked example

Suppose eight hypothetical observations, in order, are 21, 22, 21, 23, 29, 30, 31 and 30 milliseconds. With tens as stems and units as leaves, stem two contains leaves 1, 1, 2, 3 and 9, while stem three contains 0, 0 and 1. These eight leaves account for all eight observations. The right-hand time plot preserves the original order and shows that the last four values are all above the first four. The first-half mean is 87/4 = 21.75 milliseconds and the second-half mean is 120/4 = 30 milliseconds; the pooled mean is 207/8 = 25.875. Reordering the same values would leave the stem-and-leaf display unchanged but alter the time pattern. The visible separation suggests a question about changing conditions, not proof of a specific cause or a formal significance result.

Illustrative case study

Fictional case: Elm Compute reviews a short set of AI inference timings after noticing variable service performance. A digidot chart shows two concentrations of values, while the sequence side places the slower measurements near the end. Analysts check request sizes, caching, hardware load and the experiment log before attributing the pattern to a model update. They repeat the measurements under documented conditions. This invented case illustrates exploratory diagnosis and follow-up; it does not establish that the display alone identifies a bottleneck or measures a lasting performance regression.

Common misconceptions and mistakes

Do not sort the time-series values to make the line smoother: doing so destroys the very ordering information that distinguishes this display. A cluster of values in the stem-and-leaf portion is not automatically evidence of a stable distribution. A line connecting points does not prove that intermediate values occurred. If observations are irregularly spaced, an index axis and a real-time axis convey different information. The basic chart supplies no calibrated false-alarm rate, specification limits or causal explanation. Rounded digits should not be presented as preserving precision that the display has discarded.

Business use

Businesses can use digidot charts for compact reviews of modest datasets where both values and order matter. They are especially useful when teams would otherwise see only a mean or histogram. Their value comes from prompting targeted questions and improving communication, not from replacing a full monitoring design. Keep raw data available, identify any missing measurements and link unusual sequences to contemporaneous operating records. For large-scale dashboards, evaluate whether more familiar coordinated distribution and time-series views offer better readability for the intended audience.

People also ask

Is a digidot chart just a dot plot?

No. Its defining combination is a stem-and-leaf display and a time-order plot of the same data.

Can it establish that a process is statistically controlled?

Not by itself. That requires an appropriate monitoring method and assumptions.

Does the left-hand display retain original values?

It retains the digits represented by the selected stem-and-leaf scale; rounding or truncation can reduce precision.

Check your understanding

What happens if the eight example values are randomly reordered?

Answer: Their stem-and-leaf display, pooled mean and overall range stay the same. Their positions on the time plot change, potentially removing the apparent early-to-late separation. This demonstrates why the distribution alone cannot describe the sequence.

Related terms

• Stem and leaf plot
• Exploratory data analysis
• Histogram
• Control chart
• Data visualization
• run-sequence plot
• time-series plot
• process shift
• rounding
• measurement scale

Learn next

construct a stem-and-leaf display with an explicit key and compare it with a chronological plot of the same observations. Shuffle their order and note which features change. Then learn how control charts, autocorrelation checks and contextual investigation add evidence beyond exploratory visual patterns.

More in Statistics and probability

• Dichotomous variables - Dichotomous variables are variables with two possible categories or values.
• Difference scores - Difference scores are values obtained by subtracting one measurement from a corresponding second measurement.
• Differential entropy - Differential entropy is an information-theoretic quantity defined for a continuous probability distribution with a density.
• Direct sampling - Direct sampling, in the OECD glossary's survey-statistics sense, means that sampled units are actual members of the population being studied and information...
• Direct utility estimation - Direct utility estimation is a reinforcement-learning approach that estimates a state's value by averaging the observed returns following visits to that state.
• Directional test - A directional test is a statistical hypothesis test whose rejection region lies in one specified direction, commonly called a one-sided or one-tailed test.

Sources

NIST/SEMATECH, glossary, digidot chart: https://www.itl.nist.gov/div898/handbook/glossary.htm NIST/SEMATECH, Run-Sequence Plot: https://www.itl.nist.gov/div898/handbook/eda/section3/eda33p.htm NIST Dataplot, Stem and Leaf Plot: https://www.itl.nist.gov/div898/software/dataplot/refman1/ch2/stemleaf.pdf

Previous: Diffusion model Next: Digit recognition