Why choose S rather than R?

Standard deviation uses all observations and is generally more informative than the range as subgroup size increases, although the choice depends on the monitoring design.

Does an ordinary mean cancel an S-chart signal?

No. The two statistics monitor different aspects of behavior.

Can the lower S limit be negative?

A standard deviation cannot be negative, so conventional negative lower-limit calculations are truncated to zero.

AI Glossary / Statistics and probability / Intermediate

X-bar-S chart

Definition

An X-bar-S chart is a paired process-control display that monitors subgroup means and subgroup standard deviations over time. The X-bar part tracks changes in average level, while the S part tracks within-subgroup spread. It is commonly used for continuous measurements when standard deviations provide a suitable summary of variation within rationally formed subgroups.

What it means

The chart combines two views of process behavior. A batch can have an ordinary average but unusually inconsistent measurements, or a shifted average with an ordinary spread. These situations can require different investigations, so displaying both statistics gives a more useful account than a mean alone.

Unlike the range, a standard deviation uses every observation's distance from the subgroup mean. This makes the S approach attractive as subgroup sizes increase. Neither chart proves that output meets customer requirements: stability relative to a baseline is different from capability relative to specifications.

How it works

For equal subgroup size n, compute each mean and sample standard deviation. Let the grand mean be the average subgroup mean and s̄ the average subgroup standard deviation. Under the classical normal-process model, estimate process standard deviation by s̄/c₄, where c₄ depends on n.

The X-bar limits are grand mean ± A₃s̄, with A₃ = 3/(c₄√n). The S-chart limits are B₃s̄ and B₄s̄, where B₃ = max[0, 1 − 3√(1 − c₄²)/c₄] and B₄ = 1 + 3√(1 − c₄²)/c₄. Review spread stability before interpreting mean changes. Establish and document a credible baseline rather than assuming any historical data are automatically stable.

Three practical examples

1. Production measurement: Engineers sample several comparable items during each inspection period. An S-chart signal indicates a change in within-period consistency even when average dimensions remain close to the baseline.

2. Laboratory operations: A team monitors repeat measurements under comparable conditions. The paired display separates shifts in typical reading from increased measurement variability, provided the sampling design matches the process being monitored.

3. AI benchmarking: Developers repeatedly measure a fixed workload under controlled conditions. They evaluate whether standard chart assumptions are defensible before using subgroup means and standard deviations to monitor the runtime process.

Think of it this way

Think of checking both the average landing position and the scatter of repeated throws. The center can remain on target while the throws spread out, or the group can shift without becoming more scattered. X-bar and S show those two features separately, using limits based on a reference process rather than the desired target alone.

Worked example

Consider an original baseline-summary example with subgroups of ten measurements, grand mean 100 units and average subgroup standard deviation s̄ = 2 units. For n = 10, c₄ is approximately 0.972659. This gives A₃ ≈ 0.975350, B₃ ≈ 0.283706 and B₄ ≈ 1.716294.

The X-bar limits are 100 ± 0.975350 × 2, or approximately 98.0493 and 101.9507. The S-chart limits are 0.283706 × 2 ≈ 0.5674 and 1.716294 × 2 ≈ 3.4326. Its center line is two.

A new subgroup has mean 100.5 and standard deviation four. The mean lies within its limits, but the standard deviation exceeds the S-chart upper limit. The spread signal deserves investigation even though the average looks ordinary. The example assumes a suitable baseline has already been established; it does not infer process stability from the supplied summary numbers alone.

Illustrative case study

In this fictional case, Elm Instruments monitors repeated calibration measurements. One inspection period produces an S-chart signal with no corresponding mean signal. Staff review the readings and find inconsistent fixture placement rather than a uniform change in the instrument's level.

They correct the fixture procedure, document the investigation and continue monitoring against the established limits while assessing the effect. The team preserves the original measurements and avoids removing inconvenient points solely to improve the appearance of the chart. A separate specification review determines whether affected measurements are acceptable for the intended application.

Common misconceptions and mistakes

The average of subgroup standard deviations is not generally the same as the standard deviation of all pooled observations. Pooling can mix within-subgroup variation with changes between subgroup means, changing the quantity being estimated.

Do not reuse constants for the wrong subgroup size. Unequal sizes require an appropriate varying-size approach. Strong autocorrelation, mixed populations or unsuitable distributional assumptions can distort signals. A low S value may also deserve investigation, for example if measurement rounding or data duplication has changed. More consistent recorded values are not automatically evidence that the real process improved.

Business use

X-bar-S monitoring can support quality control when decision-makers need to distinguish changes in average performance from changes in consistency. Its effectiveness depends on a stable measurement procedure, rational subgrouping and a defined response to signals.

Record subgroup size, collection timing, operating conditions and any changes to instruments or software. For AI-related measurements, keep task mix and configuration visible so that a chart is not interpreted as a fixed-process monitor after its workload changes. Evaluate specification compliance and operational consequences separately; statistical stability does not establish that the baseline is good enough.

Check your understanding

In the worked example, which statistic signals?

Answer: The subgroup standard deviation of four exceeds approximately 3.4326; the mean of 100.5 is within its limits.

Why is c₄ included?

Answer: Under the normal-process framework it adjusts the relationship between expected sample standard deviation and process standard deviation.

Can a chart replace a measurement-system investigation?

Answer: No. Its signals can motivate that investigation, but they do not identify the mechanism by themselves.

Related terms

X-bar-R chart
Control limits
Rational subgroups
Special cause
Subgroup standard deviation
control limit
rational subgroup
process stability
measurement system
process capability

Learn next

Next, calculate subgroup means and standard deviations from time-ordered data and compare them with a pooled summary. Study how subgroup design affects the interpretation. Then explore individual-observation charts, unequal subgroup sizes and monitoring methods for correlated or strongly nonnormal measurements.

Sources

NIST/SEMATECH e-Handbook, “Shewhart X-bar and R and S Control Charts,” gives the standard-deviation-based limits and c₄ relationship: https://www.itl.nist.gov/div898/handbook/pmc/section3/pmc321.htm

The ten-measurement baseline summary and Elm Instruments case are original teaching illustrations. Constants and limits were independently calculated from the classical normal-process formulas, with a nonnegative lower S limit.

Previous: X-bar-R chart Next: XML database