Which SPC Control Chart Should You Use? A Practical Decision Guide

You’ve decided to put a characteristic under statistical process control. Good. Now someone asks: which chart?

If you’ve spent any time with SPC literature, you’ve seen the flowcharts. “Is your data continuous or attribute? Is your subgroup size greater than 1? Is your sample size constant?” They’re not wrong, but they can feel abstract when you’re staring at a real process and real measurement data.

This guide walks through the decision the way a quality engineer actually makes it — starting with what you’re measuring and what question you’re trying to answer.

---

The one question that matters first: what kind of data do you have?

Every chart selection starts here. Your data is either variables data (a measurement on a continuous scale — a dimension in millimeters, a torque in Newton-meters, a fill weight in grams) or attribute data (a count or classification — how many units were defective, what fraction failed inspection).

Variables data gives you more information per measurement. A dimension that’s 0.003 mm over nominal tells you something useful. A pass/fail result just tells you it failed. That’s why, when you have a choice, variables charts are preferred — they detect process shifts earlier, with smaller sample sizes.

If you’re measuring something you can put a number on, you’re in variables territory. If you’re counting defects or classifying units as conforming or not, you’re in attribute territory.

---

Variables data: Xbar-R or Individuals/MR?

For variables data, the next question is: how many measurements do you take per time period?

Xbar-R chart: two or more measurements per subgroup

The Xbar-R chart is the workhorse of manufacturing SPC. It plots two things simultaneously: the subgroup mean (Xbar) and the subgroup range (R). The mean chart tells you where your process is centered; the range chart tells you whether your within-subgroup variation is stable.

Use Xbar-R when:

• You take 2 to 8 measurements per subgroup (subgroup sizes of 3 to 5 are most common in parts manufacturing)
• Those measurements represent a rational subgroup — meaning they were produced close together in time, so within-subgroup variation reflects only common cause, not process shifts
• Your data is continuous (dimensions, weights, temperatures, pressures)

A 90-person injection molder measuring cavity dimensions on a 4-cavity tool is a natural Xbar-R candidate: four measurements per cycle, one subgroup per cycle, the chart separates within-cavity variation from shift-to-shift drift.

What “rational subgroup” means in practice: A subgroup should answer the question “was the process stable during this short window?” If you pull five parts from a two-hour production run and measure them, you’ve created a subgroup that spans two hours of potential process drift — that’s not rational. Pull five consecutive parts from the same setup, same operator, same material lot. Now within-subgroup variation reflects only the process’s natural short-term noise, and the Xbar chart can detect when something changed between subgroups.

Individuals/MR chart: one measurement per time period

When you can only get one measurement per time period, the Individuals and Moving Range (I-MR) chart is your tool. It plots each individual measurement and the moving range between consecutive measurements.

Use Individuals/MR when:

• You produce one unit at a time and measuring more isn’t practical (batch chemical processes, low-volume machined components, destructive tests)
• Your process is slow enough that taking multiple measurements per subgroup would span a meaningful time gap
• You’re monitoring a process parameter rather than a product characteristic (oven temperature, bath concentration, cycle time)

A 40-person contract manufacturer making custom titanium implants might run one part per setup. There’s no subgroup to form — each part is its own data point. The I-MR chart is the right tool.

One note: the I-MR chart is more sensitive to non-normal distributions than the Xbar-R chart. The central limit theorem smooths out subgroup means even when individual measurements aren’t perfectly normal; individual measurements don’t get that benefit. If your data is heavily skewed or bounded (like a measurement that can’t go below zero), be aware that the standard 3-sigma control limits may generate false signals.

---

Attribute data: p, np — and when each applies

Attribute charts are for when you’re counting. The two most common in manufacturing are the p chart and the np chart. They’re closely related — the difference is whether your sample size varies.

np chart: constant sample size, counting defectives

The np chart plots the number of nonconforming units in each subgroup. It requires a constant sample size — the same number of units inspected every time.

Use np when:

• You inspect the same number of units every period (50 units per shift, 100 per lot)
• You’re counting whole units that are either conforming or nonconforming (not counting defects per unit)
• Reporting a raw count is more intuitive than reporting a proportion

A food manufacturer doing end-of-line visual inspection on 200 units per hour, classifying each as pass or fail, is a natural np chart candidate — as long as the 200-unit sample size stays constant.

p chart: variable sample size, proportion defective

The p chart plots the proportion of nonconforming units (defectives / total inspected). Because it normalizes by sample size, it handles variable lot sizes.

Use p when:

• Your sample size varies from period to period (different lot sizes, variable inspection volumes)
• You want to compare defect rates across periods with different inspection volumes
• You’re already thinking in terms of proportions (a 2% defect rate is more meaningful to your team than “14 defectives”)

One practical note: attribute charts require larger sample sizes to be statistically useful. A rule of thumb is that your average count (np̄ for np charts, or n × p̄ for p charts) should be at least 5 per subgroup. If your defect rate is 1% and you’re only inspecting 50 units, your average count is 0.5 — far too low for the chart to detect anything meaningful. You’d need to inspect at least 500 units per subgroup, or reconsider whether attribute data is the right choice for this characteristic.

---

Two more chart types worth knowing

Run chart: trend detection before you have enough data for control limits

A run chart plots individual measurements over time against a centerline (the median), without control limits. It’s not a control chart in the strict sense — it won’t detect the specific violation patterns that Western Electric rules identify — but it’s useful when you’re early in data collection and don’t yet have enough subgroups to compute meaningful control limits.

Use a run chart when:

• You’re in the early stages of monitoring a new process or characteristic
• You want to see directional trends without committing to control limits based on sparse data
• You need a simple visual for an operator who doesn’t need to interpret UCL/LCL lines yet

Once you have enough data (typically 20–25 subgroups), recompute and switch to the appropriate variables or attribute chart.

Histogram: distribution shape, not time-ordered stability

A histogram shows the distribution of your measurements — where they cluster, how wide the spread is, whether the shape is roughly normal. It doesn’t show time order, so it can’t detect process drift or special-cause signals the way a control chart can.

Use a histogram when:

• You want to see whether your process is centered relative to spec limits
• You’re evaluating whether your data is approximately normal before interpreting capability indices
• You’re presenting process performance to someone who finds time-series charts harder to read

The histogram is a complement to control charts, not a substitute. A process can look centered on a histogram while hiding a drift that only shows up when you plot measurements in time order.

---

A quick decision summary

Your situation	Chart to use
Continuous data, 2–8 measurements per subgroup	Xbar-R
Continuous data, 1 measurement per time period	Individuals/MR
Attribute data (pass/fail), constant sample size	np chart
Attribute data (pass/fail), variable sample size	p chart
Early data collection, no stable limits yet	Run chart
Distribution shape and spec centering	Histogram

---

What about capability indices — Cp, Cpk, Pp, Ppk?

Capability indices tell you whether your process, when stable, can consistently produce within spec. They’re calculated from your control chart data, not a separate analysis.

A quick distinction:

• Cp and Cpk measure short-term (within-subgroup) capability. They answer: “If only common-cause variation were present, could this process hold the spec?” Cp ignores centering; Cpk accounts for it.
• Pp and Ppk measure long-term (overall) capability using total observed variation. They answer: “Over the full history we’ve collected, how well did this process actually perform?”

Cp/Cpk are only meaningful when you have a rational subgroup basis — a defensible reason to believe your within-subgroup variation represents the process’s true short-term noise. Without that, Cp/Cpk can be misleadingly optimistic. Pp/Ppk make no such assumption; they use the overall standard deviation from all measurements.

A general target: Cpk ≥ 1.33 for established processes, Cpk ≥ 1.67 for critical characteristics or new processes. These aren’t universal — your customer or your regulatory context may specify different thresholds.

---

One thing you can do today

Pull one characteristic from your control plan that you’re currently monitoring with a run chart or a simple spreadsheet trend line. Ask:

1. Is this continuous or attribute data?
2. How many measurements do I take per time period?
3. Do I have 20–25 subgroups yet?

If the answer to (3) is yes, you have enough data to compute control limits and switch to the appropriate chart. If not, keep the run chart running and set a reminder to revisit when you hit that threshold.

The chart type doesn’t matter nearly as much as the discipline of collecting data consistently, in rational subgroups, and actually reviewing the chart when a signal fires. A well-maintained Individuals/MR chart beats a neglected Xbar-R chart every time.