How to Choose the Right Control Chart: Decision Tree

By Hélène · May 12, 2026

Picking the wrong control chart is one of the most common SPC mistakes in manufacturing. The math behind each chart assumes a specific data type and subgroup structure. When the chart and the data do not match, the control limits are wrong, the signals are wrong, and the team either chases ghosts or misses real shifts. Getting the choice right takes about thirty seconds with a clean decision tree and a basic understanding of what your data actually represents.

This guide is the decision tree, with worked manufacturing examples for each branch. It is written for quality engineers, process engineers, and supervisors who already understand the basics of SPC and need a reliable map from data type to chart type.

The decision tree, in one pass

Before any math, answer four questions in order:

1. Is the data variable (measured) or attribute (counted)?
2. If variable, what is the subgroup size?
3. If attribute, are you counting defective units or counting defects per unit?
4. Is the sample size constant across periods, or does it vary?

The answers map cleanly to a chart.

Variables data: continuous measurements

Variables data is anything you measure on a continuous scale: diameter, weight, temperature, pressure, fill volume, torque, surface finish. The chart choice depends on subgroup size.

Subgroup size of 1 - use I-MR

When you collect one reading at a time, use the Individuals and Moving Range (I-MR) chart. The Individuals chart tracks each measurement; the Moving Range chart tracks the absolute difference between consecutive readings.

Common manufacturing cases:

• Continuous chemical processes where each measurement represents the whole batch.
• Slow-cycle measurements where collecting a subgroup is impractical.
• Inline gauging where one reading per cycle is the natural unit.

Consider a heat treat furnace recording one temperature per cycle. I-MR is the right choice because there is no rational subgroup; every reading is independent.

Subgroup size of 2 to 8 - use Xbar-R

The classic SPC chart. Xbar tracks the subgroup mean; R tracks the subgroup range. Use it when you can collect a small rational subgroup at each sampling interval.

Common manufacturing cases:

• CNC machining where you measure 5 consecutive parts each hour.
• Injection molding where you sample 3 parts per cycle.
• Filling lines where you measure 4 bottles per minute.

The R chart catches changes in short-term variation. Always read the R chart first; if range is unstable, the Xbar limits are not trustworthy. Our walkthrough on building an SPC dashboard in Excel shows the full structure with formulas.

Subgroup size of 9 or more - use Xbar-S

When subgroups are large enough that the range becomes a poor estimator of variation, switch to the standard deviation chart. Xbar-S replaces R with S (sample standard deviation) for tighter control limit estimates.

Common manufacturing cases:

• Automated test cells where 20 parts per hour are normal.
• Continuous monitoring with high sampling rates.
• Pharmaceutical or food processes with mandated large sample sizes.

The choice between Xbar-R and Xbar-S is not a hard line; many teams use Xbar-R up through n equals 10 by convention. The math is more efficient with S as n grows.

Attribute data: counts and pass/fail

Attribute data is anything you count: defective units, defects per unit, scrap counts, holds, customer complaints. The first split is whether you are counting defective units or counting defects.

Defective units (yes/no per unit)

A unit is either acceptable or not. You count the number of defective units in a sample. There are two charts:

• p chart: percent (or proportion) defective. Use when sample size varies between subgroups.
• np chart: number defective. Use only when sample size is constant.

Common manufacturing cases for p and np:

• Final inspection pass/fail rates by shift.
• Supplier lot acceptance rates.
• First-pass yield on assembly lines.

If your daily inspection sample size varies (some days you inspect 200 units, others 350), use the p chart. The np chart simplifies the math but only works when sample size is locked.

Defects per unit (multiple defects possible per unit)

A unit can have multiple defects. You count the defects, not the defective units. There are two charts:

• c chart: count of defects. Use when the inspection unit (the area of opportunity) is constant.
• u chart: defects per unit. Use when the inspection unit varies in size.

Common manufacturing cases for c and u:

• Defects per painted panel (c chart if every panel is the same size).
• Defects per square meter of fabric (u chart, because rolls vary in length).
• Solder defects per circuit board (c chart if boards are identical, u chart if they vary).

In quality reviews, this is where teams most often pick the wrong chart: they use a p chart for defect counts when they should use c or u. The result is incorrect control limits and false signals.

Quick reference table

Three worked examples

Example 1: CNC shaft diameter

The team measures 5 shafts every hour. Diameter is variables data, subgroup size 5. Choice: Xbar-R. Read the R chart first to confirm short-term variation is stable, then judge the Xbar chart for shifts in the process mean.

Example 2: Final inspection yield by shift

Each shift inspects between 180 and 320 units. Each unit is pass or fail. Attribute data, defective units, varying sample size. Choice: p chart. Plot the proportion defective per shift with control limits that adjust for sample size.

Example 3: Solder defects on PCB assembly

Boards are identical in size and complexity. Each board may have multiple solder defects (cold joints, bridges, missing components). Attribute data, defect counts, constant inspection unit. Choice: c chart. Plot defect count per board with constant control limits.

Common mistakes

• Using a p chart for defect counts. p charts assume each unit is binary; they cannot handle multiple defects per unit.
• Using Xbar-R when subgroup size is 1. The range cannot be calculated from a single reading.
• Reading the Xbar chart before the R chart. R instability invalidates Xbar limits.
• Using arbitrary spec limits as control limits. Spec limits and control limits answer different questions.
• Ignoring rational subgrouping. If your subgroup spans two operators, you cannot detect operator effects as special causes.

If your team is layering run rules on top of the chart choice, our guide on control chart rules in medical device manufacturing covers which rules add value and which mostly add noise. For the broader interpretation work that turns charts into action, our structured root cause approach walks through the next step.

Practical action block

What to check before picking a chart:

• Is the data continuous or counted?
• If counted, are you counting defective units or defects?
• Is the sample size or inspection unit constant?
• What is the natural rational subgroup for this process?

What practitioners should verify before publishing the chart:

• Did you compute control limits from a stable baseline period (typically 20 to 25 subgroups)?
• Are the limits locked, not recalculated on every refresh?
• Does the team know which run rules trigger a response?

If you are also reporting capability alongside the chart, our breakdown of Cp, Cpk, Pp, and Ppk covers how chart choice and capability calculation interact.

Key Takeaways

• Variables data uses I-MR (n=1), Xbar-R (n=2 to 8), or Xbar-S (n=9 or more).
• Attribute data uses p or np for defective units and c or u for defect counts.
• p and u handle varying sample sizes; np and c require constant sample sizes.
• Always read the R or S chart before the Xbar chart.
• Lock control limits to a stable baseline; do not recalculate on every new data point.