Quality & InspectionFree referencePDF, Excel, Word export

Control Chart Type Selection Guide

Choosing the wrong control chart is the most common reason SPC fails: limits are too wide, too tight or just meaningless. This guide maps your data type and subgroup size to the correct chart and shows how each chart's limits are calculated.

Download:

Chart	Data type	Subgroup	Center line and limits	Use when
X-bar and R	Variable (measured)	2 to 10, constant	UCL = Xbarbar + A2Rbar; LCL = Xbarbar - A2Rbar; R chart UCL = D4Rbar, LCL = D3Rbar	Rational subgroups of 2 to 9 parts taken close together, such as 5 consecutive parts per hour
X-bar and S	Variable (measured)	10 or more, or varying	UCL = Xbarbar + A3Sbar; S chart UCL = B4Sbar, LCL = B3*Sbar	Larger subgroups where the range loses efficiency, or automated gauging
Individuals and moving range (I-MR)	Variable (measured)	1	UCL = Xbar + 2.66MRbar; MR UCL = 3.267MRbar, MR LCL = 0	One reading per batch, per day or per lot; slow processes, chemical batches, monthly costs
Median and R	Variable (measured)	Small, odd, 3 to 5	Median limits use the average median and Rbar with their own constants	Operators plot by hand and want to avoid arithmetic
p chart	Attribute: fraction defective	Variable n allowed	pbar +/- 3sqrt(pbar(1 - pbar)/n)	Count of nonconforming units per lot or per day, where the sample size changes
np chart	Attribute: count defective	Constant n	npbar +/- 3sqrt(npbar(1 - pbar))	Same as p chart but with a constant sample size and counts easier to read
c chart	Attribute: defects count	Constant area of opportunity	cbar +/- 3*sqrt(cbar)	Defects per unit, such as blemishes per panel, where each inspected unit has equal area
u chart	Attribute: defects per unit	Variable area allowed	ubar +/- 3*sqrt(ubar/n)	Defects per board, per meter or per square foot when the inspected amount varies
Laney p prime and u prime	Attribute with overdispersion	Large, variable	Standard p or u limits scaled by a within-process sigma estimate from moving ranges of z scores	Large subgroups where standard p or u limits flag nearly every point
EWMA	Variable or attribute	1 or more	Weighted average with lambda about 0.05 to 0.25; limits about L = 2.7 to 3 times sigma times sqrt(lambda/(2 - lambda))	Detecting small sustained shifts of about 0.5 to 1.5 sigma faster than a Shewhart chart
CUSUM	Variable	1 or more	Accumulates deviations from target; common design k = 0.5 sigma, decision interval h = 4 to 5 sigma	Detecting small persistent shifts, tracking tool wear or drift against a target
Z-MR or short-run chart	Variable	1	Plot (x - target)/sigma for each part number; apply I-MR limits to the standardized values	High-mix job shops where each part number has a few readings
Three-way (between and within)	Variable	2 or more	I-MR on subgroup means plus a range or S chart for within-subgroup variation	Processes with real between-subgroup variation, for example batch-to-batch plus within-batch
Hotelling T squared	Variable, several correlated	1 or more	UCL from the F distribution on the multivariate distance from the mean vector	Several correlated characteristics monitored together, such as length and diameter from the same cut
Run chart	Any	1	Median line only, no control limits	Early look at a new process before enough data exist for control limits
Pareto chart	Categorical counts	Not applicable	Not a control chart; ranks causes by frequency or cost	Choosing which defect category to attack first, not monitoring stability

Constants A2, A3, D3, D4, B3, B4 and d2 depend on subgroup size; look them up in any SPC table. Always confirm that the subgroup is rational: variation inside it should reflect common causes only.

Put SPC on live ERP data

Optional: a short call about generating control charts from ERP inspection results instead of spreadsheets.

Optional. The tool and every export work without this. No spam.

How to use the guide

Start with the data type. If you measure a dimension, weight or temperature it is variable data; if you count defects or defective units it is attribute data. Then look at the subgroup: how many readings you take at each sampling point. Search the table for variable, attribute, small shift or short run and read the Use when column. Export the list to Excel to paste into a control plan template so every characteristic names its chart type.

Variable data with subgroups of 2 to 9: X-bar and R
Variable data, one reading per point: individuals and moving range
Defectives (units): p or np. Defects (counts): c or u

The logic of each family

Shewhart charts set limits at three standard errors from the center line. For variable data, the standard error is estimated from the within-subgroup range or standard deviation, which is why rational subgrouping matters so much. For attribute data, the binomial model drives p and np charts and the Poisson model drives c and u charts. EWMA and CUSUM are not replacements but complements: they remember previous points and respond to small shifts that a Shewhart chart takes many points to notice.

Worked example

An injection molding cell measures one shot weight every 15 minutes: that is individuals data, so use I-MR. If the cell instead pulls five consecutive shots every hour, use X-bar and R with n = 5, where A2 = 0.577, D3 = 0 and D4 = 2.114. If the inspector counts rejected parts per lot with lot sizes varying from 200 to 800, use a p chart because the sample size changes, and the limits will move with each lot.

Common mistakes

Teams often use X-bar and R with subgroups that mix machines or shifts, which inflates the within-subgroup variation and hides real signals. Others use a c chart when the inspected area varies, where a u chart is correct. Using specification limits as control limits is another classic error: control limits describe the voice of the process, specs describe the voice of the customer.

Do not mix streams in one subgroup
Use u instead of c when the area of opportunity changes
Never draw spec limits on an X-bar chart as if they were control limits

Frequently Asked Questions

What is the difference between a defect and a defective?

A defect is a single nonconformity, and one unit can have several. A defective is a unit with at least one defect. Count defectives with p or np charts, and count defects with c or u charts.

When should I use EWMA instead of X-bar?

Use EWMA when the process is stable and you care about small shifts, roughly half to one and a half standard deviations, that matter commercially. Keep a Shewhart chart alongside it for large shifts, since EWMA reacts more slowly to them.

Can I use a control chart with non-normal data?

Shewhart charts for subgroup means are fairly robust because of the central limit theorem. Individuals charts are less robust. For skewed data consider a transformation, a chart based on the right distribution, or a time-between-events chart.

How many points do I need to set limits?

A common guideline is at least 20 to 25 subgroups collected while the process runs normally. Fewer points give unreliable limits, so treat them as trial limits and review once you have more data.

If you want SPC charts generated straight from ERP or MES measurement tables instead of spreadsheets, Netray can scope that with you.

Talk to an Expert Explore More Free Tools

Related Tools

Quality & Inspection

More Quality & Inspection tools

See all

Back to All Tools