Which Control Chart Should You Use?

To pick a control chart, answer two questions: is your data measured or counted, and how do you sample it? Measurements go on I-MR, X-bar & R, or X-bar & S charts by subgroup size; counts go on p, np, c, or u charts by defectives-versus-defects and sample size. That is the whole decision.

Choosing the wrong chart is the most common and most avoidable mistake in statistical process control. The wrong chart gives you limits that are too wide, so real problems hide, or too tight, so a stable process screams every shift. The good news is the decision is short and mechanical. Walk the two questions below in order and the chart picks itself, every time.

Which control chart should you use?

Start with the data type, then the sampling. Answer these in order:

1. Is the data variable or attribute? Variable (or continuous) data is measured on a scale: diameter, weight, temperature, torque, time. Attribute data is counted: pass or fail, number of flaws. Variable data carries far more information per sample, so prefer it whenever the gauge gives you a real number.
2. For variable data, how many pieces are in a subgroup? One value at a time points to an I-MR chart. Small subgroups of two to nine consecutive pieces point to X-bar & R. Subgroups of ten or more point to X-bar & S, where the standard deviation replaces the range.
3. For attribute data, are you counting defective units or defects? A unit is defective when it passes or fails as a whole (one verdict per unit). A unit carries defects when it can have several flaws and still be counted (scratches on a panel, errors on a form). Defective units point to p or np; defect counts point to c or u.
4. Is the sample size constant or does it vary? Constant sample size allows the simpler np (defectives) or c (defects) chart. Varying sample size requires p (proportion defective) or u (defects per unit), which recompute the limits for each sample.
5. Check your subgrouping logic. A subgroup must capture only short-term variation within one stream: consecutive parts from one line, not a mix of four cavities or three shifts. Get this wrong and no chart type will save you; the limits will be meaningless.

Variable or attribute: why the first question matters most

The variable-versus-attribute split decides more than which formula you use; it decides how much the chart can see. A measured value tells you not just whether a part passed but by how much, and how the process is trending toward a limit before anything fails. A pass/fail count throws that away. In practice a variables chart can detect a process shift from a handful of parts, while an attribute chart may need hundreds of units per sample to notice the same shift. So the rule of thumb is blunt: if the gauge gives you a number, chart the number. Only drop to attribute charts when the characteristic really is a judgment (present or absent, conforming or not) or when measuring every feature is genuinely impractical. The deeper trade-off is laid out in attribute vs. variable inspection.

That information gap is the reason the first question matters more than any other. Choosing attribute data when a measurement was available does not just change the chart; it blinds the chart, forcing you to inspect far more units to learn far less. The rest of the decision tree only refines a choice you have already half-made by deciding what to record at the gauge.

How do you choose among the variable charts?

Once you know the data is measured, subgroup size is the only lever. If you can only get one value at a time, because the test is destructive, the batch has a single parameter, or the process is slow, use an individuals and moving-range (I-MR) chart. It plots each value and the moving range between consecutive points. It is the safe workhorse for one-at-a-time data, at the cost of being slower to catch small shifts.

If you can pull a small handful of consecutive parts at intervals, use an X-bar & R chart. Averaging two to nine pieces tightens the signal and catches shifts faster than individuals. Once subgroups reach ten or more, the range stops using the data efficiently, so switch to X-bar & S and let the standard deviation track spread. One principle spans all three: every variables chart is really two charts, one for location and one for spread. Read the spread chart first, because if the spread is unstable the location limits cannot be trusted.

How do you choose among the attribute charts?

For counted data, two questions decide it. First, are you counting bad units or bad spots? If each unit is simply good or bad, you are tracking defectives, which follow p and np charts. If a unit can carry several independent flaws, you are tracking defects, which follow c and u charts. Second, does the amount you inspect stay the same each time? A fixed sample size lets you use the count-based np or c chart. A changing sample size forces the rate-based p or u chart, which adjusts the limits for each sample so a bigger sample does not fake a signal. When defects are what you count and the inspection unit is constant, the c-chart is the natural home; when the inspected area changes, move to u.

What are the three most common selection mistakes?

Most bad charts come from three errors, none of them about the math:

• Charting averages on an individuals chart. Plotting daily averages of many parts on an I-MR chart hides the real variation and makes the limits so tight everything signals. Chart the raw values with the right subgrouped chart instead.
• Mixing streams in one subgroup. One part from each of four cavities in a single subgroup blends four processes. Chart streams separately, or subgroup within one stream, so the limits reflect one process's short-term variation.
• Using an attribute chart when a measurement exists. Recording a fill weight as pass/fail discards the number you already took. Keep the measurement whenever the gauge gives you one; an X-bar & R chart sees in five parts what a p chart needs hundreds to notice.

What do the numbers say about control charts?

The method is old, settled, and still the backbone of SPC:

• Walter Shewhart proposed the control chart at Bell Telephone Laboratories in a May 16, 1924 memo and set the limits at three standard deviations as an economic balance between chasing false alarms and missing real trouble (ASQ, Walter A. Shewhart).
• The seven Shewhart chart types split cleanly into three variables charts (I-MR, X-bar & R, X-bar & S) and four attribute charts (p, np, c, u), a taxonomy documented by ASQ.
• Control limits are computed from the process's own data; specification limits never belong on a control chart, because mixing the two turns a process-behavior tool into a pass/fail tally (ASQ, Control Chart).

What comes after you pick the chart?

Picking the chart is step one; running it is the rest. Once a variables chart shows sustained control, the next question is whether the stable process actually fits the spec, which is where Cp and Cpk take over. The control chart is one of the seven basic quality tools and it only pays off if the data reaching it is captured cleanly. That last part is where most plants lose the plot: readings live on paper check sheets that never become chartable data, so the chart is always a week behind the floor. Harmony connects machines, software, and paperwork into one operational layer with no rip-and-replace, so a reading typed at the station becomes structured, chartable data the moment it is written, feeding the same quality trends your reporting needs. CLS made that move for production logging, and the live visibility that follows turns a paper backlog into a signal you can act on the same shift. Pick the right chart, capture the data cleanly, and the statistics do the rest.