Control Chart vs Run Chart: When the Median Line Is Enough and When You Need Control Limits
You have two weeks of measurements off a new line and a manager asking whether the process is stable. Plotting the points over time is the obvious move — but do you draw a median line and call it a run chart, or calculate control limits and call it a control chart? With fourteen data points, the control chart vs run chart decision changes what you are allowed to conclude.
The control chart vs run chart question comes up constantly because the two look almost identical: dots connected over time with a horizontal line through the middle. The difference is what that middle line means and whether the chart carries statistically calculated limits. Get it wrong and you either claim control you cannot support or miss signals you should have caught.
What Each Chart Actually Is
A run chart is the simpler of the two. It plots a measurement over time, connects the points, and draws a median as the center line. That is the whole structure — no upper or lower limits. You read it by looking for non-random patterns about the median: long runs on one side, trends, and obvious shifts.
A control chart plots the same time-ordered data but draws the mean as the center line. It also adds an upper and lower control limit, typically at ±3 standard deviations from that mean. Those limits are calculated from the process data itself, not chosen. The control chart is the foundational tool of statistical process control because the limits turn "that looks high" into a defined statistical signal.
The Center Line Tells the Whole Story
The choice of center line is not cosmetic. A run chart uses the median because the median is robust — it does not need the data to be stable or normally distributed to be a sensible middle. That is exactly why a run chart is safe to use early, before you know whether the process is behaving.
A control chart uses the mean and three-sigma limits, and both assumptions carry weight. The limits are only meaningful if the process is stable enough that the standard deviation estimate represents real common-cause variation. Calculate limits from an unstable process and you bake the instability into the limits, which then fail to flag the very problem you are looking for. This is why control limits need history: roughly 25 or more subgroups before the limits are reliable.
What Each Chart Can Detect
A run chart detects patterns through simple run rules — too many consecutive points on one side of the median, or a trend of steadily rising or falling values. These tests catch gross shifts and trends without any statistical limit. What a run chart cannot do is separate common cause from special cause variation, because it has no statistical boundary for "normal."
A control chart adds that boundary. A point beyond a control limit, or a pattern that violates the Western Electric or Nelson rules, is a defined special-cause signal that warrants investigation. The control chart is what lets you decide, with statistical backing, whether to leave the process alone or hunt for an assignable cause. Reacting to ordinary noise as if it were a signal — tampering — is the mistake control limits are designed to prevent.
One caution applies to both charts: control limits are not specification limits. A point inside the control limits can still be out of specification, and vice versa. The chart describes what the process does, not what the customer requires.
Control Chart vs Run Chart: Side-by-Side
| Criterion | Run Chart | Control Chart |
|---|---|---|
| Center line | Median | Mean |
| Limits | None | UCL / LCL at ±3 sigma |
| Data needed | Works with few points (even under 20) | ~25+ subgroups for reliable limits |
| Assumes a stable process? | No | Limits assume common-cause-only baseline |
| Detects | Runs, trends, shifts about the median | Special-cause signals via control limits + pattern rules |
| Separates common vs special cause? | No | Yes |
| Typical use | Early analysis, rapid improvement tracking | Ongoing monitoring, capability, audits |
| Effort to build | Low — plot and draw the median | Higher — calculate limits, pick chart type |
The Run Chart Is a Precursor, Not a Lesser Chart
Quality engineers treat the run chart as the precursor to the control chart, and that ordering is deliberate. Early in a Lean or Six Sigma project — the Measure phase of DMAIC — you often have a handful of points and no idea whether the process is stable. A run chart lets you see the data honestly without pretending you can calculate trustworthy limits yet.
It is also the better chart when you are driving a rapid improvement and want to watch the effect of a change land. You are not monitoring a stable process; you are deliberately moving it, so three-sigma limits calculated from the old behavior would be misleading. The run chart shows the shift cleanly.
Once the process settles and you have accumulated enough subgroups, you graduate to a control chart. From there you choose the right chart type — an X-bar and R chart for subgroups, an I-MR chart for individuals, or an attribute chart for pass/fail data. The run chart got you to the point where the limits mean something. That sequence — explore with a run chart, then control with a control chart — is standard statistical process control practice.
Verdict by Where You Are
| If you... | Use | Because |
|---|---|---|
| Have fewer than ~20–25 data points | Run chart | Not enough history for reliable control limits |
| Are in the early Measure phase of a project | Run chart | You do not yet know if the process is stable |
| Are tracking a deliberate, rapid improvement | Run chart | Old-process limits would mislead during a change |
| Monitor an established, stable process | Control chart | Limits flag special causes before they make scrap |
| Need to separate common vs special cause | Control chart | Only control limits define that boundary |
| Need capability or audit evidence | Control chart | ISO 9001 and IATF 16949 expect control + capability data |
The short version: a run chart asks "is anything obviously changing over time?" and a control chart asks "is this process in statistical control?" Start with the run chart when you are early or short on data, and move to the control chart once the process has enough history for the limits to mean something. When you reach that point, you can plot your data and let the tool calculate the control limits and flag out-of-control signals instead of computing three-sigma limits by hand.