Autocorrelation and SPC: When Time Order Changes the Story
Learn what autocorrelation means, why time order matters in SPC, and how serial dependence can change the way you interpret control charts.
Why does a stable process sometimes look like it is drifting?
Sometimes the process has memory. When one reading is partly connected to the previous one, the chart can show long smooth patterns even when nothing dramatic changed.
What autocorrelation means
Autocorrelation means nearby values are related in time. Instead of each point being mostly new information, part of one point carries into the next point.
Mostly independent process
Each new point moves up or down with only a small connection to the previous point. Routine variation looks more random.
Autocorrelated process
Each point partly follows the last one. That creates smoother runs, waves, and trends because the process has time-related memory.
Same average, different time behavior
These two charts have similar center and spread, but the second chart has much more point-to-point memory.
Mostly independent measurements
Positively autocorrelated measurements
Why SPC cares
Control charts are easiest to interpret when each point brings mostly fresh information. Strong serial dependence can change what common-cause behavior looks like on the chart.
Points can stay on the same side of the center line for longer, making rule-based signals more likely.
A smooth rise or fall may reflect process memory rather than a sudden external disruption.
If you react to every smooth-looking pattern without understanding the physics, you can make the process worse.
Autocorrelation often points to real process dynamics such as temperature, wear, or filtering. That is useful engineering information.
Common sources of autocorrelation
Autocorrelation often comes from the process itself, not from bad charting.
| Example source | Why it creates memory |
|---|---|
| Temperature-controlled equipment | Heat changes slowly, so one reading influences the next few readings. |
| Chemical or coating processes | Material conditions evolve gradually instead of resetting instantly between samples. |
| Tool wear or slow drift | The process changes a little at a time, so neighboring points tend to move together. |
| Software smoothing or filtered gages | Averaging filters reduce abrupt jumps and create smoother sequences. |
| Sampling too close together | When samples are taken back-to-back, the process often has not had time to change much. |
What to do in practice
- Study the process physics -- If serial dependence is expected, include that in your interpretation before calling it a special cause.
- Review your sampling interval -- Spacing samples farther apart can reduce memory when the process changes slowly over time.
- Avoid overreacting to smooth patterns -- Not every long run or trend means a sudden assignable cause. Check whether the process naturally carries momentum.
- Use engineering context with chart rules -- Chart rules are useful, but they work best when paired with process knowledge and a sensible sampling plan.
Try It Yourself
Want to check YOUR process capability? Paste your data into our free tool.
Open SPC Quick CheckStill have questions? Ask QC-Coach
Tap a question below or type your own
QC-Coach
AI quality coaching assistantGet personalized AI insights on your analysis. QC-Coach explains your results, identifies issues, and suggests next steps.
Found this useful? Share it with a colleague.