Moving Average Chart
Control chart that monitors the moving average of consecutive observations to smooth process variation
A Moving Average Chart monitors the average of a fixed number of consecutive observations (moving window) over time. This smoothing technique helps reduce the effect of random variation and makes it easier to detect sustained shifts in the process mean.
Chart Overview & Applications
Moving Average & Control Limit Formulas
Moving average of window size w
= individual observation at time t
= window size (span)
Control limits for the moving average
= process mean
= process standard deviation
= window size
Averaging w points reduces the standard error by a factor of sqrt(w), tightening the limits versus an individuals chart.
Key Assumptions
- • Underlying individual observations are independent and normally distributed
- • Window size w is chosen consistently and documented (2-5 for responsiveness, 5-10 for smoothing)
- • Moving averages only begin once w observations are available, so the first w-1 points have no plotted average
Reading the Moving Average Chart
Example Chart & Signal Interpretation
The example chart above plots a 5-point moving average of a noisy process; the smoothing reveals a gradual upward drift starting around the midpoint that individual observations alone would obscure.
In-Control Signals
- • Moving average stays within UCL/LCL
- • No sustained trend or drift
- • Random fluctuation around the center line
Out-of-Control Signals
- • Moving average crosses UCL or LCL
- • Sustained trend across several windows
- • Persistent run on one side of the center line
Implementation on sixsigmacenter.com
Example Data Layout for Moving Average Chart
Individual measurements with moving averages of different window sizes. Moving averages start when sufficient data points are available. Sample Size Recommendation: Minimum 20-30 baseline measurements to establish control limits, with moving average window size typically 2-5 points for effective smoothing.
This is how your spreadsheet or CSV file should be structured before you upload it. Match the column names, keep one row per observation, and leave the analysis to the platform.