Run Charts

Run charts are the quality engineer's first-pass stability check. Before setting up formal control charts with statistically derived limits, a run chart quickly reveals whether the process shows obvious time-based patterns: upward trends (tool wear), sudden shifts (material lot change), cyclic behavior (temperature effects), or clustering (mixed process conditions).

The advantage of run charts over control charts is simplicity. They require no distributional assumptions, no subgroup decisions, and no limit calculations. The analysis is based on counting runs above and below the median — too few runs suggest trends or shifts, too many suggest oscillation. Standard tests (number of runs, length of longest run) provide objective criteria.

Run charts are especially valuable during process development and for preliminary data analysis. They answer the question "is this process reasonably stable?" before you invest effort in fitting distributions and computing capability indices. If the run chart shows a clear trend, no amount of statistical sophistication will produce a meaningful Cpk — you need to stabilize the process first.

EntropyStat's process monitoring capabilities build on the run chart concept by adding distributional context. Where a run chart shows time-ordered data against a simple median line, EntropyStat can overlay EGDF-derived probability bands that indicate how extreme each measurement is relative to the process distribution — without assuming normality.

The EGDF-based approach is particularly valuable when run chart patterns are ambiguous. A sequence of measurements above the median could indicate a genuine process shift or could be within the expected range for a skewed distribution (where more than half the area is above the median). EntropyStat's membership scoring quantifies how "surprising" each run pattern is given the actual process distribution, reducing subjective interpretation.

For trend detection, EntropyStat can compare EGDF shapes across time windows, revealing not just mean shifts (which run charts detect well) but distributional shape changes — such as increasing spread or developing bimodality — that a simple run chart against the median cannot capture.