A c or u chart assumes a defect rate high enough that a Poisson distribution behaves reasonably close to normal — which is the assumption the standard three-sigma control limit math is actually built on. As the true defect rate gets low enough, and zero-defect subgroups become the majority rather than the exception, that assumption quietly breaks down, and the standard chart stops being a reliable detector right when a process needs monitoring most: after a genuine quality improvement has driven defects down toward rare-event territory.
Why Low Counts Break the Math
The Poisson distribution is markedly skewed at low average rates and only approaches something close to symmetric, normal-like behavior as the average count per subgroup climbs into higher ranges — a commonly cited rule of thumb puts that transition somewhere around an average of two or more defects per subgroup. Below that, a standard c chart calculated with the usual center-line-plus-or-minus-three-sigma formula can produce a lower control limit that sits below zero, which gets floored at zero, and an upper limit that's a poor approximation of where a genuinely unusual cluster of defects would actually fall.
The practical result: a process running at a low, stable defect rate will show mostly zero-defect subgroups with occasional single-defect points, and the chart has very little ability to distinguish "the process just had its normal rare defect" from "something changed and defects are now happening more often." Both look almost identical against limits that weren't built for data this sparse.
Time-Between-Events as the Alternative
Rather than counting defects per fixed sampling interval, an alternative approach tracks the time, production count, or number of opportunities between successive defect occurrences — this is the basis for g charts (counting opportunities or units between defective events) and t charts (counting time between events). Instead of asking "how many defects occurred in this subgroup," the chart asks "how long did we go before the next defect," which uses every single unit of good production as information rather than compressing an entire run of zero-defect subgroups into a single unhelpful data point.
This reframing works well specifically because it doesn't waste information on runs of zeros. A process producing one defect every 3,000 units generates a rich, continuously updating signal under a g chart — every unit produced either extends the current run or ends it — while the same process under a c chart with subgroups of 500 units mostly just generates zeros with an occasional one, and detecting a shift in the underlying rate takes many more subgroups to become statistically evident.
Recognizing the Trigger Point
The signal that it's time to reconsider chart type isn't a fixed threshold so much as a pattern: when the majority of subgroups on a c or u chart show zero defects, and the chart has effectively stopped producing meaningful variation to interpret, that's the practical indicator that the process has improved past the point where standard attribute charts add useful information. This is, counterintuitively, a good problem — it means a quality improvement effort actually worked — but it requires switching monitoring approach to keep getting value from ongoing surveillance rather than continuing to run a chart that's lost most of its sensitivity.
Where This Gets Missed in Practice
The most common failure is simply not noticing the transition — a c chart set up when a process was running at a moderate defect rate keeps running unchanged after a successful improvement project drives the rate down by an order of magnitude, and nobody revisits whether the original chart type still fits the new reality. The chart keeps producing points, keeps looking like it's doing its job, and quietly stops being able to detect anything but a fairly large regression.
The second common failure is the reverse: teams reach for rare-event charting methods prematurely, on data that's actually still frequent enough for standard c or u charts to work fine, adding complexity the underlying data doesn't require. The decision should follow the actual observed rate, not a preference for one method over the other.
Building the Right Chart for the Data
Standard c and u charts remain the correct choice across the moderate-to-high defect rate range where the majority of attribute monitoring actually happens. SigmaDesk's attribute control chart builder runs p, np, c, and u charts with a flexible data model, free in the browser, as part of the complete SigmaDesk SPC platform.
A defect rate low enough to make most subgroups read zero isn't a chart that's stopped finding anything to say — it's a chart that's outgrown the assumptions it was built on.