The PDSA cycle — and why the Study phase is where most improvement fails
Plan, Do, Study, Act. The PDSA cycle is the operating mechanism of every serious improvement programme. Most organisations run the Plan and Do phases reasonably well. Almost all of them fail in the Study phase — because they use measurement tools that cannot tell the difference between genuine structural change and noise. Bootstrap CUSUM is the tool that makes the Study phase honest.
What is the PDSA cycle?
The PDSA cycle — Plan, Do, Study, Act — is the improvement cycle developed by Walter Shewhart, refined by W. Edwards Deming, and formalised by Langley et al. in The Improvement Guide (1996). It is the operating framework for improvement in healthcare, manufacturing, education, and public services worldwide. The NHS uses it extensively through the Model for Improvement.
The cycle is deliberately simple. Its power is not in its complexity but in the discipline it imposes: no change without a prediction, no result without a measure, no conclusion without evidence. Each cycle is small. Each cycle builds on the learning from the last. Improvement compounds over time through repeated cycles rather than through single bold interventions.
Define the objective. State the prediction. Specify who does what and when. Decide how data will be collected. The plan is a commitment to learning — not a commitment to the change being right.
Carry out the change at small scale. Document what actually happens, including surprises. Begin collecting the data specified in the plan. Resist expanding the change before the Study is complete.
Analyse the data against the prediction. What worked? What did not? What was unexpected? This is the most important stage and the most frequently skipped. Bootstrap CUSUM belongs here.
Decide what to do next based on evidence: adopt and expand, adapt and retest, or abandon and try a different theory. Three options — all valid. Only one of them is chosen without evidence.
The prediction discipline — the Plan phase done properly
The most important thing about the Plan phase is writing a specific prediction before implementing the change. Not a goal. Not an aspiration. A testable prediction with a number, a direction, and a timeframe.
Without a prediction, the Study phase has nothing to study against. Any result can be rationalised as success or explained away as special cause. The prediction is what makes the Study phase meaningful — because it commits you to what you expected to see before you see what actually happened.
A prediction names the change, the expected effect, the size of the effect, the timeframe, and the measurement method. Without all five elements it is not a prediction — it is a hope.
“By what method? It is not enough to do your best. You must know what to do, and then do your best.” — W. Edwards Deming
Deming’s question — by what method? — is the Plan phase in one sentence. If you cannot describe the mechanism by which the change will produce the improvement, the prediction is not a theory. It is a guess. A PDSA cycle built on a guess produces noise, not learning.
The Study phase asks one question: did the process actually change, or are we seeing variation around an unchanged mean? Most improvement practitioners answer this question with a run chart, a before-and-after comparison, or a RAG (Red-Amber-Green) status rating. None of these can reliably distinguish genuine structural change from noise. A RAG rating tells you whether a metric is above or below a threshold at a single point in time — it says nothing about whether the underlying process has structurally changed.
A run chart can show that recent data points are higher or lower than previous ones. It cannot tell you whether the underlying process mean has genuinely shifted — or whether you are seeing random fluctuation that will revert. A before-and-after comparison shows you two numbers. It cannot tell you whether the difference is statistically meaningful or whether it will persist.
This is the gap Bootstrap CUSUM fills. It applies to time-series data and asks specifically: has the process permanently shifted to a new mean — and if so, when? It gives a statistical confidence level for the answer. It will not declare a structural change until the evidence warrants it. That is what an honest Study phase looks like.
The Act phase — three options, all legitimate
The Act phase has exactly three possible outcomes. All three are legitimate results of a PDSA cycle. Only one of them — abandon — is treated as failure in most organisations, which is why the Study phase gets skipped: if abandonment is treated as failure, the temptation is to declare success regardless of what the data shows.
The prediction was correct. The Bootstrap CUSUM chart confirms a structural step change. The change is expanded to the next scale. The next PDSA cycle tests whether the improvement holds at larger scale.
The prediction was partially correct or the implementation revealed problems. The Bootstrap CUSUM shows movement but not a confirmed step change. The change is modified based on what was learned and tested in a new cycle.
The prediction was wrong. The Bootstrap CUSUM shows no confirmed change. The theory of improvement was incorrect. This is valuable information — not failure. The next cycle tests a different theory. This is Deming's method working correctly.
PDSA and the Model for Improvement — the three questions
Langley et al.’s Model for Improvement places the PDSA cycle inside three questions that must be answered before any cycle begins. The questions establish the aim, the measurement, and the theory of change. The PDSA cycle then tests the theory repeatedly until the aim is achieved.
What are we trying to accomplish?
The aim. Specific, time-bounded, stated in terms of the system rather than the activity. Every PDSA cycle serves this aim. If a proposed cycle cannot be connected to it, it does not belong in the programme.
How will we know a change is an improvement?
The measurement. Has the process structurally shifted to a new mean — or are we seeing variation around an unchanged level? Bootstrap CUSUM answers this question with statistical confidence that standard run charts cannot provide.
What change can we make that will result in an improvement?
The theory of change. The PDSA cycle tests it. Joiner’s levels classify whether the proposed change is deep enough to hold. Deming’s 14 Points describe what Level 3 changes look like in management practice.
Joiner’s levels and PDSA — classifying the change before testing it
The PDSA cycle tests whether a change produces improvement. Joiner’s levels classify how deep the change is — and therefore whether it is capable of producing the kind of improvement that Bootstrap CUSUM will confirm as genuine and sustained.
This is the most important connection on this page. A PDSA cycle applied to a Level 1 intervention will produce a Study phase result showing no confirmed structural change — not because PDSA failed, but because the intervention was not aimed at the right level. The cycle worked correctly. The theory of change was wrong.
The practical sequence is:
- Apply the five-stage diagnostic method to find the structural root cause
- Classify the proposed intervention using Joiner’s levels — is it deep enough?
- Write the prediction using Deming’s by what method discipline — name the mechanism
- Run the PDSA cycle — small scale, defined timeframe, defined measure
- Use Bootstrap CUSUM in the Study phase — the honest arbiter of whether the process has genuinely changed
- Adopt, adapt, or abandon based on evidence — not on political pressure or elapsed time
PDSA in the NHS data — what the Study phase actually shows
Every article on this site is, in effect, a completed Study phase for a real-world policy intervention. The Plan was the policy. The Do was its implementation. Bootstrap CUSUM completes the Study. The results are consistent: where the intervention was at Level 1 or Level 2, no confirmed structural change. Where something systemic changed, the chart shows it.
Dementia diagnosis — twelve years of PDSA without an honest Study phase
The 66.7% target was set in 2012. The Plan was the target. The Do was the CQUIN incentive and the awareness campaigns. The Study phase — Bootstrap CUSUM — finds one stage, mean below target, no confirmed structural improvement. The Study phase was never completed honestly. The Act phase kept adopting an approach the evidence did not support.
Never events — Level 2 PDSA, Level 3 solution available
The PDSA cycles applied to wrong-route medication administration used Level 2 interventions — guidance, training, alerts. Bootstrap CUSUM finds a flat process at 17.5 per year. The Study phase result was available. The Act phase should have abandoned the Level 2 approach and adopted the Level 3 engineering control. It did not.
ULEZ — a Level 3 intervention confirmed by Bootstrap CUSUM
The Ultra Low Emission Zone changed the economics of vehicle use in London — a Level 3 structural change. Bootstrap CUSUM confirms a structural step change in NO2 levels. The Study phase gives an unambiguous Adopt result. The theory of change was correct. The mechanism worked.
Where this fits — and where to go next
“Give people better tools and a better mental model to ask better questions about whether things have improved through structural change — and to understand why, when they haven’t.”