Using Self-Compassion Practice to enhance PDSA Cycle

What each one is actually for

Self-Compassion Practice

Neff's practice is trying to change what happens in your head immediately after something you tried did not work. Not to make the result look better - to stop the result from being converted into a report on you.

It has three moving parts, working at once. The first is registering the disappointment as a real thing that hurts, rather than skating past it into the next plan; skating past looks like resilience and is usually avoidance, and it is why the same mistake gets repeated. The second is placing the miss inside ordinary human experience - first attempts usually miss, that is what a first attempt is - so the result stops being evidence about your particular defects. The third is the tone you use on yourself while doing the first two.

The usable instruction is a substitution test. Imagine a colleague you respect brings you exactly this log, this number, this shortfall. What do you say to them, in what words? Almost nobody says "this proves you are not serious." Then look at what you said to yourself when you opened your own spreadsheet.

It is worth being clear about what this is not, since it is the standard objection. Being warm about a miss does not mean deciding the miss did not happen or does not matter. It is a claim about the conditions under which people look at things honestly, not a claim about what the honest look should conclude.

PDSA Cycle

The PDSA cycle is trying to stop you from adopting or abandoning changes on the strength of how they felt. It comes out of industrial quality work - Walter Shewhart's cycle, later reworked and popularized by W. Edwards Deming - and it has since been used heavily in healthcare improvement, where teams run small tests on one ward for one week before changing anything at scale.

Four steps. Plan: name one specific change, small enough to run soon, and - this is the part everyone skips - write down what you predict will happen, in a number, before you start. Do: run it, and note what actually went on, including the parts that were not in the plan. Study: compare the real result against the prediction you wrote, and pay particular attention to the gap. Act: adopt it, adapt it, or abandon it, and start the next cycle.

Deming was insistent that the third step be called study rather than check, and the distinction is the whole method. Checking asks whether the thing was done. Studying asks what the difference between prediction and reality tells you about your model of the situation. A cycle where the result matched your prediction teaches you relatively little. A cycle that missed by a mile is where the information is - which is exactly the cycle a self-critical person is least able to read.

The other half of the design is scale. A test small enough to be wrong cheaply is a test you will actually run, and one you can afford to have fail.

What you get out of pairing them

The cost of a fudged study step is not moral. It is informational, and it compounds.

Here is the mechanism, concretely. You try something for two weeks - a new morning structure, a different way of handling a recurring argument. The result comes in well under what you predicted. If reading that number costs you an evening of self-attack, then one of three things happens, and all three are expensive: you quietly stop logging, you soften the entry so the record no longer reflects what happened, or you abandon the whole approach and conclude that these things do not work for you. In every case you have paid for the experiment and thrown away the finding. Run that pattern four or five times and you have a personal history that says nothing works, assembled entirely from experiments whose results you never read.

What the pairing offers is the ability to keep the data. Not comfort - data. An accurate log of six changes you tried, with what each actually did, is worth more than almost any advice you could be given, and it is the one thing that cannot be bought or borrowed because it is specific to you.

In the other direction, someone who has been practicing self-kindness for months gets something the practice cannot supply for itself: an answer to the question that otherwise stays open forever, which is whether any of it is doing anything. A practice with no prediction and no measurement can be maintained for years on the strength of its plausibility. Two weeks and a number will tell you more than another book will.

What Self-Compassion Practice rests on

• Acceptance reduces struggle randomised trials
  Allowing an internal experience to be present without resisting or acting on it reduces the psychological struggle that resistance would otherwise intensify.
• Common humanity reduces isolation observational studies
  Recognizing that suffering and imperfection are part of shared human experience reduces the sense of being alone in one's pain.
• Self kindness buffers shame randomised trials
  Responding to one's own failure with warmth rather than harsh self-judgment lowers shame and threat-based physiological arousal.

What PDSA Cycle rests on

• Feedback against prediction corrects belief plausible mechanism
  Comparing the actual result of a change against the predicted result before adopting it corrects mistaken beliefs faster than adopting or rejecting changes based on intuition alone.
• Small scale testing precedes scaling observational studies
  Running a change through a small planned test cycle before wider adoption reveals real-world effects while limiting the risk of being wrong.

Where they actually meet

A fact and a verdict are different objects, and both practices here exist to keep them apart - on different material.

The improvement cycle does it to a change: the number the change produced is one thing, and what you should now do about the change is a separate thing, arrived at deliberately in the fourth step rather than felt during the third. Keeping those apart is the reason the method is written as four steps instead of two. Neff's practice does the same separation on a person: what happened is one thing, and what it says about you is a separate question that mostly should not be asked, because the answer is nearly always some version of the same tired accusation regardless of the input.

The seam sits exactly at the study step, and it is a specific second rather than a general mood. That is the second when a real result either stays information about the plan or converts into evidence about you, and which one happens is decided before you have consciously decided anything - by whatever voice gets to the number first.

Notice what that means for the fourth step. If the result has already been converted into a verdict, "act" stops being a live choice among adopt, adapt and abandon. A verdict has only one continuation: stop, and feel bad. That is why a self-critical experimenter tends to have a history of abandoned methods rather than a history of adapted ones, and why the fix belongs upstream of the decision rather than in trying harder to decide well.

The principle underneath both

A result only teaches you something if the person reading it is not on trial. That is the pattern, and it is older and broader than either of these practices - institutions that need honest data have been building around it for decades, usually after learning the hard way.

Read that way, self-kindness at the study step stops being emotional hygiene and becomes a data-quality measure. The claim is not that you deserve to feel better. It is that a reporting system which punishes the reporter gets less accurate reports, and that this holds when the reporter and the punisher are the same person, which is the hardest version of the problem because there is no one else to notice it happening.

The same logic runs the other way and is worth stating, since it is the honest limit. A system with no consequences at all gets sloppy data too. What is being asked for is not indifference to results; it is separation. The consequence attaches to the plan - adapt it, drop it, try a different one - rather than to the person, and the separation has to be maintained deliberately because under disappointment the default is to collapse the two.

That is a way of understanding the mechanism, not a finding. What it does buy you is a testable prediction about yourself: if this account is right, your logging should get more accurate and more frequent, not just more pleasant, once the aftermath stops costing you.

What that makes the real goal

Being able to read a disappointing number in the same week it happened, and still want to run the next cycle. Both halves, or you do not have a learning loop - you have a series of abandoned attempts with a gap where the finding should be.

Put like that, the goal detaches from Deming and from Neff and turns into a standard you can apply to any self-improvement system you meet. Two questions. Does it produce a real record - something specific enough that a stranger could tell whether it happened? And can you read that record on a bad day without it costing you the evening?

Most popular systems score well on the first and are silent on the second, which is why so many well-designed trackers end up abandoned in week three. Streak counters are the clearest case: they generate excellent data and load the aftermath so heavily that one missed day can end the whole record. The data collection and the emotional accounting are welded together, and when the second one breaks it takes the first with it.

Aim at both and the diagnosis is usually easy. If you have no record, your problem is measurement and you need smaller tests with written predictions. If you have a record you avoid opening, your problem is the aftermath, and better tooling will not touch it.

Other practices that serve the same goal

Whole institutions have been built around this exact problem - getting accurate reports out of the people who made the mistakes - and their methods transfer to a single person surprisingly well.

The after-action review, developed in the US Army, is four questions asked immediately after an exercise: what was supposed to happen, what actually happened, what accounts for the difference, and what will we do differently next time. The design detail that matters is that it is conducted as a group inquiry into events rather than as an assessment of individuals, and rank is deliberately set aside during it. Run on yourself, those four questions are a complete study step with the verdict engineered out of the wording.

The blameless postmortem, standard practice in software incident response, is the same idea made explicit: the written rule is that the review examines what in the system allowed the failure, and never who was at fault, on the stated reasoning that engineers who expect blame conceal the details that would have explained the outage. It is the clearest existing statement of this page's argument, and it comes from an industry that adopted it for entirely practical reasons.

The premortem, from the psychologist Gary Klein, moves the whole thing before the event: you imagine the plan has already failed badly, then write down why. Because the failure is hypothetical, nobody is on trial while describing it, which is precisely what lets people say the thing they would not otherwise say. It carries experimental support for surfacing more and better-specified risks than ordinary planning discussion.

None of those three mention feelings. They are all engineering answers to the same problem, and if warm self-talk is not your register, they are the version of this page you should be reading.

A tradition that already does this

Aviation got here first, and it built the argument into law rather than into a practice.

Confidential incident reporting in aviation - the system in the United States is run by NASA rather than by the regulator, which is itself the point - exists because the industry established that punished reporters stop reporting. Pilots and controllers file accounts of their own near misses and errors, and in return receive confidentiality and limited protection from enforcement action for what they disclose. The reasoning is not generous. It is that the reports are the single most valuable safety data available, they can only come from the people who made the mistakes, and a system that treats those people as defendants receives nothing.

Medicine has adopted versions of the same design in incident reporting, for the same reason and against considerable cultural resistance, since the instinct when something goes wrong with a patient is to find who did it.

That is a strong precedent for the claim on this page, and it is worth noticing the shape of it: nobody in aviation argues that pilots deserve to feel good about their errors. The argument is entirely about data quality. What the precedent supports is the mechanism - separate the record from the judgment or lose the record - not the specific practice of talking kindly to yourself, which is one way of running that separation inside a single head, and not the only one.

Using Self-Compassion Practice to enhance PDSA Cycle

A cycle is only worth running if the study step reports what actually happened, and that step is where the record gets quietly protected: the number softened, the entry postponed, the whole log left unopened. Build the aftermath into the cycle rather than hoping it goes well.

One: at plan time, write the change in one line and the prediction as an actual number with a date. "Phone in the other room for the first hour, starting Monday. Prediction: focused-work time rises from about forty minutes a morning to about eighty, averaged over the week." Vague predictions cannot miss, which is comfortable and useless.

Two: still at plan time, before any emotion exists, write one sentence naming what a miss would most likely mean about the plan. "If it comes in under sixty, the likely reason is that the phone was never the binding constraint." This is the highest-leverage step on this page, and it takes fifteen seconds. You are writing the interpretation while you are calm so that the disappointed version of you does not have to generate one.

Three: when the result arrives, read it out loud in the third person before you write anything. "Focused-work time came in at fifty-one minutes." Third-person description is a small, well-studied distancing move, and the effect here is immediate: it is much harder to attach a character judgment to a sentence that is phrased as a report.

Four: name the disappointment in one plain sentence - "that is a lot less than I hoped, and it is annoying" - and add the placing clause, said as fact rather than as consolation: first tests usually come in under prediction, that is why they are tests. Ten seconds. Do not extend it into a session about your relationship with achievement.

Five: write the actual number, unedited, and then choose the fourth step out loud: adopt, adapt, or abandon. Naming which of the three you are doing prevents the most common silent outcome, which is drift - neither continuing nor stopping, just not mentioning it again.

Where this goes wrong: doing step four instead of step five. If the kind sentence arrives and the number never gets written, the practice has become a way of managing how you feel about not looking. The kindness is in service of the record, and the test of whether it worked is whether the record exists.

Using PDSA Cycle to enhance Self-Compassion Practice

Treat one self-compassion move as a change to be tested rather than a stance to adopt, and the practice stops being something you either believe in or do not. "Be kinder to myself" cannot fail and cannot succeed, which is why people carry it for years without ever knowing whether it did anything. A cycle gives it an answer date.

One: plan a single specific move for a single specific trigger, defined so that a stranger watching could tell whether it happened. Not "be gentler about work." Something like: "When I catch myself rereading a sent message, I put a hand on my chest and say the three lines, once." One move, one trigger.

Two: write a prediction, with numbers. "The sting is usually about a seven out of ten and lasts most of an hour. Prediction: after the move it drops to about a five within ten minutes, on at least three of the five times it happens over these two weeks." A prediction that specific feels excessive right up until the moment it saves you from a vague impression.

Three: do it for two weeks and log within ten minutes of each occasion. Not at bedtime. Bedtime recall is unreliable and it gets colored by however the rest of the day went, which is the single most common way personal logs become fiction.

Four: study it against what you wrote. And keep the unit straight: you are rating the episode, never yourself. The entry is "sting went 7 to 6, faded by the afternoon", not "did badly at this again."

Five: act, and say which of the three you are doing. Adopt this move, adapt it - a different phrase, an earlier trigger, a physical anchor instead of words - or abandon this particular move. Abandoning one move is not abandoning self-compassion, and being explicit about that distinction is what stops a single failed cycle from taking the whole practice with it.

The obvious risk: measuring kindness can become one more thing to be graded on, which is the exact voice the practice is meant to interrupt. Two guards. The unit of measurement is the episode, not the person. And "no change" is a successful cycle - it is a real finding about a specific move, it cost you two weeks, and it tells you to try a different one. If a null result feels like a verdict, that is worth noticing before you plan the next cycle.

Sunday evening, opening the log for a week that did not work

You planned to leave your phone in another room for the first hour of the workday. You predicted your focused-work total would roughly double - forty minutes to eighty, averaged across the week - and now the week is over and the log is open on the screen. It moved by eleven minutes.

The first sentence shows up before you have finished reading the row, and it is the usual one: you cannot even do the easy stuff. Familiar tone. Extremely fast.

What you do with the next thirty seconds decides whether this week produced a finding or a bad mood. You read the row out loud instead: focused-work time came in at fifty-one minutes, against a prediction of eighty. Then the plain acknowledgment - that is well under what I wanted and it is genuinely annoying - and the placing clause, which is not consolation but an actual fact about first tests: a first test missing its prediction by this much is closer to the median outcome than to a personal failure. Then you say to yourself roughly what you would have said to a colleague who brought you this exact log, which is certainly not that they are not serious.

Then you write the study entry, and because you are not defending anything, you notice the thing that was in the data all along: two of the five mornings had video calls at nine that the plan never accounted for, and on the three clean mornings the number was closer to seventy. That is not the same finding as "the phone thing does not work." It is "the phone thing was not tested on two of the five days," which is a completely different fourth step.

You act: adapt. Next cycle runs on the three clean mornings only, with the prediction rewritten to match. Eleven minutes was never the finding. It was just the first number you saw.

When this is worth reaching for

The log has been open in a tab for three days and you have not scrolled to the bottom of it. You know roughly what it says.

That specific avoidance is the clearest signal that this pairing is aimed at your situation - not disorganization, not a lack of tracking tools, but a record that exists and is not being read. A related sign: you can list four or five things you tried this year and cannot say what any of them actually did, because you stopped logging around the point each one stopped feeling promising.

Another condition: you have a habit of running the next experiment before finishing the last one. Starting fresh is much more comfortable than reading a result, and a rolling series of new beginnings can look like enthusiasm for months while producing exactly no information.

And the reverse entry point, for someone whose practice is already the kind one: you have been doing self-compassion daily for a long time and you genuinely cannot say whether it changed anything. Not because it did not - because nothing was ever predicted, so nothing could be compared. Two weeks with a number attached will resolve a question that another year of practice will not.

Skip this until you have run at least one small change with a written prediction. Before that there is no result to be kind about, and adding an emotional protocol to an experiment you have not run yet is preparation standing in for the thing itself.

Where this does not apply

If you already look at a disappointing result without spiraling - you read the number, you note what it means, you adjust - then a formal kindness step before every study phase is overhead solving a problem you do not have. Put the effort into better predictions instead; that is where your returns are.

When the cycle is running on a high-stakes external commitment - a safety check, a clinical protocol, anything where someone else bears the consequences - the priority is an accurate answer quickly. Slowing the study step down to process how you feel about the result can cost time that matters. Self-compassion belongs to how you hold the result afterward, not to whether you look at it squarely and now.

There is also a category of thing that should not be run as an experiment at all. Promises to a person are the main one. Framing "I will not check my phone while you are talking to me" as a two-week test with a predicted compliance rate turns a commitment into a trial and, in most relationships, will be heard exactly that way. Some things are decisions, not cycles.

And a subtler misuse: the cycle as a way of not deciding. Someone who has run six clean, honest, well-logged cycles on whether to leave a job has not been experimenting for the last four of them. Measurement can be the most respectable available form of avoidance, and the tell is that the fourth step keeps coming out as adapt when the real options were adopt or abandon.

What the evidence does and does not support

One side of this page has trials behind it. The other has decades of practitioner experience and surprisingly little else, and the improvement literature sounds more scientific than it is, so the distinction is worth making bluntly.

The improvement cycle is the weaker of the two evidentially. That comparing an actual result against a stated prediction corrects a mistaken belief faster than adopting or dropping a change on intuition is mechanistic reasoning - it follows from how the method is constructed rather than from trials that isolated it. The related claim that a small planned test before wider adoption reveals real effects while limiting the cost of being wrong is observational, drawn from improvement work in healthcare and industry rather than from randomized comparison. Reviews of PDSA use in healthcare have in fact found that published cycles frequently fail to include the prediction step at all, which tells you something about how often the method is run as designed. None of this is a reason to skip it; it is a reason not to describe it as evidence-based in the sense a drug trial is.

Self-compassion is the better-tested half here, though by less than the citation counts imply. Randomized studies stand behind its central effect: warmth rather than severity toward what you got wrong lowers shame, and lowers the body's alarm reaction along with it. Behind nearly all of them sits one self-report questionnaire, written by the researcher who defined the idea in the first place, which is worth knowing when a record looks unanimous. Its shared-humanity component has correlational support only. Wider than any of this, from research with no connection to Neff, is randomized evidence that fighting an unwanted internal state costs more than allowing it.

The specific thing this page describes - self-compassion applied at the study step of a personal test cycle - has not been studied. The study would be straightforward to design: give people a change to test, randomize the aftermath script, and measure not how they felt but whether they filed the result and whether the result they filed matched what was actually recorded. That last measure is the one that would matter, and nobody has run it.