An addendum to Prediction, Control, and the Regulating Mind

Self-Injury as a Regulatory Behavior

Dr. Greg Moody · August 24, 2026

Written for clinicians. This is about mechanism and case formulation. There's no description of methods here and nothing that'd work as instruction. If you're reading this about your own situation rather than a caseload, please talk to somebody you trust or a licensed professional – I'd rather you did that than finish the article.

Part 5 described stimming as a way to buy predictability back by generating a signal the person controls in a sensory world they don't. Barca and colleagues make a related proposal for non-suicidal self-injury, except the signal comes from inside the body.

Affect regulation has empirical support. The predictive account tries to explain why the body becomes the instrument, but that part is still thin – thin enough that it matters, because ideas like this get repeated a lot faster than they get tested.

The Line Between Self-Injury and Suicidal Behavior Is Primarily Intent

The field separates them primarily on intent to die. Nock puts both inside one class of self-injurious thoughts and behaviors, then divides that class by whether there was some intent to die. He's direct that intent gets assessed by self-report, which carries bias, inaccuracy and ambivalence, and that the convention deliberately errs toward over-classifying.

Why does the exact wording matter here? Because the version in circulation is stronger than the source, and a family will hear the strong one. Nock's review draws the line on intent and treats it as a classification convention, not a demonstrated clinical separation (Nock, 2010).

It doesn't compare the two on medical lethality. That contrast gets attributed to this paper regularly and it isn't in it, which is worth knowing before you say it to a family.

The prevalence numbers in that same review are worth carrying into a parent conference. Lifetime self-injury runs somewhere between 13% and 45% of community adolescents and around 4% of adults, and rises to 40–60% in clinical adolescent samples. Onset clusters at 12 to 14.

Nock compares it to the disorders we all trained on. Even the low-end estimates put self-injury above the lifetime prevalence of anorexia and bulimia nervosa (under 2%), panic disorder (under 2%), obsessive-compulsive disorder (under 3%) and borderline personality disorder (2%). Most of us were trained more thoroughly on every one of those.

The Function Literature Has Supported This for Decades

Klonsky reviewed eighteen empirical studies covering seven proposed functions and found the evidence strongest for affect regulation, even after accounting for it being studied most often. Four findings converge: acute negative affect comes before the behavior, negative affect and relief follow it, most people who self-injure name relief as the reason, and laboratory proxies reduce negative affect and arousal.

Klonsky's paper is a narrative review, not a meta-analysis, so there are no pooled effect sizes to quote (Klonsky, 2007). Four different lines of evidence converge on affect regulation, which is usually more useful in a formulation than one tidy number.

The other six functions in Klonsky's review were anti-dissociation, anti-suicide, interpersonal boundaries, interpersonal-influence, self-punishment and sensation-seeking. Self-punishment drew strong support. The other five drew modest support, and the pattern held across clinical, non-clinical and forensic samples.

Nock and Prinstein built the model most of us were taught from 108 adolescent psychiatric inpatients, with the factor analysis running on the 89 who reported an incident that year (Nock & Prinstein, 2004). Two crossed dimensions, four functions: automatic-negative (stopping bad feelings), automatic-positive (feeling something at all), social-negative (escaping a demand), social-positive (communicating distress).

Automatic-negative scored significantly higher than everything else… 52.9% endorsed "to stop bad feelings." Automatic items got endorsed by 24% to 53% of the sample, social items by 6% to 24%. The authors call it PRELIMINARY until replicated with a larger clinical sample, and I'd keep saying that out loud, because a four-box model is very easy to teach and very easy to over-trust.

Chapman, Gratz and Brown put the same finding in learning-theory terms with their Experiential Avoidance Model – the behavior is maintained mostly by negative reinforcement, escape from or avoidance of unwanted emotional experience (Chapman et al., 2006). Theoretical paper, no new data.

The Predictive Account Adds a Mechanism Under the Function

Barca, Maisto and Pezzulo read self-injury as an active-inference strategy for reducing uncertainty. The brain builds predictive models of bodily and emotional states, and when those models are poor or the interoceptive channels are noisy, a person carries uncertainty about what they're feeling that won't resolve. Acting on the body buys certainty back, because pain is an unusually precise signal.

"By harming the body, we turn it into a very precise source of sensations that relieves us from excessive uncertainty about the present state and the future course of action" (Barca et al., 2023).

Affect regulation tells you what the behavior BUYS. The predictive account asks why the body becomes the instrument, and the authors are explicit that it complements the affect-regulation literature rather than replacing it.

It's a perspective paper with one illustrative simulation. No participants and no empirical sample. The authors flag it twice on their own initiative, calling the proposal still speculative and saying the hypothesis remains to be investigated.

Their word is "precise," not "controllable." Those are different claims and the difference matters – a precise signal is one you can read, not one you command – so keep their word when you use this.

If Uncertainty Is Doing the Work

Stimming produces an external signal the person controls. The predictive account says self-injury produces an internal signal that is unusually precise. If either behavior is reducing uncertainty, suppressing it without dealing with that uncertainty removes the person's answer and leaves the problem.

That comparison deserves suspicion. One model explaining two very different behaviors may have found a shared process. It may also be too broad to fail, and Part 7 is about telling those two situations apart.

Functions autistic adults report for stimming
Figure 1. Functions autistic adults report for stimming. Managing uncertainty is one of them, and that is the function the predictive account is reaching for.

What would separate them here? Move uncertainty about internal state without moving distress. The predictive account says urge should shift, an affect-regulation account says it shouldn't. I haven't found a direct test. Until somebody runs one, it's a hypothesis with no direct participant data under it, and I'd say that out loud every time I used it.

What This Does Not License

It doesn't license a new technique. There's no outcome data behind the predictive account, and treatment for self-injury keeps coming from the evidence-based protocols and the supervision you already have.

It also doesn't license telling a client their self-injury is "information seeking." That's a mechanism running underneath anything the person recognizes about their own experience, and delivered as an explanation it lands as being dismissed.

Where does it actually help? In formulation. What is this person uncertain about, how noisy are their reports of their own internal state, and what else could make those reports legible to them? Those are answerable, and they point at interoceptive and emotion-labeling work rather than at argument.

Use the affect-regulation finding in your formulation. Carry the predictive account as a question because the paper provides no direct participant data. On your next case, write down what the person is uncertain about before you write down what they're feeling, and see whether the two lists differ.

References

Barca, L., Maisto, D., & Pezzulo, G. (2023). Modeling and controlling the body in maladaptive ways: An active inference perspective on non-suicidal self-injury behaviors. Neuroscience of Consciousness, 2023(1), niad025. https://doi.org/10.1093/nc/niad025

Chapman, A. L., Gratz, K. L., & Brown, M. Z. (2006). Solving the puzzle of deliberate self-harm: The experiential avoidance model. Behaviour Research and Therapy, 44(3), 371–394. https://doi.org/10.1016/j.brat.2005.03.005

Klonsky, E. D. (2007). The functions of deliberate self-injury: A review of the evidence. Clinical Psychology Review, 27(2), 226–239. https://doi.org/10.1016/j.cpr.2006.08.002

Nock, M. K. (2010). Self-injury. Annual Review of Clinical Psychology, 6, 339–363. https://doi.org/10.1146/annurev.clinpsy.121208.131258

Nock, M. K., & Prinstein, M. J. (2004). A functional approach to the assessment of self-mutilative behavior. Journal of Consulting and Clinical Psychology, 72(5), 885–890. https://doi.org/10.1037/0022-006X.72.5.885

Prediction, Control, and the Regulating Mind

You are reading an addendum to this series.

  1. 1. What an Emotion Actually Is
  2. 2. Why Your Brain Guesses at the World
  3. 3. What You Predict Is What You Defend
  4. 4. What This Changes in the Therapy Room
  5. 5. The Hard Case: Autism and Stimming
  6. 6. The Loop That Won't Settle
  7. 7. How to Prove This Wrong

Addenda

Companion business articles

The full documents, free and ungated

This is the expanded reference edition, with the full framework and sources. A shorter version first appeared on today.mastermoody.com.