Neuroscience

Equal and Opposite: What I Told a Friend About Breaking Bad Habits — and What the Science Actually Says

September 19, 2026 · by Brendon Kennedy

In short: You don't delete a bad habit — you out-compete it. The evidence-backed core of what I texted a struggling friend: impulses are cue-driven circuitry, not character flaws; the reliable fix is a pre-decided, physically incompatible counter-move like a run. The 3-day fast is my n-of-1 experiment, not the science.

A friend of mine — I’ll call him Peter, because that isn’t his name — sent me a photo a few weeks ago. His bathroom scale, one of those old analog dial models, shot from above, with one word under it: “Progress.”

It was progress. He’s been putting in real work. And I knew exactly where he was standing, because I’ve stood on that scale — the late-night bag of chips, the couple of drinks that weren’t the plan, and the next-morning question that eats at you worse than any number on a dial: why did I just do that, even when I know it hurts me?

Here’s a condensed version of what I texted back:

I fixed my bad habits and still struggle with new ones, but I keep running — and that gives me a “way out” if I eat a bag of chips late at night or have a couple drinks. We need a way to reset our mind. Trust me, that’s what it’s all about.

If you find yourself battling with self-control — doing things and asking yourself right afterwards, “why did I just do that even when I know it hurts me?” — the easiest fix for me to disrupt that thinking pattern and break bad habits is an extended 3-day fast. And it’s not that difficult.

If I can do this, you can too, friend. 🌻

That text is advice from experience, sent with love. This post is me doing what I do with my own advice: dragging it into the literature to see which parts stand up, which parts are weaker than I’d like, and which parts are honestly just me. Because there’s a line I keep coming back to — it’s on my homepage now — sometimes contrary thinking is the best answer: for every action, there’s an equal and opposite reaction. Newton meant paired forces. I’m stealing it as a design principle for behavior: you don’t delete a bad habit — you push back against it with an action in the opposite direction. Most of this post is about where that metaphor is backed by real evidence, and the one place where I have to admit it runs ahead of the science.

”Why did I just do that?” is a circuit, not a character verdict

The question Peter and I both know — why did I do the thing I’d already decided not to do? — has an actual answer in the research, and it isn’t “because you’re weak.”

Your brain runs behavior on (at least) two systems. One is goal-directed: it weighs actions against outcomes and updates when the outcome stops being worth it. The other is habitual: stimulus in, response out. In rodents, these systems dissociate cleanly — lesion work reviewed by Yin and Knowlton shows the dorsomedial striatum supports action-outcome learning while the dorsolateral striatum supports stimulus-response habits, and the diagnostic test is outcome devaluation: devalue the reward, and goal-directed behavior stops while habitual behavior keeps firing. That last clause is the whole late-night-chips phenomenon in one sentence — the behavior kept running after the outcome stopped being worth it. In humans the picture is thinner: one influential fMRI study found that overtrained people kept responding to cues after the snack outcome was devalued, with rising activity in the presumed human homolog (posterior putamen) — but it’s a single small study, and later labs have had trouble reproducing lab-made habits in people. So: a strong animal framework, a suggestive but preliminary human extension.

Layer on the stopping problem. On a leading account from Aron, Robbins, and Poldrack, a sector of the right inferior frontal cortex is proposed to act as a brake on already-initiated actions. In group comparisons, people with ADHD are on average slower on exactly this — stop-signal reaction time runs about g = 0.62 slower than controls (similar deficits show up in OCD and schizophrenia, so it’s not ADHD-specific — and it’s a lab measure with heavily overlapping distributions, not a diagnostic). People with ADHD also discount delayed rewards moderately more steeply (d = 0.43) — the far-away payoff of “healthy at 60” loses to the chips that exist now — and steep discounting is transdiagnostic across many conditions, including binge-eating disorder. I’ve written about the ADHD side of my own brain elsewhere in this catalog; the short version is that if your reward-forecasting system is quiet, the immediate thing wins more often, and no amount of self-scolding changes the math.

Two more pieces complete Peter’s picture. First, negative urgency — the tendency to act rashly specifically when feeling bad — is the impulsivity facet most strongly associated with binge-type eating (r = .38, roughly double the other facets) and among the strongest correlates of problem drinking across 96 studies. These are cross-sectional correlations, not proof of causation — but they name the pattern precisely: the chips and the drinks don’t show up on good days. They show up on hard ones. Second, the timing isn’t an accident either: sleep restriction measurably impairs executive function and sustained attention (small-to-medium effects in meta-analysis — though, honestly, a direct effect on impulsive-choice tasks specifically wasn’t established). The 11 p.m. version of you is running degraded executive hardware in the exact hour your cues are strongest.

One more mechanism worth naming, carefully: Berridge and Robinson’s incentive-sensitization work suggests “wanting” and “liking” are dissociable — in rodents, depleting dopamine abolishes motivated pursuit while leaving pleasure reactions intact. It’s one influential theory among several, and the causal evidence is animal work. But it matches something everyone with a bad habit recognizes: the cue can fire off wanting for a thing you don’t even like that much anymore. That’s not hypocrisy. That’s the machinery.

Why “just stop” keeps failing

If habits were decisions, deciding would fix them. They aren’t. In Wood and Neal’s model — the leading account in the field — habits are context-response associations: the couch, the hour, the show, the preceding beer are the triggers, and the response runs without consulting your goals at all. In diary studies, roughly 45% of people’s daily behaviors repeated in the same physical location. My favorite demonstration is the cinema-popcorn experiments: moviegoers with strong popcorn habits ate just as much week-old, stale popcorn as fresh — liking didn’t matter, hunger didn’t matter — but only in the cinema, and only when eating with their dominant hand. Switch hands, and suddenly even strong-habit people noticed the popcorn was terrible. Two field experiments from one lab, so hold it as suggestive rather than settled — but the shape of the finding is the point: the habit lives in the context and the motor pattern, not in your opinions.

Which is why pure suppression is such a bad bet. A meta-analysis of 31 thought-suppression studies found that suppressed thoughts tend to rebound — coming back more after you stop suppressing. That’s a finding about thoughts, not behaviors, so applying it to urges is an extrapolation — but it’s one honest reason “just don’t think about the chips” performs so poorly, and why replacement may work better: it gives the monitoring mind an actual job.

Equal and opposite: replace, don’t erase

Here’s where my line earns its keep — and where I have to break it, because the break is the insight.

Every intervention in this literature with decent evidence works by installing an incompatible action on the same cue, not by willing the old response away:

• Competing-response training — the core of habit reversal therapy — teaches you to perform a physically incompatible action the moment the urge or first movement appears. A meta-analysis of 18 studies found a large effect (d = 0.80) — with the mandatory fine print that only one of the 18 studies met all four methodological quality criteria, and the effects come from therapist-delivered clinical treatment of tics, nail-biting, and similar disorders, not from people self-applying it to snacks. (If you have tics or compulsive behaviors, that’s professional-care territory, not blog-post territory.)
• Opposite action is literally the name of a skill in dialectical behavior therapy — deliberately acting against an emotion’s urge. The honest evidence status: the DBT skills package has RCT support in structured clinical treatment, but no trial has isolated the single skill. The clinicians got to “act opposite” long before I stole it from physics.
• Implementation intentions — pre-deciding “if X happens, I will do Y” — are the best-evidenced way to pre-load the replacement onto the cue: d = 0.65 for goal attainment across 94 tests. But the number that actually matters for breaking habits is smaller: in a meta-analysis of eating studies, if-then plans hit d = 0.51 for starting healthy eating and only d = 0.29 — small but significant — for stopping unhealthy eating.

That asymmetry is where Newton breaks down, in three ways worth saying explicitly. In physics the reaction is automatic; in behavior it never is — you have to install the counter-move in advance, which is exactly what an if-then plan does. The reaction doesn’t need to be “equal and opposite” — the evidence says it needs to be incompatible, immediate, and pre-decided. And the forces aren’t symmetric: starting a new behavior is reliably easier than stopping an old one (0.51 vs 0.29). Which is the honest, evidence-shaped reason to aim your effort at building the replacement rather than fighting the habit head-on. Contrary thinking, literally: stop attacking the thing; build its opposite.

Two more honest numbers for the road. Context change genuinely helps — a field experiment with 800 participants found behavior-change interventions worked better among people who’d recently moved house, with an estimated window of roughly three months — though at population scale the effect is small, so treat it as a modest opening, not a magic door. And the most-cited real-world habit-formation study found a median of 66 days for a new simple behavior to reach automaticity — a median from the 39 of 96 participants whose data fit the model, with a range of 18 to 254 days, so it’s an anti-rule more than a rule. The two useful facts inside it: plan in months, and missing a single day barely dented anyone’s curve. A lapse is a data point, not a reset to zero.

Running: what the “way out” actually buys

So what makes a run a good opposite action? I want to be precise here, because this is where pop-neuroscience oversells.

The strongest evidence is from smoking: a meta-analysis with individual participant data found brief exercise reduces cigarette cravings during and shortly after the bout. The standardized effect sizes look enormous (≈ −1.9 to −2.0), but they come largely from small within-subject lab designs that inflate such numbers — so trust the direction, not the magnitude. The relief is transient: minutes to about an hour. For food, the evidence is thinner but pointed the same way: a 15-minute brisk walk reduced chocolate cravings versus sitting (58 women; small crossover trials from one research team). For alcohol — the drink half of my own confession — I have to report the evidence is genuinely weak: one 20-person trial found urges dropped only during the exercise, and an uncontrolled study in people with alcohol use disorder found craving fell in 70% of participants but rose in 16%. Anyone in or near alcohol withdrawal needs medical clearance before exercising, full stop.

What a run reliably buys, on average, is mood and time. Across 158 studies and 13,101 participants, acute aerobic exercise lifts positive affect (ES ≈ 0.47) — most consistently at easy intensities, with the lift persisting roughly 30 minutes — while control conditions drift slightly down. The cognitive bump from a single bout is real but honestly small (g ≈ 0.10 overall). And the “running floods your brain with dopamine” story you’ve heard? Rodent-backed, not demonstrated in humans — the one PET study that looked found no detectable striatal dopamine release after 30 minutes of vigorous treadmill running in 12 volunteers. (The scan timing could have missed a transient release — absence of evidence, not proof of absence. But don’t let anyone sell you the flood as human fact.)

Notice that none of this needs the word “treatment” — and shouldn’t get it. The honest mechanism for the “way out” is humbler and better: a run is a pre-decided, physically incompatible, immediately available counter-move. You cannot eat chips mid-run. It changes your context (you’re not on the couch), occupies the motor system (the non-dominant hand trick, scaled up), and hands you a ~30-minute window where your mood is better and the urge has usually crested and passed. The lapse itself becomes the cue: if it’s the morning after chips-and-drinks, then shoes on before the phone. That’s my actual protocol, and every load-bearing part of it has evidence behind it — as an urge-management pattern, not a cure for anything. (Exercise programs in ADHD, for what it’s worth, show benefits in kids mostly against passive controls in the newer, larger meta-analyses — an adjunct, never a replacement for treatment.) It’s also why I log my runs publicly.

The 3-day fast: the part where I have to be honest with both of us

Now the hard one. I told Peter a 3-day fast is the easiest fix I know for disrupting the pattern. I’ve done it; I wrote up what 72 hours actually does metabolically — and something in how I relate to impulses really did feel different afterward. So what does the evidence say about that?

Less than I’d like. Here’s the honest ledger.

The fuel switch is real. Per the NEJM review by de Cabo and Mattson — using the corrected figures from its 2020 erratum — ketones sit around 0.2–0.5 mM through the first 24 hours and reach 1–2 mM by 48; the one supervised 72-hour dataset (monitored patients on diagnostic fasts, not healthy volunteers) shows ketosis still deepening in stages through hour 72. Note the scope line in that NEJM review, because fasting influencers skip it: multi-day fasting is explicitly not discussed there. By around day three, a small PET study of nine volunteers found the brain drawing about a quarter of its energy from ketones, glucose metabolism down ~26%. Your brain really is running differently on day 3. But — and this is the sentence the wellness version omits — a measured change of fuel is not a measured change of function.

The function evidence points the other way or nowhere. The best-controlled ~3-day deprivation trial (n = 23, double-blind, though confounded by 4 h/day of exercise) found mood got worse — more tension and fatigue — and cognition didn’t change in either direction. Human BDNF trials of fasting regimens are heterogeneous and net-null (5 up, 5 down, 6 no change — none of them multi-day fasts); the one study spanning a true 72-hour fast found a 10–13% rise in plasma pro-BDNF, precursor only, mature BDNF unchanged — and plasma BDNF is a poor proxy for the brain’s anyway. The ketone-BDNF mechanism you’ve heard about is rodent-and-cell work.

And the reward-circuit “reset”? There is no human study of dopamine or reward circuitry after a multi-day fast — none I could find. What short deprivation studies show runs opposite to the reset story: after a single overnight fast, reward regions responded more strongly to high-calorie food images, not less. Chronic food restriction in mice sensitized dopamine neurons. And months of semi-starvation in the Minnesota experiment made food an obsession for most of its 36 men. Different exposures from 72 hours, all of them — but every arrow in the neighborhood points toward restriction amplifying food salience, not quieting it. So I’ll say it plainly: there is no direct evidence that fasting acts as a pattern-interrupt for habits, and no one should fast expecting quieter cravings. Nothing here is a weight-loss claim either.

So why did it work for me? The most defensible reading is psychological and not fasting-specific. Experimentally raising self-efficacy reliably moves behavior (d ≈ 0.47 across 204 experiments — none involving fasting), and in Bandura’s original theory the strongest source of self-efficacy is a mastery experience: completing a hard, voluntary, clearly bounded challenge. Three days of “I decided not to, so I didn’t” — while hunger arrived in waves and left again — is a textbook mastery experience, and it rewrote something about how seriously I take my own urges. But that bridge is my inference, not a finding: no study has manipulated extended fasting and measured self-efficacy or habit change. And the same logic predicts the same payoff from a cold-water swim, a long hike, or a hard training block — cheaper, lower-risk challenges that don’t require not eating. The fast was the biggest opposite action I could take. That part is my experiment, not the evidence.

The safety part, which is not optional. “It’s not that difficult” was true for me — it is not true for everyone, and for some people a multi-day fast is genuinely dangerous. Not for you if: type 1 diabetes; any insulin or sulfonylurea (or other glucose-lowering medication) without your prescriber directly involved — clinical guidance exists for medication around even short fasts and it is emphatically not DIY; pregnancy or breastfeeding; children and adolescents; a current or past eating disorder; underweight or recent unintentional weight loss; advanced kidney, liver, or cerebrovascular disease; or any prescription that assumes food in your stomach. The reassuring safety statistics you’ll see quoted come from an inpatient clinic cohort where 1,422 people were fed 200–250 kcal/day and examined by physicians two to three times a week — the safety belongs to the supervision, not the fast. Refeeding is where the most under-appreciated harm sits — and it’s not the only risk; supervised programs reintroduce food gradually over days, not with a feast. Dizziness on standing, palpitations, confusion, fainting, persistent vomiting: those are stop signals, not milestones. A first multi-day fast is a conversation with a clinician first — same thing I told Peter when he asked.

What I’d text Peter now

Having done the reading, here’s the same advice, recalibrated to what the evidence actually supports:

1. Rename the problem. It’s not a character deficit; it’s cue-triggered circuitry running on degraded late-night hardware. The shame spiral is not just useless — it’s the negative emotion that urgency feeds on.
2. Stop attacking the habit; build its opposite. Pick the counter-move in advance and bind it to the cue with an if-then: mine is lapse last night → shoes on this morning. Make it physical and incompatible with the habit. Expect the stopping side to be harder than the starting side — the evidence says it is — so pour effort into the starting side.
3. Use the run as the way out, honestly labeled. It buys a mood window and makes the habit physically impossible for half an hour. That’s not a small thing. That’s the whole thing.
4. Give it months, and waste no time on guilt over single lapses. Median 66 days, range 18 to 254 — and one missed day barely moves the curve.
5. Hold my fast the way I now hold it: a mastery experience that worked for me, untested by science as a habit-breaker, with real exclusions — not step one. Step one is the shoes.

The dial on Peter’s scale is already moving — he sent me the photo captioned “Progress,” and he’s right. For every action, an equal and opposite reaction: the physics is borrowed, the principle is sound, and the reaction doesn’t install itself. You install it.

If I can do this, you can too, friend. 🌻

← All writing