The Human Agency Pie — What Share of People Actually Think for Themselves
The Human Agency Pie
Independent judgement is what survives seven motivational drivers. What share of people resist all seven?
This is the population question that the Cognitive-Motivational Bias Matrix implies. The matrix sets out the seven motivational drivers that produce systematic thinking error. This chart asks what share of people hold their own judgement against all seven of them.
The drivers are not seven symptoms of one weakness. They are seven distinct reasons a mind distorts, each with its own theoretical grounding, and a person can be genuinely immune to one and fold completely to the next. Someone who cheerfully holds the minority view among peers will still defer to a lab coat. Someone who tests every claim they read will still decline to know a result that frightens them.
That is why the last slice is small without requiring heroism on any single axis. Each driver is survivable alone. The rare thing is being unmoved by all seven, in seven different moments, when nothing in the environment rewards it.
M5 is the one people underestimate. The other six distort a judgement you are making. M5 stops the judgement being made at all, by keeping the information away from you — and avoidance runs highest exactly where the stakes are highest.
Share of adults by tier
Segments drawn to true angle — no scale break- M1Minimizing mental effort
Minimizing mental effort pulls you toward stopping as soon as an answer arrives, because thinking is expensive. To resist it, you have to do a second pass you already feel finished with.
Bounded rationality (Simon, 1957); cognitive load theory (Sweller, 1988)
83% missed at least one Cognitive Reflection Test item, where an intuitive wrong answer arrives first. 33% of 3,428 respondents missed all three. Frederick, 2005.
- M2Maintaining coherence
Maintaining coherence pulls you toward the answer that matches what you already believe, so nothing has to be rebuilt. To resist it, you have to let a belief you hold change shape.
Cognitive dissonance (Festinger, 1957); meaning maintenance (Heine et al., 2006)
over 90% still cited the volatile materials when explaining a warehouse fire, after being told explicitly that the closet had been empty. Johnson & Seifert, 1994.
- M3Preserving self-esteem
Preserving self-esteem pulls you toward the answer that leaves your sense of yourself intact. To resist it, you have to be wrong somewhere you can see it.
Self-affirmation theory (Steele, 1988); motivated reasoning (Kunda, 1990)
direction, not a rate identical data read correctly as a skin-rash trial became polarised and less accurate relabelled as gun control — and polarisation increased with numeracy, so capability was spent protecting identity. Kahan, Peters, Dawson & Slovic, 2013.
- M4Seeking social belonging
Seeking social belonging pulls you toward the answer the people around you already hold. To resist it, you have to hold a view your own people do not.
Belongingness hypothesis (Baumeister & Leary, 1995); social identity theory (Tajfel & Turner, 1979)
≈75% conformed to a visibly wrong majority at least once; only about a quarter never conformed on any critical trial. Asch, 1951/1956.
- M5Managing emotional comfort
Managing emotional comfort pulls you away from finding out the thing that would hurt to know. To resist it, you have to go and look anyway.
Regulatory focus theory (Higgins, 1997); emotion regulation (Gross, 1998)
≈32% avoid, or would avoid, medical information about their own health. Highest where it matters most: 41% for Alzheimer’s, 40% for Huntington’s. meta-analysis, 92 studies, 564,000+ participants.
- M6Achieving certainty
Achieving certainty pulls you toward any answer at all, rather than leaving the question open. To resist it, you have to sit with not knowing.
Uncertainty reduction (Berger & Calabrese, 1975); need for closure (Kruglanski, 1990)
65% of headlines classified correctly as real or fake, against 50% for guessing. Cambridge / YouGov, April 2023, n=1,516.
- M7Optimizing risk vs reward
Optimizing risk vs reward pulls you toward whichever reading of the same facts feels safer. To resist it, you have to give the same answer however it is framed.
Prospect theory (Kahneman & Tversky, 1979); reinforcement sensitivity (Gray, 1990)
72% vs 22% took the identical certain option depending only on framing: 72% when described as lives saved, 22% when the same outcomes were described as lives lost. Tversky & Kahneman, 1981.
Yields by default
Resists at most two of the seven, and usually the two that cost nothing in their particular life.
Not a character failure, and not chosen. It is what happens when nothing in your circumstances ever charges you for going along.
Resists three or four
Holds firm against three or four of the seven drivers.
The most misleading tier, because from the inside it feels like the top one. Four drivers held firmly reads as “I think for myself”, and on those four it is true. The others have simply never been tested, which is indistinguishable from having passed them.
Resists five or six
Holds against all but one or two.
The one still standing is almost always M5 or M3 — not knowing, or not being wrong. Both demand something the other five do not, which is accepting a hit to yourself rather than to an argument.
Resists all sevenEstimated
Holds their own judgement against every one of the seven drivers.
Not seven separate virtues. One disposition, showing up in seven different rooms.
Table view
| Tier | Cohort | Share of adults | Per 1,000 people |
|---|---|---|---|
| TIER 01 | Yields by default | 65.1% | 651 |
| TIER 02 | Resists three or four | 27.4% | 274 |
| TIER 03 | Resists five or six | 6.8% | 68 |
| TIER 04 | Resists all seven (estimated) | 0.7% | 7 |
| — | Total | 100.0% | 1,000 |
Why these seven, and not some other list
Earlier drafts of this chart assembled the axes by hand, and every revision hit the same wall: nothing governed what belonged in the set. Any list of failure modes overlaps, and arguing about which ones are “really” distinct has no end.
The CMBM had already solved it. Its seven motivational drivers were derived deductively from established models, checked for discriminant validity, and used to classify roughly 175 documented biases. Taking the axes from that framework gives this chart distinctness by construction rather than by assertion.
What is measured and what is not
Each of the seven driver figures comes from a real study, listed with the chart. They are experiments, meta-analyses and surveys — they measure how each driver behaves in a specific paradigm, not the prevalence of a trait in the population.
The tier split is different. No study measures all seven together, so the split across tiers is an estimate, and it is labelled as one on the face of the chart.
On the one row without a rate. M3 rests on Kahan et al. (2013), which establishes the mechanism and the most uncomfortable finding in the set — that polarisation on loaded data increased with numeracy, so capability was being spent to protect identity rather than to reach the answer. What it does not establish is what share of people are doing it. That row carries direction, not a percentage, and says so.
On the absence of a range. An earlier draft put a floor and a ceiling around the estimate. It was withdrawn: the ceiling rested on a student-skewed sample and the floor multiplied in an assumed rate, so the range carried the authority of a calculation while resting on a guess. A reader who wants one can build it from the driver figures and their own assumption about how correlated these are.
Companion piece: The Die Pie asks how deliberately people treat the body. This one asks whether the thinking is theirs at all.