Ikigai Statistics 2026

We analysed anonymous response data from the free ikigai assessment at ikigain.org/test. The instrument is an 18-item inventory scoring the four pillars of the modern Western synthesis of ikigai popularised by Marc Winn in 2014^[6]: Passion (what one loves), Mission (what one believes the world needs), Vocation (what one is good at), and Profession (what one can be paid for). Each answered item contributes one unit to exactly one pillar, so raw pillar scores are integer counts summing to exactly 18 (verified: mean = 18.00, SD = 0.00 across all 4,105 respondents in this window). The output is a composition — how attention is divided, not how much of it there is.

Scoring history — and why this revision exists. Our first release analysed 3,915 respondents scored under an earlier answer-to-pillar mapping, revised on 1 April 2026. Under that earlier mapping the Mission–Profession gap was 8.1 points; under the current mapping it is 3.1 points (§3). We restrict this revision to assessments completed on or after 1 April 2026 so that every response is scored under the live item set, and we flag the discrepancy openly rather than let the original number stand: an internal methods audit found it was mapping-dependent, not a stable population fact.

Compositional-data correction. Four shares summing to 100% cannot vary independently; ordinary Pearson correlations between them carry a structural negative bias from the unit-sum constraint alone — a problem known since Pearson’s 1897 work on spurious correlation and formalised in the compositional-data framework (Aitchison 1986^[9]; Pawlowsky-Glahn & Egozcue 2015^[10]). All correlations below are computed on centered log-ratio (CLR) coordinates; zero scores are handled with a standard multiplicative pseudo-count of 0.5.

The permutation-null test. CLR removes the unit-sum bias but a subtler mechanical effect remains: any two pillars with different means show some spurious negative CLR correlation purely because the constraint interacts with pillars of different average size. To isolate it, we built a permutation null: 2,000 iterations, each independently shuffling every pillar column across respondents — destroying real within-person association while preserving each pillar’s marginal distribution and the sum constraint exactly. An observed correlation counts as real signal only if it falls outside the null’s 95% band — read in both directions: more negative than the null is a genuine trade-off; less negative is a genuine complementarity.

ILR balances, cluster validation, bootstrap. As an artifact-free cross-check we report three orthogonal isometric log-ratio balances (Egozcue et al. 2003^[11]). To test the ten-archetype taxonomy we ran k-means on CLR coordinates for k = 2…10, scored with silhouette coefficients (Rousseeuw 1987^[12]). All means, correlations, and the imbalance index carry bootstrap 95% confidence intervals (10,000 resamples; 2,000 for the correlation matrix and permutation null; fixed seed).

All scores were de-identified at extraction; the dataset contains no demographic variables, no IP addresses, and no free-text responses. Aggregate results only; the companion data tables are published below (§9). Researchers may request aggregate cross-tabulations or the full bootstrap output at hello@ikigain.org.

For background on the framework itself — including why the four-circle diagram is a Western synthesis rather than the original Japanese concept — see our editorial essay on the meaning of ikigai or the What is Ikigai pillar guide.

If respondents split attention equally, each pillar would receive 25%. No pillar sits exactly there, but the deviations are modest: Mission receives the largest mean share (26.8%) and Passion the smallest (23.5%) — a gap of 3.3 points [2.9–3.8]. Profession, the pillar our first release singled out as most neglected, is essentially tied with Passion and is not the most under-funded pillar in current data.

This is a substantial, deliberate correction. The v1 estimate (8.1 points, n = 3,915) was computed under a scoring mapping retired on 1 April 2026; recomputing the identical comparison under the live mapping yields 3.1 points — roughly two-fifths of the original size. Stated plainly: the “modern broke purpose-seeker” narrative, as we first quantified it, was substantially a property of a scoring mapping since revised, not a stable feature of the population. The imbalance finding (§6) is the more robust of our original claims, and we now lead with it.

The direction of the pillar ordering still echoes what scholars of the original Japanese concept have documented. Gordon Mathews’s ethnography found Japanese informants framing ikigai in terms of contribution and self-realisation rather than income^[2]; Mieko Kamiya’s foundational 1966 monograph treated the concept as distinct from labour-market exchange^[1]. Contribution still outranks commerce in our data — just by a modest margin, not a dramatic one.

When shares are CLR-transformed and each pair is tested against its own permutation null, five of six pairs carry signal beyond the fixed-budget artifact — but not all in the same direction. Three are genuine trade-offs: Passion × Mission (r = −0.50, null band [−0.39, −0.34]), Vocation ×  Profession (−0.42, null [−0.32, −0.27]), and Mission ×  Vocation (−0.34, a small excess over its null). One pair — Passion × Profession — is statistically indistinguishable from the artifact alone.

The v1 release emphasised Vocation × Profession as the headline trade-off. That pair remains a real and second-strongest tension — the “good at it but not paid for it” pattern is not wrong — but under the current methodology the largest genuine tension is Passion versus Mission: within a fixed attention budget, what one loves and what one believes the world needs compete harder than any other pair. The talented underearner has company: the torn idealist.

Reading the permutation null in both directions surfaces a finding a one-directional analysis would miss: Mission and Profession are allies in the data. People who allocate heavily to “what the world needs” allocate more to “what can be paid for” than the budget constraint alone would produce — consistent with a population for whom contribution and livelihood are entangled rather than opposed (think: the person whose sense of what the world needs is their profession). Passion × Vocation shows the same complementary pattern: loving a thing and being good at it travel together.

Our first release claimed the opposite — that every pillar traded inversely against Profession, which we called “the universal opponent.” Under the current methodology and the bidirectional null, that framing is not merely unsupported; it is contradicted in the Mission direction specifically. We retract it, and report the Mission × Profession complementarity as a finding in its own right.

The ikigain.org scoring routine assigns each respondent a primary archetype from which pillar or pillars dominate their allocation. We tested whether those ten labels correspond to natural structure: they do not. No clustering solution between k = 2 and k = 10 reaches a silhouette score of 0.28 — well under the ~0.5 conventionally read as distinct clusters (Rousseeuw 1987^[12]) — and the best-scoring k is 4, not 10. The population is better described as a continuum in allocation space. The archetypes are a useful communication device built on a scoring rule — like compass directions on a round landscape — not a discovered typology, and we think honesty about that distinction matters.

The distribution across the ten labels is nonetheless highly non-uniform: Purpose-Driven Leaders (18.5%) and Skilled Experts (13.8%) together hold 32.3% of the sample, against 20% for any two labels under a uniform null (χ² = 570.9, df = 9, p ≈ 10⁻¹¹⁷). This concentration is far smaller than the 43.9% our first release reported under the retired mapping, but remains overwhelming: respondents are not spread evenly across the labels, even though the labels are not clusters.

Across the current mapping and a reconstruction of the earlier era, the mean imbalance index — each person’s largest pillar share minus their smallest — sits at 22–24 points. The pillar-mean story shrank when the mapping changed; this one did not move. It is, in our assessment, the single most durable empirical claim in this dataset: people who take a purpose test allocate attention lopsidedly, not evenly, whatever the diagram’s four symmetric circles imply.

Given that the four-circle diagram is itself a Western synthesis (§1 of our methods essay) rather than an artifact of the Japanese ikigai literature, we see no strong reason to treat perfect balance as the framework’s implicit target. We suggest reading the four-pillar allocation as a diagnostic of where attention currently sits — useful whether or not balance is desirable for a given person — rather than as a scorecard against an equal-split ideal.

1. Fixed-budget instruments need more than a CLR transform. The unit-sum correction is necessary but not sufficient: a permutation null — read in both directions — is what separates genuine covariation from the milder mechanical effect of unequal pillar means. Of six pairs, three are real trade-offs, two are real complementarities, one is pure artifact. Collapsing “beyond the null” into a single “trade-off” bucket — as our v1 effectively did — loses half the story.

2. A typology built on a scoring rule should be validated before being presented as discovered. Ours does not validate as clusters, and we say so. The labels remain useful for communication — people navigate by named landmarks — but the honest description of this population is a continuum with two heavily-populated regions (Mission-dominant and Vocation-dominant allocations).

3. The durable substantive claim is imbalance. Which single pillar is “most neglected” proved sensitive to a scoring revision; that people are reliably lopsided by ~23 points did not. If the four-circle diagram is read as prescribing equal attention to loving, contributing, excelling, and earning, the data are clear that almost no one lives that way — and the Kamiya tradition, which frames ikigai as orientation rather than equilibrium^[1], never asked them to.

Finally, a note on why we published a correction rather than quietly updating numbers. The point of publishing data about purpose is to be believed; the cost of being believed is being auditable. The v1 headline did not survive our own audit, so it does not survive in our publication either. The findings above are the ones that did.

Self-selection. Respondents arrived at a free online purpose test; they are not a random sample of any general population. The data describe the audience the self-help industry reaches — not the world at large.

Scoring-mapping sensitivity. As §2 details, the magnitude (though not the qualitative direction) of the pillar-mean findings changed materially between scoring eras. Point estimates from fixed-budget instruments should not be read as stable population parameters without a scoring-version sensitivity check — we recommend future releases report one, as this one does.

Archetype-assignment circularity. The per-archetype radar profiles (Figure 8) are partly a visualisation of the assignment rule. The non-trivial findings are the uneven distribution across labels and the within-archetype imbalance — neither forced by the rule.

No demographics; cross-sectional; single instrument. We collected no age, gender, country, or income data; all data are single-occasion; and all findings are internal to this 18-item allocator. Cross-validation against a validated level measure of ikigai — the Ikigai-9 (Imai, Osada & Nishi 2012^[13], validated in English by Fido, Kotera & Asano^[14] and in German by Hajek et al.^[15]) — is the highest-priority next step: pairing an allocation instrument with a level instrument would test whether imbalance predicts lower reported well-being.

Construct caveat. This is a study of the popular Western four-pillar synthesis that borrows the ikigai name — not of the original Japanese psychological construct studied by Kamiya and measured by the Ikigai-9. The two are related in spirit but are not the same measurement target. Longitudinal work linking the broader construct to health outcomes (Ohsaki cohort^[3]; JACC study^[8]) does not automatically transfer to the four-circle framework.