Teleological bias
Children explain rocks, clouds and animals by what they are for, and adults under time pressure accept purpose-based explanations of nature that they reject when given time.
Children explain rocks, clouds and animals by what they are for, and adults under time pressure accept purpose-based explanations of nature that they reject when given time.
Three claims travel under one name and their evidence is separate. The first is developmental: young children prefer explanations by purpose over explanations by physical cause, and they extend them past artifacts and body parts to rocks, ponds and clouds. Deborah Kelemen called this promiscuous teleology, in contrast to a selective teleology restricted to artifacts and biological traits.
The second is a coexistence claim: schooling suppresses the preference without replacing it, so it should resurface whenever the machinery that suppresses it is loaded. The third is methodological, and it is the one carrying most of the weight in popular accounts. It says that a speeded judgement task exposes the default, because rejecting a purpose-based explanation costs effort and accepting one does not.
Neighbouring entries cover different things. Agent detection asks whether an agent is present. Intentionality bias asks whether an act was done on purpose. Anthropomorphism attributes human traits to non-humans. This entry is about whether a thing exists in order to do something.
Repeats disagree, or the effect holds in some conditions and not others. The other 248.
Mixed Two preregistered studies reproduced the rise in teleological acceptance under speed and found the same speeding damaged accuracy on ordinary false statements more, which is the comparison the original inference rests on.
Roberts, Handley and Polito tested the paradigm rather than the phenomenon. Study 1 recruited 273 native English speakers online through Prolific, capped at undergraduate education, and analysed 200. Study 2 recruited 250 first-year psychology students at Macquarie University and analysed 168. Both used a 3,200 ms limit in the speeded condition and the standard exclusion rules, above 80% accuracy on control statements and at least 75% of test statements answered in time.
The headline effect appeared. In Study 1 accuracy on teleological test statements fell from 0.62 unspeeded to 0.45 speeded, t(198) = 5.57, p < .001, d = 0.79. In Study 2 it fell from 0.61 to 0.40, t(166) = 5.60, p < .001, d = 0.96. People accepted more purpose-based explanations of nature when hurried, as Kelemen and Rosset reported.
The control items behaved differently from the originals. Aggregate control accuracy fell with a larger effect than test accuracy did, d = 1.31 in Study 1 and d = 1.29 in Study 2, and the contrast against test statements was reliable, F(1,198) = 14.18, p < .001 and F(1,166) = 15.55, p < .001. The drop was carried by the false control statements, a category pooling false causal and false teleological items, where accuracy went from 0.95 to 0.86 in Study 1 and 0.95 to 0.85 in Study 2.
Against the true control statements the comparison ran the other way in both studies, F(1,198) = 15.62 and F(1,166) = 20.66, both p < .001. Speed cost more accuracy on teleological statements than on true ones, and less than on false ones. On the authors' reading the speeded condition raises acceptance of what is false, and teleological content adds nothing to that.
Two further results cut at the mechanism. Exclusion was lopsided, 62 dropped in the speeded condition against 11 unspeeded in Study 1, and 76 against 6 in Study 2, and in both studies the people who survived exclusion scored higher on the Cognitive Reflection Test, d = 0.65 and d = 0.47. The participants most prone to intuitive answers are the ones the speeded condition removes.
In the unspeeded conditions, response times for accepted and rejected teleological statements did not differ, t(96) = 0.49, p = .628 in Study 1 and t(110) = -0.43, p = .668 in Study 2, with Bayes factors of 0.172 and 0.077 favouring no difference. Rejecting a purpose-based explanation was no slower than accepting one, which the effort account requires.
The verdict is mixed rather than failed because the endorsement figures themselves keep coming back. Undergraduates accept about half of these statements with unlimited time, the speeded rise reproduces in American, Chinese and Prolific samples, and the developmental result reproduced in Britain. What has not survived is the inference from the speeded contrast to a suppressed cognitive default.
The rain that is there for the plantsillustration
A child explains clouds by saying they exist so that plants can drink. The explanation is by purpose rather than by cause, and it is applied to something with no maker and no job to do.
The answer given under a clockillustration
Asked quickly whether trees make oxygen so that animals can breathe, an educated adult agrees. Given time, the same adult says the oxygen is a by-product and that the sentence had the order the wrong way round.
The mountain that is there for a reasonillustration
Asked why a range sits where it does, a person reaches for what it does: it shelters the valley, it holds the snow. The answer given is what the thing is for rather than what put it there.
Cases marked as illustrations describe nobody in particular. They are there to make the pattern recognisable, not to report that a particular thing happened.
The Developmental Psychology paper used forced choice between two supplied explanations. Sixteen adults and 16 children in each of first, second and fourth grade chose why rocks are pointy, whether because bits of stuff piled up or so that animals would not sit on them. Teleological choices came from 64% of first graders' trials, 69% of second graders' and 56% of fourth graders', against 37% for adults. On non-living natural kinds the split was sharper. Only 5 of 16 adults ever gave a teleological answer, while 13 of 16 first graders and 15 of 16 second graders gave two or more.
Kelemen ran the same task on 48 British children in 2003 and got the same shape. British and American children alike preferred teleological explanations of natural object properties, 59% and 61%. On those properties age mattered and country did not. On biological properties the two countries did differ: American 7- and 8-year-olds endorsed social teleological explanations of animal traits more than British children of the same age did.
Adults show the pattern when the task is made harder. Kelemen and Rosset gave 121 undergraduates 80 sentences to mark good or bad, in a fast speeded condition at 3,200 ms per sentence, a moderate one at 5,000 ms, or unspeeded. Acceptance of unwarranted teleological explanations rose with speed, 47%, 36% and 29%, F(2,118) = 7.86, p < 0.001. Accuracy on control sentences stayed above 90% in all three conditions. A second study of 109 students found poorer inhibitory control on a Stroop task predicted acceptance, and personal belief in God did not.
Lombrozo, Kelemen and Zaitchik ran a child-appropriate version on 41 adults: 17 Alzheimer's patients, 12 healthy elderly adults and 12 young adults. Patients accepted teleological explanations 87% of the time, against 66% for the young group and 73% for the healthy elderly group. Only the gap to the young group was significant; the gap to the healthy elderly group was marginal, t(27) = 1.92, p = .065. Patients still told warranted items from unwarranted ones, 97% against 79%.
The scientist study is the one usually quoted. Kelemen, Rottman and Seston tested 80 physical scientists, 39 women, mean age 36 and mean eight years since the doctorate, publishing in chemistry, geoscience and physics departments at institutions including Boston University, Brown, Columbia, Harvard, MIT and Yale. Biologists were left out on the grounds that teleological language is routine in biology.
Comparison groups were 179 Boston-area undergraduates and 49 community members age-matched to the scientists with bachelor's degrees. Speeded scientists endorsed 29% of the unwarranted teleological sentences against 15% for unspeeded scientists. Overall the scientists sat at 22%, well under the undergraduates at 51% and the community group at 47%, F(2,302) = 58.59, p < .001. The speed effect held across groups with no interaction, F(1,302) = 22.57, p < .001. A second study of 73 humanities scholars from classics, English and history at the same institutions found no difference from the scientists in either condition.
Rottman and five colleagues took the task to Beijing with 105 Chinese students. Speeding raised inaccurate teleological endorsement by 14 percentage points and control errors by 6, an interaction of F(1,103) = 4.30, p = .041.
5 primary research4 replication and reanalysis1 background and reference
The Cognition paper names Promiscuous Teleology as a hypothesis and tests it against Selective Teleology. Study 1 asked 16 preschoolers and 16 undergraduates what 22 photographed objects were for. Children gave functional answers more often than adults, 68% against 55%, though the paper reports that age effect as marginal, F(1,30) = 3.94, P < 0.06. Study 2 asked 24 preschoolers and 24 undergraduates whether 16 photographed objects were 'made for something', and children said so of natural objects such as icebergs and clouds 73% of the time against adults' 33%. A second 1999 paper in Developmental Psychology supplied the result the idea is remembered by, the pointy rocks.
The claim, and the replication this entry reads its figures from.
The structured effect size is one number from one side of the argument, and it needs its caveat read with it. It is Roberts and colleagues' Study 1 drop in accuracy on teleological test statements between the unspeeded and speeded conditions, as the published article's Table 3 prints it. It is not a difference between a teleological effect and a control effect, and the control effect in the same study is larger.
No matching figure sits in the `original` field because Kelemen, Rottman and Seston report the condition effect as an eta squared of .07 and Kelemen and Rosset report an F with no effect size, and pairing a d with an eta squared would invent a comparison neither paper makes.
The scientist sample is smaller than its fame suggests. Eighty scientists split across two conditions leaves about 40 per cell, and the 29% against 15% contrast rests on that. Nine scientists, 51 undergraduates and 11 community members were excluded before analysis, and the undergraduate exclusion rate is the pattern Roberts and colleagues later argued biases the comparison. Three test sentences were dropped from all analyses after six independent scientific experts failed to agree that they were false, and the paper says including them does not change the pattern.
Almost every adult result comes from one task with one family of stimuli, written by the same laboratory. The Chinese replication used a translation of it, and 11 of its 100 judgement items were dropped before analysis. Only eight of those, all control items, went for ambiguous translation; the other three were test sentences already due for removal under the 2013 paper's own rule. Six items of the natural selection inventory, four of the geoscience inventory and both Mother Nature belief items were dropped for ambiguous translation as well. The paper says including everything does not change the pattern.
The children's studies are forced choices between two explanations an experimenter supplies, with 16 to 24 children per age group. They record which of two offered accounts a child prefers, not what a child would say unprompted.
The published Collabra version of Roberts and colleagues was read in full on 21 September 2026, as a PDF of the version of record posted on Vince Polito's own publications page. The publisher's own site still refuses the fetcher, returning a Cloudflare challenge to every user agent tried, at both the article page and the article PDF URL that Unpaywall gives. Every figure on this page now comes from the published article rather than from the authors' earlier manuscript. The two differ in the printed precision of their effect sizes, and the published table's values are the ones used here.
Tiililä, Schlottmann and Bechlivanidis published a linguistic challenge in Cognition in 2026. Their abstract, read at Europe PMC and at Semantic Scholar, reports two studies with 896 participants in total, finds endorsement highest for 'in order to' phrasing and lower when the connective names a purpose or a causal origin, and concludes that endorsement may reflect semantic ambiguity.
Nothing beyond the abstract is reported here, but the article is not paywalled. Crossref records it as open under a CC BY licence, and every route to the file failed for another reason on 21 September 2026. ScienceDirect answered a bot challenge with 403 to three user agents. The London School of Economics repository copy that OpenAlex and Unpaywall both list has been withdrawn, and its OAI record is marked deleted.
The SSRN accepted version sits behind a Cloudflare challenge, and University College London's repository is behind one too and holds no copy that search reaches. The Internet Archive was offline. The authors' OSF page is live and public, but it holds two processed data files and no manuscript.
Searches on 21 September 2026 covered Crossref, PubMed, Europe PMC, OpenAlex, Unpaywall, CORE, OSF, Semantic Scholar and scholar.archive.org. No meta-analysis, registered replication report or multi-site project on teleological bias was located. Rottman and colleagues' China paper mentions one failure to replicate, credited to Rottman, Greenway, Rose, Finke and Kelemen as unpublished data; no publication under those authors was found, and nothing is known here about what it did.
Mills and Frowley's study of Irish participants was verified at Crossref as The Irish Journal of Psychology, volume 35, issue 2-3, pages 121 to 132, and was not opened. Taylor and Francis served a Cloudflare challenge on 21 September 2026 and no repository copy was found. The authors' initials were wrong on this page until now, and Crossref and Roberts and colleagues' reference list agree they are R. Mills and J. Frowley.
The commonest error is the headline that scientists are as teleological as everyone else. They are not. The scientists in Kelemen, Rottman and Seston's study accepted 22% of the unwarranted teleological sentences where undergraduates accepted 51%. The finding is that speeding moves scientists as much as it moves anyone, from 15% to 29%, not that expertise fails to help. It helps by more than half.
The second error runs the other way, and the paper makes the point itself. Humanities scholars did not differ from the scientists in either condition, on less scientific knowledge, at 30% speeded and 18% unspeeded. What separates both professor groups from undergraduates is years of education, and specialising in physics or chemistry adds nothing on top.
The third is to treat the speeded task as a window onto a default. That reading needs rejecting a purpose-based explanation to cost time, and in two studies with unlimited time it did not, with Bayes factors of 0.172 and 0.077 against any difference. It also needs the speeded condition to spare ordinary false statements, and in the same two studies it did not spare them.
A fourth is to file this under religion, though the record here is mixed rather than empty. Belief in God did not predict teleological endorsement in Kelemen and Rosset's second study, did not predict it among the scientists or the humanities scholars, and did not predict it in the Chinese sample, where mean belief was 1.81 on a five-point scale.
It did predict it in the 2013 paper's pooled college and community sample, and in Roberts and colleagues' Study 1. The belief that predicted endorsement in all three of the 2013 regressions was agreement that 'Nature is a powerful being', and in the two professional samples it did so where belief in God did not.
A fifth is to read the Alzheimer's result as a return to childhood. Lombrozo, Kelemen and Zaitchik's patients still separated warranted from unwarranted items, 97% against 79%, and the authors' reading is that childhood preferences never went away rather than that new ones were built.
A sixth concerns what the sentences mean. Tiililä, Schlottmann and Bechlivanidis argue that saying yes to 'the Earth has an ozone layer in order to protect it from UV light' need not be a claim about how the ozone layer came to exist, because the phrasing also supports a reading about what the layer does. On their account a share of the endorsement is agreement with a relation, not a belief about design.
Every claim on this page was held against these sources on 2026-09-21. Nothing here is written from memory. Written and checked by Krishna Chagti; how entries are written and corrected. Found a mistake? Report it, with the sentence and a source.