ENS-05706; No of Pages 11

Evolution and Human Behavior xx (2011) xxx – xxx

1 Original Article

2 Prestige-biased cultural learning: bystander's differential attention to

3 potential models influences children's learning☆
4 Maciej Chudek⁎, Sarah Heller, Susan Birch, Joseph Henrich
5 Department of Psychology, University of British Columbia, 2136 West Mall, Vancouver, Canada
6 Initial receipt 6 January 2011; final revision received 12 May 2011

7 Abstract

8 Reasoning about the evolution of our species' capacity for cumulative cultural learning has led culture–gene coevolutionary (CGC)
9 theorists to predict that humans should possess several learning biases which robustly enhance the fitness of cultural learners. Meanwhile,
10 developmental psychologists have recently begun using subtle experimental procedures to probe the learning biases that young children
11 actually possess — a methodology ripe for testing CGC. Here we report the first direct tests in children of CGC's prediction of prestige bias,
12 a tendency to learn from individuals to whom others have preferentially attended, learned or deferred. Our first study showed that the odds of
13 3- and 4-year-old children learning from an adult model to whom bystanders had previously preferentially attended for 10 s (the prestigious
14 model) were over twice those of their learning from a model whom bystanders ignored. Moreover, this effect appears domain-sensitive: in
15 Study 2 when bystanders preferentially observed a prestigious model using artifacts, she was learned from more often on subsequent artifact-
16 use tasks (odds almost five times greater) but not on food-preference tasks, while the reverse was true of a model who received preferential
17 bystander attention while expressing food preferences.
18 © 2011 Published by Elsevier Inc.
19
20 Keywords: Prestige bias; Selective imitation; Children; Learning

21

22 Human cognition is distinct in the degree to which it is a priori predictions about such biases (i.e., predictions 35
23 shaped by cultural learning (i.e., information learned from entailed by evolutionary theory rather than explanations 36
Q1 24 others), in addition to individual experience and genetics fitted to existing empirical evidence), but so far we know of 37
25 (Boyd & Richerson, 1988; Chow, Poulin-Dubois, & Lewis, no direct tests in children of these predictions. Below we 38
26 2008; Galef & Whiskin, 2008b; Laland & Galef, 2009; review plausible candidates for evolved cultural learning 39
27 Rakoczy, Warneken & Tomasello, 2009; Wellman, Cross & biases and present direct tests of one — prestige bias — in 3- 40
28 Watson, 2001). The cognitive mechanisms that shape and and 4-year-old participants. 41
29 bias the acquisition of cultural information (which includes Several important lines of theoretical work have sketched 42
30 emotional responses, food preferences, cognitive and cognitive foundations for our species' capacity for cumula- 43
31 behavioral heuristics, etc.) have broad relevance to under- tive cultural learning. These accounts variously emphasize, 44
32 standing human cognition and behavior. A key step in for instance, intention reading and attention sharing 45
33 establishing the generality of such biases is probing their (Tomasello, Carpenter & Behne, 2005), ostensive pedagogy 46
34 developmental trajectory. Evolutionary theorists have made (Csibra & Gergely, 2009; Gergely & Csibra, 2005), 47
cognitive fluidity (Mithen, 1996) and epistemic vigilance 48
(Sperber, 2006) facilitated by mental time travel (Boyer, 49

☆
2008). Among these, culture–gene coevolutionary (CGC) 50
This research was conducted in accordance with University of British theories focus on and model the evolutionary dynamics 51
Columbia behavioral research standards, with approval of the research ethics
board. Behavioral Research Ethics Board reference number: H08-01533.
facing an emerging cultural species, in particular the 52
⁎ Corresponding author. Tel.: +1 604 827 4416. interaction of genetic and cultural inheritance systems 53
E-mail address: maciek@psych.ubc.ca (M. Chudek). (Boyd & Richerson, 1985) and the learning biases these 54

1090-5138/$ – see front matter © 2011 Published by Elsevier Inc.

doi:10.1016/j.evolhumbehav.2011.05.005
 2 M. Chudek et al. / Evolution and Human Behavior xx (2011) xxx–xxx

55 interactions select for (Cavalli-Sforza & Feldman, 1981; or, more likely, both. CGC's key prediction is that cues of 107
56 Eriksson, Enquist & Ghirlanda, 2007; Kendal, Giraldeau & differential attention and/or deference, alongside other 108
57 Laland, 2009; Mesoudi, 2009; Richerson & Boyd, 2005). sources of information about model quality, will be exploited 109
58 By bringing together empirical evidence of human by members of a cultural species because they reliably 110
59 ancestral history and evolutionary models focused on discriminate better from worse models across societies 111
60 understanding our capacities for cultural learning, CGC and epochs. 112
61 theories have derived predictions supported by a wide range Among adults, evidence for CGC's predicted biases (see 113
62 of evidence from social psychology, economics, field studies Henrich & Henrich, 2007, for a review) has emerged from 114
63 and paleoarcheology (Galef & Whiskin, 2008a; Laland, social psychology (Henrich & Gil-White, 2001), behavioral 115
64 2004; Mesoudi, 2009; Powell, Shennan & Thomas, 2009; economics (Pingle & Day, 1996), experimental anthropol- 116
65 Richerson & Boyd, 2005; for a review, see Henrich & ogy (Efferson, Lalive, Richerson, McElreath & Lubell, 117
66 Henrich, 2007; Richerson & Boyd, 2005: chapter 2). CGC 2008), field studies (Rogers, 1995) and even corollaries in 118
67 approaches suggest, among others things, that learners non-human animals (Galef & Whiskin, 2008a; Horner, 119
68 should be selective about who they attend to for the purpose Proctor, Bonnie, Whiten & de Waal, 2010). 120
69 of cultural learning. They specify a suite of hypotheses about Recent investigations into children's strategies for 121
70 which cultural learning strategies most effectively extract extracting information from their social environment (for 122
71 useful, adaptive information. a recent review see Gelman, 2009) also provide support 123
72 Strategies concerning from whom to learn are termed for CGC predictions. Predictions of an innate or rapidly 124
73 “model biases.” Some individuals are just better in certain acquired skill bias are supported by young children's 125
74 domains, or possess more relevant information, and it pays to tracking and preferential learning from more accurate 126
75 learn from them. Alongside cues based on age, sex, health models (e.g., Birch, Vauthier & Bloom, 2008; Brosseau- 127
76 and dialect (cuing ethnicity), CGC specifies three candidates Liard & Birch, in press; Clement, Koenig & Harris, 2004; 128
77 for evolved model biases: skill bias, success bias and prestige Corriveau & Harris, 2009; Koenig & Harris, 2005), more 129
78 bias. “Skill bias” entails learners selecting models by direct confident models (Birch, Akmal & Frampton, 2010; 130
79 perception of their competence, which can be inaccurate (i.e., Jaswal & Malone, 2007; Sabbagh & Baldwin, 2001), 131
80 when it is not obvious to naïve learners how to judge artifact makers about their artifacts (Jaswal, 2006) and, 132
81 competence) and costly (i.e., when careful observation of generally, adults over children, but accurate children over 133
82 many individuals is needed to gauge skill differentials). inaccurate adults (Jaswal & Neely, 2006). Similarly, 134
83 “Success bias” entails learners selecting models by the predictions about models' dialect as a cue in cultural 135
84 accumulated symbols of their success, which can vary learning (Boyd & Richerson, 1987; McElreath, Boyd & 136
85 between societies; for instance, greater wealth, fancier Richerson, 2003) are supported by developmental in- 137
86 ornamentation (Malinowski, 1922) or bigger yams (Kaberry, vestigations of selective learning (Kinzler, Dupoux & 138
87 1941). “Prestige bias” entails learners preferring information Spelke, 2007; Shutts, Kinzler, McKee & Spelke, 2009). 139
88 from models to whom other learners have preferentially While these findings fit CGC predictions prima facie, the 140
89 attended or deferred (Henrich & Gil-White, 2001). Prestige studies were typically designed to glean proximate 141
90 bias facilitates more accurate and rapid learning by insights rather than test evolutionary predictions about 142
91 capitalizing on others' knowledge about who is worthy of biased cultural learning. They are, however, quite 143
92 attention. Because others' preference for better-quality consistent with a large body of earlier theoretical work 144
93 models is, for a cultural species, fairly reliable across (e.g., Boyd & Richerson, 2005). 145
94 generations and cultures, prestige bias, as an adaptation for The present study directly tests for prestige-biased 146
95 exploiting this regularity, is a good candidate for a cultural learning in 3- and 4-year-olds. To date, the 147
96 genetically evolved, cultural learning bias. developmental evidence most relevant to prestige bias 148
97 We direct readers to Henrich and Gil-White (2001) for a comes from Fusaro and Harris (2008). Their 4-year-old 149
98 complete description of prestige bias, but include here a participants saw two models labeling the same object 150
99 caution aimed at a common misunderstanding. Prestige bias, differently while bystanders non-verbally either endorsed 151
100 a technical coinage, does not denote prestige's usual English (nodding and smiling) or denied (shaking their heads) the 152
101 meaning (an acknowledged status difference); rather, it models' statements. Children preferentially learned from 153
102 refers to learners' preference for inferring cultural informa- the affectively endorsed model — even on subsequent tests 154
103 tion from whomever receives more attention and/or freely without bystanders present. Yet this design was not 155
104 conferred deference from other learners.1 This difference intended to and consequently did not directly test prestige 156
105 results when learners prefer to attend to and hang around bias for two reasons. First, bystanders gave models 157
106 their more skillful peers, avoid attending to less skillful ones differential assent (via their affective displays), rather 158
than purely different amounts of attention. This assent 159
potentially endorsed the model and their message, making 160
1
In our experiment, the “prestigious model” is distinct only by it impossible (for the purpose of testing adaptive theories of 161
receiving bystanders' preferential gaze. cultural learning) to distinguish prestige bias from skill bias 162
 M. Chudek et al. / Evolution and Human Behavior xx (2011) xxx–xxx 3

163 or conformist bias — a distinct learning mechanism

2
2.1.2. Procedure 204
164 (Eriksson et al., 2007; Henrich & Boyd, 1998; Kendal et Participants watched an “attentional cuing” clip, where 205
165 al., 2009) which may be amenable to different develop- two models received unequal bystander attention. In this 206
166 mental investigations (for instance, some insight can be cuing scene two bystanders stood between the models, 207
167 drawn from Corriveau & Harris, 2010; Corriveau, Fusaro & attending to only one of them — the “prestigious model.” 208
168 Harris, 2009) . Second, children only learned novel labels; This prestige cuing was followed by four 10-s “test” clips, 209
169 thus the observed effects may apply only to language where those two models demonstrated different behaviors, 210
170 learning, rather than constituting the broader learning bias preferences and labels. In all test clips solitary models 211
171 predicted by CGC. Language acquisition may be a special demonstrated their preference towards an object; then 212
172 domain of learning (e.g., Pinker, 1994), particularly participants' own preferences toward those same stimuli 213
173 because it concerns coordinating with one's group rather were recorded. The order in which models appeared and the 214
174 than adapting to the non-social environment. For example, identity of the prestigious model were counter-balanced 215
175 while some fungi really are better (and safer) to eat than across participants. Participants observed the scenes in the 216
176 others (and people can learn this culturally), whether one following order (see Fig. 1). 217
Q2 177 should call them mushrooms or 茸 depends only on what Food Choice: Models made a disgusted face at either 218
178 others call them. Furthermore, linguistic disagreements large round or small square crackers and happily sampled the 219
179 between models may be a cue to ethnic differences, which other. Participants saw this scene before the cuing scene but 220
180 have also been formally predicted (McElreath et al., 2003) only later, during “free play,” were they offered a choice 221
181 and shown (Kinzler et al., 2007; Shutts et al., 2009) to bias between the two crackers; they were asked: “Would you like 222
182 children's behavior. a snack? Which of these would you like?” This element of 223
our design let us probe whether children apply prestige 224

183 1. Predictions information retrospectively. 225

Attentional Cue: Each model played with a toy in 226

184 We set out to test the hypotheses that young learners are different ways, on opposite sides of a room. Two observers 227
entered and for 10 s stood between and slightly behind the 228

185 (1) prestige-biased: they will preferentially learn from models, angled toward and watching the prestigious model. 229

186 models to whom bystanders pay more visual Artifact Use: Models interacted with a novel toy by 230

187 attention (gaze) even without explicit endorsement, delightedly using either colored balls or blocks. Participants 231

188 and were presented with the same apparatus, offered a choice of 232

189 (2) prestige-biased in various domains, including the balls or blocks and asked: “Can you show me how to play 233

190 potentially costly behavioral domains: they will with this?” 234

191 learn preferences and behaviors, not just language, Beverage Choice: Models made a disgusted face at either 235

192 including those in potentially costly domains like a cup of dyed blue or yellow water and drank from the other. 236

193 food and drink choices. Participants were offered the same two choices and asked, 237
“Would you like a drink?” or if unwilling to drink: “Which 238
do you think is better to drink?” 239
194 2. Study 1 Novel Label Preference: Each model labeled a different 240
object with the same name — a “stroop.” Participants were 241
195 2.1. Method
presented with both objects and asked “Can you give me 242
196 2.1.1. Participants the stroop?” 243
197 We measured the selective learning of 23 children (mean Explicit Questions: After a few minutes of “free play,” 244
198 age=50.4 months, S.D.=5.8 months; 12 girls) recruited from participants were shown photos of the models and asked: 245
199 a participant database at a public university. Data from one “Who would you rather play with” and “Who do you think is 246
200 boy who did not complete the experiment were excluded. more popular, who has more friends?” 247
201 The majority of participants were Caucasian or Asian; all
202 were from households where English was the main
203 language spoken. 3. Results and discussion 248

Since participants always made binary choices, with each 249

2
Note: this is again a technical term, referring to a non-linear choice representing an implicit endorsement of one model as 250
relationship between the frequency of a cultural trait and the probability of preferred, we analyzed our data using logistic regression. 251
its being learned. Readers should be careful not to confuse this with other Since each participant made multiple, related binary choices, 252
uses of the word conformity, which can involve concerns about we compensated for the non-independence of these repeated 253
coordination, signaling, explicit norms and social punishment. Conformist
bias is also indirectly supported (i.e., these were not designed to test this
observations by calculating clustered robust standard errors 254

evolutionary prediction, but their results support it) by developmental (clustering on individuals), using standard techniques 255
evidence (Corriveau & Harris, 2009). (White, 1980). To compare different logistic regressions, 256
 4 M. Chudek et al. / Evolution and Human Behavior xx (2011) xxx–xxx

Q6 Fig. 1. Scenes that children saw in Study 1 in the order they saw them.

257 we conducted log-likelihood ratio tests;3 the significance of was the prestigious model (p=.48). Since prestige (i.e., 295
258 coefficients was judged by their Z-distributed ratio to their bystander attention) significantly predicted learning across 296
259 standard errors. For each analysis, we started from a all measures pooled together, we also examined theoretically 297
260 regression containing all the predictors (reported in interesting subsets. 298
261 Table 1), then removed non-significant predictors only if
262 doing so did not significantly diminish the predictive power 3.2. Is learning prestige-biased across domains? 299
263 of the regression (these parsimonious models are reported in
264 Table 2). Participants witnessed models receiving unequal attention 300
265 We conducted two kinds of regressions. First we from two bystanders while demonstrating different tech- 301
266 regressed which model children learned from onto whether niques for using an artifact; however, our non-explicit tests 302
267 that model was the “prestigious model” (i.e., the person that fell into three domains: (1) artifact use (involving a different 303
268 child had seen bystanders preferentially watching) and our artifact from the cuing scene); (2) food and drink preferences 304
269 covariates (age, sex and the order in which models (i.e., the food and drink measures taken together as indexing 305
270 appeared). The resulting coefficient of the “prestige” a single construct); and (3) linguistic labels for novel objects. 306
271 predictor estimates the effect size of prestige bias (how In both behavioral domains (i.e., artifact and food/drink), 307
272 many times greater a model's odds of being learned from are, children were significantly more likely to learn from the 308
273 if she's the prestigious model; reported as the odds ratio, prestigious model; the odds of them doing so were 8.25 309
274 OR), controlling for the covariates' effect on participants' (CI.95=[1.15–59.00]) times greater for artifact use and 4.09 310
275 model choice. Next we regressed whether the prestigious (CI.95=[1.02–16.38]) times greater for food and drink 311
276 model was learned from onto the covariates, to analyze their preferences pooled together. When analyzed independently, 312
277 effect on participants' proclivity to prestige bias. neither food (OR=6.91, p=.10) nor drink (OR=2.59, p=.32) 313
reached significance alone. Since both measures trended 314
278 3.1. Is children's learning prestige-biased? quite strongly towards prestige-bias, this was plausibly a 315

279 Yes. Our analysis regressed which model participants consequence of our limited statistical power, which we 316

280 learned from on (1) models' “prestige” (i.e., which model address with a larger sample in our second study. Prestige- 317

281 received bystander attention), (2) the order in which models bias did not extend to language learning, the only measure 318

282 appeared, (3) participants' sex and (4) participants' age. As which trended away from prestige bias (OR=.57, p=.60). 319

283 shown in Table 1 under the “Pooled” regression, this

284 produced only one significant predictor: prestige (p=.01). Table 1 t1:1
285 Removal of the non-significant predictors did not signifi- Full models — Logistic regression coefficients (standard errors) t1:2 Q7
286 cantly diminish the regression's predictive power (P[χ2(3)N
Predictors Statistical models t1:3
287 2.1]=.55), as shown in Table 2. In this parsimonious single-
Pooled Artefact Food and drink Labelling t1:4
288 predictor regression (pLLRT=.03), the odds of a cultural
289 model being learned from were 2.37 times (CI.95=[1.22– Prestige .90 (.37)c
2.60 (1.30)b
1.42 (.67)b
−.57 (1.07) t1:5
290 4.58], p=.01) greater if she was the prestigious model. Sex .0 (.34) −1.33 (1.43) 1.30 (.73)a −2.90 (1.48)b t1:6
Order −.56 (.33)a −2.49 (1.50)a −.42 (.57) −1.94 (1.36) t1:7
291 Regressing covariates on whether the prestigious model was Age −.02 (.03) −.08 ( .14) .03 (.06) .01 ( .10) t1:8
292 learned from showed that participants' proclivity to prestige- pLLRT .16 .04 .14 .13
293 bias was not confounded by their age (p=.32), sex (p=.70), n 100 23 44 23
294 the order in which models appeared (p=.11), nor which actor Logistic regression coefficients and their standard errors. All statistical
models regress which actor participants imitated onto the listed predictors.
Prestige encodes which actor was prestigious, Sex is the sex of the
participant, Order encodes which actor appeared first and Age is the
3
That is, we compared the χ2 distributed ratio of each regression participant's age in months. pLLRT is the result of a log-likelihood ratio test
model's log likelihood to one another (reported as χ2 tests) or to that of a of the model's goodness of fit. n is the number of observations on which the
logistic regression model with only a constant predictor (reported as pLLRT ). statistical inference was based, whose non-independence was compensated
Values indicate the probability that the improvement in model fit would arise for by Huber–White clustered robust standard errors. t1:11
by random sampling alone. We refer to statistical models as “regressions” in a
pb.1. t1:12
the text to avoid confusion with the term “model” used to refer to the adults in b
pb.05. t1:13
our videos: the “cultural models”. c
pb.01. t1:14
 M. Chudek et al. / Evolution and Human Behavior xx (2011) xxx–xxx 5

t2:1 Table 2 children's prestige bias is domain-sensitive: children are 351

t2:2 Parsimonious models — Logistic regression coefficients (standard errors) more likely to learn from prestigious models in domains 352
t2:3 Predictors Statistical models more similar to those in which they received the 353
t2:4 Pooled Artefact Food and drink Labelling preferential attention of bystanders. For instance, witnes- 354

t2:5 Prestige .86 (.33)c

2.11 (1.0) b
1.4 (.7)b sing bystanders preferentially observing a model using a 355

t2:6 Sex 1.5 (.7)b −1.84 (.98)a tool could make children more disposed to learn her other 356
pLLRT .03 .02 .04 .04 tool-use techniques, but not her food preferences or other 357
n 100 23 44 23 habits unrelated to tool use. 358
Significant logistic regression coefficients and their standard errors. All Study 2 was designed to control for experimental fatigue 359
statistical models regress which actor participants imitated onto the listed and other order effects and to address the question of domain 360
predictors. Parsimonious models were developed by removing only those sensitivity by manipulating the domain in which prestige 361
non-significant predictors from the full models whose absence did not
significantly diminish the model's log-likelihood. Prestige encodes which
was cued. 362

actor was prestigious, and Sex is the sex of the participant. pLLRT is the result
of a log-likelihood ratio test of the model's goodness of fit. n is the number
of observations on which the statistical inference was based, whose non-
independence was compensated for by Huber–White clustered robust 5. Study 2 363
t2:9 standard errors.
t2:10 a
pb.06.
5.1. Method 364

t2:11 b
pb.05.
5.1.1. Participants 365
t2:12 c
pb.01.
Forty-eight children (mean age=46.6 months, S.D.=7.4 366
months; 25 girls) participated; all were recruited from a 367
320 3.3. Did prestige bias children's explicit answers? participant database. The majority of participants were 368
Caucasian or Asian; all were from households where English 369
321 Children's answers to our explicit questions about which was the main language spoken. 370
322 model they or others would prefer to play with were
323 statistically indistinguishable from chance (pLLRT=.80).
5.1.2. Procedure 371
324 3.4. Does prestige-bias operate retrospectively? In Study 2, we systematically counterbalanced the 372
domain in which prestige (specifically, differential bystander 373
325 Our participants observed models' food preferences be-
attention) was established between artifact-use and food- 374
326 fore the prestige cue, but had their own food preference
and-drink preferences (herein food preferences). After cuing 375
327 tested after. That this measure trended towards prestige-bias
a model's prestige in one of the two domains, we presented 376
328 without reaching statistical significance, though peripheral to
participants with learning tests in both domains (order 377
329 our main findings, may be of interest to researchers
counterbalanced). There were two test trials for each domain 378
330 investigating similar phenomena.
for a total of four test trials. Earlier test trials were designed 379
to minimize the feedback participants received on whether 380
331 4. Discussion: study 1 they had made a good choice to avoid any potential impact 381
this could have on subsequent test trials. To demonstrate that 382
332 Our first study constitutes the first clear experimental prestige biases decisions which children expect will have 383
333 evidence of prestige-biased learning in children, a disposi- real consequences, rather than just their answers to arbitrary 384
334 tion to learn from those to whom others preferentially attend. questions, later test trials involved tangible rewards (i.e., a 385
335 On novel tasks in the absence of any bystanders, the odds of snack to eat and a sticker from the novel machine). 386
336 our participants learning from a model to whom bystanders Subjects were assigned to one of two conditions; each 387
337 had previously preferentially attended were more than twice contained, in order, two “initial cuing scenes” (either artifact 388
338 the odds of their learning from the other model. More use or food preference) in a fixed order; two “no feedback 389
339 precisely, children more often learned from an adult model tests”; one “reminder cue” in the same domain as the initial 390
340 who had received 10 s of bystanders' attention (gaze) than cues; two “tangible consequence tests”; and, finally, two 391
341 one who had not. explicit questions. 392
342 While demonstrating the basic effect, this preliminary During prestige cuing scenes, participants saw videos of 393
343 study also raised some pressing questions. Our artifact-use two female models either interacting with artifacts or 394
344 test registered a much stronger effect than our food- demonstrating food and drink preferences, while three 395
345 preference tests, and our language-learning test registered female bystanders attended to just one of these models (the 396
346 no effect at all, despite the prior experimental evidence “prestigious model”). During tests, participants saw videos 397
347 that children do discriminate their language-learning of these models alone, demonstrating artifact-use techniques 398
Q3 348 models (e.g., Fusaro & Harris, 2008; Jaswal, 2005). or food preferences by making a choice between two options. 399
349 Since the order of our tests was fixed, this may have been The following are detailed descriptions of these components, 400
350 the result of participant fatigue. It is also plausible that in order of appearance. 401
 6 M. Chudek et al. / Evolution and Human Behavior xx (2011) xxx–xxx

402 5.1.2.1. Initial cueing scenes. In each condition, partici- Artifact, Sticker-Machine Test: Models sat behind a 454
403 pants saw two cues from a single domain (artifact or food) “sticker machine,” a grey box with two visually salient 455
404 presented in the following order. orange handles protruding from it. Each model shook their 456
405 Artifact, Drawing: Bystanders watched one of the two head at one handle and then very deliberately reached for 457
406 models drawing with colored crayons on paper. The the other. To ensure participants did not have information 458
407 drawings were indistinguishable on the video. about the success of each technique, clips terminated just 459
408 Artifact, Building: Bystanders watched one of the two before models actually manipulated the machine. Children 460
409 models connecting together small magnetized balls and sticks. were told, “You can get a sticker from this machine if you 461
410 Food, Eating: Bystanders watched models eat similar use it right. Do you want to try using it? Can you show me 462
411 looking crackers, one model selected food from a round how?” When children manipulated either handle, a door 463
412 white plate in front of her, the other from a square one. opened in the sticker machine and children were given the 464
413 Food, Drinking: Bystanders watched models pouring sticker inside. A small proportion of shy children merely 465
414 water into plastic cups from two very similar water coolers pointed to one handle and the experimenter manipulated it 466
415 and then taking a drink. for them. 467
Food, Container Test: Models sat between two white 468

416 5.1.2.2. No feedback tests. Next, all participants were containers, one square and one round. Each model opened 469

417 presented with the following two tests. They saw each model one container towards herself (so participants could not see 470

418 one at a time and then made a forced choice between what the contents) and made a disgusted face and then similarly 471

419 they had seen demonstrated. The order in which they saw the checked the other container but happily reached inside. 472

420 tests and the order in which models appeared were counter- Children were presented with both containers and told “I 473

421 balanced between participants. have two snacks here. You can have one, but you can only 474

422 Artifact, Geo-Board Test: Models sat behind a “geo- choose one. Which one would you like?” In either container 475

423 board”, a small wooden board with a lattice of small nails children found a fruit-flavored gummy candy, which they 476

424 protruding from it; to their left and right were paper plates, were allowed to eat. 477

425 one bearing green plastic blocks and the other green loops.
426 Each model shook their head at one of these plates and then 5.1.2.5. Explicit questions. Children were asked “Who 478

427 happily placed the contents of the other onto the geo-board. would you like to play with?” and then “Who were the girls 479

428 Participants were presented with the same geo-board, blocks looking at?” and answered by pointing to pictures of the 480

429 and loops and asked, “Can you show me how to use this?” If models. If subjects hesitated with the second question, they 481

430 they had not volunteered an answer within 10 s, they were were further prompted: “The girls standing in the middle, did 482

431 prompted, “Which of these do you think you should use?” they look at this girl or this one?” 483

432 Participants did not receive feedback on whether they had

433 made the right choice to avoid biasing later tests. 6. Results and discussion 484
434 Food, Drink Test: Models sat between a white and a black
435 can of otherwise identical canned beverages. Each made a Our analysis here parallels Study 1. Again, regressing 485
436 disgusted face at one and then took a drink from the other. which model was learned from on which model was 486
437 Participants were asked, “Which of these do you think is prestigious (see Table 3) yielded our effect size estimate 487
438 yummier?” Again, participants were not given any feedback for prestige bias, expressed as an odds ratio (OR) estimated 488
439 on their choice or allowed to sample the drink. by the prestige predictor. Here, we put greater emphasis on 489
regressing whether the prestige model was learned from on 490
440 5.1.2.3. Reminder Cue. Next, participants saw an addi- our predictors (see Table 4), allowing us to estimate how 491
441 tional prestige cue in the same domain as their initial cues. each predictor (particularly the domain of the tests and cues) 492
442 The same model was cued as prestigious. changed the strength of prestige bias. In this second set of 493
443 Artifact, Toys: Bystanders watched models play with regressions, the absolute effect size of prestige bias is instead 494
444 identical toys in different ways. estimated by the intercept; in Table 4 we have mean centered 495
445 Food, Disgust: Bystanders watched models make a our variables to make this interpretation straightforward. 496
446 disgusted face towards two identical-looking plates of
447 crackers. Models then crossed the room (still visually 6.1. Was children's learning prestige-biased? 497

448 tracked by bystanders) and took a cracker from the plate Yes. Regressing which model was learned from on 498
449 the other model had disliked. prestige, age, sex, order in which prestigious model 499
appeared, test domain and cue domain produced only one 500
450 5.1.2.4. Tangible consequence tests. Next, all participants significant predictor of model preference, prestige 501
451 were presented with two tangible consequence tests. Again (OR=2.11, CI.95=[1.06–4.22]) (see Table 3: Basic). Remov- 502
452 they saw models one at a time before making a binary forced ing non-significant covariates reduced the predictive power 503
453 choice. Test and model order were counterbalanced. of the regression, but did not substantially change the effect 504
 M. Chudek et al. / Evolution and Human Behavior xx (2011) xxx–xxx 7

t3:1 Table 3
t3:2 Study 2 — Prestige models
t3:3 Predictors Statistical models
t3:4 Basic Parsimonious basic Interaction Parsimonious interaction
t3:5 Prestige 0.75 (0.35) a
0.74 (0.31) b
1.56 (0.47) b
1.64 (0.43)b
t3:6 Sex is male 0.32 (0.33) 0.33 (0.35)
t3:7 Age 0.00 (0.01) 0.00 (0.01)
t3:8 Order 0.35 (0.34) 0.37 (0.35)
t3:9 Cue is food 0.25 (0.32) 0.27 (0.33)
t3:10 Test is food 0.11 (0.26) 0.12 (0.24)
t3:11 Cross-domain 1.01 (0.37)b 1.01 (0.37)b
t3:12 Cross-domaina prestige −1.54 (0.49)b −1.51 (0.48)b
pLLRT .04 .01 .002 .001
Logistic regression coefficients and their standard errors. All statistical models regress which actor participants imitated onto the listed predictors. Prestige
encodes which actor was prestigious, Sex is Male encodes participants' sex, Order encodes which actor appeared first and Age is the participant's age in months.
Cue is Food and Test is Food encode which domain the cue and test were in, respectively. Cross-Domain is true when the test and cue are in different domains
and false otherwise. pLLRT is the result of a log-likelihood ratio test of the regression's goodness of fit. All regressions were based on 191 observations whose
t3:14 non-independence was explicitly modelled by computing Huber–White clustered robust standard errors.
Prestige significantly biases learning across all tests (i.e., prestige, basic regressions). This is particularly true in the same domain as the model's prestige cue
(i.e., prestige, interaction regressions), while the effect is almost zero in the other domain (i.e., Prestige — Cross-Domain⁎Prestige, interaction regressions).
Exponentiating the prestige coefficient yields our effect size measure: the ratio of the odds of imitating the prestigious model to the odds of imitating the
t3:15 other model.
t3:16 a
pb.05.
t3:17 b
pb.01.

505 of the prestige predictor (see Table 3: Parsimonious Basic). trended non-significantly towards prestige bias when 536
506 Regressing whether the prestigious model was learned from children saw artifact cues, artifact tests showed the opposite 537
507 on these covariates did not produce any significant predictors trend when children saw food cues. Each individual test 538
508 of participants' proclivity for prestige biased learning (see showed the same domain-sensitivity effect (see Fig. 3), 539
509 Table 4: Basic).

510 6.2. Was prestige bias stronger in the domain where Table 4 t4:1
511 prestige was cued? Study 2 — Covariate models t4:2
Predictors Statistical models t4:3
512 Yes, in fact it was absent in the other domain. There are
Basic Domain Parsimonious domain t4:4
513 two ways to analyze this effect, each using a variable which
514 encodes whether the test was in the same domain (artifact or Intercept 0.42 (0.16) a a
0.81 (0.21) 0.79 (0.21) a
t4:5
Model 0.43 (0.36) 0.45 (0.37) t4:6
515 food) as attentional cues. First this variable can predict which
Sex 0.01 (0.34) 0.01 (0.35) t4:7
516 model children learned from as a multiplicative interaction Age 0.00 (0.01) 0.00 (0.01) t4:8
517 with model prestige. This approach (summarized in Table 3: Order 0.27 (0.34) 0.27 (0.35) t4:9
518 Interaction and Parsimonious Interaction) produced a Cue −0.22 (0.32) −0.23 (0.33) t4:10
519 significant interaction (b=1.54, S.E.=0.49, pb.01) and Test 0.02 (0.26) −0.03 (0.25) t4:11
Cross-domain −0.74 (0.24)a −0.73 (0.24)a t4:12
520 effect-size estimate in terms of the models: being prestigious p
LLRT .7 .07 .002 t4:13
521 improved a model's odds of being learned from by 4.75
Logistic regression coefficients and their standard errors. All statistical
522 (CI.95=[1.91–11.8]) times on same domain tests, but not
models regress whether participants imitated the prestigious model onto
523 (p=.96) on cross-domain tests. the listed predictors. Model encodes which particular actor was prestigious,
524 A conceptually simpler approach for examining this is to Sex indicates whether the participant was male, Order encodes which actor
525 add “same domain” as a predictor of whether children appeared first and Age is the participant's age in months. Cue and Test
526 learned from the prestigious model. This model (summarized encode which domain the cue and test were in, respectively. Cross-Domain
527 in Table 4: Domain Model) produced only one significant is true when the test and cue are in different domains and false otherwise.
pLLRT is the result of a log-likelihood ratio test of the model's goodness of
528 predictor and effect estimate in terms of proclivity to prestige fit. All tests were based on 191 observations whose non-independence
529 bias: when tests were in the same domain as cues, the odds of was explicitly modelled by computing Huber–White clustered robust
530 children learning from the prestigious model were 2.09 standard errors. t4:14
531 (CI.95=[1.30–3.36]) times greater. This effect shifted the For ease of interpretation, all variables are mean-centered, besides Cross-
532 probability of learning from the prestigious model from Domain. This allows the logistic intercept   the Cross-Domain
{+
coefficient}, to be directly transformed 1 + 1e − b0 to the proportion of
533 about 52% (p=.72) on cross-domain tests to about 69% participants who imitated the prestigious model on tests in the same
534 (pb.001) on same domain tests. These effects are illustrated {cross} domain. t4:15
535 visually in Fig. 2, where it is apparent that though food tests a
pb.01. t4:16
 8 M. Chudek et al. / Evolution and Human Behavior xx (2011) xxx–xxx

0.9
0.8
Probability of imitating prestigous model
0.7
0.6
0.5
0.4
0.3

Artefact Test Food Test Artefact Test Food Test

Artefact Cue Food Cue

Fig. 2. Probability of children imitating the prestigious model with 95% confidence intervals, in the same domain as the prestige cue (darker bars) and the other
domain (lighter bars).

563
562
561
560
559
558
557
556
555
554
553
552
551
550
549
548
547
546 though when analyzed independently this difference only 6.4. Did children explicitly recall who bystanders had 571
564 reached conventional levels of significance for the geo-board attended to? 572
565 test (p=.04).
No. Only 22 (46%) of our participants were able to recall 573

566 6.3. Did prestige bias children's explicit who bystanders had looked at, the remainder either said they 574

567 playmate preferences? did not know or “both”. Of those, only 12 (54%) correctly 575
identified the prestigious model. Who children explicitly 576
568 No. Regressing who children chose as their playmate on identified as having been attended to neither significantly 577
569 the full set of predictors produced no significant effects, predicted who they learned from (p=.55), nor was it 578
570 including prestige (p=.45). predicted by who had been cued as prestigious (p=.64). 579
0.9
0.8
Probability of imitating prestigous model
0.7
0.6
0.5
0.4
0.3

Geo-Board Drink Sticker Machine Containers

Fig. 3. Probability of children imitating the prestigious model on each test, when they had seen a cue in the cross domain (lighter bars) or the same domain (darker
bars). The only test for which this difference was significant when analyzed alone was the geo-board (p=.04).
 M. Chudek et al. / Evolution and Human Behavior xx (2011) xxx–xxx 9

580 7. Discussion: study 2 subjects, suggesting that it may be in place or fully 632
developed prior to 3 years of age. CGC's prediction is 633
581 Study 2 constitutes the first evidence that children's consistent with either an innate4 prestige bias or innate, 634
582 prestige-biased learning is intelligently targeted at domains evolved cognitive mechanisms which make acquiring 635
583 in which models were observed receiving differential prestige bias by individual learning substantially more 636
584 attention. This effect was consistent across multiple in- likely. Our results are consistent with either of these 637
585 stances of learning in the same domain, for choices both with possibilities, although future work could help tease them 638
586 and without immediate tangible consequences. Particularly apart. Although our study is cross-sectional, it does span a 639
587 interesting is the fact that our participants were not able to key developmental window when most children begin 640
588 reliably, explicitly recall which model had received bystand- passing classic (explicit) false-belief tests, a variable 641
589 er attention, nor were their explicit preferences for which correlated with many other cognitive changes (Wellman, 642
590 model they had liked as a playmate influenced by our cuing Cross & Watson, 2001) including social learning prefer- 643
591 of prestige. We tested the domain sensitivity of prestige bias ences (e.g., Fusaro & Harris, 2008). To the extent that our 644
592 using two instances of two learning domains likely to be of participants' age serves as a proxy for these cognitive 645
593 particular importance in an evolutionary context: artifact use developments, they seem to neither potentiate nor 646
594 and dietary preferences. This work opens the door to more strengthen prestige bias. However, further work which 647
595 systematic inquiries using multiple exemplars from many explicitly measures understanding of false beliefs and tests 648
596 different learning domains. prestige bias at other developmental stages will greatly 649
clarify this picture. 650
With regard to cross-domain effects, Henrich and Gil- 651
597 8. General discussion White (2001) predicted that prestigious individuals are 652
“influential, even beyond their domain of expertise.” Recent 653
598 Our findings provide support for the existence of a developmental research (Fawcett & Markson 2010) has 654
599 domain-sensitive prestige bias in children's learning: indicated that 2-year-olds who know a model shares their 655
600 children's learning from cultural models was biased by preferences in one domain (food or television shows) will 656
601 the mere preferential attention of bystanders, particularly only imitate that model's preferences in that same domain, 657
602 on activities similar to those the model had been engaging not the other. This suggests that mere “similarity cues” may 658
603 in when she received bystander attention. These strong not be strongly influential beyond their domain of expertise. 659
604 effects from a minimal manipulation suggest that prestige In our work with “prestige cues”, however, we witnessed an 660
605 bias may be a potent pressure on cultural evolution. As interesting domain-based asymmetry. Our subjects' food and 661
606 predicted (Henrich & Gil-White, 2001), we witnessed drink preferences were prestige-biased when they saw 662
607 biased learning in different domains, including potentially artifact-use cues; however, their artifact-use preferences 663
608 costly dietary preferences. were not prestige-biased by food cues. This raises the 664
609 Our tests bear out a behavioral phenotype predicted a interesting possibility that children's inferences about model 665
610 priori by CGC theorists, i.e., on the basis of evolutionary quality exploit an asymmetric map of the relationships 666
611 reasoning undertaken prior to this phenotype being observed. between learning domains. While future work could 667
612 Importantly, CGC predicts the behavioral consequences of document these relations, at present we merely speculate 668
613 prestige bias but remains agnostic to the cognitive mecha- that children might form cognitive representations of the 669
614 nisms that implement it. The domain sensitivity we observed correlations between skill domains. Such correlations do 670
615 in our second study speaks against the possibility of prestige exist out in the world, because 671
616 bias resulting from general attentional biases. For instance, if
(a) skill at certain tasks (e.g., foraging and dietary 672
617 children's disposition to follow the gaze of bystanders (Flom,
choices, but not weight-lifting) relies on the same 673
618 Lee & Muir, 2007) were simply creating a more salient
underlying traits (i.e., local ecological knowledge), 674
619 representation of the prestigious model in memory, one
(b) skills at certain tasks (e.g., hunting) is influenced by 675
620 would expect that representation to precipitate a bias to learn
performance in others (e.g., healthy dietary 676
621 from that model regardless of the learning domain. Although
choices), while the reverse may not be true (a 677
622 the mechanisms driving prestige bias remain largely opaque,
good hunter can make poor dietary choices), and 678
623 our results suggest that they encode and selectively exploit at
(c) competence at hard tasks implies competence at 679
624 least two pieces of information: the relative amounts of third-
easier ones but not necessarily harder ones (and 680
625 party attention that potential models receive and the domain
vice versa). 681
626 of activity in which they receive it. The fact that children do 682
627 not seem to explicitly recall the former when asked suggests
628 that these mechanisms operate outside conscious awareness
629 or direct introspective access. 4
The concept of innateness is defined differently by different groups
630 The potency of this combination of prestige bias and and easily misunderstood (Samuels, 2004). We mean: it reliably develops, Q4
631 domain specificity did not increase with the age of our insensitive to environmental variations that humans usually encounter.
 10 M. Chudek et al. / Evolution and Human Behavior xx (2011) xxx–xxx

683 If such correlations were sufficiently regular throughout for instance, why famous individuals' suicides increase 739
684 our evolutionary history, selection may have shaped suicide rates in populations of similar age and ethnicity, 740
685 cognitive mechanisms that more readily recognize them, using similar methods (Fu & Yip, 2007; Stack, 1987), or 741
686 generating regular patterns of asymmetric, domain-sensitive how it is that people can become “famous for being famous” 742
687 biased learning in children. Although developmental without any valuable skills, notable success, etc. 743
688 psychologists have given considerable attention to identi- Future laboratory investigations will clarify the proxi- 744
689 fying innate domains in children's learning and reasoning mate mechanisms underlying prestige bias and further 745
690 (e.g., Spelke & Kinzler, 2007), many questions still remain examine its widespread implications for cultural evolution. 746
691 about how children represent the relationships between In particular, we feel there is scope for further work that 747
692 different domains, including their relative difficulty (e.g., explores other domains in which prestige biases learning 748
693 Keil, Lockhart & Schlegel, 2010). Since evolutionary and the relationships between these domains. Working 749
694 speculation is notoriously error prone, confidently predict- across diverse human societies, empirical investigations on 750
695 ing evolved domain sensitivity will require formal evolu- this topic should probe the interaction of prestige with other 751
696 tionary models of the adaptive dynamics of such learning- cues of model quality such as age, self-similarity, and 752
697 domain asymmetries. Once precise predictions have been ethnicity; vary features of the bystanders, especially their 753
698 derived, future developmental investigations could test the age and cues of group membership such as ethnicity and 754
699 possibility that domain sensitivity is a coevolved adaptation accent, to see whether children are more likely to imitate the 755
700 of a cultural species. model preference of bystanders more “like them”; and 756
701 It bears mentioning that our participants saw two things measure the rate at which this effect strengthens with 757
702 simultaneously: bystanders looking towards one model and mounting evidence of preferential bystander attention. 758
703 away from the other one; either or both of these could
704 have driven our effect. We have emphasized the former
References 759
705 here for simplicity, but in fact both mechanisms are
706 consistent with our results and with the predictions made Birch, S. A. J., Akmal, N., & Frampton, K. L. (2010). Two-year-olds are 760
707 by CGC theorists, which require only that bystanders' vigilant of others' nonverbal cues to credibility. Developmental Science, 761
708 attention influence cultural learning. Further work disam- 13(2), 363–369. 762
709 biguating the relative weights of these influences would be Birch, S. A. J., Vauthier, S. A., & Bloom, P. (2008). Three-and four-year- 763
olds spontaneously use others' past performance to guide their learning. 764
710 valuable and interesting.
Cognition, 107(3), 1018–1034. 765
711 Given recent results on ostensive pedagogy (Csibra & Boyd, R., & Richerson, P. (1985). Culture and the Evolutionary Process. 766
712 Gergely, 2009; Gergely & Csibra, 2005; Topál, Gergely, Chicago: University of Chicago Press. 767
713 Miklosi, Erdohegyi & Csibra, 2008, 2009), it also bears Boyd, R., & Richerson, P. (1988). An evolutionary model of social learning: 768
714 mention that, while both our models gave ostensive learning the effects of spatial and temporal variation. In T. Zentall, & B. G. Galef 769
(Eds.), Social Learning: A Psychological and Biological Approaches 770
715 cues (they looked at the camera before each testing clip), our
(pp. 29-48). Hillsdale, NJ: Lawrence Erlbaum Assoc. 771
716 bystanders did not; they simply looked at models. That is, Boyd, R., & Richerson, P. (1987). The evolution of ethnic markers. Cultural 772
717 children seem to infer prestige information from the attention Anthropology, 2, 65–79. 773
718 of bystanders, without bystanders explicitly cueing that “this Boyd, R., & Richerson, P. J. (2005). The Origin and Evolution of Cultures. 774
719 is a learning opportunity”. USA: Oxford University Press. 775
Boyer, P. (2001). Religion Explained: The Human Instincts That Fashion 776
720 Although we tested an a priori evolutionary hypothesis,
Gods, Spirits and Ancestors. London: William Heinnemann. 777
721 we cannot conclude from this study alone that children are Boyer, P. (2008). Evolutionary economics of mental time travel? Trends in 778
722 prestige-biased cultural learners because their prestige- Cognitive Sciences, 12(6), 219–223. 779
723 biased ancestors were favored by natural selection. Chil- Brosseau-Liard P. E., & Birch S. A. J. (in press). You got that right, I bet you 780
724 dren's observed prestige bias may result from cognitive know more and are nicer too! What children infer from others' accuracy. 781
Developmental Science. 782
725 mechanisms selected for entirely different ends. Nonethe-
Cavalli-Sforza, L. L., & Feldman, M. W. (1981). Cultural Transmission and 783
726 less, in the absence of a detailed alternative account of these Evolution: A Quantitative Approach. Chicago: Princeton University Press. 784
727 patterns, this study contributes to a growing body of Chow, V., Poulin-Dubois, D., & Lewis, J. (2008). To see or not to see: 785
728 evidence supporting this CGC's prediction. infants prefer to follow the gaze of the reliable looker. Developmental 786
729 A comprehensive understanding of the role of prestige Science, 11(5), 761–770. 787
Clement, F., Koenig, M., & Harris, P. (2004). The ontogenesis of trust. Mind 788
730 bias in cultural transmission has the potential to shed light on
& Language, 19(4), 360–379. 789
731 many modern psychological and social phenomena besides Corriveau, K. H., Fusaro, M., & Harris, P. L. (2009). Going with the flow: 790
732 children's learning. Formal modeling of biased cultural preschoolers prefer non-dissenters as informants. Psychological Science, 791
733 transmission can readily generate predictions about higher- 20, 372–377. 792
734 level sociological phenomena, such as the diffusion of Corriveau, K. H., & Harris, P. L. (2009). Preschoolers continue to trust a 793
more accurate informant 1 week after exposure to accuracy information. 794
735 innovations (Henrich & Boyd, 2001), the emergence of
Developmental Science, 12(1), 188–193. 795
736 ethnic groups (McElreath et. al. 2003) or the properties of Corriveau, K. H., & Harris, P. L. (2010). Preschoolers (sometimes) defer to 796
737 modern religions (Boyer, 2001; Henrich, 2009). Prestige bias the majority when making simple perceptual judgments. Developmental 797
738 might also contribute to explanations at a sociological level; Psychology, 46(2), 437–445. 798
 M. Chudek et al. / Evolution and Human Behavior xx (2011) xxx–xxx 11

799 Csibra, G., & Gergely, G. (2009). Natural pedagogy. Trends in Cognitive Kinzler, K. D., Dupoux, E., & Spelke, E. S. (2007). The native language of 859
800 Sciences, 13, 148–153. social cognition. Proceedings of the National Academy of Sciences of 860
801 Efferson, C., Lalive, R., Richerson, P. J., McElreath, R., & Lubell, M. (2008). the United States of America, 104(30), 12577–12580. 861
802 Conformists and mavericks: the empirics of frequency-dependent cultural Koenig, M. A., & Harris, P. L. (2005). Preschoolers mistrust ignorant and 862
803 transmission. Evolution and Human Behavior, 29(1), 56–64. inaccurate speakers. Child Development, 76(6), 1261–1277. 863
804 Eriksson, K., Enquist, M., & Ghirlanda, S. (2007). Critical points in Laland, K. N. (2004). Social learning strategies. Learning and Behavior, 32, 864
805 current theory of conformist social learning. Journal of Evolutionary 4–14. 865
806 Psychology, 5, 67–87. Laland, K. N., & Galef, B. G. (2009). The Question of Animal Culture. 866
807 Fawcett, C. A., & Markson, L. (2010). Children reason about shared Cambridge: Harvard Univ. Press. 867
808 preferences. Developmental Psychology, 46(2), 299–309. Malinowski, B. (1922). Argonauts of the Western Pacific. London: 868
809 Flom, R., Lee, K., & Muir, D. (2007). Gaze-Following: Its Development Routledge. 869
810 and Significance. Lawrence Erlbaum. McElreath, R., Boyd, R., & Richerson, P. J. (2003). Shared norms and the 870
811 Fu, K., & Yip, P. S. F. (2007). Long-term impact of celebrity suicide on evolution of ethnic markers. Current Anthropology, 44(1), 122–129. 871
812 suicidal ideation: Results from a population-based study. British Medical Mesoudi, A. (2009). How cultural evolutionary theory can inform social 872
813 Journal, 61(6), 540–546. psychology, and vice versa. Psychological Review, 116(4), 929–952. 873
814 Fusaro, M., & Harris, P. L. (2008). Children assess informant reliability using Mithen, S. (1996). The Prehistory of the Mind. London: Thames & Hudson. 874
815 bystanders' non-verbal cues. Developmental Science, 11(5), 771–777. Pingle, M., & Day, R. H. (1996). Modes of economizing behavior: 875
816 Galef, B., & Whiskin, E. (2008a). ‘Conformity’ in Norway rats? Animal experimental evidence. Journal of Economic Behavior & Organization, 876
817 Behavior, 75(6), 2035–2039. 29, 191–209. 877
818 Galef, B., & Whiskin, E. (2008b). Use of social information by sodium- and Pinker, S. (1994). The Language Instinct: The New Science of Language 878
819 protein-deficient rats: a test of a prediction (Boyd & Richerson 1988). and Mind. London: Penguin Press. 879
820 Animal Behavior, 75(6), 627–630. Powell, A., Shennan, S., & Thomas, M. (2009). Late Pleistocene 880
821 Gelman, S. A. (2009). Learning from others: Children's construction of demography and the appearance of modern human behavior. Science, 881
822 concepts. Annual Review of Psychology, 60(1), 115–140. 324(5932), 1298–1301. 882
823 Gergely, G., & Csibra, G. (2005). The social construction of the cultural Rakoczy, H., Warneken, F., & Tomasello, M. (2009). Young children's 883
824 mind: imitative learning as a mechanism of human pedagogy. Interac- selective learning of rule games from reliable and unreliable models. 884
825 tion Studies, 6, 463–481. Cognitive Development, 24(1), 61–69. 885
826 Henrich, J. (2009). The evolution of costly displays, cooperation, and Richerson, P. J., & Boyd, R. (2005). Not by Genes Alone: How Culture 886
827 religion: credibility enhancing displays and their implications for Transformed Human Evolution. Chicago: University of Chicago Press. 887
828 cultural evolution. Evolution and Human Behavior, 30, 244–260. Rogers, E. M. (1995). Diffusion of Innovations. New York: Free Press. 888
829 Henrich, J., & Boyd, R. (1998). The evolution of conformist transmission Sabbagh, M. A., & Baldwin, D. A. (2001). Learning words from 889
830 and between-group differences. Evolution and Human Behavior, 19, knowledgeable versus ignorant speakers: links between preschoolers' 890
831 215–242. theory of mind and semantic development. Child Development, 72, 891
832 Henrich, J., & Boyd, R. (2001). Why people punish defectors: conformist 1054–1070. 892
833 transmission stabilizes costly enforcement of norms in cooperative Shutts, K., Kinzler, K. D., McKee, C. B., & Spelke, E. S. (2009). Social 893
834 dilemmas. Journal of Theoretical Biology, 208, 79–89. information guides infants' selection of foods. Journal of Cognition and 894
835 Henrich, J., & Gil-White, F. (2001). The evolution of prestige: freely Development, 10, 1–17. 895
836 conferred deference as a mechanism for enhancing the benefits of Spelke, E. S., & Kinzler, K. D. (2007). Core knowledge. Developmental 896
837 cultural transmission. Evolution and Human Behavior, 22(3), 165–196. Science, 10(1), 89–96. 897
838 Henrich, N. S., & Henrich, J. (2007). Why Humans Cooperate: A Cultural Sperber, D. Ed. (2006). An Evolutionary Perspective on Testimony and 898
839 and Evolutionary Explanation. Oxford: Oxford University Press. Argumentation (pp. 401-412). Lawrence Erlbaum Associates Publishers. 899 Q5
840 Horner, V., Proctor, D., Bonnie, K. E., Whiten, A., & de Waal, F. B. (2010). Stack, S. (1987). Celebrities and suicide: a taxonomy and analysis, 1948– 900
841 Prestige affects cultural learning in chimpanzees. In P. F. Ferrari, Ed. 1983. American Sociological Review, 52(3), 401–412. 901
842 PLoS ONE, 5(5), e10625, doi:10.1371/journal.pone.0010625. Tomasello, M., Carpenter, M., Call, J., Behne, T., & Moll, H. (2005). 902
843 Jaswal, V. K. (2006). Preschoolers favor the creator's label when reasoning Understanding and sharing intentions: the origins of cultural cognition. 903
844 about an artifact's function. Cognition, 99(3), 83–92. Behavioral and Brain Sciences, 28, 675–691. 904
845 Jaswal, V. K., & Malone, L. S. (2007). Turning believers into skeptics: 3- Topál, J., Gergely, G., Erdohegyi, A., Csibra, G., & Miklósi, A. (2009). 905
846 year-olds' sensitivity to cues to speaker credibility. Journal of Cognition Differential sensitivity to human communication in dogs, wolves, and 906
847 and Development, 8(3), 263–283. human infants. Science, 325(5945), 1269–1272. 907
848 Jaswal, V. K., & Neely, L. A. (2006). Adults don't always know best: Topál, J., Gergely, G., Miklosi, A., Erdohegyi, A., & Csibra, G. (2008). 908
849 preschoolers use past reliability over age when learning new words. Infants' perseverative search errors are induced by pragmatic misinter- 909
850 Psychological Science, 17(9), 757–758. pretation. Science, 321(5897), 1831–1834. 910
851 Kaberry, P. M. (1941). The Abelam tribe, Sepik District, New Guinea: a Warneken, F., & Tomasello, M. (2007). Helping and cooperation at 14 911
852 preliminary report. Oceania, 11(3), 233–258. months of age. Infancy, 11(3), 271–294. 912
853 Keil, F. C., Lockhart, K. L., & Schlegel, E. (2010). A bump on a bump? Wellman, H. M., Cross, D., & Watson, J. (2001). Meta-analysis of theory- 913
854 Emerging intuitions concerning the relative difficulty of the sciences. of-mind development: the truth about false belief. Child Development, 914
855 Journal of experimental psychology General, 139(1), 1–15. 72(3), 655–684. 915
856 Kendal, J. R., Giraldeau, L. A., & Laland, K. N. (2009). The evolution of White, H. (1980). A heteroskedasticity-consistent covariance matrix 916
857 social learning rules: payoff-biased and frequency-dependent biased estimator and a direct test for heteroskedasticity. Econometrica: Journal 917
858 transmission. Journal of Theoretical Biology, 260(2), 210–219. of the Econometric Society, 48(4), 817–838. 918
919
920
Our reference: ENS 5706 P-authorquery-v8

AUTHOR QUERY FORM

Journal: ENS Please e-mail or fax your responses and any corrections to:
Jill Shepherd
E-mail: J.Shepherd@Elsevier.com
Tel: 352-483-8113
Article Number: 5706 Fax: 352-483-3417

Dear Author,

Any queries or remarks that have arisen during the processing of your manuscript are listed below and highlighted by flags in
the proof. Please check your proof carefully and mark all corrections at the appropriate place in the proof (e.g., by using on-
screen annotation in the PDF file) or compile them in a separate list.

For correction or revision of any artwork, please consult http://www.elsevier.com/artworkinstructions.

Any queries or remarks that have arisen during the processing of your manuscript are listed below and highlighted by flags in
the proof. Click on the ‘Q’ link to go to the location in the proof.

Location Query / Remark: click on the Q link to go

in article Please insert your reply or correction at the corresponding line in the proof
Q1 The following references were not cited in the text and were temporarily inserted here. Please indicate
where to insert them.
Q2 Please check the use of the calligraphy symbol.

Q3 Jaswal, 2005, is not listed in the bibliography. Please check.

Q4 Samuels, 2004 was cited here but not in the reference list. Please check.

Q5 Please provide publisher's location and name.

Q6 Figure 1 contains blurry image. Please check and provide replacement as deemed necessary.

Q7 Please check whether the tables were captured correctly.

Thank you for your assistance.

Page 1 of 1