Little evidence that Eurasian jays (Garrulus glandarius) protect their caches by responding to cues about a conspecific’s desire and visual perspective

Corvids appear to be capable of adjusting their behaviour according to another’s perspective, knowledge and desire. For example, Eurasian jays have been found to employ a variety of cache protection strategies to minimise cache loss by responding to cues about the visual perspective or current desire of an observing conspecific. However, it is not known whether these jays (or any other corvid) can integrate multiple cues about different mental states and perform the optimal response accordingly. Across five experiments, we found little evidence that our Eurasian jays responded to either the visual perspective or current desire of another agent. In Experiments 1 and 2 we investigated whether Eurasian jays can limit the risk of cache loss by responding simultaneously to cues about the desire and perspective of a potential conspecific pilferer. Building on established paradigms, we used opaque and clear barriers to manipulate the observer’s visual access to cache locations, and specific satiety to manipulate the observer’s desire towards different types of food. Across both experiments the jays’ caching pattern provided no evidence that they could integrate information about the observer’s desire and perspective. Moreover, the results were also inconsistent with the previously reported effects that jays protect their caches by responding to either the visual access or specific satiety of the observer independently. To gain further insight into these unexpected results, we conducted three more experiments. In Experiments 3 and 4, we attempted to replicate the previous finding that Eurasian jays prefer to cache behind an opaque barrier over a clear barrier when observed by a conspecific than when caching in private. In Experiment 5, we further investigated the previous finding that jays preferentially cache a type of food that had been eaten to satiety by a conspecific over a food that had not been eaten by the conspecific. Experiments 3, 4 and 5 found no significant effects in the direction of the previously reported effects, questioning their robustness. We conclude by discussing the implications of our study for the field of corvid cognition and highlight several key issues that affect the reliability of comparative cognition research.

previously reported effects that jays protect their caches by responding to either the visual 48 access or specific satiety of the observer independently. To gain further insight into these 49 unexpected results, we conducted three more experiments. In Experiments 3 and 4, we 50 attempted to replicate the previous finding that Eurasian jays prefer to cache behind an 51 opaque barrier over a clear barrier when observed by a conspecific than when caching in 52 private. In Experiment 5, we further investigated the previous finding that jays preferentially 53 cache a type of food that had been eaten to satiety by a conspecific over a food that had not 54 been eaten by the conspecific. Experiments 3, 4 and 5 found no significant effects in the 55 direction of the previously reported effects, questioning their robustness. We conclude by 56 discussing the implications of our study for the field of corvid cognition and highlight several 57 key issues that affect the reliability of comparative cognition research. 58 59 INTRODUCTION theory of mind, because the caching jays showed this effect also when they did not know what 128 food the observer was pre-fed on and when the only cue available was the observer's 129 behaviour during the caching event itself (Ostojić et al., 2017). Thus, taken together, these 130 studies seem to indicate that Eurasian jays employ a variety of cache protection strategies to 131 limit the risk of cache loss, by responding to cues correlating with the perspective or current 132 desire of a potential pilferer. Although this evidence is not sufficient to demonstrate mental 133 state attribution, nor to pinpoint the exact underlying cognitive mechanism, it does suggest 134 that corvids are capable of behaving flexibly on the basis of different types of social cues. 135

136
The aim of the current study was to investigate whether Eurasian jays can integrate multiple 137 cues that correlate with different types of mental states to solve social problems. Previous 138 work has to some extent already, at least implicitly, tested whether animals can integrate 139 perspective cues from different sensory modalities (corvids: Shaw  tested whether jays can integrate information about a conspecific's perspective and current 146 desire to selectively protect those caches that are at most risk of being pilfered. To this end, 147 we first conducted two experiments in which we manipulated the observer's visual access to 148 caching locations and its current desire for different foods. We measured the caching pattern 149 of the birds when viewed by this observer. 150 151 Our manipulation of the observer's perspective followed the procedure by Legg and Clayton 152 (2014). The authors gave Eurasian jay cachers access to two locations, one that could and one 153 that could not be seen from an adjacent compartment and thus by an observer bird (when 154 present). One caching tray was positioned behind an opaque barrier (out-of-view tray), and 155 the other behind a transparent barrier (in-view tray). Legg and Clayton's experiment 156 encompassed three conditions: jays could cache when an observer bird was housed in the 157 adjacent compartment -the observer was either a higher ranked individual (Observed by 158 dominant condition) or a lower ranked individual (Observed by subordinate condition) -or 159 when no conspecific was present (Private condition). The authors compared the jays' caching 160 pattern in the Private condition with that in the Observed conditions (the two observed 161 conditions were merged together), and found that jays cached a higher proportion of food 162 items in the out-of-view tray in the Observed than in the Private conditions. 163 164 Our manipulation of the observer's desire for different foods followed the procedure by 165 (Ostojić et al. (2017). The authors investigated whether Eurasian jays and California scrub-166 jays can protect their caches by preferentially caching the type of food that the observer was 167 not currently motivated to pilfer. A cacher and an observer jay were housed in adjacent 168 testing compartments. In the pre-feeding phase, the observer could feed to satiety on a 169 specific type of food: maintenance diet in the baseline trial, and either food A or B in the two 170 test trials. This procedure subsequently reduces the individual's motivation for eating and 171 caching that specific food (but not different kinds of food), a phenomenon known as 'specific 172 satiety' (Balleine and Dickinson, 1998;Clayton and Dickinson, 1999;Dickinson and Balleine, 173 1994). Ostojić et al. found that the jays' preference for caching food A over food B was larger 174 after the observer was sated on food A than after the observer was sated on food B. 175 Interestingly, this pattern was exhibited not only in the Seen Condition -when the cacher had 176 witnessed which particular food had been provided to the observer -but also in the Unseen 177 Condition -when the cacher had not seen the pre-feeding of the observer -thus indicating 178 that the observer's behaviour at the time of caching may have played a key role in the 179 decision-making process of the cacher. Note that, when data were re-analysed for Eurasian 180 jays only, the effect was still statistically significant in the Seen condition -although it was not 181 statistically significant in the Unseen condition (Crosby, 2019). 182

183
In the current study we combined these two protocols. In Experiment 1, cacher jays were 184 provided with only one type of food -which on one trial was the same food on which the 185 observer sated, and on another trial was a different food from that on which the observer was 186 sated -and two caching trays, one that the observer could see and one that the observer could 187 not see. Thus, here, the jays could choose between two caching locations, one that was in-view 188 and one that was out-of-view of the conspecific, allowing them to selectively cache food out-189 of-view when it was desired by the observer. In Experiment 2, cacher jays were provided with 190 a single caching tray -which on one trial could be seen by the observer and on another trial 191 could not be seen by the observer -and two types of food, one of which had previously been 192 pre-fed to the observer. Thus, here, the jays could choose between two food types, one on 193 which the observer was sated and one on which the observer was not sated, allowing them to 194 selectively cache the less desired food when the observer could see them. Consequently  To ensure that the birds (n=9) were comfortable caching in trays when these were placed in 214 proximity of each of the two arms of the T-barrier, they initially received two familiarisation 215 trials in private, in which only a single tray was present. The tray was placed once behind the 216 opaque and once behind the transparent arm of the barrier. All birds except two reached the 217 criterion in the familiarisation, i.e., they cached at least one item on each trial, and therefore 218 proceeded to the test (leading to n=7 for the test). 219 220 Following the basic design of Ostojić et al. (2017), the test trials comprised a pre-feeding 221 phase and a caching phase (Figure 1). In the pre-feeding phase, the cacher jay could see a 222 conspecific (the observer) eat a specific type of food (macadamia nuts, M, or peanuts, P) in an 223 adjacent indoor compartment. In the subsequent caching phase, the cacher jay was presented 224 with two caching trays, each placed behind one of the two arms of the T-barrier, and was 225 allowed to cache while the observer jay was still present in the adjacent compartment. Birds 226 were tested in two conditions (one trial per condition; Figure 1). In the Different Food 227 Condition, the type of food received by the observer in the pre-feeding phase differed from the 228 type of food received by the cacher in the caching phase (e.g., M for the observer, and P for the 229 cacher). In the Same Food Condition, the observer and the cacher received the same type of 230 food (e.g., P for the observer, and P for the cacher  Figure 1) and a 263 comparison of the difference score to chance (i.e., 0) detected no statistically significant 264 difference in the Different Food condition (one-sample Wilcoxon signed rank test: n=7, W=-7, 265 p=0.61). The same comparison in Same Food condition also detected no statistically 266 significant difference from chance (one-sample Wilcoxon signed rank test: n=7, W=1, p=1). 267 Notably, the numerical pattern of the jays' caching was in the opposite direction to the 268 prediction: jays' caching was more biased towards the out-of-view tray in the Same food than 269  trials were composed by a pre-feeding phase (middle) and a caching phase (right). The cacher bird received two trials that differed in the type of food that was provided to the observer bird in the pre-feeding phase. In one trial 282 (Different Food Condition), the food provided to the observer in the pre-feeding phase differed from the food the 283 cacher bird could subsequently cache. In the other trial (Same Food Condition), the food provided to the observer 284 in the pre-feeding phase was the same as the food the cacher bird could subsequently cache. In the pre-feeding 285 phase of both trials, the cacher bird was provided with a handful of maintenance diet (md). Note that, for 286 explanatory purposes, the scheme shows the cacher as being provided with peanuts (P) in the familiarisation 287 and in the caching phase of the two test trials. However, in the experiment, cacher birds were randomly assigned 288 to one type of food (either peanuts or macadamia nuts), which was used consistently in the familiarisation and in  In Experiment 2, we used a complementary design to test whether jays can integrate multiple 296 cues to decide which type of food to hide to protect their caches from being pilfered. To do so, 297 jays had access to one caching location at a time (either in-view or out-of-view to the 298 observing conspecific) and two types of food (one on which the observer was sated and one 299 on which the observer was not sated). Following the general structure of Experiment 1, in the 300 pre-feeding phase the cacher bird was first able to see an observer eat one particular type of 301 food (macadamia nuts or peanuts) to satiety. In the subsequent caching phase, the cacher bird 302 was presented with a single caching tray and the two types of food (macadamia nuts and 303 peanuts). To manipulate the observer's visual access to the caching location, the tray was 304 placed behind an 'U-shaped' Perspex barrier (henceforth U-barrier) that consisted of two 305 lateral panels and one central panel forming two angles of approximately 45° (see Materials 306 and Methods; Figure 2). The U-barrier was either transparent, thereby allowing the observer 307 to see the caching location, or opaque, thereby preventing the observer from seeing the 308 caching location. 309 310 Birds (n=8; for details see Methods) first received two familiarisation trials in private to 311 ascertain that they were comfortable caching both types of food in a tray placed in proximity 312 of each of the barriers. All birds except one reached the familiarisation criteria, namely to 313 cache: i) at least one item of both types of food across the two trials, and; ii) at least one item 314 (of any type of food) in a tray placed in proximity of both the transparent and the opaque U-315 barrier. These birds (n=7) were subsequently tested with the transparent barrier (In-view 316 condition) and with the opaque barrier (Out-of-view condition). In each condition, the birds 317 received two trials, one in which the observer was pre-fed one type of food and one in which 318 it was pre-fed the other type ( Figure 2). 319 320 If the jays can integrate information from the different cues available and which should 321 correlate with the observer's desire and perspective, their caching pattern might be expected 322 to meet two predictions. First, the jays' preference to cache P when the observer was sated on 323 P relative to when the observer was sated on M, should be higher in the In-view than in the 324 Out-of-view condition. This is because it is in the In-view condition that the observer can see 325 the caching locations such that here the caching bird could protect its caches by caching 326 preferentially more of the food that the observer is sated on. Second, in the In-view condition, 327 the preference to cache P should be higher when the observer was sated on P than when the 328 observer was sated on M. As in Experiment 1, both proportion and difference scores were 329 used as indexes to analyse the birds' preference (for details see Analysis in the Methods The same pattern of results was found when the jays' preference was analysed using the other 346 index, namely difference scores. No statistically significant difference in the preference to 347 cache P over M when the observer is sated on P relatively to when the observer is sated on M 348 condition no statistically significant difference was detected between the trials in which 352 observer was sated on peanuts and the trials in which the observer was sated on macadamia 353 nuts (MedianPre-fed P = -1, MedianPre-fed M = 0; Wilcoxon signed rank test: n=7, W=-3, p=0.86). 354 The additional analysis of the same comparison for the Out-of-view condition also did not 355 detect a statistically significant difference between trials (MedianPre-fed P = -1, MedianPre-fed M = -356 1; Wilcoxon signed rank test: n=7, W=-1, p=1).  that was provided to the observer in the pre-feeding phase: blue denotes that the observer had been pre-fed P 413 and orange denotes that the observer had been pre-fed M. Because these birds had not recently participated in testing using the experimental set-up 444 employed here and the T-barrier, we first conducted a familiarisation that followed the same 445 procedure as that used in Experiment 1. Nine birds passed the familiarisation and proceeded 446 to the test. In this experiment only, we conducted the same analyses as for all other 447 experiments (i.e., Wilcoxon signed rank tests) but also an additional one, namely the same 448 analysis (permutation tests for paired data) that was also used in Legg and Clayton (2014). 449 Again, a strong claim of an effect would require consistent results regardless of the analyses 450 used. 451

452
In line with the original study, we found that the average number of total items cached across 453 both trays was not significantly higher when the jays were observed by a conspecific than 454 when they were in private (Permutation test, n=9, Z=0.79, p=0.43). Two birds cached no items 455 in any of the Private and Observed trials, thereby they were excluded from further analyses of 456 proportion scores because, given their performance, it was not possible to compare the 457 proportion of items cached in the out-of-view tray between conditions. In the same analysis    In Experiments 1 and 2, we investigated whether Eurasian jays can take into account two 542 types of social cues simultaneously and perform the most advantageous behavioural output 543 accordingly. Specifically, we tested whether caching birds can integrate information from cues 544 correlating with a conspecific observer's desire and perspective to most effectively protect 545 their caches. Consistently across these two experiments, we did not detect effects that would 546 support such integration of information from different cues. In Experiment 1, jays did not 547 preferentially cache in the out-of-view tray when they were provided with a food that was 548 highly desired by the observer, and not more than when the food was not desired by the 549 observer. Furthermore, in Experiment 2, jays did not preferentially cache the food for which 550 observers had a decreased desire, and not more when the observer could see them than when 551 they could not see them. preferentially cache a specific food when the observer was pre-fed on that food relative to 561 when the observer was pre-fed on a different food. experiments was surprising, especially given that they were conducted in the same lab, with 577 many if the same birds and experimenters. Specifically, in 4 out of the 7 tests of the hypothesis 578 that the jays could use social cues to protect their caches, the results were not in the direction 579 of the prediction: Experiment 1, prediction 1; Experiment 2, predictions 1 and 2; Experiment 580 5, prediction 2, No Barrier condition. In the remaining 3 tests in which we had a directional 581 prediction -Experiments 3 and 4, and Experiment 5 prediction 2, Barrier condition -the 582 effects were in the correct direction but were non-significant and much smaller than similar 583 effects previously reported. 584

585
We propose two explanations for why our studies were unable to detect effects consistent 586 with the previous literature, namely low power and the re-use of a unique bird sample. First, 587 the sample sizes (often ~10) and trial numbers (often 1 to 3) used in cache protection studies 588 are so low that the designs are only powered to consistently detect very large effects. Coupled 589 with a likely publication bias, this may be taken to suggest that published effects are 590 overestimated, and that the probability of a single replication study finding a similarly sized 591 effect is low (Farrar et al., 2020;Fiedler and Prager, 2018;Hedges, 1984). Viewed in this light, 592 it is perhaps not surprising that any one of our studies returned non-significant results. 593 However, that we found no significant results across all five of the studies was surprising.  Table 1), but they were around five years older. 596 It is possible that the behaviour of these birds has changed over time, either due to learning 597 effects, ageing, developing abnormal or stereotyped behaviour or change in motivation as a 598 result of being kept in captivity for a long duration (Garner, 2005). Clayton, 2013). We were unable to elicit the cache protection strategies that this literature 604 implies are consistently observable across corvid species, including in our Eurasian jays. 605 While it is possible our findings were local failures to find these effects, it is also possible that 606 the general research practices and methods that have produced the corvid social cognition 607 literature are liable to producing unreliable findings or overestimated effect sizes. We 608 currently do not know how many other studies have produced negative cache-protection 609 results but have not been published, and understanding the magnitude of the publication bias 610 replication attempt that also did not detect the originally reported effects (Crosby, 2019). 616 Overall, the data on these effects seem too uncertain to draw any firm conclusions about 617 Eurasian jay cognition. 618

619
Our difficulty with replicating previous research, even in the same laboratory as the original 620 findings and with many of the same birds and experimenters, highlights two ways in which 621 research on corvid social cognition could make progress. First, understanding the extent of 622 publication bias in our literatures is key to understanding their evidential value. 623 Retrospectively, this may be achieved through meta-analysis techniques, and prospectively Fourteen adult Eurasian jays from two separate colonies were tested in this study (Table. 1). 651 Most of the jays took part in multiple experiments and had previously been tested in 652 experiments that involved caching in a similar set-up as that used in the current study (details 653 about which jay participated in which experiment(s) are given in Table 1). 654 655 656

General procedures 718
In all experiments, the birds' maintenance diet was removed from the aviary approximately 719 1.5h prior to the start of each trial to ensure that the birds were mildly hungry and thus likely 720 to interact with food provided during testing. 721 722 Familiarisation. In all experiments in which the birds had not experienced the set-up and 723 apparatuses just prior to testing (i.e., in Experiments 1, 2, 4, and 5), a familiarisation 724 procedure was conducted to ascertain that birds were comfortable caching in trays placed in 725 proximity of the respective barriers (see Specific Procedures for further details). During the 726 familiarisation, each bird was tested in isolation, i.e., with no other birds present in the test 727 area. Compartments used during the familiarisation were not used in the test phase to 728 minimise the probability of carry-over effects. 729 730 Test. In Experiments 1, 2, and 5, test trials involved a pre-feeding phase followed by a caching 731 phase. Before the start of a test trial, two birds (a cacher and an observer) were given access 732 to two adjacent compartments. Subsequently, the experimenter placed a bowl containing the 733 pre-feeding food (macadamia nuts or peanuts) on the suspended platform in the observer's 734 compartment and a bowl containing a handful of maintenance diet on the platform in the 735 cacher's compartment (Figures 1, 2, and 4). Both bowls were placed in front of the mesh 736 window to ensure that the birds could see each other whilst eating and to maximise the 737 likelihood that the cacher could see on which food the observer was pre-fed. The 738 experimenter then left the test room and the birds could eat the pre-feeding food for 15 739 minutes. Next, the experimenter entered the test room again and removed the bowls as well 740 as any food remains on the platforms. In the subsequent caching phase, the caching trays, as 741 well as the barrier and the food bowl, were positioned in front of the mesh window in the 742 cacher's compartment (Figures 1, 2, and 4). The experimenter then left the test room and the 743 birds were given 15 minutes during which the cacher could eat and cache the food in the 744 trays. In Experiments 3 and 4, the test trials involved only a caching phase (Figure 3). Before 745 the start of a trial, the cacher bird was given access to the testing compartment where the 746 caching trays and the T-barrier had already been positioned. In the test trials of the Observed 747 condition, a second bird (i.e., the observer) was also induced to enter the adjacent 748 compartment. Subsequently, a food bowl was placed on the suspended platform in the 749 cacher's compartment. The experimenter then left the test room and the cacher was given the 750 opportunity to eat and cache food for 15 minutes. 751 At the end of each familiarisation and test trial, the experimenter opened the flap windows to 753 allow the bird(s) to re-join the rest of the group in the aviary and recorded the amount of food 754 eaten and the number and location of caches by manually checking the food bowls and trays. 755 Approximately three hours after each trial, the cacher was allowed to re-enter the caching 756 compartment. No other birds were present in the test area and the flap window was kept 757 open so that the bird had access not only to the test compartment but also to the adjacent 758 smaller aviary. Note that the door connecting the small aviary to the main aviary was kept 759 closed such that no other bird could enter the cacher's aviary or compartment. The cacher 760 could retrieve the hidden items and re-cache them in the compartment and in the adjacent 761 small aviary. This retrieval phase was conducted only to reduce the probability that birds 762 would stop caching in the trays, and thus these data were not analysed. Birds received a single 763 test trial per day.  Familiarisation. Birds (n = 9; Table 1) received two familiarisation trials on two separate 769 days to ensure that they were comfortable caching in trays when these were placed in 770 proximity of each of the two arms of the T-barrier. On each trial, the bird was presented with 771 the T-barrier, a single caching tray and a food bowl containing either 50 macadamia nut 772 halves (M) or 50 whole peanuts with skin (P). The type of food (macadamia nuts or peanuts) 773 was randomly assigned to birds but each bird was provided with the same type of food in 774 both trials. The bird was given the opportunity to eat and cache for 15 minutes. On one trial, 775 the tray was placed behind the opaque arm of the T-barrier and on the other trial it was 776 placed behind the transparent arm. The order in which birds experienced the tray in the two 777 locations was counterbalanced across birds. The orientation of the barrier within the 778 compartment was different from that later used during testing and was kept consistent for 779 each bird across the two familiarisation trials (Figure 1). This procedure was chosen to 780 ensure that the birds were not more familiar with one of the two orientations of the barrier in 781 a specific spatial set-up (e.g., opaque arm facing the outdoor aviary) in the subsequent test. To 782 proceed to the test, birds had to cache at least one food item in the tray on each trial. If a bird 783 did not meet this criterion, it was excluded from further testing. All birds except two (i.e., 784 Dublin and Lisbon; Table 1) passed the familiarisation and proceeded to the test. 785 Test. During the pre-feeding phase, cachers (n = 7; Table 1) could see a conspecific eat a 787 specific type of food: either the same type of food they were going to receive in the 788 subsequent caching phase (Same Food condition) or a different one (Different Food condition; 789 Figure 1). The order in which the birds experienced the Different Food and Same Food 790 conditions was counterbalanced across birds. In the subsequent caching phase, cachers were 791 provided with the same food used in the familiarisation and with two caching trays, each one 792 placed behind one of the two arms of the T-barrier (Figure 1). The food given to the observer 793 during the pre-feeding phase and to the cacher during the caching phase was either 50 794 macadamia nut halves or 50 whole peanuts with skin. All birds received one trial per 795 condition, i.e. two test trials in total. If a bird cached no items on both trials, it was paired with 796 a different observer and the two trials were repeated. If it again did not cache on both trials, 797 these data were not included in the analysis. In contrast, if the bird cached with the second 798 observer, then these data were included in the analysis. This procedure was decided during 799 data collection, after one bird (Lima) did not cache any food across both trials, but before the Familiarisation. Birds (n = 8; Table 1) received two familiarisation trials on two separate 806 days to ascertain that they were comfortable caching both types of food (macadamia nuts and 807 peanuts) in a tray placed in proximity of each of the U-barriers (transparent and opaque). On 808 each trial, the bird was presented with a U-barrier, a single caching tray and two food bowls, 809 which were presented sequentially. The food bowls contained either 50 macadamia nut 810 halves or 50 whole peanuts with skin. The bird was given the opportunity to eat and cache for 811 20 minutes: during the first 10 minutes it was provided with one type of food and during the 812 next 10 minutes with the other. The order in which the birds experienced the two types of 813 foods was counterbalanced across birds and across trials, such that each bird experienced one 814 order on their first trial and the opposite order on their second trial. On each trial, the barrier 815 was either transparent or opaque. The order in which the birds experienced the two types of 816 the U-barrier was counterbalanced across birds. To proceed to testing, birds had to i) cache at 817 least one item on each trial, and ii) cache at least one item of each kind of food across the two 818 trials. If a bird did not meet these criteria, it was excluded from further testing. All birds 819 except one (Lisbon ; Table 1) passed the familiarisation and proceeded to the test. 820 821 Test. The pre-feeding phase was conducted in the same way as in Experiment 1. In the 822 subsequent caching phase, the cacher was provided with a single tray placed within the U-823 barrier and two food bowls (one containing 50 macadamia nut halves and the other 824 containing 50 whole peanuts with skin). Birds (n = 7; Table 1) received four trials in total: 825 there were two conditions, namely the In-view condition (clear U-barrier) and the Out-of-view 826 condition (opaque U-barrier), and within each condition, there were two trials, one in which 827 the observer was pre-fed on one type of food (e.g., macadamia nuts), and one in which the 828 observer was pre-fed on the other type food (e.g., peanuts). Birds first received both trials of 829 one condition, and then the two trials of the other condition. The order of conditions was 830 counterbalanced across birds. The order in which the observer was pre-fed on the two types 831 of food was counterbalanced across birds within condition, but kept consistent across 832 conditions (i.e., the order of the two trials was the same in both conditions). If a bird did not 833 cache any items in one or two trials, those trials were repeated at the end of the experiment. 834 However, if a bird did not cache any items in more than two trials, that bird was not tested 835 any further and was excluded from the analysis. A specific trial was repeated for a maximum 836 of two times, such that a bird was excluded from the analysis if it cached no item in both 837 repeated trials. It was necessary to repeat trials for three birds (i.e., Jerusalem: two repeated 838 trials; Lima: two repeated trials; Rome: one repeated trial). The analysis included the data of 839 all seven birds. Experiment 2 was conducted from January to February 2018 by PA. 840 841 Experiment 3. 842 Familiarisation. Here, we conducted no familiarisation because this experiment took place 843 shortly after Experiment 1 (which also involved the T-barrier) and involved the same birds. 844 Thus, participation in Experiment 1 already insured that birds were comfortable caching in 845 trays next to the T-barrier. 846 847 Test. The procedure of the test was simplified from the procedure in Legg and Clayton 848 (2014). The main difference was that there was only one trial per condition. In addition, we 849 used either 50 whole peanuts with skin or macadamia nut halves (counterbalanced across 850 birds) as food for the cachers, whereas the original study used 30 peanut halves. Birds (n = 8; 851 Table 1) were given two trials in total: one with an observer present in the adjacent 852 compartment (Observed condition) and one in which no observer was present (Private 853 condition). The order of the conditions was counterbalanced across birds. On each trial, two 854 trays were positioned behind the T-barrier, one behind the opaque and one behind the 855 transparent arm. The orientation of the T-barrier was counterbalanced across birds but kept 856 constant across trials for each bird. The cacher was given 15 minutes during which it could 857 each or cache the food. If a bird did not cache on a trial, the trial was repeated. If the bird did 858 not cache again, the data were not included in the analysis. A trial was repeated for one bird 859 (Caracas). Due to timing constraints, another bird (Lisbon) was not given the possibility to 860 repeat the trials in which no item was cached. Thus, although this was not pre-specified as an 861 exclusion criterion, Lisbon's data were excluded from the analysis before it was conducted. 862 The analysis included the data of seven birds, i.e., all birds except one (Lisbon). To solicit competitive interactions among birds, higher value food items (e.g., wax worm 876 larvae, Galleria mellonella) were also presented in a bowl or scattered around on the floor of 877 the aviary. The identity of both actor and recipient involved in any displacement (i.e., Bird X 878 approaches Bird Y causing Bird Y to leave) was recorded. If necessary, to obtain data for all 879 birds, higher ranked birds were locked into separate compartments to favour interactions 880 among lower ranked birds. Observation sessions were conducted on multiple days, until data 881 were collected to establish a clear social hierarchy within each colony. Sessions lasted 882 approximately 40 min each. Dominance hierarchy data were not collected for colony 2 883 because only one bird of this group (Hunter) passed the familiarisation. 884 885 Familiarisation. Unlike birds from colony 1, birds of colony 2 had not recently had any 886 experience with the T-barrier. Thus, all birds (n=13; Table 1) received two familiarisation 887 trials on separate days to ascertain that the birds were comfortable caching in proximity of 888 both the transparent and the opaque arm of the T-barrier (see also pre-registration addition 889 from 8 Dec 2018). This familiarisation followed the same procedure as the familiarisation in 890 Experiment 1, except that here, each bird was provided with 30 peanut halves as in the 891 original study . To proceed to the test, birds were required to cache 892 at least one food item in each of the two familiarisation trials. If no item was cached in a trial, 893 then that trial was repeated for a maximum of two times. Thus, a bird could receive a 894 maximum of 6 trials in total. The repeated trials were conducted at the end (for example, if a 895 bird's first trial had to be repeated, then the bird received the second, pre-planned trial on day 896 2, and subsequently it received the first trial again on day 3). It was necessary to repeat trials 897 for two birds (i.e., Lisbon: one repeated trial; Lima: one repeated trial). Nine birds (Table 1) 898 passed the familiarisation and proceeded to the test. Due to an experimenter's error, the raw 899 data for one bird (Hunter) in the familiarisation was not archived. containing 30 peanuts halves. The bowl was placed close to the stem of the 'T' such that it was 915 equidistant from the two caching trays (Figure 4). The cacher could eat and cache for 15 916 minutes and was subsequently released back into the aviary. 917 918 All birds were tested in all three conditions, except the highest and lowest ranked bird in each 919 colony. The former could only be tested in the Private and Observed by subordinate conditions 920 and the latter could only be tested in the Private and Observed by dominant conditions. Thus, 921 most birds received six trials in total (two trials per condition), whereas the highest and 922 lowest ranked individuals in each colony received four trials in total because they could only 923 be tested in two of the three conditions. In line with the procedure of the original study, test 924 trials were not repeated if the bird cached no item. However, we decided to repeat a specific 925 trial (Dublin's first trial in the Observed by dominant condition) because the bird that served 926 as observed (Rome) appeared to experience issues with flying. This trial was repeated with a 927 different observer after the remaining pre-planned trials were completed. The analysis 928 included the data of all nine birds. This experiment was pre-registered on the Open Science 929 Framework (https://osf.io/8p4tx/). The pre-registration was conducted after the 930 familiarisation was completed but before the start of the test. Experiment 4 was conducted 931 from October to December 2018 by PA (colony 1) and Rachel Crosby (colony 2). 932 933 Experiment 5. 934 Familiarisation. Birds (n=8; Table 1) received two familiarisation trials on separate days to 935 ensure that they were motivated to cache both types of food and were comfortable caching in 936 a tray both when it was positioned close to the U-barrier and when no barrier was present. 937 Thus, the familiarisation followed the procedure of the familiarisation in Experiment 2, except 938 that here, one trial involved the transparent U-barrier and the other one no barrier. To 939 proceed to testing, birds had to i) cache at least one item on each trial (i.e. both with barrier 940 present and with no barrier present), and ii) cache at least one item of each type of food 941 across the two trials. If no item was cached in a trial, then that trial was repeated for a 942 maximum of two times. Thus, a bird could receive a maximum of six trials in total. The 943 repeated trials were conducted at the end: for example, if a bird's first trial had to be repeated, 944 then this bird received the second, pre-planned trial on day 2, and subsequently it received 945 the first trial again on day 3. It was necessary to repeat trials for two birds (i.e., Lisbon: one 946 repeated trial; Wellington: one repeated trial). All eight birds (Table 1) Figure 4). Birds first 952 received both trials (observer pre-fed macadamia nuts and observer pre-fed peanuts) of one 953 condition (e.g. Barrier condition), and then the two trials of the other condition. The order in 954 which the two conditions were conducted was counterbalanced across birds. The order in 955 which observers were pre-fed the two kinds of food within a condition was counterbalanced 956 across birds, but kept consistent across conditions such that the order of the two trials for 957 each bird was the same in both conditions. If a bird did not cache any items in one or two 958 trials, those trials were repeated at the end of the experiment. Each trial could be repeated no 959 more than two times (i.e., three attempts in total). If a bird cached no item in more than two 960 trials, that bird was not tested any further and was excluded from the analysis. It was 961 necessary to repeat trials for two birds (i.e., Quito: one trial; Wellington: two trials). The 962 analysis included the data of seven birds (Table 1) In all experiments, we recorded the number and type of food items cached on each trial by 968 manually checking the trays. The experimenters were not blind to the conditions while 969 counting the food items. These data were used to test whether the birds had a preference for 970 caching a specific type of food or for caching in a specific tray. In all experiments, we also 971 recorded i) the number of items taken from the bowl by observers (during pre-feeding) and 972 by cachers, and ii) the number of items recovered by cachers during retrieval sessions. These 973 data were collected such that all data available for each trial are archived and available, but 974 these data were not relevant to the experimental question so that they were not analysed. 975 976

Statistical Analysis 977
The birds' preference for a specific type of food or tray was analysed according to two indices: 978 proportion scores (e.g., the proportion of items cached in one location out of total number of 979 items cached in both locations) and difference scores (e.g., number of items cached in one 980 location minus the number of items cached in the other location). As stated in the pre-981 registrations of Experiments 4 and 5 (https://osf.io/8p4tx/), we originally planned to analyse 982 the data of all five experiments only through proportion scores. However, when a bird caches 983 no item in a trial, then the individual performance in that specific trial cannot be analysed 984 through the proportion scores, yet it can still be analysed through the difference scores. This 985 issue is relevant only to Experiment 4, where -in line with the procedure of the original study 986