Behavioural lateralization covaries with sex and inhibitory control in the common waxbill

preferred feeding side, contrasting with results in our population. There were also differences between individuals in our population in the direction and strength of lateralization. Females were, on average, more right-side lateralized than males, and there were larger among-individual differences in lateralization among males. None of the phenotypes studied predicted differences in the strength of lateralization. However, individual differences in the direction of lateralization were related to inhibitory control, an aspect of cognition, more strongly so in males than in females. These are among the few results indicating that the direction of behavioural lateralization, rather than its strength, is associated with cognitive differences among individuals. It is also the ﬁ rst time that a sex-dependent association between behavioural lateralization and cognitive performance has been found, suggesting that sex-speci ﬁ c functions are implicated in maintaining differences in behavioural lateralization.

Once thought to be a distinctive human feature, behavioural lateralization is currently known to be widespread across animal taxa (Frasnelli, 2013;Miletto Petrazzini et al., 2020;Stancher et al., 2018).Behavioural lateralization can differ in direction (right or left side preferences) and strength (stronger or weaker preference for either side) between species, between individuals of the same population and even within individuals, depending on the task performed (Brown & Magat, 2011;Frasnelli & Vallortigara, 2018;Ghirlanda & Vallortigara, 2004;Schiffner & Srinivasan, 2013).
An example of cerebral hemisphere specialization is the processing of visual stimuli in animals with large monocular visual fields, such as birds.Here, the lateral part of the retina of each eye is connected to the contralateral hemisphere, such that the right-side visual field (mostly input from the right eye) is mainly processed by the left hemisphere, and the left-side visual field mostly by the right hemisphere (Frasnelli & Vallortigara, 2018;Rogers & Vallortigara, 2015).In an extensively studied species regarding preferential eye use, chicks of the domestic chicken, Gallus gallus domesticus, discriminate pebbles from grain better with their right eye/left hemisphere (Mench & Andrew, 1986;Rogers et al., 1985) and, conversely, are better at perceiving predator cues that appear from their left side (Dharmaretnam & Rogers, 2005).Studies of visual lateralization conducted on other birds, such as pigeons, Columba livia (Güntürkün et al., 2000), and zebra finches, Taeniopygia guttata (Alonso, 1998), agree that the right eye/left hemisphere is specialized in object categorization and discrimination, and more used under relaxed conditions, and that the left eye/right hemisphere is important under stressful and emotionally charged situations (e.g.copulation, predator detection and escape; Rogers, 2021).
Cerebral lateralization has been linked to better behavioural performance or enhanced cognitive abilities (Frasnelli & Vallortigara, 2018;Magat & Brown, 2009;Roche et al., 2020;Rogers & Kaplan, 2019), and differences between more and less lateralized individuals can influence fitness (Brown & Magat, 2011;Dadda et al., 2010;Magat & Brown, 2009).For example, chimpanzees, Pan troglodytes, with stronger hand preferences for fishing termites, irrespective of the direction of lateralization, perform better at this task (Brown & Magat, 2011;McGrew & Marchant, 1997), and cats, Felis catus, using a preferred paw during a problem-solving task performed better than ambidextrous individuals (Isparta et al., 2020).Being strongly lateralized is also often found associated with enhanced abilities in tasks not directly linked with the lateralized behaviour (Berg et al., 2020;Rogers & Kaplan, 2019).For example, common marmosets, Callithrix jacchus, with stronger hand preferences for foraging are faster at detecting predators while feeding (Piddington & Rogers, 2013), and more strongly lateralized juvenile ambon damselfish, Pomacentrus amboinensis, are better at adjusting behaviour in the presence of predator odours (Chivers et al., 2017).Despite advantages of being strongly lateralized, there are also studies where no association was found between lateralization and cognitive performance (e.g.Benedict et al., 2023), or where more lateralized individuals perform worse in some cognitive tasks.For example, even though strongly lateralized juvenile ambon damselfish are faster at identifying predators, they are worse at competing with conspecifics and receive more aggression (Chivers et al., 2017), and more lateralized pheasants, Phasianus colchicus, were found to have lower survival rates when released into the wild (Whiteside et al., 2018).Thus, behavioural lateralization may be beneficial in some situations but disadvantageous in others (Dadda et al., 2009).
Since most research has found advantages of being lateralized irrespective of the direction of lateralization (i.e.left or right; Frasnelli & Vallortigara, 2018;Rogers, 2021; but see also Bibost & Brown, 2014;Lucon-Xiccato et al., 2020), it could be expected that an equal or a random number of left-and right-lateralized individuals exist, but usually most individuals in a population are lateralized in the same direction (Ghirlanda et al., 2009;Ghirlanda & Vallortigara, 2004).This is referred to as population level lateralization (Denenberg, 1981).Population level lateralization is common, generally with 60e90% of the individuals lateralized in the same direction (Frasnelli & Vallortigara, 2018;Ghirlanda & Vallortigara, 2004).Why population level lateralization exists, together with individual differences in lateralization, is not straightforward.There should be advantages for most of the individuals in a population to be lateralized in the same direction, such as coordination and cohesion within groups (Frasnelli & Vallortigara, 2018).However, there are also reasons why not all individuals should be lateralized in the same direction, such as frequency-dependent costs and benefits (Ghirlanda et al., 2009;Ghirlanda & Vallortigara, 2004;Tonello & Vallortigara, 2023).An example of frequency-dependent benefits is found in the giant Australian cuttlefish, Sepia apama, where the majority of individuals prefer to use the left eye to view conspecifics, and a minority of males who prefer to use the right eye win more fights because this less common behaviour can surprise their opponents (Schnell et al., 2019).Conversely, in the context of courtship and mating, where coordination with females is required, the minority of males that prefer to use the right eye to view and approach females are at a disadvantage and have lower mating success (Schnell et al., 2019).
To help explain individual and population level lateralization of behaviour, we need to understand which phenotypic differences among individuals predict differences in lateralization, as this can provide insights on advantages and disadvantages of behavioural lateralization.There are several candidate phenotypic traits that could influence, or be influenced by, behavioural lateralization.In addition to traits related to cognitive and behavioural performance, several studies have found associations of behavioural lateralization with animal personality (i.e.consistent behavioural differences between individuals across time and contexts; Stamps & Groothuis, 2010), stress susceptibility or sex.For example, right-lateralized pigs, Sus scrofa, were found to be bolder, more explorative and sociable compared to left-lateralized individuals (Goursot et al., 2019), strongly lateralized Port Jackson sharks, Heterodontus portusjacksoni, were found to be more prone to stress (Byrnes, Pouca, & Brown, 2016) and female, but not male, poeciliid fish (Gambusia hoolbroki and Girardinus falcatus) exhibited a right-side turning preference in the presence of same-sex conspecifics (Bisazza et al., 1998).Studies assessing a relation between personality, stress or sex and behavioural lateralization have, however, revealed mixed results (Austin & Rogers, 2012;Goursot et al., 2019;Ocklenburg et al., 2016).
To advance our understanding of individual differences in behavioural lateralization, here we studied side preferences of common waxbills, Estrilda astrild, when feeding in standardized conditions, and then tested whether lateralization was predicted by phenotypic differences among individuals.The common waxbill is a small and gregarious finch species that favours open habitats such as savannahs and areas containing tall grasses, where it forages for small seeds (Cardoso & Reino, 2018).The waxbills studied here were previously shown to have individual differences in cognition, personality, cues of stress and social dominance, among other traits (Beltrão, Marques, et al., 2021;Gomes et al., 2020;Guerra et al., 2020).These include phenotypic differences for which, as noted above, a relation with behavioural lateralization has been found in other species (e.g.traits related to cognition, stress, personality).We (1) quantified the direction and strength of behavioural lateralization of different individuals, (2) tested for population level lateralization and (3) investigated whether differences in lateralization are associated with differences in phenotypic traits, which may suggest possible advantages and disadvantages of behavioural lateralization.

Study System and Quantification of Behaviour
From May 2021 to May 2022, we studied a population of 68 waxbills living in seminatural conditions in a mesocosm at the precincts of our research centre in northwestern Portugal, which consisted of a large outdoor aviary (ca.235 m 2 , 1.3e2.7 m high) covered in fine net, and with abundant natural vegetation (details in Gomes et al., 2021).Most individuals were captured from the wild in 2016 and 2017, in agricultural fields within 20 km of the mesocosm (Gomes et al., 2020).In subsequent years, some pairs bred in the mesocosm and produced offspring (Beltrão, Gomes, et al., 2021).The population size remained stable throughout the years because natural deaths were compensated for by offspring produced in the mesocosm and by birds brought in from the wild in subsequent years (a total of 17 waxbills).Maintenance of the mesocosm (filling feeders with seeds, replacing water in 2 large dispensers and in 3e4 shallow containers for bathing) was done twice a week to minimize disturbance (details in Gomes et al., 2021).Each waxbill had a passive integrated transponder (PIT) tag in a plastic leg ring (Dorset Identification, Aalten, The Netherlands), which was read by an array of radiofrequency identification (RFID) antennas placed in 12 feeders on a wall, 1.20 m or 1.55 m high, and in eight long perches (15.5 cm long) in an area from 0.85 to 3.37 m from the feeders (details in Gomes et al., 2021).
On 5 May 2021, we placed a new feeder on the same wall, at a height of 0.88 m, but 1.40 m away from the closest of the 12 regular feeders.This experimental feeder was identical to the other feeders, except that it did not have a platform where the birds could land and perch frontally to the food.Additionally, we moved one of the perches equipped with RFID antennas and placed it perpendicularly to and touching the feeder (Fig. 1a).This way, waxbills had to land on the perch, then approach the food and feed with either the left or right side.The birds could fly towards this feeder coming from the large mesocosm area in front of it, from the right side, where the feeders and perches with RFID antennas were located, or from the left side, where there were bushes in which the waxbills also perched.Both the experimental feeder and the 12 regular feeders always remained available with food ad libitum throughout the study.We made video recordings of this apparatus using a small camera (Xiaomi Yi 2K Action Camera), placed in the net ceiling of the aviary, above the feeder (2.10 m high, 1.15 m above the feeder).Sessions of video recordings were evenly distributed in time from 12 May 2021 to 31 May 2022.A total of 187 video recording sessions were made, on average 3.7 per week (range 1e10), each lasting on average 73 min (range 19e95 min; the maximum duration was limited by battery capacity).At the beginning of each video recording, a PIT tag ('synchronization PIT tag') was passed by the antennas in the perch of this apparatus, so as to later synchronize the observations in the videos with the RFID data stream and identify individuals.
The same researcher (P.S.) observed all video recordings and, each time a waxbill landed on the set-up, thus initiating a feeding event, noted the time of landing, whether it landed with its right or left side towards the feeder (landing side) and whether the bird fed from its left or right side (feeding side; Fig. 1b, c).Events in which a waxbill fed using both sides alternately were not considered for scoring lateralization.For the observation of a feeding event to be considered undisturbed and valid, the bird had to approach and then stay by the feeder for a minimum of 10 s while feeding, during which time no other bird landed on the apparatus.We used 10 s because previous work showed that the vast majority of waxbills spent at least this amount of time in the feeders (see Figure 2 in Beltrão, Marques, et al., 2021).If a bird arrived at a feeder occupied by another bird, requiring it to wait for the other bird to leave or displace it (e.g.move towards the other bird or even fight) before feeding, the observation was not considered valid.When a bird arrived at an occupied feeder and the bird feeding there left simultaneously, without contact between the two, the observation of the arriving bird was considered valid.To avoid misidentifying the arriving bird, because of slight time asynchronies between the RFID data stream and the video, we noted these events so that the ID of the departing bird was not considered when identifying the arriving bird (see below).
To evaluate whether the feeding side can be interpreted as a side preference, rather than a consequence of the side from which the bird arrived and landed on the perch, we used 1647 valid feeding observations (all except those where a bird fed using both sides), to calculate the proportion of times that the landing and feeding sides matched or not.Feeding side was the same as landing side in 40% of cases and different in 60% of cases, which is close to the random 50% expected if landing side was completely unrelated to feeding side.Therefore, feeding side in this set-up should largely reflect a choice rather than be a simple consequence of which side the bird arrived at the perch.The individual identity (ID) of the videorecorded birds was retrieved a posteriori using the RFID data stream of the perch in front of this feeder.We noted the time in each video at which the synchronization PIT tag passed by the antennas and wrote an algorithm in R (version 4.2.1;R Core Team, 2022) to synchronize the RFID data stream with the video observations of feeding events, and then to identify individuals in the video observations.Briefly, for each video, the algorithm searched first for the synchronization PIT tag ID in the RFID data stream, around the time the video recording started, to synchronize the time between the video and RFID data.Then, for each valid video observation, it searched for identifications of waxbills in the RFID data stream using a backward buffer of 5 s relative to the time a bird was noted in the video observation and a forward buffer corresponding to the time interval between the focal bird and the next bird landing on the system (up to a limit of 60 s).The backward buffer was used because slight time asynchronies could exist between the video and RFID data, even after synchronization (e.g.time delay to read the synchronization PIT tag).The forward buffer was used because, depending on a bird's movements and leg position, it may take time for the bird's PIT tag to be read by the RFID antennas.If a feeding event was noted as starting when another bird was departing (without contact between the two), then the script would discard readings with the ID of the departing bird when identifying the arriving bird.Individual ID of a feeding observation was considered valid when only readings from one ID were detected within the buffer ranges described.If multiple IDs were detected, or no ID was detected, then that feeding observation was not used in further analyses.
For each waxbill with 10 or more valid feeding observations (N ¼ 22 individuals), we computed a lateralization index (LI) and an absolute lateralization index (ALI).To assess whether preferred feeding side was repeatable across the study period, we calculated repeatability using the 'rpt' function for binomial data, with ID as a grouping variable, in the R package 'rptR' (version 0.9.22;Stoffel et al., 2017).Although we had data from 34 individuals using the experimental feeder, we only used data from the 22 waxbills with 10 or more observations in order to compute more robust indexes of lateralization.LI was calculated as (R À L)/(R þ L), where R and L are the number of times the bird fed from its right or left side.LI can vary between 1, for birds always feeding with their right side, to À1, for birds always feeding with their left side, and it indicates the direction of lateralization.ALI is the absolute value of LI ( jLIj ) and can vary between 0, when an individual feeds with its left or right side an equal number of times, to 1, when an individual always uses the same side to feed.This index quantifies the strength of lateralization, irrespective of the direction.The indexes of lateralization, per individual, are given in the Appendix.

Individual, Population and Species Level Lateralization
We used a chi-square test to compare counts of left-and rightside feeding events across individuals (N ¼ 22 individuals).This tests whether waxbills differ from each other in the proportion of times they used one or the other side of the body when feeding.
To test for population level lateralization, we used a one-sample Wilcoxon test for non-normal variables on the lateralization index (N ¼ 22 individuals).This tests the null hypothesis that the mean LI in the population does not differ from 0. A significant result indicates that most individuals are lateralized towards the same direction (either left or right) when feeding.
To assess species level lateralization, beyond our study population, we collected additional data from 200 photographs of common waxbills around the world published in the citizen science Web site eBird (www.eBird.org).On 1 April 2022, the same researcher (P.S.) searched for 'common waxbill' in eBird and followed the list of results to download photographs of waxbills feeding from grass (Poacea) seeds or perching on near-vertical (>45 ) grass stems, so that it was clear on which side of the bird the grass inflorescence or seeds would be located.We included photographs perching on near-vertical grass stems because, in nature, waxbills mostly eat grass seeds by perching at the top of the stem, both in Africa (Goodwin, 1982) and in their different invasive ranges worldwide (Cardosoand Reino, 2018;Da Silvaxs & Oren, 1990;Oren & Smith, 1981;Sullivan et al., 2015).We downloaded only one photograph per photographer to avoid quantifying lateralization twice for the same waxbill.Before quantifying behavioural lateralization, half of the photographs were randomly selected and the images were mirrored, to avoid unconscious or conscious bias on the part of a second researcher (G.C.) scoring lateralization.Lateralization was then scored for each photograph, blindly to whether it had or had not been flipped, by observing whether the bird clearly had one side towards the food while eating or, when perching, had one side towards the top of the stem.For photographs of waxbills feeding, if food was located above the head, or on both sides, the photograph was classified as 'ambiguous' and was not used for further analysis.For photographs with more than one waxbill perching, a 'majority rule' was used, unless the same number of birds was scored in either direction, in which case the photograph was also considered ambiguous.From the initial 200 photographs, 35 images were classified as ambiguous, resulting in a sample of 165 photographs.After reversing the scores for the photographs that had been flipped, we compared the number of photographs scored as having the left or right side of the body towards the direction of seeds.

Relations with Phenotypic Differences and Social Dominance
To test whether phenotypic differences among individuals, such as differences in cognition, personality and stress levels, were associated with differences in behavioural lateralization, we used phenotypic data available in Gomes et al. (2020), who had earlier studied the same waxbill individuals as here, and to which we refer for detailed methods.Here, we briefly describe each phenotype.
The sex of each individual was assessed molecularly using a small blood sample as described in Gomes et al. (2020).Body size was computed as the first principal component (PC1) from a principal component analysis (PCA) on tarsus length and head þbill length, using measures from all 68 waxbills present in the mesocosm at the time (PC1 explained 72% of variation and had high loadings for both morphological measures: 0.85).
Mirror tests have been shown to reliably assay personality type along a reactiveeproactive axis in common waxbills (Carvalho et al., 2013;Funghi et al., 2018;Gomes et al., 2020;Guerra et al., 2020), and we therefore used behaviour in mirror tests to assay personality.Briefly, mirror tests consisted in filming each waxbill in a small cage with a mirror that was covered during the first 5 min of the test and exposed in the last 5 min.During the period with the mirror exposed, we quantified the time spent looking in the direction of the mirror, the number of vocalizations, changes of position within the cage and fast movements.A PCA on these behaviours resulted in a PC1 whose high values indicated a more proactive response (more vocalizations, changes of position in the cage and fast movements), while low values indicated a more reactive response (more attention towards the mirror).Mirror tests were conducted twice for each bird, 6 weeks apart.We used the average PC1 value across two mirror tests, since mirror tests in waxbills have been shown to be repeatable in the long term (Carvalho et al., 2013;Funghi et al., 2018;Gomes et al., 2020;Guerra et al., 2020).
Tonic immobility assays consisted in placing a waxbill on its back and measuring the time in seconds that it would stay still (i.e. in a state of tonic immobility) before flying away, for a maximum of 60 s.Tonic immobility may be an index of fear (Gallup, 1979;Pusch et al., 2018) and, in waxbills, it is not related to personality differences along their reactiveeproactive axis, as assessed by behaviour in mirror tests and open field tests (Carvalho et al., 2013;Funghi et al., 2018;Gomes et al., 2020;Guerra et al., 2020).We performed a tonic immobility assay immediately after each mirror test, such that the context before testing was identical and comparable across all birds.Again, for each bird, we used the average score of the two assays, since the tonic immobility test has been shown to be repeatable in the common waxbill (Carvalho et al., 2013;Gomes et al., 2020).
Differences in breath rate among individuals may reflect individual differences in their basal metabolic rate or stress (Gomes et al., 2020).We computed breath rates by counting belly movements from video recordings, during the first 5 min of the mirror test, when the mirror was covered (Gomes et al., 2020).We curated counts of belly movements using an algorithm that corrects for missing or duplicated counts (details in Gomes et al., 2020) and computed mean breath rate from these data.Again, we used the mean value of the two times each waxbill performed the mirror test (Gomes et al., 2020).Stress might also cause breath rate to be higher, such as when an individual is handled and placed in a new environment.Therefore, we also computed the change in breath rate by calculating the difference in mean breath rate between the second half and the first half of the video recordings (Gomes et al., 2020).Both these measures, mean breath rate and change in breath rate, have been shown to have high within-individual repeatability (Gomes et al., 2020).
To assay inhibitory control, an aspect of cognitive ability, we used data from a detour-reaching task.This involved a training phase, where it was confirmed that waxbills knew how to detour from an opaque cylinder with open ends to eat the seeds inside, and a test phase, consisting of 15 trials presenting an identical but transparent cylinder with seeds inside.We computed detourreaching performance as the proportion of trials in the test phase that the bird correctly detoured to feed from the seeds inside the cylinder without pecking at the walls of the cylinder, as opposed to pecking or hitting the wall of the cylinder before feeding, which would indicate a lack of inhibitory control (Gomes et al., 2020).
We also assessed social dominance, following methods in Beltrão, Marques, et al. (2021), who studied dominance hierarchies in this waxbill population, and to which we refer to for detailed methods.Briefly, we used the algorithm in Beltrão, Marques, et al. (2021) to automatically identify aggressive displacements at feeders from the data stream of RFID antennas at the 12 regular feeders.We used data from every Sunday during the period of this study (from 12 May 2021 to 31 May 2022), because there were never video recordings or other disturbances in the mesocosm on Sundays.The algorithm identifies displacements when the interval between an individual leaving a feeder and another arriving is less than 2 s, and both individuals stay at least 3 s in that feeder.Beltrão, Marques, et al. (2021) showed, using video recordings, that these criteria identify aggressive displacements accurately.We used these data to compute randomized Elo-ratings, a metric of social dominance (S anchez-T ojar et al., 2018).To measure the quality of sampling effort, we computed the ratio of interactions to individuals (i.e. total number of interactions between any two individuals, divided by the number of individuals), which was 16.7, indicating that the inferred dominance hierarchy was reliably described (!10;S anchez-Tojar et al., 2018).
To assess whether the 22 individuals with lateralization data were a representative sample of the population in terms of the remaining traits (morphology, personality, etc.), we compared their phenotypes with those of the rest of the population (47 birds).We used t tests for comparing inhibitory control, personality, cues of stress, breath rates, body size and social dominance, and we used a chi-square test for comparing the proportion of males and females in the two groups.There were no differences between the 22 individuals with lateralization data and the rest of the population, for any of the behavioural predictors (all jtj 0.58 all P > 0.5; Appendix, Table A1).Only for body size did these 22 individuals appear to be, on average, smaller than the rest of the population (t ¼ 1.96, P ¼ 0.054; Appendix, Table A1).
We computed Pearson correlation coefficients for all pairwise combinations of the above phenotypic variables.Pairwise combinations involving tonic immobility used nonparametric Spearman correlation coefficients instead, because this variable deviated strongly from normality.Sample sizes may differ in these correlations because of missing values in some phenotypes (Appendix, Table A2).All pairwise correlation coefficients were lower than 0.35 in absolute value (Appendix, Table A2).The absence of strong pairwise correlations (correlations >0.6; Dormann et al., 2013) indicates that there are no multicollinearity issues when using those traits as predictors of behavioural lateralization.
We used general linear models (GLM) and a model selection approach (Burnham & Anderson, 2002) to test which phenotypic traits were associated with individual differences in either the lateralization index or the absolute lateralization index.For these models, we used data from waxbills with at least five valid behavioural observations.We ran two separate GLMs with either LI or ALI as the dependent variable and included as independent variables the performance in the detour reaching task, scores from the mirror test, tonic immobility, social dominance, body size, breath rate and sex.The sample size for these models was 13 individuals (5 females and 8 males, all without missing values on these traits).Interactions between independent variables were not included because there were no a priori predictions for their inclusion.We computed Akaike's information criterion corrected for small sample sizes, AICc (Hurvich & Tsai, 1989), for models using all possible combinations of the independent variables.Model selection allowed us to find models with relatively few predictors, which was useful due to the limited number of statistical points (i.e.individual waxbills) in the data set.We then performed model averaging, using models within 6 AICc (DAICc < 6) from the best model (i.e. the model with lowest AICc), weighing values by their Akaike weights (Richards, 2008;Richards et al., 2011).Prior to this analysis, all continuous variables were standardized by subtracting their mean and dividing by two times their standard deviation (Gelman, 2008) in order to obtain comparable model estimates for dichotomous predictors (sex) and continuous predictors (all others).Model assumptions were checked graphically with the R package 'performance' (version 0.9.1;Lüdecke et al., 2021).To perform model selection and model averaging, we used the R package 'MuMIn' (version 1.46.0;Barton & Barton, 2015): we used the function 'dredge' for model selection, indicating that coefficient estimates were standardized by the standard deviation (argument 'beta' ¼ 'sd'), and used the function 'model.avg'for model averaging, with the argument 'subset' ¼ delta < 6.

Ethical Note
This study followed the ASAB/ABS ethical guidelines, and work with wild animals was done under permits 690/2016/CAPT and 57/ 2017/CAPT from the Instituto da Conservação da Natureza e das Florestas.

RESULTS
From the 187 video sessions, we identified 1756 valid feeding events (i.e. the bird fed undisturbed for at least 10 s): in 67% (N ¼ 1174) of feeding events the birds used the right side, in 27% (N ¼ 473) they used the left side and in 6% they alternated between sides.We could identify the individual identity of the waxbills in 884 of these 1756 observations using the RFID data, comprising a total of 34 different waxbills, with the number of valid observations per individual ranging between 1 and 102 (Appendix, Table A3).The number of waxbills with at least 10 valid behavioural observations was 22 (10 females and 11 males; see Appendix, Fig. A1 for the distribution of feeding observations across the sampling period).Repeatability of feeding side across these 22 individuals during our sampling period was moderate and significant (R ¼ 0.54, N ¼ 22, P < 0.01), which is expected given the diversity across individuals not only in the direction of lateralization but also its strength.
Among these 22 individuals, the lateralization index ranged between À0.92 (almost always feeding with the left side facing the seeds) and 1 (always feeding with the right side facing the seeds; Fig. 2a) and the absolute lateralization index ranged between 0.06 (almost no side preference) and 1 (always using the same side; Fig. 2b).Individual waxbills differed statistically in the proportion of times they used either side of the body to feed (chi-square test: c 2 21 ¼ 329.4,P < 0.001), and at the population level, there was significant right-side lateralization when feeding (one-sample Wilcoxon test: V ¼ 214, P < 0.01; Fig. 2a).
Analyses of photographs, from the citizen science Web site eBird, showed an almost equal proportion of waxbills perching or feeding with either side facing the direction of the grass seeds or inflorescence, providing no evidence for species level lateralization.Out of 165 photographs where this direction could be determined, 82 had the left side towards the seeds or the tip of the grass and 83 had the right side towards the seeds or the tip of the grass.
Using a sample of 13 individuals (5 females, 8 males) with at least five valid behavioural observations and no missing values on the various traits studied, we found that the best model (i.e. with the lowest AICc) explaining individual differences in the lateralization index included only two predictors, performance in the detourreaching task (b st ¼ À1.24 ± SE ¼ 0.35, P < 0.01) and sex (b st ¼ À1.59 ± SE ¼ 0.34, P < 0.01), and no other models were within 2 DAICc of this best model (Appendix, Table A4).Model averaging of the 13 models within 6 DAICc of the best model (none of which had more than 3 predictors; Appendix, Table A4) is consistent with those conclusions: individuals that performed better in the inhibitory control task were more left-side lateralized (effect of inhibitory control: model averaged b st ¼ À0.66 ± SE ¼ 0.33, P ¼ 0.06), and females were, on average, more right-biased than males (effect of sex: model averaged b st ¼ À0.48 ± SE ¼ 0.36, P ¼ 0.19; all other effects: model averaged jb st j < 0.15, P > 0.65; Table 1).A scatterplot of LI and detour-reaching performance suggested sex differences in the relationship between these phenotypes: detour-reaching performance of right-lateralized males was worse than that of left-lateralized males, but in females, which on average were more right-biased than males, this tendency appeared much weaker (Fig. 3).To assess this, we ran a single GLM with LI as dependent variable, using only detour-reaching performance and sex as predictors, and also including their interaction.This model suggested a strong interaction (b st ¼ À0.68 ± SE ¼ 0.34) between sex and performance in the detour-reaching task, although the null hypothesis of no interaction could not be confidently rejected (P ¼ 0.080; full model results in the Appendix, Table A5).
The best model explaining individual differences in the absolute lateralization index was the null model (Appendix, Table A6), which means that we cannot exclude the null hypothesis that the strength of lateralization is not related to the studied predictors.Thus, we did not perform model averaging for models of ALI.

DISCUSSION
Studying a wild-caught population of common waxbills living in a large open-air mesocosm, we found that most individuals consistently fed with the right side of their body towards the food, resulting in a population level right-side behavioural lateralization.Figure 2. Distribution of (a) the lateralization index (LI) and (b) the absolute lateralization index (ALI) for all individuals with 10 or more valid feeding observations.LI varies from À1 (always feeding from the left side) to 1 (always feeding from the right side); ALI varies from 0 (feeding an equal number of times from each side) to 1 (always feeding from the same side).Nevertheless, we also found significant differences between individuals, both in the direction and strength of lateralization.On average, females were more right-side lateralized than males, who showed greater among-individual differences in lateralization.Individual differences in the direction of lateralization were correlated with performance in a detour-reaching task (a cognitive assay of inhibitory control; MacLean et al., 2014).Especially in males, more right-lateralized individuals had better inhibitory control ability.Finally, despite some theoretical predictions and results from other species, we did not find relationships between the direction or strength of behavioural lateralization and either morphology, personality, behavioural cues of stress or social dominance.

Population and Species Level Lateralization
Facing the seeds with the right side may indicate a preference by most individuals for examining the seeds with the right eye (Franklin & Lima, 2001), which seems to agree with the observed feeding lateralization preferences of several species (Canning et al., 2011;Robins et al., 2005;Robins & Rogers, 2004), including birds (Güntürkün et al., 2000;Mench & Andrew, 1986;Valenti et al., 2003).Also, since the experimental feeder was placed on a wall, this lateralized behaviour could instead indicate a preference for using the left eye to look away from the wall towards the surroundings, for example while monitoring conspecifics or performing predator vigilance (Franklin & Lima, 2001;Vallortigara & Rogers, 2005).Specialization of the left eye for vigilance and predator detection has been found in amphibians, mammals, birds and reptiles (e.g.Dharmaretnam & Rogers, 2005;Lippolis et al., 2002;Lippolis et al., 2005;Martín et al., 2010).Thus, the direction of population level lateralization that we found is in agreement with the pattern of hemisphere specialization in diverse species, which could suggest that it is an ancestral trait shared by diverse taxa (Ghirlanda & Vallortigara, 2004) and was inherited by waxbills.Vallortigara (2006) hypothesized that brain lateralization first appeared in early solitary chordates to increase brain efficiency, then, with the evolution of more complex sociality, social interactions favoured the alignment of the direction of behavioural lateralization across individuals.Despite a general pattern of lateralization existing across animal taxa (Bisazza et al., 1997;Lippolis et al., 2002;Sovrano et al., 2001), some exceptions to this pattern exist, and there are examples of species, even closely related ones, that are differently lateralized (e.g. Brown et al., 2007;Franklin & Lima, 2001;Vallortigara et al., 1999).
A second, alternative hypothesis is that the waxbill population level lateralization is not inherited or innate, but results from a plastic adaptation for group coordination.Possible advantages of lateralization at the population level may include better coordinated collective behaviour (Rogers, 2021), such as in population level lateralized fish that can stay in large shoals and escape in the same direction, diluting the probabilities of being predated (Vallortigara, 2006).In the case of waxbills, it would not have to be shared lateralization in feeding per se that confers collective advantages, but perhaps it is symptomatic of population level lateralization in other behaviours where group coordination is beneficial (Vallortigara & Rogers, 2005).
In favour of this second hypothesis, we found no evidence for species level lateralization when we analysed photographs of waxbills feeding or perching in grass stems from around the globe.If the common waxbill is not lateralized at the species level, then population level lateralization could be the result of an ontogenetic process whereby most individuals copy the behaviour of the majority (Collet et al., 2023), perhaps in that way benefiting from group coordination.We need to be cautious, however, concluding an absence of species level lateralization, because the feeding task that we studied does not correspond exactly to how waxbills feed in nature.For example, in the wild, waxbills perch vertically or in near-vertical perches (grass stems) to feed, which may cause more muscular fatigue on one side of the body and, thus, encourage alternation of sides.Also, in our study system there was a wall blocking the view behind the feeder, so that the lateralization we found may in fact be a left-eye preference for vigilance, which would not be apparent when feeding from grass stems.minority of waxbills appeared not to be lateralized or were lateralized in the opposite direction.Within-population differences in behaviour can either result from adaptation to changes in selection pressures or environmental conditions over space or time (Rogers & Vallortigara, 2015;Gomes & Cardoso, 2020), or from frequency-dependent selection (Vallortigara & Rogers, 2005).For example, left-handedness in humans is thought to have been maintained by frequency-dependent selection, and even today an advantage of left-handers can be observed in competition contexts such as sports, due to their less predictable movements (Raymond et al., 1996).

Among-Individual Differences in Lateralization
The among-individual differences in lateralization were not random with respect to other phenotypic differences.More leftlateralized waxbills had better inhibitory control, assessed by a detour-reaching task (Boogert et al., 2011;MacLean et al., 2014).This suggests that possible selective advantages of, or frequencydependent selection for, being lateralized differently may be related to other associated behavioural differences rather than to lateralization per se.Interestingly, the association between inhibitory control and lateralization was not due to individual differences in the strength of lateralization, but specifically to differences along the lefteright lateralization continuum.This contrasts with the common finding that, in other species, more lateralized individuals, irrespective of direction, are better at various behaviours (Brown & Magat, 2011;McGrew & Marchant, 1997;Mench & Andrew, 1986).Similarly to our result, however, recent work with zebrafish, Danio rerio, found an association between the direction of lateralization in a social task and inhibitory control: zebrafish preferring to use the right eye to view themselves in a mirror were better at supressing attack towards prey inside a transparent tube (Lucon-Xiccato et al., 2020).Despite the type of lateralization studied in zebrafish (looking at a mirror image) and in waxbills (feeding) being different, both studies indicate that the direction, rather than the degree, of lateralization can also be associated with differences in cognition.Our result with waxbills further suggests a potential advantage of being lateralized in the minority direction, reflecting improved inhibitory control, which may help explain the persistence of some individuals in the population to be lateralized differently.
Finally, the association between inhibitory control and lateralization was stronger in males.There was also a sex difference in the mean direction of lateralization, with female waxbills being consistently right-side lateralized and males spanning the righteleft side continuum of lateralization, with a minority showing left-side lateralization.The sexes may differ in lateralization if sex differences in behaviour or life history (e.g.behaviours related to courtship or reproduction) influence lateralization (Ariyomo & Watt, 2013;Reddon & Hurd, 2008).In waxbills, males not only differed from females in the mean value of the lateralization index but were also more diverse in the direction and strength of behavioural lateralization and showed a stronger association between the lateralization index and inhibitory control.This wider variation in male lateralization agrees with the 'greater male variability hypothesis', which has been widely studied in humans but for which there is little evidence in animals (Branch et al., 2020;Harrison et al., 2022).Together, these results suggest that functions related to male behaviour may be important in explaining individual differences in lateralization.Differences among males in behaviours related to mating exist in several species, sometimes maintained by frequency-dependent selection (Shuster, 2010).For example, male goldbelly topminnow, Girardinus falcatus, have a preferred side to attempt copulation and, since females may be more vigilant of unwanted copulations from the side used by most males, the minority of males preferring the other side may have an advantage (Vallortigara & Bisazza, 2002;Vallortigara & Rogers, 2005).Likewise, if differences among male waxbills in lateralization, or in associated traits like inhibitory control, influence courtship, mating or other behaviours relevant for reproduction, then perhaps the greater diversity in lateralization among males is due to differences in mating strategies.
Apart from sex differences in lateralization and the association with detour-reaching performance, we found no further associations of behavioural lateralization with individual differences in other phenotypes: body size, personality type, cues of stress or social dominance.These results contrast with studies in some species where behavioural lateralization was associated with, for example, boldness (Found & St Clair, 2017;Reddon & Hurd, 2009), fearfulness (Rogers, 2009) or stress (Byrnes, Pouca, & Brown, 2016;Ocklenburg et al., 2016), but agree with other studies that, like ours, did not find associations of lateralization with these phenotypes (e.g.Byrnes, Pouca, Chambers, & Brown, 2016;Masilkova et al., 2022).Perhaps if studying other aspects of behavioural lateralization, or using larger sample sizes, some of these associations could be demonstrated in waxbills.Our results, however, suggest that sex and cognition (here, inhibitory control) are the traits most strongly associated with behavioural lateralization.The association with a cognitive trait supports the theory that behavioural lateralization is the result of cerebral hemisphere specialization with implications for cognition (Frasnelli & Vallortigara, 2018).In this respect, our results are among the few (e.g.Bibost & Brown, 2014;Lucon-Xiccato et al., 2020) indicating that the direction of behavioural lateralization, rather than its strength, may reflect cognitive differences among individuals.Our results are also the first indicating that associations between behavioural lateralization and cognition can depend on sex, suggesting that sex-specific functions are implicated in the maintenance of among-individual diversity in behavioural lateralization.

Figure 1 .
Figure 1.(a) Experimental set-up used to study behavioural lateralization, consisting of (1) a feeder and (2) a perch with (3) four RFID antennas underneath.Video snapshot showing (b) a waxbill feeding from its left side and (c) another feeding from its right side.

Figure 3 .
Figure 3. Scatterplot illustrating the relation between detour-reaching performance and the lateralization index.Green points and regression line denote females (N ¼ 5); orange points and regression line denote males (N ¼ 8); black dashed regression line uses data from both sexes.

Figure A1 .
Figure A1.Distribution of feeding observations for every month across the sampling period (May 2021 to May 2022) for the different individuals.

Table 1
Model averaging results for the relation between lateralization index and predictors st : standardized coefficients.N ¼ 13 birds.

Table A4
Model selection for the GLM of lateralization index showing all models within 6 AICs of the best model Body size Change in breath rate Dominance score Detour-reaching performance Mean breath rate Mirror test Sex Tonic immobility AICc DAICc Akaike weight Akaike's information criterion corrected for small samples.Bold indicates the best model (i.e.lowest AICc), which differed from the second-best model by >2AICc.Results from the GLM of the lateralization index (LI), with detour-reaching performance, sex and their interaction as predictors st : standardized coefficients; CI: confidence interval.N ¼ 13 birds.Significant P values (< 0.05) are shown in bold. b

Table A6
Model selection for the GLM of absolute lateralization index showing all models within 2 AICs, which includes the null model .M. Santos et al. / Animal Behaviour 214 (2024) 43e54 P