Toward on-farm measurement of personality traits and their relationships to behavior and productivity of grazing dairy cattle

There is growing evidence that animal personality is linked to a range of productivity traits in farm animals, including dairy cattle. To date, the methodology for assessing personality traits of dairy cattle is time consuming and often requires a test arena, which limits the opportunity for commercial farms to use personality traits of dairy cattle for individualized management. This study investigated whether personality traits of pastured dairy cattle, scored in short behavioral tests, are associated with daily behavioral patterns and milk production. Cows (n = 87) were exposed twice to each of 5 behavioral tests, where their responses to novel or putatively stressful situations were scored on an ordinal scale for investigative and reactive behavior toward a novel object and a novel human after exiting the milking parlor, response to restraint in a crush, avoidance distance from an approaching human in the paddock, and response to milking (step-kick behavior). Most behavior test scores were consistent over the 2 test repeats (using repeatability estimates, Mann-Whitney U test of difference between repeats, and Spearman rank correlation). Behavior test scores were subjected to a principal components analysis that revealed in-tertest relationships in 3 factors of correlated sets of test scores, interpreted as personality traits (fearful of humans, calm-investigative, and reactive to milking). Regression analyses determined how these traits were associated with daily grazing, ruminating, and lying behaviors, and milk production (after controlling for cow age, breed, lactation status, group, and climate variables). Cows that were more fearful of humans (high avoidance distance, reactive toward the novel human) had reduced lying time compared with cows that scored low on this trait. Cows that were more calm (during restraint) and investigative (toward the novel object) had greater grazing time, which likely contributed to their greater milk production compared with cows that scored low on this trait. Cows that were more reactive to milking produced less milk than cows that scored low on this trait. These results indicate that individual differences in daily behavior patterns and milk production of dairy cattle are associated with personality traits of cows, measured using several short behavioral tests. These methods may be useful for characterizing grazing dairy cattle on commercial farms, which could aid in understanding individual behavior patterns and provide opportunities for individualized management


INTRODUCTION
Individual cows are known to differ in their daily behavioral patterns; for instance, pastured cows spent between 6 and 13 h/d lying (New Zealand; Hendriks et al., 2019), between 5 and 15 h/d lying (Australia; Beggs et al., 2018a), and between 7 and 16 h/d lying (Brazil; Thompson et al., 2019).There are also reports of large differences in grazing and ruminating behavior of dairy cows (e.g., 7 to 10 h/d grazing, 6 to 9 h/d ruminating; Pérez-Prieto and Delagarde, 2012).It is necessary to understand the factors contributing to individual variability in feeding and lying behavior of dairy cows so that management may be tailored to the needs of individuals.Time spent away from the pasture and waiting time at the milking parlor are 2 variables known to be negatively associated with lying behavior of cows (Beggs et al., 2018b;Neave et al., 2021), but waiting time at the milking parlor only explains 14% of the individual variation in lying time (Beggs et al., 2018a).This suggests that individual cows may compensate for this challenge in different ways; while some cows reduce their lying time, others may adjust time spent engaged in other behaviors.The reasons behind these individual differences in behavioral patterns of grazing dairy cattle are still poorly understood.
Behavioral patterns that are consistent within individuals, but vary between individuals, are referred to as "personality," and specific aspects of the suite of behaviors expressed by animals are termed "personality traits" (Carter et al., 2013); for example, common traits may include fearfulness or sociability.There is increasing evidence that personality is associated with measures of growth and productivity in farmed animals, including dairy cattle (reviewed by Haskell et al., 2014;Neave et al., 2018b).Personality is also associated with milk production of dairy cows (Breuer et al., 2000;Hedlund and Løvlie, 2015), and several studies have shown a relationship between personality and dayto-day behaviors.For example, dairy cows with a lower tolerance for a novel object and an unfamiliar human had fewer lying bouts of shorter duration in their home environment (MacKay et al., 2014).Individual variation in feeding behavior was also related to personality traits in dairy calves; individuals with greater exploratory behavior in a novel arena had greater solid feed intake (Neave et al., 2018a), and calves with greater avoidance of novel object and human reached a grain intake weaning criteria earlier (Neave et al., 2019).To our knowledge, there are no reports linking personality with feeding behavior variability in adult dairy cows measured using similar approaches, but evidence in feedlot finishing beef cattle indicates reduced feed intake and feeding duration in cows that were more reactive during restraint and release from a crush (Cafe et al., 2011;Black et al., 2013).In addition, Ramseyer et al. (2009) found that less sociable beef cows (measured as having fewer affiliative partners and low group cohesion) were those that initiated herd movement to a new foraging site after a resting period; thus, specific individuals may be more influential in determining the behavior of others in the herd while at pasture.To our knowledge, only one study in dairy cattle has linked a behavioral trait (low or high residual feed intake as calves) with differences in grazing and ruminating patterns (Gregorini et al., 2015).However, no studies have examined associations between personality traits (measured as behavioral responses to novel, stressful, or social situations), daily behavior patterns, or milk production in pastured adult dairy cattle.
The objective of this study was to determine whether personality traits, measured using a series of short standardized behavioral tests, are associated with daily behavior patterns (grazing, lying, and ruminating time) and milk production of dairy cattle managed on pasture.We hypothesized that cows who are more reactive toward novel or potentially stressful situations may have reduced milk yield, possibly with reduced lying time as a contributing factor.In contrast, cows who are less reactive in these situations may have higher milk yield, which might be related to more time spent grazing or ruminating.We also determined which of the 5 personality tests were most useful in predicting behavioral patterns and productivity of dairy cows.

MATERIALS AND METHODS
This study was conducted at the Lincoln University Research Dairy Farm in Lincoln, New Zealand (−43.639latitude, 172.456 longitude) from November 11 to 30, 2019.All procedures were conducted according to the guidelines of the New Zealand Animal Welfare Act 1999 and were approved by the Ruakura Animal Ethics Committee in Hamilton, New Zealand (no. 14876).

Animal Management
The daily behavior patterns of 87 cows were monitored over a 16-d study period.This was a convenience sample of cows that were enrolled in a concurrent study examining different milking frequencies (Edwards et al., 2022).This sample size was expected to achieve sufficient individual variation in the outcome behaviors of interest (grazing, ruminating, and lying time) and in behavioral responses in the personality tests.A previous report described personality traits of adult dairy cattle using similar tests with a sample size of 22 to 26 cows (Neave et al., 2020), which is a similar to the 29 cows per group in this study.Animal management and cow enrollment were the same as reported in a companion study (Neave et al., 2021).All cows were in peak lactation (80 ± 19 d postpartum) due to the seasonal block calving system in New Zealand, with a mean ± standard deviation (SD) milk production of 22.4 ± 4.9 kg/d (range: 9.7 to 33.0 kg/d).All cows were cross-bred (Friesian × Jersey), with 48 predominantly Friesian, 8 predominantly Jersey, and 31 an equal split of Friesian and Jersey, and a mean ± SD age of 3.7 ± 2.0 yr (range: 2 to 12 yr).Cows were pseudorandomly allocated to and managed in 3 groups (n = 29 cows each) on the same farm; cows joined their assigned group the day after calving.Primiparous cows were blocked for liveweight and genetic merit, whereas multiparous cows were blocked for age, genetic merit, expected calving date, BCS, and liveweight at the end of the previous lactation, previous lactation days in milk, milk weight, and milk solids production.Cows were milked twice daily by 1 person in a herringbone milking parlor at 0500 h and 1500 h; group order of milking was maintained during the study period.All cows were returned to the paddock together as a group, so daily time spent away from the paddock was similar for all cows in a given group.Each group was on a grazing rotation of 10 × 0.75-ha paddocks with a total stocking rate of 3.5 cows/ha.Paddocks were divided into 4 strips, with 1 strip grazed after each milking, and were rotated to a new paddock every 2 d (following a 20-d paddock rotation length).Paddocks were blocked for previous management and walking distance to the milking parlor, and randomly assigned to a group.Daily pasture DM allocation was targeted at 19 to 20 kg of DM/d per cow (perennial ryegrass-white clover pasture, consisting of 80% ryegrass, 4% clover, 4% weeds, 11% dead, and 1% other grasses; with 19% DM, 17% CP, 45% NDF, and 24% ADF).Cows were visually monitored daily for body condition.Two cows from group 2 received 0.5 kg of barley supplement at each milking (1 kg/d) due to light body condition; in particular, this supplement was received throughout the study for 1 cow, and in the last 2 d of the study for the second cow.

Automated Daily Behavior Data Collection
Daily grazing and ruminating times were monitored with an electronic ear tag (CowManager, Agis Automatisering BV) attached to the existing ear tag while the cow was restrained in a crush.The electronic ear tags automatically downloaded data to a server when the cow entered the milking parlor twice per day.Daily lying time was monitored with a pedometer (IceQube data loggers, Ice Robotics Ltd.,) attached to the right hind leg while in the milking parlor.Both devices have been validated in grazing dairy cattle (CowManager ear tag: Pereira et al., 2018;IceQube pedometer: Mc-Gowan et al., 2007).The pedometer loggers stored data on the logger until weekly manual downloading using a triggering device in the milking parlor.Milk meters (DeLaval) automatically recorded milk production at each milking for each cow.Cows were monitored for paddock behaviors (grazing, ruminating, and lying time) from d 0 to 16, beginning after completion of the personality tests described below.

Personality Tests
We measured personality traits of each cow in a series of short, standardized tests adapted from previous literature; these tests aimed to characterize cows' responses to novel situations or putatively stressful situations, or both.Five tests were conducted over consecutive days with 1 test per day in the following order: novel object, restraint, novel human, avoidance distance, and response to milking tests.All tests were performed twice for each cow, once at the beginning of the study period before behavioral monitoring (repeat 1; d −4 to d −1) and again at the end of the study (repeat 2; d 13 to d 16).Two observers scored cows from video recordings (GoPro Hero7, GoPro Inc.) of the restraint, novel object, and novel human tests, and a single observer (H.N.) live-scored behavior of cows in the avoidance distance and response to milking tests.All scores were performed by the observer after establishing inter-and intra-observer reliability [Cohen's kappa (κW) > 0.70] for each test.Reliability of live scoring was performed at the research site and with the same milker for the study cows.
In the novel object and novel human tests, cows were held in the race as they exited the milking parlor.Then, each cow continued approximately 4 m along the race around a corner to encounter either a novel object attached to the railing of the alleyway or an unfamiliar woman standing immobile, wearing a high-visibility jacket and pants with her hands in the jacket pockets.Cows waiting in the alleyway before the test were unable to see the object or human.The cow was able to approach and inspect the object or human if she wanted and could move past at her own pace; however, after 30 s, the cow was gently encouraged to move past by a person positioned in the alleyway, behind the test cow, and preventing other cows from entering the test alley.The objects were a bright orange foam pool "noodle" toy with a horse-shaped head at one end (length of 140 cm) in repeat 1, and a colorful round inflatable pool "ring" toy (72 × 64 × 22 cm) in repeat 2. A different unfamiliar woman was used in each of repeat 1 and repeat 2. Investigative behavior (score 1 to 6, where 1 was no response and 6 was approach and touch without hesitation) and reactive behavior (scored as total number of reactive behaviors from: stop, avoid, increase pace, startle, step back or reverse, and turn around) of each cow during the novel object and human tests was scored from the recorded video following Gibbons et al. (2009a; see Table 1).
In the restraint test, cows walked individually down a race into the crush where their head was restrained in the head bail for 15 s and then released (body crush was not applied).The response to restraint was scored from the recorded video as either calm, uneasy, struggle, or violent following Gibbons et al. (2011) (see Table 2).
In the avoidance distance test, grazing cows were individually approached by a moderately familiar human (author H.N.; cows had previously encountered H.N. during milking and in the paddock briefly for 1 wk before study start) approximately 1 h after morning milking.The human stopped approximately 10 m away from 1 cow, and after making eye contact, slowly approached the cow with a pace of 0.5 m/step every second.The right hand was extended palm upward during the approach, and clipboard was held next to the body in the other hand.When the cow responded by moving 1 hoof backward or sideways, the test was stopped and distance from the point of the hoof just before move-ment to the human was recorded as the number of steps and given a distance classification (following Sutherland and Dowling, 2014; see Table 2).If a cow was lying down at the time of the test, she was approached in the manner described above to encourage her to stand up, and the test was performed 1 min later.Cows received a score from 1 to 6 for investigative response, and could receive one or more classifications for reactive response, which were then tallied for total number of reactive responses.Scores and descriptions are modified from Gibbons et al. (2009a).Retreats when human is within 1 m (0, 1, or 2 steps).Does not permit hand to touch muzzle. 2 Retreats when human is within 1 to 2 m (3 steps).3 Retreats when human is within 2 to 3 m (4 or 5 steps).4 Retreats when human is more than 3 m away (6 or more steps).Response to milking, during cluster attachment 4  Step Any movement of the hind hoof, excluding a kick.May be a step of any height.Score total number of steps from moment the milker places hand on cluster to moment milker removes hand from cluster or stops handling cow or udder.

Kick
Forceful and rapid movement of the hind leg backward or forward, or toward the milking cluster.Score total number of kicks from moment the milker places hand on cluster to moment milker removes hand from cluster or stops handling cow or udder.
In the response to the milking (cluster attachment) test, the number of steps and kicks were scored from the moment the milker approached the cow until the milker moved away from the cow after attaching the cluster (following Van Reenen et al., 2002; see Table 2).The test was performed for all cows handled by the same milker.
We expected this series of tests to measure 3 personality traits: reactivity to humans (using the scores in the avoidance distance test and reactive behaviors in the novel human test), curiosity (using the investigative behavior scores in the novel human and novel object tests), and reactivity to other situations (using scores in the restraint test and response to milking test).

Climate Data
Maximum daily temperature (°C) and total daily rainfall (mm) were extracted for each study date from the National Climate Database station located 3 km away from the research site (Station 17603; NIWA, 2021).

Statistical Analysis
Data Handling.All statistical analyses were performed using SAS (Studio; SAS Institute Inc.).Data from the electronic ear tags devices (reported as total min/h for each behavior) were summarized into daily durations of grazing and ruminating for each cow.Daily data from the automated behavior devices were then examined for biologically unlikely measurements (such as zero, or grazing, ruminating, or lying times above or below 3 SD from the mean) and excluded from analysis (1% of all data).All behaviors were normally distributed (assessed using PROC UNIVARIATE and model residuals).Frequency distributions of test scores were examined and score categories were combined if fewer than 3 cows were in a category (outlined in Table 1 and Table 2).This occurred for the following tests: novel object reactive behavior [revised scores: 0, 1, 2, or 3 or more (up to 6) total reactive behaviors], novel human investigative behavior (revised scores: 1, 2, 3, or 4 and 5; no cows scored 6), and novel human reactive behavior [revised scores: 0, 1, 2, or 3 or more (up to 6) total reactive behaviors].Due to few cases of kicks in response to cluster attachment, response to milking was scored as total number of steps and kicks.
Intratest Consistency of Behavioral Tests.Following guidelines outlined by Carter et al., (2013), we examined whether our behavioral tests met the requirement of consistency over time (intratest consistency) to be considered "personality tests" in dairy cattle.
Scores in each of the tests were assessed for consistency between test repeats using 3 statistical approaches, following Gibbons et al. (2009a).Each of these mathematical measures of consistency have their limitations, and so we chose to use several techniques to make an overall assessment of each behavioral measure.Repeatability estimates were calculated from the betweenand within-cow variance components computed from a mixed model with restricted maximum likelihood (PROC MIXED) with test repeat and cow fitted as random effects (as described in Gibbons et al., 2009a).Repeatability was calculated following Lessells and Boag (1987), where repeatability above 0.50 indicates more variation (Var) between than within cows, and thus test repeats are more similar than different:
Test scores were ordinal and not normally distributed, so nonparametric analyses of intratest consistency were used to test for differences in test scores between test repeats (using Wilcoxon Signed Rank; PROC NPAR-1WAY), and to test for positive association between test repeats (using Spearman rank correlation; PROC CORR).Some cows were unavailable for both test repeats in all 5 personality tests, resulting in n = 85 for the novel object, novel human, and avoidance distance tests, n = 83 for the reactivity to milking test, and n = 76 for the restraint test.
All tests except investigative and reactive behavior in the novel object test showed consistency between test repeats (see Results).To determine how scoring of the novel object test could achieve adequate consistency under on-farm conditions, scores were recategorized into 3 levels for investigative behavior (minimal response, hesitation and attentive, or approach the object) and reactive behavior (0, 1 to 2, or ≥3 reactive behaviors toward the object).This scoring approach resulted in improved repeatability (see Results), so these scores were used in further analyses.

Identifying Personality Traits (Intertest Consistency of Behavioral Tests).
A second requirement for behavioral tests to be considered "personality tests" is that behaviors should show consistency across contexts (intertest consistency; Carter et al., 2013).A principal components analysis (PCA) was used to identify common sets of behaviors across the 5 personality tests, and these correlated sets of behaviors were interpreted as describing specific personality traits.This statistical approach followed methodology used in dairy calves by Neave et al. (2018aNeave et al. ( , 2019) ) and followed the suggested approach in behavioral ecology for examining how personality traits affect life-history productivity and fitness (Carter et al., 2013).The PCA reveals orthogonal relationships among the tests, and thus whether tests expected to measure the same personality trait are correlated (demonstrating convergent validity of the tests), and whether tests expected to measure different personality traits are not correlated (demonstrating divergent validity of the tests).
The test scores from the first test repeat were used in the PCA because averaging test scores across test repeats is inappropriate for ordinal or categorical scores, and because performing one set of tests on commercial farms is most practical.The 7 test variables entered into the PCA were those that achieved adequate intratest consistency as described above: investigative and reactive behavior scores in each of the novel object and novel human tests, restraint test score, avoidance distance test score, and response to milking test score.There were 82 cows that had all 7 test scores available in the first test repeat and thus were subjected to the PCA.The correlation matrix was computed and orthogonal (varimax) rotation was applied using PROC FACTOR; factors (i.e., components) were retained after scree plot examination and if eigenvalues were ≥1.All variables met the criteria of Kaiser-Meyer-Olkin measure of sampling adequacy ( =0.52) required for conducting PCA (Budaev, 2010).Individual scores on each of the retained factors were extracted using the regression method, resulting in cows receiving a score on a continuous scale for each factor (interpreted as a personality trait).
Associations Between Personality Traits and Daily Behavior Patterns and Productivity.Each retained factor (i.e., personality trait) from the PCA was tested for its effect on daily behavior patterns (time spent grazing, ruminating, and lying), using a repeated measures regression model (PROC MIXED).Each cow was the observational unit (n = 82) and experimental day (d 1 to 16) was the repeated measure with a compound symmetry covariance structure based on lowest Akaike's information criterion.The Satterthwaite method was used to calculate degrees of freedom (df).Fixed effects included in the models were group (1, 2, or 3), cow age, breed (predominantly Friesian, predominantly Jersey, or equal mix of Friesian and Jersey), DIM, daily milk yield, maximum daily temperature, and daily rainfall.
Each factor (i.e., personality trait) was also tested for its effect on milk yield using the same model structure and fixed effects described above.Additional fixed effects included in this model were daily grazing, ruminating, and lying time.
Identifying the Best Test to Predict Behavior and Milk Yield.We were interested in determining which test scores were most useful in predicting behavior patterns and milk yield of dairy cows; as a result, this would offer insight into which personality test is most informative for future work on commercial farms.Average daily grazing, ruminating and lying times, and average daily milk yield were calculated by summarizing daily values across the 16-d study period.A best subsets regression procedure (PROC GLMSELECT) identified the best regression model out of the 7 test scores that predicted average daily grazing, ruminating and lying times, and average milk yield.A backward selection procedure was used, and selection stopped when all effects in the model were significant at α < 0.1.
All results are reported as significant at P ≤ 0.05 and tendency at 0.05 < P < 0.1, and test statistics are reported as F or t (df) .

Intratest Consistency of Behavioral Tests
Individual cows were highly variable in how they responded to the 5 standardized personality test situations; therefore, the frequency distributions of test scores at each test repeat are presented in Supplemental Figures S1 to S3 (https: / / data .mendeley.com/datasets/ y9jg7xd4fp/ 1; Neave et al., 2022).All test scores except for investigative and reactive behavior in the novel object tests were adequately consistent between test repeats (Table 3).This was demonstrated by (1) a high repeatability estimate (>0.50) for the restraint test, response to milking test, and reactive behavior in the novel human test; (2) no significant difference between test repeats for novel human, avoidance distance, reactivity to milking and restraint tests; and (3) significant positive association between test repeats for novel human, avoidance distance, reactivity to milking, and restraint tests.When investigative and reactive behaviors in the novel object test were recategorized into only 3 levels, repeatability estimates improved for both, but test repeats remained significantly different and only tended to be positively associated.The test that achieved the highest consistency between test repeats (according to all 3 analysis approaches) was the restraint test (see Table 3).

Identifying Personality Traits (Intertest Consistency of Behavioral Tests)
The PCA revealed the relationships between test scores, which could be condensed into 3 factors explaining 63.0% of the variation across tests.Factor loadings (coefficients) for each variable are presented in Table 4.Each factor was interpreted as reflecting a personality trait, and cows could score high or low on multiple personality traits.Cows scoring high on factor 1 (more reactive behavior in the novel human test, higher investigative score in the novel human test, and greater avoidance distance) were labeled as "fearful of humans."Cows scoring high on factor 2 (higher investigative score in the novel object test, and calmer during the restraint test) were labeled as "calm and investigative."Cows scoring high on factor 3 (more steps and kicks during milking) were labeled as "reactive during milking." This pattern of factor loadings revealed multiple tests that measured a common aspect of individual cow behavior (in the case of the avoidance distance and novel human tests, which loaded together on factor 1: "fearful of humans"), and identified a single test that measured multiple aspects of individual cow behavior (in the case of the novel object test, where investigative behavior, but not reactive behavior, loaded on factor 2: "calm or investigative").Against expectations, investigative behavior measured in 2 different contexts (novel human and novel object tests) were not related (loaded on factor 1 and 2, respectively), indicating that these test scores measured different personality traits.

Associations Between Personality Traits and Daily Behavior Patterns and Milk Yield
Daily grazing, ruminating, and lying time of cows in each group over the 16-d experimental period are presented in Figure 1.Mean daily milk yield of each group [±SD, (minimum to maximum)] was 22.6 ± 5.1 (9.7 to 30.7), 22.8 ± 4.8 (14.2 to 33.0) and 21.6 ± 4.6 (12.4 to 29.4) kg/d for group 1, group 2 and group 3, respectively.This individual variation in daily behaviors and milk yield among cows was associated with certain personality traits of the cows.
Cows scoring higher on the factor 2 trait ("calm and investigative") had greater daily grazing time (estimate: 21.8 ± 10.8 min/d; predicted mean ± SE: 460.9 ± 34.1 min/d (7.7 ± 0.6 h/d); F 1,75.1 = 4.1; P = 0.05) and greater daily milk yield (estimate: 1.6 ± 0.4 kg/d; predicted mean ± SE: 22.5 ± 1.2 kg/d; F 1,68 = 15.6;P < 0.001).Cows that were most (versus least) "calm Investigative behavior was first analyzed using the original scored categories (5 possible scores: minimal response, some hesitation, full stop to investigate, approach with hesitation, or approach without hesitation), but due to low consistency, scores were recategorized into 3 possible scores (minimal response, hesitation and attentive, or approach).The results for the 3-score analysis are in parentheses. 3 Reactive behavior was first analyzed using the original scored categories (0, 1, 2, or ≥3 types of reactive behaviors toward the object), but due to low consistency, scores were recategorized into 3 possible scores (0, 1 to 2, or ≥3 types of reactive behaviors toward the object).The results for the 3-score analysis are in parentheses.

Characterizing the Personality Traits of Dairy Cattle
The characterization of personality traits in dairy cattle can be time consuming and impractical to perform on a larger scale; for example, many studies perform lengthy individual tests in separate arenas (see review by Forkman et al., 2007).The personality tests  we chose were designed to be short and practical for onfarm use: they could be done at the time of milking, immediately after milking, or briefly in the paddock.Four of the 5 original behavioral test scores (novel human, avoidance distance, restraint, and response to milking tests) were similar between test repeats, demonstrating consistency over the 2-wk period and meeting one of the requirements for a test to be considered a personality test in animals (Carter et al., 2013).Previous work in indoor-housed dairy cattle using similar scoring methods also demonstrated consistency of avoidance distance in response to an approaching human (Gibbons et al., 2009a) and response to milking (number of steps and kicks; Van Reenen et al., 2002;Hedlund and Løvlie, 2015), but not response to restraint in a crush (Gibbons et al., 2011) or investigative and reactive behavior toward a novel object (Gibbons et al., 2009a).However, when dairy cattle were exposed to a novel object in an arena, behavioral responses were more consistent over time (Hedlund and Løvlie, 2015;Neave et al., 2020).
We used scores from the first test repeat (as these are most practical to conduct on-farm in the future) to identify 3 personality traits.The results of our study suggests that behavioral responses of dairy cows toward novel or stressful situations do not appear to be driven by a common underlying trait.For instance, behavioral responses to situations involving humans were most highly correlated (indicated by our factor 1, and showing consistency of response to humans across contexts) and dissimilar to the other putatively stressful or novel situations, similar to Gibbons et al. (2009).Behavioral response to milking (number of steps and kicks at cup attachment) was also not related to other test scores.Hedlund and Løvlie (2015) also found no relationship between stepping and kicking at milking and behavioral responses in a novel object test.An unexpected result was that both high investigative and reactive behavior in response to the novel human loaded on the same factor, suggesting both scores measured the  same trait.This may have occurred because no cows in our study sample approached and touched the human; instead, the highest "investigative" score observed were those cows that stopped to look at the human, with just one cow extending her head from a distance.In contrast, the lowest "investigative" score were those cows that showed no response to the human, walking past without hesitation or looking.Thus, "investigative behavior" to the human in our study context appeared to be attentive or vigilance behavior, which points to a difficulty in terminology use.Future work is needed in a larger sample of cows to determine whether cows that voluntarily approach and touch the human correlates with inspective behavior of other novelty, potentially revealing an underlying trait of curiosity.Overall, our results indicate that several behavioral tests are required to adequately measure the underlying traits driving behavioral responses to humans, and novel, restraint or milking situations.However, these behavioral tests do not characterize the full extent of the personality of dairy cattle.Other personality traits described in dairy cattle include aggression, sociability, and activity (Gibbons et al., 2009b(Gibbons et al., , 2010;;Foris et al., 2018;Neave et al., 2020), but traits (such as creativity, cooperation, and impulsivity) that have been reported in nonhuman animals (Gosling and John, 1999) remain to be examined in dairy cows.

Associations Between Personality Traits and Daily Behavior Patterns and Milk Yield
Consideration of animal-based sources of variation in behavior often focus on cow age, parity, health status, and productivity factors (Cook, 2020;Tucker et al., 2021) or genetic variation (Løvendahl and Munksgaard, 2016), whereas few reports mention personality as a key characteristic that drives between-cow variation (Moreno García et al., 2020).The time that individual cows spend feeding, ruminating, and lying is known to be highly variable in indoor-housed cows (see reviews by Beauchemin, 2018;Tucker et al., 2021), yet only a few reports are available showing the extent of individual behavior variability of pastured cows (Beggs et al., 2018a;Hendriks et al., 2019;Thompson et al., 2019).Our companion report found that farm management factors, such as time spent away from the paddock and walking distance, contribute in part to these individual differences in behavior patterns (Neave et al., 2021), but a large part of this individual variation remains unexplained.For example, waiting time at the milking parlor only explained 14% of the variation in lying time among cows (Beggs et al., 2018a).Our study showed that personality traits of cows was associated with some of the individual variation in grazing and lying behaviors, but not ruminating behavior.
Cows that were more fearful of humans (personality trait based on scores in the novel human and avoidance distance tests) had reduced lying time.Previous work showed that indoor-housed dairy cows' tolerance of an approaching human was related to daily lying bouts, but not daily lying time (MacKay et al., 2014).These authors used a similar avoidance distance test as in our study, so the difference in our results for daily lying time may be related to different management systems.For instance, the cows in our study spent on average about 9 h/d lying down, which is in line with reports in pasture-based systems where cows typically spend less time lying down (about 7 to 9 h/d in summer) compared with indoor-housed cattle (about 10 to 13 h/d; Tucker et al., 2021).Therefore, differences in lying time among pastured cows may be especially important to consider; for instance, our study estimated a reduction of about 60 min/d in lying time for the cows that were most fearful of humans compared with those that were least fearful.There is ample evidence that dairy cows are motivated to lie down, indicating that this behavior is important; indeed, pastured dairy cows will forego feeding for the opportunity to lie down after experiencing a period of lying deprivation due to an uncomfortable lying surface (e.g., Schütz and Cox, 2014;Schütz et al., 2019).There is also some evidence in pastured dairy cows that reduced lying time may be a risk factor for developing claw horn lesions in mid lactation (Sepúlveda-Varas et al., 2018) and may increase chronic stress related to affected function of the pituitary-adrenal axis (Fisher et al., 2002).Because lying times are already known to be <9 h in cows managed in pasture-based systems such as that in New Zealand (Hendriks et al., 2019), our observed reduction in lying time for cows that were scored as most "fearful of humans" may be especially concerning for the animal's welfare, given that lying time is an important time for rumination (Schirmann et al., 2012) and sleep (Kull et al., 2019) for lactating cows.This association between lying time and "fearful of humans" personality trait may arise if farm staff enter the paddock and these fearful cows abort their lying bout.Alternatively, these cows may spend less time lying down if they are a subordinate cow (Friend et al., 1977); for example, previous work has shown that subordinate cows also avoid humans during an approach test at the feeder (Waiblinger et al., 2003).However, further work is necessary to determine whether there is a link between social hierarchy and human-animal relationship that may drive behavior patterns among dairy cows in the paddock.Overall, there is merit in investigating why lying time appears to be reduced in cows that scored as more "fearful of humans" and whether changes in management practice would improve lying times in these individuals to avoid possible risks to animal welfare.
A reduction in milk yield of dairy cows is often described as a consequence of a negative human-animal relationship (Rushen et al., 1999;Breuer et al., 2000;Hemsworth et al., 2002), but this was not observed in cows that were scored as "fearful of humans."Other work in pastured dairy cattle in New Zealand found increased milk yield in cows with greater avoidance dis-tance of approaching human in the paddock, suggesting that the relationship between milk production and response to humans is unclear in grazing dairy cattle.The level of human contact is typically higher in indoorhoused versus pasture-based management systems, so negative production outcomes may be more likely in these systems due to greater likelihood of encountering negative human contact in confinement.
Cows that were most calm or investigative (personality trait based on scores in the novel object and restraint tests) spent more time grazing (by about 1 h) without a trade-off in reduction in lying time.Several studies support that calm cows differ in their feeding behavior, including greater feeding time (Cafe et al., 2011), longer feeding bouts (Llonch et al., 2018), and greater feed intake (Black et al., 2013;Llonch et al., 2018) compared with more reactive cows (also scored as response to restraint in a crush).MacKay et al. (2014) found that indoor-housed dairy cows that made more contact with a novel object (similar to our "investigative" cows) had fewer lying bouts of less variable duration; however, these authors did not report if daily lying or standing time were affected, making a comparison with our study difficult.These studies, together with our results, support the idea that a calm or investigative personality trait may have a positive influence on daily grazing and lying behavior patterns.Although our study showed that calm and investigative cows produced more milk, another study using similar tests (restraint in crush and response to novel object) failed to show a milk production benefit (Marçal-Pedroza et al., 2020).However, these authors did not consider other factors that can affect milk production such as group, age, breed, DIM, and climate variables, which our study accounted for in the model.There were also some differences in the test scoring approaches between the two studies, highlighting the need for standardization in personality test scoring across studies in the future.
Cows that were most reactive to milking (personality trait based on behavior score during cluster attachment at milking) tended to have a lower milk yield (about 1.8 kg/d less).This result is consistent with several other studies demonstrating that a cow's behavioral response during milking is related to milk yield in both pasture-based (Sutherland and Dowling, 2014) and indoor-housed dairy cattle (Bertenshaw et al., 2008;Hedlund and Løvlie, 2015;Marçal-Pedroza et al., 2020).Previous work has described how the human-animal relationship can affect milk yield (Hemsworth et al., 2000).If we interpret greater stepping and kicking as a negative emotional response to the milking process (e.g., due to a poor human-animal relationship), the observed lower milk yield may be a consequence of activation of the sympathetic nervous system inhibiting oxytocin release and milk letdown (Rushen et al., 2001).However, lower milk yield in these cows may also have occurred if they ate less or partitioned energy differently.Pasture-based dairy cattle with little feed supplement produce considerably less milk than highproducing indoor-housed dairy cattle, so the effect of behavioral response to milking on milk yield may be different in other management systems.
Overall, our study demonstrates that several personality traits are associated with daily behavioral patterns of pastured dairy cattle, particularly grazing and lying times.This suggests that groups are not homogeneous and identifying individual cow personalities could provide opportunities for tailoring pasture management; therefore, consideration for the needs of individuals within a group can help ensure that all cows are meeting their feeding, ruminating, and lying time requirements without the constraints of only "herd-level management."For instance, other authors have suggested "customizing" herds to match habitat diversity to a range of cattle personalities (with different grazing patterns) that can exploit the land in a way that is efficient, productive, and meets individual behavioral needs (Gregorini et al., 2017;Moreno García et al., 2020).Further research is necessary to investigate this approach, as well as to examine the dynamics of herds comprising cows of similar personalities.There may be a link between personality and productivity, but a larger study that uses a temporal model with more cows would be needed to establish this more conclusively.

Identifying the Best Test to Predict Behavior and Milk Yield
To further our aim of identifying suitable personality tests for use on commercial farms, we also determined which of the personality test scores best predicted behavioral patterns and milk yield.Grazing time was best predicted by the restraint and investigative behavior in the novel human test, lying time was best predicted by the avoidance distance test, and milk yield was best predicted by the restraint and response to milking tests.However, no test was suitable to predict ruminating time.Of these tests, the restraint and novel human tests have some practical disadvantages, requiring special handling of the cow and additional personnel on-farm.On the other hand, avoidance distance and response to milking tests are advantageous because they require just 1 person and can be conducted during milking or in the paddock soon after milking.However, based on our correlational patterns among the 7 test scores, these 2 test scores alone will not describe how cows respond to novelty, so inclusion of investigative behavior toward a novel object is suggested.

Limitations
We caution the use of technology to record daily time spent grazing, ruminating, and lying as it is associated with some measurement error; both devices used in this study were validated in pastured dairy cattle, with high accuracy and precision for grazing and lying time, and moderate accuracy and precision for ruminating time (McGowan et al., 2007;Pereira et al., 2018).We also caution that our results are limited to pasture-based dairy cattle fed little supplement and managed in much smaller groups than are typical of New Zealand dairy farms (total of 87 cows across 3 groups of 29 cows, compared with a typical farm of 400 cows that may be managed in several large groups; LIC and DairyNZ, 2020).The magnitude of the effects observed in our study could be larger on farms that experience longer walking distances and time away from the paddock for milking, which are factors known to affect behavior in the paddock (Neave et al., 2021), or there could be smaller effects when supplement is fed, regardless of farm size.Furthermore, the behavioral tests used to characterize personality traits were repeated over only a short period of time (20 d), so the tests may not have been as novel to the cows, as compared with tests that had been conducted further apart.Although we found that the majority of the tests were adequately consistent in this time period, consistency of behavioral responses may change over longer periods of time.Previous literature demonstrating long-term consistency in test responses was conducted in indoor-housed cows (Neave et al., 2020), so there is limitation in extrapolating that these short-term behavioral responses are indicative of personality in pasture-based cows.Nonetheless, given that such work has not been done, we suggest there is value in discussing the development of tests that can identify personality traits in such management settings.Indeed, one of our goals was to provide suggestions for practical on-farm tests.As such, the relationships between personality traits, daily behavior patterns, and milk production found in this study were derived from the first set of behavioral tests.However, we caution that these relationships could differ depending on the season, given that weather conditions are known to affect the behavior of grazing dairy cattle (Thompson et al., 2019;Hendriks et al., 2020;Cartes et al., 2021).We also encourage these behavioral tests to be repeated in different management systems (e.g., indoor-housed, larger farms, pasture-based farms providing supplementary feed) to determine whether they are suitable for

CONCLUSIONS
We characterized personality traits of dairy cows managed on pasture using a series of short, standardized behavioral tests that could be practically conducted on-farm.Behavioral responses toward novel or stressful situations, which were moderately repeatable over a short time period, do not appear to be driven by a common underlying trait and were associated with individual variability in behavioral patterns at pasture and milk yield.These results support the growing body of literature demonstrating an association between personality traits and daily behavior and productivity of cattle.Of the 5 behavioral tests, we suggest that avoidance distance, response to milking, and novel object tests are most practical to perform on-farm, are able to predict behavior patterns at pasture, and reflect several types of personalities.These short behavior tests could be used on commercial farms to characterize individual animals and aid in tailoring farm management to meet individual cow needs.Future work should determine behavioral consistency in pastured dairy cows over longer periods of time, and how this consistency relates to other behaviors and productivity.

Figure 1 .
Figure 1.Boxplots of daily duration of paddock behaviors (grazing, lying, and ruminating) of cows in each of the 3 groups (n = 29 cows per group) over the 16-d experimental period.Red, green, and blue boxplots represent data from cows in groups 1, 2, and 3, respectively.In each boxplot, the center line represents the median of the herd, the box limits represent the interquartile range of the herd, and each individual dot represents a cow within the group.

Figure 2 .
Figure 2. Boxplots showing the relationship between daily grazing time and (A) each score in the restraint test, showing on average greater grazing time in cows that were more calm during restraint, and (B) each score for investigative behavior in the novel human test, showing on average greater grazing time in cows that were more investigative.In each boxplot, the center line represents the median grazing time of the test score, the box limits represent the interquartile range, and each individual dot represents a cow.

Figure 3 .
Figure 3. Boxplot showing the relationship between daily lying time and each score in the avoidance distance test, showing on average lower lying time in cows with greater avoidance distances.In each boxplot, the center line represents the median lying time of the test score, the box limits represent the interquartile range, and each individual dot represents a cow.

Figure 4 .
Figure 4. Visual representation of the relationship between daily milk yield per cow and (A) each score in the restraint test, showing on average greater milk yield in cows that were more calm during restraint.The center line represents the median milk yield of the test score, the box limits represent the interquartile range, and each individual dot represents a cow, and (B) number of steps and kicks in the response to milking test (continuous measure), showing decreasing milk yield with increasing number of steps and kicks.Each black dot represents a cow.
Neave et al.: COWS' PERSONALITY, BEHAVIOR, AND PRODUCTIVITY characterizing and predicting daily behavior and productivity across different contexts.

Table 1 .
Ethogram of investigative and reactive behaviors scored in novel object and novel human personality tests; dairy cattle (n = 87) from 3 groups were individually tested twice (20 d apart) as they exited the milking parlor 1

Table 2 .
Ethogram of behaviors scored for 3 personality tests: dairy cattle (n = 87) from 3 groups were individually tested twice (20 d apart) as they exited the milking parlor (restraint test), in the paddock (avoidance distance), or during milking (response to milking) Test and score/classification 1 Description of behavioral response Restraint in crush 2 Calm No or minimal body movement, few head movements, no pulling on head bail (backward, forward, or both movements).Uneasy More periods of no movement than movement.Some body movement (head, body, tail, feet); may show backward, forward, or both movements on head bail.Some tail flicks.Struggle, strain More than one instance of backward and forward movements on head bail; feet stamping; head moves up and down, or side to side, or both.Straining at bail; may show some arching of back.Violent Continuous movement, no period of no movement.Violent movement, thrashing, continual head throwing, many backward/forward movements on head bail.Kneels or falls.Back arching and back feet may slip.Avoidance distance in paddock 3 0 Permits hand to touch muzzle; or permits hand to touch muzzle and other body parts; or extends head to touch hand. 1 Neave et al.: COWS' PERSONALITY, BEHAVIOR, AND PRODUCTIVITY

Table 3 .
Neave et al.: COWS' PERSONALITY, BEHAVIOR, AND PRODUCTIVITY Measures of consistency between repeats of the 7 behavior test measures; dairy cattle from 3 groups were individually tested twice in each test, 20 d apart, and received 1Behavior scores between the 2 test repeats were considered similar if the Wilcoxon Signed Rank test was not significant, and if the Spearman correlation was positive and significant.2

Table 4 .
Neave et al.: COWS' PERSONALITY, BEHAVIOR, AND PRODUCTIVITY   Coefficients (loadings)of the eigenvectors for the first 3 factors extracted by principal components analysis (PCA) of behavioral test scores recorded from 82 cows 1 exposed to the 5 personality tests: novel object, novel human, restraint in crush, avoidance distance in paddock, and response to milking 2 Of the 87 cows in the study, 82 cows had all 7 test scores available in the first test repeat and thus were subjected to the PCA.The test scores from the first test repeat were used in the PCA because averaging test scores across test repeats is inappropriate for ordinal or categorical scores, and because performing one set of tests on commercial farms is most practical.