Social interactions, feeding patterns, and feed efficiency of same-and mixed-parity groups of lactating cows

Social dynamics in group-housed animals can have important effects on their welfare, feed efficiency, and production potential. Our objectives were to: (1) evaluate the effects of parity and social grouping on competition behavior, feeding patterns, and feed efficiency, and (2) investigate cow-level relationships between competition and feeding behavior, production, and feed efficiency. Fifty-nine Holstein cows (144.5 ± 21.8 starting days in milk, mean ± SD) were housed in a freestall pen with 30 Roughage Intake Control (RIC) bins. We evaluated the effects of parity (primiparous [PR, n = 29] vs. multiparous [MU, n = 30]) and group composition at the feed bunk (same-parity [SM, n = 39] vs. mixed-parity [MX, n = 20, 50% of each parity]) with a 2 × 2 factorial design (SM-MU: n = 20; SM-PR: n = 19; MX-MU: n = 10; MX-PR: n = 10) on competition behavior, feeding patterns, and feed efficiency. Within the pen, groups of 9 to 10 cows were considered subgroups and assigned to treatments defined by sets of 5 assigned bins (2:1 stocking density). Feed bunk competition and feeding patterns were recorded via continuous video in the first hour after morning feed delivery and 24-h RIC data, respectively. Residual feed intake (RFI) was calculated as the difference between predicted and observed dry matter intake (DMI) after accounting for known energy sinks. Linear models were used to evaluate the effects and interactions of parity and group composition on competition, feeding behavior, and feed efficiency. Within-cow correlations were performed between competition, feeding behavior, and RFI. Cows in MX, compared with SM, were involved in more competitive interactions [mean (95% CI): competitive contacts: 11.5 (8.1, 16.3) vs. 7.2 (5.5, 9.3) events; displacements: 4.0 (3.0, 5.3) vs. 2.1 (1.7, 2.7) events, and replacements: 3.5 (2.6, 4.7) vs. 1.9 (1.5, 2.5) events]. Cows in MX vs. those in SM had more bunk visits/meal ( 4.3 [3.9, 4.8] vs. 3.7 [3.4, 3.9] visits/ meal) and longer meals (31.2 vs. 27.4 ± 0.9 min/meal) and tended to have higher RFI (0.41 ± 0.3 vs. −0.21 ± 0.2) and were therefore less feed efficient. Multiparous versus PR cows had greater DMI per day (29.3 ± 0.6 vs. 25.5 ± 0.4 kg/d) and per meal (4.2 [4.0, 4.4] vs. 3.4 [3.2, 3.6] kg/meal), faster eating rates (0.14 [0.13, 0.15] vs. 0.12 [0.11, 0.13] kg/min), and fewer bunk visits/d (26.6 [24.0, 29.4] vs. 32.8 [29.7, 35.9]). Regardless of grouping or parity, cows with shorter latencies to first visit the bunk after feed delivery were involved in more competition and tended to be less feed efficient. Overall, individual cow-and group-level relationships among competition, feeding behavior, and feed efficiency play an important role in feed bunk social dynamics. At a competitive 2:1 stocking density, mixed-parity groups for lactating cows may have potentially negative animal welfare and feed efficiency implications that should be considered when selecting grouping strategies on the farm.


INTRODUCTION
Social dynamics of group-housed dairy cattle can have important impacts on their welfare, feed efficiency, and production potential.The feed bunk can be a competitive environment, yet much remains unknown about the dynamics of how cows gain access to feed and how to optimize group composition.
Parity and group composition play important roles in understanding competition and feeding behavior.
In dairy cattle, individuals of low dominance rank or younger animals are often displaced from the feed bunk by higher-ranking animals (Huzzey et al., 2006).Primiparous cows received more aggressive interactions and were displaced from the feed bunk more frequently when housed with multiparous cows, compared with only other primiparous cows (Gibbons et al., 2009).Furthermore, the dynamics between specific pairs (dyads) of cows can vary within a single group.Social network analysis (SNA) has been used to investigate dyadic relationships and patterns in groups of dairy cattle (Gygax et al., 2010;Boyland et al., 2016;Foris et al., 2019).Using observed competition behavior at the feed bunk to construct networks provides an understudied opportunity to better understand dyad-level relationships.
Social dynamics at the feed bunk may also play a role in feeding patterns and feed efficiency.Given that feed is the greatest dairy farm expenditure (Alqaisi et al., 2019), increasing feed efficiency can improve dairy farm profitability and sustainability.Residual feed intake (RFI), a representation of the unexplained variation in feed intake after considering body size, change in BW, and milk production, can be used to estimate feed efficiency (VandeHaar et al., 2016).Our understanding of feed efficiency can be improved by identifying additional energy sinks that contribute to the variation in RFI, such as competition for resources.Although some studies have evaluated RFI and feeding behaviors, such as daily time spent eating (Connor et al., 2013;Xi et al., 2016;Brown et al., 2022), our understanding of how RFI relates to competition behavior at the feed bunk is limited.
Certain grouping strategies may minimize the negative impact of excessive competition, improve feed efficiency, and enhance the welfare benefits of social housing.Additionally, improving our understanding of the interrelationships among competition, feeding behavior and feed efficiency has the potential to add clarity to both the unexplained variation in efficiency and the mechanisms underlying social dynamics in group-housed dairy cattle.Therefore, our objectives were: 1) to evaluate the effects of parity (primiparous vs. multiparous) and group composition (same-vs.mixed-parity groups) on competition behavior at the feed bunk, feeding patterns, and feed efficiency and 2) to investigate cow-level relationships among competition behavior, feeding patterns, and feed efficiency.

Animals, Housing and Treatments
The study was conducted from July to October 2020 at the University of Wisconsin-Madison (UW-Madison) Emmons Blaine Dairy Cattle Research Center in Arlington, WI.All procedures were approved by the Institutional Animal Care and Use Committee (protocol # 005658-R01-A01).
Thirty primiparous and 30 multiparous lactating Holstein dairy cows (138.3 ± 21.2 vs. 150.4 ± 21.1 DIM, respectively; Table 1) were initially enrolled.All cows were housed in the same pen with 64 freestalls and 30 roughage intake control (RIC) system bins (Hokofarm Group, Marknesse, the Netherlands), which recorded individual cow feed intake continuously.Cows were milked twice daily at 0300 and 1500 h and fed thrice daily at 0900, 1500, and 2100 h.The third daily feeding at 2100 h was added at the beginning of wk 1 of the experimental period to ensure all cows were fed to approximately ≥ 1% refusals for each bin.Fresh feed was delivered during the morning feeding; additional feed mixed in morning was added to the bins in the afternoon feed deliveries.The same TMR diet was fed to all cows, regardless of treatment assignment.Diet composition and nutrient analysis are presented in Table 2. Refusals were manually recorded daily and feeding amounts were adjusted by treatment to ensure all cows were fed ad libitum (0900: 67.2 ± 3.6 kg/bin, 1500: 33.4 ± 12.5 kg/bin, 2100: 34.1 ± 16.6 kg/bin; as fed, mean ± SD).Water was provided ad libitum via 3 automatic water troughs.
Treatments were applied to cows based on bin assignments at the feed bunk.The study used a 2 × 2 factorial design, with factors of parity [primiparous (PR) or multiparous (MU)] and group composition [same-parity (SM; n = 40 enrolled, 39 retained) or mixed-parity group (MX; n = 20) with 50% of each parity] at the feed bunk, with the latter factor randomly assigned.The sample size within each level was: SM-PR: n = 20 (enrolled, 19 retained), SM-MU: n = 20, MX-PR: n = 10, MX-MU: n = 10.We applied SNA to subgroups of 9-10 cows, representing 2 networks of each group composition and parity combination (i.e., SM-PR, SM-MU, and MX; 6 networks total); the MX subgroups included cows analyzed as MX-MU and MX-PR.Each network shared a set of 5 bins (2:1 stocking density) with equidistant spacing along the feed bunk.After bin assignment, the distribution of BW and DIM were checked to ensure equal variation across networks of each type.Cow demographics by parity and group composition are summarized in Table 1.Twenty-two MU cows had previous experience with the RIC system.All cows were trained to their assigned bins during a 4-wk period and were considered trained once ≤ 20% of daily attempted bin visits were directed to non-assigned bins (mean ± SD: 8.6 ± 5.9%; range: 0.0-20.5%).Due to health issues unrelated to the study, 1 SM-MU and 1 SM-PR cow were removed; the former was replaced before training, but the latter could not be replaced due to timing in the bin training process, resulting in 19 cows in SM-PR for the remainder of the study.Once training was complete, the experimental period lasted 45 d.

Measures
Competition Behavior.Continuous video was recorded from 10 cameras (Platinum 4.0 MP Network Matrix IR Bullet Camera, CMIP9342W-28M; LT Security Inc., Washington, NY) mounted at 3.7 m high, which were set to record with 2688 × 1520 resolution at 10 frames/s through a network video recorder (Platinum Enterprise Level 64 Channel NVR, LTN8964-8; LT Security Inc.).Video was recorded for 2 d consecutively in experimental wk 1 and 6.Four h/d were observed (3 h from 0900 h, after feed delivery, starting when all cows were released from the stall area and had access to the feed bunk; 1 h from 1500 h, after milking and feed top-up, starting when all cows had returned from the parlor); these times were based on peak daily feed bunk visits.Each cow was marked with spray paint (Tell Tail, FIL Industries Limited, Mount Maunganui, New Zealand) for individual identification.Three trained observers coded the video recordings, observed using VSPlayer (Hikvision Digital Technology, Hangzhou, China), for competitive interactions based on a sequence of possible events (defined in Table 3, sequence shown in Figure 1).Inter-observer reliability was determined on a subsample of video that included all focal behaviors; Cohen's kappa ranged from 0.64 to 0.95, indicating 'substantial' to 'almost perfect' agreement (Landis and Koch, 1977).
Initial inspection of the data revealed that 65.8% (2697/4102) of competitive behaviors at the feed bunk occurred within 1 h after AM feed delivery [h 2 = 10.2% (417/4102), h 3 = 7.3% (300/4102); h 1 after PM feed delivery/milking = 16.8% (688/4102)].Because small magnitudes of events in the subsequent observation hours could skew interpretation when averaged on a per-hour basis, we only retained the first hour after morning feed delivery for analysis.Only interactions between cows within a network (i.e., assigned to the same bins) were analyzed to evaluate the impacts of parity and assigned group composition; thus, 34% of the total competitive contacts (SM-PR = 28.2%,SM-MU = 36.9%,MX = 34.9%;proportion for each network indicating the network of the actor) and 19% of the displacements (SM-PR = 22.9%, SM-MU = 45.0%,MX = 32.0%)were excluded.After exclusion, values were summarized as the average among the 4 observation days.Additionally, the event values for each cow (mean of 4 d) were used to calculate 4 behavior "ratios" reflecting proportions of behavioral subsets (replacements, displacements, competitive contacts), and 3 behavior "indexes" reflecting the proportion of events in which a cow served as an actor (defined in Table 3).
The feed intake and visit details (time of day, duration, bin location) were recorded automatically by the RIC system.A visit was defined as a single event when a cow entered an assigned bin and associated RIC data were recorded.Other variables derived from RIC data were latency (min) to first visit the feed bunk after AM feed delivery, number of visits/d, DMI/d, eating rate (kg/min), total eating time (min/d, regardless of intake), and the maximum daily non-eating interval (min/d, longest daily period without eating).Latency to the first bunk visit was available for 20 d of the experimental period because video was needed to determine when feed was delivered.For days with first-visit data, DMI and duration during this visit were calculated, along with summed eating time within the first 30 min after AM feed delivery; 30 min was selected based on the average length of a meal (DeVries et al., 2003).Finally, the proportion of first visits to each of the 5 assigned bins was identified and the highest value retained to represent the most-chosen first bin as a reflection of possible location preference.
To evaluate meal characteristics for each cow, meal analysis (DeVries et al., 2003;Horvath and Miller-Cushon, 2019) was performed using visit duration data across the entire experimental period.In brief, interval durations between each cow's bin visits were summarized and converted to log 10-transformed frequency distributions to calculate the inter-bout criteria.The inter-bout criteria were calculated by fitting a mixture of 2 normal distributions to the log 10 distributions of inter-and intra-visit intervals using exact maximum likelihood to determine the point at which the distribution curve of within-bout (intra-bout) intervals intersected the distribution curve of between-bout (interbout) intervals (R package mixdist; Macdonald and Du, 2018).A single inter-bout criterion pooled across all individuals was calculated (20.98 min).Meal characteristics were defined as the number of meals/d, number of visits/meal, average meal time (min/meal), DMI/ meal, DMI of largest meal/d, and average inter-meal interval (min).All feed, visit, and meal related variables were summarized for each cow across the experimental period.
Milk Yield and Components.Milk yields were recorded in DairyCOMP 305 (Valley Ag Software, Spencer, MA) and summarized as kg/d for each cow.Milk samples from 4 consecutive milkings/wk were collected and preserved with 2-bromo-2-nitropro-pane-1,3-diol (Advanced Instruments Inc., Norwood, MA) and   2; not used in RFI) was assessed on 1 d in wk 1, 4, and 7 in conjunction with BW by 2 trained observers using the 5-point scale (Dairy Body Condition Score Chart, Elanco Animal Health) at increments of 0.25.

Statistical Analysis.
Missing Data Daily RIC data were missing for 1 d in wk 1 due to a power outage.During the observation periods, some cows were uninvolved in competitive interactions, and thus data were not included for 2 cows (1 SM-MU, 1 SM-PR) for all competitive behaviors, and 3 others (2 SM-MU, 1 MX-MU) for displacement and replacement variables.
Statistical Models All response variables were analyzed using R software (v. 3.6.3,RStudio)and SAS software (9.4,SAS Institute).Cow was the experimental unit.Residuals were assessed visually using graphs and numerically using the Shapiro-Wilk test for normality.Linear models (R Core Team, 2019) were used to evaluate the effects and interactions of parity and group composition on feeding patterns and feed efficiency.Non-normal continuous variables were log n transformed to improve normality and meet model assumptions.Generalized linear models were used to evaluate effects on count-based competition variables and on proportions (competition indexes and ratios) using a negative binomial distribution (PROC GLIM-MIX, SAS) and a Tweedie distribution (PROC GEN-MOD, SAS), respectively.Latency to the first bunk visit was analyzed using a gamma distribution (PROC GLIMMIX, SAS).These models included fixed effects of parity (PR, MU), group composition (SM, MX), and their interactions.Significance was defined at a threshold of P < 0.05 and tendencies as P ≤ 0.10.All values are reported as least squares means.
Pearson (normal distribution) and Spearman's rank (non-normal distribution) correlations were performed between RFI and 1) competition behavior and 2) feeding patterns.To evaluate how an individual cow's first bunk visit after fresh feed delivery in a competitive environment may impact other outcomes, Pearson and Spearman's rank correlations were performed between first-visit variables (latency, DMI, duration, most-chosen first bin proportion), eating time within the first 30 min, total competitive contacts, and RFI.
Social Network Analysis To evaluate dyadic relationships, networks of cumulative competitive interactions at the feed bunk were constructed for each of the 6 groups of cows sharing sets of bins.Matrices indicating the actor and receiver of each interaction were used to characterize each network (Farine and Whitehead, 2015).In brief, a network consisted of nodes (individuals) connected by edges (interaction/relationship between 2 nodes).Each network was considered directed (edge direction was indicated between individuals; at most 2 edges/dyad) and weighted (strength of the edges, number of interactions) to depict the magnitude of competition in each dyad.Because patterns for the competition subsets (displacements, replacements) were similar over time, only initial competitive contacts were included in the networks.Network visuals were created, and metrics (network and node level) were calculated, as described by Makagon et al. (2012) using the igraph package in R (Csardi and Nepusz, 2006).Node-level metrics included degree centrality (in: number of incoming edges for a node, number of others from which the focal cow received interactions; out: number of outgoing edges for a node, number of others toward which the focal cow initiated interactions) and strength (in: total incoming edge weight for a node, number of interactions a cow received; out: total outgoing edge weight for a node, number of interactions a cow initiated; Foris et al., 2019).Node-level degree and strength measures can be used to evaluate individual roles within the network, specifically if one individual initiates and/or receives more interactions than others.Network-level metrics included degree centralization (variation in node degree centrality to illustrate the involvement of individuals in the network) and reciprocity (extent to which pairs of nodes make reciprocal connections to each other; Makagon et al., 2012) to add insight into network cohesiveness and stability.

Competition Behavior
Compared with SM groupings, cows in MX groupings had more interactions for all competitive behavior subsets (P ≤ 0.045, Table 4).Competition behaviors did not differ between parities and across parity and group composition interactions (P ≥ 0.11).
Within the MX group composition, Table 5 descriptively reports competition categorized by interaction types: PR initiating against another PR (PR-PR), MU initiating against another MU (MU-MU), PR initiating against MU (PR-MU), and MU initiating against PR (MU-PR).On a numerical basis, more competitive interactions, including competitive contacts, displacements, and replacements, occurred between inter-parity actor and receiver dyads (PR-MU or MU-PR) compared with same-parity dyads (PR-PR or MU-MU).Inter-parity dyadic interactions also yielded larger ranges across cows for event count variables compared with same-parity dyads.However, the indexes and ratios did not differ across dyad types, but showed high individual variation.
Social networks are shown in Figure 2 and associated descriptive SNA metrics are summarized in Table 6.Individuals in each network interacted with a majority, if not all other cows, resulting in similar node-level degree values for initiated (out) and received (in) interactions and similarly low network-level degree centralization for all 6 networks.Numerically, cows in MX were involved in more competitive interactions than those in SM, with large variation among cows as shown in the strength values.Reciprocity was greater in both MX and SM-PR compared with SM-MU, indicating that in the former groupings, dyads initiated and received interactions in both directions more often than in the latter.

Feeding Patterns
All feeding patterns did not differ across parity and group composition interactions (P ≥ 0.29; Table 7).For the main effect of parity, MU vs. PR cows had greater first-visit DMI, DMI/d, largest meal/d, and DMI/ meal, faster eating rates, fewer visits/d and meals/d, as well as longer inter-meal and maximum non-eating intervals (P ≤ 0.019).Additionally, MU cows tended to exhibit fewer visits/meal (P = 0.050) and longer first visit durations (P = 0.060) than PR cows; all other feeding metrics did not differ (P ≥ 0.15).For the main effect of group composition, cows in MX vs. SM groupings had more visits/meal, longer meals, and greater largest meal/d (P ≤ 0.019).Furthermore, cows in MX groupings tended to consume fewer meals/d (P = 0.060), greater DMI/meal (P = 0.062), and spend less total time eating (P = 0.054) than those in SM; all other feeding metrics did not differ (P ≥ 0.12).

Feed Efficiency
Cows in MX groupings tended to have a higher RFI (less feed efficient) than those in SM (P = 0.079), regardless of parity (P = 0.95; Table 7).Residual feed intake values did not differ across parity and group composition interactions (Figure 3; P = 0.45).Additionally, RFI was not correlated with any competition behavior variables (R s range: −0.05 to 0.18; P ≥ 0.18).However, associations with RFI and feeding patterns were present.Cows with a higher RFI consumed more feed (R = 0.39, P = 0.003) and visited the feed bunk more often (R = 0.27, P = 0.040), as well as tended to eat faster (R = 0.23, P = 0.078) and have more meals/d (R = 0.24, P = 0.064).Cows with higher RFI also had shorter maximum non-eating intervals (R = −0.40,P = 0.002) and tended to have shorter intermeal intervals (R = −0.25,P = 0.055).

Individual Cow-Level Correlations
Correlations among competition behavior at the feed bunk, first-visit feeding patterns, and feed efficiency are summarized in Table 8.Following morning feed delivery, cows with shorter latencies to first visit the feed bunk accessed the same bin for that visit more often across days (P = 0.024), were involved with more total competitive contacts (P < 0.001), and spent more time eating within the first 30 min (P < 0.001).In addition, cows with shorter first-visit latencies tended to have shorter first-visit durations (P = 0.068) and tended to have higher RFI (less feed efficient; P = 0.084).Cows involved in more total competitive contacts consumed less feed during shorter first bunk visits (P < 0.001) 6.9 (4.8, 9.9) 12.9 (7.9, 21.0)  but spent more time eating within the first 30 min after feed delivery (P = 0.004).Cows with longer first visits to the feed bunk consumed more feed at that time (P < 0.001).Finally, cows who chose the same bin more often for their first visit spent more time eating within the first 30 min after feed delivery (P = 0.009).

DISCUSSION
The purpose of this study was to evaluate the impacts of parity and group composition on competition behavior at the feed bunk, feeding behavior, and feed efficiency.We also evaluated the cow-level relationships among competition, feeding behavior after fresh feed delivery, and feed efficiency to further our understanding of factors relating to social dynamics.At a 2:1 feed bunk stocking density, we found that assigning lactating cows to feed in mixed-vs.same-parity groups resulted in greater feed bunk competition, altered feeding patterns, and tended to yield less feed efficient cows.These findings suggest same-parity groupings may have practical advantages for animal welfare and feed efficiency.

Competition Behavior
Competitive interactions are commonly measured to characterize social dynamics and hierarchy in dairy cattle.However, few studies have evaluated how parity and group composition interact to affect competition.We found greater competition in MX compared with SM groupings, regardless of parity.Similarly, in a previous study, heavy heifers (BW >250 kg) exhibited greater agonistic behaviors in a large, heterogenous-BW group than a small, homogenous one (Hindhede et al., 2010).In our study, with a 2:1 feed bunk stocking density, parity did not influence competitive behavior, similar to a previous study quantifying displacements at a feed bunk with headlocks at 80% or 100% stocking densities in lactating Jerseys (Lobeck-Luchterhand et al., 2015).In contrast, pastured primiparous cows in a same-parity group showed less aggressive behavior than those in a mixed-parity group and less than multiparous cows in a same-parity group (Phillips and Rind, 2001).A limitation of the design of our research facility is that all subgroups were housed in the same pen.We report only competitive interactions between
Within the mixed-parity groups, ours is the first study to report the interactions by parity.Previous studies evaluating agonistic behavior at the feed bunk by parity (e.g., Neave et al., 2017) did not report within-or between-parity interactions.In our study, on a numerical basis, most competitive interactions occurred between inter-parity dyads (PR-MU or MU-PR) compared with same-parity dyads (PR-PR or MU-MU).This mirrors the pattern of greater competition in MX vs. SM groups and further suggests that cows compete more with different-vs.same-parity individuals.
2 Sum of contacts initiated or received across all 4 observation periods.
3 Networks of 9-10 cows were created to evaluate parity (PR: primiparous, MU: multiparous) and group composition (SM: same-parity, MX: mixed-parity) combinations for interactions at the feed bunk; averaged between two networks of each type.

4
Degree centrality: number of edges for a node; number of other cows from which the focal individual received (r) interactions (in) or initiated interactions (out) as the actor (a).

5
Total edge weight for a node; number of interactions an individual received (r, in) or initiated (out) as the actor (a).

6
Variation in node degree centrality illustrating the involvement of individuals in the network.group composition, competition was non-transitive and bidirectional, as seen from edges directed between most individuals in a network, and further supported by the low network-level degree centralization and high reciprocity, especially within SM-PR and MX.The low level of degree centralization indicates relatively uniform distribution of involvement from most individuals in the network.Node metrics revealed that cows interacted with an average of 4.4 others out of 8 to 9 conspecifics, with high inter-individual variation as seen in the range for degree (range: 0 to 9, across all networks) and strength (range: 0 to 73, across all networks), similar to Foris et al. (2019).When assigning treatments to subgroups of cows, we balanced for similarity in DIM and BW to compare parity and group composition differences, but intra-group variation still existed that may have contributed to these findings.Other characteristics, such as social roles and personality (Krause et al., 2010), or variation in milk production (which had a relatively narrow range in our study population), may influence social behavior, and future research should investigate these factors.
In addition to reporting the counts of competitive events, previous literature has characterized social dynamics or dominance using indexes reflecting how often a cow serves as an actor vs. receiver (Mendl et al., 1992;Galindo and Broom, 2000;Gibbons et al., 2009).Ours is the first study to characterize indexes across all competitive behavior subtypes, starting from the initial attempts (competitive contacts), to displacements, to replacements.The indexes for all behavior types showed high individual variation, ranging from 0 to 1.0 (initiating 0 to 100% of interactions).On average, regardless of parity or group composition, cows initiated about half of the interactions they were involved in.Another study, which housed lactating cows in mixed-parity groups with 200% stocking density at a post-and-rail feed bunk, likewise found individual variation in displacement index values ranging from 0.1 to 1.0 (Huzzey et al., 2012), whereas pre-partum, multiparous cows housed at 150% feed bunk stocking with RIC bins showed a relatively narrower range (0.35 to 0.60; Proudfoot et al., 2009).This individual variation highlights the need for further research to evaluate characteristics, other than parity, that are associated with initiating competition at the feed bunk with consideration for comparison across stocking densities.
A weakness of the commonly used dominance, competitive, or success indexes is that they mask differences in the magnitude of the events, which underscores the need to also report the counts of events.Such indexes also fail to account for displacement attempts that are unsuccessful.To address this gap, our study is the first to include both unsuccessful displacements and replacements at the feed bunk.Many authors do not record unsuccessful competitive attempts since not all physical contacts may be performed with the intent to displace (Huzzey et al., 2012).Nonetheless, intent generally cannot be assumed during behavioral observation, and the evaluation of all types of initiated contacts, successful and unsuccessful, provides a more comprehensive account of feed bunk social dynamics.Our approach allows not only for characterizing how successful cows are at displacing or replacing others, relative to how often they initiate (successful displacement and replacement ratios), but also captures how well cows resist displacement, relative to how often others attempt to displace them (displacement resistance ratio).
The ratios calculated across all behavior types did not differ among parity and group composition combinations.On average, cows were relatively successful: 67% of initiated competitive contacts resulted in successful displacements (range: 0 to 100%), and 86% (range: 33 to 100%) of those displacements resulted in replacements (and thus, 58% of initial contacts resulted in replacements; range: 0 to 100%).In addition, cows were able to resist displacement (i.e., stand their ground) for an average of 31% of competitive contacts received (range: 0 to 100%).Interestingly, on a numerical basis, PR cows in MX were able to resist displacement more often when receiving contact from an MU vs. another PR cow, which suggests they may adapt to greater competition in MX and respond differently to certain conspecifics.All ratios showed high variation in the degree of success among individual cows, indicating a need for future research to identify individual char- acteristics associated with competition strategies and success rates.

Feeding Patterns
Our findings align with several other studies evaluating the impact of parity on feeding patterns.Multiparous vs. primiparous cows had greater daily DMI, driven by greater DMI/meal and faster eating rates, but with longer inter-meal intervals and fewer daily meals (DeVries et al., 2011;Neave et al., 2017;Crossley et al., 2018), and no differences in meal duration (Beauchemin et al., 2002;Crossley et al., 2018).Several studies did not agree with our findings of parity differences in the number of feed bunk visits or daily meals, eating rate, or size of meals; these differences could be explained by the stage of life cycle (transition cows in Neave et al., 2017), housing type, or meal definition (stanchion housing, no meal criterion analysis in Beauchemin et al., 2002).Other studies reported multiparous cows spent more, rather than similar time eating each day compared with primiparous cows (Lobeck-Luchterhand et al., 2015;Neave et al., 2017), when only evaluating a mixed-parity group composition.Finally, reporting maximum non-eating intervals and largest meal/d is relatively novel in dairy cows.We found that multiparous cows had longer maximum daily non-eating intervals than primiparous cows, which highlights an interesting parity difference that may be related to varying intake demands, which could be investigated further.Our finding that multiparous cows consumed a greater largest meal/d compared with primiparous cows aligns with numerical values previously reported (Brown et al., 2022).
Group composition impacted daily feeding patterns in alignment with previous studies.Cows in MX tended to have greater total daily eating time, similar to heavy dairy heifers in a large, heterogenous-BW group compared with a small, homogenous-BW one (Hindhede et al., 2010).We hypothesize the longer eating times in our mixed-parity groups may be explained by the greater levels of competition, including displacements, which were accompanied by more visits/meal.Additionally, as feed bunk competition increased, cows in mixedparity groups consumed less frequent, larger meals (Hosseinkhani et al., 2008;Crossley et al., 2017), which may be related to our findings of tendencies for these meal variables to be greater in MX vs. SM groups.In contrast, an older study reported primiparous cows in a same-parity group had greater feed intake and longer eating times (measured with scan sampling) compared with those in mixed-parity groups (Krohn and Konggaard, 1979), but those authors did not report findings for multiparous cows.
In addition to daily feeding patterns, we evaluated those immediately following the morning feed delivery.Latency to first visit the bunk did not differ among parity and group composition combinations, but with large variation among individual cows, ranging from approximately 5 to 120 min.In previous work on dry cows, displacement success did not impact latency to eat in mixed-parity groups, but latency likewise showed large individual variation, ranging from 0 min to 3 h (Huzzey et al., 2012).To further investigate this individual variation, we used a novel approach of calculating several feeding variables for the first bunk visit.We found MU cows consumed more dry matter in the first bunk visit than PR cows, similar to previous work showing that mature cows (≥third lactation) consumed more dry matter during the first meal than younger (≤second lactation) cows (Crossley et al., 2018).However, despite the greater competition we observed in MX vs. SM groups, all first-visit feeding variables did not differ between group compositions.
At the individual level, the magnitude of involvement in competition showed interesting relationships with first-visit feeding variables.Cows with a shorter latency to the first bunk visit were involved with more total competitive contacts, and those involved in more competition consumed less feed during a shorter first visit.Shorter latencies to the first bunk visit were also correlated with choosing the same bin more often for the visit, and the latter variable was correlated with more total time spent eating within the first 30 min after feed delivery.These findings suggest that cows are not only motivated to consume fresh feed, but may also have a preference to consume feed at a particular location.The latter topic is an area of limited research, especially related to grouping strategies, and the potential connections among preference, competition, and feeding patterns merit further evaluation.

Feed Efficiency
Feed efficiency is often measured with RFI, which is the unexplained variance in feed intake after accounting for known energy sinks, specifically metabolic BW, BW change, and secreted milk energy (VandeHaar et al., 2016).Advancing our understanding of factors that may contribute to this residual variation is important for identifying new energy sinks, such as competitive behavior, which could be included in the calculation of RFI.Previous work reported primiparous cows in same-parity groups were more feed efficient than those in mixed-parity groups, but only descriptively and using a simple calculation of efficiency instead of RFI (Bach et al., 2006).Ours is the first study to evaluate RFI between different grouping strategies.We found SM groups tended to be more feed efficient than MX groups under a 2:1 feed bunk stocking density, indicating negative implications for efficiency when lactating dairy cows feed in mixed-parity groups.This finding may be linked to the increased competition and disrupted meal patterns in MX groups compared with SM, despite similar daily DMI and visits, resulting in inefficient use of feed.
In addition, we evaluated relationships between RFI and cow-level behavioral outcomes.Evaluating these cow-level relationships can aid in our understanding of individual strategies at the feed bunk to gain access to feed in a competitive environment, as well as highlight potential variation between more vs.less feed efficient cows for further investigation.Previous work found indirect relationships between behavior and feed efficiency, with slight improvements to predictive models of feed efficiency when sensor-derived measures were included, such as activity, rumination, lying, and time spent in certain areas of the barn (Martin et al., 2021).We did not find direct relationships between competitive behavior and RFI, although lower RFI (greater feed efficiency) tended to be correlated with longer latencies to the first bunk visit after fresh feed delivery and slower eating rates.In previous studies, feed bunk competition has been shown to increase eating rate (Olofsson, 1999), and cows who ate at slower rates were also more feed efficient (Connor et al., 2013;Brown et al., 2022).Furthermore, less feed efficient cows visited the bunk more often with shorter maximum non-eating intervals, as well as tended to have more meals/d and shorter intervals between meals, which may be influenced by the aforementioned competition and disrupted feeding pattern in MX.In previous work with RIC bins, more frequent bunk visits at specific time points throughout the day also tended to be associated with less feed efficient cows at a 2:1 stocking density (Brown et al., 2022).Therefore, cows may benefit from waiting to go to the feed bunk after fresh feed delivery to reduce involvement in competition, which may reduce eating rate and increase feed efficiency.However, other cows who access the bunk first may sort the feed, resulting in a negative impact (reduced nutritional value; DeVries et al., 2005;Hosseinkhani et al., 2008) on those who wait; future studies could disentangle the efficiency implications of strategies relating to the timing of feeding.Finally, a previous study speculated that the inefficient cows may produce more metabolic heat than efficient cows (Ben Meir et al., 2018).Quantifying the energy expenditure from competition at the feed bunk could provide insight into RFI variation.Further investigation into factors that affect competition, such as grouping strategies, stocking density, and bunk management, can improve our understanding of RFI and the attributes of a feed-efficient cow.

CONCLUSIONS
Under a competitive 2:1 feed bunk stocking density, cows in mixed-parity, compared with same-parity groups, were involved in more competitive feed bunk interactions, exhibited greater total eating time, and tended to be less feed efficient.Social network analysis further illustrated heightened competition in mixedparity groups and showed that most interactions were bidirectional.Regardless of grouping, multiparous cows, compared with primiparous ones, ate more dry matter per meal and per day, had faster eating rates, and visited the bunk fewer times each day.At the individual level, cows with shorter latencies to first visit the bunk after fresh feed delivery were involved in more competition and tended to be less feed efficient.Mixedparity group housing for lactating dairy cows may have potentially negative animal welfare and efficiency implications that should be considered when selecting grouping strategies on farm.
analyzed at a commercial laboratory (AgSource, Menomonie, WI) for milk composition (fat, protein, lactose, and milk urea nitrogen) and SCC.Residual Feed Intake.Residual feed intake was calculated as a measure of feed efficiency for each cow (greater value indicates less efficient) by regressing DMI on milk energy output, median DIM, metabolic BW, and change in BW, each nested within parity.All values were summarized as an average across the experimental period for each cow.Milk energy output (kg/d) was calculated as [9.29 × milk fat (kg)] + [5.63 × true protein (kg)] + [3.95 × lactose (kg)] (NRC, 2021).Bodyweight was recorded before morning feed delivery 3 d/wk during wk 1, 4, and 7 of the experimental period using a calibrated stationary scale (EW6, Tru-Test Limited).Metabolic BW (kg) was calculated as BW 0.75 .The daily change in BW was calculated using the LINEST function in Microsoft Excel to create a simple linear regression of all 9 BW values.Body condition score (reported descriptively in Table Figure 1.Flowchart of the behavior sequence used for behavior observation of competitive interactions at the feed bunk between midlactation Holstein cows.For each behavior, an actor (cow initiating the event) and receiver (individual receiving the event) were recorded.
Reyes et al.: SOCIAL DYNAMICS AND FEED EFFICIENCYTable 4. Competition behavior 1 recorded in the first hour after morning feeding with a 2:1 stocking density at the feed bunk, reported by parity and group composition combinations 2 for mid-lactation Holstein cows, averaged across 4
Reyes et al.: SOCIAL DYNAMICS AND FEED EFFICIENCYTable5.Descriptive statistics 1 of competition behavior recorded in the first hour after morning feeding within the mixed-parity group composition with a 2:1 stocking density at the feed bunk, reported by parities (primiparous or multiparous) of mid-lactation Holstein cows involved in the interactions, averaged across 4 observation days Variable Parities of cows involved as actor and receiver 1

Figure 2 .
Figure 2. Social networks of mid-lactation Holstein cows for 3 combinations of parity (PR: primiparous, MU: multiparous) and group composition (SM: same-parity, MX: mixed-parity): A) SM-PR, B) SM-MU, C) MX (9-10 cows per network, 2 networks per combination) were constructed using igraph package in R. Each network shows the total frequency of competitive contacts at the feed bunk during the first hour after AM feed delivery, with arrows indicating the direction of the interaction and thickness representing the frequency.Red vs. blue circles represent PR vs. MU cows, respectively; in the MX networks, those colors represent the parity of the cow initiating the interaction.For MX networks, lighter colored arrows indicate interactions between cows of the same parity, while darker colored arrows indicate interactions between cows of different parities; color is based on arrow origin (parity of individual who initiated interaction).
Reyes et al.: SOCIAL DYNAMICS AND FEED EFFICIENCYTable6.Metrics 1 for social networks constructed using the number of competitive contacts 2 between mid-lactation Holstein cows with a 2:1 stocking density at the feed bunk, reported by parity and group composition to a 2 × 2 factorial design by parity (PR: primiparous or MU: multiparous) and group composition (SM: same-parity or MX: mixed-parity); SM-PR = 19 cows, SM-MU = 20 cows, MX-PR = 10 cows, MX-MU = 10 cows. 2 Back-transformed from log n values with 95% confidence interval in parentheses.

Figure 3 .
Figure3.Observed vs. predicted DMI plotted for each parity (PR: primiparous, MU: multiparous) and group composition (SM: same-parity, MX: mixed-parity) combination for mid-lactation Holstein cows.Data points above the line of unity (dashed line) represent cows consuming more feed than predicted, associated with a positive residual feed intake (RFI) value and lesser feed efficiency.Data points below the line of unity represent cows consuming less feed than predicted, associated with a negative RFI and greater feed efficiency.
Reyes et al.: SOCIAL DYNAMICS AND FEED EFFICIENCYTable8.Correlation matrix for cow-level relationships between competition and feeding behavior 1 and feed efficiency for mid-lactation Holstein first hour following AM feed delivery and feeding patterns following that feed delivery.2 Sum of competitive contacts at the feed bunk initiated and received by each cow. 3 Highest proportion of first visits to one of the 5 assigned bins.4 Sum of the time spent eating during the first 30 min after AM feed delivery.5 Estimation of feed efficiency, calculated for each cow by regressing DMI on milk energy output, median DIM, metabolic BW, and change in BW, each nested within parity.6 Pearson correlation based on normally distributed data for both variables compared; otherwise, Spearman rank correlation used for non-normally distributed data.*P < 0.05; **P ≤ 0.01; † 0.05 ≤ P ≤ 1.0.
Reyes et al.: SOCIAL DYNAMICS AND FEED EFFICIENCY

Table 1 .
Reyes et al.: SOCIAL DYNAMICS AND FEED EFFICIENCY Descriptive statistics 1 for mid-lactation Holstein cows by parity and group composition combinations 2

Table 3 .
Reyes et al.: SOCIAL DYNAMICS AND FEED EFFICIENCY Ethogram used for observing feed bunk interactions and proportions calculated from counts of competition behavior at the feed bunk during the first hour after morning

Table 7 .
Feeding patterns and feed efficiency reported by parity and group composition combinations 1 for mid-lactation Holstein cows (starting DIM: 144.5 ± 21.8 d; MY: 42.3 ±