Genetic parameters for calf feeding traits derived from automated milk feeding machines and number of bovine respiratory disease treatments in North American Holstein calves

Precision livestock farming technologies, such as automatic milk feeding machines, have increased the availability of on-farm data collected from dairy operations. We analyzed feeding records from automatic milk feeding machines to evaluate the genetic background of milk feeding traits and bovine respiratory disease (BRD) in North American Holstein calves. Data from 10,076 preweaning female Holstein calves were collected daily over a period of 6 yr (3 yr included per-visit data), and daily milk consumption (DMC), per-visit milk consumption (PVMC), daily sum of drinking duration (DSDD), drinking duration per-visit, daily number of rewarded visits (DNRV), and total number of visits per day were recorded over a 60-d preweaning period. Additional traits were derived from these variables, including total consumption and duration variance (TCV and TDV), feeding interval, drinking speed (DS), and preweaning stayability. A single BRD-related trait was evaluated, which was the number of times a calf was treated for BRD (NTT). The NTT was determined by counting the number of BRD incidences before 60 d of age. All traits were analyzed using single-step genomic BLUP mixed-model equations and fitting either repeatability or random regression models in the BLUPF90+ suite of programs. A total of 10,076 calves with phenotypic records and genotypic information for 57,019 SNP after the quality control were included in the analyses. Feeding traits had low heritability estimates based on repeatability models (0.006 ± 0.0009 to 0.08 ± 0.004). However, total variance traits using an animal model had greater heritabilities of 0.21 ± 0.023 and 0.23 ± 0.024, for TCV and TDV, respectively. The heritability estimates increased with the repeatability model


INTRODUCTION
Precision livestock farming has increased the availability of on-farm data collected from livestock operations (Baker et al., 2022;Neethirajan and Kemp, 2021;Groher et al., 2020).Information from various sensors can be captured, stored in large databases, and analyzed to determine individual differences among animals for breeding purposes, as well as for management decisions (Brito et al., 2020;Kleen and Guatteo, 2023;Montes et al., 2023).Multiple disciplines in the field of animal sciences have begun using high-throughput phenotyping platforms to investigate health and production efficiency in livestock, presenting large gains in the dairy cattle industry (Berckmans, 2017).Most of this data has been used for real-time adaptation of management, nutrition, and breeding to increase the productive efficiency and sustainability of the dairy industry.Some examples of precision livestock farming technology in dairy cattle are automatic milking robots (e.g., Pedrosa et al., 2023) that capture information on productivity per milking, as well as sensors that capture movement, laying time, and rumination (Deming et al., 2013;John et al., 2016;Pavlovic et al., 2021;Solano et al., 2022).All these data sources provide longitudinal phenotypic information that can be used to accelerate genetic and genomic selection for novel breeding goals (Brito et al., 2020).
Longitudinal calf data from automated milk feeders (AMF) have been previously used to determine patterns in consumption, drinking speed, and number of visits, and then identify possible illness.For instance, Borderas et al. (2009) reported that calves considered ill (respiratory or enteric disease) drank 2.6 L/d less and visited the feeder 2.4 fewer times in a day than their healthy cohorts.However, in the same study, it was shown that sick calves would consume a lower amount of milk (4 L/d) but maintained milk intake and had shorter visit durations during their observed illness.In addition, Svensson and Jensen (2007) found that calves that were ill had a decrease in unrewarded visits by 25%, and Knauer et al. (2017) observed that ill calves drank less milk (2.6 L/d) and at a slower rate (183 mL/ min), with 3.1 fewer unrewarded visits compared with healthy calves.Similarly, Duthie et al. (2021) reported that calves in automated feeding systems that were positive for bovine respiratory disease (BRD) had fed for lower durations, drank less milk overall, and visited the automated feeder less frequently.Therefore, the amount of milk that calves are fed could potentially be used to detect illness.
Dairy calves raised in intensively managed systems are typically removed from their mother shortly after birth and placed in either individual or group housing (Placzek et al., 2021).Within group-housed systems, dairy calves were reported to be observed until weaning (roughly 56.8 d ± 9 d) where they are then placed into group pens until calving (Jorgensen et al., 2017).Preference for group housing of dairy calves is rising and may become the preferred way of rearing dairy calves in the near future.A survey with 463 youths and 1,310 adults in Minnesota found that most people preferred group housing, citing socialization and space allowances as the driving forces for the preference (Perttu et al., 2020).Group housing has inherent challenges that must be overcome, such as allowing for diseases to spread more easily among calves (Curtis et al., 2016).
Changes in the environment can also lead to an increase in the prevalence of diseases, especially when the environment becomes colder, and the calf barn ventilation is reduced (Callan and Garry, 2002).During their growth stage, dairy cattle are most susceptible to major environmental and disease perturbations, such as BRD and enteric diseases (Callan and Garry, 2002).Bovine respiratory disease is a complex disease caused by a multitude of pathogens, and it has been reported to cause a preweaning death loss of roughly 5% to 8% in dairy cattle populations (USDA, 2014).Thus, in addition to welfare concerns, BRD has a large economic impact on the cattle industry, costing roughly $42.15 to treat each case and $444.32 per mortality incidence due to BRD in 2020 (Dubrovsky et al., 2020).
Recently, studies have been performed to estimate genetic parameters for BRD and enteric disease in calves.Respiratory illness was specifically reported by multiple authors to have a heritability ranging from 0.04 to 0.10 (Henderson et al., 2011;Gonzalez-Peña et al., 2019;Haagen et al., 2021;Zhang et al., 2022), indicating that genetic selection for calf diseases, such as BRD, is possible.Furthermore, statistically significant phenotypic associations between BRD and calf feeding behavior have been reported (Cantor and Costa, 2022), but little has been done to quantify the genetic components that explain the underlying genetic variation of feeding behavior traits recorded from AMF.It has been reported in multiple studies that fluctuations in daily milk yield for cows in automatic milking systems were heritable and could potentially be used as indicators of resilience (Elgersma et al., 2018;Wang et al., 2022;Chen et al., 2023;Pedrosa et al., 2023).These milk consumption and drinking behavioral traits could be used to elucidate the resilience to disease and other environmental influences, as well as differentiate calves based on resilience measures.Furthermore, defining the relevant traits that may be used for early selection is useful to estimate resilience using longitudinal techniques as done in other species (Putz et al., 2019;Nguyen-Ba et al., 2020).In this context, the main objectives of this study were (1) to derive novel traits related to calf feeding behavior and health based on data from AMF and BRD susceptibility, and (2) to define the statistical models and estimate variance components and genetic parameters (heritability and genetic correlations) for BRD indicators, milk intake traits, and feeding behavior in North American Holstein calves.

Animal Ethics
Animal welfare and ethics committee approval was not needed for this study because all the datasets used were obtained from a commercial dairy farm that follows standard production practices.

Phenotypic Data Summary
The data were collected between 2015 and 2021 on a large commercial dairy cattle farm in North Central Indiana.Purebred Holstein calves were reared since birth in group housing barns with 223-m 2 pens, split into 2 separate 111.5-m 2 sections containing roughly 60 animals at a time.Crossbred terminal calves and male calves were removed from the dataset because they were not retained in the herd or used for breeding purposes, leaving a total of 14,749 calves in the dataset, before data editing.The calves were reared in 4 separate barns, with 4 pens in each barn, and each Förster-Technik AMF feeding 2 pens (Förster-Technik, Engen, Germany; www .foerster-technik .com/), with 2 stalls in each pen section available for feeding.Calves were removed from their dam at d 0 and fed colostrum before being relocated to their rearing pen for 58.43 ± 8.88 d.All calves were vaccinated with an intranasal bovine rhinotracheitis-parainfluenza 3-respiratory syncytial virus vaccine (INFORCE 3, Zoetis, Parsippany, NJ) before entering the feeding program.Information regarding starter grain or DMI was not collected at the farm.Calves that were removed and treated in hospital pens were placed back into the feeding program based on their age.The treatment protocol for suspected BRD was based on gross clinical signs, which were a drop in drinking speed based on the previous day of 80% reported by the AMF, an elevated body temperature (≥39.5°C), and physical depression or rapid breathing.If calves met 2 of these 3 conditions, they were treated with subcutaneous tulathromycin (Draxxin, Zoetis, Parsippany, NJ).
The AMF devices were programmed to include a milk entitlement for each calf, which sets a threshold for the amount of milk that each calf could drink per visit and per day.The consumption entitlement changed each day based on the date the calf first entered the system, as well as carryover based on the previous day.The feeding system employed a stepdown procedure, where milk entitlement was reduced from roughly d 32 to d 60, which is when weaning occurred.This type of step-down program has been found to improve rumen development in preweaning dairy calves (Khan et al., 2007).From feeding d 1 to 10, calves were allowed to drink a maximum of 2 L/2 h; from d 11 to 21, entitlement increased to 2.5 L/2 h; and from d 22 to 31, entitlement maximized to 3 L/2 h with a maximum of 24 L/d.After d 32 to 39, the entitlement plummeted to 10 L/d with a max of 2 L/visit, beginning the weaning process.From d 39 to 60, maximum entitlement again reduced to 8 L/d with 2 L/visit.The entitlement thresholds placed on the calves created heterogeneity in the variance for milk consumption (Supplemental Figure S1; https: / / doi .org/ 10 .6084/m9 .figshare.25099910.v1;Graham, 2024), where calves were prevented from drinking more than their current entitlement.Therefore, to account for heterogeneous variances within these time frames, different approaches to model changes in variance were considered in this study as described in the Statistical Analyses section.

Weather Information
The weather information was gathered from the nearest (7 km) public weather station located in Plymouth, Indiana (station code: C65).The temperature-humidity index (THI) was calculated as follows (NRC, 1971): where T db is the dry-bulb temperature and RH is the relative humidity.Season parameters were created based on month and day as follows: winter was defined as dates between December 21 and March 20, spring was defined from March 21 and June 20, summer from June 21 to September 20, and fall was from September 21 to December 20.

Deriving Calf Traits
Data were collected from multiple information sources throughout the farm, which were pooled into the Precision Animal Science Data Ecosystem at Purdue University (West Lafayette, IN) for subsequent analyses.The datasets for this study combined Förster-Technik calf feeding information, weather station, health records, and identification information from the Dairy Comp software (Valley Agricultural Software, Tulare, CA).Two separate datasets were generated based on datatype, either per-visit (dataset 1; n = 17,073 calves) or daily records (dataset 2; n = 25,153 calves) information.The dataset 1 file contained individual recordings of calf feeding behavior at each visit, and the dataset 2 file contained daily calf feeding records.The dataset 2 information was available from 2015 to 2021, whereas the dataset 1 information was only available starting in the summer of 2019 until 2021.Each dataset was used to generate novel phenotypes for evaluating calfhealth, feeding behavior, and performance.For dataset 1, the main phenotypes recorded were per-visit milk consumption (PVMC) in liters and drinking duration per visit (DDPV) in minutes.For the PVMC trait, all zero values were removed to better account for variability in true recordings.The variance of total milk consumption (TCV) in liters squared and variance of total drinking duration (TDV) per visit in minutes squared were calculated.We also derived feeding interval (FI) by averaging the amount of time between rewarded visits in the AMF, starting the feeding day at midnight.We used rewarded visits to remove any bias from misreported visits.The main traits evaluated were daily milk consumption (DMC) in liters; daily sum of drinking duration (DSDD) in minutes; daily average of drinking speed (DS) in liters/minute, where DS = DMC/DSDD; daily number of rewarded visits (DNRV); and total number of visits per day (TNV).We also calculated the preweaning stayability trait (ST), which considers the age in which the calf was removed from the group pen for any reason.Animals that showed signs of respiratory illness were treated and this treatment was recorded as an incidence of BRD.The BRD-related trait evaluated was the number of times a calf was treated for BRD (NTT).The NTT was determined by counting the number of recorded BRD incidences starting at the first day the calf entered the system and ending at 60 d of age.Therefore, NTT is dependent on BRD recording by the farm, in which the former is a repeated record trait for each event and the latter is a single measure trait for the overall number of times a calf was treated.

Data-Editing Procedures
To edit dataset 2, individuals with missing birthweights and those that were above 4 standard deviations (SD) for birth weight were removed (9 individuals), as well as negative consumption and duration observations (12,865 records), observations that occurred twice on the same day (116 records), observations with zero rewarded visits (1,479 records with no milk consumption), observations with more than 6 h of total daily drinking duration, which would allow for 15 min of consumption per hour to occur (1,149 records), observations with more than 144 visits or once every 10 minutes in a 24-h period (6 records), and outliers were removed based on a z score = , where x i is the observed phenotype of calf i, x is the phenotypic mean, and σ x is the SD of the trait for daily consumption, daily drinking speed, daily duration, and daily rewarded visits (12,218 records).Values above 50 unrewarded visits were recoded as 50.Individuals with less than 5 observation days and contemporary groups with less than 5 individuals were removed (1 individual).Observations with entitlements less than 0 and more than 24 were filtered out, with entitlements >24 being recoded as 24, and dams with missing parity information were fixed by taking the maximum parity and adding one, or by adding one if no previous parity existed (220 obser-vations).Parities were labeled as 1, 2, and those that were ≥3 were recoded as 3+.The final file for dataset 2 comprised 566,156 observations from 10,076 individuals.The descriptive statistics for each trait are presented in Table 1.
To clean dataset 1, individuals that did not have unique identification (426 individuals), observations above 5 L, per-visit duration above 25 min, calves with less than 5 observation days, and contemporary groups with less than 5 individuals were removed (7,939 individuals, 2,466,717 observations).Additionally, the FI for each day was calculated, and individuals with negative values were removed (6 individuals, as these animals were assumed not to be correctly identified).The DMC and DSDD records that were outside 4 SD from the mean were also removed (5 individuals, 36,727 observations).All traits were filtered to remove outliers were based on a z-score as described previously.Due to the heterogeneity of the residuals across time, we could not assume a normal distribution across time.Therefore, we applied the z-score outlier removal process described above, using only within-day variation.This process to remove outliers is similar to the approach followed by Pietersma et al. (2006) and Hurst et al. (2021).The final dataset 1 consisted of 1,197,425 observations from 4,572 individuals.The descriptive statistics for each trait are presented in Table 1.An initial linear model was fitted to the DMC trait.The residuals from a model (described later) fitted to the data were plotted against the predicted values (Supplemental Figure S1).Heteroscedasticity was observed, which is indicative of heterogeneity of variance, or potential noise from drinking entitlement.The decrease in variance could be caused by the reduction of milk feed entitlement in a stepwise fashion from around d 32, which prevents calves from drinking ad libitum, thus reducing the milk consumption of calves that would otherwise be drinking.Consequently, we investigated the feasibility of the first 32 d being a distinct trait.After filtering, dataset 1 contained 699,832 observations from 4,428 calves, and dataset 2 had 311,733 observations from 10,076 calves, all occurring within the first 32 d.

Pedigree and Genomic Information
Pedigree information was reported directly from the farm using the VAS Dairy Comp software (2022, VAS, Tulare, CA).A total of 23,395 animals were included in the pedigree, which included 11,975 calves and their ancestors.Out of 11,975 phenotyped calves, 6,542 had both dam and sire reported, 5,229 had one or the other, and 204 calves had no parental information.Pedigrees were trimmed using the "trimPed" function implement-ed in the "pedigree" package (pedigree v. 1.4.2;Coster, 2022) in the R software (R Core Team, 2022) based on the relationship with individuals with phenotypes.Pedigree branches that did not end in a calf with data were trimmed out.After trimming, 18,256 animals remained in the pedigree dataset.
There were 10,703 calves with genomic information before the quality control.Genetic material was collected by the producers, and genotype information consisted of 4 commercial SNP arrays containing 11,405,17,557,18,819,or 30,754 SNPs (Illumina,San Diego,CA), which were imputed to a common set of 72,820 SNPs using the FImpute software (Sargolzaei et al., 2014), as reported by Chen et al. (2023) and Pedrosa et al. (2023).Quality control of the genomic information was done by removing the following: (1) duplicated SNP (markers at the same genome location); (2) duplicated individuals; (3) SNP located on non-autosomal chromosomes; (4) SNPs with an extreme departure from expected heterozygosity levels (>0.15); (5) SNP with call rate <0.90; (6) SNP with minor allele frequency <0.05; and (7) animals with call rate <0.90.A total of 57,019 SNP and 10,076 animals with genomic and phenotypic information remained for further analyses.

Systematic Effects
Both datasets shared similar explanatory variables.The systematic effect of time of day of feeding was split into 4 factor levels (early morning, late morning, early evening, and late evening), based on a 24-h clock starting at 0000 h and ending at 2400 h, which were split evenly on a 6-h basis.The effect variable of feeding day of calf started at d 1 and ending at d 60.The effect of feed entitlement is described in detail in the Phenotypic Data Summary section.PVMC and DMC followed a non-normal distribution due to calves having a change in entitlement every 10 d, increasing from 2 to 3 L every 2 h, then at age 32 d dropping to 2 L/d at weaning.Therefore, the reported feed entitlement was included as a main effect for all traits previously described.The systematic effect of birth weight was reported in kilograms.The effect of BRD incidence was determined by counting the number of treatments given during the recording period.The effect of pen density was calculated based on the number of calves per pen on the specific date of the observation using the precleaned dataset.Because dataset 2 only had information about the feeder that serviced 2 pens, the 2-pen average was taken.Although most animals stayed with their feeder throughout the entire period, others were moved to another pen with new pen mates, and this was accounted for in the analyses.The effect of dam parity was coded as 1, 2, or 3+ (due to a limited number of dams above  third parity).The effect of average THI was reported for the observation day.The variance traits and daily datatype contained the same systematic effects, except the time of day of observation.Finally, contemporary groups were created by concatenating birth-year, season, and barn.Model Selection.Fixed effects were chosen based on their relationship with the trait of interest and were selected based on the finding in Montes et al. (2023), which studied the same cohort of calves.Traits that were on a per-visit basis were investigated using different effects than on a daily basis.Model selection was performed via a backward stepwise regression, where each explanatory variable was removed based on significance, and the potential linear model was tested for each trait of interest.The testing of the general linear models included only fixed effects, and it was performed using the "lm" function from the "stats" package in R (stats version 3.6.2;R Core Team, 2022).After defining the fixed effects to be used in subsequent analyses based on ANOVA, the mixed models for each trait (with the same fixed effects) were compared based on the Bayesian information criterion (BIC), and the model with the lowest BIC were considered as the bestfit model (Supplemental Table S1; https: / / doi .org/ 10 .6084/m9 .figshare.25099910.v1;Graham, 2024).
Statistical Analyses.Multiple analyses were performed with the goal of understanding the genetic background of the traits of interest.All the analyses were performed using the Blupf90+ suite of packages (Aguilar et al., 2018).First, the datafiles were prepared with the renumf90 package.A REML analysis was performed using the Blupf90+ package (Aguilar et al., 2018) with a convergence threshold of 1 × 10 −12 .The models were evaluated using the expectation-maximization algorithm REML for the first 200 iterations and then reverted to the average information REML if convergence was not attained (Aguilar et al., 2018).The single-step genomic BLUP method was used to estimate breeding values based on an inverted H matrix as previously defined by Christensen and Lund (2010) and Legarra et al. (2009).The linear animal models had a general form: where y is the vector of the phenotypic records for each trait; b is the vector of fixed effects, including time of day of feeding only for dataset 1, number of visits only for dataset 2, feeding day, birth weight in kilograms, dam parity, treatment incidence, pen density, THI, feed entitlement, and contemporary groups as shown in Table 1; u is the vector of direct additive genetic effects where u H ~, ; ) p is a vector of animal permanent environment effects where p I ~, ; N p 0 σ 2 ( ) X is the inci- dence matrix for the fixed effects; Z is the incidence matrix for the direct additive genetic effect, W is the incidence matrix for the permanent environment effect, and ε is the vector of residuals Genetic correlations between all traits were estimated through a bivariate model fitting the same systematic effects described for the univariate models, but excluding effects that would adjust for the trait that the correlation was being made with.Genetic correlations were calculated as where y 1 represents the first trait, y 2 represents the second trait, σuy y 1 2 is the additive genetic covariance between both traits, and σu 2 is the additive genetic variance.
The RRM were also evaluated.The RRM fitted firstand second-order Legendre orthogonal polynomials in both the fixed and random regressions and second order for all effects including heterogeneous residuals.The general matrix model for the RRM was of form where b i is the fixed effect with subclass i, β includes the fixed regression coefficients, defined as , where n i is the number of animals in each feeding-day group, ε i is the residual variance estimated for each feeding-day residual group, and n t is the total number of animals across the groups (Khanal et al., 2019).
The theoretical (individual) accuracy was calculated to determine the accuracy of the predicted breeding values following the formula: σ where i = 1, 2, … is the number of individuals, se is the standard error of prediction, F is the genomic inbreeding coefficient, and σ u 2 is the additive genetic variance (Mrode, 2014).The genomic inbreeding coefficient from the combined additive genetic relationship matrix (F H ) was calculated by subtracting 1 from the diagonal of the H matrix.The mean (± SD) F H was 0.02 ± 0.025, minimum of 0, and a maximum of 0.35.Reliabilities of the estimated breeding values were then calculated as reliability i = r 2 and the mean reliability as r 2 (Mrode, 2014).

Phenotypic Data
In this study, we derived 11 calf feeding and BRD treatment traits and then estimated their genetic parameters to assess the feasibility of using these traits as indicators for animal health.The summary statistics for all phenotypic variables evaluated are provided in Table 2. To our knowledge, this is the first study presenting genomic-based genetic parameters for calf feeding behavior traits derived from AMF data recorded in preweaning Holstein calves.In dairy cattle, more focus has been given to understanding traits relevant to lactation performance and health in robotic milk-ing systems (Wethal and Heringstad, 2019;Dechow et al., 2020;Piwczyński et al., 2021;Chen et al., 2023;Pedrosa et al., 2023).However, early in life variables may be used to predict future performance and health in a dairy animal once they begin lactation.Making efforts to improve early-in-life measured traits is then paramount for sustainable dairy production.Therefore, understanding the genetic background and architecture of health and feeding traits at an early stage could result in more effective selection for young stock, especially for novel traits involved with later performance in robotic dairy farms.In addition, for predicting future performance, early-in-life measurements can help make decisions related to producing or buying replacement heifers.Rearing heifers is a major cost to producers, where a review study by Overton and Dhuyvetter (2020) found the cost of rearing a replacement heifer to range from $1,700 to $2,400, whereas the average market value for prelactation heifers is $1,300.This means that the cost of rearing is greater than the price one would get for sale.Therefore, decisions to keep specific heifers must be made with care to prevent incurring unnecessary costs and increase the profitability of dairy farms.

Milk Consumption
The mean per-visit intake across time for a dairy calf was 1.88 L/visit ± 0.93 with a range of 0.001 to 4.97 L/ visit, which is greater than the maximum entitlement provided of 3.00 L/visit.This may be due to the artifact of carryover, where a portion of the milk not consumed on the previous day can be allocated to the next day's feed entitlement.Further, the mean daily intake was 8.89 L/d ± 3.30, while the range of intake was 0.001 to 23.90 L/d, where the maximum is much greater than the daily milk intake found in other recent studies, with a reported maximum of 14.  2023).Despite the differences among studies, the values retained in this study are within a reasonable biological limit, especially considering that there could be carry-over milk consumption.

Variance and FI
The consumption and duration variance traits represent the total variation across all observation days for each calf.This may be useful to compare calves that have significant variation across days and may relate to resilience as found with milk production traits in lactating cows (Poppe et al., 2020;Chen et al., 2023).The mean TCV was 0.79 L 2 ± 0.20 SD and the TDV trait had a mean of 1.70 min 2 ± 0.20 SD.It should be noted that this is the raw variance, i.e., not corrected for any type of consumption curve.Calves that have irregular drinking patterns will have higher milk consumption variance.It has been suggested that sick calves will have lower milk intake per visit, slower drinking speed, and less frequent visits (Knauer et al., 2017;Conboy et al., 2021;Montes et al., 2023).Comparatively, the same could be said for animals with high across-day duration variance.We hypothesize that calves that visit less frequently and drink for shorter periods are more likely to experience detrimental environmental stress, and therefore, are less resilient.We also found a mean (± SD) FI of 275.6 ± 121.84 min, with a range of 1 to 1,409 min, which means calves, on average, feed every 5 h across all feeding days.This finding is almost double of what Perrtu et al. (2023) found in their calf feeding behavior of 142.4 min, however, the SD is similar of 125.7 min.The explanation for this difference can be that in our study, only rewarded visits were considered for calculating FI, whereas Perrtu et al. ( 2023) included both rewarded and unrewarded visits (i.e., the calf could return more frequently and reduce the FI).We excluded unrewarded visits because we could observe a bias due to calves repeatedly entering the feeding system in a short amount of time looking for a reward, as well as to reduce the number of false positive visits.

Drinking Speed and Feeder Visits
The drinking speed had a mean (± SD) of 0.6 L/min ± 0.20 with a range of 0.003 to 1.67 L/min.Studies have reported that calves maintaining higher levels of drinking speed are less susceptible to BRD (Knauer et al., 2017;Morrison et al., 2022).We found that the drinking speed increased linearly across the age of the animal on average, which was reported by Montes et al. (2023) using the same dataset, showing that as the calf grows, it can consume more milk in a shorter period of time.The mean TNV was 9.55 ± 6.50 daily visits, with an observed range of 1 to 81 visits, whereas the mean DNRV was 6.15 ± 3.67 visits, with a range of 1 to 31 visits, which is in line with other studies that found fewer rewarded visits than TNV.Knauer et al. (2017) reported that the number of unrewarded visits increased as the calves began the weaning period, which may explain a large amount of the total number of visits.We found that the maximum TNV was >50, which is much greater than anticipated.This may be explained by calves continuously entering the feeder in a pacing manner, being pushed out and back into the feeder, or possible errors with the radio-frequency identification reader.Another reason for the large distribution is due to calves being given starter feed, which was not recorded.Some calves may begin consuming dry feed at an earlier age, lowering their number of visits compared with their peers.However, this information is not available in the datasets used.The number of rewarded visits was lower than the total number of visits.We found that DNRV had a maximum value of 30, most likely due to some calves only drinking a small amount and then resting, or due to errors with the recording system.It is difficult to discern the true causes, and therefore the authors were less conservative on which information to remove to only keep what is realistically feasible.More studies using AMF-derived data in calves will help in understanding the limits to define cleaning rules for this type of data.

Number of BRD Treatments
When heifers develop diseases early in life, such as BRD, they require additional inseminations to become pregnant compared with those that were not diagnosed with BRD before 6 mo of age (Abuelo et al., 2021).The lack of disease early in life, also increases the odds of the animal staying in the herd over 24 mo by 60.47% (Abuelo et al., 2021;Closs and Dechow, 2017).In this sense, genomic selection of more robust calves can lead to healthier cows with greater longevity and reproductive performance.However, early detection and treatment of BRD and other diseases is still difficult.Early detection of the disease before observable symptoms is imperative to prevent severe illness that will affect animals long term.The use of AMF-derived data may also assist in the early and subclinical detection of sick calves and, consequently, to derive disease resilience indicators in dairy calves.We assessed BRD treatment stemming from self-reporting by the farm employees.This may be a limitation due to the farm protocols that are used to begin treatment for BRD, which are based on 3 qualifications: an 80% decrease in DS based on the previous day reported by the AMF, an elevated body temperature (≥39.5°C), and physical depression or rapid breathing.It is presumed that the farm workers used obvious signs of illness such as lethargy, elevated respiration rate, or discharge to investigate and begin treatment instead of considering only DS data.This may lead to underreporting of illness.However, a study performed in midwestern US farms found agreement between illness reported from AMF and by trained technicians (Perttu et al., 2023), suggesting that diseases such as BRD traits could be accurately reported by AMF systems.
Interestingly, BRD treatment had an incidence of 55%, where over half the calves were sick at least once in the observation period.This incidence proportion is much higher than the prevalence reported by Dubrovsky et al. (2020) from 6 California preweaning dairy calf herds, which had an average incidence of 22.8% and a maximum incidence in one herd of 27%.It must be noted that the calves in Dubrovsky et al. (2020) were not group housed but were raised in individual hutches.A similar study evaluated in the US Midwest preweaning dairy calves on farms that used group housing, and AMF found a BRD incidence rate of 37% (Perrtu et al., 2023).Therefore, group housing may contribute signifi-cantly to BRD prevalence as the disease may spread more easily among calves kept in the same barn and shared pen space.Another contributor to the disease spread is the pen stocking density.Svensson and Liberg (2006) found that the optimal pen density for group housed dairy calves are no more than 10 animals.In the study by Perrtu et al. (2023), pen density was no more than 15, whereas the average for the Midwest United States was 17.8, as reported by Jorgensen et al. (2017).Our dataset had a pen density mean of 48.57± 6.58 animals, and the maximum pen density was 60 animals (Table 2).This may have contributed to the greater disease prevalence in the population.We also noted a difference in the incidence of BRD by year, where calves reared in 2015 had a lower incidence of BRD of 20% compared with 2019, which had an incidence of 81%.Another contributing factor may be the season, where winter generally has the highest incidence of BRD.However, we found that the herd in 2019 had an exceptionally high incidence rate during both winter and summer months of 83% and 86%, respectively (Supplemental Figure S2; https: / / doi .org/ 10 .6084/m9 .figshare.25099910.v1;Graham, 2024).Therefore, a noticeable disease outbreak, together with high stocking density, may have been the causes of increased BRD incidence in the herd evaluated.
We observed a mean of 1.71 ± 0.75 for the NTT trait, which indicates that many calves received more than one treatment during the observed period.It must be stated that these are the number of treatments for BRD, not positive cases, and therefore the number of treatments may be inflated due to the possibility of the technicians treating more animals out of caution.

Variance Components and Genetic Parameters
Variance component estimates are presented in Tables 3, 4, 5, 6, 7, 8, and 9. To compare models for each trait, we assessed the BIC value from each model provided in Supplemental Table S1 and ANOVA values.It must be noted that it is not practical to compare Akaike information criterion (AIC) or BIC values between the full and truncated datasets.This is due to function BIC = −2logL + p*ln(n), where p represents the number of parameters and −2logL represents the log-likelihood function that is dependent on sample size and n is the number of observations.Therefore, we compared the BIC values between models in each dataset group.

Milk Consumption Traits
The traits most influenced by the truncation of the dataset were the milk consumption traits, where the estimate of heritability of PVMC increased from 0.025 ± 0.002 to 0.040 ± 0.003 and from 0.069 ± 0.005 to 0.090 ± 0.009 for DMC, when only the data from the first 32 d using a repeatability model were considered (Figure 1).The reason for the marked heritability difference is most likely because of a change in the correlation of the phenotypic variance over time due to the step-wise weaning methods mentioned previously.This causes the data distribution to fit a more parabolic shape that does not follow a linear pattern.When comparing the reliability of these 2 traits, we observe little change in how well the models perform when contrasting the changes in reliabilities for the repeatability models from 0.24 to 0.25 for PVMC and from 0.31 to 0.34 for DMC, respectively.These poor reliabilities and the differences in heritability presented indicate that repeatability models may not model consumption traits well, and this is supported by the higher AIC values.Therefore, more robust models are required.
Random regression models are more robust when estimating genetic parameters using longitudinal data and have become the norm when estimating lactation and growth traits (Oliveira et al., 2019).All traits NTT is a daily trait, whereas the other health traits are on a per-calf basis.Rel.= reliability calculated as r se , where i = 1, 2, …, n, where n is the number of individuals for each trait, se is the standard error of prediction, F is the inbreeding of animal i, and σ u 2 is the additive genetic variance (Mrode, 2014).were analyzed using 3 types of Legendre orthogonal polynomial RRM: quadratic, linear, and quadratic with heterogeneous residuals.We fitted both the full dataset and the 32-d truncated dataset.For the PVMC trait, we observed that the quadratic polynomial with 4 heterogeneous classes for the residuals fit the data best, with a mean heritability of 0.07 (range: 0.02-0.11)and lowest AIC value.We also observed a wider range in the additive genetic variance when assuming heterogeneous residuals (range: 0.009-0.07)compared with considering homogeneous residuals (range: 0.013-0.05).This indicates that the RRM with quadratic polynomials may capture more of the genetic variation when heterogeneous residuals are included, possibly due to better biological fit.Care must be taken in interpreting the results for this methodology due to the difficulty of estimating boundary points using random regression models and could potentially be remedied in future studies by using B-spline models (Meyer, 2005).Comparing the heritability of the PVMC to DMC, there was a marked difference found, where DMC had much higher estimates.We observed a heritability of 0.46 (range: 0.05-0.68)and a reliability of 0.42 using a quadratic polynomial RRM with 4 heterogeneous residual classes (Figure 2).This result is most likely due to daily consumption being influenced less by environmental noise compared with PVMC.Moderate-to-high heritability for feeding traits have been reported in similar feeding traits, such as DMI in beef heifers, which was reported to be 0.84 ± 0.12 (Freetly et al., 2020).However, heritability of DMI of mature cows was reported to be lower, 0.34 ± 0.11 by Freetly et al. (2020) and 0.11 by Tarakegn et al. (2021).Therefore, DMC would be the preferred trait for genomic selection in DMC over PVMC.

Drinking Duration and Speed Traits
We observed a greater heritability for the duration traits compared to the milk consumption traits using repeatability models.DDPV had a heritability estimate of 0.024 ± 0.002 when fitting a repeatability model in the full dataset and increased to 0.04 ± 0.003 when evaluated using the 32-d truncated dataset.We observed similar improvements for DSDD with a heritability of 0.067 ± 0.004 for the full dataset and an increase to 0.10 ± 0.005 when truncating the data.A marked change in variance was noted for DSDD from 3.27 ± 0.18 to 6.05 ± 0.34 and 4.67 ± 0.11 to 8.64 ± 0.20 for additive genetic and permanent environmental variances, respectively, indicating that permanent environment is more important in early stages in life compared to across the full preweaning period.This is further confirmed by the repeatability of the trait, which increased from 0.16 ± 0.003 to 0.25 ± 0.004 when the data was truncated.This may indicate that there are differences in calves' ability to acclimate to AMF.This also may be a direct effect of the step-wise weaning program due to the calves being fed less and therefore spending less time suckling.The average DS had an estimated heritability of 0.08 ± 0.004 when the full dataset was analyzed with the linear repeatability model, and had a marked increase to 0.15 ± 0.007 when considering the 32-d truncated dataset with the same model.This trait is derived from the total milk consumption divided by the daily duration of drinking sum and therefore is a ratio influenced by these 2 terms.We observed a linear behavior of this trait across feeding days, where DS increased as the calves aged.This is due to calves increasing their suckling strength as they age and are more capable of drinking more milk Rel.= reliability calculated as r se , where i = 1, 2, … is the number of individuals, se is the standard error of prediction, F is the inbreeding of animal i, and σ u2 is the additive genetic variance (Mrode, 2014).
quickly, resulting in larger calves spending less time drinking because they can drink more quickly.This is also characterized in the change in permanent environmental variance of DSDD.The change in heritability indicates that there is more variation in the strength and speed of which calves drink due to their genetic background at a younger age.The RRM show a greater estimated heritability of all 3 traits.For both the DS and DDPV traits, the optimal model based on AIC was a linear Legendre polynomial RRM.We estimated a heritability of 0.10 (range: 0.03-0.24)for DDPV and 0.19 (range: 0.01-0.51)for DS.Interestingly, for DDPV using an RRM, 32-d truncated had a greater heritability estimate.This may be due to a change in the linear trajectory of the trait that is interpreted due to the weaning process mentioned previously.We would expect a decrease in the drinking duration across time as the calf grows and an increase in DS, which our results support.Drinking speed is most likely less influenced by per-visit variation, because it is a parameter based on the daily sums of milk consumption and drinking duration, and therefore, estimation across the full length of the observation period is feasible.This can be seen in the DSDD trait, which had a heritability estimate of 0.18 (range: 0.01-0.34)and the optimal model was the RRM with a quadratic Legendre orthogonal polynomial and 3 classes of heterogeneous residuals based on AIC.Because this trait is less influenced by the per-visit variation, we are better able estimate differences in calves with differing genetic backgrounds.

Milk Consumption and Drinking Duration Variance Traits
The TCV trait had a heritability of 0.21 ± 0.023, which is similar to the findings of Putz et al. (2019), who reported heritability (± SE) 0.21 ± 0.07 for the root-mean square error of feed intake in wean-to-finish pigs, while the TDV trait had a similar heritability of 0.23 ± 0.024.More studies are needed to investigate longitudinal variance traits, such as estimating variance around an expected curved such as illustrated in Poppe et al. (2020), and also the relationship with future lactation performance and reproductive traits.

Feeder Visit Traits
The DNRV and TNV traits were assessed to indicate behaviors with a propensity to feed more often.The estimated heritability for the full dataset was 0.032 ± 0.0021 and 0.051 ± 0.004 for DNRV and TNV, respectively, for the repeatability model.However, the change was negligent or even decreased when the dataset was truncated at 32 d, with the heritability for DNRV , where i = 1, 2, … is the number of individuals, se is the standard error of prediction, F is the inbreeding of animal i, and is the additive genetic variance (Mrode, 2014).

Graham et al.: GENETIC ANALYSES OF CALF TRAITS
Table 7. Heritability estimates for feeding duration traits using Legendre polynomial random regression models; 2 polynomial (quadratic and linear) and a complete or 32 d of age subset dataset were assessed, and quadratic polynomial with heterogeneous residuals using multiple classes was also assessed on the complete dataset 1 Trait 2 where i = 1, 2, … is the number of individuals, se is the standard error of prediction, F is the inbreeding of animal i, and is the additive genetic variance (Mrode, 2014).

Graham et al.: GENETIC ANALYSES OF CALF TRAITS
Table 8.Heritability estimates for behavior and lifetime bovine respiratory traits using Legendre polynomial random regression models; 2 polynomial (quadratic and linear) and a complete or 32 d of age subset dataset were assessed, and quadratic polynomial with heterogeneous residuals using multiple classes was also assessed on the complete dataset 1 Trait 2 The mean, minimum (Min), maximum (Max), and SD for variance components are provided in supplementary materials.Boldface signifies model with superior performance.2 DNRV = daily number of rewarded visits (observations); TNV = total number of visits per day (observations); FI = feeding interval.
, where i = 1, 2, … is the number of individuals, se is the standard error of prediction, F is the inbreeding of animal i, and σ u 2 is the additive genetic variance (Mrode, 2014).and TNV being 0.020 ± 0.002 and 0.050 ± 0.003, respectively.Another trait that was assessed to predict behavior was FI which had a negligent heritability of 0.008 ± 0.0012 with the full dataset and 0.023 ± 0.0026 when the data was truncated at 32 d.We did find low but reasonable heritability for these traits when using RRM.The RRM with quadratic Legendre polynomials and heterogenous residuals for DNRV and FI did not converge.This is most likely due to the environmental variance of these traits not changing dramatically over time.Fitting the quadratic Legendre orthogonal polynomials with the full dataset was observed to have the best fit for all 3 traits.The greatest heritability was observed for FI of 0.14 (range: 0.02-0.59),while DNRV had a heritability of 0.12 (range: 0.03-0.45;Table 8) and TNV had a similar heritability of 0.10 (range: 0.03-0.25).The difference between DNRV and TNV is that TNV includes all the unrewarded visits.Therefore, environmental noise is added to this trait due to potentially false visits as well as calves, which may be entering and leaving the AMF out of boredom.

BRD and Preweaning ST
We estimated genetic parameters for potential health traits NTT and ST, which were calculated based on if the calf was removed and did not return for any reason.We found that NTT had a higher heritability estimate (0.09 ± 0.01).These estimates are on the upper end of previously reported estimates of 0.4 to 0.10 ( Henderson et al., 2011;Gonzalez-Peña et al., 2019;Haagen et al., 2021;Zhang et al., 2022).It must be remembered that NTT is simply the reporting of treatment records, which makes them be dependent on how routinely and accurately the farm staff are diagnosing and treating the animals, also affecting the accuracy of the phenotypes.

Genetic Correlations
One of the aims of this study was to determine phenotypes that could be used as genetic indicators for BRD resistance in young calves.Therefore, we estimated genetic correlations between milk consumption, feeding duration, and behavior traits with derived health indicator traits.The NTT trait did not have a high a correlation with ST (r NTT-ST = −0.27± 0.11), which indicates that animals were leaving the herd for reasons other than BRD.A multitude of management factors influences ST, and there is no protocol on the removal of calves that have been treated for BRD, and thus may not affect these traits significantly.Therefore, ST may be less influenced by mild cases of reported BRD.The NTT had a moderate genetic correlation with DDPV of 0.46 ± 0.08.The NTT also had a low correlation with TDV of 0.24 ± 0.080.This indicates that the TDV and the amount the calf drinks per visit have implications in calves more susceptible to BRD, such as animals that are more susceptible may change their drinking duration drastically throughout early life.Both NTT and DMC had a greater genetic correlation of −0.59 ± 0.02.The NTT had a lower observed heritability of 0.09, and the simplicity of the animal model is considered better for index selection compared to indirect selection (Martin-Collado et al., 2018).However, any of these 4 traits (DMC, DS, DDPV, TDV) could be used to promote genetic improvement for BRD resilience.The NTT has a favorable genetic correlation with DMC of −0.59 ± 0.02, which is the trait with the highest heritability, h 2 = 0.46 (range: 0.05 -0.68) when using a RRM with quadratic polynomial and heterogenous residuals.This negative correlation signifies that animals genetically selected for greater DMC will present lower BRD incidences in their lifetime.Drinking speed has a favorable moderate negative correlation with NTT of −0.44 ± 0.05; therefore, calves that drink slower overall will be more susceptible to illness.This result agrees with many of the phenotypic reporting of the relationship with BRD and DS, which found either no relationship (Borderas et al., 2009;Swartz et al., 2020;Duthie et al., 2021) or calves with BRD-positive cases presenting slower DS (Johnston et al., 2016;Knauer et al., 2017;Cramer and Ollivett, 2020).The DNRV has a negative genetic correlation with NTT of −0.44 ± 0.05, which means calves that visit the AMF fewer times will be more likely to be treated for BRD.These results agree with previous studies that observed a negative relationship between rewarded visits and illness, where sick calves generally have fewer rewarded visits (Conboy et al., 2021;Perrtu et al., 2023).Per our study, it seems that TNV could be a good indicator for BRD compared to DNRV.However, the mean total heritability for the TNV trait using a quadratic Legendre polynomial RRM was 0.10 (range: 0.03-0.25),which is not much different than directly selecting for the NTT itself.Another trait that could potentially be used as an indicator for BRD is DDPV, which is positively correlated with NTT (r = 0.46 ± 0.08).The mean heritability for DDPV is 0.10 (range: 0.03-0.24),which is similar to directly selecting for NTT itself.The TDV had a lower correlation with NTT (0.24 ± 0.080), but greater heritability (0.23 ± 0.024) compared to DDPV and therefore could be selected upon to make genetic improvements in NTT.The best model to attain that heritability was an RRM with a linear Legendre polynomial.Therefore, the RRM model is more complex than the simpler repeatability Graham et al.: GENETIC ANALYSES OF CALF TRAITS model and is more practical for index selection as discussed earlier.However, selection for this trait will positively influence resistance to BRD in dairy calves.The ST is different than the BRD trait in that it captures more environmental influences that can cause a calf to be culled.Drinking speed was observed to have a lowto-moderate favorable positive genetic correlation (r = 0.31) with ST.A confounding of the linear behavior of DS over time may be an issue.Where animals that remain in the herd longer will have increased DS overall.The DDPV has a very strong negative correlation with DS of −0.79 ± 0.03, while also a low-moderate positive value with ST of 0. 0.31 ± 0.08.Therefore, positive selection for drinking speed will decrease DDPV, which subsequently selects animals less likely to be treated for BRD, and, consequently greater ST.Drinking speed is the third-highest heritability trait we found (h 2 = 0.19; range: 0.07-0.43);therefore, upward selection for this trait will ultimately result in animals with lower BRD instances and increased ST.However, the trait with the greatest potential to indicate resistance to BRD would be DMC, which has favorable correlations with NTT (r = −0.27± 0.11).The DMC also has the greatest heritability of all traits when using an RRM, and the use of the heterogeneous residuals allows one to account for the step-wise weaning program in these systems.

Future Studies
Future studies should focus on leveraging computer assisted algorithms to clean datasets and reduce the complexity caused by erroneous values.Use of artificial intelligence or machine learning modeling can find trends in data with more precision and would help reduce the amount of noise within the data to better perform genomic prediction (De Vries et al., 2023).The addition of video imaging technologies would allow for capturing behaviors and better identifying false positive visits to the AMF.This data may also be used to predict overall resilience in calves using deviations in predicted trajectories to capture subclinical and misdiagnosed illnesses in calves.Furthermore, genetic correlations with future performance traits would help elucidate the importance of these traits observed at a young age on traits that are economically important.Direct testing for BRD instead of farmer reported BRD treatments would be better in confirming deviations that we observe and remove false positives.We may also perform GWAS to identify QTL that influence these traits and include them in our genomic marker arrays.Further research must be performed to develop better genomic estimates for these traits utilizing data from multiple farms.

CONCLUSIONS
This study presents the first genomic-based genetic parameters for feeding behavior traits in preweaning Holstein heifer calves derived from AMF machine data.Our results show that these traits can be genetically improved and be used as genetic indicators for calf health.Random regression models fitted for all the traits were able to capture more of the genetic variation than repeatability models based on model fit.The heritability estimates ranged from low to high, with DMC, DSDD, and DS traits having the highest heritability estimates.The DMC trait is the most promising indicator for genetic selection against bovine respiratory disease, as it presented the highest heritability, is simple to collect and has a strong negative genetic correlation with bovine respiratory disease.In addition, DS, TDV, and BRD treatment are promising traits that can be used in combination with daily milk consumption to promote selection for greater bovine respiratory disease resistance and overall calf health.
from 2 data sources.Per-visit data were recorded beginning summer of 2019, whereas daily information was recorded beginning summer of 2015.2Dataset 1 = per-visit information beginning summer of 2019; dataset 2 = daily information beginning summer of 2015.3TOD = time of day, beginning at midnight of the observation day and categorized into 6-h increments; Age = calf age at feeding day, beginning at d 0 and ending at d 60; ET = entitlement allotted daily, or per visit basis; BRW = recorded birth weight; DP = parity of dam; PD = pen density at day of observation; TI = recorded treatment of illness; recorded as 0 through 5, where 1 is added for each subsequent treatment; THI = average temperature-humidity index of that day(NRC, 1971); TNV = number of visits observed during observation day; PMCA = period milk consumption average for each calf, PPDA = period pen density average for each calf; PDDA = period drinking duration average for each calf; CG = contemporary group effect, concatenation of year, season, and barn.
Graham et al.:  GENETIC ANALYSES OF CALF TRAITS Table6.Heritability estimates for milk consumption traits using Legendre polynomial random regression models; 2 polynomials (linear and quadratic) and a complete or a subset of the dataset (60 and 32 d of age, respectively) were assessed regression with heterogeneous residuals using varying classes were also assessed on complete dataset.Boldface signifies model with superior performance based on Akaike information criterion.2 PVMC = milk consumption per visit (L/visit); DMC = total milk consumption per day (L/d).3 QFull = quadratic Legendre polynomial with full dataset; LFull = linear Legendre polynomial with full dataset; Q32 = quadratic Legendre polynomial with subset data at 32 d; L32 = linear Legendre polynomial with subset data at 32 d; QHR = quadratic Legendre polynomial with full dataset using heterogenous residuals.
minimum (Min), maximum (Max), and SD for variance components are provided in supplementary materials.Boldface signifies model with superior performance based on Akaike information criterion.QFull = quadratic Legendre polynomial with full dataset; LFull = linear Legendre polynomial with full dataset; Q32 = quadratic Legendre polynomial with subset data at 32 d; L32 = linear Legendre polynomial with subset data at 32 d; QHR = quadratic Legendre polynomial with full dataset using heterogenous residuals, with (2) indicating second-degree polynomial or (3) indicating third-degree polynomial. 2 DDPV = drinking duration per visit (min); DSDD = daily sum of drinking duration (min); DS = daily average of drinking speed (L/min).3 QFull = quadratic Legendre polynomial with full dataset; LFull = linear Legendre polynomial with full dataset; Q32 = quadratic Legendre polynomial with subset data at 32 d; L32 = linear Legendre polynomial with subset data at 32 d; QHR = quadratic Legendre polynomial with full dataset using heterogenous residuals.

Figure 1 .
Figure 1.Change in heritability across observation days for the daily milk consumption trait with a quadratic Legendre orthogonal polynomial, including 4 heterogeneous residual (HR) random regression model.

Figure 2 .
Figure 2. Change in heritability across observation days for the daily number of rewarded visits trait with a linear Legendre orthogonal polynomial random regression model.

Table 1 .
Graham et al.:GENETIC ANALYSES OF CALF TRAITS Models used to estimate genetic parameters for feeding and health traits; fixed effects were tested using bi-directional stepwise regression, and lowest Akaike information criterion value was used as final model 1 Trait

Table 2 .
Graham et al.:GENETIC ANALYSES OF CALF TRAITS Descriptive statistics for fixed effects for per-visit and daily traits; factorial levels are set as numeric to show distribution, and contemporary groups represent distribution of individual animals found within each level 1 Age = calf age at feeding day, beginning at d 0 and ending at d 60; CG = contemporary group effect concatenated by year, season, and barn; ET = entitlement allotted daily, or per visit basis; PD = pen density at day of observation; DP = dam parity, either 1, 2, or >3; TI = recorded treatment of illness, recorded as 0 through 5, where 1 is added for each subsequent treatment; THI = average temperature-humidity index of that day(NRC, 1971); TOD = time of day observations per category, beginning at midnight of the observation day and factored by 6-h increments, 1 being earliest, 4 being latest; TNV = number of visits observed during observation day; PMCA = period milk consumption average for each calf; PPDA = period pen density average for each calf; PDDA = period drinking duration average for each calf.
1 Time of day and sex were not used in daily traits; data source and total visits were not used in per-visit traits. 2

Table 3 .
Graham et al.:GENETIC ANALYSES OF CALF TRAITS Descriptive statistics for traits either on a per-visit or daily basis; variance traits were calculated from per-visit information and are a daily or total observational period value 1

Table 4 .
Variance components and heritability estimates for per-visit and daily traits using an animal/repeatability model with full information (age 1-60 d) 1

Table 5 .
Graham et al.:GENETIC ANALYSES OF CALF TRAITS Variance components and heritability estimates for per-visit and daily traits using a repeatability model with subset of information (age 1-32 d) 1