Scope for a Subjective Assessment of Milking Speed

Transcription

1 Scope for a Subjective Assessment of Milking Speed K. MEYER and E. B. BURNSIDE Centre for Genetic Improvement of Livestock Department of Animal and Poultry Science University of Guelph Guelph, Ontario, N1G 2W1, Canada ABSTRACT Farmers' subjective evaluations of their cows' milking speed, scored as "very fast", "fast", "average", "slow", or "very slow", were recorded between 1982 and 1985 under Agriculture Canada's Record of Performance milk recording scheme. For the analysis, scores taken at the beginning (score 1) and the end (score 2) of the lactation were distinguished. Edited data comprised 20,909 records for score 1 and 19,012 records for score 2 for Ayrshires and 190,455 and 148,536 for Holsteins. Data were analyzed in eight subsets. The effects of herd-year-seasons (of calving), calving age, and stage of lactation when scored were evaluated and variance and covariance component estimates were obtained for sires, cows, and residuals. Heritabilities were.21 and.14 for score 1 for Holsteins and Ayrshires, respectively, and.17 and.16 for score 2. Repeatabilities were.42 and.37 for the two breeds. The scope of a sire evaluation for milking speed based on subjective scores is discussed. INTRODUCTION Ease of milking or milkability of dairy cows is of economic importance because it can have considerable impact on the successful management of a dairy unit. It is therefore included among the auxiliary traits considered in numerous dairy selection programs (e.g., Canada and The Netherlands). With automation in modern milking parlors, milking speed is of particular interest. Sivarajasingam et al. (10), for instance, found milking speed to be the third most important variable, after milk yield and fat content, affecting net dairy profits. Various measurements of milking speed have been Received October 21, Accepted April 11, investigated, including 2-min yield, peak and average milk flow rates, and total milking time (1, 7, 13). In Canada, Sharaby et al. (9) studied records of 2-rain milk yield, collected with a milk-ometer digital totalizer (8), and its proportion of the total milk yield, giving intermediate heritability estimates of 24%. Williams et al. (14) related these traits to total milking time (TMT), reporting genetic correlations of unity among the different measures of milking speed but lower heritability (14%) for TMT. Although TMT, using a stopwatch rather than a milk-o-meter, was easier and cheaper to measure than 2-min milk yield, considerable resources were needed to obtain such records. McClelland (2) studied subjective evaluations of cows' milking speed, given by the farmer as "very slow", "slow", "average", "fast", or "very fast", as well as corresponding TMT. Heritability estimates of.25 and.24 for TMT and the subjective scores, respectively, and a genetic correlation estimate of.92 indicated that such farmer-supplied subjective scores could provide an inexpensively "measured", alternative selection criterion for milking speed. As a result, subjective milking speed scores have been collected routinely under the Canadian Record of Performance (ROP) milk recording program operated by Agriculture Canada. This paper describes the analysis of these data and considers the influence of environmental effects and the genetic parameters for subjective milking speed scores. MATERIAL AND METHODS Data consisted of subjective scores for milking speed, scored in five categories from very fast to very slow, collected by Agriculture Canada under the ROP milk recording programme from 1982 to May Of 550,422 records with valid milking speed score, 48,150 were for Ayrshire and 460,823 for Holstein cows. Basic edits eliminated records with missing herd, sire, or breed identity, invalid test 1987 J Dairy Sci 70:

2 1062 MEYER AND BURNSIDE or dry dates, and a.m. or p.m. test day milk yield outside the range of 0 to 60 kg. To establish sire identity, calving date, and age, milking speed records had to be matched with the corresponding ROP production records, also supplied by Agriculture Canada. Each milking speed record was paired to the production record for the cow with the calving date earlier than and closest to the respective test date. Further edits then excluded records with invalid calving date, missing sire identity, or an age at calving outside the range 18 to 156 mo. Subjective scores were to be collected twice per lactation for each cow. Score 1 (MS1) was to be taken at the second test after calving. After inspection of the distribution of records over stage of lactation at test, this time was defined to span d 1 to 110 of lactation. All MS1 records were required to be regular test day records with nonzero a.m. and p.m. yields. Correspondingly, score 2 (MS2) was defined as the farmer's evaluation at the end of lactation and had to be recorded at the test after drying off with a nonzero dry date earlier than the respective test date, zero day yield, and a stage of lacation between 260 and 439 d. This yielded 20,909 records for score 1 and 19,012 records for score 2 for Ayrshires and 190,455 (MS1) and 148,536 (MS2) records for Holsteins. To estimate a "pooled" repeatability (within and across lactations), one analysis of Holstein data included all scores (MS), i.e., regular tests later than 110 d and scores at dry earlier than 260 d after calving, as well as MS1 and MS2 as defined, 375,072 records in total. Figure 1 gives the distribution of scores over stages of lactation for these data. To estimate variance components, daughter records for the most widely used sires were extracted. To make efficient use of the information available, data were analyzed in various subsets (see Table 1), depending on parameters to be estimated and computational limits of the respective programs available. Only sets 1, 4, and 5 included more than one record per cow. For Holsteins, 599 sires with 25 or more records for MS1 were identified, yielding set 2. Considering heifer records only, defined -(3 [_ 0 g Z t2@o@ 8OO0 0! I:~ SO ~6 St a9e of lactation 42 [week] Figure 1. Distribution of Holstein records for milking speed (h = 375,072) over stage of lactation at test.

3 O~ m c3 t~ [- Ox TABLE 1. Characteristics of data sets. Data set Breed Holstein Hotstein Holstein Holstein Holstein Ayrshire Comment All scores Score 1 Score 1 Score 2 Score 1 Score 2 Score 1 Repeated All cows Heifers only All cows Repeated records records No. of records 236, ,341 43,553 96, ,199 42,404 18,302 No. cows 131, ,341 43,553 96, ,199 42,404 14,131 No. sires No. herds No. HYS 1 16,151 14,647 11,751 14, % single sire df 2 for residual 88, ,091 31,199 80, MS a ~s S My S Stage.X S Age S Ayrshire Score 1 All cows 13,951 13, ,035 Ayrshire Score 2 All cows 11,036 11, t~ < ~q Z O 0o Herd-year-seasons. Degrees of freedom. s Milking speed score. 4 Daily milk yield (a,m. plus p.m.) in kilograms. 5 Overall mean. 6 Phenotypic standard deviation.

4 1064 MEYER AND BURNSIDE as first available record for an age at calving between 18 and 36 mo, gave set 3 as a subset of set 2. Set 1 comprised scores throughout the lactation, MS, for these 599 sires, including repeated records but excluding MS2 scores if there was no corresponding MS1 record. The MS2 records for 596 Holstein sires with at least 20 daughters made up set 4. To estimate the genetic correlation between milking speed at the beginning and end of lactation, the first available pair of MS1 and MS2 scores was extracted for each Holstein cow. If a cow did not have a lactation with both scores, the first available record for score 1 was extracted, provided the respective calving was earlier than July Records for 310 sires with 30 or more records for MS1 then formed Set 5. Set 6 for Ayrshires corresponded to 1 for Holsteins, considering, however, only MS1 for 338 sires with at least 8 daughter records. Sets 7 and 8 consisted of MS1 and MS2 records for 318 Ayrshire sires with 7 or more and 288 sires with at least 6 daughters, respectively. For the analysis, milking speed scores were assigned values from 1 to 5 as shown in Table 2. All estimates were obtained by restricted maximum likelihood (REML) (5). Analyses were carried out ignoring the categorical nature of the trait under consideration. A transformation to "objective scores" (11) to normalize the distribution of residuals is sometimes advocated for such data (12). McClelland (2) compared estimates of genetic parameters for milking speed scores on the original and transformed scale and reported little difference. The model of analysis for all data sets included herd-year-seasons (HYS) of calving as fixed, sires as random effects, and stage of lactation at scoring and age at calving as linear and quadratic covariables. Seasons were defined as is standard in the analysis of dairy production records, as March to August and September to February. To estimate variance components between and within sires (all data sets except 1 and 6), the model was then: Yijk=hi + sj + bll (Xlijk -21) + b~l (Xlijk -21) 2 b12 (X2ijk- X2) +b22 (X2ijk-Xz) 2 +eijk [1] where h i denotes the effect of the i th HYS, sj the effect of the jth sire, Xlijk and Xzijk the stage of lactation and age at calving pertaining to Yiik, respectively, 21 and X2 the corresponding means, blm and b2m the linear and quadratic regression coefficient of Y on Xm, and eij k the residual error. For sets 1 and 6 with repeated records, the variance between cows was estimated in addition, extending the model to: Yijkl =hi + sj + djk +bh (X~ijkl--Xt) + b21 (Xlijkl - X1) 2 + ba2 (X2ijkl- X2) + b22 (X2ijkl - X2) 2 + eijkl [2] where djk denotes the effect of the k th daughter of sire j. Sires and cows were assumed unrelated. TABLE 2. Distribution of records over milking speed scores for Holsteins with class means for stage of lactation, age at calving, and test day milk yield. Phenotypic mean Numeric No. Milk Class code records Frequency Stage Age yield (%) (d) (too) (kg) Very fast Fast 2 120, Average 3 195, Slow 4 44, Very slow l Score 1 records only.

5 SUBJECTIVE MILKING SPEED SCORES 1065 Univariate analyses were carried out for data sets 1, 3, 4, 6, and 8. A specialized REML algorithm for a model with two random effects and a large number of fixed effects has been described by Meyer (unpublished data). For sets 2 and 7, milking speed (MS1) and test day milk yield (a.m. plus p.m.) were considered simultaneously using a multivariate algorithm for traits with equal design matrices (3). Univariate analyses for MS1 in set 2 included and excluded test day milk yield as an additional linear and quadratic covariable. Variance component estimates were identical for both models. This implied that all variation in milking speed at the beginning of the lactation due to test day yield was explained by age and stage of lactation. A multivariate analysis treating MS1 and MS2 as different traits was performed for data set 5 (4). The respective procedures of analysis gave lower bound standard errors of variance component estimates for data sets 2, 3, 4, 7, and 8. For each analysis, sire solutions and estimates of regression coefficients were obtained at the last round of iteration. RESULTS AND DISCUSSION Table 2 shows distribution of records over milking speed scores for Holsteins (n = 375,072), phenotypic means for age, stage of lactation, and test day milk yield for each score. Slower milking cows tended to be older and, disregarding the extremes, to have higher test day yield. Average stage of lactation for cows classified as slow or very slow was considerably less than for cows average and faster, identifying some culling for milking speed, or alternatively, the need to adjust subjective scores for stage of lactation. Environmental effects Fitting HYS as fixed effects explained 17.2 (set 2) and 19.5% (set 7) of the total sums of squares (SS) for MS1 (considering all cows) in Holsteins and Ayrshires, respectively. For heifer records only, this proportion doubled (34.4%, set 3). With respective proportions of 21% for set 4 and 25% for set 8, HYS were slightly more important for MS2. Calving age and stage of lactation affected Ayrshire records considerably more than Holstein records. Fitting covariables after HYS explained 1.0% of SS for MS1 (sets 2 and 3) and 3.3% of SS for MS2 (set 4) in Holsteins, and 5.0% (set 7) and 6.3% (set 8) in Ayrshires. For MS1 (set 2), 5.3% of SS and for MS2 (Set 4) 3.8% of SS could be attributed to sires after HYS and covariables. Corresponding values for Ayrshires were 1.8% (set 7) and 1.7% (set 8). TABLE 3. Estimates of regression coefficients (X 104) of milking speed scores on stage of lactation (in points per day and day, 2 respectively) and age at calving (in points per month and month, 2 respectively). Stage Data set Trait Linear Quadratic Linear Quadratic 2 MS MS1 ~ ±.070 ±1.179 ± MS ±3.043 ~.124 ± ± MS ±.817 ±.015 ±1.280 ± MS MS MS MS1 t 8, ±.406 ±.062 ±1.593 ± MS ±2.354 ±.044 ±3.850 ± Estimates from univariate analyses. Age Journal of Dairy Science Vol. 70, No, 5, 1987

6 1066 MEYER AND BURNSIDE Table 3 summarizes the estimates of regression coefficients for Model [1]. Subjective scores increased, i.e., milking speed decreased, with stage at the beginning of the lactation (MS1), whereas cows dried off were considered faster milkers (MS2) as lactation length increased. Figure 2 shows the effect of stage of lactation on milking speed scores for Holsteins. The influence of age on various milking speed traits has been established (9, 13). As also reported by McClelland (2), calving age had a major effect on subjective milking speed scores. Calving age appeared considerably more important in Ayrshires than in Holsteins and to affect scores at the end of the lactation markedly more than at the beginning. Figure 3 illustrates the relationship between calving age and milking speed in the Holstein data. Not including test day yield in the model of analysis, fitting calving age as a covariable accounted for some variation due to yield. This may have caused the difference in estimates of the quadratic regression coefficients from a multivariate analysis of MS1 and MY and a corresponding univariate analysis of MS1 only (sets 2 and 7). Genetic Parameters Estimates of variance and covariance components and the resulting genetic parameters are given in Table 4. After accounting for fixed effects, there was consistently less variation for Ayrshires than for Holsteins. Variance components for score 2 were smaller than for score 1 in both breeds. A heritability estimate of 21% for MS1 (set 2) considering all cows in Holsteins was slightly lower than the 24% reported by McClelland (2). Utilizing first lactation records only (set 3), heritability increased slightly to 23%, the difference being well within the range of sampling errors. Genetic, residual, and phenotypic correlations between MS1 and MY were essentially zero. Other studies found a slight to moderate antagonistic genetic relation between milking speed and milk yield (7, 9, 13). However, as the low heritability estimate of 13% for test day milk yield suggests, both the genetic correlation between MS1 and MY and the heritability of MY may have been biased downward, as sires in the analysis have been subject to selection for milk yield (6). Subjective scores at the end of the lactation in Holsteins appeared somewhat less heritable than at the beginning (h , set 4). A bias of the estimate due to culling of slower milking cows during the lactation is suspected. A multivariate analysis treating MS1 and MS2 as separate traits that would acount for such selection bias increased the heritability estimate for MS2 to 20%. However, the estimate for MS1 was also higher (24 vs. 21%), indicating that this increase may merely reflect sampling. An estimate of the genetic correlation between 3.~o q ~L 2.87 ~U~ 2.8O ~ S~oge of" Ioc~at ion (weeks) Figure 2. Effect of stage of lactation on milking speed for Holsteins. Plotted are phenotypic class means (X) for all Holstein data (n = 375,072) and predicted regression lines for score 1 set 2) and score 2 (set 4). o 3. O0 i 2.9o! 1 ~, 2.~o t 2 2,70 l I 2.60 I,,,,,,,,,, ~ 0 2, B.O calvin~ age [year~s] Figure 3. Effect of calving age on milking speed scores for Holsteins. Plotted are phenotypic class means (~) and the predicted regression line for score 1 (set 2).

7 SUBJECTIVE MILKING SPEED SCORES 1067 /x2 TABLE 4. Estimates of variance and covariance components between (~) and within (aw) sires, phenotypic standard deviations (~p), heritabilities (~2), and genetic correlations for milking speed: all scores (MS), score 1 (MS1), and score 2 (MS2) and daily milk yield (MY). All standard errors given are approximate lower bounds A2 A2 h2 Data set Trait(s) a s a w ap 1 MS MS MY !.125 ±0610 ± MS1/MY MS ± MS ± ± MS MS MS1/MS MS MS ± MY ± ±.031 MS1/MY ± MS ± ± ±.035 * ~ Estimates of covariance components and genetic correlation. MS1 and MS2 close to unity (.96) confirmed that farmer-supplied milking speed scores at different times in the lactation measured the same genetic trait Considering scores throughout the lactation and including repeated records per cow gave a heritability estimate of 19% (set 1). The variance component due to cows was estimated as 18901, yielding a repeatability estimate of 42% (between and within lactations), which, as expected for a genetic correlation of unity agreed with the estimate of the phenotypic correlation between MS1 and MS2 of.43 (set 5). Estimates for Ayrshires were based on considerably less data. Heritabilities of 14% (set 7) and 16% (set 8) for MS1 and MS2, respectively, suggested that subjective scores collected for this breed were slightly less efficient in identifying genetic differences between animals. Including repeated records for MS1 gave estimates of the heritability of 13% and the repeatability (MS1 across lactations) of 37% (set 6). Heritabilities of test day yield and correlation, at test day yield to milking speed scores were higher than in Holsteins. Sire Evaluation Estimating variance components by REML gave best linear unbiased procedure (BLUP) proofs of the sires in the data as a by-product. For Holsteins, sire solutions were obtained for data sets 1, 2, and 3. Figure 4 shows the distribution of proofs using all available records per cow (set 1) and using first lactation records for score 1 only (set 3). Product-moment correlations between proofs were.96 (1 and 2),.74 (1 and 3), and.76 (2 and 3) with expected values of.88,.73, and.82, respectively. The latter were calculated as: ~/(1 + X/n±)/(1 + X/nj) averaged over sires with nj < ni and using a variance ratio of X = Here n i denoted the number of effective daughters for proof i, derived from the direct inverse of the coefficient matrix for sires in the respective analysis. Corresponding regressions were 1.01 (set 1

8 1068 MEYER AND BURNSIDE.5 xx x x based on subjective scores in early lactation, together with a BLUP sire evaluation scheme, will be implemented in Canada. _~.2[ L. t u_ -.3[.~ xx ACKNOWLEDGMENTS This study was financed by the Canadian Association of Animal Breeders, the ROP Division of Agriculture Canada, Holstein Canada, and the Ontario Ministry of Agriculture and Food. Data were supplied by Agriculture Canada. Thanks are due to J. Chesnais for his helpfulness in making information available. -.5[ -.~[ x '_.~,_.h ~_.~ ~_.~,,~, ~ ~.~ ol I records Figure 4. Relationship between Holstein sire proofs for milking speed based on all scores (MS, Data set 1) and on first lactation records for score 1 only (MS1, Data set 3). on 2),.86 (set 1 on 3), and.94 (set 2 on 3), all with an expectation of unity. CONCLUSIONS The heritability of a farmer-supplied subjective assessment for milking speed in five categories was of order.2 with a coefficient of variation of the subjective scores around 27%. As they were highly correlated with total milking time, an objective measurement of milking speed (2), such scores provide a suitable criterion to distinguish between bulls with transmitting abilities for fast and slow milking daughters. Sire proofs utilizing repeated records per cow and scores throughout the lactation were highly correlated with proofs considering one score per cow only, collected early in the lactation. As the latter also appeared more heritable, a sire evaluation for milking speed based on the first available score per cow recorded at the second test after calving is recommended. Recent research (Monardes and Moore, personal communication) suggests that milking speed is genetically independent of lactation somatic cell count, an indicator of udder health. Then selection for an increased milking speed is desirable and will improve economic returns (10). As a result of this study, a national milking speed recording programme REFERENCES 1 Blake, R. W. and B. T. McDaniel Relationships among rates of milk flow, machine time, udder conformation and managemental aspects of milking efficiency. A review. J. Dairy Sci. 61: McClelland, L. A A comparison of objective and subjective measures of milking speed in Canadian Holstein-Friesians. M. Sc. Thesis, Univ. Guelph, Ont. 3 Meyer, K. t985. Maximum likelihood estimation of variance components for a multivariate mixed model with equal design matrices. Biometrics 41: Meyer, K Between algorithim: a "short cut" restricted maximum likelihood procedure to estimate variance components. J. Dairy Sci. 69: Patterson, H. D., and R. Thompson Recovery of inter-block information when block sizes are unequal. Biometrika 58: Robertson, A The effect of selection on the estimation of genetic parameters. Z. Tierz. Zuchtungsbiol. 94: Schneeberger, M. and C. Hagger Sire evaluation for milkability traits in Swiss Braunvieh. Livest. Prod. Sci. 13: Sharaby, M. A., E. B. Burnside, and R. R. Hacker Accuracy of an automated technique for determining individual milking rates under field conditions. J. Dairy Sci. 60: Sharaby, M. A., L. R. Schaeffer, and E. B. Burnside Variance components and sire evaluation for milk flow rates. J. Dairy Sci. 62: Sivarajasingam, S., E. B. Burnside, J. W. Wilton, W. C. Pfeiffer, and D. G. Grieve Ranking dairy sires by linear programming dairy farm models. J. Dairy Sci. 67: Snell, E. J A scaling procedure for ordered categorical data. Biometrics 20: Tong, A.K.W., J. W. Wilton, and L. R. Schaeffer Application of a scoring procedure and transformation to dairy type classifications and beef ease of calving categorical data. Can. J. Anim. Sci. 57:1. 13 Williams, C. B., E. B. Burnside, and L. R. Schaeffer Genetic and environmental parameters of two field measures of milking speed. J. Dairy Sci. 67:1273.