Breeding Programsof Dairymen Selecting Holstein Sires by Computer M. L. McGILLIARD and J. S. CLAY Department of Dairy Science Virginia Polytechnic Institute and State University Blacksburg 24061 ABSTRACT
A computerized linear program (MAXBULL) selects groups of sires to attain or exceed goals of dairymen and maximize weighted average Predicted Difference for milk. Goals for 196 dairymen with Holsteins in the spring of 1982 averaged - . 0 5 Predicted Difference fat percent, .85 Predicted Difference type, $17.52 per unit, and percents net superiority 29, 35, 36, 27, 33, and 46 for descriptive traits udder support, rear udder, fore udder, teats, hind legs, and feet. Percent net superiority is the percentage of units from sires having a significant number of superior daughters in a trait, reduced by the percentage of units from sires having a significant number of inferior daughters. Associated Predicted Difference milk, fat, and dollars were 789 kg, 25 kg, and $215. All goals exceeded averages of artificial insemination sires available, and Predicted Difference dollars exceeded Virginia service sire averages by $44 and the 90th percentile sire by $6. Predicted Difference fat percent was the most frequently binding production trait, whereas udder support was the most frequently binding type trait. Predicted Difference fat percent and price were most influential on maximum Predicted Difference milk. Breeding programs of dairymen using bulls from one organization differed from those using bulls from several organizations.
reported by Helgren et al. (1). None of these programs has been adapted for use by dairymen or extension personnel. A field application of the linear program MAXBULL described by McGilliard and Clay (5) maximizes average Predicted Difference (PD) milk in the breeding program subject to constraints of average minimum PD fat%, PD type, six descriptive type traits, and maximum average direct herd semen price. Several other features, such as bull editing and balance in number and distribution of bulls suggested, improve the utility and acceptability of the program. MAXBULL was developed to release dairymen from the arduous task of selecting sires from among 400 candidates, where sires recognized as superior are generally expensive. Use of MAXBULL encourages dairymen to concentrate on herd breeding goals, leaving the selection of sires to a computer. Additionally, MAXBULL not only maximizes PD milk for any set of goals but encourages planning and stability of goals over time as well as maintenance of balance among traits, minimization of risk, and control of expense. Breeding recommendations from MAXBULL are mailed after each semiannual USDA sire summarization to dairymen who have submitted goals. This manuscript describes and evaluates requests from 196 dairymen with Holsteins and subsequent MAXBULL recommendations given them in spring 1982. Because most dairymen had been contacted through group meetings, their goals were influenced by educational discussion and adjustments necessary to obtain solutions.
INTRODUCTION
Computerized linear programming techniques for minimizing semen expense while selecting artificial insemination (AI) sires were published by McGilliard (4) and Shanks and Freeman (6). A nonlinear technique to control risk was Received June 30, 1982. 1983 J Dairy Sci 66:654-659
BULL EDITING
Independent culling levels are offered in MAXBULL before PD milk is maximized. Table 1 indicates average culling limits applied by dairymen to remove bulls from consideration. This procedure reduces variation among bulls entering the breeding program, but it cannot increase the average PD milk obtained. 654
OUR INDUSTRY TODAY Most dairymen preferred to place limits on PD milk and price for individual bulls. Although the mean was $73 per unit, most price limits were nearer the harmonic mean of $32. Repeatability was used more to eliminate bulls proven in few herds than to obtain a battery of high repeatability sires. Ironically, restriction to high PD milk reduces the maximum PD milk that can be obtained. This is because bulls lower in PD milk, but acceptable, often are superior in other traits and release a larger percentage of units to bulls higher in PD milk. Occasionally, independent culling levels from dairymen must be relaxed to obtain bulls for the program. Mean number of bulls surviving edits and entering the linear program was 91 from 411 Holsteins. Thirty percent of the dairymen used bulls from only one organization, and these dairymen averaged only 54 bull survivors with a standard deviation of 20. Percentages of dairymen having various numbers of survivors were extremely symmetric. Of those using more than one stud (seven studs maximum), 20% culled to fewer than 50 bulls whereas another 22% retained more than 150 bulls. Nearly a quarter of the dairymen were willing to use bulls from all seven studs. Half of the dairymen using bulls from only one stud had as many as 50 to 100 bulls remaining after edits. These dairymen were using from the largest studs. Table 2 contains the distribution of numbers of bulls requested and recommended. Dairymen request a m i n i m u m number of bulls rather than an exact number. Dairymen using bulls from only one stud requested fewer bulls in general, but a disproportionate (P<.10) number of these dairymen received 7 to 10 bulls in their solutions. Where smaller numbers of bulls are considered, few bulls match the goals which are, therefore, more likely binding so more bulls must be mixed in the solution to obtain the goals and maximize PD milk.
351 bulls. Descriptive type was scored +1, 0, - 1 according to strengths and weaknesses in the Holstein Red Book (2). The average of the descriptive codes is the percentage of net positive ones where a - 1 cancels a +1. Strong feet predominated with 18% net ones, and bulls siring desirable hind legs were in the minority, reflected by - 2 % net ones. Price per unit averaged $10.70 but was variable. Traits of the 90th percentile bull are the last column of Table 4. Except for price which is more skewed in distribution, the 90th percentile was 1.25 standard deviations above the mean.
TABLE 1. Average independent limits used to cull bulls from consideration. Trait
N
.~
SD
Repeatability (%) PDa milk (kg) PD F~t (%) PD Dollars ($) PD Type (points) Price/unit ($) No. of bulls surviving independent culling
134 147 139 131 135 196
39 484 --.20 117 --.16 73
11 103
To evaluate MAXBULL breeding programs, it is useful to know characteristics of the population from which bulls were chosen. Averages and standard deviations for 411 active AI Holstein bulls are in Table 3. The PD dollars were based on $13.00 per 45.4 kg milk with a $.172 fat test differential. Type means were for
196
91
.07 25 .64 103 69
apredicted Difference.
TABLE 2. Percentages of dairymen with various numbers of bulls requested in programs and subsequently recommended by MAXBULL. Minimum requested by dairymen No. of bulls 1-3 4
A.I. B U L L P O P U L A T I O N
655
5 6 7-10 11-15 Total
One stud a
More studsb
5
6
33 24 16 22 0 100
18 25 16 31 4 100
Recommended in solution One stud a
More studsb
2 2
0 2
7 9 70 10 100
6 7 55 30 100
a58 dairymen restricting bulls to one stud. b138 dairymen using bulls from more than one stud. Journal of Dairy Science Vol. 66, No. 3, 1983
6 56
McGILLIARD AND CLAY
TABLE 3. Means and standard deviations for artificial insemination Holstein bulls, dairymen goals, and MAXBULL recommendations. Mean Active bulls a
Trait Repeatability (%) PD e Milk (kg) PD Fat (%) PD Fat (kg) PD Dollars ($) PD Type (points) Udder support (net %) Rear udder (net %) Fore udder (net %) Teats (net %) Hind legs (net %) Feet (net %) Price/unit ($)
Standard deviation
MAXBULL goalb
73 612 --.06 18 161 .42 14 14 12 0 --2 18 10.70
MAXBULL resultsb
Active bulls
67 789 --.05 25 215 .85 29 35 36 27 33 46 17.52
-.05
.58 23 20 21 14 22 22 18.72
MAXBULL goal
16 158
MAXBULL results 6 90
.10
.05
6 38
.04 3 22
.92 77 86 79 78 81 78 18
.56 17 16 17 14 17 16 13
.41 19 29 23 22 21 24 12
a411 bulls whose semen was for sale by one of seven major artificial insemination organizations; 351 with PD type and 338 with descriptive type evaluations. b196 dairymen with Holsteins. Cpredicted Difference.
G O A L S A N D RESULTS
Averages and s t a n d a r d deviations o f goals s u b m i t t e d b y d a i r y m e n are in Table 3. All goal averages e x c e e d e d averages o f available bulls. Given a choice, d a i r y m e n w e r e willing or a c c u s t o m e d t o p a y i n g $10 to $20 per u n i t for
semen. Descriptive t y p e goals occasionally w e r e n o t specified; 21% o f d a i r y m e n , f o r instance, c h o s e to ignore t e a t size and p l a c e m e n t . U d d e r s u p p o r t a t t r a c t e d t h e m o s t interest, with 94% o f d a i r y m e n specifying a goal. T h e PD fat% goal and s t a n d a r d deviation w o u l d have b e e n h i g h e r
TABLE 4. Means for goals and MAXBULL recommendations according to number of studs from which bulls were requested. One stud a Trait PD e Milk (kg) PD Fat (%) PD Fat (kg) PD Dollars ($) PD Type (points) Price/unit ($)
Goal
--.05 .38 14.50
Result 747 -.05 23 203 .70 12.18
More studsb Goal
--.05 .66 20.59
a58 dairymen using bulls from only one stud. b138 dairymen using bulls from at least two studs. CDairy Herd Improvement records. d90th percentile for each trait independent of the others. epredicted Difference. Journal of Dairy Science Vol. 66, No. 3, 1983
Result 806 --.05 25 220 .92 19.76
Virginia service siresc
90th percentile AI bull d
658 --.07 19 171
809 .06 25 209 1.54 20.00
OUR INDUSTRY TODAY if the goal had not been reduced in several instances to acquire feasible solutions. Table 3 also contains averages and standard deviations of 196 MAXBULL breeding programs. The PD milk exceeded the average of available bulls by 177 kg. The PD dollars, at $215, were $6 above the 90th percentile bull for PD dollars (Table 4). Repeatabilities were slightly below average and largely were determined by an average maximum of 22% of semen acceptable from low-repeatability (less than 50%) bulls. The traits most frequently binding were PD fat% followed by udder support and price. This is verified by the closeness of goals and results for these traits in Table 3. PD fat% was uncorrelated with other goals, but price was correlated with PD type and udder attachment goals. Thirty programs priced less than $10 per unit averaged 703 kg PD milk, - . 0 5 PD fat%, 21 kg PD fat, 189 PD dollars, and .49 points PD type for $7.44. Virginia service sire averages reported by the Dairy Herd Improvement Association (DHIA) for winter 1982 are in Table 4. These include all reported service sires having Predicted Differences. MAXBULL solutions were sub-
65 7
stantially above these averages at +131 kg PD milk and +$44 PD dollars. This $44 PD superiority would be worth an increment of $11.00 per unit of semen (3). Although the cost of service sires was not recorded, it averaged more than $6.52 per unit ($17.52 minus $11.00) which would make equivalent the value of MAXBULL solutions and Virginia service sires. Distinct differences were in Table 4 between dairymen using bulls from one stud and those using bulls from more than one stud. Single-stud dairymen requested and received .28 points less PD type and $6 less cost than other dairymen. Because of their restricted bull numbers, these one-stud dairymen obtained 59 kg less PD milk and $17 less PD dollars. For production alone, that difference is worth $4.25 in semen price, less than the actual differential of $7.58. For those using bulls from more than one stud, PD dollars were $220 with a point of PD type for less than $20 per unit. PREDICTING MILK FROM GOALS
Average PD milk for the breeding programs was regressed on linear and quadratic goals of PD fat%, PD type, and price (Table 5) to
TABLE 5. Partial regression coefficients to predict Predicted Difference (PD) milk for bulls and MAXBULL goals. Trait/ goal Intercept PD Fat% (L)d (03 PD Type (L) (03 Price/Unit (L) (Q) PD Fat% × PD Type PD Fat% × Price PD Type × Price SD (PD Milk) R-Square N
AI bulls 469.86 --959.05 * 29.10 --61.51 * --6.47 12.55" -.06* --117.61 9.43* .18 159 .61 351
All 604.42 -- 1386.09 * --2709.29" --43.13 * --8.58 10.00" -.12" --93.12 --8.87 1.05" 89 .69 185
MAXBULL goalsa One stud b 563.80 -- 1332.00 --6120.43 -25.21 --30.47* 13.94" -.20* --27.22 --31.96 --.41 83 .76 56
More studs c 622.56 -- 1553.03 * --2890.33" -- 32.33 2.83 8.42* -.07* --185.97 2.05 --.09 86 .72 129
awhere all goals specified by dairymen. bDairymen requesting bulls from only one stud. CDairymen requesting bulls from more than one stud. dL, linear; Q, quadratic. *Different from zero (P<.05). Journal of Dairy Science Vol. 66, No. 3, 1983
658
McGILLIARD AND CLAY
identify goals influencing PD milk obtained and to establish an equation to predict PD milk. The coefficients are partial regressions based only on dairymen specifying all three goals. For comparison, PD milk of the AI bulls was regressed on their PD fat%, PD type, and price, although it would be incorrect to assume a variable such as price is influencing PD milk. It was of interest whether the traits most closely related to the bulls' PD milk were those having most influence on subsequent PD milk averages in the breeding programs. All multiple correlations were between .78 and .87. The most predictive measures in the bull population were linear PD fat%, PD type, and price as well as quadratic price and the product of PD fat% and price. The PD milk of bulls declined at a constant rate as PD fat% increased from bull to bull. As price increased, PD milk was not only higher but it also declined less quickly with increasing PD fat%. For the 185 dairymen who submitted all three goals, all linear and quadratic goals significantly influenced PD milk average except for quadratic PD type. The only significant product was between PD type and price, indicating that higher price goals reduced losses in PD milk from increasing PD type. For programs using bulls from only one stud, increases in PD type significantly lowered PD
milk. Regressions on PD fat% were not different from zero, probably because those using bulls from only one stud were forced to limit PD fat% goals to obtain solutions. Evidence of this limitation is the reduced variation in the PD fat% goals, which was 25% of the variation of those using bulls from more than one stud. Because only binding constraints restrict PD milk in individual solutions, the same multiple regression was fit using 55 breeding programs with all three goals binding. This regression had a higher multiple correlation (-94) and fewer significant goals. Linear PD fat% and linear and quadratic price goals influenced PD milk. The increase in PD milk relative to increases in price was steeper than in the full set of goals but declined in rate of increase at higher prices. The PD type, as a binding constraint; did not influence PD milk in these restricted data. Table 6 contains PD milk and PD dollars obtained from nine sets of MAXBULL goals. It demonstrates typical changes for different goals. Responses also have been predicted in Table 6 for the same sets of goals based on regressions on 185 goals in Table 5. Although the correlation between predicted and actual responses was .88, both the predicted PD's milk and differences among them were underestimated. Reasons for higher responses than predicted
TABLE 6. Sample goals illustrating changes in concurrent Predicted Difference (PD) milk and dollars.
PDFat%
PDType
(%) -.05 -.05 -.05 -.05 --.05 --.05 -.05 0 --.10
(points) 0 0 0 1.0 1.5 0 0 0 0
Goalsa Descriptiveb (%) 0 0 0 0 0 25 50 0 0
Price
PDMilk
($) 10 15 5 10 10 10 10 10 10
(kg) 875 938 743 780 699 840 792
816 922
Results PDDollars Deviationc ($) 239 258 201 212 189 230 216 236 238
($) 19 -38 -27 -50 -9 -23 --3 --1
~d (kg) 759 797 716 723 698 692
813
abased on 402 sires with price ~$50; at least 4 bulls in each solution, maximum of 25% low R bulls, maximum of 15% from each. bDescriptive type. CDeviation in PD dollars from first goal. dpD milk predicted from all dairymen regression in Table 5. Journal of Dairy Science Vol. 66, No. 3, 1983
OUR INDUSTRY TODAY
6 59
2
sires available. Even when d a i r y m e n are willing to pay more than average prices, the use of i n d e p e n d e n t culling levels restricts n u m b e r s of acceptable bulls severely. Selection indexes tend to highlight a few bulls superior in several traits including high price. The M A X B U L L program capitalizes on the ability to use one ................ 2 ~ , bull to compensate for inferior traits of a n o t h e r ~ ~oi ~',~.... ~ to o b t a i n acceptable averages and m a x i m u m PD .... milk. Negative relationships a m o n g several ;;:,...........~;;...........;;~...........;;~...........;;~..........;~;;......... ;;;~- traits, coupled with current industry pricing PD talk (k,) customs, make the MAXBULL approach a useful m e t h o d of selecting sires. Average Figure 1. Distribution of MAXBULL breeding programs by average Predicted Difference (PD) milk, M A X B U L L breeding programs are balanced in Predicted Difference fat%, and price category, secondary traits and obtain PD dollar averages far exceeding usual programs and equaling the highest 10% of available AI sires. In Virginia, M A X B U L L breeding programs are a consequence of the goals from dairymen, are available free of charge through B. G. from which the regressions were derived, being Cassell, extension geneticist, dairy science. less variable than the designed nine sets, with Goals are retained on disk and resolved each 6 the d a i r y m e n requesting more descriptive type m o after new PD's are available. M A X B U L L is and more bulls from fewer studs. These charalso available in Iowa for $5.00 through T. E. acteristics would lower the PD milk as seen in Aitchison, extension dairy Specialist, Iowa the predicted numbers. The regressions, thereState University. Several state extension spefore, indicate the influence of various goals on cialists have expressed interest in using the subsequent PD milk b u t underestimate difprogram. ferences in PD milk available from changing The educational benefits of establishing goals, written genetic goals, the balance and stability The designed goals in Table 6 illustrate the of goals over time with c o n c u r r e n t improven o n l i n e a r i t y in changing price or PD type, m e n t in PD milk, and the impartial selection of where the first i n c r e m e n t of 1 p o i n t costs sires to fit the program m a k e M A X B U L L an 27 PD dollars and the next .5 p o i n t increase attractive alternative to conventional sire costs an additional 23 PD dollars. Because selection procedures. M A X B U L L is obtaining m a x i m u m PD milk, some changes in PD fat% leave the resulting PD REFERENCES dollars u n c h a n g e d when a wide assortment of bulls are available. 1 Helgren, R. E., G. E. Shook, and R. A. Schoney. 1980. A portfolio model for dairy sire selection. J. Figure 1 demonstrates the increase in PD Dairy Sci. 63(Suppl. 1):109. milk with reductions in PD fat% goals and 2 Holstein Association. 1982. Sire summaries 1982. increases in price goals. Points in the scatter Volume 1. Holstein-Friesian Assoc. Am., Brattleboro, VT. diagram represent M A X B U L L breeding pro3 McGilliard, M. L. 1978. Guidelines for purchasing grams of individual d a i r y m e n where the n u m b e r s semen. J. Dairy Sci. 61:1680. of the points represent the price category. 4 McGilliard, M. L. 1979. Relationships among Fifty-two observations are masked by those in genetic goals and herd breeding expenses. J. Dairy the scatter plot. Sci. 62:85. 5 McGilliard, M. L., and J. S. C l a y . 1 9 8 3 . Selecting groups of sires by computer to maximize herd CONCLUSION breeding goals. J. Dairy Sci. 66:647. Given an objective o p p o r t u n i t y , d a i r y m e n 6 Shanks, R. D., and A. E. Freeman. 1 9 7 9 . Choosing progeny-tested Holstein sires that meet genetic establish goals superior to the averages of AI goals at minimum cost. J. Dairy Sci. 62:1429. 2, ........... ~ 23 " ' ' 22~. . . . zz
23
22 2 2
3s
33
Journal of Dairy Science Vol. 66, No. 3, 1983