All-Breed Evaluation (AIPL Res. Rep. AB1 (4-07)
You are here: Beltsville Area Home / Beltsville Agricultural Research Center / Animal Genomics and Improvement Laboratory / Animal Improvement Program

AB1 (4-07)

All-breed evaluation

P.M. VanRaden, M.E. Tooker, G.R. Wiggans, J.B. Cole, and J.H. Megonigal, Jr.
Animal Improvement Programs Laboratory, ARS-USDA, Beltsville, MD 20705-2350
301-504-8334 (voice) ~ 301-504-8092 (fax) ~ ~

Impact of model change || Background || All-breed animal model || Inbreeding and heterosis || Genetic bases || Genetic trends || Evaluation breed || Additional resources

Records from all breeds, including crossbreds, are now combined and analyzed together in one animal model. All relatives, regardless of breed composition, contribute to each animal's genetic evaluation, and more cows are compared within management group in herds containing multiple breeds and crossbreds. Previously, most crossbred cows were excluded unless they were part of a breed-sponsored herdbook expansion program. Traits processed with the all-breed model are yield, productive life (PL), somatic cell score (SCS), and daughter pregnancy rate (DPR). Multibreed models have been applied for U.S. goats since 1988 and for Holstein and Brown Swiss calving ease since 2005. Evaluations are calculated initially on an all-breed base and then are converted to traditional within-breed genetic bases for release to the dairy industry. The effect of heterosis (hybrid vigor) is subtracted from each trait in the all-breed model, but when evaluations of crossbred animals are converted to a purebred evaluation scale, the heterosis expected when crossbreds are mated to purebreds is included in the predicted transmitting ability (PTA).

Impact of model change

To assess the impact of changing to an all-breed model that includes crossbreds, all-breed evaluations were calculated from data available for official February 2007 evaluations. Within-breed averages for bulls in active artificial-insemination (AI) service were compared by evaluation model:

Trait Evaluation
(21 bulls)
Brown Swiss
(46 bulls)
(22 bulls)
(656 bulls)
(108 bulls)
Milking Shorthorn
(6 bulls)
PTA milk (lb) Within-breed 379 694 820 877 634 433
All-breed 332 416 666 850 450 427
Change* −47 −278 −153 −27 −184 −6
PTA fat (lb) Within-breed 14.4 30.7 33.4 33.0 39.3 19.3
All-breed 12.8 18.9 28.8 31.6 34.4 16.0
Change* −1.6 −11.8 −4.6 −1.4 −4.9 −3.3
PTA protein (lb) Within-breed 13.1 26.3 23.5 27.4 27.3 11.8
All-breed 11.9 16.9 18.8 26.9 20.8 12.5
Change* −1.2 −9.3 −4.7 −0.5 −6.5 0.7
PTA PL (mo) Within-breed 0.4 1.9 −0.1 0.9 1.8 2.6
All-breed −0.2 1.4 −0.2 0.9 1.4 2.7
Change* −0.5 −0.6 −0.2 0.0 −0.4 0.2
PTA SCS Within-breed 3.0 2.9 3.0 2.9 3.0 2.9
All-breed 3.0 2.9 3.0 2.9 3.0 2.9
Change* 0.0 0.0 0.0 0.0 0.0 0.0
PTA DPR (%) Within-breed −0.1 −0.1 −0.8 −0.2 0.0 −0.1
All-breed −0.2 −0.3 −1.1 −0.5 −0.3 −0.5
Change* −0.1 −0.2 −0.2 −0.3 −0.3 −0.4
Net merit ($) Within-breed 104 256 162 247 277 182
All-breed 77 167 118 238 217 181
Change* −27 −89 −44 −10 −60 −1
Yield (milk-fat-protein) reliability (%) Within-breed 74.4 83.7 79.9 88.6 85.4 79.8
All-breed 76.0 85.0 80.3 88.7 85.2 86.3
Change* 1.7 1.3 0.4 0.1 −0.2 6.5
Daughters (no.) Within-breed 105 194 81 695 278 77
All-breed 112 226 86 698 300 116
Change* 8 32 5 2 22 39
Herds (no.) Within-breed 44 80 42 207 72 33
All-breed 49 97 45 208 80 58
Change* 5 17 3 1 8 24
*Change may differ from difference of within-breed and all-breed PTAs because of rounding.

The all-breed model should enable progeny-test programs to increase the number of matings of young bulls to cows of other breeds because all daughters contribute to the PTA. Milking Shorthorn and Brown Swiss bulls had substantial increases in daughter numbers (50% and 16%, respectively) for February 2007 all-breed data, whereas Ayrshires, Guernseys, and Jerseys has modest increases (6 to 8%). Some calculated reliabilities decreased because 1) some management groups that previously spanned 4 months included only 2 months, 2) the variance of permanent environmental effects for yield traits increased for Jerseys and Brown Swiss (0.18 of total variance now used for all breeds), and 3) a mistake in reliability calculations for breeds with small populations was corrected (contributions from progeny without records were being doubled, which affected reliabilities but not PTAs).

Evaluations from the all-breed model applied to February 2007 data are available for bulls in format 38 and for cows (including both elite and high-ranking grades) in format 105. Most of the high-ranking grade cows for some evaluation breeds are actually crossbreds. Changes in PTA and reliability were summarized for the top previously evaluated bulls of each breed based on net merit.

For some individual bulls, PTAs changed because genetic estimates from ancestors of other breeds that had not been previously included in the evaluation now are used by the all-breed model. In addition, some information from crossbred animals that had been included in within-breed evaluations because of herdbook expansion programs contributes differently to evaluations with the all-breed model. For example, the Jersey herdbook includes crossbreds with Jersey sires but dams of other breeds. Those dams were grouped with Jersey dams without known information for within-breed evaluation. Because most Holsteins have a higher genetic merit for milk yield than do Jerseys, Jersey crossbreds with Holstein dams had been underevaluated for milk yield by the within-breed model but now are credited properly for that advantage by the all-breed model, which groups the dams of crossbreds according to appropriate breed. In contrast, Jersey sires of those crossbred daughters no longer receive inappropriate credit for additional yield merit from Holstein dams and, therefore, PTAs for milk yield have declined for those bulls. The reverse situation would apply for PL and DPR because Jerseys have a higher breed estimate for genetic merit for those two traits than do Holsteins, and PTAs for PL and DPR of Jersey bulls with crossbred daughters from Holstein dams improved compared with their February 2007 within-breed evaluations.

Changes between all-breed and February 2007 within-breed evaluation averages and standard deviations for yield traits were documented for base cows and bulls born during the last 10 years:

Statistic Trait Population Ayrshire Brown Swiss Guernsey Holstein Jersey Milking Shorthorn
Change in average PTA (lb) Milk yield Base cows 2 14 4 <1 9 182
U.S. bulls −101 −220 −191 −34 −183
Foreign bulls −60 −111 −54 −22 −94
Fat yield Base cows 0.1 0.4 0.1 <0.1 0.2 8.0
U.S. bulls −4.0 −10.0 −7.0 −1.0 −5.0
Foreign bulls −2.0 −5.0 −3.0 −1.0 −6.0
Protein yield Base cows 0.1 0.3 0.1 0.0 0.2 6.0
U.S. bulls −3.0 −7.0 −7.0 −1.0 −7.0
Foreign bulls −3.0 −4.0 −4.0 −1.0 −5.0
Change in PTA SD (%) Milk yield Base cows −4 0 0 1 1 13
U.S. bulls −8 1 −1 0 −1
Foreign bulls −10 −2 −4 0 −3
Fat yield Base cows −2 −5 −2 −1 −4 15
U.S. bulls −3 −6 −2 −2 −3
Foreign bulls −2 −8 −10 −1 −5
Protein yield Base cows −5 −4 0 0 −1 16
U.S. bulls −6 −4 0 0 −4
Foreign bulls −3 −6 −6 −1 −3

Average yield PTAs of base cows increased only slightly, and standard deviations generally decreased slightly or remained the same for all breeds except Milking Shorthorn. The large changes for Milking Shorthorns likely were the result of small population size and large numbers of crossbred cows. As expected, Holstein evaluations remained fairly stable with little or no change. Average PTAs generally decreased more for U.S. bulls than for foreign bulls, but PTA standard deviations decreased more for foreign bulls than for U.S. bulls for milk yield and for Brown Swiss fat and protein yields.

Differences between PTA and parent average for bulls born since 2000 also were compared for all-breed and February 2007 within-breed evaluations:

Trait Evaluation
PTA − parent average (lb)
Ayrshire Brown Swiss Guernsey Holstein Jersey Milking Shorthorn
Milk Within-breed 66 91 46 103 1 72
All-breed 75 150 92 111  4 75
Fat Within-breed 2.8 6.0 1.4 3.7 1.1 2.0
All-breed 2.7 8.1 2.8 4.0 0.4 2.1
Protein Within-breed 1.7 3.3 1.1 3.2 0.2 1.4
All-breed 1.5 4.8 3.0 3.4 0.2 1.5

The relationship between PTA and parent average was not affected by the change from a within-breed model to an all-breed model.

All-breed evaluations were tested by the International Bull Evaluation Service (Interbull; Uppsala, Sweden) in March 2007, and average genetic correlations of the United States with other countries generally changed by less than 0.01. Exceptions were correlation increases of 0.02 for PL and 0.01 for SCS of Jerseys and 0.01 for PL of Ayrshires and correlation decreases of 0.01 for fat yield and SCS of Brown Swiss.

Interbull evaluations on foreign scales were converted to the U.S. scale with conversion formulas from the March 2007 Interbull test evaluation to allow direct comparison of foreign evaluations with U.S. all-breed evaluations. Foreign and U.S. evaluations of the top four bulls for February 2007 net merit were examined for Brown Swiss bulls with large evaluation changes and sufficient foreign daughters for comparison:

Bull Daughters that were crossbreds in official February 2007 evaluation (%) Trait Change U.S. all-breed evaluation Converted evaluation
Germany-Austria Italy Switzerland
Agenda B194882 17.0 Daughters (no.) 21 56 63
Milk (lb) −945 594 68
Fat (lb) −40 51 23
Protein (lb) −31 29 18
Even B186040 0.5 Daughters (no.) 64 466 860 400
Milk (lb) −118 39 196 163
Fat (lb) −5 3 3 6
Protein (lb) −3 14 17 18
Excite B191914 6.0 Daughters (no.) 52 196 208
Milk (lb) −419 310 68
Fat (lb) −19 28 23
Protein (lb) −15 24 18
Pronto B191184 9.0 Daughters (no.) 68 237 2,229 171
Milk (lb) −583 544 1,102 388
Fat (lb) −20 19 45 19
Protein (lb) −17 16 37 9

Although U.S. all-breed PTAs for all yield traits declined substantially from official February 2007 within-breed PTAs for Agenda and Excite, their all-breed PTAs were still higher than their converted PTAs from Switzerland and Germany-Austria. Even had fewer crossbred daughters and smaller PTA declines, and his U.S. all-breed PTAs were almost at the same levels as his converted Italian and German PTAs. Pronto also had large PTA declines; his all-breed PTAs are further from his converted German-Austrian PTAs but closer to his converted Italian PTAs. Genetic correlations with the United States are 2 to 3% higher for Italy than for Germany-Austria. The all-breed model is more effective than the within-breed model for including information from crossbred daughters without bias.


Prior to May 2007, genetic evaluations in the United States compared animals within the same breed. The USDA animal model was implemented in 1989 to calculate PTA by comparing records of cows for each breed within herd, lactation group (first vs. later), and management group (calving period of 2 months or more); Holstein records were also compared within registry status. With that system, separate pedigree files, contemporary groups, and genetic bases were used by breed. Holsteins were compared with Holsteins, Jerseys with Jerseys, and so forth. Records from crossbred animals, unless they were part of a herdbook expansion program of a breed association, were not included. Daughter records for sires of other breeds such as Swedish Red, Montbeliarde, and Normande also were excluded from evaluation. Increased crossbreeding in the United States led to dairy industry requests to evaluate the genetic merit of crossbred animals that were being excluded. Therefore, the all-breed animal model evaluation was developed to include purebreds and crossbreds from all breeds together in routine genetic evaluations.

All-breed animal model

The all-breed animal model includes lactation records from 1960 to the present, including crossbred records. The total number of cows with records in the national database as of 2006 ranged from 10 to 22 million for milk, fat, protein, SCS, PL, and DPR.

Purebred status is based on an animal's heterosis coefficient, a measure of hybrid vigor. Animals with heterosis coefficients of 0 to 25% were defined as purebreds, with coefficients of 26 to 50% as backcrosses, and with coefficients of 51 to 100% as crossbreds. Breed composition of the first-lactation cow population in 2004 was 90.9% Holstein, 6.2% Jerseys, 0.8% Brown Swiss, 0.4% Guernsey, 0.3% Ayrshire, <0.1% Milking Shorthorn, 1.2% crossbred, and 0.3% backcross. Nearly all crossbred cows had one parent that was a Holstein, and contributions from other breeds were proportional to population size. The number of crossbreds has doubled during the last 3 years. Recently semen from Scandinavian Red and French breeds was imported, and the resulting daughters are nearly all crossbreds. Since 1987, over 5,000 herds had at least 1 cow coded as a crossbred, and currently 1,377 herds are coded as mixed-breed herds with more than 25% of cows of different breeds or crossbred.

Pedigrees for over 46 million dairy cattle were traced to their earliest ancestors with computer records, with a lower birth-year limit of 1950 because records for earlier ancestors had not been included in the national database. Most animals (99%) had ancestors of only one breed, but 431,000 had ancestors of more than one breed. Of those, more than 350,000 had breed compositions with less than 94% of one breed and greater than 6% of another breed because the crossbreeding occurred within the most recent four generations. The percentage of primary breed currently is reported for bulls and cows with pedigrees that contain more than one breed.

Grouping of unknown parents is needed when calculating genetic evaluations to account for changes in genetic merit across time. Unknown-parent groups in the animal model are separated by breed, pedigree path (dams of cows, sires of cows, and parents of sires), national origin (U.S. or foreign), and birth year. Groups are formed when they include at least 500 animals within a time period and at least 2,000 animals across all years. The grouping pattern is similar to that for Dutch evaluations except that those require only 40 animals per group. Larger numbers are needed for traits with lower heritability. Crossbred ancestors with no records and only one progeny are treated as known parents so that animals with records can be linked back to purebred ancestor groups.

Multiplicative factors to adjust for age, parity, and calving season prior to all-breed analysis are set to a 36-month age standard as was done for within-breed analysis. Adjusted yields are lower than mature equivalents by about 5% for Guernseys; 10% for Holsteins, Jerseys, and Ayrshires; and 15% for Brown Swiss and Milking Shorthorns. Sire breed is used to adjust crossbred records. Holstein factors are applied if the sire was crossbred or if the cow’s breed was not Ayrshire, Brown Swiss, Guernsey, Holstein, Jersey, or Milking Shorthorn. Additional age-parity-region-time factors that have been included in the animal model to account for gradual changes that have occurred after multiplicative preadjustments were implemented in 1995 are estimated as uniform effects for all-breed evaluations.

Herds and breeds differ in the variance of phenotypic records and genetic effects. Adjustment for heterogeneous variance of herds was modified for all-breed analysis. Adjustments are applied to all milk, fat, protein, and DPR records, but the estimated variance ratios are calculated only from complete records. Previously, records with a data collection rating (DCR) of 95% or higher were used, but the DCR requirement now is 92%. Some herds on a.m./p.m. testing with long test intervals previously had few or no cows included in the herd variance estimate and instead received the breed-region average variance. Records with a DCR of less than 92% still do not contribute to estimation of herd variance ratios, because separate procedures are used to adjust variances and weights for records in progress. For mixed-breed herds, variance within herd would be overestimated if no account were taken of breed differences. Variance adjustments for milk, fat, and protein previously had been based on ratios of milk variances, but variances of fat yield are used in the all-breed analysis because crossbred cows are more similar to purebreds for fat yield than for milk or protein yield. The two revisions to the adjustments for unequal herd variance may cause evaluations to change for cows with extreme yield deviations or in herds with unusually high or low variation. Variance adjustments for PL and SCS had not been used for previous within-breed evaluations and are not included in all-breed evaluation. Data for breeds other than Holstein are adjusted to make genetic variances equal to that for Holstein base cows.

Separate management groups in the within-breed evaluation were defined for registered and grade Holsteins if at least 5 cows of each type were present, whereas cows within other breeds were grouped together regardless of registry status. In the all-breed evaluation, crossbreds are grouped together with registered or grade cows to allow estimation of breed differences. Crossbred cows sired by Holstein bulls are treated as grades, but all cows sired by bulls of other breeds are treated as purebreds and grouped with purebred cows.

Heritability of yield traits in the within-breed animal model was higher for Jerseys and Brown Swiss (0.35) than for other breeds (0.30). For the all-breed model, all heritabilities are set at 0.30, and the higher effective heritability for Jerseys and Brown Swiss is achieved by adjusting their lactation-length weights upward.

Inbreeding and heterosis

The all-breed model provides more accurate evaluations because genetic interactions such as inbreeding and heterosis are included. However, selection and mating (rather than just selection) should be considered when using those evaluations. Because inbreeding and heterosis are genetic effects, PTAs are adjusted so that they measure the average genetic merit of animals when mated to the current population.

Inbreeding adjustments were introduced for within-breed evaluations in February 2005 and include a regression on inbreeding in the model as well as the regression multiplied by expected future inbreeding (EFI) in the PTA. The population used for calculation of EFI was revised and is more current. The sample size increased from 600 to 1,000 animals and now includes both heifers and cows born during the latest 4 years instead of only 3-year-old cows. This change ensures a wider genetic sample, especially for breeds with smaller populations that may have only a few popular sires in a given year. Crossbred females are excluded from the sample by requiring that the dam, maternal grandam, and maternal great-grandam all have the same breed code as the animal sampled. Thus, EFI predicts average inbreeding when mated to recent purebred animals.

Heterosis adjustments were introduced with all-breed evaluations. The model includes a regression on heterosis coefficients. When PTAs are converted to a purebred evaluation scale, the heterosis expected when an animal is mated to a recent purebred of that breed is included in the PTA.

Genetic bases

Evaluations from the all-breed model can be reported on a single base or converted to separate bases for each breed. An all-breed base was calculated using the trait average of all cows born in 2000, and then within-breed bases were calculated from the PTA of cows with a heterosis coefficient of less than 50% (that is, all crossbred and most backcross cows were excluded). For breeds with the highest percentages of crossbreds, within-breed bases under the all-breed model changed from previous within-breed bases because first- and second-generation crossbred cows were no longer included as base cows; however, cows that were 7/8 purebred or more were included. The PTA for each breed is adjusted to the within-breed base as had been done for previous within-breed evaluations.

Conversions between all-breed and within-breed bases involve both a breed average and the standard deviation (SD) ratio for traits with variance adjustment that differed by breed:

within-breed PTA = (all-breed PTA − breed average) × (breed SD/Holstein SD);

all-breed PTA = [within-breed PTA × (Holstein SD/breed SD)] + breed average.

Genetic trends

A separate correction to the calculation of parity adjustments for cows that changed herds resulted in about 3% lower genetic trend for all traits except PL. That lower trend frequently resulted in considerable change in PTAs of animals that were born in the 1960s and 1970s and less change in the opposite direction for most animals of current interest. Although the impact on bull and cow rankings is minimal, the decline in estimated genetic merit for yield traits, especially for active AI bulls, likely is disappointing for Brown Swiss, Guernsey, and Jersey breeders.

Estimated genetic trend for DPR is now more negative because estimated repeatability for DPR was increased. The previous repeatability estimate was 0.13, and the permanent-environmental variance of 0.05 was obtained by subtracting the heritability of 0.04 and the herd × sire interaction variance of 0.04. That estimate for permanent environmental variance was too low because herds often contain many paternal half-siblings, but the repeatability estimate did not include the interaction term. Estimates of genetic trend for DPR are sensitive to those parameters, and new parameter values of 0.12 for permanent environmental variance and 0.20 for repeatability improved Interbull trend tests 1 and 2. Trends for other traits were much less sensitive to repeatability changes. For cows with multiple lactation records and for bulls with daughters with multiple lactation records, reported DPR reliability decreased, and reliability of net merit also decreased slightly because of the higher repeatability. Past genetic declines in fertility were larger than originally reported, and the revised estimates are more consistent with trends reported by researchers in other countries.

Evaluation breed

Evaluation breed for cows has been added to the cow evaluation record (format 105, bytes 369–370) as has the cow's heterosis coefficient (bytes 366–368). For most animals, evaluation breed is simply the breed code in the animal's identification. However, evaluations for crossbred animals with breed code XX are reported on the base of the sire breed because that generally is the breed of interest. That breed assignment may cause some confusion because evaluations of animals from reciprocal crosses are on different bases. Animals with a breed code other than for traditionally evaluated breeds (Ayrshire, Brown Swiss, Guernsey, Holstein, Jersey, or Milking Shorthorn) also are reported on the base of the sire breed. If neither the animal's breed code nor the sire's breed code is for one of the traditional breeds, Holstein is the default evaluation breed. An exception is that animals with breed codes NR, SR, and RE (Scandinavian Reds) have evaluations reported on the Ayrshire base to be consistent with Interbull policy. Red and Whites (WW) continue to be reported on the Holstein base, and other breeds such as Montbeliarde (MO) or Lineback (LD) are also reported on the Holstein base because cow numbers are insufficient to determine separate bases. All data fields that have a base are reported on the bull's evaluation breed for format 38 and on the cow's evaluation breed for format 105 except for the cow's contribution to sire for milk, fat, and protein (format 105, bytes 324–336). The cow's contribution to sire (which is twice the cow's PTA minus the dam's PTA) is reported on the sire's evaluation breed so that a weighted average of those contributions will match the sire's evaluation.

Breeders and advertisers must be sure to compare only PTAs on the same base, which may be difficult if reciprocal crosses (for example, Jersey × Holstein and Holstein × Jersey) exist in the same herd. In the future (perhaps at the next base change), evaluations of all breeds and crossbreds could be reported on the all-breed genetic base to simplify selection.

Additional resources