EPD's - Problems and Solutions
The Problems with EPD’s and the Solutions to EPD’s
In the cattle breeding business, most breed associations calculate Expected Progeny Differences (EPD’s). Currently, EPD’s are calculated primarily for measurable characteristics related to size and carcass characteristics. EPD’s are calculated for traits such as Birth Weight (BW), Weaning Weight (WW), Intramuscular Fat (IMF), etc…
There are three major problems with EPD’s. If corrected these three problems are corrected, perhaps EPD’s could be beneficial for breeds rather than detrimental.
1.
EPD’s assume mean regression. In other words, in the absence of sufficient data, it is assumed that an animal will perform like the breed average. When a small amount of data indicates superior performance, the “mean regression” coefficient means that an individual’s EPD’s will be closer to the breed average. By far the most informative discussion of EPD’s that I have read is located in an old newsgroup discussion on the Advantage Cattle message board. You can find the discussion at http://www.advantagecattle.com/forum/topic.asp?TOPIC_ID=301 .
2. “Accuracy” is double-counted in the presentation and application of EPD’s. The argument for including mean regression in EPD’s is to account for a lack of accuracy. As explained by Dr. Richard Willham, one of the creators of EPD’s,
“EPDs are predictions of FUTURE progeny compared to the average animals in the breed. EPDs are regressed toward the average for heritability and NUMBERS of records by themselves and related animals. Thus, the EPDs of animals can be fairly compared even though the amount of information for each is different. So, study the EPDs as they are and select those with the desired values. This means if a young bull with a low accuracy beats an older bull with a higher accuracy, select the bull on EPDs. Do not try to consider the accuracy because this has already been accounted for in the EPDs. As statisticians, we have included accuracy, but really did not need to do so. As I can see, it has bothered some of you breeders.”
Dr. Willham is well aware that breeds publish accuracy figures for EPD’s that are already formulated to adjusted for accuracy. The method by which EPD’s adjust for accuracy is Problem #1, mean reversion. EPD’s were created by statisticians, and statisticians approached EPD’s with the cautious approach as assuming mean reversion until statistically significant variances appeared. Breeders have now spent enough time studied EPD’s and inputs to be able to “honestly” create statistically significant variances. A prime example is a breeder having a contemporary group of 30 cows, five of which he knows are outstanding, five of which should be culled, and twenty of which he has some opinion about based upon lineage, potential outcross heterosis, etc, but does not differentiate as good or bad in his own mind. The simple exercise is to take the five bad cows and ten mediocre cows with the highest EPD’s and breed them to the top performing AI bull in the country. At the same time, breed your own most promising bull to the five best cows and the five mediocre cows with the worst EPDs. Have four other breeders follow the same program. The contemporary group legitimately and honestly will consist of 30 calves among two bulls in a total of five herds. If the breeders are worth their salt, even a considerably under-performing bull will become “proven” as an outlier versus the breeds’ top proven bull over a couple of years.
Those in favor of the current program have to argue that either breeders are not smart enough to figure this out, that breeders lack the organization and resources to carry out such a scheme, or that the breeders are incapable of identify their five best cows and five worst cows from a group of 30.
3.EPDs do not adequately account for inbreeding coefficients. In fact, I do not believe that Hereford EPD’s are adjusted at all for homozygoity. It is well-known among cattle breeders that there is such a thing at “inbreeding depression.” The corollary, heterosis (“hybrid vigor”), is understood by almost all breeders. “Inbreeding depression” is merely a method of explaining that animals with little variance in genetics exhibit less heterosis.
4.EPD’s do not account for Disposition, Vigor, Conformation, Doablity, and Frame Score.
The Solution
Problems 1 & 2 – Mean Regression and Accuracy
I propose to cure the problem of legitimate data manipulation and double-counting accuracy by augmenting the assumption of mean reversion. The style of data manipulation as described above exists because breeders possess more knowledge and data than is asked for by breed associations. Breeders are aware of the reasons that outliers exist in contemporary groups and whether the outlying performance is likely to recur. Keep in mind, I am not talking about lying, and I am not referring to breeders inaccurately reporting contemporary groups by counting sick calves among well ones, etc. I am talking about a breeder recognizing that they have enough inputs after only 3-4 calves know that a particular cow will ratio well with a particular bull line, especially compared to her EPD’s. Other cows will ration poorly, especially compared to their EPD’s. I am a relatively novice (perhaps bordering on intermediate breeder), and I have this knowledge among cows in my herd.
The augmentation that I suggest is to let breeders set EPD’s for 10% of the cattle in their herd each year. The accuracy assigned to the “new” EPD would be (0.25 + 0.25*(previous reported accuracy)). The maximum or minimum value that could be assigned would be the top or bottom 1% for any given trait.
In actuality, the breed would report final EPD’s twice a year as it does now. Breeders would then submit their changes. The changes would be treated as if exactly enough data had been submit through new progeny to create the newly assigned accuracies. Then, the EPD program would be rerun, and all other non-adjusted animals data would change according to the new data, and “Final Adjsuted EPD’s” would be reported.
I can hear the comments already, “Heresy! Have breeders openly encouraged to manipulate their own EPD data! Wouldn’t that destroy the entire purpose of EPD’s by biasing them?” My answer is to reread my description of the problem before. The problem is that EPD’s are already biased and inaccurate, and any breeder that cares to do so can manipulate EPD’s to his advantage within the existing system and within the confines of honest and accurate reporting. Breeders who wish to lie are already free to do so in the data that they report. Allowing an individual breeder to adjust EPD’s up and down and then re-running the calculations with the accuracies on the adjusted numbers being between 0.25 and 0.5 would likely improve the data integrity, EPD accuracy, and ultimately the overall quality of the breed. The reason that breed quality would improve is that a breeder could adjust a low-EPD, high performing cow to the EPD’s he believe are accurate and retain her in the herd. Over time, her progeny would cause EPD’s to fluctuate according to performance.
In addition to the rule of only 10% of the herd having EPD changes in any one year, there would likely need to be a couple of other simple ground rules. First would be that EPD’s could only be changed on wholly-owned animals that had reported at least two progeny over two breeding seasons with weaning weights under the current owner. Second would be that one owner could only adjust the EPD’s of a given bovine twice in that bovine’s life, and all changes would have to be made to living bovines expected to produce additional progeny. In reality, the two rules would mean that most EPD adjustments would be made to active dams, although a few older but lesser proven bulls would likely also be candidates for adjustment.
Problem 3 - Heterosis
Two adjustments would be necessary to account for heterosis, an adjustment to accuracies and an adjustment to the actual numerical values. First, accuracies of EPD’s would need to be adjusted for prepotency. An overly simplistic numerical would to do this would be to weight all data inputs by a bovine’s inbreeding coefficient, preferably using at least a 12 generation pedigree to calculate the inbreeding coefficient. As for the adjustment to numerical values, the Line 1’s inside Miles City and once used as outcrosses probably provide data to quantify the impact of heterosis (as defined by the inbreeding coefficient) on BW, WW, YW, and all other objectively measured traits. A “Heterosis Discount”, once quantified, needs to be universally applied to EPDs.
Problem 4 – Important Cattle Traits Not Receiving Enough Attention
Off hand, I can think of Five important criteria that I select for that are not currently accounted for in EPD’s: Disposition, Vigor, Conformation, Doability, and Frame Score. Good business practice generally involves establishing a plan, establishing criteria, setting objectives, measuring criteria relative to objectives, and persistent pursuit of objectives.
I believe that many breeders have chose to maximize existing EPD’s because they are the only objective benchmarks on which they are publicly compared outside of show winnings. Breeders are advised to select for Disposition, Vigor, Conformation, Doability, and Frame Score in addition to existing EPD’s. Most breeders have the concepts in mind when selecting, but the pull of a “curve-bending” EPD outweighs a slight conformation, vigor, or doability fault in some cases. Prior changes to measured EPD’s and accuracies would reduce mongrelization of the breed and create rewards for prepotency. Five new EPD’s, however, would help in returning the focus of breeders to good, practical cattle.
EPD’s calculations for each of the new categories would be simple. Breeders would assign a 0-5 score for each of the 5 new EPD’s on all calves at weaning and to dams and sires at the end of each breeding season when new calves are reported to the AHA. Actual EPD’s would be on a different scale, perhaps 0-100, and would be based upon historic reported data, data of relatives, inbreeding coefficient, and objective correlated factors. For example, consistently producing 17-year old dams would receive high scores for conformation and doability. Consistently and longevity would have significantly influence over scores.
Impact of the New EPD Scheme
Assuming that breeders agreed with the EPD changes and utilized them to at least the degree current EPD’s are utilized, four behavioral changes should be observed:
•Breeders would pay more attention to their cattle in the field rather than EPD’s because breeders would have a direct role in setting EPD’s based upon their own observations that would be understandable.
•Breeders would observe and attempt to quantify on a relative basis the important attributes of Disposition, Vigor, Conformation, Doability, and Frame Score.
•More input from men in the field regarding EPD’s of measurable traits such as BW, WW and YW should lead to greater accuracy of EPD’s and more buy-in from breeders. A virtuous cycle and increased confidence in the numbers should emerge.
•Adjustments for in-breed heterosis and prepotency’s impact on accuracy would reduce the genetic variability of the aggregate Hereford herd over time and would like stimulate the creation of line-breeding programs, some of which would contribute to the progress of the breed and others which would not contribute genetically but would contribute to the knowledge of Hereford breeders.
With the aforementioned modifications, I believe that wise commercial cattlemen would be able to more effectively utilize Hereford in their herds. Since EPD’s would be tracking the attributes that commercial cattle care about, ultimately Herefords would regain market share for two reasons. First, commercial cattlemen would have more information on the traits they care about, especially when compared to other breeds. Second, by focusing on the traits that create great cattle, by giving breeders a greater role in EPD calculations, and by creating incentives for genetic concentration of the Hereford breed, Hereford cattle should steadily migrate towards the kind of cattle that made beef in America great in the first place.
