Contention point #10:   The newest generation is always the best

Holstein type summaries have built within them an aging factor for scores based upon the year of scoring.   In this way, the higher scores of older bulls that might still show up as contemporary sires are regressed so as not to hurt the ability of current sampling sires to generate positive type proofs.   Why did we need to do this?    It seems that when the SET programs (allowing grade heifers to be scored in young sire herds to get type proofs on sampling sires) began, the average two year old scored 78 points.   After a couple decades, average had dropped to 76 points.   There are individual AI systems whose sampling herd heifers average maybe 74 points.   Without this adjustment factor, hardly any new bull would be "plus" for type compared to the standards of the classification system, and every breeder would be paying $100 or more per ampule to scrounge semen on great old bulls.   Catchall term for these kind of adjustments is called "merit of mates".    The higher merit we assume the mate has, the less credit we give the sire for what he accomplishes in that mating.   That's why all famous old bulls end up "minus" in those fancy sale catalogues of famous herds with high BAA%s.

Wait, you say, are today's classifiers just harder on cows than they used to be?   Not a chance,   They get a monthly report showing how their distribution of scores compares to all classifiers.   They are expected to find 2% of all cows "Excellent", 13% "Very Good", 35% "Good Plus", 35% "Good", 13% "Fair" and 2% "Poor", approximately, no matter what they are actually seeing in your barns.   So they find them where they can, to avoid getting called out of line.  

Do you think AI studs would keep paying $3000 to $10,000 per bull for type proofs if all their bulls came out minus for type?   Not a chance.   Design the system so half the bulls are always plus, and everyone stays happy (except the poor guy using all those sampling sires, who ends up with the 55% cull rate).    To be really sure, we also adjust the score of any heifer who scored 52, or 55, or 58, up to a minimum of 65 points for sire evaluation purposes only.    Obviously, it was bad management, not genetics, that makes a heifer look that bad...  right??   Not a high pedigreed AI young sire!!

Contention point #11:  AI sired contemporaries are more accurate  

My friend Joe manages his Uncle's commercial purebred dairy in northern Michigan.   His veterinarian is also a purebred Holstein breeder with a unique, high BAA% type homebred herd. and Joe will often borrow one of Doc's bulls to use for "cleanup" after a couple AI services.   The herd also participates in AI young sire sampling.   One day the classifier shows up with a list of heifers to classify, in order to complete a type summary on an AI sampling sire from whom they had two daughters.   These two AI heifers scored 76 and 77 points.    The system requires seven ID'd contemporaries for each daughter, so a list of other cows to score was included.   These other cows ended up averaging 75 points, which was going to put this bull on track to look like a +1.50 to +2.00 bull.

Wait a minute,,, Joe asks, "half of these contemporary cows are already fresh second calf.   Why not use the actual age herdmates that grew up with these heifers?"    No can do, says the classifier.   Why not? asks Joe.   Because they are sired by a non-AI bull (Doc's cleanup bull, actually a bull who was used in three herds and would have enough daughters in total to get an evaluation if given a chance).   What if I pay you to score them?    OK, so they scored a dozen daughters of Doc's bull and those end up averaging 80 points, which would make the AI sampling sire a -2.50 bull.   However, even though he was a registered bull, because he was not an AI bull (ie, not entered in the SET program as needing type data), they would not be used for type evaluation purposes as contemporaries... even though they fit the parity definitions on all possible points:  (1) born same farm, (2) same age, (3) calve same season, (4) milked same environment.    

If you are wondering why high Rel% sires (Whittier Farms Ned Boy was one of many examples of this phenomenon) can be +2.00 PDT on first sampling daughters, and end up -0.50 PDT at 99% Rel on first use as a "proven sire", this is why.   Captive AI sampling programs create subpopulations of contemporaries that do not always match the attained breed average.    This is why a bull whose daughters average 74 points (-3.00 to the breed average) can be a +2.00 "Type" bull on the sire summary: the herdmates that AI system produces are even worse.      Caveat emptor.

Contention point #12:   Incomplete lactations extended to 305d helps us  

Everybody wants good news fast.    Hot new bull's daughters are really milking, why wait for them to finish lactation, let's project those 60-90 day records out to 305d and get an evaluation!

OK, I'll buy that we have decades of data on which to develop extension factors to estimate 305d yield.   I also know that the evaluations will get adjusted once those heifers actually make their 305 days.   They might change a little, not a lot.    The point is, the extension factors were developed from completed lactations of cows capable of surviving to complete the lactation.

But why, if a heifer calves and dumps her udder, gets a test in after 15 days before she gets shipped, do we extend that lactation out to 305 days and include it in an evaluation as a 305 day effort?   The cow is not there-- she was culled.   She only had maybe 100 days in lactation, not 305 days.   That is all her sire should get credited.   If he made a lot of such daughters, we need to know it now, not three years later after 250,000 straws of semen were used on the basis of "estimated" PTA values.

Contention point #13:  Productive Life ratings answer the questions on longevity

USDA bases all PTA calculations upon the first five lactations, if available.   That means calvings through six years of age, which is considered maturity (Jerseys mature a bit quicker, in their five year old lactation).   Lactations made at later ages have no bearing upon any genetic evaluations.

The biology of bovines suggest a 20 year life span in nature, and experience tells us there are cows productive up to fifteen years of age in commercial dairy settings, just that they have become rarer as the industry has progressed.   In any other specie, if your goal was to increase length of life, research says to pick longer-life ancestors.   In dairy, we only measure the first five years of productivity, or in other words, we truncate measurement to the first half of the potential productive lifespan only.

I think, as a result, PTA Productive Life's usefulness is limited to telling us which sires are the worst for productive life length (ie, let's avoid bulls with a "minus" PL rating).   But if we are trying to breed in a positive manner to extend the productive life length of a herd of cows, we don't have useful data.

The best sires of Productive Life-- those bulls, like Paclamar Bootmaker (Holstein) or Schultz Performing Legend (Jersey), whose daughters would live for ever-- look no different than any other bull who has a few daughters reach a fifth lactation.

Contention point #14:  PTAs cannot be blamed for inbreeding effects

The "Animal Model" system is front-end loaded with Parent Averages and weighted sibling PTAs.   Since "Animal Model" was adopted with the 1988 sire summary (1986 genetic base) we have seen an increase in the rate of increase in inbreeding coefficient in every major (ie, AI focused) dairy breed.

It is logical to assume some increase in "ibc" from genetic selection.   After all, the whole process is one of discarding lower value mates and extending the use of higher value animals.   This means we will have more pedigree relationships as we accumulate selection generations.   In fact, until recently we even valued "linebreeding" as a positive tool in genetic selection, because we believed it would maintain the influence of a prepotent ancestor into later generations.

The trouble with "Animal Model" is that we continue to give the pedigree weight after we have data from progeny.   The only purpose a bull has is to give us milkable daughters.   The fact his grandam was a supercow has no meaning, if he is not transmitting her characteristics in total to his offspring, and unless he is a transsexual clone, it is impossible for him to be that.   The whole purpose of sampling and evaluation is to determine if the pedigree potential is transmitted in reality.

Because sire analysts recognize pedigree factors continue to contribute to sire PTA values at least to the levels of first sampling Rel%s, they are reluctant to sample any meaningful volume of "outcross" sires.   This hurts dairymen in two ways: (1) they have less sire variety available, (2) a large number of sires who look superior on first evaluations will not maintain their relative ranking after wide usage (ie, once they get enough AI offspring to cancel out all the pedigree inputs).

Our "inbreeding effects" are in fact the fault of using overevaluated sires.   If a bull cannot live up to his parent average, he is by definition inferior to his parentage generation.   This may be in production, in type, or in something less obvious and more crucial, like stamina, fertility, immunity, ease of calving, or whatever adds up to the package of mediocrity his offspring represent.   Multiply this effect of over evaluation of sires over several generations, and these "lack of vigor" traits will accumulate, just as the inbreeding coefficients escalate.   You can call it "inbreeding depression", or you can call it "selection depression" (the Dutch conclusion after a research published in 1996).   Either way, our geneticists are in love with a faulty system, and have little incentive to scrap it in favor of something new, because they promised the AI industry something that would "take the fluctuation out of bull proofs".

Contention point #15:  Composite indexes are functional for rapid genetic gains

My confusion relates to trait heritabilities.   Say you take six traits that individually appear to correlate with current economic needs, some are yield traits (ex PTA Protein), some are physiology traits (ex SCC), some are linear type traits (PTA Foot Angle).    They might have individual heritabilities of 10% to 30%.   We weight each trait by its "relative economic importance" and combine them all into some composite index, like "Net Merit $" or "TPI" and proceed to rank sires on that basis as everybody is currently doing so enthusiastically around the world.

In a multiple trait selection scheme, if I am using sires at random (as is implied by a composite ranking where you start at the top of the list), is not the "heritability" of the composite a multiplicand of all the individual trait heritabilities?   Therefore it might be 30% x 20% x 15% x 12% x 10% in a five-trait combination = .0108% composite heritability???

Here is another point of confusion.   Say we have an eight-trait composite that is based upon our local dairy economic realities.   Of course, if all eight of these traits relate to our ability to breed profitable and productive cattle, would we not want a "positive" value for each of the individual traits?   So why (as seems to be true so often in composite sire rankings) will we put a sire at or near the top of the list on the composite index, who is clearly negative for one or two of those eight priority traits??   Is not the point of all the research that, if a cow's udder falls off, she is no longer productive, but a cull?  Why then is her sire given the enconium as "number one TPI bull" in place of a bull who actually sires strong udder traits??   How do I continue to produce pounds of Protein from a blown udder??!!

Contention point #16:  Traits of low heritabilities should be ranked on parent averages longer

This has never made much sense to me.   We have geneticists now suggesting we need sampling sizes of 200 daughters to have the level of accuracy we need (currently, commercial AI sampling systems generate anywhere from 50 to 100 daughters).   This is because (a) "proofs" fluctuate too much from "sampling use" generation to "public use" generation, (b) some of the more important functional trait areas, feet and legs for example, are lower in heritability [as measured], requiring larger numbers of offspring to achieve "accuracy".

However, when you look at the evaluation model methodology, a curious fact emerges:  when calculating a trait score, more "pedigree" is included on a "low" heritability trait than on a high heritability trait, at a given sampling size.   Is not the "pedigree effect" the least reliable source for information on any trait that has a LOW heritability?   Who do we inherit from, other than our parents??   Why do we continue to depend on an unreliable [low heritability] averaging of parents, when we have the actual offspring in front of us???

The major problem in fact with trait comparisons of a sire offspring group vs unrelated contemporary group is that very absence of pedigree relationship.   This goes beyond the basic problem of defining the "accurate" way to" measure" a foot and leg structure, such that the true nature of genetic transmission for mobility is defined by the evaluation model (obviously, with feet and legs being a three decade point of contention in breeding, we still do not have the right measurement).    By measuring the change that occurred within the mating (ie, dam to daughter) you can tell which direction the sire took the trait, and over a normal sampling size, have accuracy,    Without that knowledge, just comparing cows to unrelated herdmates in fact is probably just random number generation, and a comparison of "parent averages" between the sire evaluated and the sires of contemporaries would likely produce the same data we actually see as his "evaluation".  

Some conclusions and suggestions

What would be wrong with going back to a daughter-dam comparison for sire evaluation, but instead of comparing "pounds", which inserts a management bias, we compare percent relative values? 

For example, we sample bull "A" and his resulting list of daughters produces at 102% Relative Value (for whatever yield you are measuring, milk, fat, protein).   The dams of those daughters averaged at 96% Relative Value.   That gives the bull a +6% rating for yield.

Likewise, we sampled bull "B" at the same time, his resulting list of daughters produces at 103% Rel Value, against dams that were 101% Rel Value.   That gives this bull a +2% rating for yield.    On this basis, bull "A" is superior to bull "B" but both offer some level of improvement in mating.

In comparison, we sampled bull "C" as a highly touted super sampler at a premium price, used him on better cows, and his resulting list of daughters produces at 105% Relative Value (which, under current evaluation methodology, makes him rank higher than bull A or B) --- yet because the dams of these daughters were actually 107% Relative Value, this bull should have a -2% rating for yield.

Using percents cancels the biases both related to management improvements over time, and those differences in management level from one herd to another.    Likewise use of Relative Value implies the in-herd merit of the mates, not based upon historical pedigree, but upon actual performance.   I think 99% of all dairymen get paid on actual production (the other 1% sell bulls to AI studs).   Genetic evaluation should be based upon their needs, not upon preserving a gravy train into the future for a small group of "elite" breed/propogators. 

If a composite index ranking is desired, why do we not first translate all trait values into a percentile ranking from the total population, and then only rank sires who are at the 50 percentile (maintenance) level or better on all the traits selected for the economic index?   Your goal would be the sire who has the highest average percentile rank across all traits considered:

Bull "A"    90% pr   95% bf   70% SCC   95%UDC   85%FLC   80% DPR   =  97.5 composite      Bull "B"    95% pr   80% bf   90% SCC   80%UDC   90%FLC   85% DPR   =  86.7 composite     Bull "C"    99% pr   95% bf   50% SCC   55%UDC   75%FLC   55% DPR   =  71.5 composite     Bull "D"    50% pr   75% bf   95% SCC   90%UDC   99%FLC   90% DPR   =  83.2 composite

Better yet, why not give up the idea of broad ranking indexes altogether?   Until we do, the "Animal Model" inclusion of pedigree/parent averages into evaluation values will continue to reduce the sire lines available in AI (and the cow lines accessed by AI) to a level where "inbreeding" really will be the problem we are being led to believe it could be.