BLUP, What Does It Mean? Is It a Valid Prediction Method?
BLUP is the acronym for Best Linear Unbiased Prediction. The BLUP is a method of statistical analysis and estimation. Numerical scores are given to traits and compiled as predictions for future use.
Is It a Valid Prediction Method?
"BLUP is a mathematical model that has been used for predicting future
generations in certain agricultural crops - I seem to remember reading
examples where it's been used to predict protein content in soybeans,
butterfat in cow's milk, the mature height of certain coniferous
trees...that sort of thing. Simple traits that can accurately and
objectively be measured, and possibly predicted.
For BLUP to work, the following all have to be true. The characteristic to
be monitored/predicted has to be simple - only one trait or one goal can be
predicted in a model. It has to be something that can be objectively
measured with excellent accuracy - like butterfat in cow's milk. And you
can only measure a trait that is known to be heritable - we assume that
Elsie's calf will grow up and will produce butterfat somewhat affected by
Elsie's own genetic tendencies. And, it has to be applied to huge numbers
of the animal/plant studied, in order to be statistically meaningful - by
huge numbers, normally maybe hundreds of thousands of cases - but
multiple-millions of individuals would be better.
BLUP, when applied to Icelandics, meets none of those criteria. The input
is evaluation scores. We all know that evaluation scores are affected by
who's trained the horse, who's ridden the horse, where the rider sit on the
horse's back, what kind of shoes they use, how willing they are to push the
horse, and how hard, what kind of nosebands, bell boots...and none of these
things are heritable. Strike one.
And, of 100,000-200,000 Icelandics
worldwide, only a fraction are evaluated - nowhere nearly the number of
individuals needed to attain statistical accuracy. Strike two.
And - are
evaluations scores "simple" and are they objective - like measuring the
protein content of soybeans? Do you put the same priorities on what makes
a good riding horse as I do? Is measuring a good riding horse as objective
as measuring butterfat? No way.
It's not even applied towards predicting
one single gait - which alone is probably too complicated for BLUP to work
for, since your idea of a "good trot" might not be the same as mine - and
that's not even considering the different priorities different people put on
temperament, friendliness, intelligence, conformation... The traits we
value in horses are completely subjective, and you can't apply a simple,
objective model to a complicated, subjective being. Strike three.
I e-mailed a professor at an Agricultural College in Canada who had written
an Agricultural genetics textbook, just to see if I was missing anything.
He confirmed that I wasn't. I can probably still find the e-mails, if
anyone is interested. I also bought his book. I also requested information
from the guy in Iceland who is the "father" of BLUP as applied in these
horses - he never responded. In this case, I do feel like I can make these
statements with confidence as I have.
The model just won't work on mammals - they
are WAY too complicated. Soybeans maybe - if you just apply it to one
It will not work when
applied to characteristics that are subjective (as in "willingness") in
characteristics that are not known to be inheritable, or if the samples are
not large enough. And it certainly can't be used to predict multiple traits, and
there are certainly many desirable and not-so-desirable traits in any
mammal...My horses are much more complicated than soybeans."
The theory behind BLUP is comparatively complex, and the BLUP method demands a lot of calculations.
The following example was given:
>>I've been researching genetics with respect to some giant breed dogs and have
seen pedigrees where (with respect to hip problems where 1 = no problem, 3 =
some problems) a perfect hipped dog mates with another
perfect hipped dog and produce a puppy with some hip problems, and then to
confuse things more, at least for me . . . a #1 dog is mated with a #3 dog and
the puppy is a #1.
I think BLUP is fun from a mathematical standpoint and I enjoy reading my
horoscope too but when it comes right down to it, really how accurate is it???<<
Which resulted in this response:
>>In the example you gave, you were looking at one particular trait and
the link wasn't quite as predictable as you might think. BLUP looks
at many, many characteristics. From what I've read, or more
accurately, what I haven't been able to find, I've come to think your
comparison to horoscopes is pretty appropriate. Funny, yes, but sad
too. I have looked for the sound science behind BLUP as used by
Icelandic horses and I can't find it so far. For a mathematical
model to work, the input data has to be 100% relevant. If the data
is not relevant, no matter how sound the formula and / or model, the
results are skewed. The more parameters with less relevance, the
more skewed, to the point of becoming useless. The expression I
learned in my first year of Computer Science was "GIGO" : garbage in
= garbage out. My first engineering mentor taught me to question the
relevance of any formulas or studies.
Many of the input parameters to BLUP are very subjective and not
scientifically definable, unlike Hip Dysplasia, which has some
established guidelines in its diagnosis. Has anyone ever seen a
established scientific definition of "character?" Is there a
universally accepted equine IQ test? Those are very subjective
descriptions, certainly not anywhere near 100% definable. Is there
an absolute definition of "pretty head?" How about the influence of
a rider on the evaluation of gaits? Riders aren't hereditary. And
horses are allowed to have slightly longer feet with shoes when they
are evaluated than is considered appropriate by many trained American
farriers. Shoes and trimming are not inheritable traits. In fact,
the use of shoes in evaluating breeding horses could even push the
breed towards less natural gaitedness. Genetics themselves are
complicated enough, but when you let factors such as handling,
training and management come into the equation, any value that might
possibly have been in the BLUP theory is thrown out the window.
I've sent out questions to some in the Icelandic community to show me
the independent scientific organizations who have validated that BLUP
is appropriate for evaluating future breeding potential. I haven't
yet got that request answered. There are Schools of Genetics at many
universities, but I can't find published research on BLUP as its
being used with Icelandic Horses. If any independent group with
valid scientific credentials does indicate that the BLUP model itself
is appropriate, then the next step would be to determine if the input
parameters and data being used are valid. That hasn't been answered
I went into this assuming that BLUP would be relevant for my tiny
breeding program. However, as I've searched for substantiating
background information, I've instead come to feel like I'm on a snipe
hunt. My original questions arose from wondering why the
Thoroughbred racing industry doesn't use BLUP. Nowhere is there more
motivation to improve breeding predictability. BLUP is not used by
other breeds in the USA. Is the Icelandic world ahead of the curve -
or following a few without questioning? I'd love for someone to show
me how BLUP is more relevant than horoscopes.<<
Along with this followup response:
I have another problem with BLUP. When using BLUP the "goal" is to produce
horses that will do well at the Evaluations. That is not the stated goal but
determining BLUP based the horses' performance at the evaluations, that is
what becomes the mathematical goal. (I hope that makes sense). Just how
relevant is this to an individual breeder? For myself - some things are
relevant (i.e. good conformation). Some things are not (i.e. lift). Some
things are not measured at all (i.e. ability to be a sensible trail horse).
The problem is that BLUP gives you one number (Katina x Gymir = 110%). Does
this mean that my foal will have 110% better chance of being a successful
trail horse (when compared to the average Icelandic). Absolutely not! It
means that my foal has a 110% better chance of being successful at the
Evaluations than the average Icelandic. (If the mathmatics are accurate).
Is that important to me? No.
They use a similar system in dairy cows - the characteristics that are
important and measurable are used (i.e. milk production, solids) and these
are weighted depending upon the "hereditablity" of that trait (and the
influence of a particular farm is taken into account (the cow is compared to
other cows in her herd)) and then the statisticians wave their magic wands
and - poof - a highly successful program is launched.
BLUP measures things that are not objective and I don't know that the actual
"hereditablity" of the traits that are being measured is known. (You need a
huge number of animals and very, very accurate rcord keeping). Some trait
are determined almost entirely by genetics (i.e. coat clour) and others are
not (how does "character" get measured?). The higher the number of traits
being measured - the less useful the number is.<<
>>It's a huge jump between
mathematical theory and practice and that's what I'm asking about. I
understand that BLUP may have been used to predict some simple traits in
both livestock and field crops regarding traits that are known without doubt
to have a genetic connection. However, Icelandic evaluation scores take
into account many, many characteristics, and some are also influenced by
training and management and the strength of a genetic link is not always
established. Many evaluation criteria have both genetic and management
links, but management is not hereditary. What I haven't been able to find
is any research that suggests that the Icelandic breeding evaluations are
appropriate to use as data into this model. Can anyone point me to a study
by a well known independent scientific organization (a university, etc.)
that shows where shows a strong cause and effect relationship between the
data input and the output related to horse breeding? Despite many
questions, and much searching on the Internet I can't find it.
There are a lot of statistical models that are valid when used in the
appropriate situation, but if the correct models is not chosen for the
appropriate situation and the data is not determined to be purely related to
the traits being predicted, the results are very suspect if not totally
So far, I can't even find out why this particular model was chosen, who
decided that it was the appropriate model to use, and if the model is valid
when so many traits are being considered, or who validated that these input
parameters can be successfully used for predicting futures breedings. In
fact, I can't even find what parts of the evaluation numbers are used as
field data for determining BLUP. Most research methods and studies have
this sort of background documentation available to prove their merit. My
engineering training makes me suspicious as to why more background
documentation details are not available. If this is valid science, then why
aren't other breeds using it?<<
Adding to the discussion:
>>First let me qualify this post by saying: I'm not a horse breeder in
any capacity. So, that will explain my ignorance on this subject. : )
However, it is my sincerest hope that the breeding of Icelandic horses
will always include the variety of horses we find available now.
I tend not to like "systems". Mother nature does not seem to like
them either. And, for the most part, that's who REALLY gave us the
Icelandic horse we know and love. Man has only been meddling in the mix
for the last 80 years or so.
A system like BLUP and to some extent, the evaluations, is geared
towards one "perfect type" of horse which may or may not be ideally suited
to all kinds of riding and riders who fancy Icelandic horses. My prayer
is that there will be enough "maverick" breeders out there who will not be
swayed by the fashion of the show ring in terms of turning out the "perfect"
Icelandic and instead, breed what they like. This will ultimately keep
the variety of the breed in tact. Yes, it will also likely produce some
real toads, but if Iceland is still culling 25% of it's foal herd, I don't
think this idea is any worse.
While the current trend might be for elegant, refined, narrow bodied,
long-legged, highly animated, miniature Saddlebreds or Fresian look-a-likes,
there will still be smaller breeders offering something that resembles a
horse that looks like it survived the millenium and will likely make it to
the next. Give me a usin' pony over a hothouse flower any day.<<
>>The only other breed for which I've seen any BLUP references is the Swedish
Warmblood. I couldn't find any details on that breed either, but I know its
numbers are pretty small too, especially when you compare to the large
numbers of American breeds like AQHA, TB, Apps, etc.<<
In regard to looking at the total number, or the numbers assigned to individual traits:
>>I absolutely agree that each individual trait should be considered,
not a total score, with or without a formal evaluation, and I think
that was the point.
In an ideal scenario, one would hope that the sire's recessive traits
would align with the mare's superior dominant traits, and vice
versa. We use the best judgement we can, pick the breed breeding
candidates we can, but we still have to keep our fingers crossed for
a good roll of the genetic dice, followed by a healthy pregnancy,
normal delivery, and then hope we pick the best practices for
raising, handling, and training each individual foal. It's a long
and complicated process and I can't see how it can be reduced to a
single number, or even a series of numbers. There has to be a little
art and intuition used with some genuinely sound science.<<
>>Great post. You have re-stated the same concerns that other vets,
mathematicians, and people trained in statistical and scientific research
have pointed out when I've asked them about this issue. I took the liberty
of re-listing six of your points below.
 1. Some things are not measured at all (i.e. ability to be a sensible trail
 2. If the mathematics are accurate
 3. BLUP measures things that are not objective and I don't know that the
actual "hereditability" of the traits that are being measured is known.
 4. Some traits are determined almost entirely by genetics (i.e. coat color)
and others are not (how does "character" get measured?).
 5. The higher the number of traits being measured - the less useful the
 6. When using BLUP the "goal" is to produce horses that will do well at the
Evaluations. That is not the stated goal but determining BLUP based the
horses' performance at the evaluations, that is what becomes the
With so many basic premises of this model in question, it can't be taken
seriously. BLUP as used with Icelandic horses seems to be pretty darned
flawed from a scientific point of view from the beginning.<<
In regard to a stallion and his BLUP score and individual traits:
>>You may possibly have found a good example, but I'd have to have to ask more
questions to say. I think we agree on the unimportance of BLUPs, so this
isn't to you personally, just another attempt to make people aware of
potential pitfalls of using a questionable method.
There could be very little link between BLUPs and pace here. For example,
to draw the conclusion that HE and he alone passed on the pace, you'd have
to carefully study the mares he was bred to and their genes - each of them
contributed 50% to the offspring too. Is it possible that some (many?) mare
owners purposely selected him to sire offspring to their mares who had
strong pace? For example, might a breeder be tempted to breed a good mare
with a slightly stronger pace than he's comfortable with to a stallion that
scored a 9.0 on trot and tolt, but only 8.0 on pace? If that was the case
very often, the pace might be showing up DESPITE his genes instead of
BECAUSE of them...? What were these breeders' goals - strong pace,
four-gaits, or an even distribution of five gaits? Did they all have the
same goal? Different horse breeders can have very different goals, so it's
not like with soybeans where for instance, this year's genetic manipulation
might aim for a given protein range and next year, draught-resistance. But
what really concerns me (statistically anyway) is that only 28% of his
offspring were evaluated, even thought that is probably a pretty high
percentage for any stallion in the real world. What about the other 72%?
Any hard-pacers or hard-trotters among those 492? That's a lot more of his
offspring that we know nothing about compared to the ones we do know
If we could show that many or most of the mares he bred had no or weak pace,
then it would appear more likely that he passed on pace more strongly than
he showed himself.
A good study shows how extraneous data, coincidences, etc. are explained.
Isn't it likely that the people who had his offspring evaluated purposely
chose the "best" of his offspring and possibly selected more offspring of a
given type? (I'm not going into the subjectivity of "best", but assume
we're talking "best" relating to "normal" judging of traits scored in
evaluations that are evenly applied everywhere.) If you only chose to study
what might the "cream of the crop" and that "cream" is only 28% of the
available data, and many or most of those owners were after a particular
type horse, then the data is again very skewed from the get-go. Further,
we'd have to know if the owners of the offspring tested were more (or less)
likely to use any special (harsher or less harsh) training, bitting,
trimming, or shoeing methods, to see to see the role training and
environment might have played.
I confess don't understand the whole idea of 100% accuracy relating to
BLUPs. If the BLUP is determined by feeding in the evaluations of his
offspring, wouldn't the results always be 100% just by definition? I
haven't seen any definition of that "percent accuracy" relating to Icelandic
BLUPs and again, since that's not readily available, that omission of
information concerns me.
I don't know enough about how evaluations are actually conducted to have a
strong opinion on them yet. However, I can at least accept and respect what
I understand they claim to be: a measurement of how that particular horse
looked and/or performed on that particular day, by stated rules (even if I
question some of the rules), hopefully by judges that are not overly
subjective or biased. But it's really pushing it to try to predict the
future by these scores by projecting them into BLUPs.
We'd all like as many tools as possible when making breeding decisions, but
I'm afraid BLUP could actually be misleading in more situations than it's
helpful. That's not a tool I want to use.<<
>>I spent many, many an hour learning all about the system for breeding dairy
cows (BS and MS in Animal Science, large animal track through vet school);
finally got some use out of it!
The other point to consider is that in dairy cows the goals are clear,
universally accepted, easily measured, influenced strongly by genetics and
few in number - (more milk/higher solids/can't remember the rest - but it's
a small number). There are a lot of cows in the measurements (zillions,
more or less) This system has been highly effective because of those
The BLUP system has none of these attributes.
I don't want a system for producing Icelandics that will do well at the
evaluations to be effective because I don't think that I want that horse.
We all have different horse preferences and I think it's great that the
Icelandics come in many different packages. Parts of
the evaluation are useful for matching up mares and stallions - however - I
question the validity of the final score.<<
>>BLUP - the actual % that is given as
a final score: IMO that % score is not useful and may, indeed lead some
people down a path they did not intend to go.<<