Special Forces & Dogs: A Clone In Time Saves Nine?

Decision analysis of whether cloning Special Forces dogs is a profitable improvement over standard selection procedures, compared to improving measurement/forecasting methods. If training is extremely expensive or heritability is relatively high, dog cloning is hypothetically profitable. (genetics, decision theory, R)
created: 18 Sep 2018; modified: 14 Jan 2019; status: in progress; confidence: possible;

Cloning is widely used in animal & plant breeding despite steep costs due to its advantages; more unusual applications have opened up recently, including entire polo horse teams of clones and reported trials in police/Special Forces war dogs. Given the cost of cloning, however, can this ever make more sense than standard screening methods for selecting from working dog breeds? I model the question as one of cost per dog with the trait of successfully passing screening being a dichotomous liability threshold polygenic trait, and look at some reported costs and success rates and dog heritabilities. Since none of the relevant parameters are known with confidence, I run the cost-benefit equation for many hypothetical scenarios, and find that in a fraction of them, dog cloning would improve training yields enough to be profitable.

Military and police dogs are specially trained for their jobs. Only some dogs are appropriate, but like training seeing-eye guide dogs, it’s difficult to know in advance and many dogs will wash out of training as expensive failures; then they may get injured on the job, develop hip dysplasia or cancer, making for a short career, and leading to perennial shortages. This is despite the best efforts of the (mostly European) breeders who raise the Malinois, German Shepherd, Belgian Shepherd, and Labradors preferred for war dogs.

In 2014, Bloomberg reported on an interesting aspect of Sooam Biotech, the famous South Korean dog cloning company: they were cloning a Special Forces dog. If it’s hard to be a K9, it’s even harder to be a SF dog, able to jump out of airplanes (they have special parachute harnesses), go on raids, carry cameras with them, even (reportedly) wear little doggie hoods with infrared camera goggles for night work. If you have a successful SF dog… maybe the clone will be much more likely to succeed than a random puppy picked from one of the usual breeders, and you can make as many clones as necessary long after the original has gone to Dog Heaven. A striking example of this approach is the world polo champion Adolfo Cambiaso, who has cloned his prized polo horse not once but >10 times, and has rode entire teams of clones to repeated victory. On the other hand, dog clones are still extremely expensive (~$100,000) and prices have not yet come down to the eg ~$10-20k of cattle.

There may be cheaper alternatives to improving SF dog yield: training is probably well-refined and can’t be watered down without risking lives, but that leaves an obvious place for improvement of selection into training - better prediction of SF potential means fewer dogs washing out means less total money spent to produce a successful SF dog. The predictions don’t work well, but the descriptions of screening suggest there’s a lot of room for improvement: the research literature supports the generalization that dog and cat behavioral measurements are not all that predictive. They may be badly designed or testing the wrong things, or there may be inherent noise which can be fixed by doing multiple measurements. (Even something as apparently mechanical as offering catnip to a cat can have different results from occasion to occasion and may have rater-specific effects, because perhaps the cat is fearful and distrusts the person offering the catnip that day, with the anxiety shutting down any response or play.) Many described measurements in the literature measure a dog once, on one day, by one person, for example, measuring aggressiveness by taking away food and seeing if the dog snaps at the person, and that’s the whole test. Such a test will be hindered by day-to-day variation (perhaps he is stressed that day), different levels of liking for that particular food, disliking of the person taking the food, sheer randomness in the particular split-second decision of whether the dog decides to express their aggression and likely would be much stabler and predictive if they were done multiple times in multiple ways by multiple people etc, although such extended testing would increase the cost of testing.

Rationale

Some of the potential benefits of dog clones:

1. lower total cost: since dog clones are much more likely to succeed in training given any reasonable heritability, they may reduce washout costs enough to more than compensate for the expense of cloning.
2. greater scalability in dogs: there are only a few dog breeders, and they have only a relative handful of bitches at any time; even if demand spiked in a war and 10,000 more dogs were needed yesterday, they simply wouldn’t be able to deliver them - dogs take a certain amount of time to reach sexual maturity, have only so big litters, mating in inbred/narrow pedigrees like German Shepherds/Malinois must be managed carefully to avoid exacerbating existing genetic issues (thus eating the seed corn), training takes a while, and so on. Use of cloning can break part of the bottleneck by enabling surrogacy in female dogs of other breeds which are not scarce, and by enabling unlimited reproduction of a particular individual. (This doesn’t require cloning, since one could create the necessary embryos with standard IVF, but since the IVF/surrogacy is necessary, why not use cloning as well?)

• greater scalability in facilities: another bottleneck might be not the number of dogs, but the infrastructure for housing/training/testing the dogs. There might be only so many dog kennels and experienced dog trainers at any point, and increasing the number could take a while. (You probably want the trainers and program management to have SF dog handler experience themselves, but it might take decades for a new recruit to become an experienced trainer.) So given the inelastic and relatively fixed throughput, here it would be valuable to improve the quality of inputs, which will increase the total yield, simply because it means less dilution or waste of scarce fixed housing/training/testing slots on dogs less likely to succeed.
3. greater predictability:

• response to training: yield might be increased simply by the inherent homogeneity of clones allowing improved training by greater experience, rather than any specifically genetic merit. One of the reasons Adolfo Cambiaso gives for investing so heavily in clones of a single polo horse is that he has learned from experience how best to train them: each new clone can be given personalized training which he knows it’ll respond best to, because he’s trained many clones before them. If there is some consistent weakness the clones are prone to, he can start addressing it before it even shows up. He also has gained long experience with their injury propensities, preferences, and other behavior, instead of starting from scratch with each new colt. Their similarity avoids the need for learning or wasted pedagogy. Dogs presumably vary as much as horses do, and training of clones could benefit from this sort of homogeneity. (Since dog trainers will have never encountered clones before, and identical twin dogs are vanishingly rare, there’s no way to know how useful this would be in practice until large numbers of dog clones have been trained by individual trainers.)
• reduced variance: given the choice between a small group of clones and a much larger group of regular dogs, such that they have ostensibly identical average costs and the same number of expected successes, which should a manager prefer? The small group of clones, of course. The large group will, by the law of small numbers, have larger absolute fluctuations due to randomness, especially with a base rate like 1%. It’ll be feast or famine. Sometimes there will be considerably more, sometimes considerably less in absolute numbers. This will complicate planning greatly, stress facilities/trainers, risk delivering too few dogs each year, and so on. Switching to clones with a higher base rate will make the overall process more controllable and predictable, and this is worth something.
4. use in selective breeding for more elite dogs:

• a clone can have a higher genetic potential than the average SF dog if intensive selection is done among SF dogs: if the best SF dog is selected for cloning, it’ll have a higher genetic potential than the default calculation of a truncated normal+regression toward the mean would imply.
• elite clones can be useful in breeding programs in allowing particular individuals to keep contributing genetically long after the original has become infertile or died, or contribute far more (as mentioned before, female dogs are highly limited in reproductive fecundity compared to males, but they could be cloned & born via surrogacy)
5. Value of Information: dog cloning may or may not be worthwhile, but if it is, the total returns from cloning hundreds of dogs per year indefinitely (plus the additional benefits) could be large. It would be valuable to know if it would work. Since heritabilities are (apparently) unavailable, the only way to know is to try it out experimentally. (The clones’ realized performance would also provide additional valuable information as it would estimate heritability, which would be useful for the regular kinds of breeding & selection as well - because they give an idea of how much one can predict a dog’s performance based on known relatives, and how fast a breeding program can/should proceed.)

Since, to be profitable, the success rate of clones need to be >=9% (see later), this is reasonably easy to estimate: a sample of ~50 clones would give a reasonably precise estimate as to the success rate and enable better decision-making as to whether to keep pursuing cloning (in which case more information will come in and firm up the decision) or drop it as a dead end due to too high costs and/or low heritability.

Screening problem

Of course, that would take more time and would cost a lot more, and it’s unclear the increase in predictions is worth it. It’s a variant of the old mammography screening scenario in Bayesian statistics: for a rare case of low prior probability, even a good test will increase the posterior probability to a surprisingly small probability, because there are just so many false positives along with the occasional true positive.

So both approaches could wind up being expensive and there’s no a priori answer about which one would be more cost-effective. To a certain extent, they are also mutually exclusive approaches: dog cloning is so expensive that unless it results in high probability of success, it probably won’t be cost-effective at all, and if the probability is sufficiently high, then testing is no longer useful (because you would save money by simply trying to train all clones), so better testing is unlikely to then pay for itself. Testing to gain information is only profitable in a certain intermediate region of probabilities & costs/benefits.

Modeling the SF selection problem

How could we estimate the benefit of cloning? A SF dog is highly selected among candidate dogs, and it is either an acceptable SF dog or not. Being a SF dog requires a package of traits, ranging from physical health to courage to finely-controlled aggression (attacking if the handler orders, immediately stopping when counter-ordered), which sum up to an overall quality: somewhat poorer health can be made up by better smelling skills, say.

So a natural approach is to treat it as a logistic model, or more specifically, a liability threshold model: if a bunch of random variables all sum up to a certain high score, the dog becomes SF, otherwise, it is a normal dog. These random variables can be split into genetic variables and everything else, the environmental variables.

Then the benefit of cloning can be estimated based on how much the genetic variables contribute to a high score, how high the genetic variables of a cloned SF dog might be (remembering that they are highly selected and thus imply regression to the mean), and this provides an estimate for increased probability that the clones will achieve a high score too.

Once the probability a clone will succeed versus a random candidate dog is calculated, then one can get the cost of screening candidate dogs for a SF dog versus cloning+screening clone dogs for a SF dog.

Liability threshold model

This requires us to estimate two things: the threshold and the heritability on the liability scale.

For common police dogs and other working dogs, training appears to be not that hard, and estimates of 30-50% are seen. This gives a threshold of 50%, or in standard deviations, 0SD.

A SF dog is much more selective, and the only specific estimate given is 1%, which in standard deviations, means each dog would be >=2.33SD. (For a threshold like 50%, the mean of everyone >=0SD/>=50% is actually more like 0.8SD/75% - which is very different from 0SD/50%! - and we need to use the truncated normal distribution to get it right.)

The clone of the SF dog shares only genetics with it, it doesn’t benefit from the unique luck and environment that the original did which helped it achieve it success, so it will regress back to the mean. If genetics determined 100% of the outcome, then the clones would make the 1%/+2.33SD cutoff 100% of the time, as they have the same genetic potential and zero environmental input (although that is extremely unlikely a scenario, due to measurement error in the testing if nothing else). While if genetics contributed 0% to the outcome, then the clones will make the 1% cutoff just as often as if they were a random dog sampled from their breeders ie. 1%. And in between, in between.

Under a more plausible case like genetics determining 50% of the variability (a common level of heritability for better-studied human traits), then we could only expect the clones to be above-average by +1.16SD; if the clones are distributed around a mean of 1.16SD from their genes, what’s the probability they will reach a total of +2.33SD with help from the environmental variables? Half the variance is used up, and the environment has to contribute another $2.33-1.33$ = +1SD, despite causing differences of only 0.5SD on average. In that case, the clones will have ~9% chance of being successful - which is 9x greater than a random dog, but also far from guaranteed:

## exact mean:
truncNormMean <- function(a, mu=0, sigma=1, b=Inf) {
phi <- dnorm
erf <- function(x) 2 * pnorm(x * sqrt(2)) - 1
Phi <- function(x) { 0.5 * (1 + erf(x/sqrt(2))) }
Z <- function(beta, alpha) { Phi(beta) - Phi(alpha) }

alpha=(a-mu)/sigma; beta=(b-mu)/sigma

(phi(alpha) - phi(beta))/Z(beta, alpha) }
## If we select the top percentile, the cutoff is +2.32SD, but the mean is higher, +2.66SD:
qnorm(0.99)
# [1] 2.32634787
truncNormMean(qnorm(0.99))
# [1] 2.66521422

cloningBoost <- function(successP=0.01, heritability=0.5, verbose=FALSE) {
threshold <- qnorm(1-successP)

originalMean <- truncNormMean(threshold)
cloneMean <- 0 + (heritability * originalMean) ## regress to mean
regression <- originalMean - cloneMean

## Monte Carlo estimate to double-check:
# cloneP <- mean(rnorm(10000000, mean=cloneMean, sd=1-(heritability)^2)>threshold)
cloneP <- pnorm(cloneMean - threshold, sd=1-(heritability)^2)

if(verbose) { print(c(threshold, originalMean, cloneMean, regression, cloneP)) }
return(cloneP) }

cloningBoost(successP=0.01, heritability=1.0, verbose=TRUE)
# [1] 2.32634787 2.66521422 2.66521422 0.00000000 1.00000000
# [1] 1
cloningBoost(successP=0.01, heritability=0.8, verbose=TRUE)
# [1] 2.326347874 2.665214220 2.132171376 0.533042844 0.294812629
# [1] 0.294812629
cloningBoost(successP=0.01, heritability=0.5, verbose=TRUE)
# [1] 2.3263478740 2.6652142203 1.3326071102 1.3326071102 0.0925876179
# [1] 0.0925876179
cloningBoost(successP=0.01, heritability=0.2, verbose=TRUE)
# [1] 2.3263478740 2.6652142203 0.5330428441 2.1321713763 0.0308792184
# [1] 0.0308792184
cloningBoost(successP=0.01, heritability=0.0, verbose=TRUE)
# [1] 2.32634787 2.66521422 0.00000000 2.66521422 0.01000000
# [1] 0.01

The real price per dog may be lower. Their practice seems to be to engage in overkill by implanting multiple clone embryos to ensure the minimum specified number of healthy clones, and offer all resulting healthy live-births to the customer - for example, Vangemert is not mentioned as being charged $150k instead of$50k for the 3 puppies, as would be the case if Viagen charged $50k for each success and she requested 2 and got 3. So the per-clone cost appears to have become surprisingly reasonable:$50k-$100k for 1 dog, and potentially <$16k (if a purchase results in 3 puppies as in Vangemert’s case), suggesting that the price has a large fixed cost to it and the marginal costs might be quite small, which is not helpful in the pet-replacement scenario but would be important to large-scale cloning of specific animals like elite drug sniffers. For comparison, cattle and horse cloning have become industrialized at ~$10-15k; dog cloning is apparently more difficult (harder to control estrous, Viagen notes), so that may be a lower bound for the foreseeable future. Cost-benefit Does cloning minimize loss? My cost-benefit below takes the cost per final dog without cloning, computes the implied per-dog-candidate cost, and then computes the increased success rate for a given threshold+heritability, and sees if the expected cloning+training cost is less than the original total cost. dogCloningCB <- function(successP, heritability, totalTrainingCost, marginalCloningCost, verbose=FALSE) { normalLoss <- -totalTrainingCost marginalTrainingCost <- totalTrainingCost / (1/successP) cloningP <- cloningBoost(successP=successP, heritability=heritability) cloningLoss <- - ((1/cloningP) * (marginalTrainingCost + marginalCloningCost)) if(verbose) {return(list(Boost=cloningP, Cost.normal=normalLoss, Cost.clone=cloningLoss, Profitable=normalLoss<cloningLoss, Profit=cloningLoss-normalLoss)) } return(cloningLoss-normalLoss) } ## Example: dogCloningCB(0.01, 0.8, 50000, 50000, verbose=TRUE) #$Boost
# [1] 0.294812629
#
# $Cost.normal # [1] -50000 # #$Cost.clone
# [1] -171295.24
#
# $Profitable # [1] FALSE # #$Profit
# [1] -121295.24

Scenarios

As the key heritability trait is almost completely unknown and heritabilities of dog behavioral traits are all over the map and seem to suffer from severe measurement error issues, we might as well consider a wide range of scenarios to get an idea of what it would take. For success/threshold, we continue with 1%; for heritability, we’ll consider the most plausible range, 0-90%; for training cost, we’ll do the full $50k-$283k range since while it’s unclear what these numbers mean, treating them as a total per-dog cost is being conservative and makes it harder for cloning to be profitable, and for cloning costs we’ll consider the Vangemert case up to Viagen’s list price of $50k (since there doesn’t seem to be any good reason to pay twice as much to Sooam). scenarios <- expand.grid(SuccessP=0.01, Heritability=seq(0, 0.9, by=0.10), trainingCost=seq(50000, 283000, by=10000), cloningCost=seq(15000, 50000, by=10000)) scenarios$Profit <- round(unlist(Map(dogCloningCB, scenarios[,1], scenarios[,2], scenarios[,3], scenarios[,4])))

## Plot relationships:
library(ggplot2); library(gridExtra)
p1 <- qplot(cloningCost,  Profit, color=Heritability, data=scenarios) + geom_hline(yintercept=0, color="red")
p2 <- qplot(trainingCost, Profit, color=Heritability, data=scenarios) + geom_hline(yintercept=0, color="red")
grid.arrange(p1, p2, ncol=1)

## All profitable scenarios with reasonable heritabilities:
scenarios[scenarios$Profit>0 & scenarios$Heritability<=0.6,]
#     SuccessP Heritability trainingCost cloningCost Profit
# 146     0.01          0.5       190000       15000   7470
# 156     0.01          0.5       200000       15000  16390
# 166     0.01          0.5       210000       15000  25310
# 176     0.01          0.5       220000       15000  34230
# 186     0.01          0.5       230000       15000  43150
# 196     0.01          0.5       240000       15000  52070
# 206     0.01          0.5       250000       15000  60990
# 216     0.01          0.5       260000       15000  69910
# 225     0.01          0.4       270000       15000   4898
# 226     0.01          0.5       270000       15000  78830
# 235     0.01          0.4       280000       15000  13400
# 236     0.01          0.5       280000       15000  87750

## All profitable scenarios:
scenarios[scenarios$Profit>0,] # ... The subset of profitable scenarios for SF dog cloning (typically requiring high heritabilities, and higher training costs / lower cloning costs). Success Probability Heritability Training cost Cloning Cost Profit 0.01 0.9 50000 15000 26092 0.01 0.8 60000 15000 7085 0.01 0.9 60000 15000 35937 0.01 0.8 70000 15000 16746 0.01 0.9 70000 15000 45783 0.01 0.8 80000 15000 26407 0.01 0.9 80000 15000 55629 0.01 0.7 90000 15000 3198 0.01 0.8 90000 15000 36067 0.01 0.9 90000 15000 65475 0.01 0.7 100000 15000 12652 0.01 0.8 100000 15000 45728 0.01 0.9 100000 15000 75320 0.01 0.7 110000 15000 22106 0.01 0.8 110000 15000 55389 0.01 0.9 110000 15000 85166 0.01 0.7 120000 15000 31560 0.01 0.8 120000 15000 65050 0.01 0.9 120000 15000 95012 0.01 0.6 130000 15000 2576 0.01 0.7 130000 15000 41014 0.01 0.8 130000 15000 74711 0.01 0.9 130000 15000 104858 0.01 0.6 140000 15000 11795 0.01 0.7 140000 15000 50468 0.01 0.8 140000 15000 84371 0.01 0.9 140000 15000 114703 0.01 0.6 150000 15000 21013 0.01 0.7 150000 15000 59922 0.01 0.8 150000 15000 94032 0.01 0.9 150000 15000 124549 0.01 0.6 160000 15000 30231 0.01 0.7 160000 15000 69377 0.01 0.8 160000 15000 103693 0.01 0.9 160000 15000 134395 0.01 0.6 170000 15000 39449 0.01 0.7 170000 15000 78831 0.01 0.8 170000 15000 113354 0.01 0.9 170000 15000 144241 0.01 0.6 180000 15000 48668 0.01 0.7 180000 15000 88285 0.01 0.8 180000 15000 123015 0.01 0.9 180000 15000 154086 0.01 0.5 190000 15000 7470 0.01 0.6 190000 15000 57886 0.01 0.7 190000 15000 97739 0.01 0.8 190000 15000 132675 0.01 0.9 190000 15000 163932 0.01 0.5 200000 15000 16390 0.01 0.6 200000 15000 67104 0.01 0.7 200000 15000 107193 0.01 0.8 200000 15000 142336 0.01 0.9 200000 15000 173778 0.01 0.5 210000 15000 25310 0.01 0.6 210000 15000 76322 0.01 0.7 210000 15000 116647 0.01 0.8 210000 15000 151997 0.01 0.9 210000 15000 183624 0.01 0.5 220000 15000 34230 0.01 0.6 220000 15000 85541 0.01 0.7 220000 15000 126101 0.01 0.8 220000 15000 161658 0.01 0.9 220000 15000 193469 0.01 0.5 230000 15000 43150 0.01 0.6 230000 15000 94759 0.01 0.7 230000 15000 135555 0.01 0.8 230000 15000 171319 0.01 0.9 230000 15000 203315 0.01 0.5 240000 15000 52070 0.01 0.6 240000 15000 103977 0.01 0.7 240000 15000 145009 0.01 0.8 240000 15000 180979 0.01 0.9 240000 15000 213161 0.01 0.5 250000 15000 60990 0.01 0.6 250000 15000 113195 0.01 0.7 250000 15000 154463 0.01 0.8 250000 15000 190640 0.01 0.9 250000 15000 223007 0.01 0.5 260000 15000 69910 0.01 0.6 260000 15000 122414 0.01 0.7 260000 15000 163917 0.01 0.8 260000 15000 200301 0.01 0.9 260000 15000 232853 0.01 0.4 270000 15000 4898 0.01 0.5 270000 15000 78830 0.01 0.6 270000 15000 131632 0.01 0.7 270000 15000 173371 0.01 0.8 270000 15000 209962 0.01 0.9 270000 15000 242698 0.01 0.4 280000 15000 13400 0.01 0.5 280000 15000 87750 0.01 0.6 280000 15000 140850 0.01 0.7 280000 15000 182825 0.01 0.8 280000 15000 219623 0.01 0.9 280000 15000 252544 0.01 0.9 50000 25000 10667 0.01 0.9 60000 25000 20513 0.01 0.9 70000 25000 30358 0.01 0.9 80000 25000 40204 0.01 0.8 90000 25000 2148 0.01 0.9 90000 25000 50050 0.01 0.8 100000 25000 11808 0.01 0.9 100000 25000 59896 0.01 0.8 110000 25000 21469 0.01 0.9 110000 25000 69742 0.01 0.8 120000 25000 31130 0.01 0.9 120000 25000 79587 0.01 0.8 130000 25000 40791 0.01 0.9 130000 25000 89433 0.01 0.8 140000 25000 50452 0.01 0.9 140000 25000 99279 0.01 0.7 150000 25000 5330 0.01 0.8 150000 25000 60112 0.01 0.9 150000 25000 109125 0.01 0.7 160000 25000 14784 0.01 0.8 160000 25000 69773 0.01 0.9 160000 25000 118970 0.01 0.7 170000 25000 24238 0.01 0.8 170000 25000 79434 0.01 0.9 170000 25000 128816 0.01 0.7 180000 25000 33692 0.01 0.8 180000 25000 89095 0.01 0.9 180000 25000 138662 0.01 0.7 190000 25000 43146 0.01 0.8 190000 25000 98756 0.01 0.9 190000 25000 148508 0.01 0.7 200000 25000 52600 0.01 0.8 200000 25000 108416 0.01 0.9 200000 25000 158353 0.01 0.7 210000 25000 62055 0.01 0.8 210000 25000 118077 0.01 0.9 210000 25000 168199 0.01 0.6 220000 25000 7367 0.01 0.7 220000 25000 71509 0.01 0.8 220000 25000 127738 0.01 0.9 220000 25000 178045 0.01 0.6 230000 25000 16585 0.01 0.7 230000 25000 80963 0.01 0.8 230000 25000 137399 0.01 0.9 230000 25000 187891 0.01 0.6 240000 25000 25803 0.01 0.7 240000 25000 90417 0.01 0.8 240000 25000 147060 0.01 0.9 240000 25000 197736 0.01 0.6 250000 25000 35021 0.01 0.7 250000 25000 99871 0.01 0.8 250000 25000 156720 0.01 0.9 250000 25000 207582 0.01 0.6 260000 25000 44240 0.01 0.7 260000 25000 109325 0.01 0.8 260000 25000 166381 0.01 0.9 260000 25000 217428 0.01 0.6 270000 25000 53458 0.01 0.7 270000 25000 118779 0.01 0.8 270000 25000 176042 0.01 0.9 270000 25000 227274 0.01 0.6 280000 25000 62676 0.01 0.7 280000 25000 128233 0.01 0.8 280000 25000 185703 0.01 0.9 280000 25000 237119 0.01 0.9 60000 35000 5088 0.01 0.9 70000 35000 14934 0.01 0.9 80000 35000 24780 0.01 0.9 90000 35000 34625 0.01 0.9 100000 35000 44471 0.01 0.9 110000 35000 54317 0.01 0.9 120000 35000 64163 0.01 0.8 130000 35000 6871 0.01 0.9 130000 35000 74008 0.01 0.8 140000 35000 16532 0.01 0.9 140000 35000 83854 0.01 0.8 150000 35000 26193 0.01 0.9 150000 35000 93700 0.01 0.8 160000 35000 35853 0.01 0.9 160000 35000 103546 0.01 0.8 170000 35000 45514 0.01 0.9 170000 35000 113391 0.01 0.8 180000 35000 55175 0.01 0.9 180000 35000 123237 0.01 0.8 190000 35000 64836 0.01 0.9 190000 35000 133083 0.01 0.8 200000 35000 74497 0.01 0.9 200000 35000 142929 0.01 0.7 210000 35000 7462 0.01 0.8 210000 35000 84157 0.01 0.9 210000 35000 152774 0.01 0.7 220000 35000 16916 0.01 0.8 220000 35000 93818 0.01 0.9 220000 35000 162620 0.01 0.7 230000 35000 26370 0.01 0.8 230000 35000 103479 0.01 0.9 230000 35000 172466 0.01 0.7 240000 35000 35824 0.01 0.8 240000 35000 113140 0.01 0.9 240000 35000 182312 0.01 0.7 250000 35000 45278 0.01 0.8 250000 35000 122801 0.01 0.9 250000 35000 192157 0.01 0.7 260000 35000 54732 0.01 0.8 260000 35000 132461 0.01 0.9 260000 35000 202003 0.01 0.7 270000 35000 64187 0.01 0.8 270000 35000 142122 0.01 0.9 270000 35000 211849 0.01 0.7 280000 35000 73641 0.01 0.8 280000 35000 151783 0.01 0.9 280000 35000 221695 0.01 0.9 80000 45000 9355 0.01 0.9 90000 45000 19201 0.01 0.9 100000 45000 29046 0.01 0.9 110000 45000 38892 0.01 0.9 120000 45000 48738 0.01 0.9 130000 45000 58584 0.01 0.9 140000 45000 68429 0.01 0.9 150000 45000 78275 0.01 0.8 160000 45000 1934 0.01 0.9 160000 45000 88121 0.01 0.8 170000 45000 11594 0.01 0.9 170000 45000 97967 0.01 0.8 180000 45000 21255 0.01 0.9 180000 45000 107812 0.01 0.8 190000 45000 30916 0.01 0.9 190000 45000 117658 0.01 0.8 200000 45000 40577 0.01 0.9 200000 45000 127504 0.01 0.8 210000 45000 50238 0.01 0.9 210000 45000 137350 0.01 0.8 220000 45000 59898 0.01 0.9 220000 45000 147195 0.01 0.8 230000 45000 69559 0.01 0.9 230000 45000 157041 0.01 0.8 240000 45000 79220 0.01 0.9 240000 45000 166887 0.01 0.8 250000 45000 88881 0.01 0.9 250000 45000 176733 0.01 0.7 260000 45000 140 0.01 0.8 260000 45000 98542 0.01 0.9 260000 45000 186578 0.01 0.7 270000 45000 9594 0.01 0.8 270000 45000 108202 0.01 0.9 270000 45000 196424 0.01 0.7 280000 45000 19048 0.01 0.8 280000 45000 117863 0.01 0.9 280000 45000 206270 At the lab, the staff is preparing for a group of Americans coming to pick up two special puppies, cloned from the DNA of a Belgian Malinois that’s currently deployed with a unit of the U.S. Army Special Forces (which Sooam isn’t permitted to name). The donor dog was chosen because he was a standout among Special Forces canines-elite even among the elite, something like the soldier-dog equivalent of LeBron James-and these three-month-old puppies were heading to the U.S. to undergo training as part of an experiment. Photo of two Belgian Malinois puppies, cloned from the DNA of a dog that’s currently deployed with a unit of the U.S. Army Special Forces At this point, cloning a pet is straightforward for Sooam. Given fresh cells, Hwang says, we have never failed cloning a specific dog, regardless of its size or breed. In turn, that part of the business is fairly mature. Orders are healthy. There’s a waiting list. What’s most intriguing to Hwang now is the study of clone performance, particularly among what Sooam calls special purpose dogs. He wants to know if a puppy cloned from a truly exceptional working dog will end up performing at that job as well as his genetic twin. If he does, it could seriously disrupt the process of breeding and training police dogs, explosives detection dogs, and others that serve in jobs that help save human lives. Recently, Sooam secured a contract to provide 40 cloned special purpose dogs to the South Korean national police, and several are already in service at the Incheon International Airport near Seoul. But Hwang’s scientists lack proof that the donor dogs were truly special. That’s why they sought out the Americans, to find empirically great dogs to clone. …Eventually, they settled on Shallow Creek Kennels, a small facility north of Pittsburgh that trains elite dogs for numerous police departments and U.S. government agencies, including Special Operations. The owner, John Brannon, loved the idea and had just the dog in mind. He arranged for fibroblasts to be collected from the dog, which is currently working in Afghanistan and whose identity is classified. Sooam cloned him, resulting in Ghost and Echo, the adorable clone brothers that the Americans had all come to Seoul to collect. Because every day matters when your goal is to turn a puppy with potential into a dependable, battle-ready working dog, Brannon had given Sooam staffers a strict training and socialization regimen to follow from birth, but it isn’t until the dogs are bounding around on the front lawn after a short adoption ceremony that Brannon is able to get his first good look at them. I’m impressed. They seem advanced for their age. But you don’t really know until a dog is 12 months what you have physically and mentally, Brannon says, which is why he doesn’t bother with the imprecise and wasteful process of breeding. It’s far more effective for him to travel to Europe a few times a year to source year-old dogs from one of several kennels he knows and trusts. One of the most challenging things about great police dogs, Badertscher says, is finding the right puppies and then training them, only to have to retire them eight or nine years later. Now we have a chance, an idea-it’s only a theory, he says. Every time you breed a dog naturally, you lose some portion of its greatness, because the genes are diluted by the contribution of the mate. And you’re lucky if one or two dogs out of a litter of eight might have the drive and focus to become the kind of dogs who can find bombs, take fire, and work independently on command-let alone jump out of airplanes at night. Ghost and Echo are the first research study to see if this idea works: Can we reproduce these top-quality dogs through cloning and eliminate most of the margin for error, Badertscher says. Beyond that, he believes, the next step is giving these dogs a chance to live longer by using cloning to eliminate problems such as cancer, hip dysplasia, and bad eyesight that can prematurely end a working dog’s career. Two extra years of work would be an incredible boost in productivity, keeping the best dogs working longer and offsetting the increased costs of cloning. The biggest thing we’ll have to fight, he says, is the word cloning. ghost and echo have been joined by specter; 3 of 3 clones are successful enough to be police K-9s like the original: Brannon says cloning seems to take the guess work out of normal breeding procedures. Meaning, you have an excellent male an excellent female, and maybe out of a litter of eight only four would be police service dogs or military dogs, according to Brannon. Specter is the third clone that the kennel has trained, and the other two are now working with federal SWAT units. Right now were are three for three and they’re all successful, said Brannon. The Sooam Biotech Research Foundation has started a pilot program aimed at replicating the crème de la creme of military and police dogs. We’re looking for the top 1 percent, says John Brannon, a trainer in Pennsylvania working with Sooam. Some dogs are genetically predisposed to be superstars. … The K-9 cloning would fall under the foundation’s mandate to help human welfare, and the cost would be negotiable and likely lower if the dogs are sold in bulk, a spokesman said. Sooam went on to work with partner organizations to find and clone other highly skilled canines with proven track records in working with quick-response teams to hunt criminals and uncover narcotics. So far its biggest recipient is South Korea’s national 119 rescue service, but it also has sent three cloned dogs to a U.S. police dog training facility in Pennsylvania. Demand is on the rise for K-9s, elite dogs that play important roles in combat, bomb detection, narcotics investigations and other operations….But breeding and training programs are costly and often inefficient. For example, the school that trains K-9s for the Department of Defense has found that the suitability rate runs around 50 percent, so the program tries to train about 200 dogs per year to produce 100 that are serviceable. …Brannon, who also trains dogs for police departments around the U.S. as well as the military, said he was skeptical about cloning in the beginning but is now convinced it is more efficient than natural-breeding programs. He’s expecting another clone next year - this one the twin of a dog that has helped agents find millions of dollars in narcotics and apprehend many suspects. However, this European solution turned out to be only temporary, as rejection rates continued to remain high, and continue today in the range of 25 to 50 per cent (Andersen, Burke, Craig, Hayter, McCathern, Parks, Thorton). http://www.dtic.mil/dtic/tr/fulltext/u2/a229000.pdf Former Navy SEAL Mike Ritland, who now trains dogs for U.S. Special Forces, wrote about training Malinois in his book Trident K9 Warriors. The 200-step training program the military uses costs$50,000 per dog. Not all Malinois make the cut. According to Ritland, only 1 percent make it into the U.S. Special Forces. The dogs we deploy have to be unflappable in all circumstances, he wrote. They have to perform their activities willingly and with a single-minded purposefulness that few, if any, humans possess. https://www.washingtonpost.com/news/morning-mix/wp/2014/09/23/the-belgian-malinois-the-dog-the-white-house-didnt-use-on-fence-jumping-intruder/

Special Operations Forces canines are overwhelmingly chosen from one breed, the Belgian Malinois. Only 1% of candidate dogs make the cut for training.

https://www.nytimes.com/2011/06/12/us/12dogs.html :

When she costs $230,000, as Julia did, the preferred title is executive protection dog. This 3-year-old German shepherd, who commutes by private jet between a Minnesota estate and a home in Arizona, belongs to a canine caste that combines exalted pedigree, child-friendly cuddliness and arm-lacerating ferocity. Julia and her ilk have some of the same tracking and fighting skills as the dogs used in elite military units like Navy Seal Team 6, which took a dog on its successful raid of Osama bin Laden’s compound in Pakistan. In fact, Julia was sold by a trainer, Harrison Prather, who used to supply dogs to Seal Team 6 and the British special forces. But then Mr. Prather switched to a more lucrative market. Either rich people discovered me or I discovered them - I can’t remember which happened first, said Mr. Prather, the president of Harrison K-9 Security Services in Aiken, S.C. He and others in the high-end dog training business say prices have shot up thanks to the growing number of wealthy people around the world who like the security - and status - provided by a dog with the right credentials. Moguls and celebrities now routinely pay$40,000 to $60,000 for a well-bred German shepherd that is certified as an expert in the sport of Schutzhund, which means protection dog. The price can go much higher if a dog does well at an international championship, as Julia did. …Mr. Prather’s dogs are trained for three years in Germany before they go to South Carolina, where they receive further training and are put to the test of family living. Before her sale, Julia lived for four months in the home of November Holley, the company’s vice president and head trainer. https://www.bloomberg.com/news/features/2017-08-28/military-dogs-are-becoming-an-increasingly-precious-weapon The armed services have had dogs since about day one. At the moment, roughly 1,600 Military War Dogs (MWDs) are either in the field or helping recuperating veterans. That’s approximately one dog for every three U.S. soldiers currently in Afghanistan. These animals are, however, an increasingly precious resource. With terrorists targeting public transportation and tourist sites all over the world, global demand for bomb-sniffing dogs has surged. Canines with finely trained noses now fetch$25,000 and up on the open market, where border patrol units, the State Department, and private security firms go for canine talent. Even the war on bedbugs scoops up some of the best noses in the business. And that’s just U.S. demand.

We thought maybe we’d sell 50 [training models], but it has just grown overwhelmingly, said KForce Vice President Carolyn Hollander, who added that the project was originally considered just doing the right thing-a give back.

The K9 Hero-Trauma’s proxy pooch-is fully articulated, weighs 50 pounds, and costs around $20,000. It has a pulse and an internal, inflating bag that mimics breathing, plus a host of potential afflictions. Push a button on a remote control and the rubbery pet even bleeds profusely. Next month, the company will deliver Hero’s successor, Diesel. Developed specifically for special forces dogs, the animatronic soldier has multiple gunshot wounds, amputatable limbs, and bowels that bloat. It also barks and whimpers. …The high demand for trained dogs may play a role in their depleted ranks. The Air Force says the U.S. military’s dog allotment is about 38 percent lower than it was at the height of the war in Afghanistan. Exacerbating the economics for the Pentagon, the U.S. still isn’t very good at producing war dogs. Although the Air Force has begun a breeding program in Texas, most of the country’s working dogs are imports, primarily from Central and Eastern Europe-where dog-training culture runs deep. Military procurement officers make four trips a year to stock up on European puppies. In a 2016 Senate hearing, Cynthia Otto, a veterinarian and executive director of the Penn Vet Working Dog Center, said the availability of good dogs was becoming a critical challenge. The risks of relying on foreign sources of dogs to support our national security are high, she said. The U.S. military spends up to$283,000 to train a working war dog.

Once it has a promising pup, the Pentagon spends an additional $42,000 to train a K9 unit, a process that starts with obedience and drug and/or bomb detection at Lackland Air Force Base in San Antonio, Texas. Some of the dogs get a second round of training in how to patrol, detain an enemy and attack. A dual-purpose dog spends about 120 days completing both training cycles. When all is said and done, a fully trained military dog costs about as much as a small missile. Keeping them in the field as long as possible is increasingly good business. (The Air Force declined to discuss canine casualty rates.) https://www.townandcountrymag.com/society/a12108750/personal-protection-dogs/ How a Former Navy Seal Turns an Attack Dog Into Your$100,000 New Best Friend: For a hefty price, Mike Ritland trains Personal Protection Dogs to keep your family safe

The perfect guard dog knows when to bark and when to bite, and it can turn back into a docile pet after an incident…Most amazingly of all, after an incident they must be able to mellow out and morph back into a docile pet. If this sounds practically impossible, it is. [Mike] Ritland estimates that around 1 percent of all dogs have this capability.

https://www.nytimes.com/2011/05/12/world/middleeast/12dog.html The Dogs of War: Beloved Comrades in Afghanistan

American troops may be starting to come home this summer, but more dogs are going in. In 2007, the Marines began a pilot program in Afghanistan with nine bomb-sniffing dogs, a number that has grown to 350 and is expected to reach nearly 650 by the end of the year. Over all, there are some 2,700 dogs on active duty in the American military. A decade ago, before the Sept. 11 attacks, there were 1,800.

Most of the public isn’t aware of what these dogs add to national security, said Gerry Proctor, a spokesman for training programs at Lackland Air Force Base in Texas, including the Military Working Dog School. Dogs are used for protection, pursuit, tracking and search and rescue, but the military is also increasingly relying on them to sniff out the homemade bombs that cause the vast majority of American casualties in Afghanistan. So far, no human or human-made technology can do better.

Within the military, the breeds of choice are generally the German shepherd and a Belgian shepherd, or Malinois, but Marines in Afghanistan rely on pure-bred Labrador retrievers because of the dogs’ good noses and nonaggressive, eager-to-please temperaments. Labs now accompany many Marine foot patrols in Helmand Province in southern Afghanistan, wandering off-leash 100 yards or more in front as bomb detectors. It is the vital work of an expensively trained canine (the cost to the American military can be as high as $40,000 per dog), but at the end of a sweltering day, sometimes a Lab is still a Lab. Ground Dog Day: Lessons Don’t Have To Be Relearned In The Use Of Dogs In Combat, Hammerstrom 2005 http://www.dtic.mil/dtic/tr/fulltext/u2/a442891.pdf Vietnam: Not surprisingly, the expansion of the scout dog program strained the procurement process’s ability to acquire the sufficient numbers. This problem could be attributed to a high rejection rate of 30 to 50 percent of the potential canine recruits. Competition with civilians and private security firms also hampered military procurement (Lemish, 1996, p. 184). …The USALWL contracted a civilian company to establish a mine detection program. The civilian company that was contracted by USALWL was called Behavior Systems, Inc. (BSI) which, according to Perry Money, a former Marines Corps handler of a BSI dog, deployed 56 Army dogs in 1969 and 28 Marine Corps dogs in 1970. The training doctrine was written a nd administered by two civilians who, at the time, held Master’s Degrees in Animal Behavioral Psychology. BSI initially trained fourteen dogs to detect mines, booby traps, and trips wires, and another fourteen to detect and locate tunnels only. Each dog produced by BSI cost approximately$10,000 (Lemish, p. 201). … Perry Money’s assessment of the BSI program is that, You get what you pay for, which was approximately $15,000 per dog, an amount somewhat different from Lemish’s figure. [~1974, so ~$51-$76k now] …Other programs evolved as offshoots of the Vietnam Scout Dog Program. One was the Superdog Program as part of the Biosensor Research project. This program was an attempt to selectively breed dogs with fewer health problems, thereby increasing the length of use of the dog along with the development of a superior ambush detection dog (Lemish, p. 216). The program involved a range of people from different career fields involved. Nothing conclusive appears to have been published or disseminated about the experiment. At first glance, it might appear that Lackland AFB’s puppy program has similar objectives today. However, the puppy program seems much more a response to continual procurement issues. Appendix Dog heritabilities Canine Behavioral Genetics - A Review, Mackenzie 1986 Variable Proportion Posture in Pavlov stand 0.43 Investigative behavior in Pavlov stand 0.46 Escape attempts while in Pavlov stand 0.56 Human avoidance and vocalization at 5 weeks 0.59 Playful fighting at 13-15 weeks 0.42 Leash fighting 0.77 Docility during sit-training 0.48 Running time for long barrier 0.78 Vocalization on U-shaped barrier 0.47 Table 2: Proportion of total variance due to breed differences between Basenjis and Cocker Spaniels (after Scott and Fuller, 1965) …G. Geiger investigated the breeding-book of Dachshunds in Germany in 1973 and found the scores better distributed than the data studied by Sacher, perhaps due to the 12-point system used as opposed to the 4-point system used in the pointer prize classes. He conducted a three-level nested 379 analysis of variance on 1463 full- and half-sib progeny of 21 sires. In contrast to the earlier findings of Humphrey and Warner (1934), King (1954) and Mahut (1958), his results showed maternal effects but no effect due to sex. The heritabilities are shown in Table III (Geiger, 1973, cited in Pfleiderer-Hogner, 1979). Trait Sire Dam Hare tracking 0.03 0.46 Nose 0.01 0.39 Seek 0.00 0.41 Obedience 0.01 0.19 Table 3: Heritability estimates in Dachshunds (after Geiger, 1973) A second study of additive genetic variation in 1973 came from the Army Dog Training Center in Solleftea, Sweden. C. Reuterwall and N. Ryman reported on their study of 958 German Shepherds from 29 sires. The 8 behavioral traits studied were labeled A-H: • Trait A was termed Affability (tested by having an unknown person con front the dog); • Trait B was termed Disposition for Self Defense (tested by having an unknown person attack the dog); • Trait C was termed Disposition for Self Defense and Defense of Handler (tested by having an unknown person attack the dog and handler); • Trait D was termed Disposition for Fighting in a Playful Manner (tested by asking the dog to fight for a sleeve or stick); • Trait E was termed Courage (tested by having a man-shaped figure approach the dog); • Trait F was termed Ability to Meet with Sudden Strong Auditory Disturbance (tested by firing shots at some distance and making a noise with tin cans just behind the dog); • Trait G was termed Disposition for Forgetting Unpleasant Incidents (tested by scaring the dog at a certain place and then asking the dog to pass the place again); • Trait H was termed Adaptiveness to Different Situations and Environments (tested by observations during the other parts of the test). In contrast to Geiger’s findings, Reuterwall and Ryman reported significant differences between the sexes, males handling noise (Trait F) better and exhibiting more controlled defense (part of Trait C) and playful fighting (Trait D). Sex differences had also been noted by Humphrey and Warner (1934), King (1954) and Mahut (1958). Reuterwall and Ryman noted that, in all 380 the traits studied, the additive genetic variation was small (Reuterwall and Ryman, 1973). The heritability estimates listed in Table IV were reported by Willis based on the information found in Reuterwall and Ryman (Willis, 1977). It should be noted that the scores used by Reuterwall and Ryman were transformed and extremely complex. Some workers in Sweden today, working on the genetics of the breeding program at the Statens Hundskola, feel that the findings of Reuterwall and Ryman’s study are based on scores too complex to have much meaning (L. Falt, personal communication, 1982). Trait Males Females A [Affability] 0.17 0.09 B [Disposition for self-defense] 0.11 0.26 C [Disposition for self-defense and defense of handler] 0.04 0.16 D [Disposition for fighting in a playful manner] 0.16 0.21 E [Courage] 0.05 0.13 F [Ability to meet with sudden strong auditory disturbance] -0.04 0.15 G [Disposition for forgetting unpleasant incidents] 0.10 0.17 H [Adaptiveness to different situations and environments ] 0.00 0.04 Table 4: Heritabilities in German Shepherds (after Reuterwall and Ryman, 1973) The next year, M.E. Goddard and R.G. Beilharz stated their belief that fearfulness and dog distraction were heritable in Australian guide dogs (Goddard and Beilharz, 1974). In 1982, Goddard and Beilharz reported further on the genetics of Australian guide dogs…. Fearfulness emerged as the most important and most highly heritable component of success. Estimates of heritabilities based on scores of 394 Labrador Retrievers computed from sire components, dam components and the two combined are listed in Table V (Goddard and Beilharz, 1982). In contrast to reports by Scott and Bielfelt (1976), Geiger (1973) and Scott and Fuller (1965), no strong maternal effects were evident (Goddard and Beilharz, 1982) Trait Sire Dam Combined Success 0.46 0.42 0.44 Fear 0.67 0.25 0.46 Dog distraction -0.04 0.23 0.09 Excitability 0.00 0.17 0.09 Health 0.40 0.10 0.25 Hip dysplasia 0.08 0.20 0.14 Table 5: Heritability estimates in Australian Labradors (after Goddard and Beilharz, 1982) …Estimates of heritabilities based on scores of 249 Labrador Retrievers, calculated from combined sire and dam components, are listed in Table VI (Goddard and Beilharz, 1983). Nervousness had the highest heritability and was the only trait with a significant sire component. Estimates of genetic correlations between the traits are listed in Table VII (Goddard and Beilharz, 1983). In contrast to other workers (Castleberry et al., 1976; Bartlett, 1976; Rosberg and Olausson, 1976), Goddard and Beilharz (1983) found no negative correlations between important traits. However, they did not list correlations for hip dysplasia. They also noted the importance of sex; females being more fearful and distracted by scents but less aggressive and distracted by dogs than males. Sex differences were also noted by Humphrey and Warner (1934), King {1954), Mahut (1958), Reuterwall and Ryman (1973) and Pfleiderer-HSgner {1979). G. Queinnec, B. Queinnec and R. Darre reported on their work with French racing greyhounds (Queinnec et al., 1974). Breeding values for greyhounds were based 40% on the animal’s own performance and 60% on the performance of its progeny, both over 3 racing seasons to account for repeatability Trait Heritability Nervousness (N) 0.58 Suspicion (S) 0.10 Concentration (C) 0.28 Willingness (W) 0.22 Distraction (D) 0.08 Dog distraction (DD) 0.27 Nose distraction (ND) 0.00 Sound-shy (SS) 0.14 Hearing sensitivity (HS) 0.00 Body sensitivity (BS) 0.30 Table 6: Heritability estimates in Australian Labradors (after Goddard and Beilharz, 1983) In 1975, the U.S. Army Biosensor Project reported a heritability estimate of 0.70 for their intermediate temperament evaluations. They also stated their intention to use heritability estimates of both hip dysplasia (previously estimated in their colony as 0.22) and temperament in the calculation of breeding values (Castleberry et al., 1975). The following year, they reported the first known estimate of the genetic correlation between temperament and hip dysplasia (considered by many to be the two major problems in breeding dogs for military or police work). Before listing the estimate, they noted that previous dysplasia-free litters had shown undesirable temperaments. Their estimate of the phenotypic correlation between the two traits was -0.25 and that of the genetic correlation was -0.35 (Castleberry et al., 1976). In 1976, C.R. Bartlett reported heritabilities and genetic correlations between traits studied in American guide dogs. The traits listed were hip dysplasia, body sensitivity (judged by how hard a jerk on the choke-chain leash the new dog could tolerate; low scores indicating a lack of sensitivity), ear sensitivity (judged by how loud a vocal correction the new dog required; low scores indicating lack of sensitivity), nose (olfactory acuity leading to distraction problems for all but the best trainers; low scores indicating greatest use of the nose), intelligence (the ability of the dog to understand things from its own viewpoint, not implying a willingness to obey; low scores indicating great intelligence, which may be a problem to all but the best trainers), willingness {willingness to do what the dog’s master asks of it, regardless of distractions; low scores indicating the most willing dogs), energy (activity versus laziness; low scores indicating active, energetic dogs), self right (the belief of the dog that it has a right to be where it is; negative scores indicating a tendency to give way to another), confidence (confidence shown with strange people or in strange environments; low scores indicating more confident dogs), fighting instinct (tendency to fight; low positive scores indicating the tendency to avoid fights, negative scores indicating even less tendency to fight, passing into submission) and protective instinct (a desire of the dog to protect its own; low positive scores indicating a dog which will speak if a stranger approaches its master with menace, but will not fight to protect the master). Heritability estimates of these traits, based on over 700 records for males and over 1000 records for females, both calculated by paternal half-sib analysis, are listed in Table VIII (Bartlett, 1976) Trait Males Females Combined Hips 0.72 0.46 0.54 Body sensitivity 0.26 0.05 0.10 Ear sensitivity 0.49 0.14 0.25 Nose 0.30 0.05 0.12 Intelligence 0.17 -0.07 -0.06 Willingness -0.14 -0.04 -0.03 Energy -0.03 0.06 0.05 Self-right 0.15 0.25 0.22 Confidence 0.04 0.26 0.16 Fighting instinct -0.05 -0.08 -0.04 Protective instinct -0.21 -0.13 -0.12 Table 8: Heritability estimates in American guide dogs (after Bartlett, 1976) Rosberg and Olausson reported low heritability estimates for mental traits in the dogs at the Swedish Army Dog Center in Solleftea, Sweden. All dogs included in the study were German Shepherds. Phenotypic correlations between the mental traits they were studying and hip dysplasia were small, but negative. Genetic correlations were negative, ranging up to -0.55, but the authors felt they were unreliable due to problems with the material studied (Rosberg and Olausson, 1976). A study of the genetics of American guide dogs was completed in 1976 by C.J. Pfaffenberger, J.P. Scott, J.L. Fuller, B.E. Ginsburg and S.W. Bielfelt. They followed up Scott and Fuller’s (1965) work in behavior and obtained estimates of heritability for their puppy tests. The traits reported by Scott and Bielfelt (1976} in their chapter on analysis of the puppy-testing program included the following: sit (three repetitions of a forced sit with a vocal command}; come (five repetitions of the handler moving away, kneeling down, calling the puppy by name, followed by the command come while clapping the hands); fetch (three repetitions of playful retrieving with vocal command); trained response (a complex score, indicating if the puppy was afraid of the tester or not, was over-excited or cooperated calmly, did or did not pay attention to moving objects, adjusted slowly or readily to the new environment, showed no curiosity or was curious about new objects and people, did or did not remember previous experience, tried to do what the tester wanted or not, and showed persistence or not in performing a task); willing in training (also a complex score, indicating if the puppy was fear386 ful or at ease, afraid to move or moved freely, was indifferent or friendly to the tester, was unresponsive or responsive to encouragement, urinated or was continent, was upset by the new situation or was confident, and was obstinate or willing in its responses); body sensitivity (another complex score, indicating if the puppy stood erect or cowered, turned head away or not, looked at or away from the tester, showed pain by action or not, came back after pain or attempted to escape, tucked in the tail or not, wagged tail or not after pain, and growled or not when in pain); ear sensitivity (similar to body sensitivity, except in relation to sound instead of pain); new-experience response (similar to trained response, but this time an emotional response to novel stimuli, not training); willing in new experience (similar to willing in training, except related to novel stimuli instead of training); traffic (indicates if puppy can avoid a moving and stationary cart without becoming fearful); footing-crossing (indicates if puppy noticed differences in footing between curbs and metal patches in the sidewalk); closeness {how close the puppy passed to obstructions); heel (how well the puppy accepted leash training). Eleven of the 13 traits, whose heritability estimates are listed in Table XI, had dam components much larger than the sire components, indicating strong maternal effects (Scott and Bielfelt, 1976). This agrees with the findings of Scott and Fuller {1965) and Geiger {1973). As part of the same study, J.L. Fuller examined the relationship between physical measurements and behavior. Once again, no substantial correlations were found (Fuller, 1976). Trait Heritability Sit 0.06 Come 0.14 Fetch 0.24 Trained response 0.08 Willing in training 0.12 Body sensitivity 0.16 Ear sensitivity 0.00 New-experience response 0.06 Willing new experience 0.24 Traffic 0.12 Footing-crossing 0.06 Closeness 0.04 Heel 0.10 Table 11: Heritability estimates for California guide dogs (after Scott and Bielfelt, 1976) Comparing Scott and Fuller’s 1965 estimates with those of the U.S. Army Biosensor project (Castleberry et al., 1975), it seems possible that certain components of behavior may be highly heritable. The failure of other workers to find high estimates may indicate that such estimates are quite sensitive to the quality of the tests, size of the samples and statistical methodology. In 1979, M. Pfleiderer-HSgner estimated heritabilities of Schutzhund scores in Germany. She analyzed 2046 test results in 1291 German Shepherds from 37 sires, all tested animals being born in 1973. The four criteria studied were tracking, obedience, man-work and character. She found sex and the number of dogs competing in a given trial to be significant, but not age or month of trial. Sex differences were previously noted by Humphrey and Warner (1934), King (1954), Mahut (1958) and Reuterwall and Ryman (1973). Estimates of heritabilities from sire components, dam components and their combination are listed in Table XII (Pfleiderer-HSgner, 1979). Trait Sire Dam Combined Tracking 0.01 0.20 0.10 Obedience 0.04 0.13 0.09 Man-work 0.04 0.07 0.06 Character 0.05 0.17 0.12 Table 12: Heritability estimates for German Schutzhund scores (after Pfleiderer-H&gner, 1979) In 1982, L. F~ilt, L. Swenson and E. Wilsson reported their unpublished work on heritability estimates for behavioral traits studied at the National Dog School (Statens Hundskola) in Solleftea, Sweden. The traits studied in 8-week-old German Shepherd puppies included: yelp (time from first separation from litter to first distress call); shriek (time from the same separation to the first serious, emphatic distress call); contact 1 (tendency to approach a strange person in a strange place after separation); fetch (pursue a ball and pick it up in the mouth); retrieve (bringing the ball back after picking it up); 389 reaction (to a strange object in a strange place); social competition (actually a form of tug-of-war); activity (number of squares entered when left in a marked arena); contact 2 (time spent near a strange person sitting passively in a chair in the middle of the marked arena); exploratory behavior (number of visits to strange objects placed in the corners of the marked arena). Estimates of heritabilities for the traits, calculated from sire components and dam components separately, are listed in Table XIII (F~ilt et al., 1982). Although some specific behaviors had low heritability estimates, others had quite high estimates. Trait Sire Dam Yelp 0.66 0.73 Shriek 0.22 0.71 Contact 1 0.77 1.01 Fetch 0.73 0.10 Retrieve 0.19 0.51 Reaction 0.09 1.06 Social competition 0.11 0.76 Activity 0.43 0.76 Contact 2 0.05 1.11 Exploratory behavior 0.31 0.83 Table 13: Heritability estimates for Swedish German Shepherds (after Felt et at., 1982) …They felt that improved training and upbringing were as important as genetics in producing good behavior. Since the first-generation hybrids performed better than either of their pure-bred parents in problem-solving situations, Scott and Fuller recommended that cross-breds be considered as working dogs, provided that the pure-bred lines were properly maintained. Maintenance of the pure-bred lines seems important since they stated that the heterosis (hybrid vigor) lasted only for one generation. Consequently, inter-breeding of the hybrids should not result in any improvement in problem-solving ability. They also recommended against breeding one champion sire to many bitches, since they felt that good breeding programs need to consider multiple criteria to be effective (Scott and Fuller, 1965). measurement error and heritability low repeatability: http://aura.abdn.ac.uk/bitstream/handle/2164/11022/MS_revised_2nd_revision_FINAL.pdf?sequence=1 Heritability of behavioural traits in domestic dogs: A meta-analysis, Hradecká 2015 /docs/genetics/heritable/2015-hradecka.pdf : global heritabilities: 0.15/0.10/0.15/0.09/0.12 Psychical traits: Belgian Shepherd Dog, 0.13; German Shepherd Dog: 0.12; Labrador Retriever: 0.07 Moreover, evaluations of the behavioural traits are often difficult due to the lack of testing repeatability between and also within judges. Performance testing is usually subjective as significantly different scores are given by the judges as shown, for example, in Finnish Spitz (Karjalainen et al., 1996). - Karjalainen et al 1996. Environmental effects and genetic parameters for measurements of hunting performance in the Finnish Spitz. J. Anim. Breed. Genet. 113, 525-534. https://www.gwern.net/docs/genetics/correlation/1996-karjalainen.pdf - test-retests are all r=0.10-0.20! terrible! • Willis 1995, Genetic aspects of dog behaviour with particular reference to working ability /docs/genetics/heritable/1995-willis.pdf • Houpt, K.A., Willis, M.B., 2001. Genetics of behaviour /docs/genetics/heritable/2001-houpt.pdf • Takeuchi & Houpt 2003, Behavior genetics /docs/genetics/heritable/2003-takeuchi.pdf • Houpt 2007, Review article genetics of canine behavior • Hall & Wynne 2012, The canid genome: behavioral geneticists’ best friend? • Jones , A. C. & Gosling , S. D. ( 2005 ). Temperament and personality in dogs (Canis familiaris): a review and evaluation of past research . Applied Animal Behaviour Science , 95 : 1 - 53 . • van den Berg , L. , Schilder , M. B. H. & Knol , B. W. ( 2003 ). Behavior genetics of canine aggression: behavioral phenotyping of golden retrievers by means of an aggression test. Behavior Genetics , 33 : 469-83 . /docs/psychology/2003-vandenberg.pdf • Van den Berg , L. , Schilder , M. B. , de Vries , H. , Leegwater , P. A. & van Oost , B. A. ( 2006 ). Phenotyping of aggressive behavior in golden retriever dogs with a questionnaire. Behavior Genetics , 36 : 882 - 902 . • Liinamo , A.-E. , van den Berg , L. , Leegwater , P. A. J. et al . ( 2007 ). Genetic variation in aggression-related traits in golden retriever dogs . Applied Animal Behavior Science , 104 : 95 - 106 • Lequarré , A. S. , Andersson , L. , André , C. et al . ( 2011 ). LUPA: a European initiative taking advantage of the canine genome architecture for unravelling complex disorders in both human and dogs . Veterinary Journal , 189 : 155-9 van den Berg 2017, Genetics of dog behavior /docs/genetics/heritable/2017-vandenberg.pdf The dog genetic studies reviewed in this chapter used more subjective phenotypic measures. Most heritability studies used phenotypes based on the behavior of dogs in test batteries. Jones and Gosling (2005) have reviewed studies of canine personality and noted that, In theory, test batteries were the closest to achieving objectivity, but in practice the levels of objectivity actually attained varied substantially. The molecular genetic studies mostly used even more subjective measures such as owner-report questionnaires and expert ratings (experts being veterinarians, trainers, or dog obedience judges). Owner and expert ratings may be influenced by a variety of factors other than the behavior of the dog, e.g. owner personality and expectations of typical dog behavior. Intuitively, the use of specific and objective metrics in genetic studies seems preferable. However, behavior of dogs in a test battery may not be representative of their behavior in everyday life and it is often unclear what exactly is being measured. Van den Berg and colleagues used three methods for measuring canine aggressive behavior: a behavioral test of the dog (van den Berg et al ., 2003), a questionnaire for the dog owner (van den Berg et al ., 2006 ), and a personal interview with the dog owner (van den Berg et al ., 2003 , 2006 ). The most promising heritability estimates (i.e. high heritability with low standard errors) were obtained for the owner impressions collected during the personal interview (Liinamo et al ., 2007 ). This is rather surprising because of the subjectivity of these phenotypes. Large coordinated projects, such as the European LUPA consortium, make an effort to clarify dog behavioral phenotypes by following standard procedures to describe dog behavior (Lequarré et al ., 2011 ). This is of great value for progress in canine behavioral genetics. Quantification and description of individual differences in behavior, or personality differences, is now well-established in the working dog literature. What is less well-known is the predictive relationship between particular dog behavioral traits (if any) and important working outcomes. Here we evaluate the validity of a dog behavioral test instrument given to military working dogs (MWDs) from the 341st Training Squadron, USA Department of Defense (DoD); the test instrument has been used historically to select dogs to be trained for deployment. A 15-item instrument was applied on three separate occasions prior to training in patrol and detection tasks, after which dogs were given patrol-only, detectiononly, or dual-certification status. On average, inter-rater reliability for all 15 items was high (mean=0.77), but within this overall pattern, some behavioral items showed lower interrater reliability at some time points (<0.40). Test-retest reliability for most (but not all) single item behaviors was strong (>0.50) across shorter test intervals, but decreased with increasing test interval (<0.40). Principal components analysis revealed four underlying dimensions that summarized test behavior, termed here object focus, sharpness, human focus, and search focus. These four aggregate behavioral traits also had the same pattern of short-, but not long-term test-retest reliability as that observed for single item behaviors. Prediction of certification outcomes using an independent test data set revealed that certification outcomes could not be predicted by breed, sex, or early test behaviors. However, prediction was improved by models that included two aggregate behavioral trait scores and three single item behaviors measured at the final test period, with 1 unit increases in these scores resulting in 1.7-2.8 increased odds of successful dual- and patrol-only certification outcomes. No improvements to odor-detection certification outcomes were made by any model. While only modest model improvements in prediction error were made by using behavioral parameters (2-7%), model predictions were based on data from dogs that had successfully completed all three test periods only, and therefore did not include data from dogs that were rejected during testing or training due to behavioral or medical reasons. Thus, future improvements to predictive models may be more substantial using independent predictors with less restrictions in range. Reports of the reliability and validity estimates of behavioral instruments currently used to select MWDs are scarce, and we discuss these results in terms of improving the efficiency by which working dog programs may select dogs for patrol and odor-detection duties using behavioral pre-screening instruments. … In many selection and training programs for police and detection dogs, more than half of the candidate dogs are rejected for behavioral reasons (Wilsson and Sundgren, 1997b; Slabbert and Odendaal, 1999; Maejima et al., 2007). Wilsson, E., Sundgren, P.-E., 1997b. The use of a behaviour test for the selection of dogs for service and breeding. I: Method of testing and evaluating test results in the adult dog, demands on different kinds of service dogs, sex and breed differences. Appl. Anim. Behav. Sci. 53, 279-2 - Slabbert & Odendaal 1999, Early prediction of adult police dog efficiency - a longitudinal study > Up to 70% of dogs that were bred at the South African Police Service Dog Breeding Centre (SAPSDBC) were not suitable for use. • Maejima et al 2007, Traits and genotypes may predict the successful training of drug detection dogs wiki/docs/genetics/selection/2007-maejima.pdf In Japan, approximately 30% of dogs that enter training programs to become drug detection dogs successfully complete training. … While the improvements in prediction observed here were small (2-7%), given the costs of purchasing, importing, housing, and training (approximately$18,500US per dog), this small percentage improvement results in a substantial potential savings.

The use of a behaviour test for selection of dogs for service and breeding. II. Heritability for tested parameters and effect of selection based on service dog characteristics https://www.appliedanimalbehaviour.com/article/S0168-1591(96)01175-6/abstract

Shyness-boldness predicts performance in working dogs, Svartberg http://www.svartbergs.se/pdf/Personality_workingdogs.pdf