This post replaces an early one that contained several errors. Its purpose is to explain my Primate IQ Scale. It also incorporates data from an important 2016 paper by Sandra et al, to it.
Deaner et al. 2006, were the first to establish that there is a universal general intelligence present across primate species. Although they were working with rank order information across many different tasks, they used a Bayesian approach that could render probable z scores. Unfortunately, there have been few follow-ups to this seminal meta-analysis by other authors, the next best being a meta-analysis of behaviour in the wild by Reader et al. 2011. This paper didn’t utilise laboratory tested cognitive performances under identical conditions, but it did confirm that a general intelligence factor was present, and the results could also be rendered into z scores. Of similar importance to Reader et al., is a 2016 paper by Sandra et al. Although this last paper used only one test (they observed the manipulation complexity in captive primate species), they looked at a wide variety of primates. The results correlated highly with those of Deaner et al. In the last case, data also includes humans. All three studies correlate so highly with absolute brain size (r = .8, .6 and .9 respectively), as to imply that log brain weight can also be used as an indirect measure of primate intelligence.
For thirteen genera, all four metrics are available. Eight of these are selected below, by way of example.
Primate (z(1), z(2), z(3), z(4))
1 Orangutans (1.75, 1.71, 1.12, 1.67)
2 Gorillas (.96, .87, 2.03, 1.71)
3 Macaque (.56, 1.62, .90, .73)
4 Guenon (.23, .14, .30, .33)
5 Gibbon (.11, -.79, -.31, .59)
6 brown lemur ( -.47, -.71, -.90, -.47)
7 squirrel monkey (-.94, -.79, -.48,-.45)
8 marmoset (-1.22, -.86, -.98, -1.14)
There is a significantly different set of primate genera used for each of these four metrics, and some sets are normed by species not genus, so even if they were in perfect agreement those z scores would still differ. For that reason it is also worth examining rank order. For the above eight genera it differs only slightly set by set, being 12345678, 13246(5=7)8, 21345768 and 21354768 for each of the four sets over the above primate sample respectively.
A meme has been going around that researchers are guilty of trying to measure other animals intelligence too much in the way they test humans. Unfortunately for us, the exact opposite is true, with few ever using a common test on both nonhumans and humans. Fortunately, the last two of our datasets are among the few objective metrics of cognition that contain us. This allows me to place atop the above list…
0 Humans (na, na, 2.36, 2.41)
The first dataset is a meta-analysis of all direct laboratory tests of primate intelligence that involve more than one genera. As such this acknowledges the importance of comparisons being made when the test conditions were identical. They found that the tests that met their conditions could be pooled into 30 procedures testing 9 different cognitive paradigms. These involved 24 genera.
The placings of some genera were calculated from only a single procedure (night monkey, brown lemur, mouse lemur, slow loris, fork-marked lemur, sirili, ruffed lemur), other placements were particularly secure with between 4 and 23 procedures (gracile Capuchin, guenon, gorilla, dwarf gibbon, ring-tailed lemur, macaque, chimpanzee, orangutan, squirrel monkey). Some day I will weight the values of each genera in this dataset accordingly, but for now I weighted all values derived from this meta-analysis as five times more significant than dataset 3 or 4.
The second dataset has never been claimed to be superior to the first, yet we have three reasons to use it. These are: it confirms the existence of a pan-primate general intelligence factor; it covers several genera not covered in the first dataset; and it uses animals in their natural environment, there being reason to believe that the cognitive abilities of primate species are differentially impacted by captivity, and the details of their raising. One complication is that the first paper only examined the gracile genus of Capuchin, but this second one lumps together both gracile and robust species. This is a problem, since all robust species are tool users, but few gracile species have ever been observed using tools, and those that do use them sparingly. This is hardly surprising since graciles have brains only half the size of robust species. Because this meta-analysis is most effected by the most intelligent behaviours observed, I have applied this figure to the robust genus only (Sapajus). All in all, this dataset still has great utility, and I have given it twice the weighting of those in the third and fourth dataset for my calculations.
The third dataset uses a non-linear scale, that is compiled in a way that gives a pseudo-linear result. A symptom of this is that the lowest theoretically possible z score is -1.25 by it. However, the correlation of this set to log brain weight is so high (r=0.9), that I treat the data as if it is a good linear reflection of IQ. That 0.9 value for r is so high as to require explanation, which I interpret as follows. The characteristic manipulation complexity of a primate can be thought of as a combination of two factors: an animal’s dexterity, which tends to increase with body size; and an animals intelligence, which we have strong reason to believe is closely aligned to the number of neurons in the forebrain. In primates, brain size is a descent proxy for each of those factors, thus it is possible for this metric to be more highly correlated to brain size than it is to either intelligence or body size alone. This is, however, only a single test, so I can only give it a weighting of 1.
The last dataset is simply log absolute brain size. Unfortunately, I have gained most of these values ad hoc, but they are averaged over many sources. Because very few of these values deviate more than 20% from the medium.value, I do not believe the nature of their origin problematic. The reasons I use brain size as a proxy for intelligence are twofold: in theory, if each neuron is equally active, then brain neuron number should dictate the maximum intelligence of a species, and brain volume reflects that number; and in practice, all the other datasets show that absolute primate brain size explains at least twice as much variance in intelligence as every other considered factor combined.
Setting the Units
A third way is to take the standard deviation within a species, and assume that the species average is the same as the human average (ie. 15 IQ points). Most importantly, this method also assumes that the primate g versus G debate is resolved, as I think it will be, in favour of g. This is a highly technical debate, that I may detail in another post, but also may not as it takes us far from our interest in cetacean intelligence. Currently I have no data on this method, but it introduces the fourth way.
The fourth way is already explained in the footnote on THAT page on my website, on which I may elaborate with further examples in the future. All I will say now is that this method does seem to point to a slightly lower conversion value than the first two methods mentioned.
In trying to attain units true to the human scale, the value I use here has effectively become frozen. With my current dataset, which now including 40 of the 72 primate genera, its value equates to 74 IQ/SD.
Setting the Zero
I find it incredible how few people understand the IQ system. They say a little knowledge is a dangerous thing, and with that in mind many know that ratio IQs were used in the early years of the previous century. That reinforces many wrong ideas, such as an IQ of zero equating with no intelligence. In theory, if testing categories always exist whereby some that failed at the previous highest level pass the next one down, then IQ can go to negative infinity. However, there really should be an absolute zero of intelligence, which I can explain as follows…
Primate brain size is almost proportional to its neuron number and, presumably, its synaptic density. If we equate one synapse with the capacity to make one regulatory decision, then a bacterium that has only one on-off regulatory gene would have an IQ close to -1500 on the human IQ scale. All bacteria have multiple regulatory genes (ie. are far smarter than this), but it gives us an idea of where the true zero of decision making may lie on such a scale. Adding 1500 to each figure, however, would also confuse our intuitive feel for the world’s smartest animals, and this is, after all, what we are most interested in here. For that reason, I have added 150 points to the human scale. This should put it in line with our intuition, since on it any animal over 100 tends to be one of those considered among the smart ones, and one over 150 is in line with our perceptions of an exceptional species. This makes it analogous to the way we tend to think IQ relates to humans, thus giving it a familiar feel. I hope that this scale may give three points in line with our expectations.
>150 an animal of exceptional intelligence
>100 an animal of high intelligence
0 an animal of average mammalian intelligence
That brings me to our final list with selected primates that illustrate their cognitive class…
My Primate Intelligence Scale
|300||modern Western adult|
|250||modern Western seven year old|
|250-230||humans of 60,000 BC|
Now we have the skeleton of an absolute intelligence scale, we can extrapolate out to non-primate groups and see how many are in desperate need of direct cognitive testing.