Preliminary evidence suggests that the aliens test has little, if any, correlation with IQ, at least among readers of this blog. The most likely reason for this is that the test so overloaded working memory that people were essentially just guessing.

As a result I have substantially revised the test in the hopes of reducing the demands on working memory and attention to detail, since the point of the test was to get at higher level, more conceptual ability.

I have reduced the number of aliens you have to compare from 12 to nine and also reduced the number of ways people can be tricked into miscategorizing the aliens.

Because the point of this test is not just to measure your intelligence, but more importantly, to show that there is a correct way to organize the natural world, whether we’re dividing categories as broad as plants and animals or as narrow as racial groups.

[Take the new version of the aliens test here. You can register with a fake name; email optional]

Like the Pairs test, Aliens is a very culture reduced test of conceptual reasoning. It is culture reduced because it’s not about America or any other culture on Earth, but rather about 12 “men” on an imaginary planet. Because no one has ever been to this planet, it hopefully measures largely novel problem solving or fluid ability as it is sometimes called.

Like the Pairs test, this test measures the ability to categorize but unlike the Pairs test, one must also identify the number of categories and even subcategories, a skill I’ve never before seen measured on an IQ test, but one that is dear to the author’s heart.

You can take the test here. Feel free to register with a fake name (email optional).

Over 60 people have taken the new version of the Pairs and of these, nine reported scores on the cubes test. The mean Pairs score of this sub-group was 7.6 (SD = 1.74) and the mean cubes IQ was 129 (SD = 16); U.S. norms. By equating the mean and SD, I was able to create a rough equivalency between between Pairs raw score and IQ (see chart below).

A random sample of all people who got valid scores on the new version of the Pairs test (excluding those who clearly misunderstood the instructions or quit after less than about a minute) showed their scores to be slightly lower and more variable than the subset who reported cubes scores. By assuming they had the same IQ distribution as all who took the defunct version of the Pairs test, I was able to create an equivalency between both versions which allowed me to equate the old version to IQ too.

The old version has a lower floor and much higher ceiling but this came at the cost of having invalid items (hard to hit the ceiling when certain questions don’t have a valid answer). But even after revising the invalid items, the ceiling remains ridiculously high.

While I’d like to think this test can discriminate above the one in 30 million level, realistically the high ceiling is likely an artifact of low reliability. When test items are flawed, they tend to correlate weakly making it freakishly rare for anyone to hit the ceiling.

Another red flag is the test only correlated 0.24 with self-reported scores on the cubes test though given the small sample size (n = 9) it isn’t quite panic time. On the bright side, the old version of the Pairs correlated perfectly with self-reported PATMA scores (n = 4).

Pairs raw score	Old version (taken before 6 PM Eastern Time, on April 10, 2024); IQ equivalent (U.S. norms)	New version (taken after 6 PM Eastern Time, April 10, 2024); IQ equivalent (U.S. norms)
0	50	59
1	61	68
2	73	77
3	84	87
4	96	96
5	107	105
6	118	114
7	130	123
8	141	133
9	152	142
10	164	151
11	175	160
12	187	169
13	187+	178
14	187+	188
15	187+	188+
16	187+	188+

[UPDATE: 8:09 PM EASTERN STANDARD TIME: After correcting some errors on this test and article discovered by Teffec and Melo, it is now back online. If you did not get a chance to take it, you can do so now. You can register with a fake name if you want (email optional)]

I created the Pairs test because of a void I had noticed in the field of psychometrics. Although there are tests that are very culture reduced (performance subtests on the Wechsler) most of these seemed to load heavily on spatial ability. What if your testing someone who doesn’t speak English or may have never learned to read in any language, but who nonetheless has a naturally high verbal IQ?

What is needed, it seemed to me, is a test that gets at verbal, semantic, or symbolic modes of thinking but without using language. I immediately began thinking of some of the more fluid tests of verbal ability and wondered, what if we could recreate these tests with pictures instead of words.

At first I began thinking of the odd man out test (which of these is not like the others: yellow, red, blue, seven, green?) and thought, this could easily be made more culture fair by translating it into picture form. Indeed there are IQ tests like this discussed in one of Jensen’s book, but I quickly became discouraged by the error introduced by guessing. To solve this problem, I created the Pairs test where instead of guessing which of the 5 pictures is not like the others (1 in 5 chance of guessing right), you have to guess which pair of the 5 pictures ARE alike (1 in 10 chance of guessing right).

It’s unclear whether this test should be considered a measure of Verbal or Performance IQ since the medium is visual but the type of thinking required is more semantic. Either way it nicely complements the Information subest which is arguably the most crystallized subtest on the PAIS, while Pairs is the most fluid. A large Pairs > Information gap is probably indicative of cultural or educational deprivation while a large Information > Pairs gap might suggest schizophrenia, autism, dementia or brain damage though far more research is needed to validate such speculation.

The test is by no means 100% culture fair. You couldn’t give it to a hunter-gatherer. But it’s arguably culture fair enough to give to someone who never attended school, as long as they grew up in a city.

As mentioned, you can take the test here.

Of the 118 or so readers who have taken the cubes test, 14 have reported credible Wechsler full-scale IQs in Canada/the U.S./the U.K. within the last decade. These are are their scores:

cubes raw score	Wechsler IQ	Age when taking the cubes test
10	144	Thirties
6	120	Twenties
9	133	Late teens
7	133	Twenties
8	137	Fifties
11	149	Twenties
10	160+	Twenties
17	160+	Twenties
8	93	Below 16
5	143	Undisclosed
2	103	Thirties
7	120	Twenties
9	142	Twenties
11	124	Twenties

There mean cube score of the sample is 8.57 (SD = 3.46) and the mean Wechsler IQ is 133 (SD = 19.5). By equating these statistics, cube scores can be converted to IQ equivalents:

cubes raw score	IQ equivalent (U.S. norms)
17	181
16	175
15	169
14	163
13	158
12	152
11	147
10	141
9	135
8	130
7	124
6	119
5	113
4	107
3	102
2	96
1	91
0	85

When I ran a regression equation, I found that each year older than 25 decreased cubes IQ by 0.5 points holding Wechsler IQ constant, thus I tentatively suggest older people add a bonus of 0.5 IQ points for each year above 25. However this suggests either massive age related decline and/or the Flynn effect and more research with older subjects is needed because this bonus seems a little generous.

The correlation between cubes and self-reported Wechsler is 0.61 however this needs to be corrected for extended range (since the IQs of the sample are more variable than the general U.S. population).

Grady Towers supplied a formula for correcting for range restriction, which I assume I can use since extended range is simply negative range restriction:

Correcting for extended range reduces the correlation from 0.61 to 0.52. However we should also correct for the fact that the Wechsler IQs were taken as much as a decade ago and thus would only correlate perhaps 0.8 with what their Wechsler IQ would have been on the day they took the cubes test. Dividing 0.52 by 0.8 gives 0.65 which is probably a good estimate for the cube test’s g loading.

So what the above chart is showing us is that Ivy League students in general have about a 12% chance of getting a very high paying job after college, however for the ones with high IQ, the odds jump to 16%. Surprisingly being a legacy or having strong non-academic traits (charisma) doesn’t help any. Being an athlete provides a small benefit, probably because it makes you better looking and more energetic, but IQ is a better predictor than all three other factors combined.

This shows that even among Ivy League grads, smart people are getting ahead naturally instead of just artificially getting propped up by the colleges they attend.