Comment by somenameforme | Hacker News Reader

Comment by somenameforme | Hacker News Reader

Parent Full thread

There was a famous study that tried to test this - the Perry Preschool Study. [1] Basically they enlisted a number of high risk children - black, low iq, low income children. Half were placed into a high quality specialized preschool program (that lasted two years for 2.5 hours a day) with small class sizes, half were not, and they followed what happened over the next 40 years. The results were definitely impactful, but not the sort of major turn around one might hope for.

So for instance 55% of the control group ended up being arrested 5+ times by age 40, while 'only' 36% of the experiment group did. I think the thing this demonstrates is that intervention can help, but is also insufficient alone. Students who are in a sufficiently high risk scenario need ongoing support and treatment that they're not going to receive at a normal public institution. And not only that but they will remain disproportionately disruptive to other student's educations at normal institutions, even with years of ongoing care.

[1] - https://en.wikipedia.org/wiki/HighScope (overview)

[1] - https://highscope.org/wp-content/uploads/2024/07/perry-presc... (detailed paper)

I'm surprised that 2.5 hours a day for 2 years was enough to make that big a difference on outcomes through age 40. Like... damn, that's a big effect!

In Germany children only spend between 5.5 to 6 hours at school per day. You‘ve raised that amount to 8 hours now and the outcomes are not that much better since the number represents being arrested at least five times. If you get arrested four times, you would be considered a model student.

Reading the actual study, this appears to be a preschool program of 2.5 hours minimum, not adding on to an existing school day. There are also a lot more details about outcomes and they're wildly positive for an intervention period of just two years. The authors estimate the ROI (from increased productivity and savings on various costs) at an astounding 16x.

There are way more metrics in there, including more crime stats. The one somenameforme chose to highlight has a ton of ambiguity, leaving it open to the reader to guess that maybe all the program participants were arrested merely four times by age 40, so in fact this program sucks (plus somenameforme's scare-quotes on "only"), but the paper itself contains far more information and paints a clear picture of outstanding success for a relatively small intervention. Somenameforme's characterization of the study doesn't match the contents.

If that's the evidence a person's citing, the evidence they've cited is screaming "this works great", not the opposite, as implied. It may still not be true, but if so... cite different evidence to support that, because this study says this intervention was wildly successful.

Make sure you're reading the study and not just glancing at their charts. They try to present their data positively to the point that it can be quite misleading. For instance you might see things like 67% of the experiment group having an IQ of 90+ at age 5, contrasted against only 28% of the control group.

But read further down on the details and that difference disappeared almost immediately after the end of the intervention. It follows in line with a well known fact that childhood IQ is primarily driven by environmental factors whereas adolescent and especially adult IQ is primarily driven by the IQ of your parents - paradoxically, strengths or deficiencies in earlier life notwithstanding.

And their decision to set the baseline for arrests at 5+ is obviously doing something akin to p-hacking. It makes it clear that near 100% of the entire sample (males at least) ended up in prison, likely multiple times. The ROI from the program had nothing to do with increased productivity - it was driven almost entirely by less time spent in prison. It led to the interesting fact that 93% of the ROI came from males, precisely because the females had a much lower baseline criminality rate.

In a nutshell, the main benefit of the program was reducing the criminality rate of the experimental group to a level that is still orders of magnitude higher than for society at large. That is a good thing, but it also emphasizes that something like this would only be the beginning of special care needed to try to ensure these sort of people could live remotely decent lives.

http://bactra.org/weblog/494.html

The person who wrote that site spent quite a lot of time writing, yet unfortunately little reading. Heritability is, by definition, the degree of variation in a trait, within a population, due to genetic variation. The heritability of an accent is zero.

One clever way this is measured is twin studies, which also are not what most people, particularly those who prefer to write more than read, think. You don't search for twins separated at birth, but instead compare the differences in a trait between identical and non-identical twins. If the variation is greater, then the trait is generally significantly heritable. So for example - height would be an obvious one. By contrast the variation in accent between identical and non-identical twins would be zero.

The person who wrote that site is Cosma Shalizi, who very certainly knows what "heritability" is. Unfortunately, you appear not to. "Heritability" is simply the ratio of genetic variance to phenotypical variance. It's not genetic causality. Whether or not you wear lipstick: highly heritable. The number of fingers on your hands: not heritable.

So it's a blog from some guy with no background in genetics. Your definition is correct, as is your statement that it's not genetic causality. But to discuss heritability you need to understand the most typical, and reliable, way it's assessed. That would immediately clarify to you why lipstick wearing (or your accent) is not heritable, yet the number of digits you have (at least at birth) most certainly is. Here [1] is Wiki's take. You can also pick up any textbook on genetics.

[1] - https://en.wikipedia.org/wiki/Twin_study

I don't think "Cosma Shalizi doesn't know what he's talking about" is a good hill to die on, and you've now expanded your portfolio of opponents to Ned Block, from who I shoplifted the heritability point.

Direct genetic causality is not the only mechanism through which genes select for phenotypical traits. Genes also select and interact with the environment.

A person you respect in one field is not necessarily all-knowing within that field and, most certainly, not outside of it. This is especially true on topics that become politicized. This is not just because of the 'our side' vs 'their side' stuff, but because these issues can and have destroyed the careers of high profile people who adopt the wrong opinion.

Unlike the individuals you have cited, James Watson is a geneticist, spent his entire life studying and working on genetics, and in fact was even the person who discovered the structure of DNA. But because of his views on the genetic aspects of IQ (which inherently becomes intertwined into race, as race is just shared genetic ancestry), he was completely demonized, his career destroyed, and various honors revoked. Higher profile people speaking on these topics publicly know this all too well, so it mostly just turns into cheap virtue signaling as opposed to adding some genuine insight.

In your case, the examples they've offered are simply wrong, as would be immediately apparent with the most typical method of measuring heritability!

You're irritated because I gave you an output of the broad-sense heritability statistic that conflicts with your intuitive understanding of what "heritability" means. Now you understand how people feel when commenters randomly throw around the term "heritability" with respect to cognitive ability.

This is a "not even wrong" situation. Is cognitive ability significantly genetically determined? Maybe, maybe not. A broad heritability statistic from a twin study isn't going to resolve the question.

Here's a good link for you:

http://bactra.org/weblog/520.html

I promise, the author has studied and thought more carefully about the question than we have.

Fair warning: you would not be happier if I cited a molecular geneticist on this subject. Your argument gets even harder to sustain once you bring GWAS into the picture.

I'm not at all irritated besides the fact that you're relying on examples that simply are incorrect, and instead of responding to this issue in any way you're linking to walls of text from somebody who (1) has made plainly false statements on the topic already and (2) has literally 0 qualification in the field whatsoever.

It'd be akin to arguing to somebody who wants to claim the Moon landing was faked, and after the rather straight forward rebuttal of their argument links to some blog in the tens of thousands of words from some statistician they claim is "very smart." It's silly.

Imagine the moral dilemma of having to choose which kid goes in which group

For the experiment, you don't want it to be a "moral dilemma" at all.

If the group-splitting decisions are made by humans, it inevitably introduces a systematic bias. That bias then will show up in the outcomes, and confound the very data you got out of your way to gather.

The easiest way to avoid that is to split the groups randomly.

Reply on news.ycombinator.com