site banner

Culture War Roundup for the week of October 24, 2022

This weekly roundup thread is intended for all culture war posts. 'Culture war' is vaguely defined, but it basically means controversial issues that fall along set tribal lines. Arguments over culture war issues generate a lot of heat and little light, and few deeply entrenched people ever change their minds. This thread is for voicing opinions and analyzing the state of the discussion while trying to optimize for light over heat.

Optimistically, we think that engaging with people you disagree with is worth your time, and so is being nice! Pessimistically, there are many dynamics that can lead discussions on Culture War topics to become unproductive. There's a human tendency to divide along tribal lines, praising your ingroup and vilifying your outgroup - and if you think you find it easy to criticize your ingroup, then it may be that your outgroup is not who you think it is. Extremists with opposing positions can feed off each other, highlighting each other's worst points to justify their own angry rhetoric, which becomes in turn a new example of bad behavior for the other side to highlight.

We would like to avoid these negative dynamics. Accordingly, we ask that you do not use this thread for waging the Culture War. Examples of waging the Culture War:

  • Shaming.

  • Attempting to 'build consensus' or enforce ideological conformity.

  • Making sweeping generalizations to vilify a group you dislike.

  • Recruiting for a cause.

  • Posting links that could be summarized as 'Boo outgroup!' Basically, if your content is 'Can you believe what Those People did this week?' then you should either refrain from posting, or do some very patient work to contextualize and/or steel-man the relevant viewpoint.

In general, you should argue to understand, not to win. This thread is not territory to be claimed by one group or another; indeed, the aim is to have many different viewpoints represented here. Thus, we also ask that you follow some guidelines:

  • Speak plainly. Avoid sarcasm and mockery. When disagreeing with someone, state your objections explicitly.

  • Be as precise and charitable as you can. Don't paraphrase unflatteringly.

  • Don't imply that someone said something they did not say, even if you think it follows from what they said.

  • Write like everyone is reading and you want them to be included in the discussion.

On an ad hoc basis, the mods will try to compile a list of the best posts/comments from the previous week, posted in Quality Contribution threads and archived at /r/TheThread. You may nominate a comment for this list by clicking on 'report' at the bottom of the post and typing 'Actually a quality contribution' as the report reason.

20
Jump in the discussion.

No email address required.

Phenotypic null hypothesis isn't an obscure methodological limitation, it is a fundamental property of concepts like "heritability" and "genetic correlation"

To be blunt: «Phenotypic null hypothesis [for the genetics of personality]» is a 2013 review paper by Eric Turkheimer, with 162 citations.

Turkheimer, like Lewontin before him with his «apportionment» and, more to the point, with his corn plants (cited by Ned Block here), is making a mountain out of a common-sensical molehill, in his case to bury the monster of HBD implications that have sprang forth (as he asserts, unreasonably) from his own First Law that he now seeks to reframe (as much is stated in the paper). Moreover, you both are sufficiently obscurantist and informal that it's hard to tell if you're gesturing at the same mountain:

The phenotypic null hypothesis integrates what we have learned from multivariate behavior genetics: Until demonstrated otherwise, complex heritable behavioral traits should be the result of psychological processes defined at a high level of analysis, rather than at the level of genes or neurons. Although such complex traits are never independent of genetic variation, they cannot be defined by genetic processes. Genes and behavior are a single entity, a single organism observed at two levels of analysis. Some traits are better understood using low-level concepts (genes, neurons, structures), whereas others require high-level constructs (organizations, algorithms, beliefs).

vs yours:

Put simply, the phenotypic null hypothesis is this: Heritability tells you that if you go up through the chain of causation, then you will often end up with genes. However, there may be many ways that variables can be connected to each other, and there’s no particular reason to expect that every step along the chain of causation from genes to outcomes is best thought of as biological.

Ok, and?

….The problem you talk about in your Substack post is real, although not nearly as damaging with regards to the sort of HBD beliefs most salient in the culture war and pro/anti-HBD debate (i.e. not the issue of bullied gays) as one could assume from careless reading of your initial post. (Besides, we have plenty of data such as admixture and, as of late, GWAS confirming simple additive model for the group difference – qualitatively similar to the case of height, not similar to gay-bullying and personality research). Measurement error is a fundamental problem; though as better-informed people remind us, there are methods, e.g. common pathway models, which help against it, and HBD research using those methods has yielded largely the same conclusions. It is also possible to directly test for AE vs ACE, precluding the sort of erroneous assumptions you warn against.

This is what is recommended is a textbook, literally, in crypto-HBDer Plomin's «Behavioral Genetics» 2012 edition (that is, anticipating Turkheimer's paper by a year). Consider p.157:

«In addition, recent work suggests that careful attention should be paid to claims of causality, measurement error, and environmental factors that can influence both the endophenotype and the final outcome (Kendler & Neale, 2010). … Another issue is that the goal of behavioral genetics is to understand pathways among genes, brain, and behavior. Genes found to be associated with brain phenotypes are important in terms of the brain level of analysis, but their usefulness for behavioral genetics depends on their relationship with behavior (Rasetti & Weinberger, 2011). In other words, when genes are found to be associated with brain traits, the extent to which the genes are associated with behavioral traits needs to be assessed rather than assumed».

Or in a section about model fitting, pp. 380-390:

« Although we will not follow the proof here, researchers have demonstrated that we cannot ask about additive genetic effects, dominance genetic effects, and shared environmental effects simultaneously if the only information we have is from MZ and DZ twins reared together. In virtually every circumstance, we will wish to retain the nonshared environmental variance component in the model. We wish to retain it partly because random measurement error is modeled as a nonshared environmental effect and we do not wish to have a model that assumes no measurement error (it is unlikely to fit very well). Most commonly, we would then model additive genetic variance and shared environmental variance. As mentioned earlier, such a model is called the ACE model […] Table A.2 shows that the AE model is unable to account for this particular set of observed values. Such a model is said to be underidentified. This condition is not necessarily problematic: In general, underidentified models are to be favored. Because a saturated model will always be able to fit the observed data perfectly, the goodness of fit does not really mean anything. However, if an underidentified model does fit the data, then we should take notice—it is not fitting out of mere statistical necessity».

Plomin practices what he preaches, too – here's an example of a paper.

So yeah, those are real problems with real countermeasures. «Phenotypic null hypothesis» is a meme, and there are apparently only two people forcing this meme.

Again, what Turkheimer is doing here is not different from Lewontin's tricks, which are known to have been motivated by politics. It's uncanny how similar these situations are. In the entry-level «Making sense of Heritability, pp. 60:-62:

Lewontin’s criticism of ANOVA is often presented as making a purely the- oretical point about inherent limitations in any attempt to derive causal conclusions from statistical data by using the analysis of variance. This interpretation is supported by the way he himself describes the upshot of his argument at the outset: “I will begin by saying some very obvious and elementary things about causes, but I will come thereby to some very annoying conclusions” … The strange thing about (a) is that, despite defending an extreme methodological claim that measurement of human norms of reaction is impossible, Lewontin neither discusses nor so much as mentions what were then most important contributions to the literature on that issue. At the time, the most sophisticated analysis of methodological prob- lems in human behavior genetics was undoubtedly the paper by Jinks and Fulker (1970), which was an important step in the development of powerful model-fitting methods that dominate the contemporary scene. Moreover, John Jinks and David Fulker did suggest how G–E interac- tions could be empirically detected without testing the same genotype in a variety of environments (see below), which directly contradicted Lewontin’s impossibility claim. It is hard to explain why Lewontin fails to address their argument, or at least inform the reader about this “landmark paper”…

By all means, do promote statistical literacy and demand intellectual honesty among HBDers, do shame us for our sloppiness. Keep discussing this in terms of phenotypic null hypothesis, if you must. But do not gaslight. This argument, and in this specific formulation, is not Behavioral Genetics 101, but your recent contribution to the debate, and it's more of a rhetorical nature.

Ok, so regarding this paper:

Plomin practices what he preaches, too – here's an example of a paper.

According to Plomin, the goal of the paper is to test the plausibility of evolutionary theories are about environmental sensitivity by using twin studies to look for heritability:

According to the recent evolutionary-inspired theories (i.e., differential susceptibility [1], biological sensitivity to context [2]), humans, like many other species [3], differ substantially in their sensitivity to contextual factors, with some more susceptible to environmental influences than others. Importantly, these theories suggest that heightened sensitivity predicts both the reactivity to adverse contexts as well as the propensity to benefit from supportive features of positive environments. In other words, sensitivity is proposed to influence the impact of environmental influences in a “for better and for worse” manner [4]. These prominent theories converge on the proposition that genetic factors play a significant role in individual differences in Environmental Sensitivity (ES) [1, 2, 5].

This is very much the sort of nonsense the phenotypic null hypothesis is an objection to. Everything is heritable, and we have good theoretical understanding of why that is. It is thus of no evidentiary value to find that things are heritable, and this shouldn't be treated as a confirmation of evolutionary theories, which destroys the whole point of the paper.

For example, children who scored higher on the HSC scale were found to benefit significantly more than less sensitive children from schoolbased resilience [16]

Not so relevant to the phenotypic null hypothesis and I haven't looked at this in detail as it's a citation of a different study, but the cited study makes me suspicious: They didn't find any main effect of the treatment, so this was a subgroup analysis of exactly the sort that Scott Alexander has warned me about.

[common pathway model for HSC]

I acknowledge that common pathway models/factor models can control for some types of measurement error in some scenarios, but it doesn't really seem to work for personality traits (and therefore not for HSC either, unless HSC is an unusual personality trait). The appropriate way to do this for personality data is multi-informant data, which tends to lead to way higher heritabilities for personality, indicating that a substantial proportion of the nonshared environment component is measurement error, even with naive common pathways.

[correlation matrix for personality traits]

Plomin finds that all the "good" personality traits are correlated, i.e. emotional stability, extraversion, openness, conscientiousness and agreeableness are all positively correlated with each other. The model he chooses to apply to those correlations assumes that these correlations are substantive, but I believe that is an inappropriate model.

Correlations between the Big Five personality traits within a single rater appear to reflect a "Halo"/"social desirability" bias factor. The way we can tell this is because it fails to correlate across informants. I.e. while it's true that you rating yourself as more extraverted correlates with you rating yourself as more conscientious, you rating yourself as more extraverted does not correlate with others rating you as more conscientious. See for instance this paper.

Also I believe it's well-established that the different subscales of HSC differ from each other in their correlations with the Big Five, and indeed he replicates that finding in the study. However, this pattern of correlations is incompatible with the notion that the correlations between HSC subscales and Big Five is mediated by the HSC common pathway, which makes his later models very strange.


In conclusion, the Plomin study you linked is a fractal of bad study design. In many ways it's a good example of the necessity to further popularize the phenotypic null hypothesis. However, the study also has severe flaws beyond the phenotypic null hypothesis. This is cruxy to me: if you can convince me that Plomin's study is good, then I will likely grant that I was wrong about my point about the phenotypic null hypothesis, but conversely I think Plomin's study is really bad and I think Turkheimer has to deal with an endless stream of studies that are equally as bad as Plomin's here, so I think this serves as an excellent case study that explains why Turkheimer is so bothered by behavioral genetics.

I rest my case.

Wait is the study even by Plomin? The authors listed are:

Elham Assary, Helena M. S. Zavos, Eva Krapohl, Robert Keers & Michael Pluess

None of whom seem to be Plomin.

Plomin practices what he preaches, too – here's an example of a paper.

I don't have time to respond to this right now, as I'm on my phone, but just quickly skimming it, it looks cruxy to me. I will respond once I get home.

Ok, and?

….The problem you talk about in your Substack post is real, although not nearly as damaging with regards to the sort of HBD beliefs most salient in the culture war and pro/anti-HBD debate (i.e. not the issue of bullied gays) as one could assume from careless reading of your initial post.

I'm not claiming it to be damaging to HBD beliefs, I'm claiming it to be damaging to certain types of arguments and findings HBDers often make. There can be bad arguments for correct conclusions, and people who make those arguments should stop making them because it introduces noise and makes it harder to find the good arguments.

Again I posted various examples of people not properly applying the phenotypic null hypothesis. Let's zoom into one of them to understand the problem:

https://twitter.com/tailcalled/status/1475441032292667394

If one doesn't understand the phenotypic null hypothesis, then this is an exciting study. Researchers have shown jealousy and restricted sociosexuality to be genetic! And to be biologically linked to each other! In the past I would have been really interested in these sorts of results, as tying into all sorts of evo psych theories.

However, when appreciating the phenotypic null hypothesis, it's a boring study. What are we even supposed to learn from it? Obviously these variables are going to be heritable and genetically correlated, but this doesn't really tell us much.

If you disagree with this notion, then feel encouraged to make your case for why this is such an important and meaningful finding.

Turkheimer, like Lewontin before him with his «apportionment» and, more to the point, with his corn plants (cited by Ned Block here), is making a mountain out of a common-sensical molehill, in his case to bury the monster of HBD implications that have sprang forth (as he asserts, unreasonably) from his own First Law that he now seeks to reframe (as much is stated in the paper).

I don't think Turkheimer is being an obscurantist here. He's a leading behavior geneticist and an editor for a behavior genetics journal; he has to deal with an endless stream of papers that proudly talk about how they've shown this and that to be genetic. He's got excellent reasons to try to make people accept the phenotypic null hypothesis, since it's a huge piece of missing knowledge.

Measurement error is a fundamental problem; though as better-informed people remind us, there are methods, e.g. common pathway models, which help against it,

But these models are rarely used. Even from the "better-informed people", I have had trouble getting it for e.g. testing the causal effect of g.

It is also possible to directly test for AE vs ACE

This is badly powered when C is smallish, e.g. try computing the power requirement for C^2=0.01.

(Besides, we have plenty of data such as admixture and, as of late, GWAS confirming simple additive model for the group difference – qualitatively similar to the case of height, not similar to gay-bullying and personality research)

I don't see how GWAS additivity defends personality research, can you expand? In particular I don't see how phenotypic null hypothesis predicts nonadditivity.