The Classical Twin Design (CTD) has always been criticized for being oversimplistic, and consistently overestimating heritability estimates due to not accounting for GxE, GxG, rGE and equal environmental effects.
This is why I never understood why all these researchers who used quasi-experimental studies to assess the "causal" impact of scholastic intervention on IQ often concluded it had an impact while the intervention programs always failed but they always ignore these past research.
Similarly, I think most researchers who study twin data seem to be overly positive, sometimes unfairly so. I think the bias is more serious some believed, but likely much less serious than the harsh critics would have it. More importantly, there is no conclusive evidence of a systematic bias that would inflate heritability. Only indirect "evidence" and "suggestion" that heritability might be upwardly biased.
I can't answer much about the missing heritability. If my memory serves me well, during the 2012-2014, I could recall that people were excited about this new GWAS design. So I thought maybe 10 years later, we would close the gap in the missing heritability debate. Now in 2024, it doesn't appear so.
Their sample sizes are still terrible. To rule out rare variants, they need to 10x height sample sizes and more for everything else. The fact that so much is missing for height is also going ignored -- IBD regression results are in line with twin estimates and downwardly biasing confounders for height in twin estimates are not plausible. Yet they're missing half the heritability of height. Maybe it's because they sample <1% of the genome?
Yes, there are new methods like GREML-IBD. And I agree with you here. My point was more general though. In the past, researchers were also arguing that this missing heritability may be due to GxG and GxE interactions etc but that it is extremely hard to detect, and looking at the debate today it feels like some crucial aspects of the debate hasn't changed much compared to 10 years ago. Hence my disappointment.
To my knowledge height is at ~70% additive heritability when RDR/sib-regression are adjusted for assortative mating, we also get roughly the same with machine learning at 70%(https://journals.plos.org/plosgenetics/article?id=10.1371/journal.pgen.1009141) in the UK Biobank. Admittedly the results are still below twin studies showing 80% or more common 90% adjusting for measurement error in adulthood(polderman 2015, you can see the results on MaTCH by age group). And in the case of young's paper differ from the results are incongruent with an equilibrium heritabilty of almost 1 and a negatively correlated indirect and direct genetic effect. And there's some rare variants missing from any machine learning approaches published thus far.
Totally agree.
This is why I never understood why all these researchers who used quasi-experimental studies to assess the "causal" impact of scholastic intervention on IQ often concluded it had an impact while the intervention programs always failed but they always ignore these past research.
Similarly, I think most researchers who study twin data seem to be overly positive, sometimes unfairly so. I think the bias is more serious some believed, but likely much less serious than the harsh critics would have it. More importantly, there is no conclusive evidence of a systematic bias that would inflate heritability. Only indirect "evidence" and "suggestion" that heritability might be upwardly biased.
I can't answer much about the missing heritability. If my memory serves me well, during the 2012-2014, I could recall that people were excited about this new GWAS design. So I thought maybe 10 years later, we would close the gap in the missing heritability debate. Now in 2024, it doesn't appear so.
Much like the twin debate by the way...
Their sample sizes are still terrible. To rule out rare variants, they need to 10x height sample sizes and more for everything else. The fact that so much is missing for height is also going ignored -- IBD regression results are in line with twin estimates and downwardly biasing confounders for height in twin estimates are not plausible. Yet they're missing half the heritability of height. Maybe it's because they sample <1% of the genome?
Yes, there are new methods like GREML-IBD. And I agree with you here. My point was more general though. In the past, researchers were also arguing that this missing heritability may be due to GxG and GxE interactions etc but that it is extremely hard to detect, and looking at the debate today it feels like some crucial aspects of the debate hasn't changed much compared to 10 years ago. Hence my disappointment.
To my knowledge height is at ~70% additive heritability when RDR/sib-regression are adjusted for assortative mating, we also get roughly the same with machine learning at 70%(https://journals.plos.org/plosgenetics/article?id=10.1371/journal.pgen.1009141) in the UK Biobank. Admittedly the results are still below twin studies showing 80% or more common 90% adjusting for measurement error in adulthood(polderman 2015, you can see the results on MaTCH by age group). And in the case of young's paper differ from the results are incongruent with an equilibrium heritabilty of almost 1 and a negatively correlated indirect and direct genetic effect. And there's some rare variants missing from any machine learning approaches published thus far.