When listening to todays (day 1) lectures I was wondering about the relationship between missing heritability and measurement error. Hight for instance, does not have much noise in the measurement. But most complex traits, questionnaire items, administrative data etc. all have their own source of measurement error, some sometimes accounting for up to 50% of variance. Twin models seem to be rather unaffected even though I dont really know why (I was told one could just add another level in the structural equation model: one could just use 3 items to inform an laten factor/trait error free). In GWAS often nobody cares about measurement error, so I wonder how that effects the heritability we miss.
Thanks, Henrik, for another great question. The missing heritability problem is mostly about the gap between GWAS and twin-based studies. Measurement error would in principle affect both but you’re right to underline that the impact would be trait-specific. In our recent paper (https://www.nature.com/articles/s41586-025-09720-6) we focused on comparing SNP-based heritability (generalising GWAS-based estmator of heritability in unrelated people) and family-based heritability (generalising twin-based although different assumptions might hold) in the same cohort. This is the best way to make sure you don’t have differential measurement error. I think this paper should help answering other questions you might have.
Thanks. I will give it a read. I realized that the question on twins relates to the common pathway / psychometric model.