6 Comments

Maybe a look at measurement invariance for other SIRE group could be a future blog post/paper, given that there's beenn some controveresy to the extent of the asian-white difference in early samples(James flynn found that in the mid 20th century east-asian immigrants performed equivalent to white americans in the US, but academically and socioeconomically performed significantly above, but cremieux thinks the difference is mostly due to language/test bias(believable,and with some evidence in Jensen 1973,https://twitter.com/cremieuxrecueil/status/1661449979116806148) in samples like Project Talent(and the unbiased gap is the same as modern samples), as well as geographical factors impacting their success.

Expand full comment
author
Sep 12, 2023·edited Sep 12, 2023Author

I'll see if I can unhearth more good data. The PT has lots of students but the great majority of them belong to the catagory "unknown". As for asians, the category "Oriental" has 999 students. I'll check if other old data sets have enough asian people. And compare with more recent data with asians. If the argument is tenable, then I believe we should expect invariance to hold for recent data but not old data.

Expand full comment

https://rpubs.com/JLLJ/AIQ and https://www.researchgate.net/publication/335239154_A_Surfeit_of_Evidence_on_Spearman's_Hypothesis have some analysis indicating that invariance is tenable for modern east asian-white comparisons in the US and internationally(at last for the Chinese, Japanese and Korean WAIS/WISC standardisations).

Expand full comment

If there's subtest data and a large sibling/twin sample, could you have tested between-group heritability directly(using a common pathway model, and if it fits including group differences)?

Expand full comment
author
Sep 3, 2023·edited Sep 3, 2023Author

I would have, of course. Given the description of the variables in the documentation, I could only find questionnaires like how many siblings dropped out, or if you have a twin at school etc. But nothing like sibling/twin IDs which are necessary for this kind of analysis. It's likely another data, the official number is 400,000+ people and my sample is 377,000 because it doesn't include those additional familial samples. I just wasn't able to get it, perhaps it's not publicly available. But I know someone is working on this twin/sibling data right now. I'll eventually ask more about it soon once I'm done with my current projects.

Expand full comment

I was interested because cremieux apparently did this analysis a while ago in the NLSY 79/97 and found a 77-84% BGH(90-92% WGH), but I don't think it was posted anywhere. Jordan lasker did a analysis a few years back and found mixed support for a common pathway model(becausse of small sample sizes and not the best cognitive tests), so not enough to try a model with multiple groups. The TEDS sample(early waves are public with no SIRE data), has a very large sample size for testing common/independent models, and it's also longitudinal(i've started analysing this but there's some loss with followup waves).

Expand full comment