@aazuur.bsky.social
Predicting bean consumption with knee X-rays is more under-analyzing data, if anything. And, if the association holds in a large sample, why could we say it's fake?
The issue isn't whether "shortcutted" associations are real or not (they are), but rather if we can use them for decision making.
The issue isn't whether "shortcutted" associations are real or not (they are), but rather if we can use them for decision making.
December 17, 2024 at 9:36 PM
Predicting bean consumption with knee X-rays is more under-analyzing data, if anything. And, if the association holds in a large sample, why could we say it's fake?
The issue isn't whether "shortcutted" associations are real or not (they are), but rather if we can use them for decision making.
The issue isn't whether "shortcutted" associations are real or not (they are), but rather if we can use them for decision making.
But, isn't a point of the paper that the association between knee X-rays and bean consumption is actually, somehow, through who-knows-what backdoor paths by confounders, true, and not just an artifact of overfitting? The authors point out that the sample is "large" (25k).
December 17, 2024 at 3:10 PM
But, isn't a point of the paper that the association between knee X-rays and bean consumption is actually, somehow, through who-knows-what backdoor paths by confounders, true, and not just an artifact of overfitting? The authors point out that the sample is "large" (25k).