I vaguely remember there being some stats trick where like if you know distributions random variables X, Y, and X*Y or something you can find stuff about the relationship between X and Y
If so interested in this because let X = total EA spending, Y = EA meta spending, then we could elicit the effectiveness of EA meta spending (+$1 to meta = +$X in overall funding) from prediction markets or superforecasters
idk if I’m just making this up tho
@TetraspaceGrouping I don't know how this would work, but if you find out, please post, bc it sounds interesting
@niplav The NLRG answered on twitter: https://twitter.com/NLRG_/status/1440994547316793345
E[XY] - E[X]E[Y] is the covariance (https://en.wikipedia.org/wiki/Covariance), and the covariance divided by the variance of X squared is the gradient of the linear regression (https://en.wikipedia.org/wiki/Simple_linear_regression#Fitting_the_regression_line)
