search

ijyliu / ECMA-31330-Project

Econometrics and Machine Learning Group Project
2 stars 1 forks source link

Good parameter values to test #19

Closed ijyliu closed 3 years ago

ijyliu commented 3 years ago

We need to pick interesting parameter values for:

N (sample size) rho (correlation between true covariates x) p (number of covariates) kappa (coefficient linking x_1 to the instrument value) beta (true beta for the covariates x) x_measurement_errors (variance for the measurement error for each covariate in x, we are assuming mean zero noise)

https://github.com/ijyliu/ECMA-31330-Project/issues/12#issuecomment-835498637

This will help inform #16

ijyliu commented 3 years ago

@nicomarto I ran some (panic-inducing) sims on this and the preliminary evidence is that you need pretty high ME for x_1 and basically none for the other covariates. it also seems like beta has to be kind of large for the other covariates.

it would be nice to have theoretical results so we can motivate the choices of parameter values to try.

ijyliu commented 3 years ago

@marionoro has already picked some pretty OK parameter values, though we should bump up the p probably. discussion continues in #44