You can't use the summary of the data to get your confidence intervals. You
have to use all the simulations of \hat{\beta} and \hat{\sigma^2} and put
them in order and take the appropriate percentiles of the simulations.
----- Original Message -----
From: "Alex Liebman" <liebman(a)fas.harvard.edu>
To: <olau(a)fas.harvard.edu>
Sent: Wednesday, September 29, 2004 4:34 PM
Subject: problem #c
hey olivia,
on problem c, are we meant to calculate the confidence intervals for the
aggregate (ie, the means) of all the procedures we ran?
in other words, to test the biasedness in part b, i took the mean of the
constants, the mean of all the slopes, and the mean of all the variances.
can we use this aggegate data to check confidence intervals? that just
seems strange to me, because of course in the real world we'd only "run"
the data once. i know this is a simulation problem, but there still seems
to be something strange about this, even granted that we're trying to
mimic the DGP. do you see what i mean?
alex