I hate to think I have discovered a bug because it usually is something
I have done wrong, but not this time, I don't think. After many hours
with this command it struck me that the reported standard error is much
too small; for one, the difference between four times the standard error
and the width of the 95% confidence intervals is too big, even with big
sample sizes.
Has anyone experienced this? My suspicion is that there is an error in
the formula on page 106 of the [ST] manual. The formula includes the
square-root of a g-hat, identified as the standard error from Major
Greenwood's formula. Either that should be the variance from Major
Greenwood, or the square-root is out of place. I suspect that one too
many square-roots exist.
Anyone had any experience with this? My phone call to StataCorp was not
useful.
