Hi,
when calculating the Cox model for a factor with n levels
(using treatment contrasts), i noticed that the off diagonal elements
of the estimated covariance matrix are always nearly (but not exactly)
equal.
On the one hand, this is plausible to me: If i could obtain estimates
x_i, 1<=i<=n, for the log hazard ratios relative to the baseline hazard,
then of course the off diagonal elements of the covariance matrix for
the estimated contrasts y_i = x_i - x_1 would be exactly equal.
On the other hand, the coxph model estimates the contrasts directly,
and numerically the off diagonal elements are just very close, but not
equal.
Any explanation, hint or reference would be highly appreciated.
Christian
