SE for marginal mean

SE for marginal mean should always be calculated by cvovG, because vcovHC won't work (vcovHC only works for "difference" contrast and linear model without treatment by covariate interactions). This needs to be fixed as discussed in our meeting today.

Thanks!
Ting