The mean of a binary variable is totally interpretable;
in this case the interpretation is not that an individual
has half a degree, but that on average half the individuals
have degrees.
Logit, probit and much else would not make sense otherwise.
Nick
[email protected]
Exactly. Anova is technically incorrect here, but in the few large sample
cases where I've tried it both the "right" and the "wrong" way, Anova got
the significance tests almost exactly right. Of course, that may not
always be true, and there is no reason not to do it the "right" way, other
than maybe convenience. Especially back when computing power was more of
an issue, doing things a quick but technically wrong way (e.g. run an OLS
regression with a binary dependent variable) was sometimes a way of
narrowing down what you would estimate with a more time-consuming correct
procedure.
