Could anybody direct me to a proof of the approximation of the difference of the proportion estimates of the two samples? (p. 24, chapter 9, first expression)
You're using a normal approximation to the binomial and then doing the subtraction of two normal distributions (see p41 of chapter 4).
Makes sense. Really useful. Thank you! Is the standard MLE theta_1,2 ~ X_1,2/n_1,2 in the denominator always reliable (i just noticed that this is what really bothers me)? I mean the simplification of the nonlinear equations here must be harder to justify than in the one-sample case (as we would need to back-solve for the expression (theta_1 - theta_2)).