Thank you!

That's very helpful. Alant mentioned the logit of probability equation to me in my BC thread about this, but at the time I didn't see how it related to Rind's "right-hand side of the equation." Your detailed explanation is at least putting me on the right track, I think. I don't recall encountering natural logarithms in high school math. On a college prep track I apparently should have. I do remember I only got to trigonometry in a junior college class I took years after college. I can get a logit of 0.24 (and thus Rind's 56 percent) by setting the coefficient for the survey year in Rind's Sandfort example to 0.01181 instead of .01, which tends to confirm my speculation that rounding is at least one of the issues here. But when I then add the 0.89 for participant initiated, the logit rises to 1.13, yielding a probability of 76 percent, not the 81 percent Rind gives. (The whole idea of the value for survey year actually being the survey year does seem to stand out as a possible source of difficulty, though as you noted Rind is quite specific about this. Other characteristics for which there are three possible values, like the age ranges, are handled by treating one as a reference and the other two as either present or absent.) The other thing that keeps grabbing my attention is the negative coefficient (-0.72) for "not coerced." The presence of characteristics with negative coefficients will make the outcome less likely, since they make the logit smaller, right? But we don't expect lack of coercion to make a positive reaction less likely. And elsewhere Rind says that for both boys and girls lack of coercion was associated with positive reactions. (In the logit equation for girls, the coefficienct for "not coerced" is 1.5.) And of course getting a logit of 0.24 still requires treating the minus sign with which the model for boys begins as a typo -- a pretty bold assumption, but one I am not seeing much way of avoiding. hugzu ;-p