Hi Huiying,
Question 1: MO wasn't really designed for binary variables with measurement error and so there isn't really a proper way to specify the prior mean/SD. If you have validation (gold-standard) data, you include both the validated and misclassified/mismeasured variable in the same amelia model and get imputations that way. Otherwise, you could try to force amelia to handle this situation by setting the prior mean to either 0 or 1 with some SD that reflects your uncertainty and then manually force the imputations to be either 0 or 1 after the imputations.
Question 2: moPrep is a convenience function that sets up the priors and overimp arguments in the right way. So you should get the same imputations either way (except for the randomness of the imputations themselves).
~~~~~~~~~~~
Matthew Blackwell
Assistant Professor of Government
Harvard University