It looks like you need to model sex and age predicting donut liking using a multi-level logistic regression where you have additional random effects and intercept nested within household. e.g.

    lmer(donutLiking ~ sex * age + (sex * age | household), data = A)


From that you can get parameters that allow you to make probabilistic statements about the donut liking in data set B both taking into account cross household variability and random and fixed effects of sex and age