Now which PhD candidate will bite :)
Not to mention that unless the adoption of kids are done blind by parents, there has to be some biases there. As an example out of my ass, kids with behavioral problems might have way less adoption rate, and potentially those are the one with below median outcome in any case.
That gets you two categories of people do definitely want kids, and only one is subject to the classes, interviews, and checkups.
Really the ideal test would be, in order of preference:
1. randomization (some parents get the classes, some do not, chosen randomly)
2. geographic (parents in state/area A get classes, rest do not)
3. temporal (parents after year X get classes)
4. twins (parents of twin A get classes, twin B do not)