The same way you control for variables in any research.

You have a significant sample size.
You do profiles, you evaluate the subjects, you control conditions, etc. etc.

There are a host of ways to do it and different studiess to evaluate different factors which then go into metaanalysis to further filter them into quality data.

That doesn't mean I liked it, but we have to remember: infants and children have temperaments too- they are not 100% formed, realized in response to their parents. Humans have innate qualities from birth. It's not all nurture.