If you go the interview approach you'll run into many potential confounds. If the average age of first exposure is 13, and someone says they waited until 17, it

If you go the interview approach you'll run into many potential confounds. If the average age of first exposure is 13, and someone says they waited until 17, it's possible that a third variable delayed them from seeking porn and also will have an impact on other life incomes. This is why random assignment is the gold standard. With a large enough sample size random assignment will account for all confounds. Here, we can only measure likely confounds and attempt to 'account for' them with statistics. All of these introduces more noise and generally reduces the validity of the study. In general, I dismiss all scientific studies with these looking back questionnaires as little more than well documented hunches.