Your above exercise is silly because it doesn’t control for genetics. How can you run a robust experiment without accounting for major confounds?

Even more we don’t even remotely understand how those confounds work - much less be able to control for them. This is why most psych / sociological experiments are weak.