Which is why you need high-quality and large-scale data on the relationship between income and internet usage and the relationship between income and happiness so you can chain those together and subtract the whole thing out. Yes, error bars stack up rapidly. That's why you need 2 million data points to get a reasonable conclusion, when you only need ~1000 to get high-quality information on things like consumer preferences or political affiliations. Good thing that we have plenty of people that are very interested in things like "how much happier does money make you" and full-scale datasets like that are produced on a regular basis!
Really, the math isn't any harder than what's necessary to land a rocket screaming along at mach 8 or reverse a couple terabytes of voltage readings into a nice clean volumetric medical image.