I think the answer to this is roughly ... the content of an applied statistics degree, or a research methodology practicum. There are so, so many ways that quantitative measurement of human phenomena can mislead, from measurement error to conflation between trends/sample sizes/effect sizes, to p-hacking, to good old bias (in interpretation or in deciding what tools to use to smooth or normalize data), and thousands more.