You've probably seen it before: a headline that goes something like "Study: caffeine gives you brain cancer." And you think, What, really? Such results are often followed by a similar headline a few months later with the exact opposite result. How do we sort through this endless stream of science?
Aaron Carroll's most recent video suggests some useful tools for this purpose. He tells us about meta-analyses, which are studies built on previously done studies. By combining data, one can increase statistical power and perhaps uncover new results. Many of the most counterintuitive studies — and therefore most widely reported — are meta-analyses. Here are some tips to keep in mind:
* Conducting a meta-analysis requires a judgment call. One has to pick which studies go in and which do not. This can and should be done on rigorous grounds, but it means that meta-analyses will always be a bit more vulnerable to cherry-picking than a randomized controlled trial.
* Meta-analyses are for weak relationships. If an effect is strong, then it will be easy to find. Smoking's relationship with lung cancer, for example, practically explodes out of the data. Meta-analyses, by contrast, are mostly done on questions where the effect is not nearly so obvious, like whether organically-grown food is more healthy.
Therefore, one should generally approach a new meta-analysis with a wary eye. They're an important tool, but not infallible. Watch the full video explanation below. --Ryan Cooper