Minimizing the role of statistics in research

I submitted a paper for publication today. Working my way through the 9503 word long Guide to Authors, I eventually got to this:

Statistics The authors are requested to pay particular attention to the use of appropriate statistics throughout the text. A detailed description of all statistical methods should be included in the Methods section. The Journal encourages authors to seek expert statistical advice prior to submitting any manuscript that has a statistical content.

50 words, or 0.5% of the instructions.

At the risk of overstating the importance of this one observation, to me this reflects a scientific culture that limits the role of statistics to some kind of specialist, technical input that comes at the very end of the research process. This of course varies widely across fields, but I have felt this way about epidemiology for a long time. The reality is that if you are involved in research that involves sampling, then any reasonable conclusions you draw from your research will rely on statistics.

If you are training for a career in scientific research, you have to make choices about what you learn. Nobody can be an expert on everything, even in very niche fields. With this limitation in mind, I implore you, don’t short change yourself on statistical methodology. While statistics can’t correct for poor theory or data collection (well, maybe a tiny bit), you can ruin an entire research project with poor statistical analysis. This is why the idea of doing all the work leading up to an analysis (finding an important, testable question; reading existing research on the subject; collecting the necessary data) and then just turning it all over to a statistician (or witch) always made me uncomfortable. If you can make the time, statistical training is widely available for people at all career stages. It might be a painful process, especially if you don’t have  strong background in math, but for me it’s really been worth it.