Austria     Belgium     Brazil     Canada     Denmark     Finland     France     Germany     Hungary     Iceland     Ireland     Italy     Luxembourg     The Netherlands     Norway     Poland     Spain     Sweden     Switzerland     UK     USA     

Minding your p-values

Statistics help us understand data, make better decisions, and make scientific discoveries. But Brexit, followed by Trump’s election, shook a lot of the public’s faith in statistics. In both situations, the polls were wrong. But for most people, polls are the most relevant and publicised statistics they will interact with in their daily lives. Leading up to voting day, a vast majority of polls showed that Remain would win the day. Regarding the 2016 United States Presidential Election, the final analysis of America’s most famous statistics website, fivethirtyeight.com, showed that Hilary Clinton had a 71.4% chance to win the election. Just a few weeks earlier, Clinton’s chances of victory were closer to 90%. So, the world was shocked when Britain voted Leave, and Trump won by a significant margin. But one unexpected result was the significant backlash against statistics in general.

An Imperfect Puzzle

Statistics is an inherently intimidating science. Some people do not like numbers. Others who might like numbers are still intimidated because it is such a niche area. The truth of the matter is that unless you have taken a graduate-level statistics course, you might not speak the statistics language fluently enough to be able to fully understand even public-facing polls. And many people are forced to wield the weapon of statistics, whether or not that is proper. As aforementioned, statistics help us make scientific discoveries, but there are arguments within the scientific community that it overly or often improperly relies on aspects of statistics.

For example, in a piece entitled “Retire Statistical Significance”, the authors make a compelling argument that there are certain statistical conventions that need to be reconsidered. There is an arbitrary line surrounding what is known as the P-Value. To oversimplify it, P-values give you an idea of how likely an experiment’s effect is real. Somewhere along the way, a P-value of .05 became the barrier of significance. That is, if a study’s P-value is .049, then the study is statistically significant; if that P-value is .051, then that study is insignificant. This matter of significance can make or break whether a prestigious journal decides to publish the study. For scientists, the amount of publications they have directly affects their funding and ability to obtain tenure. This is not a trivial matter.

Practically speaking, many scientific sub-communities are filled with people who have no interest in statistics. Do many people study human psychology or anthropology because of their love for numbers? But still, statistics is such an important piece of the academic puzzle that such graduate programmes emphasise stats heavily. Thus, there is an inherent and unavoidable conflict: people are forced into wielding a weapon that they might not fully understand. And somewhere during their education, they learn these arbitrary cut-offs and limits (like P must be less than .05) that they then adhere to for the rest of their careers.

We should embrace statistics despite its flaws. It is a profession that constantly seeks to improve itself, even in the face of consistent public backlash. And with improved statistics, we can make better sense of science, business, and our world in general.