Robin Gong
The journey of a drug from research lab to patient takes an average of 12 years. New drugs go through years of development, animal testing, three rounds of clinical trials, and numerous safety committees before it is deemed safe to be placed on the market. Robin Gong, a PhD candidate in statistics, argues that statisticians need to think a little more like scientists before unleashing a statistical method onto the world.
“We have to be aware that whatever statistical method we advocate for will have consequences,” she cautions. “We’re not simply devising a mathematical toy that behaves well in the clean room. The method will go out and interact with messy human data, will be interpreted by humans, and will generate conclusions that will be taken seriously by actual human beings.”
Statistical models were designed to take in precise knowledge and produce precise results, but in the real world the input is not always precise. We as a society, however, expect unambiguous results, so there is a lot of pressure for statisticians to deliver them. “To make up for this lack of precise knowledge, statisticians make assumptions about how the data was collected and choices about how to predict the outcome,” Gong explains. “But, the more assumptions we make, the higher chance that we can end up with results that are so precisely, confidently wrong.”
During the Super Bowl, we can use statistical models to predict whether the New England Patriots will win. For most people, a yes/no answer is sufficient. But if you are a Patriots aficionado with a lot at stake in the game, you are better off analyzing their chance of winning using Gong’s higher-level statistical models since they will provide you with an idea of how uncertain the statistical results are. “These models can operate on set-valued (instead of point-valued) inputs, thus no longer require them to be precise inputs. As a consequence, they allow us to say yes/no/not sure for the result.” For many people this uncertain result seems less useful, but if you really care about your answer and don’t want to get it wrong, than you probably want to stick with an outcome that is more conservative,” explains Gong.
Gong describes the job of a statistician as doing the best we can to use what we know to make a guess about something we don’t know. Gong argues that we should start practicing what she calls judicious statistics. “It means making sensible judgments based on statistical outcomes, but keeping in mind that we’re already making judgments in the way we collect and analyze data and that influences the outcome.”