The prevalence of drawing causal conclusions from observational data
June 2021 – This summer, we’re spending time in office hours with some of Drew University’s amazing faculty to learn about what interests and inspires them and their research.
Today, we’re talking with Sarah Abramowitz, John H. Evans Professor of Mathematics and Computer Science, and Chair of Mathematics and Computer Science.
What about your field interests you most?
Why does it fascinate you?
Summarize the topic for us.
As an example, right after my daughter was born, there was a widely publicized study about nightlights. The CNN headline was, “Young children who sleep with a light on may have a substantially higher risk of developing nearsightedness as they get older, says a new study in the journal Nature.” A year later, the study conclusions were retracted after the discovery that parents who were nearsighted tended to use nightlights and also to have children who were nearsighted. In other words, even though the use of nightlights was correlated with nearsightedness, the nightlights were not the cause of the nearsightedness. Other examples of these kinds of mistakes may be found here.
How do you bring the topic into the Drew classroom?
In our introductory statistics class, which enrolls about 300 undergraduates annually, we devote time to identify when studies are observational and when causal conclusions have been improperly made from observational data. We teach students to interpret our own correlational analyses using language that is not causal. We also teach students to look carefully at samples to make sure that they are representative.