Last week we introduced Simpson’s Paradox, a phenomenon where the observed relationship between two groups/variables reverses when the data are split into subgroups by a third variable. We saw that the paradox generally occurs due to an imbalance in the third variable across the two groups we originally sought to compare.
In our example (using data from Zeisel 1981), we saw that White offenders received the death penalty slightly more often in aggregate (23% vs. 21%), but Black offenders received the death penalty much more often in cases involving a White victim, and in cases involving a Black victim. The reason was that 96% of cases with White offenders involved a White victim, while the majority of cases with a Black offender involved a Black victim, and juries were much more likely to return the death penalty for cases involving a White victim.
Today, you will be asked to research a resolution to Simpson’s Paradox.
\(~\)
Your group will be assigned to research strategies to prevent or ameliorate Simpson’s Paradox in one of two contexts:
For #1, you should focus on things that a research could do when designing an experiment in order to collect data that won’t be impacted by Simpson’s Paradox. For #2, you should assume that the data have already been collected, you’ve identified one or more variables that could cause Simpson’s Paradox, and you need a way reliable way to analyze the data.
You should try to rely upon reputable scientific sources, including peer-reviewed journal articles, published textbooks, and the writings of well-known professional statisticians.
\(~\)
During this endeavor your group will produce two things:
For #1 you should include:
For #2:
\(~\)
\(~\)