UCB Admissions

In 1973, data was collected from UC Berkley’s admissions that showed a gender bias in its six largest graduate departments. It seemed that males were more likely to be admitted into these graduate departments than females were. Looking at the graph below that was created using the data collected, it appears that males saw more admittances than females saw.

However, when taking another look at the data the bias shown in the first graph isn’t as blatant as it may have seemed. Taking a look at the admittance and rejection proportions shows that the bias originally found is not true. Across the departments the proportion of males that were admitted was similar to the proportion of females admitted. Some departments, like Department A, there was even a higher proportion of females admitted than the proportion of males admitted.

Simpson’s Paradox occurs when data shows a certain trend or result that isn’t necessarily true. This occurs when a lurking variable is present. A lurking variable is a hidden variable that splits the data a certain way. In this scenario with UC Berkley Admissions, the way the data is separated in the first graph, it appears that there is a greater proportion of men admitted than females. However, when the data gets separated by departments in the second graph we see that this isn’t necessarily the case. The lurking variable would be the departments that separates the data.

Bibliography

Grigg, Tom. “Simpson’s Paradox and Interpreting Data.” Towards Data Science, 9 Dec. 2018, towardsdatascience.com/simpsons-paradox-and-interpreting-data-6a0443516765. Accessed 3 Sep. 2024.