A useful plot for examining multiple measures over a range of a continuous predictor is a shaded area plot. These are especially valuable for compositional data (data on multiple categories that always add up to 100 percent), and are often used with time-series data, to visualize how the relative frequencies in multiple categories change over time. But they can also be useful for seeing how those relative frequencies change over values of any more-or-less continuous variable.
Returning to the ABA’s employment data for 2014, we can use a shaded area plot to see how the various employment outcomes vary as we move through the U.S. News rankings. Each shaded area represents the proportion of a school’s graduates who achieved a particular type of employment outcome, with the schools ordered by their 2014 U.S. News ranking. A few details:
In the 150-200 range, we substitute a ranking based on U.S. News’ “Academic Peer Score” for the overall ranking.
We combine three categories (“Unemployed – Deferred,” “Unemployed – Not Looking for Work,” and “Unemployed – Looking for Work”) to create the “Unemployed” category. The numbers in the first two categories are generally quite small; for those interested in the plot with the unemployment numbers disaggregated, it is here.
We include dashed vertical lines at rankings of 10, 50, 100, and 150.
Finally, because the data are noisy (more on that in a future post), we smooth the data across the range of rankings. This masks some of the school-to-school variability, but also allows us to see the general patterns better.
I suspect everyone will find their own interesting things in this plot, but to me the most striking characteristic is the steep drop in Bar Passage Required employment between the teens and the 50-60 range, together with the much less consistent (and less radical) decline from there to around 150. Of course, not all Bar Passage Required jobs are created equal, something we’ll investigate further in a future post.