We can see the Will Rogers phenomenon at work in this instance because the overall slope was -0.015 standard deviations per year. For those with less than twelve years of education the slope was -0.039, whereas for high school graduates, the slope was -0.041. For those attending college the slope was -0.034 versus -0.027 for people who had some college but didn’t graduate; the slope was -0.019 for college graduates, -0.018 for those currently in graduate or professional school, and -0.024 for those with graduate or professional degrees.
The overall slope is less than the slope for any individual level of education because people are sorting across the levels in exactly the way I suggested.
So, are we getting dumber? Unfortunately, this wouldn’t show it. If the authors ever release or return to the data to analyze whether the changes are measurement invariant and how they appear after post-stratification, then we can know. Until then, we’ll be left guessing.