I'm going to focus on my senior thesis, work out, and binge a bunch of TV shows while waiting lol
The claims on page 1 that GRE scores do not predict performance is at odds with the empirical evidence. Athey et al. find that GRE scores strongly predict first year grades, and that first year grades predict job placements. (It is not surprising that controlling for the intermediate outcome, the correlation between GRE scores and predicted grades is small.) NB: economics departments care about first year grades because passing comps is a necessary condition for receiving a PhD, so even if GRE scores only predicted first year grades, it would be rational for admissions committees to consider them.
Could be, but significant coefficients and "strongly predict" aren't at all the same. I bring this up because (1) I'm always interested in what Prof has to say and (2) learning the difference between the two concepts is critically important for folks heading off to graduate school.
In this case, the magnitude of the effects is important. I don't have GRE percentiles for the time frame of the data in the paper, but averaged over 2014-2017, a 160 on the quantitative section is the 74th percentile, and a 170 is the 96th percentile (https://www.ets.org/s/gre/pdf/gre_guide_table1a.pdf). Therefore, moving from a 160 to a 170 is associated with a 9.5 percentile (or 0.34 SD) increase in the micro grade (using estimates from Table 2, column 1). Note that I share the authors' interpretation that specifications controlling for admissions rank are less informative for this exercise because it is likely a function of GRE scores. This is as large as the premium associated with attending a top-15 university, for example.
To the person who down-voted the previous post, why? What about the post was offensive or inaccurate?
These things change over time. I was able to find an old ETS guide that provides percentiles for 2011-2014: http://dbbs.wustl.edu/PortalDocs/GRE...ion%202014.pdf. In that time frame, a 160 was the 78th percentile and a 170 was the 98th percentile, which doesn't change the way I interpret the results.