Gee Arr Eee

I took the GREs yesterday, a very weird test. It's like the SATs (math and verbal sections worth 800 points each, plus writing) but computer-based, and uses how well you answer the first questions to determine how difficult the later questions will be. People generally do better on the math section than the verbal section -- a perfect math score is only 92nd percentile, but you only need about a 720 to achieve that on the verbal.

I think this may be because the math is more precise. Broadly speaking, if you understand the concept and know how to reduce a hard problem to smaller solvable steps, you can answer any problem in that format -- it doesn't ask you to know Euler's constant off the top of the head, for example. The verbal, on the other hand, depends heavily on your vocabulary: by my estimate (with a healthy margin of error), less than a third of the questions I recieved in the practice and real tests were reading comprehension. The rest were antonyms, word choice, relationships, and other vocab-dependent questions.

My problem with this is that the English language is large enough that two random people may not have much overlap in the list of "difficult" words they know, even if they are both have extensive vocabularies; as a rough guess, the test might require you to know about 50 "difficult" words. I found one webpage that lists over a thousand words that the GRE uses -- I'm not convinced that the adaptive scoring's statistical methods are really overcoming the role of chance. I took two practice tests two or three times each, and found my verbal score had a range of about 80 points (out of 800), with no clear pattern. I had a similar range on the math section, but I was actively reviewing concepts I wasn't clear on -- and that score showed a general upward trend.

Incidentally, on the real test, my verbal score was only 10 points higher than my lowest practice score, and my math was 150 points higher than my highest math score. For whatever that's worth.