I wonder how acceptably accurate vocabulary totals can be determined. If it could be done at a rate of one word per second for ten hours, 36,000 words could be administered. But nobody could give that many responses at such a rate for so long. But if you gave the subject two seconds, it would take two days.
So, obviously, some type of sampling has to be used, and extrapolated to give total.
But, how in hell can the dependability of the sampling method the ascertained?

There is a hell of a discrepancy between the magazine's 40,000 . and the URL's 20,000.