I think the test is based off of a statistical model that looks at the probabilities on some form of tree of knowing related derived words from certain combinations of roots.
That said, such a thing will be, by nature, imprecise for an individual but quite precise for a large population of test takers.
Which is why I am not to worried about the apparent ridiculous ranges of word knowledge offered -- the test can only examine a small subset of words considered, so it is probable that some of us have gotten a batch of words we'd never know despite our knowledge of other batches, while others will get nearly all of them from a favorable batch, possibly through a set of initial conditions or chance.
I am more interested in the implementation of this test, if the model is at or greater complexity than what I've speculated. Would be cool.
That said, such a thing will be, by nature, imprecise for an individual but quite precise for a large population of test takers.
Which is why I am not to worried about the apparent ridiculous ranges of word knowledge offered -- the test can only examine a small subset of words considered, so it is probable that some of us have gotten a batch of words we'd never know despite our knowledge of other batches, while others will get nearly all of them from a favorable batch, possibly through a set of initial conditions or chance.
I am more interested in the implementation of this test, if the model is at or greater complexity than what I've speculated. Would be cool.
