Yes, I think that sentence is talking about how you might get a higher score for answering one difficult question correctly but getting two easy questions wrong than somebody who answered the easy questions correctly and failed the hard one.
The IRT stuff looks really clever -- it simultaneously figures out how difficult each question was and also how good each candidate was from the raw input data, so you don't have to manually set question weightings in advance or anything. The only downside seems to be that the algorithm is a total black box that you can't manually score yourself against.
The IRT stuff looks really clever -- it simultaneously figures out how difficult each question was and also how good each candidate was from the raw input data, so you don't have to manually set question weightings in advance or anything. The only downside seems to be that the algorithm is a total black box that you can't manually score yourself against.