3 ms·
By necessity, any test can realistically test only a subset of a student's knowledge. If the patterns that the tested subset follows are easily predictable or h
by rbehrends 12y ago
By necessity, any test can realistically test only a subset of a student's knowledge. If the patterns that the tested subset follows are easily predictable or have a simpler structure than the actual knowledge, you can get better performance by teaching those patterns than by teaching the actual skills.
Multiple choice tests are particularly egregious examples, as for them you only have to be able to verify an answer instead of having to derive the answer.
Standardized tests are rife with multiple choice tests and questions that follow predictable patterns.
- danielweber 12y agoGood test design is hard, especially for an amateur. But it seems the remedy ought to be "have better tests" instead of "don't test." There are good tests out there; the AP tests have a significant multiple-choice component that resists simple trickery.
- rbehrends 12y ago> But it seems the remedy ought to be "have better tests" instead of "don't test." Nobody is saying "don't test". The only place that I know of that doesn't test in the traditional sense are Waldorf schools, and even they still evaluate their students. Concerns are being raised about standardized testing in a NCLB-influenced environment, both in how that limits test design, tests for the wrong things, and how this kind of testing creates the wrong incentives. > the AP tests have a significant multiple-choice component that resists simple trickery. It isn't just about trickery. I remember encountering a pumping lemma question during the GRE that was simply easier to answer in multiple choice form because I didn't have to prove the correctness of the right answer, but only the incorrectness of the wrong ones (and even there I didn't have to work out the fiddly details, I just needed a modest amount of confidence).