N1Grammar- studying differences between similar expressions

(2016-03-24, 5:52 pm)pm215 Wrote:

(2016-03-24, 2:24 pm)zx573 Wrote: You can lose points for correct answers still as far as I understand (and get points for incorrect answers).

Why do you think that? The Item Response Theory stuff that the JLPT info about scaling refers you to is a pile of maths that I don't really understand, but I wouldn't expect it to result in a correct answer increasing your score or an incorrect answer decreasing it...

After rereading the IRT explanation again, I think I might have assumed the "possible to miss points on correct questions" bit, but this is what I was thinking of:

Quote:Scaled  scores  are  determined  mathematically  based  on  “answering  patterns”  of  how  examinees  answer  particular  questions  (correctly  or  incorrectly).  For  example,  a  test  consisting  of  10  questions  (items)  has  a  maximum of 1024 answering patterns (210 patterns). For the scaled-score calculation process for the new test, based  on  these  answering  patterns  Japanese-language  proficiency  of  examinees  is  positioned  on  a  scale  between    0    and    60    points    for    one    scoring    section    (0-120    points    for    Language    Knowledge    [Vocabulary/Grammar]・Reading  for  N4  and  N5).  Because  the  maximum  1024  answering  patterns  of  a  test  consisting  of  10  questions  (actual  exams  have  more  questions)  are  categorized  into  61  groups,  scaled  scores  sometimes become identical for different answering patterns. Therefore, scaled scores can be identical for two examinees  even  when  the  number  of  questions  they  correctly  answer  or  their  answering  pattern  does  not  match.

If their scaled scores are identical but the number of questions they answer correctly differ, then that implies to me that they either weigh certain questions differently (probably the most likely case after rereading it), or you sometimes get assigned to a matrix where you might lose (or gain) some points somewhere.

Without knowing the exact details of how they calculate things, I think either of those are possible.

Edit: I read up some papers on IRT a little bit. So I was wrong about being able to lose points for incorrect answers (most likely). It seems like the way points for each problem is determined is parameters like difficulty of question and chance of guessing question correctly, and probably a few other parameters that I don't know of (different models exist, and I don't know what they're using exactly).

This paper has some good examples that should be easy to understand even if you aren't good at math (the first half, second half gets more mathy): http://www.creative-wisdom.com/computer/sas/IRT.pdf

Sorry for spreading bad information based on my misunderstanding.