mcelroytutoring wrote:
Anyone who claims to know the exact specifics of the GMAT algorithm (and corresponding score conversions) vs. the GMAT Prep algorithm (and corresponding score conversions) is either lying, or in serious breach of a GMAC confidentiality agreement.
I gather this is directed at me. I don't know where you've gotten your information about the scoring algorithm, but you are misinformed on some important questions.
It is true, in a literal sense, that some of the algorithm is secret. If someone is administering an IRT-based test, there are certain parameters he or she can freely set without affecting the integrity of the test. Only the designers of the GMAT would know how they've set those parameters. But those are really just technical details. The mathematical basis of the algorithm, which is what is important, is well-known. The GMAT uses a three-parameter IRT model, a fact confirmed in many official research reports. I invite you to research that on your own if you are unsure if what I am saying is true, but someone repeating that is neither lying nor in breach of any confidentiality agreement.
The important information that is not public knowledge, and which you would absolutely need to reverse engineer your score from your responses on a GMATPrep or real GMAT test, are the statistics associated with each test question - the difficulty, discrimination and guessing parameter values. Because those question statistics will vary from test to test, a particular response pattern (right/wrong/right/wrong/etc, for example) will typically produce different scores each time. But those statistics have nothing to do with the underlying algorithm, in the same way that the difficulty levels of questions on a traditional multiple choice test, where your grade is the percentage of questions you answer correctly, have nothing to do with how the test is graded. The GMAT's underlying algorithm is just based on probability theory.
I've read dozens of academic papers in this area, and have programmed an IRT-based algorithm using the same model as the GMAT, so while I'm not a specialist in this area, (anyone doing a graduate degree in the field would know more about it than I do) I do know something about it.