ASVAB Scores

[sjard]

The problem with the ASVAB for gaming is that it is an aptitude test. It serves effectively no other purpose than to find out what a prospective recruit might be good at.

At the time I took it in the early 90s, all that was required for enlistment, if you had a high school diploma, was score a 22 or higher. With a GED you needed to score 45 or higher. This was for Army National Guard only, I don't recall other branches.

As I scored a 98, aced the engineering aptitude section, and was enlisting with an engineering unit, big surprise I got offered a combat engineer MOS as best match.

And that is the part that makes the ASVAB rather pointless from a gaming standpoint. Think of it as closer to the way D&D used to be. You rolled your stats in order (no rearranging) and then figured out which class you could be based on what your stats were. The ASVAB is just to figure out what MOS you might be good at, and really nothing more.

[Kromm]

<Moderator>

Purple Haze, your insinuations that soldiers are "retarded" and such really is not helping the OP. It is merely generating complaints. Please drop it. Thanks in advance.

</Moderator>

[malloyd]

Originally Posted by Dragondog

I know they renormalize them. But as I explained, I'm not after a single number, though that 65% is useful. It seems you do not understand my question and I’ll leave it at that.

Assuming I'm understanding your question, you may be missing the point of renormalizing the scores.

I can't speak to the ASVAB, but for most standardized tests, this process means there *is* no relationship between how many questions right produces what score that holds for more than one test. The way these things work is each pool of tests has some set of questions that are substantially similar to questions in a previous pool. You use the number of those answered right or wrong relative to the total number of questions gotten right or wrong to determine how hard this particular test was relative to previous tests. That is, you assume people who get a pool question right should on average have the same score as people who got it right on the prior test. If last time the people who successfully answered it got 30% of the questions right overall, and this time they got 80%, this test was (vastly) easier, and so you need (lots) more right answers to be awarded the same score. Realistically you use multiple overlapped questions and normalize over lots of these relationships, but the principle holds - it's perfectly reasonable for 10 out of 25 right answers to map to a really bad score last year, and a really good score this year, depending on how hard that particular batch of tests was.

[Dragondog]

Originally Posted by malloyd

Assuming I'm understanding your question, you may be missing the point of renormalizing the scores.

I can't speak to the ASVAB, but for most standardized tests, this process means there *is* no relationship between how many questions right produces what score that holds for more than one test. The way these things work is each pool of tests has some set of questions that are substantially similar to questions in a previous pool. You use the number of those answered right or wrong relative to the total number of questions gotten right or wrong to determine how hard this particular test was relative to previous tests. That is, you assume people who get a pool question right should on average have the same score as people who got it right on the prior test. If last time the people who successfully answered it got 30% of the questions right overall, and this time they got 80%, this test was (vastly) easier, and so you need (lots) more right answers to be awarded the same score. Realistically you use multiple overlapped questions and normalize over lots of these relationships, but the principle holds - it's perfectly reasonable for 10 out of 25 right answers to map to a really bad score last year, and a really good score this year, depending on how hard that particular batch of tests was.

I understand the meaning of renormalizing the scores, but I find it strange that they would be constructed with so different levels of difficulty. I figured that the difficulty of the test would be about the same no matter when you took it. If the difference is that big, the data I was looking for is less useful than I had thought, but could still be of some minor use.

[malloyd]

Originally Posted by Dragondog

I understand the meaning of renormalizing the scores, but I find it strange that they would be constructed with so different levels of difficulty. I figured that the difficulty of the test would be about the same no matter when you took it. If the difference is that big, the data I was looking for is less useful than I had thought, but could still be of some minor use.

Undoubtably the test designers have a target center, since you'd like to be able to resolve the high and low ends of the range too, so you don't want to squeeze the distribution too far toward either end. But it's undoubtably expressed in something much more complex than "number right". Hm, naively it looks like on purely theoretical grounds the ideal target is a breakpoint you are most interested in (whatever that is here, it's not clear to me even what the ASVAB is *supposed* to be measuring) should fall at exactly half the answers right, while two equally interesting points should be set equidistant from it and at least one standard deviation above and below 50% it but I wouldn't be too surprised if the actual math worked out differently.

In any case the point of renormalization is that you don't have to get the test design right on that target for the test results to be comparable across editions, and more or less by definition the more interesting zones for a template design (that is the ones far from the median that might translate into substantially better or worse than average scores in something on a template) are going to be the ones with the highest variability. I can't see what you could be using this data for that that actual score wouldn't be more useful than the number of correct answers anyway - that's the *point* of renormalizing after all, to obtain a number comparable across tests. One can argue this number is equally irrelevant, but it certainly isn't any less useful than the raw correct answer count would be.

[Dragondog]

Originally Posted by malloyd

Undoubtably the test designers have a target center, since you'd like to be able to resolve the high and low ends of the range too, so you don't want to squeeze the distribution too far toward either end. But it's undoubtably expressed in something much more complex than "number right". Hm, naively it looks like on purely theoretical grounds the ideal target is a breakpoint you are most interested in (whatever that is here, it's not clear to me even what the ASVAB is *supposed* to be measuring) should fall at exactly half the answers right, while two equally interesting points should be set equidistant from it and at least one standard deviation above and below 50% it but I wouldn't be too surprised if the actual math worked out differently.

In any case the point of renormalization is that you don't have to get the test design right on that target for the test results to be comparable across editions, and more or less by definition the more interesting zones for a template design (that is the ones far from the median that might translate into substantially better or worse than average scores in something on a template) are going to be the ones with the highest variability. I can't see what you could be using this data for that that actual score wouldn't be more useful than the number of correct answers anyway - that's the *point* of renormalizing after all, to obtain a number comparable across tests. One can argue this number is equally irrelevant, but it certainly isn't any less useful than the raw correct answer count would be.

It seems I'm not explaining myself very well. I'm looking for the percentage of test takers who get the various raw scores in the various subtests and what ASVAB scores those raw scores are transformed into.

With the former it is possible to calculate the latter, but if both sets of numbers are already available, I didn't want to recalculate them. And you obviously cannot get the raw scores from the ASVAB scores.

My plan called for both sets of numbers, but if they’re not available they’re not available, and I’ll have to figure something else out.

[cosmicfish]

Originally Posted by Dragondog

My plan called for both sets of numbers, but if they’re not available they’re not available, and I’ll have to figure something else out.

I still don't see what your plan actually is or why you need this level of detail! Let's say that you had a set of numbers like you are looking for - what then? And if you really DO need these numbers for a game, remember that you only need results from one test or year, not a rule that covers EVERY test.

[Akicita]

Originally Posted by Dragondog

I figured that the difficulty of the test would be about the same no matter when you took it. If the difference is that big, the data I was looking for is less useful than I had thought, but could still be of some minor use.

The test changes all the time, and the level of difficulty changes, too. The data you want would be different every time the test is given.

Originally Posted by Dragondog

It seems I'm not explaining myself very well. I'm looking for the percentage of test takers who get the various raw scores in the various subtests and what ASVAB scores those raw scores are transformed into.

Again, different every time the test is given. The ASVB just doesn't do what you want it to.

[Langy]

Originally Posted by Akicita

The test changes all the time, and the level of difficulty changes, too. The data you want would be different every time the test is given.



Again, different every time the test is given. The ASVB just doesn't do what you want it to.

I don't think it matters that the data changes every time - I think he's likely looking for one set of data from one single test. Still might not be publicly available, though.

[sjolly75]

Originally Posted by Dragondog

Working on some military templates for my campaign world, I've been looking at the ASVAB test. Are there any statistics for the percentage of correct answers test takers usually get for each subtest and what ASVAB score that gives them?

Dragondog,

I'm a Navy Chief Petty Officer (Nuclear trained) with over 18 years time in service. I've served as a recruiter (about 8 years ago), and was wondering why you want to (what it seems to me) to correlate ASVAB to stats. Are you trying to ensure they meet "minimum requirements" to be a part of that unit?

Honestly, since the ASVAB consists of several different sections, running from Verbal Expression (VE) and Math Knowledge (MK), both kind of like sections on an SAT test, to General Science (GS) and Mechanical Comprehension (MC), which are more helpful to determine strengths in various areas - it is probably easier to just have them create the characters and ensure they meet certain minimums in the various attributes.

Hopefully that helps a little, and if you have any specific questions, please feel free to ask!

Steve

Stephen C. Jolly
EMC(SW), USN