“Seems to me, based on Definition B. that say 32,000 kids are being told they are 99% instead of 16,000 kids using SI% Table …”
Again, it just shifts the table by 1 row. Let’s use this year’s SI table as an example.
In this year’s table, there are three 98 percentile ranks, and nine 99 percentile ranks. So around the bottom of the 99s, each row is probably 0.2% - 0.3% (Assuming a normal distribution, the step sizes for each rank increase as you go down towards the middle of the distribution. Maybe @dallaspiano will compute the exact step sizes in this area).
The new table inflates the percentages by one row’s worth. So, we might be telling 1.3% of the kids that they are in the top 1%. Thus it doesn’t double the number that are told they’re in the top 1%, but it might increase it by 30% - which is still a lot.
Note, though, that as you get higher in the table, the inflation declines. So we might be telling 0.54% of kids that they’re in the top 0.5%, for an increase of only 8%.
@Speedy2019 - if by Definition B you’re referring to the percentile differences, see what DoyleB has said. It just means that the equivalent score gets moved by the smallest amount possible. For SI, that’s 1 point. (For total score, that’s 10 points.) So for any NMSF SI cutoff calculated based on percentiles before, just add 1. Literally. It’s as simple as that. (For a percentile-based estimate.)
Disclaimer:
I am still a junior. My numbers was crunched two weeks ago, they were based on PSAT data tables (2012-2014). The numbers were expressed in a subjective, a speculative way, a biased angle or many assumptions and etc… Make it short, there are many flaws, missing points… do not take my numbers seriously or personally. I mean no harm at all. Just like the guessing games of NMF cut off based on calculated and logical reasoning.
To @DoyleB, I believe you know your numbers when you reposted. Thank a lot
Since CB wants to correlate to old SAT and to ACT, is there any way we can correspond to SAT and ACT scores? IN other words, a perfect 2400 is “rare space”, so a perfect 228 would be a similar percentage of test takers and then the percentage of test takers in various scores would gradually widen as you head to the mean.
@Speedy2019 I think the preliminary concordance tables are making a LOT more sense now and my very sketchy work earlier suggests that the concordance tables are now more consistent with the page 11 SI tables. In theory there shouldn’t be any inconsistency and once I started using the CORRECT definition of the percentiles for this year I started to see more consistency than I did before.
I think that Test Master’s commended of 210 is too high. But wasn’t Test Masters trying to be conservative or overshoot the cut-offs a bit? That’s the impression I got from some of their responses to the comments.
@Mom2aphysicsgeek the “common sense” thought is that CB would not want the cut-offs to swing due to the new test (they still could, of course, for some of the individual states - as they would have under the old test). The only ones predicting big changes have been some test prep experts and CC people. If it turns out to be much ado about nothing it’ll be pretty much what we deserve ;))
@speedy2019 But remember, the inflation due to the percentile definition change was only one piece of the overall inflation. Compass showed that at the high end of scores the total inflation was a full percent (chart on page 7). So I think ultimately you will be proven correct that something like 32,000 were told they are in the top 1%
It is more than just switching from definition A to B. The larger 99% group is also attributed to the change in User/National groups (of which we have no good information on).
If you look at the middle of page 7 and top of page 8 of the document above, it explains the problem. It also says counselors are reporting 3% of students getting 99% scores (consistent with posts here) which would mean the commended cutoff is going to be more like 205 (bottom of 99%) with most states increasing a bit in their SF’s. The top states may drop a little.
To post #1645 with @micgeaux,
228 is very rare, and it associates with TS 1520. Then how do reconcile to next 1510 then 1500 to 1480. Just like we have around 500 11th graders at 1520, then 2000 above 1510, then 4000 above 1500, then 8000 above 1490, o la la la we have 16000 above 1480
@LivinProof “National” vs “User” could add the remainder. If you look at the total score chart, at the bottom of the 99s, there are two rows where the “national” is 99, and the “user” is 98. If the SI table uses “national”, then two more rows of inflation need to be added to the row of inflation we’ve already discussed. That certainly might double the number of one percenters.
@Mamelot Deflating the SI% table has to bring it closer to the concordance tables, but it does not close the gulf by all that much in my opinion. Going from low 99+ to lowest 99 doesn’t sound like much, but when the topic is nmsf it’s a deal breaker.
@LivinProof - this is what I’ve been sensing all along: that on the SI 207-220 (99% user and TS reported score sheet plus p11), you could have 35-40k kids instead of 15-17k. This I think is what Applerouth discovered but when he concorded them he got 2/3s of them down to the 98,97,96 and 95th percentiles or UNDER the 99% 2014 (218SI) GA cutoff. THIS IS MY INTERPRETATION. HAPPY TO BE CORRECTED.
@LivinProof - I also find this question interesting.
Certainly one possibility is that they screwed up when testing / weighting their “representative” sample. But that’s not a complete answer, as their sample was used for both percentiles and concordance tables. TBH, I simply don’t understand how they could release a concordance table which doesn’t agree with their percentile table.
Consider a 214 SI on the current test, which is just enough to make it into 99+. If that’s 36M/35R/36W, that concords to 72/67/70 = 209. (Also troubling, changing those numbers slightly to 36M/36W/35R - which keeps the same 214 SI - now concords to 72/71/68 → 211.) But 209 (and even 211) on the previous year shows up as 98 percentile. How can you concord a 99.5% score and get a 98.5% score??
@Mamelot wrote:
“the “common sense” thought is that CB would not want the cut-offs to swing due to the new test (they still could, of course, for some of the individual states - as they would have under the old test). The only ones predicting big changes have been some test prep experts and CC people. If it turns out to be much ado about nothing it’ll be pretty much what we deserve”
Before scores came out, I think a lot of folks assumed the cutoffs would come down because the max score came down.
Then the concordance tables indicated that cutoff would stay the same in the middle, come down a bit on the high end, and rise on the low end.
The published SI table, at first glance, indicated that cutoffs would come down substantially.
Anecdotal evidence suggested that concordance values might be reasonable - SI table not so much.
Understanding of percentile definitions, and the nebulous user vs national, indicate percentile inflation.
Maybe we’ve come all the long way around. I still think concordance is going to be close.
@LivingProof “Which might imply that SI scores are directly comparable to past SI scores, and hence cutoffs would stay about the same as in past years (and would agree with concordance tables)
I now lean toward believing this, but there is one perplexing hurdle to completely believing…if this was their intent, then why put out an SI% table that in no way at all supports it?”
I’ve been spinning wild tales for a week trying to guess why the SI table, which I never believed made sense, would get published by CB. It’s like I’m halfway through a mystery novel - now I can’t wait to find out the result. I’m still wondering if it’s based on “national” - despite the fact that publishing a “national” SI table makes no sense either.
@thshadow: How can you concord a 99.5% score and get a 98.5% score??
@DoyleB: I still think concordance is going to be close
So hard to believe they wouldn’t run a reasonableness test on the SI% table (with the actual data) before publishing. But everything seems to point to that 1% inflation at the high end.
If the commended cutoff ends up being around 208-210 (per concordance), CB will have to get some slick copy writers to explain how the top 50k scorers out of the 1.7 mill test takers can all be at or in the 99% when the 50,000th tester of out 1.7 mill has to be at or in the 97%.
“Consider a 214 SI on the current test, which is just enough to make it into 99+. If that’s 36M/35R/36W, that concords to 72/67/70 = 209. (Also troubling, changing those numbers slightly to 36M/36W/35R - which keeps the same 214 SI - now concords to 72/71/68 --> 211.) But 209 (and even 211) on the previous year shows up as 98 percentile. How can you concord a 99.5% score and get a 98.5% score??”
The whole premise of that question is that 214 is a 99.5% score. I think it is much more likely that a 214 this year is, in fact, at or just below 99%. If that’s true, the concordance works pretty nicely. It’s not meant to be perfect, just close.
@thshadow wrote"Consider a 214 SI on the current test, which is just enough to make it into 99+. If that’s 36M/35R/36W, that concords to 72/67/70 = 209. (Also troubling, changing those numbers slightly to 36M/36W/35R - which keeps the same 214 SI - now concords to 72/71/68 → 211.) But 209 (and even 211) on the previous year shows up as 98 percentile. How can you concord a 99.5% score and get a 98.5% score??"
This is what I struggle with - in our case it is 218 SI (R 35 W- 36 & M 38) – on the one hand it is an SI that is 99%+ but concords to 216 but on page 1 (total to total) his 1470 supposedly concords to 99+%ile and a SI of 223 (ha ha!). So really unclear how to interpret an SI of 218 for me.
