@BunnyBlue & @Mamelot – & others too - with the changes CB made to pg 11 Posts 2172 & 2173 – Are you now thinking the SI percentiles on pg 11 here are correct: https://collegereadiness.collegeboard.org/pdf/2015-psat-nmsqt-understanding-scores.pdf
then the question is the effect of the change in definition of percentile (at or above?) and is there a strong thought that moves each score up a notch so 99.5%ile starts at 215? This though would not really align to Test Master predictions though would it – for example NY is predicted to go down to 217 but that is in the 99.5%ile & in the past the cut off of 218-219 was in the 99%ile. It is hard for me to understand (I am NOT a math or stats person), how likely it would be that the reported (on the 2015 Understanding Score Report) Mean SI is 148 (up 6 or more points from last year although the overall scale moved down 12 points)) and yet the 99:5%ile would go down so much (range of 99.5%ile scores in 2104 understanding scores was 224-240 & now per CB it is 214-228) - even adjusting a bit for the change in definitions of the percentiles.It makes sense I suppose if the curve was set to boost more students up in the percentiles but at the high end bands nationwide there are fewer students, such that 99.5%ile is 214 or 215. Thoughts?? (Please be kind - I am just trying to make sense of this). 
Need my coffee now!! ~O)
@CA1543 the page 11 percentiles conform to the mean and std. dev. that CB finally inserted. We still don’t know if that’s based on actual, research or other data. And the mean IS higher than previous years but the standard deviation is lower (26 vs 31). It’s a tighter curve around the mean than prior.
@thShadow yes they have less flexibility to drop from the top, so to speak, especially if a whole bunch of kids miss one or two. However, because scaling for each test is separately derived, missing one or two on reading isn’t the same thing as missing one or two on math or on writing. They can still control for the difficulty of each section. However, if 10% of their test takers got perfect scores or missed no more than one, I can see where they would have a bit of a challenge fitting any kind of normal curve at that high end. In reality this is what we might see - that the actual curve would look normal for the most part with some weird stuff going on at the upper 1% but they aren’t showing that on page 11 because that curve is supposed to be representing longer-term trends rather than strictly actual data.
Guessing that in future years predictions CAN be made using historical data that starts this year - but this year all bets are off in establishing a relationship between the Cut-offs and the percentile table or historical data showing where Cuf-Offs used to lie on the percentile tables. The curve has been reset and that changes the rules of prediction but we don’t know those new rules yet.
@rb681000 and @WGSK88 I’m wondering if that SI number of 216 is the average of that tutoring/prep clientele. Again, that’s pretty high and matches the Walton data. Can we conclude that kids who prepped well scored an average of 216 on this test? If so - Yikes!!!
I also am just waking up and also need my coffee. But whether 214 or 215 was listed as the 99+% could be influence by rounding of the numbers as suggested by @thshadow or could it be due to the changing definition of percentiles as we all have discussed before. I must admit I was not following those discussions that closely but doesn’t the changing definition of percentile shift the scores down by one row. Let’s say from 215 to 214. I really don’t know. But as some one else pointed out, the percentiles for 2015 follow the normal curve far more closely than they did in 2014 at the high end of the curve. But like others have pointed out, CB controls the scoring but at the top end there is only so much room (so many questions wrong in three separate sections) to play with.
There was a curve - pretty generous I thought for math & reading so maybe CB did take into account actual test takers in setting the curve. But I am not understanding how 215 or 216 is likely to be true 99.5%ile ,
I did just notice that on the online full score report for my son, the total score is 1470 - 99Th%ile & under the User percentile description it says:Nationally Representative Sample Percentile
"Your PSAT/NMSQT User Percentile - National
“Your percentile indicates the percentage of a typical group of 11th grade College Board U.S. test takers who would have had scores at or below your score. The average score for the 11th grade College Board PSAT/NMSQT typical group is 987.”
But we have heard from the various GC reports, the actual national mean was about 1009-1010, so the User data in the reports is lower than actual by 22 --23 points. If the percentile charts (e.g pg. 11) are off the mark by a similar percent (2.3% (meaning actual of 1009-1010 is 2.3% higher than the 987 User percentile for the total Score)?), I wonder what the likely implications are for the percentile charts if CB tried to achieve results & overall curve similar to the past as much as possible. Of course this is still just dealing with the mean but I wonder if there is any way to estimate impact up the curve - I am not capable of the - even have a few cups of coffee! CB – please release the State Summary reports!
Today’s hypothesis:
The test was too easy for good students. This created a bump in the results histogram for the test that was impossible to fix with the scoring tables.
If you look at the histogram for a test with a normal distribution, the number of students with each score decreases monotonically when going from the high end to the mean. Sometimes there’s an exception for a perfect score, but lower than that it doesn’t happen. Another way of saying this - take the number of kids who made a 215. The number of kids who scored a 216 is lower than that. The number of kids who scored a 217 is lower than the number of kids with a 216. Etc, all the way to the end.
I believe this didn’t happen on this test - there’s a local peak (or cluster) in the 217-219 range. I believe we will find that results similar to those reported by @WGSK88 will not be atypical - they had more kids with a score of 218 than with a score of 217, 216, or 215, etc. This is not the expected result from a test such as the PSAT,
So this year you have a multimodal histogram rather than a unimodal histogram. The SI table was based on a research sample and assumes a normal distribution. Thus, that table is useless for predictions for this year.
Crude Example using characters, going from the high scores to the mean… This may not come out looking like what I want it to, depending on spacing. But I’ll try.
Typical histogram
**
**
**
**
This year:
**
**
*
*
*
*
**
**
**
Excuse my typos - I meant to write “increases monotonically”, not “decreases monotonically”. And technically I should have referred to it as a bimodal histogram - I believe there will only be two peaks, not more than two.
@DoyleB, interesting idea. I think it’s more likely to happen if the test as a whole was a little easy but some of the hard questions were missed as a group. Eg., if a lot of kids did well overall but say missed all 4 of the Douglass questions.
Was the test group not large enough to see what the high end curve was?
Also, this ties in to what I wrote some posts ago. I believe more kids will miss NMSF this year by a single point than ever before, because the cutoff will be somewhere in the “bump”.
I also heard few 218s. That’s one reason I think CA cutoff will be 219-220.
Based on the College Board posted percentile differentiation, California will probably be good with 214-216
< sigh >
@DoyleB – I know – hard though to “get up the curve” on this and the percentiles CB put out are rather misleading.
@exeover - http://collegeadmissions.testmasters.com/update-psat-scores-cut-national-merit-2016/
These estimates - which put Cal at 219 might or might not be realistic - many are predicting a bit higher actually for Cal & some other states.
That is 100% correct. So a 215=99+% means that most folks who scored 215 and up can be pretty confident about their prospects, almost no matter what state they are in. There have been some folks on this forum who make the strange argument that there are somehow more 99+%'s than usual. That is mathematically impossible unless a lot more Juniors took the test this year than last year. Was there a baby-boom in 1998 that caused a sudden increase of 17-year-olds in 2015-16? If so, I’m not aware of it. There are a lot of CB sub-category percentile measurements, but only one final percentile measurement, and a 215 -and up equals 99+%, no matter which way you slice it. That is equivalent to a 224 and up under last year’s percentile measurement.
@exeover You are aware that the published SI table is not based on students who actually took the test, right?
@execover 2193 if you read through this thread you will see there is a lot of skepticism about the published percentile tables. In short, there are just too much many reports coming in that have 3x as many kids as there should in the 99% range. Also, there are concordance issues.
@F1RSTredeo, your post #2153
99+ = 222-226 ===> the mid is 224? yes
99 = 221-214
98 = 213-209
97 = 208-206
I will compare with the mid of 2013 data table
2013 SAT data table (http://media.collegeboard.com/digitalServices/pdf/research/SAT-Percentile-Ranks-Composite-CR-M-W-2013.pdf)
*** (Reason I pick Y2013, since it had 11 band slots in 99+%, so it can be compared with 11 band slots in PSAT 2014 - you used) ***
2014 PSAT data https://longleafschool.files.■■■■■■■■■■■■■/2014/09/guide-to-understanding-psat-scores-2014-college-board.pdf
We have some very real data, then I project them to 2015 PSAT SI by mapping
2013 # 99+% Or Y2015 PSAT
2400 494 99+ 99.97 228 +++++ I compress data at the top since at higher, the lesser difference i
2390 245 99+ 99.95 226-227
2380 434 99+ 99.92 224-225
2370 413 99+ 99.91 223-223
2360 641 99+ 99.89 221-220
2350 660 **** 99+ 99.82 219
2340 890 99+ 99.77 218
2330 897 99+ 99.71 217
2320 1,188 99+ 99.64 216
2310 1,187 99+ 99.57 215
2300 1,237 99.+ 99.50 214
Usually at this spot (2350) we have 2487 Students, with your proposed analysis we may have 7500+. It is the widest jump and it never happened in CB history, I may miss something here?
Project like your mid 224, we have 1172, it is even wider than the widest that i have above
I am aware of what the College Board published for 2014 test takers, which was a 99+% starting at 224, and the 2015 99+% starting at 215. I certainly trust the CB numbers more than speculators on an Internet discussion board. Especially speculators who argue that there is a higher percentage of 99%'s this year when that is mathematically impossible… Unless there are many more Juniot test takers this year, of which there is no evidence. The CB percentiles are credible. The strange arguments about “inflated” percentiles are not.
< sigh >
Hi @Speedy2019
I thought your table with the 2014-2015 concordance was interesting, thanks for putting it together! But when I did it I got a different range. I only looked at 2014’s SSI of 225 and came up with a range for 2015 SSI’s of 216 to 224.
At the one end, a 225 could be achieved with scores of 80 W, 80 R and 65M. If I read the concordance tables correctly, this equates to a 38/38/32 which produces an SSI of 216.
At the other end, a 225 could be generated with an 80 W, 70 R and 75 M. This would generate 38/37/37 for an SSI of 224.
The middle of this range is 220. If I did the concordance correctly, this would equate to the Testmaster prediction of 220
Thoughts? I am missing something?