Folks, here is a list showing 2014 possible scores and what the 2015 concorded score would be. The concorded score is a range except at the very high end. Hopefully did all the math right. Next message I’ll relate the ranges to Testmasters predictions.
Notice below, how starting at 2014 score 222 the SI is always in the concorded score range. I think that is why people might conclude the cutoffs this year will be similar to last year. That is - IF you believe the concordance tables instead of the SI% table.
@F1RSTrodeo,
Since you said you do data table by “comparison of 2015 to 2014” but not by “what “theory” /“assumption” / “hypothesis””.I accept your way(s) at face value
I do mapping by comparison like you did in post #2153
I assume I can use the lowest of 99%tile range as reference point - or starting point. It will be like this
SI in SI in
2015- 2014
213 - 222-223 ** assumption: the higher it goes up, the lesser difference at this data point
212 - 220-221
211 - 219
210 - 218
209 - 217
208 - 216
207 - 215
206 - 214
205 - 213
If it’s possible, please explain why my mapping is so much different from yours. Thank you in advance
Not sure how well the columns will line up below. Anyway, Testmasters stated they were using a conservative approach to their cutoff predictions. Thought I would compare their prediction against the 2015 concorded score range.
That table is below. Looks to me with the high cutoff states, they were liberal. For example, Maryland had a cutoff last year of 222, the concorded range is 218-222. Their prediction is 218 - on the low side of the concorded range.
With the low cutoff states, they were conservative. For example, West Virginia had a cutoff last year of 202, the concorded range is 198-210. Their prediction is 210 - on the high side of the concorded range.
It is their mixture of liberal and conservative approaches that gives us high states decreasing, mid states staying the same and low states increasing. Not sure what their philosophy was on mixing liberal/conservative.
Finally, I included a complete liberal cutoff column and a complete conservative cutoff column without mixing the 2 as testmasters did. Just for the exercise of it.
The liberal approach has a cutoff range of 198-220.
The conservative approach has a cutoff range of 210-224.
The mixed approach used by Testmasters has a cutoff range of 210-220.
Testmasters cutoff range is close to the fully conservative approach. I guess that is why they said “conservatve”, but they actually used a mixture.
@likestowrite, your comments in post #2144 about this
Score 2015% Def B Converted to Def A
204 98 98% of students scored at or below this score → 97% of students scored below this score
203 98 98% of students scored at or below this score → 97% of students scored below this score
202 98 98% of students scored at or below this score → 97% of students scored below this score
201 97 97% of students scored at or below this score → 96% of students scored below this score
200 97 97% of students scored at or below this score → 96% of students scored below this score
I TOTALLY agree with you
But, I believe we are discussing the cut off for NMF - very extreme ends of SI scores. In previous post, @VandyAlum93 pointed out that at higher end the change would be insignificant (less than 0.1%)
I was interested in the theory that @AJ2017 came up with that CB might have extrapolated percentiles by assuming that the score distribution had a normal curve. According to some, 99+ percentile means at least 99.5th percentile. I don’t know if this is true, but I decided to calculate what the 99.5th percentile would be for the 2015 PSAT with mean of 148 and standard deviation of 26. I looked up what the z score is for 99.5th percentile and found it is 2.58. This means that 99.5th percentile is 2.58 standard deviations above the mean. From this I calculated that the 99.5th percentile score is 215.08. That is, 148 + 2.58(26) = 148 + 67.08 = 215.08
Your formula might be slightly more precise than the CB statistician’s formula to calculate 99+percentile . 214 is too close to what would be predicted bases on the mean and SD that the CB now provides, right? Does anybody really think that the scores of the top 1% would follow so closely what would be predicted based on a normal curve? We could probably come up with several reasons why the curve would be more elongated, thickened (whatever would be the correct term to describe behavior at the end part of the curve). Like I said, I am no statistician. I am just trying to reconcile the percentiles with the anecdotes and my gut feeling. What do you all think?
@AJ2017 they don’t fit it as best they can - they can actually make it so. They totally control the scaling and can fit the raw data to any kind of curve they want. Perhaps they are more interested in a curve that resembles what they presented on page 11 and it will shift over time to represent something like that curve. Certainly page 11 does NOT fit the anecdotal data we are hearing. But the idea that somehow a whole bunch of scaled scores accidentally became higher than they thought makes no sense because they would have just been more harsh at the top.
@SLparent the mean and std. dev. as included on page 11 now are probably now correct. They were a cut and paste error when the report first went online and so it was nonsensical.
@Lea111 all SI scores are integers. However, the relevant data from CB is the Percentiles. Only the band of SI’s are percentile integer, for instance 204-202 is 98%ile (def B). The relevant info is that the Integer 98 is attached to the scores 202, 203 and 204. In light of this, 202, 203 and 204 are Not integers but a fraction of a percentile of 1.
@dallaspiano I appreciate that my chart does not solve the issue of TX and other high scoring states. Hw, there are plenty of states whose traditional cut off for NMSF is at the lower end of 99% or the upper end of 98 % or the mid of 98% or even the lower end of 98%. Thus, this information Is relevant for NMSF in most of the states. Just not for those states with higher cut offs.
@AJ2017 I don’t know what to think. I liked your idea, but it seems that if CB were just extrapolating from a normal curve they would have designated the score 215 as the lowest 99+ score, not 214. I happened upon a more precise table that gave the z score for 99.5 percentile as 2.5758. Using that I came up with the score 214.97 as the 99.5 percentile for a normal distribution with mean of 148 and standard deviation of 26. (148 + 2.5758(26) = 214.97)
@BunnyBlue they are a bit more precise this year than in 2013 (the last SI table we have access to). With a mean score at that time of 141.9 and Std. Dev. of 30.7, using the Z-stat. should give us a 99+ score of 141.9 + 2.5758(30.7) = 220.97 so about 221. When, in fact, they characterize the 99+ range beginning at 224. Perhaps every time the test is revised they start a-fresh with a very precise normal curve and then there’s some drifting or whatnot over time after that.
@Mamelot - that’s not completely true. If there were huge swaths of test takers that only got a couple wrong for example, practically speaking there’s not really anything they can do to make it work out. There aren’t that many different raw scores at the high end, so they don’t really have total control. For example, say 0.1% got none wrong, and 0.1% got 1 wrong. That would make for a very top heavy curve. Would they scale it so that none wrong was a 228, and a single wrong answer anywhere would yield a 214?? With no possibility for a scaled score from 215 to 227? Even if they did that, it still wouldn’t be accurately normal, as all percentiles from 215 to 227 would be identical (when they should be ascending).
But maybe these pathological distributions aren’t realistic enough.
I’ll point out that even though the report says that the mean is 148 and the std dev is 26 - that’s clearly just rounded, and they presumably have a few more decimal places. For example, if the std dev is actually 25.62, then 214 is exactly 99.5%. Or maybe the mean is 147.8 and the std dev is 25.7. (Also not known - if the percentile came out as 99.48%, would they write that as 99+%, or 99%? I’d guess the latter, but I’m not 100% sure…)
@rb681000 In Illinois I don’t think it will be less than 218. Please see my posting based on the actual data one of the Schools in Illinois. Again this is just a sampling of one School but if few more Schools with consistent NMSF representation data is obtained we can be sure or close enough about our IL cutoff predictions for NMSF for the 2017 Graduating class (http://talk.collegeconfidential.com/discussion/comment/19234638/#Comment_19234638). Please understand all these are only guesses but it is an more actual data oriented intelligent guess.
I very much hope my daughter who has 218 gets in NMSF which looks like based on her School record of NMSF from 2011 unless something drastically goes wrong and her School this year has only 10 NMSF!
Thanks @Speedy2019 I find your chart on Post #2160 & 2162 very useful as ranges are likely best we can do at this point. I am hoping Test Masters is fairly accurate but without lots of state by state data, no way to get comfortable. Appreciate everything you all are pointing out.
@WGSK88 – Just curious at your daughter’s school do you think many students prep for the PSAT? You do have a lot of higher scores. I would think a 218 is fine for IL’s NMSF but of course can’t assure that.
@CA1543 Her School did not do any preparation for PSAT. My daughter though a very good student had not taken any standardized tests before like ACT/SAT (including PSAT in her sophomore year), at the last minute I did put her in Princeton Review in Sept/Oct for 6 classes for New PSAT since she had timing issues in Reading comprehension. Though first couple of these coaching classes she didn’t like but then she continued the 6 classes which might have helped her in Reading. Math she was already good. That might have helped her to secure 710 in CBRW Total score. She had a perfect Math score.