NY with 223, base on msg #4125 you are the top half of 99+%tile - very good chance to have NMF and more than 90% to get commended
@WGSK88 We live in Illinois too and have had the opposite experience as what your DD reported at her school. My DS so far has the highest score (218) out of all of the kids he knows. All of the friends of his whose scores he knows have ranged from 202-216, and these are all honors kids. He is supposed to meet with his academic adviser this week and he plans to ask how his score compares to the rest of his school since they historically end up with ~30% of the students making finalist. Hopefully, there is a similar distribution as in years past but I’ll report back what I hear. I think the school GC’s are better positioned to see upward/downward trends since they have access to larger datasets, so hopefully they are more willing to share info than CB!
Texas was 220 last year, 219 for the year prior. One of the links quoted above was accurate, but watch confusing your years. The kids who tested in 2014 would actually be the class of 2016’s NMS. Throw in the old percentile data in recent reports, and it gets complicated. This is why I typically like to use multiple years in my comments. For example, last year’s report dated “Understanding 2014 PSAT/NMSQT Scores” was given to students testing in 2014. These students were from the class of 2016. However, the percentiles in the report were from the students who tested in 2013, the previous year, and who were members of the class of 2015.
Confused yet? If not, make a mental note that reports titled 2011 or prior included data from a sample of the actual testing class (2011 testers, from class of 2013 in the 2011 report). If you have access to all of the reports, you’ll see that the reports from both 2011 and 2012 include very similar data. 2011 from a 2011 sample, and 2012 from complete 2011 data.
Thanks to everyone for researching this topic so extensively. It has been very helpful. I have read most of the discussion, but as a 1440 (217 SI) from Alabama, can I rely on making Semifinalist with a fairly high degree of certainty?
Hi Everyone – i sent a question into the Prep Scholar folks as they had estimated the state cut offs as -12 from last year- just reflecting the change in scale. I just got this response back to my inquiry:
From - Allen Cheng (PrepScholar)
Jan 17, 17:52
Hi there-
Sorry about the confusion - it is true that the concordance table presented by the College Board presents a different scaling than what we had expected.
For example, let’s try California’s 222 cutoff score from previous years, out of 240. We can estimate 74 on each section.
Now we use concordance tables to see what the section score now is:
74 → 37 writing
74 → 37 Math
74 → 36 Reading
NMSC Index: (37 + 37 + 36) * 2 = 220
The reason this is happening, contrary to our 12 point deduction, is that they’re compressing the higher end of the spectrum so top performers don’t differentiate from each other as strongly. Another way to see that is the difference between the maximum score and the cutoff score is higher now than it used to (228 - 220 = 8, 240 - 222 = 18).
We’ll be updating our guides with this analysis to be up to date. Though to be fair the College Board still hasn’t figure out its exact scaling yet, so the best we can do now is estimate.
Allen
Their blog is here - and as dated today this still has the same cut off predictions (I think as before - simply less 12 points from previous cut off): http://blog.prepscholar.com/national-merit-scholarship-cutoff-2015-2016
I imagine they’ll update their projections in the next week or so though
@CA1543. Thank
Allen Cheng is a very smart guy. I trust him very much
Can anyone shed light on how this translates to projected cut off scores - per Allen Cheng (see post 4208 above) “Another way to see that is the difference between the maximum score and the cutoff score is higher now than it used to (228 - 220 = 8, 240 - 222 = 18).” Thanks!! And thanks @dallaspiano for your comment - I think Allen & colleagues at Prep Scholar will do their number crunching & try to give some good guidance - to the extent possible with the limited info available.
At least Allen Cheng independently reached the same conclusion that is the general consensus on this thread and others. That should lend more validity to many predicted cutoff scores that seem insanely high. Hopefully, the new predictions will shed more light from an unbiased perspective (not a parent or student).
I received an email that my handle @WRUAustin was mentioned by @gettingschooled .
I believe that 220 is the Texas Cutoff for the class of 2016. These students took the test in October of 2014. @dallaspiano if you have any information that it was 219, could you please confirm. I think there must have been a mistake in the year or something else. Seems pretty clear that it was 220. I would love to be wrong, however.
@CA1543 Pretty sure that should read “the difference between the maximum score and the cutoff score is SMALLER now than it used to be.”
I am taking SAT next Saturday. I need to study.
Am trying to think this through, but probably erring. Scaled scores used to range from 20 to 80, a 60-point spread. Now they range from 8 to 38, a 30-point spread. For that reason, if one were to construct a frequency histogram with scaled subtest score on the x-axis and number of students on the y-axis, there would be higher y values than if the same number of students had taken last year’s test. In order to equate the scales so that each has a 180-point range, they are doubling the sum of scaled scores. But that still leaves twice as many students (on average) at each score but with no odd scores represented. Fewer scores and therefore more students per score implies a higher, narrower Gaussian (assuming normally distributed data). Of course, the shapes of the two years’ data would differ anyway because the tests are so different. But looking solely at scaling differences, one would predict lower cutoffs but a smaller range of cutoffs. I’m either making sense or babbling…
Can’t figure out how to edit post above, but one CAN have odd scores on Math (when doubled) and therefore on the summed and doubled subscores (i.e., total SI) because there are .5’s on Math, but even Math is squished into a 30-point range despite half steps along the way.
@MatzoBall
If you scroll down there are two percentile graphs.
https://www.■■■■■■■■■■■■■■/blog/2016/01/14/can-you-trust-your-psat-score/
https://www.■■■■■■■■■■■■■■/blog/wp-content/uploads/2016/01/Screen-Shot-2016-01-14-at-9.13.16-AM.png
https://www.■■■■■■■■■■■■■■/blog/wp-content/uploads/2016/01/Screen-Shot-2016-01-14-at-9.13.16-AM.png
Can someone PM me the PSAT questions (all of them) please? I still haven’t received my scores due to some CB delay and it’s killin me
Thanks @Plotinus - the difference is not higher but lower or smaller. ok, it makes more sense that way. I notice that Allen is using the top of the range on page 3 of the concordance tables. Page 4 table yields different results (lower) but not sure how the 2 tables are supposed to be used together or if there mean different things. Let’s see what Prep Scholar comes up with for some projected cut off scores - should be soon.
@CA1543 Allen gave the example of using the concordance tables to convert an old score that has the same subscore of 74 in each section to a new selection index. For additional examples that assume the same score in each of the three sections, see BunnyBlue #1470 p. 98.
I just read the Applerouth analysis for the first time. Thanks for this very interesting link.
According to Applerouth, CB is inflating the percentiles of the redesigned SAT scores. While I think this is a very helpful analysis and I don’t think his conclusion is wrong, I think it somewhat obscures what is happening.
We all know that College Board recentered the curve in 1995. Median and average scores were dropping, especially in verbal, and CB artificially fixed this by just adding points to scores.
If you add the same number of points to EVERYONE’S score, you don’t change the percentile ranking of any scores. If you add a DIFFERENT number of points to different scores, there is going to be some movement of percentile rankings. In this way, you can move the lower scores up. Recentering was like Christmas, but Santa loved people who were not doing so well more than the people who were doing great. That way the whole distribution of scores was shifted in a way that raised the middle and squashed the top.
Here is an analysis of what happened in 1995:
In this document, first CB claims again and again on page one that by adding points to scores, it has not changed any percentile rankings. Then, in the table page two, it admits that it HAS changed the percentile rankings, especially in Verbal. That is, the table shows that in 1995, CB produced the follow changes:
old verbal 1% = new verbal 5%
old verbal 4%= new verbal 11%
old verbal 8% = new verbal 21% (and other changes not stated by CB).
The 1995 recentering looked really bad. It looked like a cover-up of a generalized decline in academic performance. Adcoms were told that 650 Verbal=730 Verbal. They could not understand that.
This is one reason I am always a little skeptical about curved scores.
Indeed, I always suspected that when CB redesigned the test back in 2005, one of the motivations was to be able to recenter the curve (add points to middle-low scores) without admitting it.
I also suspected that when CB redesigned the test in 2016, the same motivation was at play: let’s recenter the curve again on the sly.
Well, it looks to me that Apperouth’s analysis proves that this is in fact what is happening. Applerouth shows that CB is concording a 2014 42nd percentile score to a 2015 54th percentile score, and a 2014 94th percentile score to a 2015 98th percentile score. Another way to look at this is that the concordance table is recentering the curve by artificially adding points to scores. The people in the old 42nd percentile got more points than did the people in the old 94th percentile. The people in the top of the 99th percentile got NADA.
It was more transparent back in 1995.
Another really fascinating read for all the stats people out there is the 2002 ETS research report entitled,
“The Recentering of the SAT Scales and its Effects on Score Distributions and Score Percentiles.” You can find it here:
https://www.ets.org/Media/Research/pdf/RR-02-04-Dorans.pdf
This report gives an analysis of the methodology, effects, and justification for adding points to scores.
I am not a statistician, but it looks to me as though there are a lot of questionable assumptions in it.
Some of the justifications for adding points given in the report:
- Median scores were continuously dropping.
- The population taking the test was growing and changing in character (becoming progressively less elite)
- Verbal and Math scores were out of “alignment” in the sense that mean and median Verbal scores were lower than mean and medial Math scores.
Sound familiar?
The criteria given on page 4 are basically that we should have a nice, symmetric bell curve in which the mean=the median=the midpoint of the score scale, and the midpoint scores are the same for math and verbal. Whether this kind of score distribution actually reflects the distribution of skills, ability, intelligence, etc., in the tested population is not considered relevant.
It is also stated on page 4 that we should expect to have to “correct” the curve on a regular basis.
Another interesting related issue is “score drift” produced by the equating process.
The equating process is not perfect, so when ETS saw that median scores were dropping, it hypothesized that maybe the reason for the drop in scores was that the newer tests were actually harder and this had been covered over by accumulated small equating errors.
http://onlinelibrary.wiley.com/doi/10.1002/j.2333-8504.1975.tb01048.x/abstract
A study was carried out in 1975 to determine whether there had been a “drift” in test difficulty between 1963 and 1973. The study determined that there had indeed been a drift in scores due to accumulated small equating errors, but that the drift had been UPWARD. That is, the 1963 tests were HARDER than the 1973 tests, not easier. The 1963 tests were judged 14 points harder in verbal and 17 points harder in math. So in effect, the drop in scores was even worse than had been thought.
ETS now regularly tests for score drift due to equating errors. For example, here is the study carried out for 2005-2010:
https://www.ets.org/Media/Research/pdf/RR-12-15.pdf
According to this study, between 2005 and 2010 there was an upward drift of around 5 points in math and cr, and 11 points in writing. If we assume a similar drift from 2010 to 2015, this would make the 2015 tests effectively 40 points easier than the 2005 tests. This might also explain why some people claim that the OC Tests are “harder” than the real tests: the first couple of OC Tests are QAS’s from 2005 and 2006.