The most prestigious law school admissions discussion board in the world.
|
 |
 
|
Can Stats/Math Dorks Explain NCE (Normal Curve Equivalent) re Standardized Tests
Poast new message in this thread
Date: November 20th, 2025 9:02 PM Author: ...,,..;...,,..,..,...,,,;.., wtf is this shit:
https://en.wikipedia.org/wiki/Normal_curve_equivalent |
Date: November 20th, 2025 9:05 PM Author: metaepistemology is trans yeah its a normed linear scoring scale used to make different tests comparable by avoiding non-linearities of the percentile system most standardized tests use. basically the issue with comparing percentile ranked scores is thar percentile ranks are ordinal, not interval so the density of the curve compresses percentiles near the tails making a lot of modeling on them not valid or more difficult (http://www.autoadmit.com/thread.php?thread_id=5800512&forum_id=2!#49448326) |
 |
Date: November 20th, 2025 9:07 PM Author: ...,,..;...,,..,..,...,,,;.., how does this make different tests comparable? like SAT Math and SAT verbal? (http://www.autoadmit.com/thread.php?thread_id=5800512&forum_id=2!#49448329) |
|
Date: November 20th, 2025 9:14 PM Author: metaepistemology is trans you can't really do that because those tests aren't normed to teh same reference population. the sections each have their own raw to scaled score conversions and are normed and equated separately for each section. they are measuring diff constructs and the distributions differ. you can compute NCEs for each test but they preserve the structure of the underlying z-scores. all the NCE is is a linear transforms on teh z-score which is going to preserve the underlying distribution (http://www.autoadmit.com/thread.php?thread_id=5800512&forum_id=2!#49448344) |
|
Date: November 20th, 2025 9:19 PM Author: ...,,..;...,,..,..,...,,,;.., i can look at something like this and see how they took away normal distribution and stretched it out linearly: https://www.centralriversaea.org/wp-content/uploads/2023/06/C4_Normal-Curve-Equivalent-NCE-Revised-5.22.17.pdf
i just dont know practically what that means. if ur in the 95th %ile, that means you've got an NCE of ~85. i dont know what you do with that info, and how that helps you compare a single thing. |
|
Date: November 20th, 2025 9:26 PM Author: metaepistemology is trans Percentiles are ordinal. They tell you where you are relative to other people taking the test but the spacing isn't equal. NCE is a linear transformation of the underlying z-score (assuming normality). So in your example you have an NCE of 85 which is 1.645 SD above the mean on an equal interval scale where 50 is the avg and 21.06 is one SD. That right there *is* the entire added value of NCE. It puts your position on a linear scale. (http://www.autoadmit.com/thread.php?thread_id=5800512&forum_id=2!#49448381) |
|
Date: November 20th, 2025 9:28 PM Author: https://imgur.com/a/o2g8xYK Assumption of linearity is a logical fallacy (http://www.autoadmit.com/thread.php?thread_id=5800512&forum_id=2!#49448385) |
|
Date: November 20th, 2025 9:30 PM Author: metaepistemology is trans no. its a modeling choice (http://www.autoadmit.com/thread.php?thread_id=5800512&forum_id=2!#49448388) |
|
Date: November 20th, 2025 9:32 PM Author: https://imgur.com/a/o2g8xYK Models should be driven by theory. This one is only designed to make it easier for an ape to evaluate test takers. Furthermore the accuracy of the ape's evaluation is not enhanced by this. The new numbers give you no additional information about the test takers (http://www.autoadmit.com/thread.php?thread_id=5800512&forum_id=2!#49448390) |
|
Date: November 20th, 2025 9:35 PM Author: metaepistemology is trans I'm with you on this one (http://www.autoadmit.com/thread.php?thread_id=5800512&forum_id=2!#49448398) |
|
Date: November 20th, 2025 9:45 PM Author: ...,,..;...,,..,..,...,,,;.., can you give a practical example of the added value? (http://www.autoadmit.com/thread.php?thread_id=5800512&forum_id=2!#49448426) |
|
Date: November 20th, 2025 9:53 PM Author: metaepistemology is trans Sure. So suppose you take the SAT math twice. First time you get percentile 95, NCE 85. Second time you get percentile 97, NCE 88. Here the percentile change is +2 and the NCE change is +3. The percentile change looks tiny and ambiguous because the percentiles compress more at the tails, where the NCE change is a direct mapping to the improvement in standard deviation. Consider further a comparison between two students at diff parts in the distribution. First student gets 95th percentile NCE 85. Second gets 75th percentile NCE 63. So the percentile difference is 20 vs 22 points for NCE which seems fine, but percentile spacing is non-linear so the gap between the 95th and 75th is far larger than the gap between 55th and 35th even though both are 20 percentile points apart. On NCE you are getting a difference that maps directly to standard deviation. There isn't a lot of added value IMO but it makes it easier to see the difference in SD, where the percentile is showing you the empirical shape of the distribution. (http://www.autoadmit.com/thread.php?thread_id=5800512&forum_id=2!#49448447) |
|
Date: November 20th, 2025 10:09 PM Author: CapTTTainFalcon could you use lsat data to show that the average nigger (164) at hls is life-support-machine braindead retarded compared to the average kike (173) there (http://www.autoadmit.com/thread.php?thread_id=5800512&forum_id=2!#49448467) |
|
Date: November 20th, 2025 10:12 PM Author: https://imgur.com/a/o2g8xYK I only see it being useful if you're looking for an excuse to dumpster dive and admit students with scores that might get them dinged in the past (http://www.autoadmit.com/thread.php?thread_id=5800512&forum_id=2!#49448473) |
|
Date: November 20th, 2025 10:19 PM Author: metaepistemology is trans The lsat uses a 3PL model internally. Exact theta values aren't published but public reconstructions from admissions cycle data consistently show 165 to have a theta of around +1.3 - 1.5 and 173 to be +2.2 to +2.5. So the gap between 164 to 173 is on the order of .8-1.0 SD on the latent ability scale. That difference is huge in terms of intellectual ability metrics. The raw score gap looks small but in the upper tail, each additional correct item represents a disproportionately larger jump in estimated ability. It is basically the difference between a 90th percentile and a 99th percentile on an IQ test (120 vs. 135) (http://www.autoadmit.com/thread.php?thread_id=5800512&forum_id=2!#49448488) |
|
Date: November 20th, 2025 10:38 PM Author: https://imgur.com/a/o2g8xYK No one is discriminating at the upper bounds though. A 175 gets treated about the same as a 179. To the extent there's any sorting that needs done it's in the 160s. A lot of genuinely bright people can end up in the high 160s just because they're lazy and didn't prep (http://www.autoadmit.com/thread.php?thread_id=5800512&forum_id=2!#49448559) |
|
Date: November 20th, 2025 10:42 PM Author: ...,,..;...,,..,..,...,,,;.., but doesnt everyone with half a brain already know all this already. i guess i dont see what NCE adds other than more confusion with an additional number that nobody knows what to do with. (http://www.autoadmit.com/thread.php?thread_id=5800512&forum_id=2!#49448575) |
|
Date: November 20th, 2025 10:53 PM Author: metaepistemology is trans oh yeah the stuff i just said about the lsat has nothing to do with NCE and yeah NCE doesn't really add much extra. It is just another way of modeling the data that makes certain aspects easier to see. (http://www.autoadmit.com/thread.php?thread_id=5800512&forum_id=2!#49448615) |
|
Date: November 20th, 2025 10:45 PM Author: ...,,..;...,,..,..,...,,,;.., tyft. anyone who understands percentiles is going to know that at the tails, you get percentile changes that are tiny and ambiguous.
and is anyone who doesnt understand that gonna understand this NCE shit? i guess i dont see why you'd use it and who would use it and how they'd use it. |
|
Date: November 20th, 2025 10:55 PM Author: https://imgur.com/a/o2g8xYK It can only be used to devalue scores and make the test less relevant for admissions purposes. You would never do something like this for a pass/fail like the bar exam (http://www.autoadmit.com/thread.php?thread_id=5800512&forum_id=2!#49448618) |
|
Date: November 20th, 2025 9:27 PM Author: https://imgur.com/a/o2g8xYK It's pointless. Whatever combination of factors makes someone good at the test, it's unlikely that they correspond to each other in standard deviation increments. Like having one standard deviation higher IQ doesn't correspond to one standard deviation of additional test prep and so on. (http://www.autoadmit.com/thread.php?thread_id=5800512&forum_id=2!#49448382) |
|
Date: November 20th, 2025 9:23 PM Author: https://imgur.com/a/o2g8xYK We put a standard in your standard (http://www.autoadmit.com/thread.php?thread_id=5800512&forum_id=2!#49448373) |