Stanford, Harvard, Dartmouth, Yale, Penn, Brown, CalTech, JHU, and UT-Austin to Require Standardized Testing for Admissions

Applying to College
2020

I get it, and that’s your prerogative. I’m not sure what the context is that makes you care, nor do you have to explain that. Although we are talking undergrad here, this exchange makes me think about this med school quote: “What does one call the person who graduated last in their med school class?” Answer: Doctor. Many of us have no idea where in their graduating class our physicians graduated.

I am not being dismissive, and apologies if that’s how what I said came across.

I am one of the few people on this thread who support the testing requirements that each school thinks is best for it, while also sometimes suggesting a different way to think about the data that is shared. Again, when I look at that chart I posted above, I see students with a 1200 getting a 3.2 GPA at Ivy Plus schools in their first year, and with my admissions hat on that’s positive. That’s the takeaway for me, and demonstrates why some highly rejective schools don’t put a lot of weight on test scores.

3 Likes
2021

The most difficult part about Harvard is getting in, not getting out.

At Harvard, the median grade is now an A-, and most of the classes aren’t particularly hard. I think the same is true at Yale as well. Engineering classes certainly aren’t harder here than at most state flagships. Now, there are some classes at Harvard that are known to be extremely difficult, but they are all optional. And interestingly, they have more signaling value to certain employers than the overall GPA.

When you have a median grade of A-, getting a 3.2 either shows you couldn’t handle the work at all, or that you didn’t care. Neither good attributes for an engineer.

3 Likes
2022

The study with the previously posted graph says the mean first year GPA was 3.49. The median grade was A-. 1/4 of students received a C grade during freshman year. For predicting who receives C grades:

SAT + GPA – Explains 11% of variance in who has C grades
Above + Demographics – Explains 14% of variance in who has C grades
Above + HS Name – Explains 63% of variance in who has C grades

2023

Sorry, which study are you referring to? I looked back through your posts, and you referred to a Duke study, but that was from 2011, which is extremely old at this point. Here is a more recent (20230 Harvard Crimson article that says that 79% of grades at Harvard were in the A-range.

In an attempt to bring this back to the subject of testing, the reason testing helps is because grade inflation means grade compression, which creates an inability to distinguish between good and great, and rewards perfectionism over exceptional work.

2 Likes
2024

That suggests that a B+ at Harvard means that the student is below median at Harvard if the median grade is A-.

But does that mean that a student who gets a B+ in a Harvard course would get a D or F (“couldn’t handle the work at all”) in a similar course at some other university? Or would you expect the student to get a B+ in a similar course at some other university?

2025

Most of the years of the study were pre-COVID. I expect the extremely few matriculating students getting admitted with submitted 1200 SAT are primarily recruited athletes. Under Ivy recruiting rules, a small portion of athletes may have academic stats that are 2 standard deviations below the general student body. Using Harvard as an example, academic rating 4 and 5 are defined as follows (pre test optional definition):

4 – Respectable grades and low-to mid-600 scores on SAT and
subject tests or 26 to 29 ACT.
5 – Modest grades and 500 scores on SAT and subject tests (25 and below ACT).

The distribution of matriculating students with these stats in the lawsuit analysis was:

4 Academic – 1-2% of admits, 91% are athletes
5 Academic – ~1 admit per year, 100% athletes

2 Likes
2026

I’m referring to the study that we have been discussing in recent posts, with the graph showing higher average first year GPA at Ivy+ colleges when SAT score increases, that shows 3.2 GPA at 1200 SAT. Zrt42 first referenced it. That sample group had a mean first year GPA of 3.49, median GPA = A-, and 25% receiving at least one C grade (in freshman year)..

2 Likes
2027

It seems an apt time to repeat my nephew’s quote from his experience at Yale, several years ago when grade inflation wasn’t as rampant as it is now: “It can be hard to get an A, but it’s much harder to get a C”.

When 79% of all grades at Harvard now in the A range, and getting a C pretty much impossible, it’s hard to know just how bad a student at the bottom of the distribution is. In contrast, at a state flagship, you do.

2 Likes
2028

I’m not saying there’s not pervasive grade inflation at the HS level. Yet, plenty of enrollment management peeps at highly rejective schools (with support of school leadership and trustees) have a different opinion than you do…they don’t necessarily see a 2-3 hour test all that helpful in understanding which applicants they want to offer admission to. I’m not saying these people would say there’s no value in testing either, but to get back to something we talked about much earlier in this thread, some believe the juice is not worth the squeeze.

1 Like
2029

I am not for mandatory testing either, primarily because some fraction of capable students don’t do well on standardized tests. It’s great that those students have options.

That said, I don’t believe that the goals of the those that are against testing are always as altruistic as they try to appear. When the politically appointed UC regents overturned the recommendations of the Academic Senate, they were pandering to politics, not responding to evidence. And the evidence showed that the SAT was a useful predictor of performance at the UC level (I recall that Data10 showed that we could use enough other admission factors to substitute for it, but that’s also a different admissions process).

Another reason many so colleges are test optional is that it boosts applications, something pretty important given the demographic cliff colleges are facing. To many colleges, it’s a matter of survival.

A third reason is that going test optional allows admissions officers more leeway in crafting a class to create the demographic mix they want. We know that poor students tend to have lower test scores than wealthier students. If a college wants to admit more poor students, it’s easier to do so without requiring their test scores.

6 Likes
2030

Totally agree. Your UC example is a good one, as well as the political reasons behind the Tenn and GA state systems/board of governors mandating test requirements for the public colleges in those states (all in Tenn, 7 in GA.)

I agree test optional or blind generally lead to more apps (and a lower admit rate), but it’s not clear (nor have I seen data) that when a college receives more apps they get more enrollments, or a higher yield, or a ‘stronger’ class

Yes, will be interesting to see what the clsss composition is at the schools that went back to requiring tests this year. I predict we will see more questbridge and stars admits (and likely new partnerships/admits from other CBOs…but many schools don’t report all those details.)

2 Likes
2031

First, those are FIRST YEAR GPA, not “college GPA”. Second, those graphs are wrong, since the are giving averages without error bars. The actual data points aren’t in that nice line, they are likely a huge cloud of data points, with a fairly large number of them not falling along that line.

So that graph isn’t actually showing what the title says it’s showing, and second, it is creating an illusion of a much tighter and much stronger correlation than there actually is.

I’m sorry but that graph seems purposefully deceptive. If I reviewed an article with a graph like that, I would recommend that the article be rejected.

This is not worthy of that website.

4 Likes
2032

You seem to be confused. This will be a great help to you:

2 Likes
2033

That sounds like grad school grading:

A: Good work
B: Meh – like a C in undergrad
C: Fail

1 Like
2034

That article is just an extended version. It is better in that the title is not making claims, and they at least have multiple caveats in the conclusions that the shortened version does not.

Well, one of the flaws in the title remains: they wrote “Academic performance”, while it’s actually “academic performance in the first year”.

However, the methodological flaws are still there.

Those are not actually data points, they are derivatives of groups of data points that have been normalized based in multiple factors. To claim that these are graphs of correlations between SAT and first year college GPA is misstating the actual facts.

Their data points are averages of highly skewed datasets which are, in turn, made up of different factors. The datasets are also not equally skewed. They use non-parametric statistics, yet present their graphs and results as though they were using parametric methods.

In short, there is lot of data manipulation and, added to the methods that they use, means that the statistical power of this model is very weak.

They claim that their graphs shows that

Yet they also wrote that these aren’t actually the HSGPA values or first year college GPA values of any particular students, or even the simple averages or medians of these values, but the result of normalizing, adding the effects of their definition of “academic struggle”, etc, etc, etc.

So this sentence does not actually describe what the graph is showing.

An additional problem is that even if this were an accurate representation of what is actually happening in these eight colleges, it does not prove that higher SAT scores demonstrate higher preparedness for college, or even for first year of college.

These results would still be there if SAT just indicators of student family income.

Higher income students are likely to do better in their first year of college than lower income students, because they have more support. Living away from home (these are all residential colleges) is also likely to be more difficult for low income students. Low income students are more likely to have spent a lot less time away from family. No summer camps, no long trips abroad, no times at home when one parent are both are travelling.

I am also pretty certain that students who attended boarding schools for high school will do better in their first year at residential colleges than students who attended schools with the same academic levels but lived at home. Not because they are better prepared for the classes, but are better prepared for studying while not living at home.

My point isn’t that there isn’t a correlation between SAT scores and academic preparedness. It’s that the correlation is a lot weaker than this article claims, and that their methodology is not good.

BTW, I don’t think that the authors’ agenda is keeping URMs out of elite colleges. Quite the opposite. Since their conclusions are that more needs to be done to make sure that low income students are more prepared I just think that they started with the assumption that SAT scores are a good indication of preparedness and moved backwards from there.

Again - the SATs/ACTs are useful, but their accuracy and usefulness is overstated here and elsewhere.

That being said, I think that the authors have an underlying message, with which I agree, which is that the problem is in the high schools that are serving low income kids are not preparing them well for college.

A - Average
B - Below Average
C - Confused, as in too confused to drop the class.

5 Likes
2035

Grade distributions vary quite a bit by college, but the general pattern at highly selective private colleges is the more selective the college, the higher the grade distribution. This pattern makes sense if you think of grades as a measure of % of students who master the material, rather than a means of distinguishing between students. More selective colleges are likely to have a larger portion of students who master the material than less selective colleges.

However, the rate of grade inflation is higher than I’d expect if it was primarily a function of student quality. For example, the table below shows Yale grade distribution over time. The big jump in grade distribution with a large portion getting A’s occurred just after COVID + test optional. This doesn’t fit well with some of the earlier posts alluding to a significant portion of test optional kids failing classes.

I suspect the initial jump in grade distribution at colleges following COVID is primarily due to professors being more lenient with grades, with both students and professors struggling with COVID + remote learning. Less obvious is why effects of the bulk of this grade boost still seem to be present in 2022-23. % A/A- was 73% pre-COVID, 82% in first COVID year, and 79% in 2022-23.

Changes in Yale Grade Distribution Over Time

image
image522×386 25.5 KB

Public colleges often go into more detail about grade distribution than private colleges like Yale. For example, stats for GeorgiaTech are below. There seems to be less of a post-COVID distribution boost at GT and many other publics than at Yale and in general Ivy+ colleges.

Changes in GeorgiaTech Grade Distribution Over Time
2019 – 57% A, 18% B, 5% C, 2% Below C, 7% W
2020 – 59% A, 16% B, 4% C, 2% Below C, 6% W
2024 – 59% A, 16% B, 5% C, 2% Below C, 6% W

There was also a huge variation in grade distribution for different classes, and in many cases different professors teaching the same class. Some examples are below for GT 2024-25:

Math 1111 (Algebra) --27% A, 25% B, 20% C, 28% D/F/U, 0% W
Math 1551 (Calculus) – 28% A, 40% B, 20% C, 9% D/F/U, 3% W
Math 1553 (Linear Algebra) – 41% A, 35% B, 16% C, 6% D/F/U, 2% W
Math 2552 (Diff. Equations) – 63% A, 22% B, 7% C, 4% D/F/U, 3% W

2036

This review was well discussed on the CC website when it was first published 1-2 years ago. There is some useful underlying information, but it’s difficult to decipher that useful information due to the results being presented in a biased way, likely to improve public optics in relation to the decision to switch to test required. Examples of this biased presentation include:

  1. Only Listing Relationship with First Year GPA – Having highest possible freshmen GPA is not how typical Ivy+ colleges review success of their students. Other metrics like graduation rate, cumulative GPA, major completion/switching, and post graduate outcomes likely have a greater contribution to success. The analysis sample began in 2017, so these measures are available.
  2. Ignoring That Ivy+ Admission Is Not Rack and Stack Based on SAT and HS GPA in Isolation – The critical decision for test optional vs test required is not whether SAT or HS GPA in isolation is more correlated with First Year College GPA. It is instead how much SAT score adds to the prediction beyond the measures used to admit students when scores are not available – how much accuracy is lost from the prediction when test scores are not available and students are instead admitted based on a combination of transcript, rigor of courses, LORs, ECs/awards, essays, … The analysis doesn’t review this.
  3. Looking at HS GPA in Isolation – The analysis only reviewed HS GPA in isolation. For example, a 3.8 HS GPA student taking highly rigorous classes at a HS with harsh grading such that 3.8 is towards top of class is not the same as a 3.8 HS GPA student taking easy classes at HS with lenient grading where 3.8 GPA is below median. This isn’t how Ivy+ colleges use GPA in admission decisions. They instead review the full transcript, including considering grades in the context of the particular HS.
  4. Ignoring that SAT Score Only Explained a Small Minority of Variance in First Year GPA – The technical appendix shows that SAT score explained 19% of variance in first year GPA and 10% of variance in which students received C grades. However, the main sections don’t mention the this and instead imply a stronger degree of influence.
  5. Ignoring That Including HS Effects Explain Majority of Variance in First Year GPA – The technical appendix also shows that when HS name is also considered, score + HS GPA changes from explaining a small minority of variance in first year GPA to explaining the majority of variance. If you have a measure in your review that explains the majority of variance in your target outcome, it’s worthwhile to review why. For example, the review found that when HS name is also considered, HS GPA becomes more influential in predicting college outcome. That is when you consider the HS GPA in the context of the particular HS (whether HS has harsh/lenient grading, and whether most students take more/less rigorous courses), then HS GPA becomes more influential than when reviewed in isolation, which is not discussed in the main article. How much more influential in relation to SAT is unclear since the analysis doesn’t review this.
  6. Generally Presenting Results in a Misleading Way – For example, in an earlier post I noted that Ivy+ students receiving 1200 SAT were primarily recruited athletes. There are many reasons why recruited athletes might average lower first year grades than just their SAT score. They might be generally weaker admits who are weaker across the full application, including SAT score. More useful would be a regression analysis that controls for reader scores on the rest of application such that you are reviewing the effect of SAT among students who had similar application ratings. With such an analysis you can review relative influence of scores vs other parts of application that are correlated with scores.
  7. Graph – The graph itself shows averages, without information about variance in that average, giving the impression of a stronger correlation than the 10-19% of variance explained I noted above. Looking at the graph, a casual reader might think there is near certain causality – if you have x SAT, you are likely to get y first year GPA.

That said, there is some useful underlying information. With increased grade inflation and COVID learning effects (students struggling with remote learning combined with many HSs being more lenient with grades or switching to pass/fail), test scores likely are more relevant than in past decades. These effects contribute to the analysis result. It would be great if more colleges reviewed and published results publicly.

For example, I’d like to see things like how did graduation rate, cumulative GPA at graduation, and major distribution/switching differ between students who submitted scores and students who were admitted without scores. I’d like to see regression analyses that reviewed the relative influence of different admission factors on these outcomes, as well as reviewed what % of variance is explained for test submitted admission factors vs what % of variance is explained for test optional applicants admission factors.

2 Likes
2037

If by error bars you meant confidence intervals, they are best practice if they are showing an estimated mean using a sample. But they are not required (and in fact incorrect) if they are showing an actual mean of the entire population.

Simplifying for others, if I chose 100 high schools at random throughout the country and measured every senior’s height, I can calculate the exact mean height for senior boys and girls in those schools. This is a population mean calculation, and no confidence interval bars would be shown. But if I instead use the data from these 100 schools to to estimate the height of seniors throughout the country, then confidence interval bars are required.

If you instead meant “error bars” to represent the spread around each point, I would use something like a box and whisker plot rather than confidence interval bars. But that would results in a cluttered graph with a similar upward sloping pattern. That’s really only useful with a far smaller number of bins.

Sure, but that’s obvious isn’t it? I mean, nobody really thinks that everyone with a 1500SAT is automatically going to get a 3.5-3.6 GPA.

What is misleading though is showing both a best fit line AND binned mean values, as that suggests a much higher correlation than the underlying data suggests. The actual underlying correlation is probably close to 0.3, whereas the best fit line with the binned values suggests a correlation very close to 1.0.

1 Like
2038

All models are flawed. Some are quite useful.

Not necessarily. The right data manipulation can lead to significant insights, and quite often that data manipulation is not simple. The entire field of quant finance is built on the fact that they can find profitable insights are too difficult for others to find, because of the amount of statistics and data manipulation involved.

It’s much more likely that you have jumped to a conclusion and are looking for reasons to justify it.

3 Likes
2039

Yes, but there is a difference in creating a model for explaining who succeeded, vs. what’s reasonable and ethical in admission criteria to predict who will succeed.

There are a lot of things that would be useful for predicting how a student will perform but would be inappropriate in admission criteria. For example, if you wanted students that are likely to succeed you couldn’t do much better than to choose the top students from highly ranked prep schools. Other inappropriate approaches would include filtering on the parent’s occupations, race, income level, education level, or the zip code in which they lived.

In other words, predicting who will succeed without resorting to inappropriate measures will always be messy.