Profile picture
Prof Dynarski @dynarski
, 36 tweets, 6 min read Read on Twitter
Does Harvard discriminate against Asians in admissions?

Evidence says no, by my reading of the evidence in this case. /THREAD…
Key charge of the plaintiffs: Asian applicants have high test scores & grades, so should represesent more of the Harvard class than they do 2/N
But grades/scores are NOT determinative of Harvard admission. Harvard could fill a class many times over with perfect grades & scores. 3/N
Like all the most selective schools, Harvard weighs extracurriculars, family background, geography, & a bazillion other factors in order to pick out five admits from 100 applicants.
Once you consider all these factors, there is no evidence of bias against Asians in the process.

David Card has does the statistical analysis on this for the defense and, unsurprisingly, it is airtight. 5/N
My personal nightmare: David Card presenting the opposing statistical evidence for a legal case in which I am engaged.

He is the top applied micro-economist in the world, IMO. 6/N
Card models the Harvard admissions process using all the factors the school uses in making decisions. Essay ratings, background, yada yada.

The plaintiffs also do stats modeling but leave out key variables in the process.

The Harvard brief is here…

The plaintiffs’ brief is here…

All of this is a case study in social science making a difference in the world. 9/N
David Card’s analysis of Harvard admissions is here…

Card’s rebuttal of the plaintiffs’ data analysis is here…

The plaintiffs’ analysis of Harvard admissions is here


Plaintiffs’ rebuttal of Card’s analysis is here


The social scientists on both sides of this case are kick-ass researchers whom I deeply respect.

David Card and Peter Arcidiciano are both gems with stellar research records

This court case is linked, by the way, to the effort to pack Harvard's alumni board with anti-affirmative-action candidates. I weighed in on this.

Nobody is paying me for my opinion in this case, BTW

I avoid being a paid expert witness so that I retain complete freedom to shoot my mouth off

Plaintiffs have also now released an old analysis by Harvard's office of internal research. 15/N
The internal analysis is an extremely rough cut & lacks relevant variables.

Unsurprisingly, it's not as nuanced as analyses by Arcidiciano and Card. 16/N
There are a lot of modeling choices to make here, as indicated by the exchange between the social scientists in this case. The internal analysis goes nowhere near this level.

I see the release of this internal Harvard analysis not as new statistical evidence but rather as a PR move by plaintiffs.

Personal essays & teachers’ letters of rec play an impt role in elite admissions. I don’t love this. Other countries do it differently by, for example, having far more thorough tests than our SAT/ACT (see England) 19/N
How do we acct for this stuff in a stats analysis? Harvard admissions creates a summary metric, which has the unfortunate title of “personality score.” 20/N
Plaintiffs leave out this score, saying it’s biased (Card shows it is not)

By leaving it out they toss a lot that goes into admissions. Nuance about lives and the challenges kids face. Kid lost parent at young age, sibling is incarcerated. 21/N
Yes how you weigh this stuff is subjective but please suggest a better approach that incorporates it.

Plaintiffs complain that they were not provided essays and letters of rec. Yo, maybe the applicants have some right to privacy here?
Is Harvard systematically biased against Asians in its admissions ratings?

Asians score LOWER than Whites on personal rating! 😡

Asians score HIGHER than Whites on academic & extracurricular ratings 🤔

Annoyed that first half of article is uncritical recitation of the plaintiff's case

The previous 23 tweets in this thread provide that critical context.

A lot is being made of Asians scoring“low” on the “personality” metric.

Going to explain the methodology here (& reiterate my hatred for the metric’s moniker)

Admissions boils the hundreds of application elements into a few summary ratings on academics, extracurriculars, “personality.” 🙄 26/N
To check for bias against Asians, the expert witnesses use regressions to try to predict the summary ratings using the underlying items that go into them.

They use identical models but interpret the results differently.

The regressions answer the questions:

“How much of the summary ratings is explained by the variables we see (=“objective”)?


How much of the summary ratings is NOT explained by the variables we see (=“subjective,” perhaps a channel for discrimination).

The “Asians get low personal ratings” comes from the ACTUAL personal rating being lower than the PREDICTED rating for Asians.

(Nerds: a dummy var for Asian has a negative coeff)

Plaintiff’s interpretation: bias against Asians

BUT on two other ratings (extracurriculars & academics) the ACTUAL rating is higher than the PREDICTED rating for Asians

A consistent reading of the results would say that THIS is evidence of bias IN FAVOR OF Asians.

Instead of this consistent interpretation, plaintiffs say:

Unexplained, LOW ratings for Asians indicates discrimination against them while unexplained, HIGH ratings for Asians indicates Asians must be better in ways we can’t see.


Another point: both expert witnesses concur there is evidence of bias IN FAVOR OF Asians who are female and/or from CA.

So! Harvard is apparently engaged in discrimination ONLY against the subpopulation defined as male Asians who are not from California or legacies or athletes or children of Harvard faculty/staff.

More depth on that “personality” measure. Great thread and follow.


This table from Card summarizes how various modeling decisions affect the estimates. 35/N
Instead of hypothesizing about the effect of various choices, we can quantity it

Missing some Tweet in this thread?
You can try to force a refresh.

Like this thread? Get email updates or save it to PDF!

Subscribe to Prof Dynarski
Profile picture

Get real-time email alerts when new unrolls are available from this author!

This content may be removed anytime!

Twitter may remove this content at anytime, convert it as a PDF, save and print for later use!

Try unrolling a thread yourself!

how to unroll video

1) Follow Thread Reader App on Twitter so you can easily mention us!

2) Go to a Twitter thread (series of Tweets by the same owner) and mention us with a keyword "unroll" @threadreaderapp unroll

You can practice here first or read more on our help page!

Did Thread Reader help you today?

Support us! We are indie developers!

This site is made by just three indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member and get exclusive features!

Premium member ($3.00/month or $30.00/year)

Too expensive? Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal Become our Patreon

Thank you for your support!