Thread by @jmlarkin

I'm finally set up to livetweet Day 13 of #HarvardTrial. Harvard's expert, the @UCBerkeley labor economist David Card, is on the stand, attempting to firm up his finding that there is no statistical evidence of discrimination against Asian-Americans.

Another economist, Susan @Dynarski, who has contributed to a brief in support of Harvard and looked closely at this case, has said that Card is the exact last expert she'd want to to go up against — see below:

https://twitter.com/dynarski/status/1007629578905489409

Card is looking over a chart of some of @Fair_Admissions's models, in which Asian-Americans see measurably higher academic and extracurricular ratings in Harvard admissions officers, but lower personal ratings. He calls it a "strange kind of schizophrenia."

Card clears his throat, and says, "I'm sorry—I'm not used to talking this long."

Today, as yesterday, @Harvard is relying on Card to provide a statistical sense of the hidden mechanics of the admissions office's approach to its applicants.

They've spent most of the past hour arguing that @Harvard's personal ratings are *not* determined by academic factors.

"My assessment is that white students are stronger than Asian-American students on the observable factors" that inform those personal ratings, Card says.

AN INTERPOLATION FROM ME: Most journalists should get robust statistical training. In case, like, Phil Knight is reading this, that would be a good grant program to start/expand recklessly.

Harvard has used "demonstratives" — glorified PowerPoint presentations — throughout the trial. Here's how they summarized Dr. Card's four-part argument today in the latest demonstrative:

"In light of everything that you've seen and everything that you've done in this case, is the personal rating an instrument of bias against Asian-Americans?," Card is asked. "No... I don't think that makes any sense at all," he replies.

Card is also arguing that Arcidiacono's selective exclusions — especially of ALDC applicants (athletes, children of alumni, etc.) — skewed his results. You *can* produce an apparent systemic bias against Asian-Americans by those exclusions, but it's not clear what it tells you.

Even Judge Burroughs seems slightly overwhelmed by the statistical refinement of what Dr. Card is presenting today. (They should have sent a @NateSilver538.)

We are likely to learn a lot more during the cross-examination of Card — unclear when that will begin.

Card's pre-lunch testimony will wrap with him saying that there's no statistical evidence of discrimination against Asian-Americans — that numbers bounce around from year to year: "It's exactly the kind of pattern of findings that one would expect" if there were no correlation.

Now they are running through the questions Card addressed in his testimony. As to the effect of race in @Harvard admissions, Card says it explains less than other factors: the r-squared values for student scores, teacher ratings + alumni interviews explain much more, for example.

"Race per se is a very, very small component of explanatory power," Card says. But it's still important: "Any individual factor can be important, but only for students who are highly competitive" already. So the 'tip' system is working as intended, Card's arguing.

Card is now talking over a kind of hockey-stick chart — "an s-curve relationship" — showing that the weakest 75% of applicants to Harvard are effectively "out of the money." They aren't getting in, in short.

But then for the top quartile of students, they're all "on the bubble." Take a student from the 90th percentile, for example, and favorable factors like race, geography or alumni parents can have a "very large effect on the probability of admission," Card says.

For instance, for most African-American applicants, the "marginal effect" of *being* African-American on their chance of getting into @Harvard is negligible. But for the strongest 20 percent, their race can boost their chance by as much as 50 percentage points.

Card shows the effect of being 'Hispanic or other' has much the same feature, though the boost is smaller.

Judge Burroughs asked Card if he looked at the boost from being Asian-American. "No, Your Honor... there's no 'tip' for Asians in my model," Card replies.

Now Card is showing a matrix of factors that can have a "large effect" for competitive applicants. The biggest admissions boosts are for high extracurricular ratings or academic ratings, with a smaller advantage going to "lineage applicants," or children of alumni.

"Let's consider a lineage applicant" in the eighth-highest admissions decile, Harvard's attorney asks. How much does having an alumni parent help him? For any one such student, Card says, admissions is still not a sure thing.

This is @Harvard's "tip" system laid out in clear detail: students in the second-strongest overall decile (80th to 89th) get the biggest benefit from being, for example, black, or a 'legacy admit', or a stellar student.

Card is summarizing — two points on race in Harvard admissions. First, he says, "if one looks just at race as a variable, it has a very, very small effect — almost negligible on the overall chance of admissions."

Then second, if you look at applicants who are already 'on the bubble,' being African-American or being Hispanic *can* be one more factor that puts their applications over the top.

Now, question 3: "Is there statistical evidence that Harvard has engaged in racial balancing?" (That would mean the unlawful fixing of soft racial quotes, a big part of @Fair_Admissions's case.)

Card says, "in my view, there's no evidence that that's going on."

He shows a chart showing apparently wild year-to-year changes in the racial breakdown of @Harvard admitted students. For example, the Asian-American share dropped 11 percent in 2005, then rose 17 percent the following year.

Card characterizes these as "fairly large changes... in the share of each admitted race group."

If Harvard *were* racially balancing their admitted classes, Card says (to laughs), "They're not doing a very good job."

(A similar variability manifests in the racial shares of *matriculating* students, meaning those students who actually enroll at @Harvard.)

Now we're seeing a chart showing that the Asian-American share of applications to @Harvard has climbed from below five percent in 1980 to north of 20 percent last year.

A second chart shows that Asian-American matriculation has climbed by about that much. It's the opposite of what Card would says you would expect to see if Harvard were trying to suppress Asian-American enrollment.

Abbreviated lunch break... over. I saw former Bulger attorney Jay W. Carney, wearing a bemused look, just outside the Moakley Courthouse. Odd to think Carney was here defending Bulger just five years ago.

Now that we're back in the courtroom, with Card kicking the tires on @RickKahlenberg's proposed "race-neutral" approach to admissions at Harvard. Card found that without racial consideration, the white share of the admitted class would jump from 40 to 48 percent.

If you've been following along for the past two weeks, you'll recall that Kahlenberg argued that by giving a sufficiently large "tip" to disadvantaged students, Harvard could create a racially diverse class as well as an economically diverse one.

So rather than simply removing race, Kahlenberg proposed giving a compensatory advantage to low-income students, who are disproportionately racially diverse.

Card is going to analyze those proposals of Kahlenberg's point-by-point. But because Harvard is still examining, we won't hear pushback for a while.

Proposal one: "eliminating policies that allegedly benefit white applicants." Card found that removing preferences based on race, athletics, legacy admissions, "dean's list" + children of staff results in a class that is still whiter and more Asian-American than the actual class.

You can find some of the charts we're seeing in court, like a version of the one below, in Card's report for @Harvard: projects.iq.harvard.edu/files/diverse-…

Most of the scenarios being discussed are based on statistical modeling. But when it comes to 'early action' — the process of applying early and solely to Harvard — there's empirical evidence: Harvard abolished early action from 2012 to 2015.

During that period, the matriculation rate — the percentage of accepted black applicants who enrolled — dropped, from 70 percent to 65 percent. Card and Harvard's attorneys are suggesting that ending early action has already shown a homogenizing effect on @Harvard enrollment.

Now Card is working through Kahlenberg's second proposal: placing a greater weight on the advantage given to students of low socioeconomic status (low SES).

Consider a student who has one or more of the following characteristics:
— She's considered by @Harvard to be "disadvantaged";
— She's a first-generation college student;
— She applied for application fee waiver;
— And she's coming from neighborhood family income of >$65k.

Under Card's so-called "1x" boost, the student who has *all* of the prior characteristics would see her chance of admissions to @Harvard rise from 7 percent to 36 percent.

But Card's research found that that "1x boost" would be too small to create a class that's as diverse as Harvard is today. (You can see that in the chart below.) You need to go up to a "4x boost" to recreate that level of racial diversity.

But that level of "boost" would mean that the above-mentioned student, with all of those four characteristics of "low SES," would be all but guaranteed admission to @Harvard based on her socioeconomic background alone, Card's research suggests.

Further, Card is arguing, creating a robust-enough advantage for low-SES students would result in a drop in the quality of applicants: 13 percentage points fewer top-tier students, 11 p.p.s fewer applicants with top "personal" ratings, etc.

This has been Harvard's argument all along. They have many desiderata for their student body — they want it to be academically excellent, racially diverse, devoted to the institution, selectively athletic, etc. And they say without considering race, they can't accomplish that.

Kahlenberg also proposed abandoning standardized tests, which (research suggests) tend to disadvantage most students of color. Doing that along with all the other neutralizing changes still means you have to give a v. large boost on the basis of income to achieve race diversity.

Finally, Kahlenberg proposed increasing @Harvard's transfer application process as a way of building diversity. Here was his reasoning, from his report: …2i553a2t1s05i-wpengine.netdna-ssl.com/wp-content/upl…

But Card is saying currently, the body of transfer applicants to Harvard skew whiter than the non-transfer applicants. Still seems as though, by means of targeting/recruiting, Harvard *could* develop an Amherst-style commitment to admitting community college students.

Some context here may be useful — Harvard says it has stepped up its efforts at recruitment, but according to federal data (which Harvard disputes), the share of its seats that goes to African-Americans is about the same as it was 36 years ago: around 7 percent. Chart below.

(Note that the presence of international students and Asian American students on @Harvard's campus has grown by quite a bit during that same period.)

Card is running through how the expansion of Harvard's HFAI program (tuition-free for low-income families) led to more racial diversity in the applicant pool, including a rise from 20 to 27 percentage points in the share of applicants who were black, Hispanic or "other."

But those effects turned up in the four or five years after the first implementation of the HFAI program, but further tweaks didn't show a further diversifying effect.

Finally, Card is taking issue with the "place-based" admissions policies of the kinds advocated by Kahlenberg and Sheryl Cashin. There are 33,144 ZIP codes and 41,368 high schools in America, Card says — drawing from those would "swamp the [1,719] available slots at Harvard."

@fernandesglobe is sitting right behind me, but *also* has reported well on just this phenomenon! www0.bostonglobe.com/metro/2018/06/…

Drawing students based on place — namely, pulling the top students from every high school — would result, Card found, in a "notable decline" of 34 percentage points or more in the quality of applicants.

Card says it's an "unfortunate" feature of American public education that many, many high schools are simply disadvantaged, and their graduates can't really measure up to Harvard's standards.

Now we're getting the wrap-up from David Card. Namely,
1) No stat. evidence that Harvard discriminated against Asian-Americans.
2) Race affects admissions mostly for highly-qualified students.
3) No stat. evidence of racial balancing.
4) Race-neutral alts. have problems.

We are taking a ten-minute break before Card gets cross-examined.

And we're off with a bang! SFFA lead attorney Adam Mortara is questioning David Card about the Bakke decision (1978), which applauded Harvard for its considered, "non-predominant" use of race in admissions.

Card replies that he is "in no way an expert on the law." (He's a labor economist by trade.) Mortara is going back to Card's finding that race is not close to being the top explanatory factor in admissions decisions at Harvard.

"I want to talk about some — I'm not gonna say 'mistakes,' but some 'inaccuracies' in your slides," Mortara says. He says Card labeled a chart "r-squared" when he meant "pseudo-r-squared" in a key slide. Card says he had warned the court he'd use them interchangeably.

(This is perhaps as adversarial as a discussion of statistical methods could possibly be.)

After a long excursus from Card, Mortara says, "You done?"
Harvard attorney jumps up and says questions are one thing, "snide remarks are another."
Mortara says next time he gets three paragraphs after a yes, he'll move to strike.

Judge Burroughs steps in and says, "My experience in this trial is the experts often answer questions with three paragraphs after one"—that SFFA's expert, Peter Arcidiacono, did the same, she said. Mortara says, "due respect," he didn't. We're back on track.

That Harvard attorney was Seth Waxman, BTW. Lotsa, lotsa lawyers. Meet the team: thecrimson.com/article/2018/1…

We're breaking. The judge says she's tired; Dr. Card is tired, she assumes. We'll resume tomorrow at 9:30am.

.@WBUR's office at the courthouse is full of Whitey-ana from @davidboeri's tenure of the case, including this picture of a man and his... goat. Weird.