Magnets for discrimination? Affirmative action in Maryland

My little corner of the world was rocked this March when the lists of children admitted into Montgomery County (Maryland) Public Schools’ (MCPS) Gifted and Talented (G&T) programs were posted (G&T programs take children out of their neighborhood schools and put them in schools with additional programs of accelerated learning). At a packed townhall, parent after parent told the story of their child who had straight As, scored in the 99^th percentile on the standardized test, but was still rejected. Many of them were Asians. While researchers still have a hard time finding positive effects of admission into selective programs, for many parents, these rejections were seen as a marker of a flawed and discriminatory process.

It’s a small corner of the world. It’s a small issue. And it’s (probably) not going to dramatically affect the life chances of our kids. But it’s also a window into a knotty problem that countries around the world are facing: How do we ensure that education systems adequately compensate for the multiple deprivations that children from poor families may face? Where do we want to be in that line between rewarding effort and compensating for disadvantage?

A recent article in the New York Times about the admission process into the G&T programs in MCPS fell short of the kind of data-based journalism that I, as an avid Upshot reader, have come to expect. Instead of poring over the data and highlighting how this very local story relates to a host of questions that all of us are struggling with, the New York Times article further forced parents into making a Hobson’s choice: As an Asian parent, you must choose between advocating for your children or caring about societal equity.

This was unfortunate and unnecessary, especially as the intellectual progenitor of the MCPS experiment showed that this is an evitable conflict.

The Florida Experiment

In spring 2005, a “large and diverse school district” in Florida embarked on a novel experiment. State law already allowed for a lower cutoff into the G&T program among disadvantaged children. Nevertheless, severe disparities remained. For instance, only 28 percent of children identified as gifted in Grade 3 were black or Hispanic compared to 60 percent of the student body.

This Florida district then supplemented its system of teacher and parent referral with a system where all children were screened and those above the cutoff for their subgroup were automatically admitted. The idea was ingenious: When admissions are based on applications, parental effort becomes important. Children from low-income backgrounds may be under-represented if their parents do not have the resources to apply, especially as the effort they may require could be higher compared to parents ‘in the know.’ Changing the default status from “opt-in” to “opt-out” could uncover low-income, highly able children who were previously not applying, leading to lower racial and income disparities.

David Card and Laura Giulano’s analysis of this program showed that the universal screening program sharply increased the fraction of the school population admitted into the G&T program from 3.3 to 5.5 percent (Figure 1). There were indeed many children who were scoring above the cutoff for the program but were not applying. The vast increase in intake came from children for whom English was a second language (ESOL) or who were eligible for Free and Reduced Meals (FARMS), collectively called Plan B children. Similarly, there were dramatic increases in the representation of Hispanic and Black children in the program (Figure 1b). Budget cuts in 2008 reversed these gains and ultimately the program returned to status quo by 2011.

Figure 1: Universal screening dramatically increased intake into the G&T program…

Figure 1b. ….with much greater representation among Plan B, Hispanic and Black children

Figures reproduced from Card and Giulano, available here. Children are classified as “Plan B” if English is not their native language (ESOL) or if they are eligible for Free and Reduced Meals (FARMS).

The Montgomery County Program

MCPS has evolved from a predominantly white community to one where Hispanics (30 percent), whites (29 percent), blacks (22 percent), and Asians (14 percent) made up comparable proportions of school enrollment in 2016. Yet, it remains segregated. Black and Hispanic populations concentrate along the Northeastern border of D.C. and spread north towards Silver Spring, Glenmont, and other communities (Figure 2b). Asians and whites are scattered throughout the county, with higher concentrations west of Connecticut Avenue. A background report on choice programs in MCPS notes that in 2013-2014, 45.5 percent of middle school magnet program students were Asians, 7.9 percent were black and 5.6 percent were Hispanic; this compared to their population fractions of 14.8 percent (Asians), 21.4 percent (black), and 26 percent (Hispanic).

Figure 2a: MCPS has evolved from a predominantly white county to a highly diverse community…

Figure 2b. …nevertheless, the county remains highly segregated

Figures taken from a Harvard Study on segregation in the county (the updated figure shown in 2a is available here). The Washington Post’s interactive housing map shows the distribution of houses by ethnicity between 1990 and 2016.

Into this complex and changing schooling system, MCPS introduced a version of the Florida district program for the 2018 G&T programs. But there were critical differences.

• Unlike in Florida, MCPS chose not to increase the budget and number of G&T seats.
• It instituted universal screening, but also voided the use of parental and teacher referrals. In the MCPS version of universal screening, there were two steps—the entire population was tested and then a (large) fraction were invited for a second final test.
• Instead of a clearly defined cutoff, MCPS’s screening process that, inter alia, gave less preference to children with 20 peers performing at a similar level. A child at the 98^th test percentile would be out of luck if there were 20 other children with the same score in the school.

What Happened?

The honest answer is that we don’t know. Citing privacy concerns, the county has been steadfast in its refusal to share data for analysis, denying even watered-down requests. And the story I can piece together on middle school G&T admissions into the magnet schools involved in this change (Eastern and Takoma Park) from publicly available data is unclear. Here is why:

Overall Intake: Total intake increased by 22 children or just under 10 percent (compared to 66 percent in Florida). More Hispanic and black children, as well as those eligible for Free and Reduced Meals were admitted (Figure 3). The fraction of children admitted into the program who were white remained at 40 percent, increased by 3-4 percentage points for black and Hispanic families and declined by 8.5 percentage points for Asians.

This is the second year of losses in Asian representation in these two middle-school magnet programs, from 113 children enrolled in 2016 to 70 in 2018. If the fractions reported in the choice report (which also includes a smaller middle-school that was excluded from this program) can be compared, then there has been a precipitous decline in the fraction of middle school G&T participants who are Asian from 2013.

Figure 3: Between 2017 and 2018, the fraction of Hispanic, black, and FARMS children increased, the fraction of whites remained the same and the fraction Asian declined

Population Screened: The program increased the fraction of Asians who passed the first screening (versus those who had applied in 2017) by 57 percent, Hispanics by 850 percent and blacks by 300 percent. The higher “screen-in” rates, particularly among Hispanic children, is like the Florida experience. But here is the surprise. There was also a 460 percent increase in the number of whites who passed the first screening; only 8 percent had applied in 2017 and 37 percent were screened-in in 2018.

If you find this confusing, that’s because it is. There is no simple “Florida-like” story emerging from these data because the vast increase in white screened-in populations combined with their constant share among the admitted children and the continuing declines among Asians was not seen in Florida.

In fact, these data are compatible with two very different hypotheses.

Hypothesis 1: The program was fair and unbiased. It uncovered highly able black and Hispanic children who were not applying before. The fraction of whites also increased because white households were better at judging whether their children belonged in a G&T program—among all populations, they had (by far) the highest admission rate conditional on applying in 2017, before the program started.

Hypothesis 2: The program was unfair and biased. The program had different implicit cutoffs by ability for different subgroups. In addition, program intake was biased against Asians while ensuring that representation among whites did not change.

What truly happened depends on the relative performance of different subgroups. What we know from publicly available information, including the choice report, is that the performance of Asian and White children is similar (and very high) on average. This lends credibility to Hypothesis 2, but what we really need to know is how the top 5 percent of each group is performing—and here, averages can be highly misleading.

New York Times Reporting: A missed opportunity

At this point two simple questions need answers. First, what were the (implicit) test-score cutoffs used for children from different backgrounds? Second, if you apply the peer availability rule, is there any evidence that any group was favored or disfavored in the process?

A top-notch article would have explained why parents were concerned and undertaken the analysis to understand what truly happened. But the New York Times article instead focused on the overall numbers (which, as we saw, are not easy to rationalize) and pitched the change as a battle between affirmative action and Asian (and to some extent, white) parenting.

That is a mistake.

If there was one thing to learn from the Florida and the MCPS experiments, it was this. Think of the schooling system as having three different objectives: Compensating for disadvantages, equal rewards for equal ability, and staying within the budget. The Florida experiment showed that you can compensate and provide equal rewards as long as you are also willing to increase the budget (surprisingly, the marginal cost of additional G&T enrollment may be quite small). MCPS went the other way and kept the budget (more or less) the same. The outcome was predictable—almost certainly equal rewards for equal ability went out the window and the gains for disadvantaged children were much smaller than in Florida. In addition, the process opened a can of worms about potential discrimination against Asians in the county.

This is unfortunate because even in this highly progressive county, the discussion is becoming increasingly parochial. By choosing to debate the change within a straitjacket of artificial scarcity, ethnic concerns are coming to the fore. I don’t know what the right solution is. But I do know that the correct process involves a clear and transparent analysis of the program and public discussion about what to do next.