Splitting DonorsChoose's equity label at its two cutoffs shows income moved donors most

The design is what makes the question answerable. DonorsChoose grants the label only when a school meets two conditions at once: at least half of students Black, Latino, Native American, Pacific Islander or multiracial, and at least half qualifying for free or reduced-price lunch ^[4]. A school can miss the label by falling short on either one. Sabzehzar, Hong, Burtch and Santanam put a separate regression discontinuity at each threshold, comparing otherwise similar projects at schools just above and just below the line ^[5][2]. Because schools on both sides of one cutoff already satisfy the other condition, that comparison isolates the criterion donors are answering ^[18].

Narrowing the comparison that way also narrows what it covers. Two schools a percentage point apart on the criterion are alike in most other respects, so a gap in fundraising is attributable to the label and not to the school. But the estimate holds only for schools within a narrow band of the cutoff. Projects at schools well above either threshold sit outside the comparison ^[19].

Donors responded most strongly to the income criterion. Projects that got the designation because of the share of students from low-income households had stronger fundraising outcomes ^[6]. "We see a strong effect on the poverty dimension," Sabzehzar said ^[7]. "When we look at the racial minority dimension, the effect becomes much more dependent on who the donors are and the communities they come from," he said ^[8]. The racial criterion is reported as conditional on the donor community ^[8][9]. The phys.org account gives directions of effect, but it does not give magnitudes, sample size or study period ^[20].

That community result compares donor communities as they already exist. Lower-income communities responded more strongly to signals of economic disadvantage, and more racially diverse communities responded more positively to signals involving racial minority students ^[9]. Donors were not assigned to their communities, so this half of the finding is an association between where a donor lives and how the donor reacts to the label ^[21].

Hong stated the test the platform's own purpose implies. "If the platform's goal is to alleviate inequality, then we want to understand whether these kinds of interventions actually work," he said ^[10]. DonorsChoose was set up in part to connect resources with classrooms that need them, and funding gaps still open between projects serving different populations ^[17].

Where a label alone does not meaningfully change giving, Sabzehzar said platforms could try matching donations or seed funding for projects serving disadvantaged populations ^[11]. "If signaling is effective, then use signaling," he said ^[12]. "If signaling isn't effective for a particular group, maybe there are other mechanisms the platform can use" ^[13].

Hong cautioned against extending the findings indiscriminately to other settings, and said a label identifying a classroom serving disadvantaged students operates in a different context from labels used on commercial platforms ^[14]. He and Sabzehzar said future work could test whether responses to equity interventions shift over time ^[16].