Practice Test 3

Many language revitalization programs begin in schools, where classes are easy to count and tests are easy to administer. The visibility of enrollment numbers and proficiency assessments makes such programs attractive to funders who work with short grant cycles and require measurable outputs. Yet communities that have stabilized or increased the number of fluent speakers consistently report that classroom instruction, though important, is insufficient on its own. Studies comparing cohorts across regions show that gains in vocabulary acquired during after-school programs rarely persist unless students regularly use the language in contexts unrelated to instruction. In places where elders are paired with families in mentorship arrangements and childcare settings default to the heritage language, conversational fluency rises not only among children but also among the adults who interact with them. These initiatives expand the number of hours during which the language is the default medium for basic activities, from buying groceries to scheduling appointments, and thus begin to create domains in which the language is practical rather than merely symbolic. Policymakers, however, often prefer programs that can be launched quickly and reported on comprehensively, such as increasing the number of class sections or setting higher testing benchmarks. The tension is not between schools and communities but between interventions that are easily audited and those that require slower institution building, such as training front-desk staff at clinics or creating vendors who can transact in the language. While there are examples of family pledge programs that produce incremental gains without much institutional backing, even their architects acknowledge high attrition rates when participants cannot find opportunities to use the language beyond the home. The cumulative lesson is that intergenerational transmission stabilizes when children encounter the language in multiple spheres, not just in the classroom, and when adults see practical incentives to participate in those spheres. The author's point is not that schools are unhelpful but that, by themselves, they do not alter the social ecology in which languages either thrive or become vestigial.

City-level evaluations of an employment program have often used a staggered-adoption difference-in-differences design: early adopters serve as treated units while later adopters act as controls until their turn arrives. But two empirical features compromise this approach. First, early adopters exhibited upward trends in employment even before adoption, as documented by administrative payroll data, making them poor counterfactuals for late adopters. Second, early adopters seeded program know-how among neighboring jurisdictions, which then adjusted their own services in anticipation—spillovers that contaminate the supposed controls. These features, combined with heterogeneous capacity across cities, bias naive estimates upward.

The bias is not merely theoretical. Jurisdictions with high administrative capacity adopted earlier and experienced modest gains that began before formal rollouts. Later adopters, often with weaker capacity, did not show pre-adoption gains and struggled to implement at scale. When the naive estimator pools cohorts, it ascribes to "the program" both the momentum of early adopters and the indirect benefits accruing to their neighbors. An alternative cohort-specific synthetic control method that aligns pre-adoption trajectories and downweights contaminated controls yields smaller average effects concentrated in the early, high-capacity cohort; for late-adopting, low-capacity cities, estimated effects are near zero once pretrends and spillovers are addressed.

These findings matter for policy extrapolation. Officials comparing the naive average effect to budgeted expectations risk overpromising in places unlike the early adopters. The revised estimates suggest that what looks like a generalized program success is largely a composition effect: cities most able to implement quickly also looked better beforehand and conferred some of that advantage outward. For cities coming later to the program—precisely those with fewer resources—there is little evidence of meaningful impact without complementary investments that strengthen capacity and reduce dependency on neighbors' spillovers.

The methodological lesson is straightforward: when adoption timing correlates with pre-existing trends and with an ability to influence would-be controls, naive difference-in-differences overstates average treatment effects and is least informative about the municipalities most in need of accurate forecasts.