Association for Public Policy Analysis & Management

In 2008, the Utah State Legislature passed Senate Bill 41, which provides financial incentives for Utah public schools to open or expand dual language immersion programs. The next eight years saw a rapid proliferation of these programs, from a handful in 2008-09 to 149 programs across 26 of the state's 36 district by 2016-17. As of that year, the program served 28,587 students, or about 4.4% of the 644,000 students in the state.

We capitalize on this rapid scale-up to estimate the plausibly causal effects of dual language program access and enrollment on students' mathematics and reading achievement, as well as differential effects on English learners (ELs) and native English speakers. Our identification strategy relies on time-varying access to dual language immersion across the state during this scale-up period.

The study includes longitudinal enrollment data for 1,292,572 students enrolled in Utah public schools in 2006-07 through 2017-18. It focuses mainly on 646,736 students enrolled in one of seven districts statewide that have captured historical data on student-by-year enrollment in dual language immersion. Of these districts, about 23,200 students have enrolled in such programs, about 9.5% of whom are or have been ELs. In 2016-17, the share of current or former ELs in these districts ranges from 4% to 20%, with the fraction of immersion students who are or have been ELs ranging from 1% to 24%.

In Utah, immersion programs are situated alongside monolingual English programs within neighborhood schools. They typically start in first grade and give enrollment preference to residentially zoned students who opt in. For these reasons, within-school comparisons of immersion to non-immersion students are vulnerable to selection bias. We estimate plausibly causal intent-to-treat effects of immersion access over time by capitalizing on year-to-year changes in the fraction of immersion slots available per first grader in that student's kindergarten school. Our model adjusts for vectors of cohort and school fixed effects, and for students' baseline demographic attributes including race/ethnicity, gender, free/reduced-price lunch eligibility, special education status, migrant and intergenerational poverty status, native language, and EL status at baseline.

One contribution of our paper is that, because our model leverages within-school changes in immersion access, we can distinguish the effects of immersion access from time-invariant school characteristics, including leadership, teacher, and peer effects, insofar as these are consistent over time. Still, causal identification of the intent-to-treat effect depends on the assumption that families do not change local schools before kindergarten in response to changes in immersion availability--an assumption we test by examining changes in observable student demographics as dual language immersion access changes. We then examine differential intent-to-treat effects for ELs and non-ELs by estimating the model separately for students who entered the district as ELs.

Finally, because intent-to-treat effects estimates are averaged across students who do and do not enroll in immersion, we use two-stage least squares to estimate the Local Average Treatment Effect for students whose enrollment in immersion is driven by the change in access to first-grade immersion slots in his or her kindergarten school.