Panel Paper: Dual-Language Immersion Education at Scale: An Analysis of Program Costs and Mechanisms

Thursday, November 3, 2016 : 11:00 AM
Columbia 6 (Washington Hilton)

*Names in bold indicate Presenter

Jennifer L. Steele1, Robert O. Slater2, Jennifer Li3, Michael Bacon4, Gema Zamarro5 and Trey Miller3, (1)American University, (2)American Councils for International Education, (3)RAND Corporation, (4)Portland Public Schools, (5)University of Arkansas


In a recent longitudinal study that tracked more than 1,600 kindergarteners randomized to dual-language immersion or traditional education in Portland, Oregon, researchers found intent-to-treat effects of immersion equivalent to about seven months of reading skills (tested in English) in fifth grade and nine months of reading skills in eighth grade, as well as no detriment to performance in other subjects and grades. Extending that research, the current paper examines the mechanisms by which these results may have emerged, including the effects of lottery winning on the size of one's homeroom class, the qualifications of one's teacher, the demographic attributes of one's classroom peers, and the demographics and behavior of one's schoolmates.  We also examine whether these effects of lottery-winning on the educational environment differ by one's native language (English or another language) and ethnic background.

Using instrumental variables analysis, we then examine the extent to which these differences mediate estimated treatment effects in reading. We find that winning the lottery does increase the share of one's classmates who are English learners and who are Hispanic, and it reduces the share who qualify for special education services, but it is unrelated to the share who are eligible for subsidized meals. We find no evidence that these differences drive the academic benefits of immersion.

In addition, we use instrumental variables analysis to examine the causal effect of each year of immersion enrollment on student achievement for those whose dosage is driven by their outcome in the immersion lottery. In this analysis, we find that the linear and statistically significant benefit of each additional year in immersion education is roughly 4 percent of a standard deviation in reading. We find little evidence of a statistically significant dosage effect in math or science.

To illuminate operational differences between immersion and non-immersion programs, we use data from school principal and district administrator interviews, as well as public records, to estimate differential operating costs for dual-language immersion programs, including class size, costs of curriculum and professional development, parent outreach, classroom technology, field-based learning, administrator time, and miscellaneous expenditures.

We find few differences in per-pupil expenditures at the school level, with most differences concentrated at central office in terms of curriculum development and professional development. In 2013-14, these represented about 0.1% of the district’s operating budget. However, because district-level expenditures have not been consistent over time, our estimates of cost effectiveness must also account for historic variation in district spending on dual-language immersion over time. At the school level, modest variation in resources such as classroom technology appear to be driven by individual teachers' fund-raising efforts and not by systematic differences in the distribution of district dollars.

This paper aims to unpack plausible mechanisms by which dual-language immersion appears to raise students' reading skills. It is intended to help policymakers who wish to start or scale immersion programs to understand how differences in student experiences and program expenditures may or may not mediate the benefits of immersion education.