I found this paper by Briana Ballis and Katelyn Heath first as a working paper via Dan Willingham, but looking a bit further I discovered that the study has now been published in the American Economic Journal. In their study both researchers used a unique policy change that introduced exogenous variation in SE participation in Texas, allowing them to find evidence for the long-run causal effects of special education. How come? Texas introduced a policy in 2005 which required school districts to have no more than 8.5% of their students enrolled in SE. This policy change led to an immediate drop in SE enrollment, which varied across districts depending on their initial SE rates. This allowed the researchers to analyse the differences.

What did the researchers find?

We find that SE services prepare students with disabilities for long-run success. In the average school district (with initial SE enrollment of 13 percent), fully exposed fifth grade SE cohorts experienced a 3.7 percentage point increase in the likelihood of losing SE four years after fifth grade, a 2.0 percentage point decrease in the likelihood of high school completion, and a 1.2 percentage point decrease in the likelihood of college enrollment. These outcomes are strong predictors of adult success. The magnitude of the estimates is larger among less-advantaged youth and among those attending school in districts with lower wealth and lower average achievement. Our results are robust to a number of specification checks, including student attrition from the sample and differences in trends across the types of districts that would have been closer to or further from compliance with the 8.5 percent threshold prior to implementation.

The authors also checked the effect of removing SE on long-run educational outcomes, the researchers…

…find that SE removal decreases the likelihood a student completes high school by 51.9 percentage points and decreases the likelihood of college enrollment by 37.9 percentage points. Although very large, we view these results as plausible given the potential number of and intensity of supports. SE students receive a variety of services, even within the GE classroom, including (but not limited to) teachers aides or paraprofessionals, small group instruction, additional time on assignments and tests, and specific seating assignments. Again, we find that these results are driven by less-advantaged youth. Our results suggest there are large, meaningful, long-run returns to investing in SE services in the public K-12 school setting for students on the margin of placement, especially those from disadvantaged backgrounds.

Do note that this isn’t a comparison between special education and inclusive education. Also worth noting is that the policy was deemed illegal in 2018 by the federal government.

Abstract of the paper:

Over 13 percent of US students participate in Special Education (SE) programs annually, at a cost of $40 billion. However, the effect of SE placements remains unclear. This paper uses administrative data from Texas to examine the long-run effect of reducing SE access. Our research design exploits variation in SE placement driven by a state policy that required school districts to reduce SE caseloads to 8.5 percent. We show that this policy led to sharp reductions in SE enrollment. These reductions in SE access generated significant reductions in educational attainment, suggesting that marginal participants experience long-run benefits from SE services.