Background/Context: As part of a national movement to eliminate a global tracking system that disproportionately assigned underserved students to low-level math classes, the Chicago Public Schools (CPS) mandated academic algebra for all students entering high school in 1997. The results were disappointing: low-skill students experienced elevated failure rates while high-skill students lost ground in math achievement. To rectify this failure, CPS adopted Double Dose Algebra in 2003 for students scoring below the median on a pre-test. Earlier research, using a regression discontinuity design (RDD), found that Double Dose increased algebra scores, high school graduation, college attendance, and even the college degree attainment of students attending neighborhood high schools serving predominately low-income, minority youth. However, these results are only applicable to median-skill students attending classrooms with similarly skilled peers. Median-skill students who took Double-Dose with low-skill peers did not benefit from additional instructional time. These findings pose a key dilemma. For median-skill students, the optimal policy is to increase instructional time while ensuring that their algebra classes are composed of peers having similar skills. A worry, however, is that such a policy would leave low-skill students isolated from median-skill peers. Prior studies offer competing hypotheses regarding the likely effects of such isolation. One line of reasoning suggests that concentrating low-skill students would lead to a negative stigma and such classes might also induce low expectations, perhaps low motivation among the students, and reduced outcomes. A contrasting view is that homogenous grouping would facilitate tailored instruction, thereby optimizing learning growth. This clash of perspectives generates our research questions. Purpose/Objective/Research Question: We ask whether the double-dose policy enhanced or undermined the learning of low-skill students. We assess how their classroom academic composition was affected and whether such compositional changes were related to success in algebra. Our objective is to better understand the optimal class composition when the policy goal is to help low-skill students catch up by increasing instructional time. Setting: Chicago has the third-largest school system in the country. The student population is about 50% Black, 38% Hispanic, 9% White, and 3% Asian. The FRL rates are 85%. Population/Participants/Subjects: We use 4 cohorts of first-time ninth-grade students who began high schools in 2001 through 2004. The analytic population consists of "low-skill" students, defined as those whose predicted probability of scoring below the national median is greater than .90 based on past academic performance and demographic background (N=18,140 students in 61 schools). Intervention/Program Practice: Ninth graders with 8th-grade math scores below the national median were required to take algebra plus a support course designed to help them build foundational math skills. Their teachers received curricular resources and professional development. Among low-skill students, nearly all non-special education students took Double-Dose, while only 50 percent of special education students did so, and their compliance levels varied considerably by school. Research Design: We estimate the policy impact on the treated students ("compliers") separately by special education status. Our design is a within-school version of the principal score method (Frangakis & Rubin, 2002). Specifically, we reweight pre-policy students to resemble the post-policy Double-dose takers within each school. Then, we adopt a school-specific difference-in-differences (DID) strategy to estimate the policy impact within each school. This effectively controls for fixed school effects, school-specific linear time (cohort) trends, prior academic skills, and Data Collection and Analysis: Our main outcomes are classroom compositions and algebra course failure. We analyze a two-level model after re-weighting to pre-policy students to make them resemble the population of interest (compliers or never takers). Our within-school model for student i attending school j is: Y_ij=?0_j+ ?1_j(cohort)+B_jZ +e_ij, where cohort is a linear cohort variable centered on the 2002 cohort, Z is a post-policy indicator, and e_ij is a zero-mean random error. The coefficient B_j is school specific policy effects. At the school level, we subclassified schools into three levels of policy compliance. We model the coefficients ? 1_j,B_j as a function of school compliance levels Gqj where Gq=1 is medium- and Gq=2 is high-compliance with low-complying schools as the reference group: ?(?_1j=???G_qj ?_1q ?+u_j@ ?_ATTj=???G_qj ?_q ?+b_j ) where are assumed bivariate normal in distribution with means 0, variances Tao_uu, Tao_bb and covariance, Tao_ub. Our analytic model includes the main effect of Z and the interaction between Z and compliance levels. Results: For low-skill regular education students, taking Double-Dose induced a slight decline in classroom peer skill, no change in course failures, but a substantial improvement in algebra scores. For special education Double-Dose takers, classroom peer skills and class size increased and so did their algebra course failures. In high-complying schools where most special education students took Double-Dose, peer composition remained relatively similar post-policy as they continued to take the course with other special education students. Those schools had the smallest increase in algebra course failure post-policy. In contrast, in low-complying schools, special education Double-Dose takers experienced the greatest increase in classroom peer skills and class size as they attended classes with mostly non-special education students post-policy. It is in these schools that course failure increased the most for special education Double-Dose takers. For regular education students, school compliance patterns appear to be unrelated to their classroom composition or academic outcomes. Special education non-compliers were mostly unaffected by the policy. Conclusions: This study considered how Double-Dose algebra affected low-skill students and whether concentrating these students together is potentially detrimental as suggested by some prior studies. Our results showed no such evidence. Our findings, together with previous research on Algebra reforms, guide us to sketch an optimal regime for instructional time and classroom composition for students with varying skills. High-skill students tend to benefit from Algebra instruction in high-skill classes. For median-skill and lower-skill regular education students, homogenous instruction appears to be effective with additional instructional time. Low-skill students, smaller and homogenous classes (e.g., tutoring) may be a more effective approach.