The Effects of English Language Proficiency and Curricular Pathways: Latina/os’ Mathematics Achievement in Secondary Schools

Abstract

This study analyzes nationally-representative quantitative data from the first (2002) and second (2004) waves of the Educational Longitudinal Study to examine the relationship between Latina/o secondary school students’ degree of English-language proficiency (ELP), mathematics course-taking measures, and 12th grade mathematics achievement. Employing a Hierarchical Linear Models (HLM) analytic strategy to account for students’ clustering in specific schools, this study contributes empirical evidence that course-taking and ELP are influential predictors of 12th grade mathematics achievement. Findings from this study exemplify that maximizing Latina/os’ mathematics achievement requires access to rigorous mathematics coursework and the provision of pedagogical and institutional supports that develop students’ proficiency in both the mathematics register and ELP.

Acknowledgments

The authors thank Kip Téllez for providing invaluable detailed feedback on this article as well as the reviewers for their constructive suggestions. They also thank Pedro Nava for his assistance in finalizing the charts and figures. Additionally, the authors express their gratitude to the MEDAL research assistants: Abigail Lopez, Lucia Juarez, Magali Molina, and Alma Martinez.

Notes

1. We define migrant generational status as such: first-generation as immigrant children of immigrant parents; second-generation as U.S.-born children of immigrant parents; and third plus generation as U.S.-born children of U.S.-born parents.

2. Sample weights are used to adjust Latina/o participants to the population of 12th grade Latina/o students in 2004.

3. Widely cited large-scale sociological studies of immigrants using similar types of data sets have used these same self-reported English proficiency measures and find that they are relatively reliable measures of language skills (Portes & Rumbaut, 2001).

4. The principal components analysis routine in STATA yielded the following weighted composite equation: ENG_PROF = .486*UNDERSTAND + .511*SPEAK + .510*READ + .492*WRITE. This single construct of English proficiency captured 68% of the variance in the four English proficiency subscales.

5. The questionnaire weight (for F1) applies to all first follow-up respondents.

6. Guided by factors identified in our literature review, we began to build our analytic models by incrementally adding one variable at a time and refitting the model adhering to this nested design.

7. We took multiple steps in order to avoid issues of multicollinearity. We first analyzed pairwise correlation among explanatory variables to identify highly correlated covariates. We were particularly focused on potentially high degrees of correlation between ELP, SES, first-generation, second-generation, and course-taking. Bivariate correlations showed these to range from low to modest. Respectively, ELP and SES r = .19, ELP and first-generation r = .18, ELP and second-generation r = .18, ELP and third-generation r = –.01, ELP and course-taking, r = .21. In addition, we monitored changes in coefficients as variables were systematically added to our fitted models one at a time or as variables were removed from models. Our nested model approach allowed us to also examine our results for dramatic changes in coefficients’ values and signs. Lastly, we also monitored changes in t-statistics and standard errors to ensure that our nested models remained consistent.

8. The intraclass correlation (ρ) partitions the variance in the outcome due to between-school (τ₀₀) and within-institution (σ²) differences, and is calculated by the following formula: ρ = τ₀₀ / (τ₀₀ + σ²).

9. We calculated the partitioning of variance between level 1 and level 2 by ρ = 23.15 / (23.15 + 54.6).

10. The ICC estimated in Model 2 shows that 22.4% of the variance is due to between-school differences, and 77.6% of the variance is due to differences among students.

11. The ICC estimated in Model 3 shows that 21.9% of the variance is due to between-school differences, and 78.1% of the variance is due to differences among students. Model 3 greatly improves the explanation of variance in the outcome from the unconditional model, and modestly improves such explanation between Model 2 and Model 3.

12. The ICC estimated in Model 4 shows that 20.8% of the variance is due to between-school differences, and 79.2% of the variance is due to differences among students. Each model moderately improves upon the ICC estimates.