Written by Alp Doğan Yel, Roy Lahood, Nicholas Hariskos, and Shiyi Chen
I. Introduction
With 44% of US college students in four-year colleges failing to receive their degree before their sixth year[1], identifying learning difficulties at an early age helps to discern systemic and individual issues with the current K-12 education in the United States. Standardized tests have proven to be a cost-effective method of providing a measure of academic ability that is easy to compare across different cohorts, thus increasingly becoming a more important criteria for college admissions in the United States. Given that universities in the United States are trying to admit a more diverse body of students with each passing year, it is vital to differentiate between the multitude of factors that go into determining standardized test scores.
II. Literature Review
Due to the increasing emphasis on socio-economic factors that influence test scores, there exists a broad spectrum of literature pertaining to test scores and factors that affect them. One strand of literature concerning test scores utilizes biological factors in order to explain variation in test scores. In their inquiry into high-stakes quantitative standardized tests such as AP calculus tests and the GRE’s quantitative section, Niederle and Vesterlund (2010) finds stark differences between males and females at the high end of math test scores, while mean performance on the tests are almost identical for both groups. Similarly, looking at differences in reading and math scores in statewide tests administered in Texas and nationwide tests administered in the United States, Klein et al. (2000) finds that there are major gaps between scores of Caucasian students and students of color. Klein et al. (2000) points out that such differences are often not reflected in state-level examinations as the incentive of performing well on the statewide tests far exceeds those on nationwide tests. A similar study done recently by Curran and Kitchin (2018) on a national level suggests that the race-fixed effects for science scores are far greater than those for math and reading scores, which can be explained by differences in language and immigration contexts among Hispanic and Asian students As Curran and Kitchin (2018) point out, the race effects are often correlated with another major factor that affects test scores: language.
Given the significant number of immigrant communities from non-English speaking countries, a significant strand of research in the United States and Canada involves looking at the effects of speaking another language at home (ELS- English Second Language) rather than being a native English speaker (ELO- English Language Only). Using publicly available data from California, Gaastra (2017) concludes that ESL speakers have a negative peer effect on ELO speakers in English test scores, but not on Math. Similarly, investigating the differences in aptitude test scores of Hispanic children growing up in the United States to non-Hispanic children, Locay, Regan, and Diamond (2012) conclude that speaking Spanish at home has negative effect on students’ test scores, which scales with parental education.
To our knowledge, the literature in Economics has not delved much into how the effect of speaking a language other than English at home varies with parental education. Following from previous literature, which indicates that the educational attainment of biological children is highly correlated with that of the mother (Sacerdote, 2007), we hypothesize that the effect of speaking English at home, as opposed to another language, will be diminished as the mother’s educational attainment increases.
III. Description of Data
The data for this study comes from the ECLS-K dataset collected by the National Center for Education Statistics’ Early Childhood Kindergarten Longitudinal Studies (ECLS-K) Program. The ECLS-K dataset follows a nationally representative cohort of students from kindergarten, 1998-99, all the way until the end of middle school, 2007. We’ve specifically utilized the subset of the ECLS-K dataset, the 5th year follow-up in 2004. The 5th year follow-up of the ECLS-K study contains observations about 8105 5th grade students, including variables for standardized test scores, student characteristics, and characteristics of the students’ school.
The key variables in the data set that we used are the standardized test scores of the students, mother’s education, whether or not the students speak a language other than English at home, family income, number of siblings, students’ gender and race, and the type of the school (public, private, or Catholic). The standardized test scores of the students include scores in the subjects of math, science, and reading, all of which are individually graded and standardized to be mean 100 and standard deviation 10. In addition to these three test scores, we’ve constructed an average test score variable by taking the arithmetic mean of math, reading, and science scores of each student.
The distribution of frequency of observations conditional on mother’s education and home language are reported in Table One. The table shows that the distribution of educational attainment of the mother is more skewed to higher educational levels for students who speak English at home and more evenly distributed for students with a home language other than English.
IV. Model
We hypothesize that the effect of speaking English at home will be scaled down as the educational attainment of the mother increases. In order to test this hypothesis, an OLS regression of the general form will be run:
Where score is either one of the four score types: math, science, reading, and their average. Race is a vector of race dummies, Region is a vector of region dummies, and School_Type is a vector of school dummies. Siblings is the number of siblings of the student. Home_Lang_Eng is a dummy that is one if the student speaks English at home and zero otherwise. Mom_Educ is a vector of mother’s education dummies including Some_High_School_Educ, High_School_Grad_Educ, Some_College_Educ, and College_Grad_Educ. The dummy variable corresponding to the highest level of educational level of the mother is one and all else are zero, and the base group, when all the dummies are zero, is elementary school level education. In addition to the model described by equation one (model 1) we also ran an additional model excluding the interaction effects of Mom_Educ and Home_Lang_Eng (model 2), using the results to determine the joint significance of the interaction terms between Mom_Educ and Home_Lang_Eng in an F-test. School characteristics such as dummy variables pertaining to crime, gang, and drug issues in student’s school were excluded from the model since the effects of these variables would be captured by highly correlated variables such as family income and school type.
Given that the data is from a nationally representative sample, we expect that the error term, u, will average out to be zero after controlling for the covariates. Furthermore, given that the data is part of a panel consisting of multiple follow ups, we expect the measurement errors in self-reported sections, such as family characteristics, to be trivial as the surveyors can identify inconsistencies and ask for clarifications.
V. Results
Table 2 reports the results for the regressions on average, math, science, and reading scores, both including the interaction terms between mother’s education and home language (model 1) and excluding them (model 2). The estimates for the interaction terms between mother’s education and home language are, by themselves, mostly not significant up until the 10% level for all four test scores; however, results of the F-test for exclusion restrictions using the difference of sum squared residuals between model 1 and model 2, results of which are reported in table 3, indicate that the interaction terms are jointly significant down to the 1% level for all four test scores. Table 4 summarizes the results of the interaction terms between mother’s education and home language, reporting the total effect of speaking English at home conditional on mother’s educational attainment for all four test scores. Given the imprecise nature of the estimates, we cannot make a confident conclusion regarding the direction and the magnitude of the effect of speaking English at home on test scores. Although, the aforementioned F-test shows that the effects of speaking English at home scales with the educational attainment of the mother, and without more precise estimates we cannot make a confident conclusion on the magnitude and direction of this scaling effect.
With the estimates we have on hand, there seems to be a mostly positive effect of speaking English at home on science, reading, and average test scores but not on math. The lack of a positive impact of speaking English at home on math scores is in line with the findings in previous literature (Gaastra, 2017). We explain this lack of a positive effect by attributing it to the symbolic nature of math overcoming linguistic barriers in instruction, which can be supported anecdotally by the experience of ESL speakers learning math in English, but to our knowledge the current literature is lacking in this area. The fact that there is a positive impact of speaking English at home on science scores does not contradict this line of reasoning, as science curriculum in the 5th grade level is less mathematical, thus science tests are based more on terminology and comprehension rather than math. In terms of magnitude, the positive impact of speaking English, seen in Table 4, ranges from insignificant levels such as 0.4 points (0.04 standard deviations) up to moderately significant levels 2.7 points (0.27 standard deviations). The greatest effect is seen for science scores where students with a mother who has some high school education on average scores 2.7 points (0.27 standard deviations) more if they speak English at home. On the other hand, the smallest effect is seen for reading scores where students with a mother who also has some high school education on average scores 0.00247 points (0.000247 standard deviations) more if they speak English at home, which is negligible.
Although the results for the F-test (Table 3) indicate that the effect of speaking English at home scales with the educational attainment of the mother, the direction of this effect is not clear with the estimates on hand. As seen in Table 4, the total effect of speaking English at home scales in a parabolic fashion, initially scaling up until mother’s education reaches high school, but scales down as mother’s educational attainment reaches college or beyond. The downward trajectory on the scaling effects of mother’s education on the effect of speaking English at home is to be expected as the mother’s educational attainment is correlated with that of the child’s (Sacerdote, 2007), which suggests that children whose mothers have a higher educational attainment are on average more academically able, thus are less impacted by linguistic hindrances of learning in their non-native language. The initial upward trajectory, on the other hand, is harder to explain as there is no cohesive explanation that we can come up with other than the variation in the estimates.
VI. Caveats and Suggestions for Future Research
Like all research, this study has its limitations. As stated in the results section, for all four test scores, we mostly fail to reject the null hypothesis that the interaction terms between mother’s education and home language are individually zero at the 10% significance level; however, running an F-test we reject the null hypothesis that the interaction terms are jointly zero at the 1% significance level, indicating that the effect of speaking English at home scales with the educational attainment of the mother. Yet, due to the imprecise nature of our estimates, we were not able confidently determine the direction and magnitude of the total effect of speaking English at home on the test scores. Using the same model, better estimates may be obtained in the future by using a larger, yet still nationally representative, sample.
The model described in equation 1 relies on two assumptions that we have made in the absence of further data. The first assumption is that students who do not speak English at home have parents that speak a language other than English and those parents are not bilingual. The ECLS-K dataset only indicates whether the student speaks a language other than English at home, but does not provide further information regarding their own proficiency in English as well as their parents. One might assume the scaling effects of mother’s education on home language to be further affected by whether or not the parents are bilingual as well as the native language they speak at home. Linguistically, some languages are closer to English than others, therefore the effects of not speaking English at home may also vary based on the actual language spoken at home. For instance, speaking Spanish at home could have a smaller impact on test scores compared to Chinese or Arabic as Spanish comes from the same linguistic family whereas Chinese or Arabic do not. Ideally, data used in further research should include an indicator on whether or not the parents are bilingual, if the students were placed in an English as a Second Language (ESL) classroom, and include parent-reported English proficiency of their child or scores from an external English language proficiency test. Such variables would have given us more insight into what extent the language spoken at home is correlated with proficiency in English and thus allow us to further disentangle the effects of speaking a language other than English at home on test scores.
The second assumption is that all children are the biological children of the mother in the study. As with the case of home language, the ECLS-K dataset only provides data for the mother’s education and family characteristics such as the family type and number of siblings, but none on whether the children are biological or not. If it is the case that some of the mothers are not the biological mother of the child whose education is reported, then we have the issue of having to disentangle the effects of nature, inherited ability, and nurture, the effects of the home environment in learning. Assuming that the cases where the mother whose education is reported in the data are insignificant overall in the dataset helps us overcome the perennial issue of nature vs nurture as in both cases the effects of speaking English at home as opposed to another language should be scaled by the parental education. Whether a more educated mother creates a better learning environment by tutoring her child or the fact that a significant proportion of mother’s education is transmitted to the child (Sacerdote, 2007), meaning children of mothers with higher educational attainment are on average more academically able, having a more educated mother should diminish the effects of having a non-English home language on test scores. Ideally, the data used in future research should include variables for father’s education, which we’d assume would be correlated with the educational attainment of the child just like the mother, and a variable to indicate whether or not the parents are the actual biological parents of the child.
Finally, given that the ECLS-K dataset includes data for the same students across different grade levels, future research may dwell on comparing the effects of home language, and how these effects scale by parental education, as the student gets older.
VII. Conclusion
The results of this study suggest that the effects of speaking a language other than English at home scales by the parental educational attainment; however, the magnitude and direction of the scaling effect remains unclear in the absence of more precise estimators. These results are consistent with findings of Locay, Regan, and Diamond (2012) on effects of speaking Spanish at home on aptitude tests. The results at hand suggest that speaking English at home has a positive effect on reading and science scores but not on math scores, which would be consistent with the findings of Sieuwerd Gaastra, which suggests that ESL students have a negative peer effect on ELO students in English scores but not on math scores. Furthermore, the results at hand indicate a parabolic effect of mother’s education on effects of speaking English at home, increasing initially than decreasing afterwards. While the decreasing direction of this effect can be attributed to the increasing educational ability of the students as suggested by correlation between mother’s education and that of the child (Sacerdote, 2007), the initial increase is harder to account for with a single cohesive narrative. We once again caution the reader that these effects might vary in magnitude and direction as estimates are individually not significant at the 10% level.
References
Curran, F. C., & Kitchin, J. (2019). Why Are the Early Elementary Race/Ethnicity Test Score Gaps in Science Larger Than Those in Reading or Mathematics? National Evidence on the Importance of Language and Immigration Context in Explaining the Gap‐in‐Gaps. Science Education, 103, 477-502. https://doi-org.proxy.lib.umich.edu/10.1002/sce.21491
Dooley, M., & Furtado, C. (2013). ESL Policy Reform and Student Academic Achievement. Canadian Public Policy / Analyse De Politiques, 39(1), 21-43. https://www.jstor.org/stable/23594701?seq=1
Klein, S. P., Hamilton, L., McCaffrey, D., & Stecher, B. (2000). What Do Test Scores in Texas Tell Us?. Education Policy Analysis Archives, 8, 49. https://doi.org/10.14507/epaa.v8n49.2000
Locay, L., Regan, T.L., & Diamond, A.M. (2012). The Effects of Spanish-Language Background on Completed Schooling and Aptitude Test Scores. Economic Inquiry, 51(1), 1-33. https://doi.org/10.1111/j.1465-7295.2012.00458.x
Niederle, M., & Vesterlund, L. (2010). Explaining the Gender Gap in Math Test Scores: The Role of Competition. Journal of Economic Perspectives, 24(2), 129-44. https://doi.org/10.1257/jep.24.2.129
Sacerdote, B. (2007). How Large are the Effects from Changes in Family Environment? A Study of Korean American Adoptees. The Quarterly Journal of Economics, 122(1), 119-157. https://doi.org/10.1162/qjec.122.1.119
Gaastra, S. (2017). Essays in Education and Public Economics. University of California: San Diego.https://escholarship.org/uc/item/9tb068bm
[1] Lou Carlozo, Why Students Stop Short of a Degree (Reuters 2012)