Excerpt
Table of Contents
Background and Introduction
Present Study
NELS and participants’ description
Variable description and descriptive
Model Determination and Plan of Analysis
Regression Analysis
Model comparison
Final Model
Discussion and Conclusion
References
Appendix 1: Figures
Appendix 2: Models with full variables
Abstract
Due to the importance of English language proficiency in education, this study looks to examine the relationship between student performance and language (Stiefel, Schwartz, & Ellen, 2006). Especially it explores student performance on standardized tests assessing science, the first language the student learned, the language used to teach students in their first and second year of learning science in the U.S., and race. This paper used the National Education Longitudinal Study: 1988/2000 (NELS: 88) database from the National Center for Educational Statistics (NCES, 2002). The main aim of this paper is to sensitize teachers If primary language and language of instruction influences how students perform, it is imperative that teaching be adjusted for students who may not speak English as a primary language in school.
Background and Introduction
The congressional act of “No Child Left Behind” is a state standardized tests in which it mandates states to administer standardized assessments in order to receive federal school funding (Starr, 2014). According to Connor & Vagyas (2013), standardized tests are used throughout school systems in the US as a means of accountability for the academic performances of K-12 students. Standardized tests help us to judge comparative successes and competitiveness across the school in United States.
Recent national tests show significant differences in student achievement (McKinsey & Company, 2009). The student’s standardized test scores and ultimate academic success are directly influenced by racial makeup of a school. So far there are substantial scoring differentials among various population groups in many standardized tests. The students in American schools with predominant populations of Caucasian children have consistently scored higher on standardized tests than those in schools with predominant populations of African American children (Lupinski & Jenkins 2005).
According to McKinsey & Company (2009), rich students generally perform better than poor students, white students generally perform better on tests than black students, and students of similar backgrounds perform dramatically differently across school systems and classroom. Asian American students’ performances are comparable to those of white students.
The convergence test scores that took place during 1970, 1980 and 1990 shows the gaps remain unchanged that Asian and white students continue to perform significantly better than Hispanic and Black students on science achievement tests. The white eighth graders score significantly higher than Black or Hispanic students on the National Assessment of Educational Progress (U.S. Department of Education, 2003). The English language proficiency plays an important role in explaining the achievement gaps between Hispanic-white gaps than black-white gaps (Stiefel, Schwartz, & Ellen, 2006). A larger share of Asians and Hispanics live in a home where a language other than English is regularly spoken. Asians but not Hispanics are much more likely to be foreign-born. However, the learning gap between the performance of English Language Learner and non-English Language Learners students was smaller in science (Abedi, 2002)
The differences in socioeconomic background characteristics explain the magnitude of race gaps in test scores (Hedges & Nowell, 1998, 1999). Cook and Evans (2000) estimate that “…the within-school disparities are important-25% of the reduction in the black/white test score gap in that period can be attributed to shifts in family and school characteristics, while 75% can be attributed to reductions within schools”. There is a strong relationship between science scores and family with more resources and learning opportunities at home. In such case the whites tend to have more of these advantages than Blacks or Hispanics (Peng, Wright, & Hill, 1995).
There are number of potential explanations for test score gaps across racial groups in family background (Barton, 2003; Conley, 2001; Scholz & Levine, 2004). The differences in economic prospectus and social pressures faced by children of different races, regardless of neighborhood test influenced score gaps across racial groups in family background (Fordham & Ogbu, 1986), with Steele & Aronson (1995) looking at "acting white" and stereotype threat and Bertrand & Mullainathan (2004) looking at persistence of labor market discrimination. The lack of resources, fewer quality teachers in schools and lower performing peers in schools with larger shares of minority students contribute to racial disparities (Clotfelter, Ladd, & Vigdor, 2005; Lankford, Loeb, & Wyckoff, 2002; Orfield, 2001).
According to the Nation’s Report Card, (2011), the score gaps between Black and White students and, between Hispanic and White students has narrow since 2009 to 2011. The average science scores in 2011 were 3 points higher for Black students, 1 point higher for White students, and 5 points higher for Hispanic students in comparison to 2009. Likewise, there are no significant changes in the scores for Asian/Pacific Islander or American Indian/Alaska Native students from 2009 to 2011. The average scores for both female and male students are lower in 2009 than in 2011.
Present Study
Due to the importance of English language proficiency in education, this study looks to examine the relationship between student performance and language (Stiefel, Schwartz, & Ellen, 2006). Specifically, it explores student performance on standardized tests assessing science, the first language the student learned, the language used to teach students in their first and second year of learning science in the U.S., and race. If primary language and language of instruction influences how students perform, it is imperative that teaching be adjusted for students who may not speak English as a primary language or require schooling that suits their language needs. Thus the questions this study seeks to answer include:
a) is there a significant difference in students’ achievement gaps as determined by race in NELS:88 data set?
b) what effect does the inclusion of primary language or language of instruction for early years of science classes have on later academic performance in science?
c) if an effect exists, how does the magnitude of the impact of language compare to other control covariates?
These questions are descriptive in nature and do not seek to determine any causal, functional or predictive relationships, but rather are meant to explore if there is a relationship between these specific variables. As a result of this design, no specific hypothesis regarding the relationship will be provided and the models analyzed are exploratory in nature. However, it is expected to find a relationship between race and standardized test scores in science that is consistent with previous research.
NELS and participants’ description
This study used the National Education Longitudinal Study: 1988/2000 (NELS: 88) database from the National Center for Educational Statistics (NCES, 2002). The NELS, a nationwide longitudinal dataset, was designed to provide five waves of assessment of the various developmental and psychological aspects of students from eighth grade in 1988 through 2000 (Chang, Singh & Mo 2007). The database consisted of five student component data files, and what we used for this study was from the base year. In other words, we analyzed the first wave of student component data to examine the factors that influenced the performance in science-standardized tests during their eighth grade year 1988.
The NELS applied a two-stage stratified probability sampling design to select a nationally representative sample of schools and students. The school-stage resulted in 1,734 school selections with 1,052 participating schools, including 815 publics and 237 private schools (NCES, 2002). And the participants of NELS were randomly selected from the sampled schools, resulting in participation by 24,599 eighth-grade students. The student-stage sample included 16317 White, 3009 Black, 3171 Hispanic, 1527 Asian/Pacific Islander, and 299 American Indian/ Alaska Native students. Additionally, race data of 276 students was missing.
Variable description and descriptive
The primary outcome variable of interest is the respondent's science standardized score. This is a composite score based on the cognitive tests administered to the students and then standardized. Variables being explored in relation to science standardized scores include race of the respondent, the primary language of the respondent, and in what language the first and second year science was taught in the U.S. Race will be dummy coded using the categories Asian Pacific Islander, Hispanic, Black, and American Indian with the reference category being White. Primary language and language instruction will also be dummy coded similarly into English with Other as a reference category. A comparison between languages and how they affect science standardized scores cannot be made for the respondents who did not give a response to the question. To run the analysis, all cases that did not respond were omitted. Coding Other as the reference category is arbitrary, as either could serve as a reference, and the interpretation requires being mindful of this coding system. There are likely multiple variables that will influence the outcome measure, so these variables will be included to control for their influence. These variables include sex, student science ability, the belief in science being useful, afraid to ask questions in a science class, participation in science fair, and participation in science club. All of these control variables are categorical and will be dummy coded as well. All are binary variables that will be coded with the negative response as the reference, male serving as the reference for sex, and science ability group has 3 categories of high, medium, and low, with low serving as the reference category.
Included here are histograms showing the distribution of the variables of interest. The N for the sample was 4415. Figure 1 shows the distribution of science standardized scores with a mean of 49.202 and a standard deviation of 10.134. Figure 2 shows the categorical variable of race, while Figure 3 and 4 show first language and language of instruction split into English and Other. Figure 5, 6, and 7 show a comparison of science standardized scores by race, first language, and language of instruction respectively. Of the 4415 cases included, 1052 were Asian/Pacific Islander, 2139 were Hispanic, 169 were Black, 121 were American Indian, and 1088 were White. Also, 476 learned English as their primary language, with 4149 learning Other languages and 3672 were taught their first and second year of science class in English, with 953 being taught in Other language. The mean science standardized score for Asian was 53.001(10.282), Hispanic was 46.146(8.861), Black was 46.202(8.976), American Indian was 44.452(9.919), White was 52.706(10.055), English as a first language was 51.014(10.628), Other language as a primary was 48.994(10.056), English as the language of instruction was 49.244(10.186), and Other as the language of instruction was 49.041(9.932).
Model Determination and Plan of Analysis
Abbildung in dieser Leseprobe nicht enthalten
Regression Analysis
To conduct an analysis, all the categorical variables needed to be recoded. Most questions were posed as Yes/No answers and coded into values of 1 for Yes, 2 for No, and others for Missing or No Response. For the control variables, No and No Response were recoded to 0 to serve as a reference category. Group No response into the No category was considered justified as No was the more likely response and a failure to answer was a akin to a negative response. Race was recoded into the individual categories with White serving as a reference. This was done to keep typical white male as a reference, but ultimately is an arbitrary decision. Due to the significant number of missing or no response values for first language and language of instruction, the data could not be effectively re-coded and analyzed while keeping these missing values. Lumping ‘no response’ with the negative response to the question “Was English your first language learned?” would prevent any inferences to be made regarding the difference between languages and their relationship with science standardized test scores. Thus any cases where the respondent failed to answer ‘yes’ or ‘no’ either question asking about first language being English or instruction being in English were omitted prior to analysis.
Other variables originally considered included a composite of socioeconomic status, which incorporated income, parents’ education, and other factors. However, this proved to be too correlated with the standardized science score and inclusion of the SES for this model skewed the outcome variable and violated the assumption of normality and it also violated the assumption of homoscedasticity, we tried several transformation of the outcome variable to address these issues but none of the transformation worked. So we had to remove SES from our model even when it contributed significantly to the R square of the model.
[...]