1. Information included in the dataset
The dataset includes pooled information on the academic achievement and achievement gaps across various geographic public school districts in the United States. There is no data on individual schools. The data is pooled across grades (3rd through 8th), subjects (Mathematics and Reading Language Arts), and school years (2008-09 through 2018-19). Our chosen data is reported on a cohort standardized (CS) scale, allowing for comparisons against a national reference cohort. The variables in the dataset include the average test scores, the average learning rate (slope across grades), and the average trend in test scores (slope across cohorts), along with the corresponding standard errors for each variable. Also included are the Empirical Bayes (“eb”) versions of each variable, which are a different method of pooling the data that shrinks noisy estimates towards the mean, compared to the overall pooled (“ol”) values.
2. Information the dataset illuminates
The data can reveal the disparities in average test scores, learning rates, and trends in test scores between different geographic public school districts in the United States. Whilst it cannot reveal specific years in which disparities are more severe, having pooled information also has the benefit of reducing the noise of anomalous events to see how these differences have persisted over the entire period. Legislation and programs often require some time to take effect; therefore, the data can potentially reveal specific programs or methods that have worked between districts, or uncover other underlying reasons behind these differences, such as gender, economic status, or race, since the dataset does include these subcategories.
3. What the dataset cannot reveal
The dataset cannot reveal individual student achievement level data on the standardized test scoring due to the data being pooled across all grades, years, and subjects, and instead reports the overall geographic school district achievement estimates. Another major limitation is that the dataset does not extend beyond the limitations of the school environment. Several external factors exist, such as the relationship dynamics within the family (with the parents) involving the living circumstances/environment conditions at home, whether the parents are separated or married as this can have an impact on student stability in their work-life balance, and even the level of educational background and economic income bracket of the parents directly affecting the likelihood of prioritizing receiving tutoring services for their children, which indirectly plays a significant role in shaping the academic prosperity and high-achieving potential in students.
4. How the data was generated
The Stanford Education Data Archive (SEDA) dataset was generated using raw data from the EdFacts database provided by the Department of Education’s National Center for Education Statistics (NCES), where states report test results for schools and districts. This data comes from annual standardized tests in Math and Reading Language Arts (RLA) for students in grades 3–8 across all 50 states from 2008–09 through 2018–19. EdFacts reports the number of students scoring at each proficiency level for nearly every U.S. school. These proficiency thresholds are then translated to a common, nationally standardized scale using results from the National Assessment of Educational Progress (NAEP), allowing scores from different states, years, and grades to be compared on the same metric.
5. Original sources
The dataset uses NAEP (National Assessment of Educational Progress) test scores as its foundational source.
Specifically: HLM using SEDA HETOP est, EdFacts, and SEDA LEA ID
Source Type: Standardized test results from NAEP (a congressionally mandated project administered by the U.S. Department of Education).
Cohorts Included:
4th Graders: Nationally representative test scores from Spring 2009, 2011, 2013, and 2015.
8th Graders: Scores from Spring 2013, 2015, 2017, 2019 (inferred via grade progression)
6. Creators of the dataset
Stanford Education Data Archive (SEDA) team, led by Reardon at Stanford University.
Founders: The construction of SEDA has been supported by grants from the Institute of Education Sciences (R305D110018), the Spencer Foundation, the William T. Grant Foundation, the Bill and Melinda Gates Foundation, the Overdeck Family Foundation, and by a visiting scholar fellowship from the Russell Sage Foundation. Some of the data used in constructing the SEDA files were provided by the National Center for Education Statistics (NCES).
7. Information that is left out
The dataset only includes test scores for Math and “Reasoning Through Language Arts” (RLA), therefore excluding the science and social studies sections of standardized tests. This negatively impacts our understanding of the students’ educational opportunity by narrowing the scope of learning to just two subjects. While the research team presents this data as comprehensive, likely referring to its large breadth of data throughout every state, it lacks holisticity in what it includes and defines as a part of grade school education. If all four sections of standardized testing were included in the dataset, it would provide more insight as to whether the trends in educational opportunity can be attributed to particular subjects or the whole of education that can be represented by the test. This greater level of detail would help researchers analyzing this dataset pinpoint their observations and analyses to more specific influences, and therefore propose advancements to the education system more efficiently.
8. Ideological Effects
The data is displayed through a spreadsheet of 30 variables, including geographic location, school district, the number of Math and RLA tests, and standard deviations of different statistical variables. The data can also be viewed through an interactive map data visualization, where filters can be manipulated to display trends among educational opportunity metrics, region, subgroup/gap (comparison between two subgroups), and other customizable data metrics. These allow researchers using the data to more digestible to view trends in the large amount of data based on different types of racial groups or geographic information that is of their particular interest.