The world’s leading publication for data science, AI, and ML professionals.

The Impacts of COVID-19 on the Achievement Gap

With COVID-19 placing the U.S. economy in turmoil, how could decreasing economic prospects along racial lines impact the achievement gap?

Data for Change

Abstract

Given the widespread and unequal effects of the COVID-19 pandemic on the socioeconomic health of the American population, the economic implications of the pandemic along racial lines and their potential repercussions on the academic success of students of all levels have gained new interest in the United States. Examining the impacts of the pandemic on the Achievement Gap is especially relevant now due to the disruptions in education and the transition towards remote teaching by most educational institutions. This work focuses on the correlations between poverty and income with high school academic achievement, then examines how family incomes and unemployment rates have changed among the U.S. population once that correlation is established. Analysis was carried out on publicly available datasets provided by the National Center for Educational Statistics High School Longitudinal Study and the Bureau of Labor Statistics Current Population Survey. The analyses done in this study suggest significant statistical correlations between poverty and high school academic achievement indicators, between family income and high school academic achievement indicators, and between race and high school academic achievement indicators. This study also revealed that people of color have typically experienced higher levels of unemployment than their white counterparts have since the first pandemic lockdowns in March, and that there has been an exacerbation of income inequality in the United States along racial lines. In aggregate, these findings imply that, as far as socioeconomic indicators such as race, poverty, and income are concerned, the pandemic may have a profound impact on the academic achievement of students of color and poverty and widen the already pronounced racial achievement gap.

This is part of a larger project by Ursatech Berkeley: Impacts of the COVID-19 Pandemic on the American Socioeconomic Academic Achievement Gap Through the Perspective of Race, Income, Unemployment, and Poverty.

At Ursatech Berkeley, we give back to our campus community by providing pro-bono consulting services within the UC Berkeley community and lending our expertise to university-affiliated groups. This study is part of a greater independent project by Ursatech Berkeley to examine the academic impacts of the coronavirus pandemic along racial lines.

Introduction

Several socioeconomic indicators are relevant in predicting students’ academic success in secondary Education. In fact, the correlations between national incomes, socioeconomic status (SES), race, and educational attainment are well documented by existing literature (the influence of maternal education and family wages on children’s math and reading scores, for example [7]). As the coronavirus pandemic and its relevant effects continue in the United States, it becomes pertinent to assess the potential changes in education caused by its economic and social implications. Most importantly, the negative impacts of the pandemic are not felt equally across the American population. Recent evidence suggests profound "racial, economic, and health inequality in the population infected by and dying from COVID-19," with people of color, particularly African Americans, as well as individuals of poverty or with disabilities, experiencing higher rates of mortality due to the virus [1].

This study seeks to examine the extent to which income, poverty, and race influence high school academic performance. Once correlations between economic indicators and academic performance have been established, this study will then examine how income and unemployment have been impacted in the United States along racial lines; tracking socioeconomic developments by race should give a clear picture of changes in the American racial achievement gap. In fact, multiple studies emphasize the importance of socioeconomic factors with regard to academic success. The Coleman Report, for instance, argues that a student’s family SES is a greater indicator of educational success than, for example, school quality and academic resources [4]. Therefore, the initial hypotheses of this study are that there exist significant correlations between income and educational attainment, and between poverty and educational attainment. This study further hypothesizes that populations of color, particularly the African American and Hispanic populations, are more severely impacted in terms of income and unemployment rates than the general U.S. population is because of the pandemic. Hence, while verifying the hypotheses, this study seeks to answer the following questions about students’ SES, race, and academic performance:

  • Question 1: To what extent are academic achievement gaps among students of different family incomes or poverty levels statistically significant?
  • Question 2: Are there any correlations between family income and a student’s academic performance, and if so, how significant are those correlations?
  • Question 3: Are there any correlations between poverty and a student’s academic performance, and if so, how significant are those correlations?
  • Question 4: Are there any correlations between student race and academic performance, and if so, how significant are those correlations?
  • Question 5: How have unemployment rates changed during the course of the COVID-19 pandemic along racial lines?
  • Question 6: How have family income distributions changed during the course of the COVID-19 pandemic along racial lines?

Methodology

Two empirical datasets were analyzed: the High School Longitudinal Study (HSLS) conducted by the National Center for Educational Statistics (NCES) in 2009 and the Basic Monthly Current Population Survey (CPS) conducted by the Bureau of Labor Statistics. The HSLS surveyed fall 2009 ninth-grade students with follow-ups in 2012 and 2016 and tracked their progress through high school and college. Four academic variables from the HSLS were considered, referred to in this report as "academic indicators": mathematics theta scores from the first follow-up, mathematics quintile scores from the first follow-up based on mathematics test scores, overall high school honors-weighted GPAs, and college enrollment status as of November 1, 2013 (variables X2TXMTH, X2TXMQUINT, X3TGPAWGT, and S3CLGFT respectively). Two economic variables from the HSLS were considered: students’ 2011 family incomes and 2011 poverty levels relative to 100% of the Census poverty threshold (variables X2FAMINCOME and X2POVERTY, respectively). (For reference, as of 2011, a family of four with income less than $22,350 per year would be under the 100% poverty threshold in the contiguous United States and Washington, D.C.) Respondents’ race was also considered throughout the examination of both the HSLS and CPS data. From the CPS, analysis was done on respondents’ 12-month family incomes and month labor force recodes (variables hefaminc and pemlr, respectively) [14]. (For reference, the income brackets for the HSLS and CPS are listed in Figures 0a and 0b, respectively.)

Figure 0a - High School Longitudinal Study income brackets and corresponding dollar values.
Figure 0a – High School Longitudinal Study income brackets and corresponding dollar values.
Figure 0b - Current Population Survey income brackets and corresponding dollar values.
Figure 0b – Current Population Survey income brackets and corresponding dollar values.

There were some limitations to this data, however. Due to the small numbers of mixed-race respondents in the CPS, and to be consistent with the HSLS, only respondents who were of a single race were included. However, some racial groups in the CPS, such as those of Hawaiian and Pacific Islander descent, only had a few hundred respondents, which resulted in widely variable aggregate statistics. Some variables from the CPS and HSLS also had high numbers of non-respondents which may skew the data, although these variables were not used in the final iteration of this study. Because the HSLS was a publicly released dataset containing some confidential information, variables including SAT and other standardized test scores have been suppressed, limiting the scope of this study. Additionally, the income data from the CPS asked respondents to report their total income for the past 12 months rather than for the current month, which may make income statistics appear more optimistic.

The data was cleaned of non-respondents, then visualized and modeled using standard Python data analysis libraries (e.g. Pandas, NumPy, Matplotlib, Seaborn, Sklearn, Statsmodels, and Tensorflow). Relevant data were plotted and graphed. One-way Analysis of Variance (ANOVA) testing was carried out on mathematics theta scores and high school GPA with income and poverty as the independent variables.


Phase 1 Analysis: NCES High School Longitudinal Study

Answering Question 1 – Hypothesis Test: One-way ANOVA

Before gathering and analyzing any further data, hypothesis testing was carried out on mathematics theta scores against economic variables. Using a 5% significance level, ANOVA testing was carried out twice: once with mathematics theta scores and income as the dependent and independent variables, respectively, and again with mathematics theta scores and poverty levels as the dependent and independent variables, respectively. The sample was the mathematics theta scores of high school students surveyed by the NCES. The null hypothesis assumes that the mean mathematics theta scores for students of different income brackets are equal, and that the mean mathematics theta scores for students above the poverty threshold and students below the poverty threshold are equal. (See Figures 1a-2b for tables and visualizations.)

Following are the visualizations, implementations, and results of the two ANOVA tests:

Figure 1a: Mathematics theta score statistics based on family income. Income brackets are based on the definitions used in the HSLS. 1 indicates the lowest income bracket surveyed (≤$15000 per year) and 13 indicates the highest income bracket surveyed (>$235,000 per year). See Figure 0a for corresponding income bracket dollar values. Figure by author based on NCES data.
Figure 1a: Mathematics theta score statistics based on family income. Income brackets are based on the definitions used in the HSLS. 1 indicates the lowest income bracket surveyed (≤$15000 per year) and 13 indicates the highest income bracket surveyed (>$235,000 per year). See Figure 0a for corresponding income bracket dollar values. Figure by author based on NCES data.
Figure 1b: Mathematics theta score means and confidence intervals based on family income. Income brackets are based on the definitions used in the HSLS. 1 indicates the lowest income bracket surveyed (≤$15000 per year) and 13 indicates the highest income bracket surveyed (>$235,000 per year). See Figure 0a for corresponding income bracket dollar values. Figure by author based on NCES data.
Figure 1b: Mathematics theta score means and confidence intervals based on family income. Income brackets are based on the definitions used in the HSLS. 1 indicates the lowest income bracket surveyed (≤$15000 per year) and 13 indicates the highest income bracket surveyed (>$235,000 per year). See Figure 0a for corresponding income bracket dollar values. Figure by author based on NCES data.
Figure 2a: Mathematics theta score statistics based on poverty status. The poverty level is relative to 100% of the Census poverty threshold. Figure by author based on NCES data.
Figure 2a: Mathematics theta score statistics based on poverty status. The poverty level is relative to 100% of the Census poverty threshold. Figure by author based on NCES data.
Figure 2b: Mathematics theta score means and confidence intervals based on poverty status. The poverty level is relative to 100% of the Census poverty threshold. Figure by author based on NCES data.
Figure 2b: Mathematics theta score means and confidence intervals based on poverty status. The poverty level is relative to 100% of the Census poverty threshold. Figure by author based on NCES data.

Below are the implementation and results of the ANOVA tests:

# ANOVA testing mathematical theta scores against family income
math_income_str = math_income.copy()
math_income_str['X2FAMINCOME'] = math_income['X2FAMINCOME'].apply(str)
math_income['X2FAMINCOME'] = math_income['X2FAMINCOME'].apply(str)
math_income_mod = ols('X2TXMTH ~ X2FAMINCOME', data=math_income_str).fit()
math_income_aov_table = sm.stats.anova_lm(math_income_mod, typ=2)
print(math_income_aov_table)
                   sum_sq       df           F  PR(>F)
X2FAMINCOME   2833.967430     12.0  198.623833     0.0
Residual     24470.831239  20581.0         NaN     NaN
# ANOVA testing mathematical theta scores against poverty
math_poverty_str = math_poverty.copy()
math_poverty_str['X2POVERTY'] = math_poverty['X2POVERTY'].apply(str)
math_poverty_mod = ols('X2TXMTH ~ X2POVERTY', data=math_poverty_str).fit()
math_poverty_aov_table = sm.stats.anova_lm(math_poverty_mod, typ=2)
print(math_poverty_aov_table)
                 sum_sq       df           F         PR(>F)
X2POVERTY    987.871369      1.0  772.971973  4.899952e-167
Residual   26316.927301  20592.0         NaN            NaN

Both tests produced p-values well lower in magnitude than 10⁻¹⁰⁰. Hence, the probability of observing the sample data (Figures 1b and 2b) given that the null hypothesis is true is very low. Since the p-values are below the significance level of 0.05, the null hypothesis can be rejected. These tests suggest the probability that the mean mathematics scores are equal for students among different income brackets or different poverty levels is extremely low.

ANOVA testing was also carried out on overall high school honors-weighted GPA. Using a 5% significance level, ANOVA testing was carried out twice: once with overall high school honors-weighted GPA and income as the dependent and independent variables, respectively, and again with overall high school honors-weighted GPA and poverty as the dependent and independent variables, respectively. The sample was the overall high school honors-weighted GPA of high school students surveyed by the NCES. The null hypothesis assumes that the mean overall high school honors-weighted GPA for students of different income brackets are equal, and that the mean overall high school honors-weighted GPA for students above the poverty threshold and students below the poverty threshold are equal. (See Figures 3a-4b for tables and visualizations.)

Following are the visualizations, implementations, and results of the two ANOVA tests:

Figure 3a: Overall high school honors-weighted GPA statistics based on family income. Income brackets are based on the definitions used in the HSLS. 1 indicates the lowest income bracket surveyed (≤$15000 per year) and 13 indicates the highest income bracket surveyed (>$235,000 per year). See Figure 0a for corresponding income bracket dollar values. Figure by author based on NCES data.
Figure 3a: Overall high school honors-weighted GPA statistics based on family income. Income brackets are based on the definitions used in the HSLS. 1 indicates the lowest income bracket surveyed (≤$15000 per year) and 13 indicates the highest income bracket surveyed (>$235,000 per year). See Figure 0a for corresponding income bracket dollar values. Figure by author based on NCES data.
Figure 3b: Overall high school honors-weighted GPA means and confidence intervals based on family income. Income brackets are based on the definitions used in the HSLS. 1 indicates the lowest income bracket surveyed (≤$15000 per year) and 13 indicates the highest income bracket surveyed (>$235,000 per year). See Figure 0a for corresponding income bracket dollar values. Figure by author based on NCES data.
Figure 3b: Overall high school honors-weighted GPA means and confidence intervals based on family income. Income brackets are based on the definitions used in the HSLS. 1 indicates the lowest income bracket surveyed (≤$15000 per year) and 13 indicates the highest income bracket surveyed (>$235,000 per year). See Figure 0a for corresponding income bracket dollar values. Figure by author based on NCES data.
Figure 4a: Overall high school honors-weighted GPA statistics based on poverty levels. The poverty level is relative to 100% of the Census poverty threshold. Figure by author based on NCES data.
Figure 4a: Overall high school honors-weighted GPA statistics based on poverty levels. The poverty level is relative to 100% of the Census poverty threshold. Figure by author based on NCES data.
Figure 4b: Overall high school honors-weighted GPA means and confidence intervals based on poverty status. The poverty level is relative to 100% of the Census poverty threshold. Figure by author based on NCES data.
Figure 4b: Overall high school honors-weighted GPA means and confidence intervals based on poverty status. The poverty level is relative to 100% of the Census poverty threshold. Figure by author based on NCES data.

Below are the implementation and results of the ANOVA tests:

# ANOVA testing high school GPAs against family income
gpa_income_str = gpa_income.copy()
gpa_income_str['X2FAMINCOME'] = gpa_income['X2FAMINCOME'].apply(str)
gpa_income_mod = ols('X3TGPAWGT ~ X2FAMINCOME', data=gpa_income_str).fit()
gpa_income_aov_table = sm.stats.anova_lm(gpa_income_mod, typ=2)
print(gpa_income_aov_table)
                   sum_sq       df          F  PR(>F)
X2FAMINCOME   1753.831408     12.0  191.50842     0.0
Residual     15105.334589  19793.0        NaN     NaN
# ANOVA testing high school GPAs agains poverty
gpa_poverty_str = gpa_poverty.copy()
gpa_poverty_str['X2POVERTY'] = gpa_poverty['X2POVERTY'].apply(str)
gpa_poverty_mod = ols('X3TGPAWGT ~ X2POVERTY', data=gpa_poverty_str).fit()
gpa_poverty_aov_table = sm.stats.anova_lm(gpa_poverty_mod, typ=2)
print(gpa_poverty_aov_table)
                 sum_sq       df          F         PR(>F)
X2POVERTY    732.730336      1.0  899.82634  2.229521e-193
Residual   16126.435661  19804.0        NaN            NaN

Once again, both tests produced p-values well lower in magnitude than 10⁻¹⁰⁰. Hence, the probability of observing the sample data (Figures 3b and 4b) given that the null hypothesis is true is very low. Since the p-values are below the significance level of 0.05, the null hypothesis can be rejected. These tests suggest the probability that the high school honors-weighted GPA means are equal for students among different income brackets or different poverty levels is extremely low.

Answering Question 2 – Visualizing Academic Measures by Total Family Income: Overall Honors-Weighted High School GPA, College Enrollment, and Mathematical Proficiency

Figures 5a-d partition students based on race and, for each income bracket, illustrate the distribution of high school students based on students’ overall high school honors-weighted GPA, student college enrollment status as of November 1, 2013, and HSLS mathematics quintile based on student mathematics scores. The datasets were downloaded as CSV files from the NCES HSLS website [15]. The datasets were then filtered to only include survey respondents, and grouped by the 13 family income brackets used by the HSLS. The rest of the code used to process and analyze the data is in the Github repository listed in the references section [16].

Figure 5a: Distribution of student overall high school GPAs based on family income. Note that GPA values are categorical. See the NCES documentation for more. Figure by author based on NCES data.
Figure 5a: Distribution of student overall high school GPAs based on family income. Note that GPA values are categorical. See the NCES documentation for more. Figure by author based on NCES data.
Figure 5b: Distribution of college enrollment status based on family income. Note that the college enrollment status is categorical. See the NCES documentation for more. Figure by author based on NCES data.
Figure 5b: Distribution of college enrollment status based on family income. Note that the college enrollment status is categorical. See the NCES documentation for more. Figure by author based on NCES data.
Figure 5c: Distribution of students in mathematics quintiles based on family income. 5 is the highest-achieving mathematics quintile and 1 is the lowest-achieving mathematics quintile. Quintiles are based on students' math scores (X2TXMTSCOR). Figure by author based on NCES data.
Figure 5c: Distribution of students in mathematics quintiles based on family income. 5 is the highest-achieving mathematics quintile and 1 is the lowest-achieving mathematics quintile. Quintiles are based on students’ math scores (X2TXMTSCOR). Figure by author based on NCES data.

Based on these findings, it is apparent that a greater proportion of students achieve greater overall high school honors-weighted GPAs, go to college full-time, and reach higher mathematics quintile scores as family income increases. Considering the breadth of academic indicators analyzed, the data suggest a positive association between family income and overall academic performance. In some extreme cases, a $20,000 increase in family income can improve a student’s overall high school honors-weighted GPA by 0.203 standard deviations on average. Most notably, in Figure 5c, there is a negative association between the mathematics quintile and the percent of students within that quintile for the two lowest income brackets surveyed, indicating that students in lower income brackets tend to be academically underachieving. Figure 5d in particular illustrates how higher incomes typically correlate with higher mathematics theta scores. Conversely, at least half of students in income brackets in excess of $115,000 per year achieve overall high school GPAs of 3.50 or higher. Interestingly, Figure 3b reveals that for incomes above the seventh income bracket, improvements in average honors-weighted high school GPA become far less substantial, suggesting that correlations between income and academic performance may not be as strong once family incomes grow in excess of $115,000 per year.

Figure 5d: Mathematics theta score distribution based on family income bracket graphed using a violin plot. Income brackets are based on the definitions used in the HSLS. 1 indicates the lowest income bracket surveyed (≤$15000 per year) and 13 indicates the highest income bracket surveyed (>$235,000 per year). Income brackets are at $20,000 intervals. Figure by author based on NCES data.
Figure 5d: Mathematics theta score distribution based on family income bracket graphed using a violin plot. Income brackets are based on the definitions used in the HSLS. 1 indicates the lowest income bracket surveyed (≤$15000 per year) and 13 indicates the highest income bracket surveyed (>$235,000 per year). Income brackets are at $20,000 intervals. Figure by author based on NCES data.

Answering Question 3 – Visualizing Academic Measures by Poverty Level: Overall Honors-Weighted High School GPA, College Enrollment, and Mathematical Proficiency

A similar analysis was done along the lines of poverty. Figures 6a-d partition students based on the poverty threshold and illustrate the distribution of high school students based on students’ overall high school honors-weighted GPA, student college enrollment status as of November 1, 2013, and HSLS mathematics quintile based on student mathematics scores. The datasets were downloaded as CSV files from the NCES HSLS website [15]. The datasets were then filtered to only include survey respondents and students were grouped by poverty levels. The rest of the code used to process and analyze the data is in the Github repository listed in the references section [16].

Figure 6a: Distribution of student overall high school honors-weighted GPAs based on poverty level relative to 100% of Census poverty threshold. Note that GPA values are categorical. See the NCES documentation for more. Figure by author based on NCES data.
Figure 6a: Distribution of student overall high school honors-weighted GPAs based on poverty level relative to 100% of Census poverty threshold. Note that GPA values are categorical. See the NCES documentation for more. Figure by author based on NCES data.
Figure 6b: Distribution of college enrollment status based on poverty level relative to 100% of Census poverty threshold. Note that the college enrollment status is categorical. See the NCES documentation for more. Figure by author based on NCES data.
Figure 6b: Distribution of college enrollment status based on poverty level relative to 100% of Census poverty threshold. Note that the college enrollment status is categorical. See the NCES documentation for more. Figure by author based on NCES data.
Figure 6c: Distribution of students in mathematics quintiles based on poverty level relative to 100% of Census poverty threshold. 5 is the highest-achieving mathematics quintile and 1 is the lowest-achieving mathematics quintile. Quintiles are based on students' math scores (X2TXMTSCOR). Figure by author based on NCES data.
Figure 6c: Distribution of students in mathematics quintiles based on poverty level relative to 100% of Census poverty threshold. 5 is the highest-achieving mathematics quintile and 1 is the lowest-achieving mathematics quintile. Quintiles are based on students’ math scores (X2TXMTSCOR). Figure by author based on NCES data.

Similar to the findings regarding family incomes, there is a distinct positive association between being above the poverty level and the proportion of students who earn higher overall high school honors-weighted GPAs, attend college full-time, and achieve greater mathematics quintile scores. In fact, being above the poverty level improves a student’s high school honors-weighted GPA by 0.541 standard deviations on average. The percentage of students above the poverty threshold who attend college full time by November 1, 2013 is 88.28% compared to 76.24% of students below the poverty threshold. Most notably, while there is a positive correlation between a higher mathematics quintile and the percentage of students within that quintile for students above the poverty line, there is a strong negative correlation between a higher mathematics quintile and the percentage of students within that quintile for students below the poverty line, implying that students below the poverty line are academically underachieving. Note in Figure 6d how mathematics theta score distributions tend upwards for students above the poverty level and downwards for students below the poverty level. These findings suggest there is a considerable achievement gap between students above and below the poverty threshold and a strong association between poverty and high school academic performance.

Figure 6d: Mathematics theta score distribution based on poverty level. Figure by author based on NCES data.
Figure 6d: Mathematics theta score distribution based on poverty level. Figure by author based on NCES data.

Answering Question 4 – Visualizing Academic Measures by Race: Overall Honors-Weighted High School GPA, College Enrollment, and Mathematical Proficiency

Academic indicator analysis was also carried out along racial lines. Figures 7a-7c illustrate the distribution of high school students based on students’ overall high school honors-weighted GPA, student college enrollment status as of November 1, 2013, and HSLS mathematics quintile based on student mathematics scores. Data from the HSLS were used to generate the aforementioned visuals. Eight racial groups were of particular interest: American Indian and Alaskan Native, Asian, Black and African-American, Hispanic (no race specified), Hispanic (race specified), more than one race (non-Hispanic), Native Hawaiian and Pacific Islander, and White.

Figure 7a: Distribution of student overall high school honors-weighted GPAs based on race. Note that GPA values are categorical. See the NCES documentation for more. Figure by author based on NCES data.
Figure 7a: Distribution of student overall high school honors-weighted GPAs based on race. Note that GPA values are categorical. See the NCES documentation for more. Figure by author based on NCES data.
Figure 7b: Distribution of student college enrollment status based on student race. Note that the college enrollment status is categorical. See the NCES documentation for more. Figure by author based on NCES data.
Figure 7b: Distribution of student college enrollment status based on student race. Note that the college enrollment status is categorical. See the NCES documentation for more. Figure by author based on NCES data.
Figure 7c: Distribution of students in mathematics quintiles based on student race. 5 is the highest-achieving mathematics quintile and 1 is the lowest-achieving mathematics quintile. Quintiles are based on students' math scores (X2TXMTSCOR). Figure by author based on NCES data.
Figure 7c: Distribution of students in mathematics quintiles based on student race. 5 is the highest-achieving mathematics quintile and 1 is the lowest-achieving mathematics quintile. Quintiles are based on students’ math scores (X2TXMTSCOR). Figure by author based on NCES data.

There are distinct differences between the racial groups’ overall high school honors-weighted GPA distributions, college enrollment distributions, and mathematics quintile distributions. A greater percentage of White and Asian students earn GPAs of 3.0 or above, attend college full-time, and achieve higher mathematics quintile scores as determined by the HSLS when compared to other racial groups. For instance, while 89.50% and 90.05% of White and Asian students, respectively, attend college full-time, those rates are only 81.97% and 77.55% for Black and Hispanic students, respectively. Furthermore, based on the mathematics quintile statistics provided by the HSLS, 27.32% and 50.50% of White and Asian students, respectively, achieved the highest mathematics quintile based on their mathematics scores, while those proportions were only 9.51% and 14.19% for Black and Hispanic students, respectively. These findings suggest that there is a sizeable achievement gap within the American secondary education system based on race [16].

Phase 1: Explanations for Observed Correlations

Clearly, families with higher incomes can provide students with greater sources of academic enrichment including academic tutoring, educational mobility (e.g. the ability to move to a higher-quality school district), and exposure to literature at a young age. However, many studies emphasize the importance of behavioral factors as predictors for academic success. The poor academic performance of low income students can be partially explained by higher rates of suspension and expulsion; missing school increases student absence and can reduce their performance in academic assessments [2]. Other studies point to both external factors, such as parental investment and cognitive stimulation, as well as internal and cognitive factors, such as task persistence, to explain the gap in achievement among low-income students [13]. In terms of poverty specifically, there is a high prevalence of mental health issues such as depression, impulse control, and decreased cognitive ability among children living in poverty, due to their high exposure to chronic stressors (e.g. unsafe neighborhoods, financial strain, separated or divorced parents, overcrowded households, etc.) [9] that may influence academic achievement.

Moreover, it is easy to illustrate how the socioeconomic achievement gap could transfer to race. As an example, previous government policies, such as the Fair Housing Act of 1968, have led to the geographic segregation of people of color, in particular African Americans, resulting in economic isolation, the segregation of school districts, and, therefore, inequality in terms of academic success [12], revealing that the racial achievement gap is closely related to poverty and socioeconomics. Other social biases, such as the fact that students of minorities are more likely to experience negative interactions with teachers than their white student counterparts [2], may also contribute to differences in achievement among different racial groups.


Phase 2 Results and Analysis: Basic Monthly Current Population Survey

Phase 2 of this study concerns tracking economic indicators across the COVID-19 pandemic for different racial groups in the U.S. The primary dataset used was the Bureau of Labor Statistics Current Population Survey, which tracks the employment status of all individuals of a household who are of working age. Of particular interest for this study are household incomes and respondents’ employment status (i.e. employed, unemployed, or outside of the labor force).

In this study, monthly household incomes and labor force recodes were isolated from the dataset along with respondents’ race. Non-respondents were filtered from the dataset, and changes in income and employment were aggregated, then graphed along racial lines. Six racial groups from the CPS were considered: Hispanic, White, Black, American Indian and Alaskan Native, Asian, and Hawaiian and Pacific Islander. The time period of interest is January to September 2020.

Answering Question 5 – Unemployment Rates During the Pandemic: January to September, 2020

Employment statistics were aggregated along racial lines and used to produce unemployment numbers (Figures 8a-b).

Figure 8a: Unemployment rates by race from January to September 2020. Figure by author based on BLS data.
Figure 8a: Unemployment rates by race from January to September 2020. Figure by author based on BLS data.
Figure 8b: Unemployment rates by race from January to September 2020. Figure by author based on BLS data.
Figure 8b: Unemployment rates by race from January to September 2020. Figure by author based on BLS data.

Between March and April 2020, the unemployment rate for all races combined tripled and all racial groups experienced rapid increases in unemployment. The high spike in the unemployment rate during that time period is corroborated by other evidence, including drops in electricity usage, changes in mobility, and 30 million new unemployment insurance claims in the U.S. six weeks after the pandemic’s start [3]. However, people of color, especially those of American Indian and Alaskan Native, Hawaiian and Pacific Islander, Hispanic, and Black descent, experienced greater changes in their unemployment rates than their white counterparts following the first coronavirus lockdowns in March. Of the six racial populations sampled, only the White population managed to stay under the unemployment rate for all races after April 2020, and is the only racial group to have an unemployment rate under 7% as of September 2020 (6.25%, specifically). On the other hand, Blacks, Hispanics, and Asians experienced unemployment rates of 11.38%, 9.61%, and 9.36% respectively as of September 2020. Overall, the data suggest that unemployment due to the coronavirus pandemic disproportionately impacted people of color during the time period of interest.

Answering Question 6 – 12-month Household Incomes During the Pandemic: January to September, 2020

Household incomes were aggregated along racial lines and percentages of each racial group within each income bracket were calculated and graphed (Figures 9a-g).

Figure 9a: Changes in percent population distribution in 16 income brackets for all races. Figure by author based on BLS data.
Figure 9a: Changes in percent population distribution in 16 income brackets for all races. Figure by author based on BLS data.
Figure 9b: Changes in percent Hispanic population distribution in 16 income brackets. Figure by author based on BLS data.
Figure 9b: Changes in percent Hispanic population distribution in 16 income brackets. Figure by author based on BLS data.
Figure 9c: Changes in percent White population distribution in 16 income brackets. Figure by author based on BLS data.
Figure 9c: Changes in percent White population distribution in 16 income brackets. Figure by author based on BLS data.
Figure 9d: Changes in percent Black population distribution in 16 income brackets. Figure by author based on BLS data.
Figure 9d: Changes in percent Black population distribution in 16 income brackets. Figure by author based on BLS data.
Figure 9e: Changes in percent American Indian and Alaskan Native population distribution in 16 income brackets. Figure by author based on BLS data.
Figure 9e: Changes in percent American Indian and Alaskan Native population distribution in 16 income brackets. Figure by author based on BLS data.
Figure 9f: Changes in percent American Indian and Alaskan Native population distribution in 16 income brackets. Figure by author based on BLS data.
Figure 9f: Changes in percent American Indian and Alaskan Native population distribution in 16 income brackets. Figure by author based on BLS data.
Figure 9g: Changes in percent Hawaiian and Pacific Islander population distribution in 16 income brackets. Figure by author based on BLS data.
Figure 9g: Changes in percent Hawaiian and Pacific Islander population distribution in 16 income brackets. Figure by author based on BLS data.

Despite the rapid increases in unemployment from March to April 2020 (Figure 8b), overall distributions of all races combined in the 16 income brackets used by CPS have remained fairly stable (Figure 9a). Surprisingly, income bracket distributions trended towards higher incomes from March to April 2020, the same period of time in which the unemployment rate for all races tripled. While some of that wage growth may be attributed to the stimulus packages offered by the U.S. federal government (the 2020 stimulus packages were found to have improved U.S. stock market returns [11], for example) and the Paycheck Protection Program, it is important to note that the majority of pandemic-related layoffs were concentrated in relatively low-paying jobs [6].

In fact, the pandemic seems to have worsened the already abysmal racial income gap in the United States. While the proportion of the white population in the highest BLS income bracket tended to vary between 15% and 17% since January (Figure 9c), as of September 2020, the proportion of Black and Hispanic populations in the highest income bracket were at 7.30% and 7.88% percent, respectively, and were on downward trajectories (Figures 9d and 9b). As of September 2020, the only population of color with favorable income statistics was the Asian minority with 27.53% of the group in the highest CPS income bracket.


Conclusions and Importance

Correlations Between Socioeconomic Indicators and Academic Performance of U.S. High School Students, Impact of the COVID-19 Pandemic on Family Incomes and Unemployment Rates Along Racial Lines, and What it Means for the Future

The overall results of this study support the initial hypotheses posed at the beginning of this report. There exist significant correlations between socioeconomic indicators (specifically race, income, and poverty) and academic achievement, and the COVID-19 pandemic has caused considerable income and unemployment disparities along racial lines. These findings conform to previous literature emphasizing the importance of poverty and income in community-level trends as explanations for both the racial and socioeconomic achievement gaps – race and ethnicity often determine a person’s SES (U.S. Census Bureau, 2009). Through the examination of individual respondents, this study offers specific evidence for the existence of an achievement gap between students of different incomes, poverty levels, and racial groups in terms of overall high school honors-weighted GPA, college enrollment, and mathematical ability (Figures 5a-d, 6a-d, and 7a-c). This study also reveals that the economic repercussions of the COVID-19 pandemic disproportionately affect people of color. The great disparities in unemployment among different racial groups are particularly worrisome; after America’s first "lockdown" phase of March and April 2020, all populations of color analyzed in this study experienced unemployment rates higher than those of the White population (Figure 8b). Moreover, the pandemic has also exacerbated the great inequity in household income between White/Asian populations and other people of color (Figures 9a-g). Given the disproportionate socioeconomic changes caused by the pandemic and the observed correlations between a student’s SES and academic performance, it is expected that the COVID-19 pandemic will continue to widen the U.S. racial achievement gap.

Unfortunately, previous longitudinal analyses done with students in the latter half of the twentieth century reveal that the achievement gap has been an issue in American education since at least the 1950s [5]. While this study will not discuss measures to alleviate the economic disparities in the United States, there are several behavioral solutions that could help close the socioeconomic and racial achievement gaps without engaging in extensive socioeconomic reforms. McKenzie argues that in order for students of poverty to conquer academic challenges, "teachers need to develop strong relationships with their students, embody respect in their interactions with students, embed social skills in lessons, promote inclusive classrooms, recognize the signs of poverty, empower students, alter classroom environments, build core skills, provide accurate assessments, and recruit caring and empathetic staff" [9]; the development of social and behavioral skills among students, in particular, has notable positive effects on their academic success. Furthermore, educating teachers and educational leaders to be more aware of their internal racial biases would go a long way towards bridging the racial achievement gap. Hence, while it is important to confront the economic disparities shaping the socioeconomic and racial achievement gaps, embracing social and behavioral reform in the classroom can help mitigate their impacts.

Nevertheless, the COVID-19 pandemic and its serious socioeconomic inequalities represent a dangerous intersection between race, education, economics, and mortality. Given the disparities in the incidence of COVID-19 infection and death among underrepresented racial groups [10], and as socioeconomic disparities continue to widen through the course of the pandemic, the findings of this study suggest a dangerous precedent for the future racial achievement gap in the American education system.


References

[1] Abedi V, Olulana O, Avula V, et al. Racial, Economic, and Health Inequality and COVID-19 Infection in the United States. Journal of Racial and Ethnic Health Disparities. 2020:1. doi:10.1007/s40615–020–00833–4.

[2] Bjorklund-Young A, Plasman JS. Reducing the Achievement Gap: Middle Grades Mathematics Performance and Improvement. Research in Middle Level Education Online. 2020;43(10):25–45. doi:10.1080/19404476.2020.1836467.

[3] Chen S, Igan DO, Pierri N. Tracking the Economic Impact of COVID-19 and Mitigation Policies in Europe and the United States. [Electronic Resource]. International Monetary Fund; 2020. Accessed November 30, 2020. https://www-elibrary-imf-org.libproxy.berkeley.edu/doc/IMF001/29155-9781513549644/29155-9781513549644/Other_formats/Source_PDF/29155-9781513550589.pdf.

[4] Coleman, J. S., Campbell, E. Q., Hobson, C. J., McPartland, J., Mood, A. M., Weinfeld, F. D., et al. (1966). Equality of Educational Opportunity. Washington, DC: Government Printing Office.

[5] Hanushek EA, Peterson PE, Talpey LM, Woessman L, Harvard University P on EP and G. Long-Run Trends in the U.S. SES-Achievement Gap. Program on Education Policy and Governance Working Papers Series. PEPG 20–01. Program on Education Policy and Governance; 2020. Accessed December 4, 2020. https://files.eric.ed.gov/fulltext/ED606001.pdf.

[6] Industry Today. Average Wages During The Coronavirus Pandemic – Industry Today. [Electronic Resource]. 2020. Accessed November 30, 2020. https://industrytoday.com/average-wages-during-the-coronavirus-pandemic/.

[7] Isaacs J, Magnuson K. Income and Education as Predictors of Children’s School Readiness. TheSocial Genome Project. [Electronic Resource].; 2011. Accessed November 29, 2020. https://search.ebscohost.com/login.aspx?direct=true&db=cat04202a&AN=ucb.b22512360&site=eds-live.

[8] Johnson AD. Implicit Bias of Education Leaders and the Achievement Gap between Black and White Students in Suffolk and Nassau County, New York High Schools. ProQuest LLC. January 2018. Accessed November 30, 2020. https://www-proquest-com.libproxy.berkeley.edu/docview/2054025669.

[9] McKenzie K. The Effects of Poverty on Academic Achievement. BU Journal of Graduate Studies in Education. 2019;11(2):21{26. Accessed December 4, 2020. https://files.eric.ed.gov/fulltext/EJ1230212.pdf.

[10] Moore JT, Ricaldi JN, Rose CE, et al. Disparities in Incidence of COVID-19 Among Underrepresented Racial/Ethnic Groups in Counties Identified as Hotspots During June 5–18, 2020–22 States, February-June 2020. MMWR: Morbidity & Mortality Weekly Report. 2020;69(33):1122–1126. doi:10.15585/mmwr.mm6933e1.

[11] Narayan PK, Phan DHB, Liu G. COVID-19 lockdowns, stimulus packages, travel bans, and stock returns. Finance Research Letters. January 2020. doi:10.1016/j.frl.2020.101732.

[12] Wages M. The Achievement Gap: A Poverty Crisis, Not an Education Crisis. Rowman & Littlefield Publishers; 2018. Accessed November 30, 2020. https://eds-b-ebscohost-com.libproxy.berkeley.edu/eds/ebookviewer/ebook/bmxlYmtfXzE4Nzk2NDFfX0FO0?sid=f944fb9e-8e9b-479e-8b7f-14013ad9e644@pdc-v-sessmgr03&vid=0&format=EB&rid=1.

[13] Whipple SS, Genero CK, Evans GW. Task Persistence: A Potential Mediator of the Income-Achievement Gap. Journal of Applied Research on Children. 2016;7(1). Accessed December 5, 2020. https://files.eric.ed.gov/fulltext/EJ1188422.pdf.

[14] See https://nces.ed.gov/OnlineCodebook/Session/Codebook and https://www2.census.gov/programs-surveys/cps/datasets/2020/basic/2020_Basic_CPS_Public_Use_Record_Layout_plus_IO_Code_list.txt for detailed variable lists, variable explanations, and documentation for the HSLS and CPS, respectively.

[15] Downloads are available at https://nces.ed.gov/OnlineCodebook/Session/Codebook/462c2ac2-6f64-47ec-af42-9a80a2aa6f72 and additional documentation for the base year and first follow-up can be found here: https://nces.ed.gov/pubs2014/2014361.pdf.

[16] For the source code, data frames, tables, and additional visualizations, please visit the following Github repository: https://github.com/adamzuyang/Ursatech-Adam.


Related Articles