Table 1: 1960 Summary
=======================================================================================
Statistic N Mean St. Dev. Min Median Max
---------------------------------------------------------------------------------------
Absolute Income Mobility 50 0.639 0.053 0.519 0.644 0.760
High School Attainment % 50 41.722 7.333 27.600 42.050 55.800
Proportion of Wealth in Top 1% 50 0.197 0.045 0.108 0.194 0.320
Public School Expenditure (unadjusted dollars) 50 769.180 159.567 501 734 1,327
---------------------------------------------------------------------------------------
Table 2: 1970 Summary
======================================================================
Statistic N Mean St. Dev. Min Median Max
----------------------------------------------------------------------
Absolute Income Mobility 43 0.613 0.048 0.450 0.615 0.701
High School Attainment % 43 52.300 8.165 37.800 52.800 67.300
Proportion of Wealth in Top 1% 43 0.219 0.014 0.193 0.222 0.247
----------------------------------------------------------------------
Table 3: 1980 Summary
======================================================================
Statistic N Mean St. Dev. Min Median Max
----------------------------------------------------------------------
Absolute Income Mobility 50 0.507 0.043 0.379 0.505 0.623
High School Attainment % 50 67.468 7.570 53.100 67.950 82.500
Proportion of Wealth in Top 1% 50 0.196 0.011 0.180 0.197 0.218
----------------------------------------------------------------------Application of Great Gatsby Curve to United States Education
Objective
The Great Gatsby Curve is a recently studied curve tracking the relationship between inequality and economic mobility. Countries with higher levels of inequality, measured in the Gini Coefficient, tend to have lower economic mobility, crippling the later generations. The aim of this study is to apply the Great Gatsby Curve to just the United States to see if the relationship stays the same, looking through the scope of education. The models used will test if education attainment and spending has any significant effect to the economic mobility. Applications beyond this will cover the validity of the American Dream, already studied by Raj Chetty.
Introduction
Land of Opportunities
The Great Gatsby Curve has already been studied in the past, both focused in the United States and globally. It centers around higher Gini Coefficients that measure income inequality having threatened economic mobility. It plots several countries on one graph, indicating the success of some countries compared to others based on their inequality. The United States is expected to have higher intergenerational mobility, becoming the ‘land of opportunity’ with both relative and absolute upward mobility (Chetty, Hendren, Kline, and Saez 2014). The reality is that it sits near the center of the curve, begging the question of whether it brings enough opportunity. In several trends, inequality seems to be both brought upwards and downwards, evening out mobility to the center of the curve. The upbringing of children is one of the main explanations for the variation of this intergenerational income mobility, especially shown in college attendance and teenage birth rates. Over generations, almost all families in the United States have opportunities to become educated, no matter how wealthy they are. However, with the larger access to higher education from individuals whose parents already were supported by upward mobility, inequality may seem to spread when compared to the amount of individuals whose parents could not be affected by this mobility. Factors including race, inequality, education quality, social capital indices, and family structures all play into how much intergenerational mobility changes every year.
Educational Inequalities
Specifically for education, it is expected that individuals with experience larger socio-economic investments from their families tend to score higher PISA scores along with having better future social mobility (Blanden, Doepke, and Stuhler 2022). However, a divergence in the 1960s showed a new trend where despite larger socio-economic gaps being present, educational inequality decreased. On the other hand, this difference took the form of an increase in income inequality (Bennett 2011, 13-14). This data challenges my study, indicating that higher levels of educational attainment are negatively correlated with income inequality, indicating missing independent variables in my study at first glance. This discrepancy is nothing new, concluding with how there is no clear consensus over whether the distribution of education actually affects income inequality. While a poverty trap can be made with families not being able to afford higher education to escape poverty, the increase in overall earning opportunity for the lower income families suggest a smaller gap in income inequality (Checchi 2001, 5). Although an individual’s upbringing greatly affects their earning potential with education, a large amount of variation must also be attributed to the other factors stated from Chetty, Hendren, Kline, and Saez along with other factors regarding labor inequality and fiscal policies.
Methodology
Because Gini Coefficients measure a whole country’s income inequality, this study uses a different metric being the proportion of wealth within the top 1% of the population. This will act as the main independent variable. The data will come from a study from Suss, Connor, and Kemney, titled GEOWEALTH-US: Spatial wealth inequality data for the United States, 1960-2020. To make factor in education, high school attainment rates from the US Census Bureau will be a control variable. Additionally, per pupil spending for education from Table 194 of the Digest of Education Statistics will also be a control variable and completes the robust model. The dependent variable will be measured in absolute income mobility from Raj Chetty’s study, The Fading American Dream: Trends in Absolute Income Mobility Since 1940.
Original Models
For the original regressions, only the cohort years 1960, 1970, and 1980 will be analyzed due to data availability. Later in this study, the robust model will only be created for the 1960 cohort for accuracy purposes.
Below are the summary statistics for each cohort’s variables:
The models in question will abide by the following regressions, moving from original to robust:
Absolute Income Mobilityi = β1 + β2 * Proportion of Wealth in Top 1%i + ei
Absolute Income Mobilityi = β1 + β2 * Proportion of Wealth in Top 1%i + β3 * EducationAttainedi + ei
Absolute Income Mobilityi = β1 + β2 * Proportion of Wealth in Top 1%i + β3 * EducationAttainedi + β4 * PerPupilSpendingi+ ei
For the first two models, it seems that the year 1960 is the most accurate with how its significance holds at a higher confidence level than the others, as shown below:
Table 4: Summary of original models
=================================================================================================================
Dependent variable:
--------------------------------------------------------------------------------
`Absolute Income Mobility` `Absolute Income Mobility` `Absolute Income Mobility`
1960 1970 1980
(1) (2) (3)
-----------------------------------------------------------------------------------------------------------------
`Proportion of Wealth in Top 1%` -0.354
t = -2.189**
`Proportion of Wealth in Top 1%` 0.066
t = 0.122
`Proportion of Wealth in Top 1%` -0.571
t = -1.017
Constant 0.709 0.599 0.619
t = 21.711*** t = 5.068*** t = 5.609***
-----------------------------------------------------------------------------------------------------------------
Observations 50 43 50
R2 0.091 0.0004 0.021
Adjusted R2 0.072 -0.024 0.001
Residual Std. Error 0.051 (df = 48) 0.048 (df = 41) 0.043 (df = 48)
F Statistic 4.790** (df = 1; 48) 0.015 (df = 1; 41) 1.034 (df = 1; 48)
=================================================================================================================
Note: *p<0.1; **p<0.05; ***p<0.01
Table 5: Summary of second models
=================================================================================================================
Dependent variable:
--------------------------------------------------------------------------------
`Absolute Income Mobility` `Absolute Income Mobility` `Absolute Income Mobility`
1960 1970 1980
(1) (2) (3)
-----------------------------------------------------------------------------------------------------------------
`High School Attainment %` -0.003
t = -2.950***
`Proportion of Wealth in Top 1%` -0.414
t = -2.734***
`High School Attainment %` -0.002
t = -1.926*
`Proportion of Wealth in Top 1%` -0.033
t = -0.063
`High School Attainment %` -0.003
t = -2.891***
`Proportion of Wealth in Top 1%` -1.479
t = -2.425**
Constant 0.835 0.710 0.968
t = 15.880*** t = 5.540*** t = 6.108***
-----------------------------------------------------------------------------------------------------------------
Observations 50 43 50
R2 0.233 0.085 0.169
Adjusted R2 0.200 0.039 0.133
Residual Std. Error 0.048 (df = 47) 0.047 (df = 40) 0.040 (df = 47)
F Statistic 7.132*** (df = 2; 47) 1.862 (df = 2; 40) 4.774** (df = 2; 47)
=================================================================================================================
Note: *p<0.1; **p<0.05; ***p<0.01To check the functional form of each regression, a RESET was tested for each cohort’s first two models. Below are the results of the test at the 95% confidence level. The tests will be ordered as such: 1960-1, 1960-2, 1970-1, 1970-2, 1980-1, 1980-2.
RESET test
data: lm(data1960$`Absolute Income Mobility` ~ data1960$`Proportion of Wealth in Top 1%`, na.action = na.exclude)
RESET = 7.9973, df1 = 2, df2 = 46, p-value = 0.001045
RESET test
data: lm(data1960$`Absolute Income Mobility` ~ data1960$`Proportion of Wealth in Top 1%` + data1960$`High School Attainment %`, na.action = na.exclude)
RESET = 7.7518, df1 = 2, df2 = 45, p-value = 0.00128
RESET test
data: lm(data1970$`Absolute Income Mobility` ~ data1970$`Proportion of Wealth in Top 1%`, na.action = na.exclude)
RESET = 0.19155, df1 = 2, df2 = 39, p-value = 0.8265
RESET test
data: lm(data1980$`Absolute Income Mobility` ~ data1980$`Proportion of Wealth in Top 1%`, na.action = na.exclude)
RESET = 1.6673, df1 = 2, df2 = 46, p-value = 0.2
RESET test
data: lm(data1980$`Absolute Income Mobility` ~ data1980$`Proportion of Wealth in Top 1%` + data1980$`High School Attainment %`, na.action = na.exclude)
RESET = 3.4286, df1 = 2, df2 = 45, p-value = 0.04112It should be noted that 1960 had consistent issues with functional form, 1970 had no issues, and 1980 had inconsistent issues. In this study, we will rework the 1960 model since it has the most consistent data. While 1970 has consistency with its functional form, its relationship is not accurate enough, given its positive relationship, as seen in Table 4 and the graph below.
Robust Model
The robust model will add another independent variable, per pupil spending. We see that with the 1960 cohort, the wealth inequality and high school attainment maintain their significance at the 99% confidence level. In addition, the per pupil spending becomes significant at the 90% confidence level.
Table 6: Summary of robust model
============================================================================
Dependent variable:
---------------------------
`Absolute Income Mobility`
----------------------------------------------------------------------------
`High School Attainment %` -0.004
t = -3.432***
`Proportion of Wealth in Top 1%` -0.573
t = -3.257***
`Public School Expenditure (unadjusted dollars)` 0.0001
t = 1.687*
Constant 0.840
t = 16.250***
----------------------------------------------------------------------------
Observations 50
R2 0.278
Adjusted R2 0.230
Residual Std. Error 0.047 (df = 46)
F Statistic 5.891*** (df = 3; 46)
============================================================================
Note: *p<0.1; **p<0.05; ***p<0.01Upon testing the robust model, it appears that adding the last variable of public school expenditure fixes multiple issues within the original regression. Based on the results below, any functional form issues have been cleared, along with any issues revolving heteroscedasticity.
RESET test
data: model1960r
RESET = 2.5133, df1 = 2, df2 = 44, p-value = 0.09256
studentized Breusch-Pagan test
data: model1960r
BP = 6.0435, df = 3, p-value = 0.1095Furthermore, the robust model does not violate multicollinearity, suggesting that there is not a strong correlation between wealth inequality and education:
data1960$`High School Attainment %`
1.515869
data1960$`Proportion of Wealth in Top 1%`
1.430674
data1960$`Public School Expenditure (unadjusted dollars)`
1.787734 This denotes a key difference between wealth inequality and income inequality from the Gini Coefficient in the typical Great Gatsby Curve. Because the type of inequality tested does not possess enough similarity to income inequality, the Great Gatsby Curve in this study is not true to the actual, explaining the difference as a negative relationship between inequality and absolute income mobility.
Findings
Based on the results, the relationship is still correlational rather than causal. While the inverse relationship between inequality and economic mobility is captured, the R2 being so low indicates that our model does not explain enough about the economic mobility in the United States. Other variables or controls that may have been able to affect the true impact are GDP, median household income, school quality, and segregation or other types of inequalities other than wealth and education. Once these independents are added, omitted variable bias will be resolved, yielding larger coefficients across all other variables.
Conclusion
This study finds that there is indeed a negative relationship between wealth inequality and economic mobility, honed in through the scope of education. While the correlation may not be particularly strong, it does suggest that the American Dream is possible. This is indicated from lower inequality states with higher attainment rates of education having more wealth towards the upcoming generations. By closing barriers to higher education, such as lowering tuition or utilizing tactics for diversity, equity, and inclusion, the United States will be able to sustain future generations with higher wealth.
References
Bennett, Daniel. Educational Inequality in the United States: Methodology and Historical Estimates of Education Gini Coefficients. Center for College Affordability and Productivity, 2011. https://ssrn.com/abstract=2134646
Blanden, Jo, Matthias Doepke, and Jan Stuhler. Unequal education and the ‘Great Gatsby Curve’. Centre for Economic Policy Research, 2022. https://cepr.org/voxeu/columns/unequal-education-and-great-gatsby-curve
Checchi, Daniele. Education,inequality and income inequality. Distributional Analysis Research Programme, 2001. https://researchonline.lse.ac.uk/id/eprint/6566
Chetty, Raj, Nathaniel Hendren, David Grusky, Maximilian Hell, Robert Manduca, and Jimmy Narang. The Fading American Dream: Trends in Absolute Income Mobility Since 1940. National Bureau of Economic Research, 2016. https://opportunityinsights.org/paper/the-fading-american-dream/
Chetty, Raj, Nathan Hendren, Patrick Kline, and Emmanuel Saez. Where is the land of opportunity? Intergenerational mobility in the US. Centre for Economic Policy Research, 2014. https://cepr.org/voxeu/columns/where-land-opportunity-intergenerational-mobility-us
Connor, Dylan, Tom Kemeny, and Joel Suss. GEOWEALTH-US: Spatial wealth inequality data for the United States, 1960-2020. Inter-university Consortium for Political and Social Research, 2024. https://doi.org/10.3886/E192306V4
World Bank Group. World Development Indicators. https://databank.worldbank.org/source/world-development-indicators#