Public Policy, Higher Education, and Income ...

7 downloads 253677 Views 1MB Size Report
policies intended to make college more affordable by reducing credit constraints or ..... period, but 15 states (Alaska, California, Colorado, Connecticut, Hawaii, ...
Public Policy, Higher Education, and Income Inequality in the United States: Have We Reached Diminishing Returns? Daniel L. Bennett1 Richard K. Vedder2

Abstract Public policy designed to promote greater college enrollment rates has often been justified as a means to reduce income inequalities, yet there is very little evidence that higher college attainment is associated with less inequality. Economic theory at best suggests that the relationship between college attainment and inequality is ambiguous. An overview of some of the unintended consequences of public policies designed to promote greater enrollment is described. One such consequence is that the growth in college completion, which is at least partially attributable to public policy, may have actually contributed to rising income inequality. We hypothesize the existence of a U-shaped relationship between college attainment and income inequality, and using panel data for the 50 U.S. states over the period 1970-2004 to provide empirical evidence in support of the curve. Prior to the mid 1990’s, increases in attainment were associated with less inequality for most states. Rapid growth in attainment since then has moved most states to the right of the inflection point such that attainment gains are associated with more inequality in most states.

Working paper version. Please cite as: Bennett, Daniel L. and Richard K. Vedder (2015). Public Policy, Higher Education, and Income Inequality in the United States: Have We Reached Diminishing Returns? Social Philosophy and Policy, 31(2): 252-280.

1

Daniel L. Bennett is Assistant Professor of Economics at Patrick Henry College. The majority of this research was completed while he was a graduate student in the Department of Economics at Florida State University. 2 Richard Vedder is Director of the Center for College Affordability and Productivity, Distinguished Professor of Economics Emeritus, Ohio University, and an Adjunct Scholar at the American Enterprise Institute.

I. Introduction An integral part of American exceptionalism has been the fact that persons from all walks of life could succeed. Barriers to intergenerational income mobility have historically been perceived to be small. Observers beginning famously with Alexis de Tocqueville and continuing to modern times have spoken of American egalitarianism, manifesting itself in what is often called the American Dream. With the passage of time, Americans increasingly saw the completion of a good education as being very useful, if not an absolute prerequisite, to the pursuit of economic success. That manifested itself with the common school movement associated with Horace Mann and others, and, with respect to higher education by government involvement with universities, an effort present since colonial times that increased substantially after the passage of the Morrill Act in 1862 that granted federal lands to the states for the purpose of establishing public universities.3 Nonetheless, higher education still remained open only to a small percentage of the population, and the most financially successful men in America before World War II were largely not college graduates - Cornelius Vanderbilt, John D. Rockefeller, Andrew Carnegie, J.P. Morgan, and Henry Ford are good examples of fabulously wealthy persons with no college background, some of whom (e.g. Rockefeller and Vanderbilt) came from relatively humble family backgrounds. The GI Bill of 1944 and a frenetic increase in the number and size of universities in the postwar era, however, led a marked expansion in the number of college graduates, and the development of the view that a college education was a prerequisite for economic success. Beginning with the Higher Education Act of 1965, taxpayer-funded grants and loans became available to low and middle income college students, helping to further promulgate the idea that college is the key to social mobility.4 While as late as 1970, only about 10 percent of adult Americans had bachelor’s degrees or more, that proportion roughly tripled over the next four decades to around 30 percent, as depicted in Figure 1. This rapid rise in college attainment is attributable to a number of factors, including public policies intended to promote greater

3

Daniel L. Bennett, “Myth busting: The Laissez Faire Origins of American higher education,” The Independent Review 18, no. 4 (2014): 503-526. 4 See Scott Winship’s article “Has Rising Income Inequality Worsened Inequality of Opportunity” in this issue of Social Philosophy & Policy for a more thorough discussion of the relationship between income inequality and economic mobility.

attendance, a rising wage differential between college and high school educated workers, and a growing perception that a college degree is necessary for economic success. It has become an article of faith that higher education is a major vehicle for promoting a path to the middle class and income equality in America. Although those promoting greater government in financing college had done so for many reasons, providing a taxpayer-funded ticket to a middle class, prosperous life to those who cannot afford the cost of attending college as a means to reduce inequality has been a persistent and oft-repeated argument. Both sides of the political aisle have been supportive of these endeavors. For instance, more than 85 percent of the House of Representatives voted in favor of the 1965 Higher Education Act, including 83 percent of the minority Republican members. The bill also easily passed the Senate with 82 percent of the members voting in favor.5 The highest level policymakers in the U.S. government have made proclamations of this nature recently in advocating for greater government support of higher education to promote increased attendance. Speaking at Harvard in 2008, the Chairman of the Federal Reserve Board Ben Bernanke said that “the best way to improve economic opportunity and to reduce inequality is to increase the educational attainment and skills of American workers."6 In pushing to further increase the proportion of adults with college degrees, President Barack Obama, in his 2009 State of the Union address, argued that this “will open the doors of opportunity for our children."7 The Obama Administration reiterated similar claims in its August 2013 higher education policy agenda by declaring that a "college education continues to be the ticket to the middle class."8 In this paper, we explore the economic theory and empirical evidence on the relationship between college attainment and income inequality in the U.S., finding that the theoretical relationship is ambiguous and empirical evidence seems to support the existence of a U-shaped inequality-attainment curve. Next is a discussion of related theory, followed by empirical 5

Govtrak, “House Vote #126 in 1965” and “Senate Vote #215 in 1965,” retrieved December 25, 2013 from http://www.govtrack.us/congress/votes/89-1965/h126 and https://www.govtrack.us/congress/votes/891965/s215, respectively. 6 Quoted in William G., Bowen, Matthew M. Chingos, and Michael S. McPherson, Crossing the Finish Line: Completing College and America’s Public Universities (Princeton, N.J.: Princeton University Press, 2009): 1. 7 Barack Obama, “State of the Union Address: February 24, 2009,” retrieved March 27, 2012 from http://www.whitehouse.gov/the\_press\_office/Remarks-of-President-Barack-Obama-Address-to-Joint-Sessionof-Congress 8 White House Office Hours, “College Affordability,” August 21, 2013, http://www.whitehouse.gov/blog/2013/08/21/white-house-office-hours-college-affordability

evidence in section 3. The penultimate section offers some policy implications and section 5 concludes.

II. Economic Theory Advocates for a greater government role in promoting college attendance have often argued that a college education provides greater economic opportunities, and that by reducing inequality of educational opportunity, greater equality of income will follow.9 This implies that there is an inverse relationship between inequality of income and the proportion of the population with college degrees. If greater numbers of persons go to college, it is reasoned, the educational and skill advantages previously largely the domain of the richer members of society will expand to a larger segment of the population. With greater educational equality will follow greater economic equality, manifested in a reduced dispersion in the distribution of income. Policymakers and other public intellectuals espousing such views seem to be echoing the assertions of some academics who have addressed this issue. The most prominent and comprehensive affirmation come from Claudia Goldin and Lawrence Katz, who observe that income inequality has risen sharply in the U.S. since 1980, and that this was largely explainable by increasing inequality in the distribution of work-related income.10 Economists have often explained this rise in wage inequality as an outcome of skills-biased technological change, hereafter SBTC. The SBTC theory suggests that the incomes of workers with high levels of technological skills rise disproportionately from technological change relative to low skill workers, acting to increase inequality.11 Some authors, including Goldin and Katz, contend that the sharp rise in the income differential between high school and college graduates is attributable to SBTC, suggesting that college graduates learn skills that make them better equipped to take advantage of technological change than high school graduates. The implication is that the college-high school income differential is substantially a consequence of inadequate growth in college degree attainment, and that faster growth in college attainment would have spawned greater income equality since more 9

See e.g. Arthur M. Okun, Equality and Efficiency: The Big Tradeoff (Washington D.C.: The Brookings Institution, 1975). 10 Claudia Goldin and Lawrence F. Katz, The Race Between Education and Technology, (Cambridge, MA: Harvard University Press, 2008). 11 For a review of this literature, see: Daron Acemoglu, “Technical change, inequality, and the labor market," Journal of Economic Literature, 40 (2002): 7-72.

workers would have been equipped to adapt to technological change that emerged beginning in the 1980s. This argument has been accepted relatively uncritically by several other authors of significant works relating to the economics of higher education in promoting greater equality through an expansion of government finance of higher education.12 While SBTC as an explanation for rising inequality may be considered conventional wisdom, a few economists have nonetheless challenged the view that college education imparts the skills necessary to adapt to technological change, showing that rises in educational attainment can lead to greater inequality when education is used by employers as a signaling mechanism. Paul Krugman, for instance, notes that this “could easily be misinterpreted as exogenous skillbiased technological change."13 Building on the screening/signaling theory of Michael Spence,14 several authors have constructed models to show that as the number of high ability workers earning a college education rises, the number of high ability workers in the uneducated pool is reduced. Given that wages equal the average expected marginal productivity in equilibrium and education is the only signal of productivity, the average wage level of the uneducated pool declines with a rise in the attainment level. Meanwhile, the pool of educated workers rises along with their average wage level. Such a process would lead to a widening wage differential between the two groups, leading to an increase in inequality. In discussing education as a screening mechanism, Joseph Stiglitz suggested that it “tends to increase inequality."15 Krugman notes: “Anything that encourages good workers to get educated can set in motion a cumulative process of growing inequality."16 Igal Hendel, Joel Shapiro, and Paul Willen show how this process works through

12

William G. Bowen, Matthew M. Chingos, and Michael S. McPherson, Crossing the Finish Line: Completing College and America’s Public Universities (Princeton, N.J.: Princeton University Press. 2009). Robert Archibald and David H. Feldman, Why Does College Cost So Much? (New York: Oxford University Press, 2011). 13 Paul Krugman, “And Now for Something Completely Different: An Alternative Model of Trade, Education, and Inequality," in R.C. Feenstra, ed., The Impact of International Trade on Wages (University of Chicago Press, 2000), 16. 14 A. Michael Spence, “Job Market Signaling," Quarterly Journal of Economics 87, no. 3 (1973): 355-374. 15 Joseph E. Stiglitz, “The Theory of Screening, Education, and the Distribution of Income,” The American Economic Review 65, no. 3 (1975): 299. 16 Paul Krugman, “And Now for Something Completely Different: An Alternative Model of Trade, Education, and Inequality," in R.C. Feenstra, ed., The Impact of International Trade on Wages (University of Chicago Press, 2000), 28.

policies intended to make college more affordable by reducing credit constraints or subsidizing tuition.17 J.B. Knight and R.H. Sabot argue that the overall effect of an increase in educational attainment on inequality depends on the relative strength of the composition and wage compression effects.18 The former is the change in inequality resulting from a change in the composition of the educational composition of the labor force evaluated at the original educational wage structure, while the latter is the change in inequality resulting from compression of the wage structure evaluated at the original composition. As the relative size of the educated group grows, the compression effect unambiguously generates greater equality while the composition effect initially results in greater inequality, but upon reaching a critical attainment level, additional increases in the attainment rate reduce inequality. The critical attainment rate is higher the lower the education wage premium.19 The human capital model of income distribution suggests that an increase in average attainment has an ambiguous effect on inequality, depending on the evolution of rates of return to education.20 Most economic theory related to education and inequality has focused on income differences between education groups. Caroline Hoxby and Bridget Terry find however that income inequality among college graduates has also grown over time, suggesting that both within and between education group distributional differences contribute to rising income inequality.21 The signaling model of Andreas Bergh and Günther Fink, who show that the emergence of private institutions as a means of providing a costly elite signal of ability has an ambiguous effect on inequality as the share of educated workers rises, provides one theoretical avenue through which within-group inequality could rise with the attainment rate if aggregate

17

Igal Hendel, Joel Shapiro, and Paul Willen, “Educational Opportunity and Income Inequality," Journal of Public Economics 89 (2005): 841-870. 18 J.B. Knight and R.H. Sabot, “Education Expansion and the Kuznets Effect," American Economic Review 73, no. 5 (1983): 1132-1136. 19 Sherman Robinson, “A Note on the U Hypothesis Relating Income inequality and Economic Development," American Economic Review 66, no. 3 (1976): 437-40. 20 José De Gregorio and Jong-Wha Lee, `”Education and Inequality: New evidence from Cross-Country Data," Review of Income and Wealth 48, no. 3 (2002): 395-416. 21 Caroline M Hoxby and Bridget Terry, “Explaining Rising Income and Wage Inequality among the CollegeEducated," NBER Working Paper 6873 (January 1999).

attainment rates do not account for heterogeneity of the education signal.22 This additional dynamic further confounds how increases in college attainment will affect inequality. Thus is not clear a priori what the effect of increasing the college attainment rate will have on income inequality, and empirical evidence has also been mixed.23 Nor is it evident that the relationship between the two is linear, as it is possible that the marginal effects depend on the current level of attainment, composition of the labor force and wage structure, and other education-related factors contributing to heterogeneous incomes.24 Next we discuss basic economic principles and some rather fundamental observations about human behavior that collectively suggest that in small to medium-sized doses, the notion that greater college attainment promotes income equality is broadly true, but it is clearly untrue when higher education expands beyond a certain point. Pictorially, what may exist is something more like a U-shaped curve that contains both a negatively and positive sloped relationship between higher education attainment and income inequality. This is very similar to a famous graph in economics, the Laffer Curve, which argues that raising tax rates increases tax revenues for a while, but at some point further rate increases become high enough that taxpayers substitute away from production and labor to leisure, shrinking the tax base and reducing tax revenues. But does a U-shaped inequality-college attainment curve make any sense on broader theoretical grounds? Two basic behavioral realities suggest that the non-linear, parabolic nature of the equality-higher education relationship makes sense. First, there is a huge gap between the rhetoric of American egalitarianism found in the Declaration of Independence (“all men are created equal”) and elsewhere, and the reality with respect to the innate and acquired abilities of individuals to be economically productive and thus derive income. Cognitive abilities, as measured by IQ tests and other widely used aptitude exams such as the SAT, GRE and Armed Forces Qualification Test, exhibit a great amount of variation. With economic growth, demand has risen sharply for those possessing cognitive skills relative to those whose “comparative 22

Andreas Bergh and Günther Fink, “Higher Education, Elite Institutions and Inequality,” European Economic Review 53 (2009): 376-384. 23 See e.g. Rati Ram, “Can Educational Expansion Reduce Income Inequality in Less-Developed Countries?” Economics of Education Review 8, no. 2(1989):185-195. José De Gregorio and Jong-Wha Lee, `”Education and Inequality: New evidence from Cross-Country Data," Review of Income and Wealth 48, no. 3 (2002): 395-416. 24 See e.g. Andreas Bergh and Günther Fink, “Higher Education Policy, Enrollment, and Income Inequality," Social Science Quarterly 89, no. 1 (2008): 217-235. Andreas Bergh and Günther Fink, “Higher Education, Elite Institutions and Inequality,” European Economic Review 53 (2009): 376-384.

advantage” derives from other skills, especially those requiring physical strength, endurance, or dexterity. Related to this, the ability of students to successfully master higher order forms of knowledge or critical thinking skills, such as those requiring advanced mathematics, is highly dependent on cognitive ability. This point ties in with the second behavioral reality: the Law of Diminishing Returns. When university attendance is low, say five percent of the 18 to 22 year population, a doubling of that attendance to 10 percent will likely lead many bright persons (with high cognitive skills) to go to college that previously were deprived of that opportunity. Even when 10 percent of the relevant cohort is attending college, most college students are likely to be from the top quartile of the distribution of cognitive skills, and the average IQ of college students may be perhaps 120 – well above the national average of 100. The incremental students benefit hugely from a college education and they generally fulfill relatively high paying jobs requiring university training. When attendance is already very high and a similar 5 percent increase is experienced, from say, 50 percent of the college aged cohort to 55 percent, by mathematical necessity the number of students admitted of average and even below average levels of cognitive skills rises. The incremental students almost certainly have average IQs of 100 or lower. They are generally far less capable of mastering higher order skills requiring intellectual rigor. Universities are no longer dominated by, relatively speaking, the cognitively elite. The incremental students no longer generally get high paying jobs that will advance them, and their families, up the economic ladder. To deal with this situation and avoid massive dropouts, universities offer remedial instruction, lower their grading standards, and water down the curriculum. Students, aware of their academic limitations, increasingly avoid “analytically challenging” subjects such as physics, mathematics, or mechanical engineering, and major in less rigorous subjects such as communication, education, marketing, or recreational management. Empirical evidence supports this. There is pretty solid evidence that the typical grade point average of undergraduate students has risen from about 2.5 (on a 4 point scale) around 1950 to over 3.0 today, despite the vast increase in the number of students, which almost inevitably required some lowering in the average cognitive abilities of the college age cohort. 25

25

Stuart Rojstaczer and Christopher Healy, “Where A Is Ordinary: The Evolution of American College and University Grading, 1940-2009,” Teachers College Record 114, no. 7 (2012): 1-23.

Similarly, the proportion of college students majoring in academically challenging and demanding fields such as the STEM disciplines –science, technology, engineering and mathematics, has declined. For example, since the mid-1960s, the proportion of college freshmen indicating that they intended to pursue a career in scientific research has fallen by about 50 percent.26 Still other evidence suggests that recent college students spend less time on academic pursuits than students did half a century ago – on average, less qualified students work less but receive much higher grades.27 We would add that personality traits other than declining average levels of cognitive skills similarly show changes that point in the direction of less rigorous academic preparation. Human beings vary widely in their self- discipline, in their ambition, and in their interest in promoting long-run potential gains to their utility (satisfaction) as opposed to very short term forms of satisfaction. When college attendance is low, one would suspect that those attending college on average have relatively high quantities of such non-cognitive attributes as ambition and a strong work ethic, but as enrollments expand to encompass a majority of the population, that is distinctly less so, particularly as the curricula become less rigorous and require less effort on behalf of the student. As the proportion of adults with a college degree rises, college completion becomes a less reliable screening device because the ability of employers to predict the likely productivity attributes of college graduates declines as those graduates become less distinctive from the general population. To overcome this challenge, employers begin to use additional criteria to screen applications such as reputation of the university attended and possession of an advanced graduate degree. Students attending elite private colleges or pursuing advanced degrees are likely to be brighter, more industrious, etc., than those attending non-selective state institutions or those with a terminal bachelor degree. In response, students increasingly pursue admittance to elite private colleges and/or graduate education as a means to enhance their signal to prospective employers. A good deal of evidence suggests that the perceived quality gap between typical state universities and the elite private institutions has widened a good deal. For example, in 1988, eight of the top 25 national universities in the US News & Report rankings were public 26

Derek Bok, American Higher Education Today, (Princeton, N.J.: Princeton University Press, 2013). Philip S. Babcock and Mindy Marks, “The Falling Time Cost of College: Evidence from Half a Century of Time Use Data,” Review of Economics and Statistics 93, no. 2 (2011): 468-78. 27

universities; today, only three are public. Resources per student have grown faster for the private schools, particularly as state appropriation growth for public universities has slowed significantly from earlier generations. Recipients of Pell Grants are typically from relatively low income families, and government statistics show that the proportion of Pell Grant recipients at prestigious private schools is dramatically lower than at state schools. For example, at a sample of 25 elite private schools that we examined (Harvard, Yale, Princeton, Dartmouth, Columbia, Pennsylvania, Cornell, Brown, Williams, Swarthmore, Pomona, Amherst, M.I.T., Stanford, Duke, Chicago, Northwestern, Notre Dame, Emory, Vanderbilt, Cal Tech, Washington University in St. Louis, Bowdoin, Washington and Lee, and Wellesley), on average 15 percent of their undergraduates received Pell Grants (with a range between 6 and 23 percent). At 25 distinctly less prestigious state universities (schools ranked below 500 among the 650 institutions included in the 2013 Forbes rankings) surveyed, by contrast, the average proportion of students on Pell Grants was 40 percent, with a range between 22 and 76 percent. Thus, as enrollments grow and merely having a college diploma means less, students from upper income families disproportionately circumvent that problem by attending elite schools still known for having bright, highly motivated students –that disproportionately get the best jobs. This could lead to greater income inequality, although Bergh and Fink develop a model depicting a similar situation, showing that this process exerts a theoretically ambiguous impact on inequality, depending on the educational composition of the workforce, education premium and relative cost of attending an elite private versus public college.28 The general lowering of the intellectual capital of college graduates that has accompanied expanding enrollments is also verified by some empirical evidence. The Adult Literacy Survey administered roughly once each decade by the U.S. Department of Education has shown declining literacy among college graduates over time. Similarly, the evidence of little growth in cognitive learning skills among college graduates in modern day America was vividly demonstrated recently by Richard Arum and Josipa Roksa.29 Besides the decline in cognitive and other highly productive skills by the average college graduate, another factor is at work. Underemployment by the college educated is on the rise, 28

Andreas Bergh and Günther Fink, “Higher Education, Elite Institutions and Inequality,” European Economic Review 53 (2009): 376-384. 29 Richard Arum and Josipa Roska, Academically Adrift: Limited Learning on College Campuses, (Chicago: The University of Chicago Press, 2010).

aided undoubtedly by some of the aforementioned factors such as grade inflation and a decline in average cognitive ability and work effort among graduates, but also a disconnect between the supply and demand conditions of the labor market. The growth in what might be termed high skilled jobs typically associated with a college education has been much slower than the growth in the supply of college graduates, forcing an increasing proportion of those graduates to take relatively low paying unskilled jobs. As of 2010, 1,048,000 retail sales persons held a bachelor’s degree—nearly one-fourth of the total and five times the proportion prevailing in 1970. Less than one percent of taxi drivers had degrees in 1970, a figure that now exceeds 15 percent. Additionally, more than 115,000 “janitors and cleaners” have four year degrees or more.30 Broader based data support this conclusion. For example, from 2008 to 2011, the college/high school earnings differential for young college graduates (25 to 34 years of age) fell noticeably, following earlier periods of increase.31 As college graduates increasingly staff lower-paying positions that were previously staffed by persons without postsecondary credentials, the value of many college degrees and the career options for those without a college credential is diminished, potentially acting to increase inequality. Thus it is theoretically possible that public policy designed to increase the college attainment rate as a means to promote greater equality could result in achievement of the opposite effect, an unintended consequence to say the least. Therefore, we now examine the empirical relationship between college attainment and income inequality in the U.S.

III. Empirical Evidence A casual inspection of historical data over the six decades since the mid-twentieth century tells two different stories. Figure 2 plots the national college attainment rate against two measures of family income inequality over the period 1950-2010.32 The family income Gini coefficient, a measure of relative income inequality among families that takes a value between 0 and 1 where the former represents perfect equality and the latter perfect inequality, is the measure used in panel A, whereas the ratio of the shares of income received by the top to bottom quintiles (80/20 30

Richard Vedder, Christopher Denhart and Jonathan Robe, Why are Recent College Graduates Underemployed? (Washington, D.C.: The Center for College Affordability and Productivity, January 2013). 31 Sandy Baum, Jennifer Ma, and Kathleen Payea, Education Pays 2013: The Benefits of Higher Education for Individuals and Society (New York: College Board, 2013). 32 College attainment data source: U.S. Census Bureau, Current Population Survey. Retrieved September 19, 2013 from http://www.census.gov/hhes/socdemo/education/data/cps/historical/index.html.

ratio) of families is indicated in panel B.33 Both figures reveal a similar trend, namely a continual rise in the college attainment rate over the entire period and a general decline in inequality into the early 1970s, a trend that reversed during the middle of the decade. The growth in college attainment in the post-war era has been influenced by public policies such as the GI Bills of 1944 and 1952 that provided tuition subsidies for military veterans, the Higher Education Act of 1965 that provided low income students with taxpayerfunded grants and scholarships to attend college, and the 1978 Middle Income Student Assistance Act that extended government loans to all students, regardless of need.34 Several of these policies are indicated in Figure 2. There was also an explosion in state and local government subsidies for public colleges and universities over the time period, as real per capita expenditures by state and local governments on higher education increased from $67 to $659, a nearly nine-fold increase between 1950 and 2008. Expressed in per-student terms, real state and local government expenditures grew from $4,392 to $10,474, a real increase of 140 percent.35 We observe a negative relationship between income inequality and college attainment at the national level during the two decades spanning 1950-1970, but a positive relationship since then.36 It is interesting to note that inequality generally exhibited a downward trend prior to 1965 when the federal government entered the student finance sector on a large scale, and that it began to rise as the federal government became increasingly involved in financing student tuition. Income inequality and college attainment both increased in all 50 states since 1970, although the growth in each varied by state. In general the states exhibiting the largest gains in college attainment over the period, such as Connecticut, Massachusetts and New Jersey, also exhibited

33

Income inequality data source: U.S. Census Bureau, Current Population Survey, Annual Social and Economic Supplements. Retrieved September 19,2013 from http://www.census.gov/hhes/www/income/data/historical/inequality/index.html. 34 Daniel L. Bennett, Adam R. Lucchesi, and Richard K. Vedder, For-Profit Higher Education: Growth, Innovation and Regulation (Washington, D.C.: The Center for College Affordability and Productivity, July 2010). 35 Government expenditure and population data are from the U.S. Census Bureau’s Annual Surveys of State and Local Government Finances. Enrollment data are from the U.S. Department of Education. Nominal dollar figures adjusted using the CPI-U. 36 Between 1950 and 1970, the 80/20 ratio declined from 9.5 to 7.6, a decrease of 20 percent, and the family income Gini coefficient declined from 0.379 to 0.353, a 6.9 percent decline. Meanwhile, the college attainment rate nearly doubled, growing from 6.2 to 11 percent. Since 1970, the college attainment rate has nearly tripled reaching 30 percent in 2010. Over this same period, the 80/20 ratio grew to 12.5 and the Gini coefficient to 0.44, increases of 64.5 and 24.6 percent, respectively.

the largest increases in inequality. Figure 3 depicts this trend, plotting the change in the family income Gini against the change in college attainment over the period 1970-2004 by state.37 We must be careful not to draw too strong of conclusions from these data, as many other factors likely impacted the distribution of income, including changes in the structure of the economy and characteristics of the population. To gain a better understanding of the relationship between college attainment and income inequality, we turn to multivariate regression analysis as a means to control for some of these other factors that likely influenced the distribution of income. We constructed a panel dataset spanning the period 1970-2004 using the fifty states as the unit of analysis to accomplish this task. We use fixed effects regression techniques which allow us to exploit variation within states over time to examine the correlations between multiple independent variables and the dependent variable, income inequality. Equation 1 gives the general form of the marginal effects to be estimated, where s and t denote the state and year, respectively, Y is the dependent variable, X is an (𝑁 × 𝑇) × 𝐾 matrix of independent variables, 𝛽 a 𝐾 × 1 vector of partial effects, 𝑐𝑠 an unobserved fixed state effect and 𝑒𝑠,𝑡 an idiosyncratic error term.38 The term 𝜈𝑡 in the equation represents a deterministic time trend and controls for the effect of time on inequality, as the data exhibit a general upward trend.39 𝑌𝑠,𝑡 = 𝛼 + 𝑋′𝑠,𝑡 𝛽 + 𝜈𝑡 + 𝑐𝑠 + 𝑒𝑠,𝑡 James Galbraith and Travis Hale estimate annual family income Gini coefficients for all 50 states over the period 1970-2004.40 We use this data as our primary measure of income inequality in the analysis below. The independent variable of interest is the college attainment rate (college). State-level attainment data are from the 1970-2000 decennial Censuses of the Population and 2006-2010 American Community Surveys. Missing years are interpolated using the compound rate of growth between observable periods. In addition, we control for a number of additional economic and demographic variables that have been associated with income inequality in the literature. This includes the female labor force participation (female LFPR) and 37

We use 2004 in lieu of 2010 as the end period here due to data availability. State level inequality data not available prior to 1970. 38 N, T and K are the number of states, time periods and independent variables, respectively. 39 𝜈 is a parameter to be estimated, while t is the period of observation, rescaled such that t=1 for 1970. 40 James K. Galbraith and Travis Hale, “State Income Inequality and Presidential Election Turnout and Outcomes,” The University of Texas Inequality Project, Working Paper 33 (March 2006).

unemployment rates, median income, percentages of the population that are African American (black), Hispanic (Hisp) and senior citizens (senior), the share of persons employed in the manufacturing and finance, insurance and real estate (FIRE) sectors, union density (union), and urbanization (urban). Details about these data and sources are given in the Appendix. Descriptive statistics for all of the state-level data are given in Table 1. Table 2 reports the estimates using the entire panel of data. Column 1 includes college and the deterministic time trend (time), both of which are positive and statistically significant at 5 percent or better. Column 2 adds the college squared term, allowing for a non-linear relationship. The linear and squared college terms have negative and positive coefficients as necessary to form a U-shaped inequality-attainment curve, although only the squared term is statistically significant.41Column 3 adds a set of three demographic variables (black, Hispanic and senior) to the equation from column 1. Each has a positive sign but only the latter two are statistically significant. The linear college term remains positive and significant, although the magnitude of the partial effect is reduced relative to column 1. Column 4 adds a set of economic variables (FIRE, female LFPR, log median income, manufacturing, urban, unemployment and union) to column 1 and the linear college term remains significant. Column 5 includes both the demographic and economic variables. The partial effect of college on inequality becomes insignificant in this specification. Columns 6-9 return to the non-linear college-inequality specification from column 2. Column 6 includes the set of demographic variables. The estimates for the demographic variables are similar to those in column 3, but both college terms are positive and neither significant. Column 7 adds the set of economic variables to the quadratic specification. The partial effects of the economic variables are qualitatively similar to those from column 4. Both the linear and squared college terms are significant at 10 percent or better in this specification, with the former negative and the latter positive, indicating a U-shaped inequality-attainment curve. Column 8 adds squared terms for each of the economic variables to account for potential non-linear effects of these variables. The college and college square terms are both significant at 41

One interpretation of the result from column 2 of Table 2 is that the relationship between inequality and attainment is exponential. This is potentially misleading however. A rejection of the null hypothesis that the coefficient on the linear college term equals zero does not preclude that there is in fact no relationship, particularly when the model may be misspecified. Column 2 does not include any additional covariates that likely impact the income distribution. As such, the coefficient estimates likely suffer from omitted variable bias. It is nonetheless worth retaining in the results for comparative purposes.

1 percent in this specification and maintain the U-shaped relationship. Column 9 adds the set of demographic variables to column 8. The college-inequality curve holds, as both college terms are significant at 1 percent. The regressions from Table 2 explain 81 to 90 percent of the within-state variation in income inequality, depending on the specification, with column 9 having the highest 𝑅 2 value. Columns 7, 8 and 9 provide statistical evidence that there is a U-shaped relationship between college attainment and income inequality. This suggests that increases in college attainment from an initially low level act to reduce income inequality, but there are diminishing social returns such that there is an inequality-minimizing attainment rate, above which additional increases act to increase inequality. Using the partial effects from columns 7-9, we estimated the college attainment rate at which partial effect of college attainment on inequality changes from negative to positive. We refer to this as the inequality-minimizing attainment rate and denote it as college* in the results. College* ranges from 0.193 to 0.246 in columns 7-9, suggesting that the current college attainment rate of 30 percent is to the right of the inequality-minimizing curve such that additional increases in attainment are associated with more income inequality. We further examine the college-inequality curve in Table 3 by examining subsets of the period. Columns 1-5 only include the linear and squared college terms. Column 1 reproduces the estimates from column 2 of Table 2, which utilized the entire 1970-2004 time period. Columns 2 and 3 repeat this specification but are restricted to the 1970-1989 and 1990-2004 periods, respectively. The U-shaped curve is present in both equations, but the college terms are only significant (at 5 percent or better) in column 3. Columns 4 and 5 repeat this specification but are restricted to the periods 1970-1984 and 1985-2004, respectively. In both columns 4 and 5, the linear and squared college terms are significant (at five percent or better), with the former negative and the latter positive. Columns 6-10 add additional demographic and economic control variables. Column 6 serves as the baseline, as it includes all observations spanning the 1970-2004 time period. Columns 7 and 8 repeat the baseline for the sub-periods 1970-1989 and 1990-2004, respectively. As was the case in columns 2 and 3, both college terms are significant only in the latter time period, although the linear term is negative and significant in column 7. Columns 9 and 10 repeat the baseline for the sub-periods 1970-1984 and 1985-2004, respectively. Similar to columns 4 and 5, both college terms are significant at 5 percent or better for each period.

The results from Table 3 provide additional evidence of the existence of an inequalityminimizing college attainment rate.42 These results suggest however that the inequalityminimizing attainment rate remains constant for relatively long periods of time –up to 35 years. Figure 4 plots the family income Gini against the college attainment rate for five year intervals beginning in 1970. Casual observation of these graphs suggests that the relationship between the two variables has changed over time and that the inequality-minimizing attainment rate may have increased over time. This seems practical given that the structure of the economy changes over time and changes in the education level of the workforce influence the composition of workers and their wages, as suggested by economic theory. Next we estimate the inequality-minimizing college attainment rate by five-year period. We do so by interacting the linear and quadratic college terms with a series of seven dummy variables equal to one if the observation occurs during the five-year interval 𝑑𝑗, 𝑗 = [1970 − 1974, 1975 − 1979, … ,2000 − 2004], and zero otherwise. Equation 2 describes the regression to be estimated. Table 4 gives the results, including the number of states below and above college* for each period. It also includes the results from a separate regression using annual state household income Gini measures from the American Community Survey as the dependent variable over the 2006-2010 period in column 8. All of the college terms are significant at 5 percent or better and the equation explains about 84 percent of the within-state variation of income inequality. 7 2 𝐺𝑖𝑛𝑖𝑠,𝑡 = 𝛼 + ∑[𝛽𝑗 (𝐶𝑜𝑙𝑙𝑒𝑔𝑒𝑠,𝑡 × 𝑑𝑗) + 𝜇𝑗 (𝐶𝑜𝑙𝑙𝑒𝑔𝑒𝑠,𝑡 × 𝑑𝑗)] + 𝛾𝑡 + 𝑐𝑠 + 𝑢𝑠,𝑡 𝑗=1

Prior to the latter part of the 1990s, most states had an average attainment rate below the inequality-minimizing point, suggesting that the states could have experienced additional increases in attainment without exacerbating inequality (indeed, moderately reducing it).

42

We estimate this rate when both the linear and quadratic college terms are statistically significant. For instance, college* is 0.23 in column 6 for the specification spanning the period 1970-2004. The average state attainment rate over the 35-year period is greater than 23 percent in four states–Colorado, Connecticut, Maryland and Massachusetts. College* is 0.23 and 0.24 in columns 9 and 10 for the 1970-1984 and 1985-2004 periods, respectively. None of the states have an average attainment rate greater than 23 percent over the 1970-1984 period, but 15 states (Alaska, California, Colorado, Connecticut, Hawaii, Maryland, Massachusetts, Minnesota, New Hampshire, New York, Utah, Vermont, Virginia, Washington) have an average attainment rate above 24 percent for the 1985-2004 period.

College* was 0.136 during the first half of the 1970s and only 10 states had an attainment rate above 13.6 percent.43 By the latter part of the decade, the inequality-minimizing rate increased to 0.154 and another 8 states moved to the right of college*.44 The inflection point increased to 0.184 during the first part of the 1980s, although 3 fewer states achieved an attainment rate above college*.45 The inequality-minimizing rate grew to 0.22 in the latter part of the 1980s and first half of the 1990s before rescinding slightly to 0.21 in the late 1990s and first five years of the new millennium. College attainment continued to grow and by the latter part of the 1990s, the majority of states surpassed the estimated inflection point. In the early 2000s, 39 states had achieved an attainment rate above college*, suggesting that the inequality-reducing dimension of higher education had largely been exceeded and diminishing returns set in. The inequalityminimizing rate for the 2005-2009 period pertains to household income inequality as opposed to family income inequality. Although measures of the former typically suggest greater inequality than the latter, the estimates nonetheless suggest that approximately half of the states exceeded college* during the period. 46

IV. Policy Implications Public policy designed to increase college enrollments and completion may be done with good intentions, but it can lead to unintended consequences. As discussed above, one consequence has been a watering down of the curriculum to accommodate less academically prepared students. This has coincided with grade inflation. These trends have reduced the potential human capital gains of a college education as well as distorted the signal that a college credential sends to employers about the ability and work effort of a prospective employee. The upward shift in the supply of college-credentialed workers has exceeded the increase in demand from the labor force. Rather than drive down the wage level of college workers and raise the wages of the non-college-educated workers, working to decrease inequality as a static 43

The 10 were Alaska, California, Colorado, Connecticut, Delaware, Hawaii, Maryland, Massachusetts, Utah and Washington. 44 Arizona, New Hampshire, New Jersey, New Mexico, New York, Oregon, Vermont and Virginia joined the 10 states indicated in the previous footnote. 45 Arizona, Delaware and New Mexico 46 For more information on various inequality measures and their properties, see Daniel L. Bennett, “The Concept and Measurement of Economic Inequality," in Essays on Institutions, Economic Development, and Inequality, Electronic Theses, Treatises and Dissertations. Paper 8728 (2014): 62-85.

analysis would predict, this largely government-induced disequilibrium has instead resulted in the displacement of non-college-educated workers with college-credentialed workers. Given the choice between hiring college and high school educated workers to fill relatively low paying and low skill positions, employers are increasingly hiring the former as evidenced by rising underemployment of college graduates. As a result, the latter group has a harder time finding employment in sectors that have historically been open to them. Thus policies designed to induce greater college attendance as a means to open the doors of opportunity and promote more equality may result in the opposite effect. Another factor contributing to fundamental changes in the equality-higher education attainment relationship over time is the rising cost of college. Over the past 30 years or so, tuition fees have risen at roughly double the rate of inflation. More importantly, the growth in tuition fees greatly exceeds the growth in median family income. To be sure, net (after tuition discounts) fees have risen somewhat less dramatically, but there is evidence that students applying for college are heavily influenced by “sticker” prices –undiscounted tuition fees, since the amount of scholarship aid is unknown at the time of original application. Sensitivity to prices generally falls as income rises. In this context, the increase in tuition fees, even relative to income levels, has no doubt disproportionately led some lower income Americans to avoid college on the grounds that it is too expensive. Federal data painstakingly analyzed by Thomas Mortenson suggests that in 1970, about 12 percent of 24 year olds with college degrees came from the bottom quartile of the income distribution, compared with a meaningfully smaller proportion (about 10 percent) in 2011.47 Colleges are becoming far more unequal economically, the rhetoric of college presidents notwithstanding. Space does not permit an elaborate analysis of the determinants of rising college costs. One very likely culprit, however, is the explosive growth in federal student assistance programs, a view commonly accredited to former Secretary of Education William Bennett who was the first to publicly express it in an editorial more than 25 years ago.48 While there is definitely divided opinion on the veracity of the Bennett Hypothesis, the best recent studies we have seen are

47

Thomas G. Mortenson, “Family Income and Unequal Educational Opportunity, 1970 to 2011,” Postsecondary Education Opportunity, no. 245 (November 2012). 48 William J. Bennett, “Our Greedy Colleges,” New York Times, February 18, 1987.

generally at least partially supportive of it.49 No one questions that federal student aid increases the demand for higher education, and given restraints on educational supply (e.g., selective admissions, accrediting rules restraining new college entrants), on theoretical grounds the net effect of enhanced federal student financial aid is almost certainly to increase prices (tuition fees) as well as enrollments. The federal government makes low cost loan money available to young persons, who feel they now can “afford” to go to college. The colleges, seeing this, raise their fees from what would otherwise be the case, pushing up prices. Since much of the lending is being done to middle class and even moderately affluent students and parents, the net effect to lower income student enrollments may be that the government’s impact on pushing fees up may more than offset any positive impact of student financial aid to lower income students. Thus the federal student financial aid program’s explosive growth (5.4 percent real annual compound rate since 1971) may have had the unintended consequence of making higher education more a haven for the affluent, reducing intergenerational income mobility and actually increasing income inequality.50

V. Conclusions The analysis presented here indicates that the relationship between college attainment and income inequality probably has changed over time, which is plausible since changes in the structure and composition of the workforce exert an impact on the dynamic real economy. Theory and empirical evidence suggest that there may be a higher education Laffer Curve. Expanding college attainment may act to reduce income inequality to some extent, but there may exist a critical attainment rate beyond which additional growth in college attainment may exacerbate income inequality, particularly if the supply of college educated workers grows faster than the demand. We estimated this critical point and find that prior to the mid-1980s, most states were able to sustain increases in college attainment without increasing inequality. Since 49

Stephanie R. Cellini, and Claudia Goldin, “Does Federal Student Aid Raise Tuition? New Evidence on For-Profit Colleges,” NBER Working Paper 17827 (2012). Andrew Gillen, Introducing Bennett Hypothesis 2.0 (Washington, D.C.: The Center for College Affordability and Productivity, February 2012). Dennis Epple, Richard Romano, Sinan Scarpca, and Holger Sieg, “The U.S. Market for Higher Education: A General Equilibrium Analysis of State and Private Colleges and Public Funding Policies,” NBER Working Paper, 192988 (August 2013). 50 Federal aid figures include grants, loans, work study and education tax benefits. In constant 2012 dollars, this number grew from $22.2 billion in 1971-72 to $178.8 billion in 2011-12. On a per full-time equivalent student basis, the total federal aid rose from $3,106 to $11,074, a real annual compound growth rate of 3.2 percent. College Board, Trends in Student Aid 2013 (Washington, DC, 2013).

then the attainment rate has soared nationally, influenced significantly by the vast expansion of federal student financial assistance, particularly the provision of government loans to middle and high income families. By the latter part of the 1990s, the majority of states achieved an attainment rate exceeding the critical point such that additional increases in college completion may be acting to exacerbate inequality. Higher education is becoming more and more an oasis for higher and middle income kids. Whereas at one time a degree from, say, Slippery Rock State College was not vastly less impressive to employers than one from, say, Harvard, that is no longer the case. Relatively fewer lower income students graduate, and they increasingly attend colleges perceived as qualitatively inferior. Egged on by propaganda from colleges, high school guidance counselors, political leaders and some foundations, students increasingly borrow large sums of money, and yet a large portion fail to graduate or, if they do, take low paying jobs previously staffed by non-college graduates, contributing to rising student loan default rates. The less affluent students increasingly are finding college a bad deal –contributing to the probability that increased college attainment actually promotes greater income inequality. While there are certainly many factors contributing to the rise of income inequality, the analysis presented here should at least cause us to pause and rethink the role of public policy designed to promote greater college attainment as a means to achieve greater equality. The decades-long expansion of federal subsidization of higher education enrollment could be triggering the opposite effect.

References Daron Acemoglu, “Technical Change, Inequality, and the Labor Market," Journal of Economic Literature, 40 (2002): 7-72. Robert Archibald and David H. Feldman, Why Does College Cost So Much? (New York: Oxford University Press, 2011). Richard Arum and Josipa Roska, Academically Adrift: Limited Learning on College Campuses, (Chicago: The University of Chicago Press, 2010). Philip S. Babcock and Mindy Marks, “The Falling Time Cost of College: Evidence from Half a Century of Time Use Data,” Review of Economics and Statistics 93, no. 2 (2011): 468-78. Sandy Baum, Jennifer Ma, and Kathleen Payea, Education Pays 2013: The Benefits of Higher Education for Individuals and Society (New York: College Board, 2013). Daniel L. Bennett, Adam R. Lucchesi, and Richard K. Vedder, For-Profit Higher Education: Growth, Innovation and Regulation (Washington, D.C.: The Center for College Affordability and Productivity, July 2010). Daniel L. Bennett, “The Concept and Measurement of Economic Inequality," in Essays on Institutions, Economic Development, and Inequality, Electronic Theses, Treatises and Dissertations. Paper 8728 (2014): 62-85. Daniel L. Bennett, “Myth busting: The Laissez Faire Origins of American higher education,” The Independent Review 18, no. 4 (2014): 503-525. William J. Bennett, “Our Greedy Colleges,” New York Times, February 18, 1987. Andreas Bergh Andreas and Günther Fink, “Higher Education Policy, Enrollment, and Income Inequality," Social Science Quarterly 89, no. 1 (2008): 217-235. Andreas Bergh Andreas and Günther Fink, “Higher Education, Elite Institutions and Inequality,” European Economic Review 53 (2009): 376-384. Derek Bok, American Higher Education Today, (Princeton, N.J.: Princeton University Press, 2013). William G., Bowen, Matthew M. Chingos, and Michael S. McPherson, Crossing the Finish Line: Completing College and America’s Public Universities (Princeton, N.J.: Princeton University Press, 2009). Stephanie R. Cellini, and Claudia Goldin, “Does Federal Student Aid Raise Tuition? New Evidence on For-Profit Colleges,” NBER Working Paper 17827 (2012).

José De Gregorio and Jong-Wha Lee, `”Education and Inequality: New evidence from CrossCountry Data," Review of Income and Wealth 48, no. 3 (2002): 395-416. Dennis Epple, Richard Romano, Sinan Scarpca, and Holger Sieg, “The U.S. Market for Higher Education: A General Equilibrium Analysis of State and Private Colleges and Public Funding Policies,” NBER Working Paper, 192988 (August 2013). James K. Galbraith and Travis Hale, “State Income Inequality and Presidential Election Turnout and Outcomes,” The University of Texas Inequality Project, Working Paper 33 (March 2006). Andrew Gillen, Introducing Bennett Hypothesis 2.0 (Washington, D.C.: The Center for College Affordability and Productivity, February 2012). Claudia Goldin and Lawrence F. Katz, The Race Between Education and Technology, (Cambridge, MA: Harvard University Press, 2008). Igal Hendel, Joel Shapiro, and Paul Willen, “Educational Opportunity and Income Inequality," Journal of Public Economics 89 (2005): 841-870. Caroline M Hoxby and Bridget Terry, “Explaining Rising Income and Wage Inequality among the College-Educated," NBER Working Paper 6873 (January 1999). J.B. Knight and R.H. Sabot, “Education Expansion and the Kuznets Effect," American Economic Review 73, no. 5 (1983): 1132-1136. Paul Krugman, Paul, “And Now for Something Completely Different: An Alternative Model of Trade, Education, and Inequality," in R.C. Feenstra, ed., The Impact of International Trade on Wages (University of Chicago Press, 2000), 15-28. Thomas G. Mortenson, “Family Income and Unequal Educational Opportunity, 1970 to 2011,” Postsecondary Education Opportunity, no. 245 (November 2012). Arthur M. Okun. Equality and Efficiency: The Big Tradeoff (Washington D.C.: The Brookings Institution, 1975). Rati Ram, “Can Educational Expansion Reduce Income Inequality in Less-Developed Countries?” Economics of Education Review 8, no. 2(1989):185-195. Sherman Robinson, “A Note on the U Hypothesis Relating Income inequality and Economic Development," American Economic Review 66, no. 3 (1976): 437-40. Stuart Rojstaczer and Christopher Healy, “Where A Is Ordinary: The Evolution of American College and University Grading, 1940-2009,” Teachers College Record 114, no. 7 (2012): 1-23. A. Michael Spence, “Job Market Signaling," Quarterly Journal of Economics 87, no. 3 (1973): 355-374.

Joseph E. Stiglitz, “The Theory of Screening, Education, and the Distribution of Income,” The American Economic Review 65, no. 3 (1975): 283-300. Richard Vedder, Christopher Denhart and Jonathan Robe, Why are Recent College Graduates Underemployed? (Washington, D.C.: The Center for College Affordability and Productivity, January 2013).

Appendix: Data Sources In the statistical analyses, we control for a number of other demographic and economic variables that may influence the distribution of income. These data and their sources are described in this appendix. Summary statistics for all of the data are presented in Table 1 above. The economic variables include the female labor force participation rate (LFPR), the shares of employment in the finance, insurance and real estate (FIRE) and manufacturing industries, real log median income, the annual average state unemployment rate, the urbanization rate and union density. The female LFPR is defined as the share of the civilian, non-institutional females above sixteen years of age in the labor force in each state. The data come from two sources. The primary source is the U.S. Bureau of Labor Statistics’ (BLS) Current Population Survey, which provides the annual average of monthly labor force participation rates over the 1976-2010 period, although several years (1977, 1980, 1981, 1989, 1997, 2006) were not available.51 Next, the 1970 Census of the Population provides state-level female LFPRs. State-level female LFPRs were not available for the 1971-1975 period in addition to the individual years indicated above. National female LFPRs were however available annually for the entire 1970-2010 period.52 Data for the missing years was interpolated using equation 3, where 𝑋𝑖,𝑡 is the interpolated female LFPR for state I in year t, 𝑌𝑡 is the national LFPR in year t, 𝜒𝑖,𝑡 = 𝑋𝑖,𝑡 /𝑌𝑖,𝑡 is the ratio of the LFPR in state i in year t to the national LFPR in year t, and j and k are the number of years since the last and next, respectively, actual state-level female LFPRs are available.53 𝑋𝑖,𝑡 = [

𝜒𝑖,𝑡−𝑗 + 𝜒𝑖,𝑡+𝑘 ] 𝑌𝑡 2

The FIRE and manufacturing shares of employment ratios are computed using annual state-level data from the Bureau of Economic Analysis, with both having total employment as the denominator. Due to the change in industry classification codes in 1997, industry employment data are not directly comparable over the entire period of study, although the total employment figures are identical under both systems. U.S. Federal statistical agencies adopted 51

The BLS data were collected from the annual Statistical Abstract of the United States reports. BLS reported a national female LFPR figure for 1970 that differed significantly from the Census ratio. As such, the BLS data are used for the 1971-2010 period, but the national rate for 1970 is from the Census of the Population. 53 For instance, state-level female LFPR’s are interpolated for 1980 using j=1 and k=2 since the actual rates are available for 1979 and 1982, but not 1981. Similarly, the figures interpolated for 1981 using j=2 and k=1. Note that 𝑋𝑖,1980 /𝑌1980 = 𝑋𝑖,1981 /𝑌1981 and the 𝜒𝑖,𝑡 is unchanged for t=1971,..,1975. 52

the North American Industry Classification System (NAICS) in 1997, replacing the Standard Industrial Classification (SIC) system that had been in place since 1937. The BEA provides SIC data from 1969-2001, and NAICS data from 1990-2010. The total employment data for the two classifications matches for the overlapping years, but the FIRE and manufacturing employment data differ. Because our analysis covers the 1970-2004 period, we need to make an adjustment in order to make the data comparable over the period of study. We do so using data from the 19902001 period since it is available for both systems. The SIC FIRE employment data are given under the broad industry heading “Finance, Insurance and Real Estate.” The corresponding NAICS figure used in the present analysis is the sum of the “Finance and Insurance” and “Real Estate and Leasing” industry categories in an effort to match the corresponding sectors encompassing the respective classification systems. Both the SIC and NAICS systems report manufacturing employment numbers, but the two figures differ due to reclassification of sectors. The correlation between the FIRE employment share under the two classification systems for the overlapping years is 0.956 for the entire 1990-2001 period, with the correlation coefficients for the individual years ranging from 0.921 in 1994 to 0.985 in 2001.54 The correlation between the manufacturing employment share under the two classifications for the 1990-2001 period is 0.990, with the correlation for single years ranging from 0.984 in 1991 to 0.994 in 1997. Thus, the ratios for both variables between the two datasets are highly correlated. As such, we employ a fixed effects regression model to predict the SIC employment ratios for post-2001 years using the NAICS employment ratios and deterministic trend as a control variable to account for employment trends. The model is given by equation 4, where 𝑋𝑠𝑖𝑐 and 𝑋𝑛𝑎𝑖𝑐𝑠 are the employment ratios for the SIC and NAICS systems, respectively, and t is deterministic trend such that t=1 represents 1990. The results from the regression are reported in Table 4. Nominal state median household income data for 1975 and the 1984-2010 period come from U.S. Census Bureau's Current Population Survey, and are augmented by decennial state median income measures for 1969 and 1979. Figures for the missing years spanning 1970-1974, 1976-1978 and 1980-1983 were interpolated using the average annual compound growth rate between observation years to estimate the missing measures. Nominal median income figures 54

With the exception of the correlation of 0.921 in 1994, the correlation for all years was above 0.95.

were adjusted to constant 2011 dollars using the CPI-U. The natural log of the real median income figure is used. Median income is included as a measure of the level of income in a state. Annual state unemployment rate data spanning 1970-2010 were collected from the Statistical Abstract of the United States.55 Urban population data come from the 1969-2009 decennial Censuses of the Population. We use the annual compound growth rates between Censuses to linearly interpolate the between-Census figures.56 Union density data are from Hirsch, MacPherson, and Vroman (2012), who estimate the share of non-agricultural unionized labor annually by state. We include age, race and ethnic characteristics of the population to control for state demographic trends. Age data are reported annually by the Population Distribution Branch of the U.S. Census Bureau. We use the share of the population above age 65 (senior) as a control variable. We also control for racial and ethnic characteristics of the population by including the percent of the population that is black and Hispanic. Data for these two variables was obtained from the decennial Censuses, with between Census observations interpolated using the annual compound state growth rate for each variable.

55

BLS is the orignal source of the data. The figures are reported monthly. An arithmetic average of the monthly figures is used as the annual rate. 56 Population density was used as an alternative to urbanization. Doing so does not substantially change the results.