THE ECONOMICS OF RECENT GENERATIONAL CONFLICT IN THE U.S.: ANALYZING TRENDS IN AGE-BASED WAGE INEQUALITY,

THE ECONOMICS OF RECENT GENERATIONAL CONFLICT IN THE U.S.: ANALYZING TRENDS IN AGE-BASED WAGE INEQUALITY, 1976-2015 Nicholas Rogness Abstract: Ballooning student debt and a tepid job market have fueled a heightened sense of economic anxiety among younger Americans. Meanwhile, older generations most of which now form the baby-boomer generation have aroused disdain for the greater volume of economic opportunities they seem to have enjoyed throughout their life-cycle. Indeed, a likely consequence of these disparities in opportunity has been an observed increase in economic inequality between older and younger age groups. Specifically, this paper focuses on the growth in wage inequality between age groups in other words, a phenomenon I define as age-based wage inequality. In 1976, according to Current Population Survey data from the Bureau of Labor Statistics, the average worker aged 55-64 earned approximately the same hourly wage as the average younger worker aged 25-34 by 2015, the older worker earned more than 25% in hourly wages than his/her younger counterpart. I then attempt to examine the cause of this growth in age-based wage inequality through a relative labor supply and relative labor demand framework based on the analysis of prominent education economists Claudia Goldin and Lawrence Katz. This adapted framework reveals that labor supply for older workers relative to younger workers exhibited a steady growing trend beginning in the 1990s after contracting across the 1970s and 80s. This trend is consistent with the unexpected turn-around in labor-force participation rates among older Americans observed in the 1990s (Harvard School of Public Health, 2004). The growth in relative old-young labor supply also appears to have gone in tandem with an even greater increase in old-young relative labor demand. After all, this conclusion, as reflected in my own calculations, is the only one that accounts for the relative growth in older worker wages and agebased wage inequality in accordance with the laws of supply and demand as found in introductory economic textbooks. Thus the last part of my analysis focuses on the plausible contributors to the growth in relative demand for older workers. I test the effects of differences in college completion rates across generations (specifically comparing groups born between selected years, known formally as birth cohorts ), improving technological productivity in the workforce (a phenomenon known by economists as skills-biased technological change ), and differences in unionization rates across generations. The rapidly increasing college completion rate of birth cohorts from the late 1930s to the early 1940s appear to play a crucial role in the growth of age-based wage inequality. Drawing from other economics literature on higher education, the growing size of the birth cohort observed at age 30 over time also appears to have some empirical link with age-based wage inequality. Keywords: wage inequality economic inequality generational conflict generation gap labor economics higher education economics of education Nicholas Rogness 2017. Originally published in Explorations: The UC Davis Undergraduate Research Journal, Vol. 19 (2017). http://explorations.ucdavis.edu The Regents of the University of California.

I. Introduction The well-documented rise in income inequality and stagnant wages in the United States since the 1980s has become a topic of growing concern in the wake of the 2008-2009 so-called Great Recession. Yet the causes of growing unequal wages have been debated by economists and other interested social scientists for nearly two decades perspectives generally differ from an emphasis on the role of public policy and financial deregulation (such as Piketty s (2014) changing social norms argument) to a more conventional viewpoint that attributes differences in wages to shortages in socalled skilled labor. Economists taking this less political approach to the inequality discussion generally support the concept of skills-biased technological change the widely-accepted idea that computerization and capital investments in labor-saving technology taking place throughout the 1970s & 1980s favored the circumstances of more skilled, highly educated workers at the expense of those unskilled (Autor, Katz, and Krueger, 1998). Skills-biased technological change (often abbreviated to SBTC for short) has been affirmed by most economists as the primary contributor to wage inequality in the United States, a consequence of the invention of the microcomputer. In fact, according to prominent National Bureau of Economic Research economists David Card and John DiNardo (2002), the recent inequality literature [of the 1990s and early 2000s] reaches virtually unanimous agreement that relative demand for highly skilled workers increased in the 1980s, causing earnings inequality to increase as skilled workers were more likely to use a computer at work, and thus receive better compensation in proportion to the additional value that such technology provided to the economy. Skills-biased technological change has been a welltrodden subject by researchers in the economics discipline, so I take little interest in exploring it further in this paper. Rather, I examine a specific facet of income inequality which I had hypothesized was perhaps influenced by SBTC as it pertains to the issue of generational conflict. I define this phenomenon with a new term included in the title of this paper: age-based wage inequality. The inequality stems from the observed disproportionate growth in the wages of older working Americans (which I define for my purposes as those aged 55-64) compared to their younger counterparts (defined as aged 25-34). Between 1976-2015, employment-weighted hourly wages grew by approximately 35% for older workers by contrast, younger worker hourly wages grew only briefly by roughly 16% over the late 1990s and early 2000s and have since stagnated (Figure 1). 1 As a result, an agebased gap in wages has formed whereas older and younger workers yielded approximately the same level of wages in 1976, the former earned 25% more per hour than the latter in 2015. I offer several well-tested factors associated with age-based wage inequality. I first test the expected role of skills-biased technological change, for which I fail to find empirical evidence. Differential experiences in union participation rates across age groups are associated with growing older worker wages, although to a limited extent. Fluctuations in the college completion rate across birth cohorts appear to act as the single largest contributor, especially with the rapid growth in college graduation rates for birth cohorts born between 1935 and 1950, and its decline beginning with the 1980s birth cohorts (see Figure 10). As expected from the findings of Card and Lemieux (2001), I find that the estimated average cohort size of those considered young (for the purposes of this paper) is the single-most statistically significant factor that is positively associated with relatively unequal growth in older-worker wages. Unemployment in the local labor market is also positively correlated with age-based wage inequality, as well as the proportion of employment within individual states in industries that exhibited the largest volumes of job growth during the 1990s. However, the significance of these relationships vary across various specifications of old-young relative wage and labor supply measures (referred throughout the rest of the paper simply as old-young relative labor measures ). I also test the plausible role of rising college tuition cost; however, the price of in-state tuition at 4-year public universities demonstrates a surprising statistically insignificant relationship. The objectives of this paper are twofold and roughly divides this paper into two parts. The first objective apparent throughout Sections II & III is to offer a definition of the age-based wage inequality phenomenon by calculating and illustrating various agecomparative labor statistics. Part of this descriptive analysis relies on a modification of the supply, demand, and intuitions (SDI) framework used by Goldin and Katz (2008) in their research on the relationship between education and wage inequality. The second 1 In other words, each observation of wage data was adjusted according to the number of hours worked in a year by each survey respondent. Vol. 19 (2017) Nicholas Rogness p.2

objective of Sections IV attempts to uncover the causal mechanisms, through multivariate regressions that utilize standard econometric techniques, behind agebased wage inequality, as briefly described earlier. This part includes reference to a broad range of related economics literature, as well as an analysis of the regressions that follow. In accomplishing all this, I wish to establish a valuable empirical base from which other researchers may follow. It would useful to note that alternative phrases are used throughout the paper to describe age-based wage inequality these include oldyoung relative wages and the age wage premium, yet they all describe the same phenomenon. II. Age-Based Wage Inequality in the United States Recent Trends in Age-Based Wage Inequality from 1976-2015 A number of macroeconomic trends attest to a growing sense of economic anxiety among younger people, most recently the millennial generation born between 1981 and 1999. A growing body of research implicitly attributed this to a growing disparity in labor market outcomes between older and younger age groups, with the former group having received a disproportionate volume of aggregate wage growth and employment opportunities. The Pew Research Center (2011) found that the age gap in economic outcomes grew considerably between 1984 and 2009, with heads of households aged 65 and older having experienced a 42% increase in median net worth and a 109% increase in total household income. Meanwhile, householders under age 35 had an increase of 29% in total income since 1967 and a contraction in median net worth of 68% since 1984, well underperforming the near retirement 55-64 age bracket, whose incomes grew by 54%. Wage data tabulated from the Current Population Survey s Annual Social and Economic Supplement (CPS ASEC) further illustrates the disparity in economic outcomes that arose over the period of 1976-2015. Figures 1 and 2 explain two different halves of this story of rising inequality, with a dividing line that can be imagined being set in the early 1990s. In 1976, older workers enjoyed roughly the same hourly wage as their younger counterparts yet, by 2015, the average older worker earned (25%) more in hourly wages than the average younger worker, as shown in Figure 2. A closer inspection of the difference in trends in wages across the age groups in Figure 1 reveals the changing nature of these inequality trends before and after the crucial period of the early 1990s. Prior to 1994, the majority of the growth in age-based wage inequality nearly 65% of it, in fact is attributable to an 11% drop in younger worker wages starting from 1976. Wages for older workers, meanwhile, grew by 6% over this period. Yet the economic gap between the old and young is not a simple matter of divergence. While the substantial drop in younger wages over the 1976-1994 period brings attention to the key role of differential educational attainment across birth cohorts (as will be explained later in Section V), the trends in wages of both age groups reflect the same fluctuations in the business cycle. This is apparent, for instance, in the peaks of the economic expansion of the mid-1980s followed by the troughs resembling the after-effects of the recession of the early 1990s. The economic boom of the 1990s and early 2000s, referred to colloquially as the period of the dot com bubble, is reflected similarly in the rising wages for both older and younger workers. However, it is at this point that the second half of the inequality story develops that of substantially higher growth in wages among older workers. From 1994 to 2002, older worker hourly wages increased by 28% an increase about 47% greater than the younger worker s raise of only 19%. The early 1990s are also notable in that nearretirement age workers (again, still the age 55-64 group) began to participate in the labor force again at higher rates, in a reversal of the growing early retirement trend that began in the 1960s. In fact, the labor force participation rate (often referred to as LFPR by labor economists) among the aged 55-64 population grew from its low of 55% in 1987 to a period high of 66% in 2010 (see Table 1). Meanwhile, the LFPR among those aged 25-34 peaked at 83% in 2000 and has since declined. These two simple measures hourly wages and LFPR by age groups point to rising old-young relative wages, or wage inequality across age groups. It also points to an increase in the relative labor supply of older workers compared to younger workers. Thus I also introduce the concept of age-based relative labor supply (to which this term will be used frequently throughout the analysis of this paper), which can be calculated rather simply once the necessary microdata from government survey data is collected. When economists refer to labor supply or otherwise measure it, they measure this in the number of hours one has worked in a given time period. In my analysis, I have added up the Vol. 19 (2017) Nicholas Rogness p.3

number of hours estimated to have been worked in a year by each worker in each age group; 2 in other words, I added up all the hours worked by the older 55-64 age group as well as the younger 25-34 age group in the period of a calendar year. I represent the sum of hours worked in a year by all older workers with the letter O and the sum of hours worked in a year by all younger workers with the letter Y. In accordance with standard econometric techniques for calculating any form of relative labor supply (i.e. college-educated vs. noncollege-educated workers), I utilize the following formula for estimating relative labor supply: log(o/y) In other words, a natural logarithm is taken of the sum of older worker hours O divided by the sum of younger worker hours Y. The results allow one to compare the sum of hours worked by each age group by proportions and percentages. For instance, according to Figure 2, older workers contributed about 70% (or a proportion of -0.7) fewer hours of worked compared to younger workers in 1976. By 2015, that difference shrunk to less than 30% fewer hours of work. Another interpretation of this finding is that older workers increased their number of annual working hours from around 30% of that of younger workers to 70% of that of younger workers over the approximately 40-year period of analysis. Thus, Figure 2 illustrates a dramatic increase in the relative age-based labor supply beginning in the 1990s, fueled by increases in older worker labor force participation. Yet when one considers the principles of the laws of supply and demand, the growth in relative labor supply in the face of rising relative older-worker wages presents a puzzle. As supply is increased, or shifted outward in any market, this almost always results in an increase in price for market s product. 3 In the case of the old-young relative labor market, we use old-young relative wages as the price. Thus, old-young relative wages will be found at the quantity at which the old-young relative supply is in equilibrium with oldyoung relative demand. Assuming the relative demand for older workers had remained constant and that relative labor supply is perfectly inelastic (unaffected by 2 Survey respondents estimated the number of hours they worked on an average week in the previous year, as well as the number of weeks worked in the previous year. I multiplied these two measures together to estimate the annual number of hours worked. 3 The only exception to this is in the case of the perfectly competitive market where demand is perfectly elastic and price remains the same regardless of quantity supplied. changes in relative wages) in the short run, 4 this would have reduced relative older worker wages. In other words, the gap between older and younger worker wages age-based wage inequality would have shrunk. For this supply-and-demand based explanation for age-based wage inequality to be supported, then a key finding must be yielded from the data: relative demand for older worker labor must be shown to have outpaced the growth in relative old-young labor supply after the 1990s. I present such evidence in the age-based relative labor supply and relative labor demand calculations of the following section. III. The Role of Relative Demand Connections Drawn from Goldin and Katz s The Race Between Education and Technology Serving as the basis of my presented analysis is a supply, demand, and institutions (SDI) framework based on Goldin and Katz s education and skills shortage analysis of historical trends in wage inequality in their 2008 book The Race Between Education and Technology. Of interest in its application toward the study of age-based wage inequality is their comparison of the wages of skilled workers (those with a baccalaureate degree, graduate degree, and top half of wage earners with some college) to those considered unskilled (lower half of those with some college, high school diploma holders, and high school dropouts). Goldin and Katz proceeded to compute skilled-unskilled relative labor supply and relative labor demand (2008). In the process, they provided an explanation for the U- shaped pattern in wage inequality observed throughout the twentieth century. That is, a trend of rising inequality at the beginning of the century, falling inequality in the middle period from 1940-70, followed afterwards with a return to inegalitarian wage growth. They asserted that this two-sided story could be attributed mainly to fluctuations in skilled-unskilled relative supply. This in turn was affected by the volume of skilled workers in the workforce; thus, when wage inequality between skilled and unskilled workers fell from 1940-70, this was caused by an excess supply of skilled workers in the economy that outstripped the level of relative labor demand, which in turn could be plausibly attributed to the efficacy of the American education system. The growth in wage inequality after 4 All short-run measures of labor supply (relative or not) are generally accepted to be perfectly inelastic by the economics discipline. For reference, see Goldin and Katz (2008). Vol. 19 (2017) Nicholas Rogness p.4

1970 would then seem to suggest a shift in the ability of the American education system to supply skilled workers to the labor market. Yet there is a more relevant purpose for discussing Goldin and Katz s findings, aside from its implications for education policy. My own analysis of age-based wage inequality relies on a translation of their relative labor supply and demand framework. I modify two of Goldin and Katz s key equations in computing oldyoung relative labor supply and demand changes. I begin with the following relative supply and demand equations used to explain the premium for skilled labor: log(d su) = log(s/u) + σ su log(w s/w u) log(s/u) = log(w ss/w uu) log(w s/w u) Where S represents the total sum of efficiency units, or total sum of annual hours worked by skilled labor, U represents the total sum of annual hours worked by unskilled labor (using the same definitions for skilled and unskilled as mentioned above), w s and w u represents the weighted average hourly wages of skilled and unskilled workers, respectively, and σ su represents the elasticity of substitution between skilled and unskilled labor. Essentially, σ su accounts for the trade-offs firms face in choosing their combination of skilled and unskilled labor to maximize productivity given a fixed wage cost that they can expend. Mathematically, the negative inverse of this measure determines the slope of the skilled-unskilled relative demand curve. Goldin and Katz estimated a skilledunskilled elasticity of substitution of approximately 1.64; taking the negative reciprocal of this reveals the slope of the demand line, which they estimate at -0.61. With some modification, I create the following relative supply and demand equations for my own purposes. Old-young relative labor demand is thus computed using the following equation: log(d oy) = log(o/y) + σ oy log(w o/w y) Old-young relative labor supply, which is assumed to be fixed in the short run, is computed by subtracting the natural logarithm of the old-young relative wage bill from the old-young wage inequality measure: 5 log(o/y) = log(w oo/w yy) log(w o/w y) Where O represents the total sum of efficiency units, or total sum of annual hours worked by older labor (workers aged 55-64), Y represents the total sum of annual hours worked by younger labor (workers aged 25-34), w o and w y represent the average hourly wages of older and younger workers, respectively, and σ oy represents the elasticity of substitution between old and young labor. In other words, my elasticity represents the trade-off firms face in determining their ideal proportions of older and younger labor given a fixed labor cost that they can expend. Thus, the slope of the demand line between older and younger labor which presented difficulties in computing given the changing relationship between old-young relative wages and relative supply that begins in the 1990s (see Figure 3) can be found by taking the negative reciprocal of this elasticity. These figures are calculated for every year (which I represent with the letter t) in the data set. The results of the old-young relative supply calculations using this formula are rather pronounced. Age-based relative supply exhibits a rather dramatic U- shaped pattern over the 1976-2015 period, as suggested by the differing trends in labor-force participation rates (LFPR). Finally, I attempt to estimate the elasticity of substitution between young and old labor. In once again mimicking Goldin and Katz s methodology, I attempt to accurately estimate σ oy using two different linear regressions. For this, I utilize old-young relative wages as my dependent variable and old-young relative supply as the independent variable, as well as some controls. Perhaps the clearest opportunity to measure the relationship is revealed when plotting old-young wage inequality against old-young relative supply, as was done in Figure 3. By doing so, the significance of trend changes begun in the early 1990s is further underlined when restricted to data points prior to 1996, the relationship between age-based wage inequality and old-young relative supply is found to be significantly negative, even when controlling for a relative demand measure. The full regression includes old-young relative wages regressed against old-young relative supply, controlled with a time variable (observation year minus value of base year 1976) and 5 Wage bill is simply any measurement of wages interacted or multiplied with a number of hours worked in a given time period (the labor supply). In a sense, the term wage bill is meant to evoke the idea of any other bill that one would receive from purchasing a good or service, which is often broken down into the quantity per product used multiplied by the price per unit of product. In the case of age-based wage inequality, old-young relative wage bill refers to the natural logarithm taken of the result of average weighted younger worker hourly wages multiplied by the sum of annual hours worked by younger individuals over an analogous product found with older worker wages and labor supply. Vol. 19 (2017) Nicholas Rogness p.5

restricting the observations to those before 1996. 6 This results in an estimate of the slope of the oldyoung relative demand curve at -0.275. Taking the negative reciprocal of this yields an elasticity of substitution by age of approximately 3.6. Put in proper statistical notation, this means the elasticity was found using the following regression model with β 1 representing the slop of the relative demand curve and, in statistical terms, the coefficient of interest: 7 log(w o/w y) = β 1ln(O/Y) t + β 2(Time) t + ε t I obtain an additional estimate using dummy variables restricting for groups of years where the relative supply appears to move along the same relative demand line in the short run for the sake of comparability. This was necessary because of the dramatic growth in old-young relative supply that took place after the mid-1990s; by employing dummy variables that take advantage of more stable data points found in the short term of 5-year groupings, this prevents an otherwise spurious result due to the obviously nonlinear manner the wage inequality-relative supply data points are distributed across a scatterplot (Figure 3). Thus, I test this additional model to estimate σ oy: log(w o/w y) = D 1995-1999+ D 2000-2005+ D 2006-2010+ D 2011-2015+βln(O/Y) t+ε t Where β represents the coefficient of interest in this case, the estimated slope of the relative demand line. The results are presented in Table 2 and show a statistically significant coefficient of -0.195. Thus, taking the negative reciprocal of this and that of β 1 from the previous regression, my estimates of σ oy range from roughly 4 to 5, which is comparable to the 4 to 6 range found in Card and Lemieux s (2001) analysis of the college wage premium between younger and older age groups. It should be noted, however, that Card and Lemieux s analysis looked at slightly different age groups (those aged 26-30 and 56-60) in order to determine the degree of labor demand substitutability across all of the individual age groups at the same level of education (2001). Given the apparent novelty of these specific age-based calculations, I compromised between 6 The use of this variable as a proxy for relative labor demand is consistent with the methodology utilized in Goldin and Katz (2008). For reference, see page 299. 7 β2 represents a coefficient estimating the controlling effects of the time, or relative demand measure, and so is not of much interest insofar as it affects the result of β2. the two differing figures by including calculations for relative demand when σ oy=4 and σ oy=5. The numbers presented in Table 3 are consistent in telling the two distinct halves of the old-young wage inequality story: falling relative supply working in tandem with substantial relative demand growth before the early 1990s, followed by large growth in relative supply offset by even greater growth in relative demand. As illustrated by the relative wage/relative supply plot presented in Figure 3, old-young relative supply contracted from 1976-1994 by an annual average of - 1.76% when σ oy=4. Demand shifts, however, appear to have contributed more significantly to the rising wage inequality, especially after 1994: it averaged 1.79% in growth before and accelerated to 5.51% in the anteceding 1995-2015 period. Meanwhile, relative supply contracted up to 1994 before rebounding to an unprecedented 3.35% annual rate of growth. Thus the simplified supply and demand framework presented in Table 3 tells the two halves of the old-young inequality story: in the first half leading up to the early 1990s, moderate relative demand growth for older workers complemented contracting relative supply. This resulted in an annual age-based wage inequality growth rate of 0.89% until 1994, and accounts for roughly half of the growth in the inequality measure for the entire historical period. Post-1994 trends, on the other hand, provide a very different account while relative demand climbed, rising relative supply acted as a buffer on inequality, resulting in a slower annual relative wage growth rate of 0.54%. Thus, two logical conclusions are reached: not only does the enormous post-1994 shift in demand deserve explanation in its relation to sustaining growth in age-based wage inequality; whatever economic forces responsible for the growth in demand are also likely responsible for the growth in supply. I present evidence in Section IV based on differences in educational attainment across birth cohorts (groups of individuals born in the same given time period) that give me sufficient reason to imagine age-based wage inequality as primarily a side effect of skills-biased technological change. However, as will be explained, this relationship is not found and leads me to conclude that rising educational attainment across cohorts born throughout the 1930s and 1940s allowed older worker wages to rise independently of SBTC and even in the face of declining employment in the routine-task intensive occupations, such as those in manufacturing and clerical work, as documented by Autor, Katz, and Kearney (2006) and Autor and Dorn (2009). I explain this argument and test additional contributors to age-based Vol. 19 (2017) Nicholas Rogness p.6

wage inequality, including differences in the age gap in union-supplied labor and in inter-cohort educational attainment, in the coming section. Before moving onto this task, I test the robustness of the supply and demand calculations using a simple regression analysis of time-series data measuring both changes in the age-based wage premium and age-based relative supply: log(w o/w y) t = α (time) t + β log(o/y) t + ε t Where α represents the coefficient of all time effects over the period of analysis for every year t and β represents the coefficient of the effects of changes in relative supply. As evidence suggests that age-based relative supply increases over this period, I would expect to find a significantly negative result for the latter coefficient. Meanwhile, the expected positive effects of the time variable, which estimated as a proxy for older-young relative demand, were controlled. I estimate this simple regression model in Table 4. The results confirm the expected significantly negative relationship between old young relative supply and the age-based wage inequality measure once the significantly positive relationship of the relative demand variable is controlled. (That is, a simple time variable that takes the difference between the observation year and the base year 1976.) The addition of a post-1994 interaction variable in a second model (column 2) demonstrates that the acceleration of relative demand growth after 1994 is strongly associated with the sustained growth in age-wage inequality as observed. Both models demonstrate R 2 values of 0.917 and 0.929, respectively. It is not possible, however, to speculate as to how well these findings establish their robustness, as they encompass a relatively small number of observations. IV. Testing Hypotheses Examining the Role of Skills-Biased Technological Change, Old-Young Differences in Rates of Unionization, and Old-Young Differences in College Completion Rates Aside from Goldin and Katz, an analysis of several other research papers was needed to throw additional light on the mysterious age-based wage inequality phenomenon. More specifically, I considered what macroeconomic changes since the late 1970s highlighted by previous labor economics literature could have induced the drastic shifts in relative supply and demand in the post-1994 era. After all, there exists no previous academic research that has specifically measured and interpreted economic inequality across age groups. 8 As such, the methodology for this paper was based entirely on tangentially-related research. The modification of Goldin and Katz s relative supply and demand equations is the first such instance of methodology that reflects this challenge. To recall: the calculations from the previous section highlight post-1994 relative demand growth as the crucial factor allowing sustained age-base wage inequality growth, despite the growth in old-young relative supply. This leads one to wonder what began to make older workers relatively more employable starting in the mid-1990s. A survey of other economics literature provides a multitude of plausible factors, including skills-biased technological change, differential declines in unionization across age groups across industries, and differentials in educational attainment across birth cohorts. This section presents the results of empirical models developed from a survey of relevant economic literature. a. The SBTC Hypothesis It would seem age-based wage inequality could be plausibly explained by skills-biased technological change (SBTC), whose spread throughout the 1980s and 1990s displaced manual laborers with computerized machines and smaller numbers of well-educated operators while still boosting manufacturing productivity. SBTC was well-documented by Autor, Katz and Krueger (1998) as an economic phenomenon far greater in impact on the American labor force than global outsourcing. Indeed, they found a significant relationship between college graduate relative wage bill share and growing computer use in the workplace across nearly all industries, including those in manufacturing. 9 Those born throughout the 1930s, who reached primeworking age around the mid-1990s when accelerated age-based wage inequality began to touch off, may have stood to gain from continuing SBTC at that time. As Figure 10 illustrates, these birth cohorts saw some of the 8 It should be fairly noted, however, that Fry, R., Cohn, D., Livingston, G., & Taylor, P. (2011) made some measurement of differences in wealth and income inequality across age groups in their Pew Research paper. However, it speculated little as to its causes. 9 This version of relative wage bill refers to the natural logarithm of the ration of wage bills of college graduates over noncollege workers. See Footnote 5 for further reference on the general definition of wage bill. Vol. 19 (2017) Nicholas Rogness p.7

largest increases in college completion rates by age 30 ever recorded. In 1985, they would have been in their late career stage between ages 45-54. Yet evidence on computer use in the workplace from Card and Dinardo (2002) has made the case for SBTC-induced age-based wage inequality less certain after all, by the time the well-educated 1935-1945 birth cohorts began rapidly reversing the early retirement trend in the early 1990s, only 38.6 percent of workers aged 50 and older used a computer directly at work in 1993, compared to the 51.3% of their aged 40-49 counterparts that used a computer at that time. Perhaps those early 1950s cohorts aged in mid-40s in 1993 held onto SBTC s effects and fueled sustained growth in older wages throughout the late 1990s and early 2000s. However, falling college completion rates among baby boomers born between 1950 and 1965 suggests otherwise (Figure 10). Regardless, I test an empirical model attempting to link age-based wage inequality and SBTC based similarly on that presented in Autor, Katz, and Krueger (1998). I test the role of SBTC using the following regression: Y = β(%δ, computer use) i + ε i Y represents the two dependent variables used in the models presented in Table 7. These two measures include old-young overall relative wage bill and an absolute measure of older skilled worker wages. 10 The independent variable is the proportional change in the number of all intra-industry workers who reported using computers at work between 1989 and 2003 for every industry i for which data was available using the 1990 Current Population Survey list of classifications. However, neither of these models find a significant positive relationship. This is especially apparent when examining a scatterplot of old-young relative wage bill plotted relative to proportional change in intra-industry computer use as was done in Figure 4. After all, its line of best fit is nearly flat, further illustrating the lack of linear relationship. How to attempt to account for this failed SBTC hypothesis? It is helpful to return to Card and DiNardo s 10 The use of the words skilled and unskilled in this paper refer to measures associated with workers defined similarly as they are in Goldin and Katz (2008) I make one difference with the division of those workers with some college education by dividing them into skilled vs. unskilled based on their place in the 1976 income distribution for those at that education level. Those individuals at or above the 50 th percentile of the distribution were marked as skilled, whereas those below the 50 th percentile were marked as unskilled. comments on the problems and puzzles presented with the commonly held belief by economists in SBTC as a primary explanation for growth in overall wage inequality (2002). For one, they pointed out that much of the growth of computerization and greater investments in labor-saving machinery was prevalent throughout the 1970s and 1980s. Although computer workplace investment continues throughout the 1990s, wage inequality as found in Goldin and Katz s calculations still trended upward, but slowed considerably (Goldin and Katz, 2008). In other words, SBTC s effects on rising economic inequality reached its peak in the 1980s and trended downward across the 1990s. Yet this observation is inconsistent with the acceleration of age-based wage inequality observed in the late 1990s. Moreover, Card and DiNardo s observations suggest that the timing of SBTC tended to benefit younger, more technologically-adept workers. If anything, the evidence suggests that SBTC acted as a buffer on age-based wage inequality, not as a catalyst. b. Unskilled Older Workers and Differential Rates of Unionization SBTC, although it may not have an empirically plausible link with age-based wage inequality, is found to have induced significant changes in the structure of the labor market throughout the 1980s. A 2006 paper authored by Autor, Katz, and Kearney points to a major consequence of SBTC: a polarization of the U.S labor market between low-skill and high-skill occupations, and thus, low-pay and high-pay occupations, with middle-skill professions disappearing altogether. Reflected in the changing labor market structure is the decline of middle-skill, middle-wage occupations, many of which were intensive in routine tasks such as bookkeeping and repetitive production work, especially those in manufacturing, which tend to have higher levels of unionization. It is plausible that older workers remaining in these middle-skill occupations would have disproportionally benefited from union protections, which in turn would displace younger individuals entering the workforce into less-unionized occupations. A simple model testing the effect of differential rates of unionization between older and younger workers within industries does offer some additional explanatory power when the age-based relative labor measures used are restricted to that of this paper s specification of unskilled workers. Although no relationship is found when relative wage bill is used as the dependent variable, a weak yet significant Vol. 19 (2017) Nicholas Rogness p.8

relationship is found between relative wages and change in unionization once a positive effect of relative unskilled labor supply is controlled. (The independent variable used is the percentage change in unionization between 1990-2015 in each of the CPS-classified 194 industries for which data was available.) It is estimated that age-based wage inequality among unskilled older and younger workers increases by approximately 0.34% for every 1% increase in unionization in an average industry; however, this figure is associated with a standard error of 0.12% (Table 6). While differences in unionization between younger and older unskilled workers may explain their inequality, a plausible explanation for age-based wage inequality among skilled workers which in turn could provide answers for the problem of overall age-based wage inequality and other measures of inequities in labor market participation across age groups has not yet been presented. I offer an additional hypothesis in the following sub-section. c. The Old-Young Educational Attainment Differential Hypothesis The improbable link between SBTC and age-based wage inequality would seem to leave behind an unsolved puzzle. However, an alternative hypothesis is offered in comparing college completion rates across birth cohorts, which is supported by previous literature, CPS data, and U.S. Census data. Figure 10 presents an estimation of college completion rates for cohorts born between 1910 and 1985. College completion rates in the United States grew fastest among cohorts born approximately between 1935 and 1950 compared to any other birth cohort on record. This tail-end portion of the so-called Silent Generation would seem to be the single greatest contributor to age-based wage inequality. Blau and Goodstein (2010) attributed approximately these same birth cohorts to the reversal of the early retirement trend observed in the 1990s. In fact, they found that the growing levels of educational attainment among birth cohorts throughout the 1920s-1930s was the single strongest contributor to the reversal of declining labor-force participation rates that took place across most of the latter half of the twentieth century. Meanwhile, other plausible contributors were found to have a far weaker effect. Policy changes that reduced the generosity of Social Security benefits in the 1980s, for instance, were only found to have a significant relationship under certain measurement specifications. Their evidence from the Survey of Income and Program Participation data found that the transition from defined benefit (DB) guaranteed pension plans towards less employee-optimal defined contribution (DC) plans had an even weaker impact on rising LFPR, with an effect that measured at less than a twelfth of that attributed to rising inter-cohort educational attainment. The logic for the deferred retirement of more skilled older workers beginning in the 1990s, as Blau and Goodstein posited, can be explained using the labor-leisure model. This foundational labor economics model assumes all workers are interested in balancing all available time within a given period between work and leisure in such a way that maximizes the utility received from an optimal combination of consumption income (which includes total wage received dependent on number of hours worked added with any guaranteed non-labor, or non-work-related income) and leisure time. Then the worker will rationally choose a consumption-leisure bundle at the point where an indifference curve of a given utility level is tangent to his/her budget constraint, the slope and intercept of which is determined by the worker s hourly wage and a given level of non-labor income (Borjas, 2013). Per this model, the retired worker finds his/her optimal point where consumption income is equal to non-labor income (presumably in the form of a retirement pension, Social Security benefits, etc.) and all available time is spent on leisure. Thus, the early retirement trend prevalent until the mid-1990s can be thought of as greater numbers of the older population having found complete leisure as their optimal point. However, as the college wage premium rose throughout the 1980s and 1990s, the slope of the budget constraint for the average older worker at least for those skilled, of which status older workers became increasingly composed must have theoretically steepened in the 1990s such that the opportunity cost (simply put, measured as the value of whatever other activity could have been done in the time spent instead of the activity actually done) of early retirement exceeded the benefit derived from enjoying full-time leisure. While other factors may have also shifted older worker preferences to be more favorable towards work and reduced leisure (improving health, having a working spouse, etc.), the increasing opportunity cost of leisure for older workers driven by a shift towards a greater proportion of skilled labor was demonstrated as the single largest contributor to increasing LFPR (Harvard School of Public Health, 2004) and thus, as this paper has pointed out, overall old-young relative labor demand. Thus, it seems that the reversal of the early retirement trend and the growth in old-young relative Vol. 19 (2017) Nicholas Rogness p.9

labor supply was primarily driven by rising labor market outcomes for skilled older workers. CPS ASEC data also suggests that this class of skilled older workers has been growing faster than that of skilled younger workers since 1976, at least when measured in terms of the proportion of each group with baccalaureate and postbaccalaureate degrees. In 1976, less than 7% of older workers had just a bachelor s degree and nearly 5% carried just a graduate degree. Of all younger workers, by contrast, 15% held a bachelor s degree and 10% had a graduate degree. By 2015, older workers managed to catch up quite a bit in boosting education attainment. Attainment of a bachelor s degree for older workers increased nearly threefold to over 20%, and 15% held a graduate degree. Younger workers also boosted their educational attainment, but to a considerably smaller extent. 25% of them held a bachelor s degree as of 2015, and just 12% of them had a graduate degree. While older workers nearly tripled their proportion of graduate degree consistently over the decades, younger workers experienced a drop in this same number across the 1980s and 1990s, and have only been gradually recovering since then (Figure 5). We see this significant growth in educational attainment among the older workers also reflected in the growth of the old-young relative wage bill, especially among those skilled. Indeed, while skilled old-young relative wages did not rise as rapidly over this period (rising from 100 natural logarithmic value of 20.1 to 26.7), the post-1994 rise in inequality was more strongly driven by the rapid increase of skilled older workers choosing to remain in the labor force during early retirement age (Figures 6a and 6b). This difference from pre-1994 trends is demonstrated in Figure 7: whereas skilled older workers commanded only half of their younger counterparts portion of the total wage bill in 1976, the figures became roughly even by 2015. A breakdown of old-young relative labor supply further illustrates the crucial role played by growing numbers of skilled older workers in explaining agebased wage inequality. Older worker relative labor supply growth was fastest in highly skilled, non-routine, abstract-reasoning occupations such as those in computer science, management, or in financial work another consequence of rapidly growing educational attainment. 11 But I add a few caveats to this observation: 11 I modify Autor and Dorn s (2009) definition of a skilled, nonroutine worker to be limited to a worker whose occupational routinetask intensity is in the bottom two-thirds of the index s 1980 distribution and define highly skilled as earning an hourly wage of old-young relative labor supply also grew considerably in those occupations considered repetitive, or routinetask intensive jobs that include manufacturing and clerical work. It also grew in those jobs considered manual, or low skill and non-routine work. 12 This finding is interesting in the context of the observations of Autor and Dorn (2009), who found that the polarization of the labor market that took place across the late 20 th century resulted in the disproportionate displacement of workers aged 55-64 into low-skill jobs. As Figures 9a and 9b explain, this goes back to the catch-up effect in educational attainment exhibited among older workers, as demonstrated by similar trends of convergence of average number of years of schooling with younger workers across all different categories of occupations. 13 Perhaps unskilled older workers also saw their average hourly wages rise relative to younger workers not just due to their higher rates on unionization, but also by virtue of the fact that more of those with a college education went into those occupations and were given higher wages on the basis of their education level. Other research on historical trends in educational attainment further suggests a link between it and agebased wage inequality. Bound, Lovenheim, and Turner (2010), using a sophisticated simulation of longitudinal data collected from students, primarily attributed their observed decline in college completion between samples collected in 1972 and 1988 to educational supply-side factors such as student-faculty ratios over 102%, according to their full sample. 14 In other words, their findings suggested colleges and universities failed to expand academic resources at a rate that kept pace with demand pressures from the growing college-going population, especially among public colleges and universities. Other researchers have also illustrated the role of stretched educational resources in establishing a tight empirical link between birth cohort size and college attendance, retention, and graduation rates (Card and Lemieux, 2001; Bound and Turner, 2006). at least the 75 th percentile in the 1980 occupational wage distribution. 12 See Autor and Dorn (2009) for definitions of abstract, routine-task intensive, and manual occupations. 13 I limited this calculation to those who carried at least some high school education. High school dropouts were coded as having 8 years of education, high school graduates as having 12, some college as having 14, college graduates as having 16, and graduate degree holders as 18. 14 See Table 6 of Bound, Lovenheim, and Turner (2010). Vol. 19 (2017) Nicholas Rogness p.10

1. Empirical Results of the Cohort-Based Approach The hypothesized causality suggested by the previous literature is as follows: growing birth cohort size results in falling college completion rates, which in turn results in declining or stagnating wages and a growing age gap over time. I employ various weighted ordinary least squares (OLS) models, while using pooled (combined) Census, CPS, and ACS data to establish these links. 15 The variables used were grouped by 51 states (including District of Columbia) and by observation year ranging from 1977-2015. The first model, presented in Table 9, follows from the observed fluctuations in the birth cohort s ratio of college completion over college attendance by age 30 as plotted in Figure 12 in a reflection of trends observed in Figures 10 and 11. Similar to observed trends found in Census records from 1970-2000 (Bound, Turner, and Lovenheim, 2010), the ratio declines considerably across those born throughout the 1950s, spikes with late 1960s birth cohorts, and later declines with the start of the millennial generation in the early 1980s. In this regard, the ratio closely reflects the college completion rates presented in Figure 10. In employing the relative measure of the ratio, I account for the effect that fluctuating college attendance may have in confounding the model s results. When using the completion/attendance ratio and controlling for individual state and year fixed effects, 16 a significantly negative relationship is found for all variations of the dependent variable used in Table 7, including overall age-based wage inequality, skilled old-young relative wage bill, and skilled old-young relative labor supply. Intuitively, the completion/attendance ratio has the most statistically significant marginal effect on skilled oldyoung relative supply, with every 1 percent increase in the ratio decreasing skilled relative supply by an expected 0.06 log points. Transitively, this positive contribution to skilled relative supply also exacerbates overall age-based wage inequality for the reasons clear in Figure 6a as labor market demand transitioned to more skilled labor, older workers were relatively educationally well-equipped to follow suit. Even though age-based wage inequality among skilled workers remained relatively stable over the 1976-2015 period, 15 See Data Appendix for further descriptions on the data used. 16 The use of fixed effects in weighted OLS models is standard in econometric regression models that use panel data. In fact, this model followed a very similar procedure used in Card and Lemieux (2001), which can be looked at for reference. the larger proportions of older workers able to command higher wages by undertaking skilled labor starting in the late 1980s contributed to the continued growth of overall age-based wage inequality that took place after 1990. I next extend this analysis to a model similar to that presented in Card and Lemieux (2001). Namely, I test the role of birth cohort size of the young group, local prime-age male unemployment rates, and tuition cost at public 4-year colleges and universities. I estimate young cohort size as the average of the natural logarithm of the size of all young cohorts captured within a given observation year. Prime-age male unemployment rates are calculated from CPS data and refers to the unemployment rate of men aged 25-54. This measure is lagged one year in order to estimate the effect of the wage-deflating portion of the business cycle on age-based wage inequality; after all, macroeconomic theory asserts that price and wage level will decrease once an economic trough and its peak unemployment rate is overcome. However, in following the logic of Card and Lemieux (2001) that local unemployment rates measured at age 18 will affect college completion rates measured in subsequent years, I later present an alternative measure of unemployment that carries a lag of 12 years. Tuition is measured as the average log of reported in-state undergraduate tuition cost at public 4-year colleges and universities from 1980-2015. I also include a variable that attempts to capture the unique effect of the 1990s economic expansion in exacerbating old-young wage differentials: that is, a measure of the percentage of those employed in industries numbering among the top 20 gaining the largest numbers of employment throughout the 1990s. 17 Further information on the sources and methodology used to collect and manipulate this data in establishing these models can be found in the Data Appendix. Table 8 presents the results of the first regression used with this set of variables. It excludes the effect of log tuition, and includes a dependent variable of skilled age-based wage inequality in addition to the other dependent variables employed in Table 7. Young cohort size demonstrates a consistently positive effect on overall age-based wage inequality, skilled relative wage bill, and skilled old-young relative supply. The marginal effect of young cohort size results in estimated increases of 3.8%, 12.0%, and 17.6% respectively. Lagged 17 These Top 20 industries are derived from Bureau of Labor Statistics data. Further description of the data and a list of these industries are included in the Data Appendix. Vol. 19 (2017) Nicholas Rogness p.11

unemployment has a statistically significant positive effect on the skilled relative wage bill and skilled relative supply at the 0.05 level. The percentage employed in 1990s large hiring volume industries only positively significantly contributes to skilled relative supply. As also observed in Table 7, the relationships are generally found to be strongest in the case of skilled old-young relative supply, and for intuitive reasons. The skilled relative supply measure acts as a direct reflection of the differences in college completion rates between the young and old, whereas the skilled relative wage bill interacts this number with the less-variant measure of skilled age-based wage inequality. Overall age-based wage inequality is a step further removed from the strong causality afforded to skilled relative supply, as it also encompasses the growth that has taken place in unskilled age-based wage inequality. Later regressions presented in this section illustrate the differing nature of unskilled age-based inequality growth measures compared to those observed among the skilled. A slightly altered model in Table 9, which includes a 12-year lagged measure of the average natural logarithm in-state 4-year public college tuition, 18 creates rather perplexing results. When comparing the dependent variables used which in this case is limited to overall age-based wage inequality, skilled relative wage bill, and skilled old-young relative supply the relationships that are estimated with this added variable are reversed from those found in Table 8. Log young cohort size in the Table 9 model are characterized as having a significantly negative relationship. Lagged unemployment carries a significantly positive relationship across all three specifications at the 0.10 level, and employment percentage in the 1990s large hiring volume industries is weakly positively significant in the case of skilled relative wage bill. Most crucially, however, 12-year lagged log tuition carries no statistically positive relationship with any of age-based inequality measures used in the model. These unexpected results are likely a consequence of the large lag placed on the measure of college tuition cost, as this restricts the model to observations after 1992. In order to address the probability of spurious correlations possible throughout Tables 7-9, I test some of these same models on measures of unskilled agebased relative labor measures. Namely, I use unskilled 18 A 12-year lag was used so as to estimate the effect rising tuition may have had when the average-aged young person in our analysis (when he or she would be at age 30) was 18 and presumably entering college around that age. age-based wage inequality, unskilled relative wage bill, and unskilled old-young relative supply as dependent variables for the models presented in Tables 10 and 11. Table 10 acts as an imitation of Table 8, which presents a surprising positive relationship between log young cohort size and all three dependent variables on a scale similar to Table 8. Employment percentage in the 1990s large hiring volume industries have a relatively large positive marginal effect on unskilled age-based wage inequality and unskilled relative wage bill. Lagged unemployment rate has no significant effect on any of the measures. The puzzling relationship between young cohort size and unskilled old-young relative labor measures may raise doubts of the relationships found in Tables 7-9. Table 11 regresses the various unskilled old-young relative labor measures to the young completion/attendance ratio in order to address these concerns. Unlike in the case of skilled old-young relative labor measures, no significant relationship is found among the pairings of these variables, suggesting that the correlation between young cohort size and unskilled old-young relative labor measures is spurious. The final model employed in this section uses an alternative unemployment variable that reflects Card and Lemieux s hypothesized causality between local prime-age unemployment rates and college attendance and retention (2001). They reason that an increase in localized employment will induce greater incentives towards college enrollment and retention when measuring unemployment for every observation year when the birth cohort reaches age 18. Thus, for my purposes, this requires a 12-year lagged variable so as to measure unemployment s impact set at the age midpoint of the young group at age 30. Table 12 extends the model used in Table 9 to include unskilled relative wage bill and unskilled old-young relative supply as additional dependent variables, as well as the substituted 12-year lag unemployment rate. A significantly positive relationship under the unskilled old-young relationship measures is found; however, this comes with a negative effect found in log young cohort size. Similar to Table 9, log young cohort size in Table 12 s model appears to suffer from a downward bias when the observation years included in the model are restricted, a consequence of the recording of CPS data in groupings of state prior to 1977. 2. Inferences and Conclusions from the Cohort- Based Approach Vol. 19 (2017) Nicholas Rogness p.12

Based on all these results, I reach the following conclusions. The first is that the CPS data suggests that young cohort size acts as the single-most significant contributor to age-based wage inequality, or at least across its dimension of skilled old-young relative labor measures. This is plausibly caused by the stretching of educational resources induced by the growing size of the young birth cohort over time and declining state expenditures on higher education per student. As a result, while attendance for incoming students continued to rise, college completion rates struggled to keep pace. Differing trends in educational attainment over time put older workers in the better position to take advantage of growing labor demand for skilled labor throughout the 1980s and 90s, and older worker relative wages grew subsequently. The other variables employed in the models of Tables 7-12 may have some role in the growth of agebased wage inequality and other old-young relative labor measures, but these are subject to specification. This is particularly the case with prime-age unemployment, which demonstrates weak positive significance to growth in overall age-based wage inequality, skilled relative wage bill, and skilled relative supply. This finding suggests that skilled younger workers tend to be relatively worse off when overall wages decline at the beginning of an economic recovery. The positive effect of employment percentage in the large hiring volume industries of the 1990s also tends to be subject to specification, which could be viewed either as inconsistent or intuitive. Among skilled old-young relative labor measures, employment in the 1990s large hiring volume industries only has a significantly positive effect on skilled old-young relative supply. This is perhaps a reflection of those 1990s large hiring volume industries that disproportionately required workers with college degrees: hospital, colleges and universities, personnel supply services, management and public relations services, and computer and data processing process services (Hatch and Clinton, 2000). Among unskilled old-young relative labor measures, a different pattern emerges. Instead of contributing to growth in relative supply, employment percentage in large new-hire volume industries is strongly associated with the growth in rising old-young relative wages among the unskilled, which in turn leads to the positive relationship found with the unskilled relative wage bill. This finding is consistent with the rising age-based wage inequality among the unskilled observed in Figure 6a, and suggests older workers were also attracted to employment in relatively unskilled high-growth industries of the 1990s as higher wages were offered as an inducement. However, rising unskilled older worker relative wages at this time may simply have become a reflection of their greater experience working in jobs they already held in such industries prior to the expansion of the 1990s. The nature of these presumably less-skilled, high-growth industries also notably demonstrates that their employment and wage growth grew out of rising consumption afforded by the 1990s expansion: these included eating and drinking places, trucking and courier services, department stores, and grocery stores. The fluctuation of the completion/attendance ratio adds a final caveat to the unique role of the 1990s in exacerbating age-based wage inequality. With those born throughout late 1960s, this ratio and the college completion rate shot up considerably (Figures 12 & 10). This is likely a reflection of the growing computer and technology industry at that time, which incentivized disproportionately larger numbers of entering undergraduate students to pursue computer science and other related bachelor s degrees (Hatch and Clinton, 2000). Presumably, this caused a spike in the returns to college education, a momentary return towards higher graduation rates among the 1965-1970 birth cohorts, and growth in younger worker wages. However, given the rising college completion rates of the 1930-50 birth cohorts and the more rapid transition of older workers towards higher educational attainment, older wages still outpaced younger wages and age-based wage inequality was furthered. Since the end of the 1990s expansion, the completion/attendance ratio exhibited an overall downward trend with the 1970s and later birth cohorts, contributing significantly to the stagnant younger worker wages observed since 2002 (Figures 12 and 1). V. Conclusions Will Age-Based Wage Inequality Be of Concern Much Longer? As presented here, my explanation for rising relative wages for older workers points primarily to differences in educational attainment across cohorts. Over the period of analysis, the growth in age-based wage inequality shifted dramatically towards being driven by demand growth for older skilled workers beginning in the 1990s. Even by this token, skills-biased technological change throughout the 1980s and 1990s could not have played a significant role in the old-young wage inequality phenomenon. Relatively lackluster Vol. 19 (2017) Nicholas Rogness p.13

educational attainment growth among younger birth cohorts over this period primarily exacerbated this wage differential. A less obvious contribution of SBTC, or another long-term economic factor, towards age-based wage inequality was its downward role in worker unionization as greater numbers of younger workers were displaced from routine, often heavily unionized occupations, this left the remaining older unskilled workers with more room for economic opportunities and a mid-level version of wage earnings. A much tighter empirical link seems to lie in examining fluctuations in the college completion/attendance ratio across birth cohorts, as explained in Section V. Overall age-based wage inequality, skilled old-young relative wage bill, and skilled old-young relative supply are all negatively correlated with the completion/attendance ratio of birth cohorts measured at age 30. Growing cohort size acts as the single largest contributor to age-based wage and labor supply inequality, especially among skilled workers. Lagged unemployment also appears to be associated positively with age-based inequality measures, but the causality of this relationship is unclear, requires further analysis, and is not as strongly supported by previous literature. Employment percentage in high-growth industries of the 1990s has limited success at capturing the unique effects of the 1990s economic expansion in exacerbating age-based wage inequality, with findings of significant contributions to skilled old-young relative supply and unskilled old-young relative wages. Inferences in estimating the effect of tuition cost and alternative unemployment measures are problematic due to the limited data available. Historic trends in the college completion rates and the completion/attendance ratio carry important implications. The decline and stagnation of both measures among early-1950s to 1960s birth cohorts in other words, most of the baby boomer generation and some of Generation X suggests that age-based wage inequality may stabilize and may even reverse in the near future. It has already been observed that old-young relative wages have largely remained unchanged in the 2005-2015 period (Table 3, row 7) as the latter half of the baby boomer cohort has reached early retirement age. Discouraging to this possible development, however, is the observed decline in the college completion/attendance ratio since the start of the millennial generation with the 1980s birth cohorts. Given the rising attendance/completion ratio plotted over the course of those birth cohorts considered Generation X, or those born approximately between 1965 and 1980, an economic generational gap could emerge between it and the generation following the millennials, known widely as Generation Z. Regardless of the true trends of the age-based economic gap observed over time, a sense of economic frustration held by the millennial generation against baby boomers is likely to have social and political consequences for the foreseeable future as accessibility to higher education and training for skilled labor continue to lack especially for college programs that result in onerous student loan debt loads and few guarantees of post-graduation job prospects. These circumstances, combined with millennials lack of experience with the capitalism-communism divide that shaped the Cold War, has influenced nearly an entire generation to carry more favorable views of policies in line with European-style social democracies but have been historically unpopular with the general public. A 2015 YouGov poll found that 43% percent of under-30 respondents held a favorable view towards socialism, whereas only 27% of an older category of 30-44 year olds gave this same response. While this may simply indicate changing politico-cultural attitudes towards the socialism label, the college-student base of support for the Occupy Wall Street movement and other movements stressing greater economic redistribution and equality suggest that millennials differ significantly in their policy preferences compared to previous generations. How this will result in concrete policy changes in subsequent decades, however, remains uncertain. Registered Democrats under age 30 supported self-labeled democratic socialist presidential candidate Bernie Sanders by an overwhelming proportion of 70% in the party s state primary and caucus contests between January-April 2016 (The Data Team, 2016). At the very least, America s shifting demographics are more favorable to change: the millennial generation was projected to make up the plurality of the country s electorate starting with the 2016 presidential election. In the wake of the potential political consequences, I wish that these underline the salience of this particular labor market issue and its interest to future researchers. Future papers on this topic may wish to provide additional insight with models providing panel data across different specifications aside from the state-level. Studies of longitudinal data on this topic from sources such as the National Longitudinal Survey of Youth (NLSY) may also bear fruit. As understanding on the topic of age-based wage inequality is presently limited Vol. 19 (2017) Nicholas Rogness p.14

by this single study, its advancement as a popular field of economics research would benefit from a variety of various approaches and perspectives. Data Appendix The majority of my analysis applies microdata from the U.S. Annual Social and Economic Supplement (ASEC) collected over the 1976-2015 period using the IPUMS database. This initial microdata provided information on age, sex, annual wages, weeks worked per year, usual hours worked per week, level of educational attainment, union membership/coverage, and other variables of interest. The accuracy of ASEC is virtually uncontested by researchers in most of its aspects, as it consists of tens of thousands of respondents across the country and produce reasonable statistical estimates at the state, commuting zone, and metropolitan area levels of analysis ( Frequently Asked Questions, N.d.). Moreover, its primary purpose of examining labor force trends in relation to poverty and income makes the database particularly relevant to this paper s motivation. Still, the nature of ASEC data collection methods does warrant a few critiques as it relates to income inequality measurement its practice of top-coding fails to capture where some of the most rampant growth of the phenomenon has taken place in recent decades at the very top end of the national income distribution. 19 (Of course, recent research by Thomas Piketty in his seminal tome Capital in the Twenty-First Century and in earlier research papers use tax-return data to capture a truer sense of nature of growing income inequality.) Thus, it is possible that age-based income inequality is actually quite higher than is reflected in the ASEC microdata (also possibly lower, although it seems a more distant possibility). For the presented industry-level analysis, I utilized other CPS data. In particular, I merged labor supply and relative wage data from the ASEC dataset with survey data on computer use in the workplace from the October CPS in 1989 and 2003. As for my unionization analysis, I merged industry-level labor supply and wage inequality data from ASEC with October CPS data on unionization from 1990 to 2015. My analysis also relies substantially on data from U.S. Censuses and the American Community Survey (ACS). In particular, my estimates of birth cohort size 19 See arguments in American political economic literature from McCarty, Poole, and Rosenthal (2006) (pages 122-123). rely on data from the 1910-2010 U.S. Censuses and the 2014 ACS collected through USA IPUMS. The methodology employed for estimating cohort size is followed analogously from that used in Card and Lemieux (2001). 20 Estimates of college completion and attendance are obtained from 1940-2010 U.S. Census and 2014 ACS data. These estimates utilize the same methodology outlined by Acemoglu and Autor (2012). 21 A few notes are also offered on the definitions used throughout the paper. I define young and old workers (aged 25-34 and aged 55-64, respectively) as used throughout this paper was applied in analysis of ASEC, Census, and ACS data. The microdata in all datasets concerning wage data was restricted to those who reported working a positive number of weeks within the last year. Records from respondents living in group quarters were also omitted. The initial supply and demand regressions consisted of collapsed time series data. Mean hourly wage was calculated by dividing total wage earnings for the survey year by the number of reported hours worked by the respondent during the previous year (number of weeks worked times average number of hours worked per week) and adjusting for inflation according to CPI figures. Finally, hourly wages for both the time series and panel models was weighted to adjust the sample to resemble the trends of current national demographics (age, gender, race, etc.). The models in Tables 9-14 differ in that they employ weighted OLS regression, using the number of observations counted for each state-year cell as a weight. They also control for state and year fixed effects. Unemployment rate in all cases is the average unemployment rate of men age 25-54 in the state in the year of observation. Tuition is the average amount of instate undergraduate tuition at state 4-year public colleges and universities and is collected from IPEDS from 1980-2015. Percentage employed in high-growth industries of the 1990s includes all individuals identified as working in any of the top 20 industries gaining the most jobs between 1989-99 as identified by Hatch and Clinton (2000). This list includes those presented in Table A. References 20 See Chapter 9, Page 468 of Risky Behavior Among Youths: An Economic Analysis. 21 See Figure 7 of Acemoglu and Autor (2012) for explanation of methodology. Vol. 19 (2017) Nicholas Rogness p.15

Autor, D. H., Katz, L. F., & Kearney, M. (2006, May). The Polarization of U.S. Labor Market. American Economic Review, 96(2), 189-194. Autor, D., & Acemoglu, D. (2012, July). What Does Human Capital Do? A Review of Goldin and Katz's The Race Between Education and Technology. Journal of Economic Literature, 50(2), 426-463. Autor, D., & Dorn, D. (2009, May). This Job is "Getting Old": Measuring Changes in Job Opportunities using Occupational Age Structure. American Economic Review, 99(2), 45-51. Autor, D. H., Katz, L. F., & Krueger, A. B. (1998, November). Computing Inequality: Have Computers Changed the Labor Market? The Quarterly Journal of Economics, 113(4), 1169-1213. Blau, D., & Goodstein, R. M. (2010, March). Can Social Security Explain Trends in Labor Force Participation of Older Men in the United States? Journal of Human Resources, 45(2), 328-363. Borjas, G. J. (2013). Labor Economics (6 th ed.). New York, NY: McGraw-Hill. Bound, J., & Turner, S. (2006, June). Cohort Crowding: How Resources Affect Collegiate Attainment. Journal of Public Economics, 91(5-6), 877-899. Bound, J., Lovenheim, M., & Turner, S. (2010, July). Why Have College Completion Rates Declined? An Analysis of Changing Student Preparation and Collegiate Resources. American Economic Journal: Applied Economics, 2(3), 129-157. Card, D., & Lemieux, T. (2001, May). Can Falling Supply Explain the Rising Return to College for Younger Men? A Cohort-Based Analysis. The Quarterly Journal of Economics, 116(2), 705-746. Card, D., & Lemieux, T. (2001). Dropout and Enrollment Trends in the Post-War Period: What Went Wrong in the 1970s? In National Bureau of Economic Research, & J. Gruber (Ed.), Risky Behavior among Youths: An Economic Analysis (p. 439). 482: University of Chicago Press. Aaron, J. M. Lindsay, & P. S. Nivola, Agenda for the Nation (pp. 17-60). Washington, DC: Brookings Institution Press. Frequently Asked Questions. (N.d.). U.S. Census Bureau. Retrieved from http://www.census.gov/programsurveys/cps/about/faqs.html. Fry, R., Cohn, D., Livingston, G., & Taylor, P. (2011). The Rising Age Gap in Economic Well-Being. Pew Research Center, Social and Demographic Trends. Washington, DC: Pew Research Center. Goldin, C., & Katz, L. K. (2008). The Race Between Education and Technology. Cambridge, MA: Harvard University Press. Harvard School of Public Health. (2004). Reinventing Aging: Baby Boomers and Civic Engagement. Cambridge, MA: Center for Health Communication. Hatch, J., & Clinton, A. (2000, December). Job growth in the 1990s: a retrospect. Monthly Labor Review, pp. 3-18. McCarty, N., Poole, K. T., & Rosenthal, H. (2006). Polarized America: The Dance of Ideology and Unequal Riches. Cambridge, MA: Massachusetes Institute of Technology. Piketty, T. (2014). Capital in the Twenty-First Century. (A. Goldhammer, Trans.) Cambridge, MA: The Belknap Press of Harvard University Press. The Data Team. (2016, April 27). Young v old votes for Bernie and Hillary in the 2016 primaries. The Economist. Retrieved from http://www.economist.com/blogs/graphicdetail/ 2016/04/daily-chart-19 YouGov U.S. (2015, March 11). YouGov Socialism Poll Results. Retrieved from YouGov U.S.: http://cdn.yougov.com/cumulus_uploads/docum ent/3csd07d2dd/tabs_opi_socialism_20150508. pdf DeLong, J. B., Goldin, C., & Katz, L. F. (2003). Sustaining U.S. Economic Growth. In H. J. Vol. 19 (2017) Nicholas Rogness p.16

Acknowledgments I would be remiss as a budding researcher not to take an opportunity to thank all those who have mentored and inspired me along this first journey of mine into the realm of economics research. My immediate thanks goes to Professor Giovanni Peri, whose dedication towards guiding myself and my honors program colleagues went above and beyond his call of duty given his concurrent commitments as a graduate-level course instructor and 2015-16 Chair of the UC Davis Department of Economics. It is gratifying to reflect at how, since I started the process of this senior honors thesis late in my junior year at UC Davis, my image of Professor Peri has transformed from that of a mere public university administrator into that of a true mentor and passionate investor in my work. undergraduate feat (including one who suggested that I include this acknowledgement section in the first place). Much appreciation also goes to my friends turned surrogate-family in Agape, whose dance performance practices and social hangouts were a blessed reprieve from my countless hours of number-crunching these past several months. (Moreover, I feel it my duty to see that their prayers for this project do not go to waste.) My final and most important acknowledgement goes to my father, mother, and the rest of my eight-sibling family for supporting my college education and personal development with unconditional love and support at every turn of my life journey. I must also thank Professor Janine Wilson for graciously taking the time to meticulously review, critique, and ultimately co-sponsor my work. Combined with comments from Professor Adrienne Hosek of the Department of Political Science, I was able to receive crucial feedback on my methodology and conclusions from a variety of perspectives despite these individuals own tremendous time constraints. That they shared in my enthusiasm for the paper s motivation furthers my gratitude for their contribution. I also thank all the community college instructors who introduced me to the process of college-level reasoning and problem solving. I especially wish to acknowledge Professors Noelle Oliver and Steve McFaul, whose classes taught me the basic writing and rithmatic (statistical) skills that I came to use day-today as a UC Davis upperclassmen and built up to my writing of this paper. An especially special mention goes to Professor David Reese, whose quirky and unconventional teaching style took my three semesters of introductory political science classes with him far beyond simple rote memorization of key terms and persons, but into normative in-class discussions of truth, justice, and occasionally the very meaning of life. Moreover, his particular emphasis on political economy, income inequality, and the well-being of the millennial generation sparked my interest in the problem of agebased wage inequality presented here. Lastly, I must thank the friends and family who provided crucial emotional and spiritual support throughout my research process. I particularly thank my few, proud colleagues in the UC Davis Economics Honors Program for encouraging one another in this Vol. 19 (2017) Nicholas Rogness p.17

Table A: List of 1990s Large Hiring Volume Industries Hatch and Clinton (2000) CPS Industry Code Industry Name Rank: 1 731 Personnel Supply Services 2 641 Eating and drinking places 4 732 Computer and data processing services 6 741 Misc. business services 7 812 Offices and clinics of medical doctors 8 831 Hospitals 9 810 Misc. amusement and recreation services 10 892 Management and public relations services 11 832 Nursing and personal care facilities 12 840 Home health care services 13 870 Residential care services 14 410 Trucking and courier services 15 862 Individual and family services 16 591 Child day care services 17 850 Department stores 3, 5, 18 900-31 State and local government 19 842 Colleges and universities 20 601 Grocery stores Vol. 19 (2017) Nicholas Rogness p.18

Figure 1: Log Hourly Wages for Old and Young Age Groups, 1976-2015 (Employment Weighted) Table 1: Labor Force Participation Rates Compared Between Younger and Older Age Groups, Selected Years Year LFPR, Ages 25-34 LFPR, Ages 55-64 1976 73.7% 57.2% 1980 78.5% 56.4% 1985 81.1% 55.2% 1987 81.4% 54.6% 1990 82.2% 55.6% 1993 81.7% 56.8% 1995 82.0% 57.8% 2000 83.8% 60.4% 2005 81.0% 64.0% 2010 81.1% 66.4% 2015 79.6% 65.1% Vol. 19 (2017) Nicholas Rogness p.19

Figure 2: Old-Young Wage Inequality and Relative Supply, 1976-2015 Figure 3 Vol. 19 (2017) Nicholas Rogness p.20

Table 2: Estimation of Elasticity of Substitution Between Older and Younger Workers (σoy) (1) Relative Supply -0.195 ** (-2.90) Dummy, 1995-99 0.128 *** (6.26) Dummy, 2000-05 0.123 *** (6.26) Dummy, 2005-10 0.207 *** (6.38) Dummy, 2010-15 0.296 *** (6.71) Constant -0.0716 (-1.18) Observations 40 t statistics in parentheses * p < 0.05, ** p < 0.01, *** p < 0.001 Table 3: Changes in Old-Young Relative Wages, Relative Supply, and Relative Supply (100 Annual Log Changes) Period Relative Wage Relative Supply Relative Demand (σoy=4) Relative Demand (σoy=5) 1976-1994 0.89-1.76 1.79 2.68 1995-2015 0.54 3.35 5.51 6.05 1976-85 1.70-2.74 4.05 5.75 1985-95 0.56-0.37 1.88 2.44 1995-2005 0.98 3.60 7.54 8.52 2005-15 0.00 3.03 2.99 2.98 Vol. 19 (2017) Nicholas Rogness p.21

Table 4: Statistical Significance of Old-Young Relative Supply and Relative Demand Calculations (Old- Young Relative Wages as Dependent Variable) (1) (2) Relative Supply -0.0794 ** -0.121 ** (-2.99) (-3.21) Time 0.00814 *** 0.00600 *** (17.16) (6.30) Time Post-1994 0.00223 * (2.23) Constant -0.0424-0.0651 (-1.60) (-2.01) Observations 40 40 R 2 0.917 0.929 t statistics in parentheses * p < 0.05, ** p < 0.01, *** p < 0.001 Table 5: Relationship Between Age-Based Wage Inequality and Increased Computerization Change in % of Workers w/ Computers (1) (2) Change in Relative Wage Bill Growth in Older Worker Wages 0.232 0.329 (0.48) (0.79) Constant 0.756 *** 0.199 ** (8.40) (3.25) Observations 171 175 R 2 0.002 0.009 t statistics in parentheses * p < 0.05, ** p < 0.01, *** p < 0.001 Vol. 19 (2017) Nicholas Rogness p.22

Figure 4: Relationship Between Old-Young Relative Wage Bill Growth and Proportional Change in Level of Computerization (Change Across Industries, 1989-2003) Table 6: Relationship Between Unskilled Age-Based Wage Inequality and Differential Intra-Industry Unionization Trends (Change between 1990-2015) % Change in Unionization 100 Change in Old-Young Unskilled Relative Supply (1) (2) Unskilled Relative Wage Bill Old-Young Unskilled Relative Wages 0.000924 0.00369 ** (0.50) (2.97) 0.145 * (2.07) Constant 0.908 *** -0.0500 (10.63) (-0.55) Observations 194 194 R 2 0.000 0.039 t statistics in parentheses Vol. 19 (2017) Nicholas Rogness p.23

* p < 0.05, ** p < 0.01, *** p < 0.001 Vol. 19 (2017) Nicholas Rogness p.24

Figure 5: Percentage Attainment of Baccalaureate and Post-Baccalaureate Degrees by Age Group Vol. 19 (2017) Nicholas Rogness p.25

Figure 6a: Trends in Skilled and Unskilled Old-Young Wage Inequality (Log Ratio), 1976-2015 Figure 6b: Trends in Skilled and Unskilled Old-Young Relative Supply (Log Ratio), 1976-2015 Vol. 19 (2017) Nicholas Rogness p.26

Figure 7: Old-Young Relative Wage Bill Comparison (Log Ratio): Overall, Skilled, and Unskilled Figure 8: Change in Old-Young Relative Labor by Occupation Type (Log Ratio) Vol. 19 (2017) Nicholas Rogness p.27

Vol. 19 (2017) Nicholas Rogness p.28

Figures 9a and 9b: Mean Years of Schooling by Age Group and Occupation Type Vol. 19 (2017) Nicholas Rogness p.29

Figure 10 Figure 11 Vol. 19 (2017) Nicholas Rogness p.30

Figure 12 Table 7: Regression of Old-Young Relative Wage, Skilled Relative Wage Bill, and Skilled Relative Supply Measures on Ratio of Young Completion/Attendance Ratios Young Ratio, Completion/Attendance 100 Dependent Variable (1) (2) (3) Skilled Relative Wage Bill Overall Age- Based Wage Inequality Skilled Old- Young Relative Supply -0.0094 + -0.057 *** -0.0638 *** (0.0055) (0.0094) (0.0116) Constant 0.476 2.275 *** 2.027 *** (0.290) (0.493) (0.605) Observations 1989 1989 1989 R 2 0.522 0.806 0.829 Standard errors in parentheses Vol. 19 (2017) Nicholas Rogness p.31