Earnings Mobility: Determining What Measure to Use

Transcription

1 Center for Urban Studies October Working Paper Series, No. 10 Earnings Mobility: Determining What Measure to Use

2 Center for Urban Studies October Working Paper Series, No. 10 Earnings Mobility: Determining What Measure to Use Stephen J. Rose Senior Research Economist, ORC Macro A version of this paper was presented on February 27, 2002 at a seminar co-sponsored by the Center for Urban Studies and the Fraser Center for Workplace Issues, College of Urban, Labor and Metropolitan Affairs, Wayne State University

3 Introduction Mobility is a dynamic concept that looks at changes over time. As such, it is central to the concept of opportunity and the ability to rise beyond one s initial circumstances. In many research papers, change has been measured by comparing two snapshots at different points in time. These analyses present a sense of overall change but do not reveal anything about individual experiences. It is quite possible for overall growth to be slow while most people individually are experiencing gains in their standard of living as they age: think of a escalator in which the distribution of the heights of the stairs do not change but individuals pass through and move up to the next floor. To measure mobility, a longitudinal data set that follows the same people over many years is needed. There are fewer data sets available and there are many methodological and conceptual issues to work out. Gottschalk and Danziger (1998) reflect the current conventional wisdom that mobility is substantial, does not seem to change that much over time, and is of roughly similar magnitude in different countries. Cox and Alm (1995, 1999) and Armey (1995) take these findings a step further by arguing that rising inequality (as measured in comparative static studies) is not as important an issue as it may appear. If those who are poor in one period increase their earnings dramatically in following years, then the issue is one of volatility rather than permanent low income. In this short paper, I will show that there is a big difference in looking at long periods of labor market experiences (up to 15 years) than just using a single year s data. In particular, the experience of male and female workers is much different, and much of the reason for this has to do with labor force interruptions as workers who are out of the labor force for a year or more have substantially lower earnings when working than comparable workers who have no interruptions. And, women are twice as likely as men to experience interruptions. In longitudinal data, there is a lot of yearly volatility and it is important to develop methodologies that take this into account. For example, many workers have isolated bouts of very low earnings (less than $15,000) even though very few men in their prime-earning years with strong long-term labor force ties average under $25,000 (in 1999 dollars) a year. Women workers, on the other hand, tend to be in gender-segregated occupations and make up 90 percent of those with a persistent labor force presence but very low earnings. Therefore, the mobility experiences of women and men are very different because many women are starting over and then have strong gains after their re-entry. For male workers, however, the notion that most workers get steady earnings gains as they build seniority and job-specific skills is not now true as 40 percent of male workers had either flat or negative real earnings gains over the 15-year period from 1981 to 1995 (the last years for which there is good longitudinal data). Furthermore, although there is not good data about earlier years, there is indirect evidence that mobility for male workers has not stayed the same but declined over time. This work is part of a larger book project, and I can only present some summary tables and conclusions. In order to cover more ground, the text will be kept to a minimum. The data come from the Panel Study on Income Dynamics (PSID), the longest ongoing longitudinal data set in the world. Because of declining funding, the most recent year of the final release data is 1992! I have been able to use the early release data to do computations through There is 3

4 new early release data through 1998, but I have yet to solve all difficulties in making this data usable. However, the initial results indicate that the experiences of 1995 through 1998 are not considerably different from the 1981 to 1995 period that I study here. Conceptual, Data and Methodological Issues The term mobility has many popular usages. On the one hand, it is used to reflect movements between generations. In particular, as immigrants have come to this country, they have faced economic hardships but have seen their children become more successful. In this analysis, this issue will not be addressed. On the other hand, there are movements within a single person s lifetime, e.g., starting at the bottom and moving up. This kind of social movement can be looked at two ways. First, there is the peer or cohort dimension: relative to those in your own generation, do you change places? In other words, if you start at the bottom (or middle or top), do you remain in the same relative place when you are older? Second, there is the life cycle effect associated with earnings changes (hopefully increases) as workers age and gain experience and seniority. At the beginning of one s career, young people move from temporary to permanent employment as they finish their education. Their first years may be rocky before they find their best labor market position. After they find a good niche, they gain seniority, learn job-specific skills, and move up the occupational ladder. Thus, the conventional wisdom is that earnings gains are usually fastest from 20 to 30 years old, slow down (but remain positive) between 30-50, and flatten out or turn negative as workers approach retirement. The difference between life cycle and cohort effects is often presented as the difference between absolute and relative changes. But this can be confusing because over one s career, real earnings tend to rise but so too does one s relative position in the overall economy. This is not the relative change that is most commonly used in the mobility literature in which relative change refers to movements within the same cohort. These cohort changes are usually measured by arraying all individuals from lowest to highest earners and dividing the population into equal groups of five quintiles ; movements are quantified by transition matrices that show beginning and end quintiles. The percent that are on the diagonal represent those who stay in the same quintile and have no mobility. While this may seem straightforward, there are several difficulties that undermine the usefulness of this approach and have led many researchers to misrepresent their findings: their numbers are correct, but their conclusions are incorrect because they are capturing different effects than they proclaim. First, there is the problem of confounding cohort and life cycle effects. In a Treasury Department study in 1992, a longitudinal IRS file of tax records was constructed and revealed that most people in the lowest income quintile were no longer there 10 years later. They concluded that this showed that mobility was an overlooked factor in discussing the conditions of rich and poor. However, their data had the anomalous characteristic that 20 percent of their study group began in the bottom quintile but only 8 percent were in the bottom quintile (and 32 percent in the top one) at the end. This occurred because their methodology stacked the deck. In determining who was in the bottom quintile at the end of the period, the authors looked at 4

5 the universe of all taxpayers during that year. Some of these taxpayers were young people who were not in the original year and had low incomes that put them in the bottom quintile; the earnings paths of these people were not included in the analysis. Thus, the quintile transition matrix was not a true comparison of relative changes within the same cohort. In the later period, the quintile cut-offs were determined by including people from the adjoining younger cohorts. Because of the life cycle effect, these added people filled the lower quintiles, leaving the older cohort for which the transitions were being measured to be disproportionately in the higher quintiles. Cox and Alm (1995, 1999) run into another problematic life cycle effect. In their approach, they track male workers over a long period. However, the earnings of the bottom quintile in the initial year are remarkably low (only $4,000); in the last year, this group has approximate average earnings of $30,000. This remarkable gain is based on including teenagers in the study group. Because they are mainly part-timers who have not really begun their career, the earnings gain presented merely captures their entrance into the labor market rather than high levels of mobility throughout one s career or movements up within the same cohort. Generalizing the movement of teens into the labor force as a sign of a dynamic American labor market seems inappropriate. This study highlights the importance of choosing appropriate age groups to avoid the strong growth during the initial entrance into the labor force and the earnings decline during the onset of full or partial retirement. Second, there is the problem of reversion to the mean. Sawhill and Condon (1992) reported that the poor did better than the rich in the 1980s. This was a surprising finding from liberal researchers given that virtually all other studies had found that poor had done worse than the rich in the 1980s. Their approach looked at the average earnings growth of workers over 10 years based on their starting quintile position in For those who started in the lowest quintile, their average earnings grew by 80 percent while those who started in the top quintile had earnings gains of only 5 percent. The issue here is how to determine who is rich and poor. If the stratifying principle had been based on the last year rather than the first year, the results would have been reversed and the top quintile would have had the high growth level while the bottom quintile would have had negative growth. The sensitivity to which year is chosen is due to the volatility of income. In any given year, a substantial number of people are having an unusually bad year while some are having a good year. In succeeding years, these people revert back to the more normal income level. Thus, for those who are having a single temporary bad/good year, their earnings will jump/decline in the next year. In an earlier work, I showed that fully a third of people in the bottom quintile were there temporarily and had very strong growth in succeeding years. If these people were excluded from the calculations, then the growth rate of the bottom quintile would have been substantially less. The reversion to the mean issue can be easily solved by stratifying people into rich and poor on the basis of income over the entire period. This does away with the differing results based on using the beginning or end year. Sawhill came to realize the problem or reversion to mean and stopped referring to these earlier results in her later writing on mobility (McMurrer and Sawhill, 1998). But, the volatility issue also has ramifications in determining how to measure mobility. In Rose (1994), approximately 15 percent of the population in any given year were shown as having an unusual earnings level compared with adjacent years. Therefore, transition matrices that compare relative standing in two different years may contain a quarter of people whose incomes or earnings 5

6 are atypical. Since these atypical years will almost always move the person to a different quintile, a sizable percent of off-diagonal movements will be based on volatility rather than more permanent movements. Gottschalk and Danziger (1998) use multiyear averages to minimize, but not eliminate, this problem. Third, there is measurement error. Despite the best effort of researchers, these surveys are long and often filled out by a single member of the family giving individual information for all members. Not surprisingly, in a 1977 validation of the Current Population Survey (CPS), Mellow and Sider (1983) found that over half of respondents reported wage rates different from the ones reported by their employer to the Social Security Administration. Further, many of these erroneous answers were fairly large (the average error term equaled the standard deviation of overall earnings). Akerlof, Dickens, and Perry (1996) showed that this could have a big effect on earnings changes found in longitudinal data. It is generally agreed that monetary wages are sticky downward for employees who stay with the same company. Due to established custom, employers have responded to economic hardships by reducing their work force rather than reducing wage levels. In some circumstances, they have let wage gains fall behind the rate of inflation permitting a decline in real wages without a decline in nominal rate. Yet, various studies using the PSID show that about 10 percent of workers have lower nominal wages relative to the previous year. Ackerlof et al. used a distribution of yearly earnings changes from a private survey of Washington, D.C. firms, but they dirtied their results by adding in a simulated error term based on the CPS validation study. While their original data found virtually no workers who remained with the same employer had lower wage rates in consecutive years, the simulated data set had slightly more wage rate losers than found in the PSID studies. If up to one-quarter of reported earnings in the PSID have substantial measurement error, then transition matrices that rely on comparing two years of data would have almost half of respondents with an error in one or both years. This error problem is probably responsible for many of the cases that were classified as unusually good or bad years. So the problems raised by the combination of volatility and measurement error are substantial and can only be minimized by using as many years of data as possible. The fourth problem concerns boundary and size effects. In the quintile transition matrix approach, some small changes (e.g., moving from the 59th to 61st percentile) count as mobility while other larger changes (e.g., moving from the 41st to 59th percentile) count as non-mobility. Whether this odd asymmetry cancels out is certainly unknown, but does call into question the validity of the numbers. Recognizing this problem, Fields proposes that a better metric of relative change might be the absolute size of the percentile change. A related problem to tracking percentile changes is that equal absolute changes do not lead to similar percentile changes. Because earnings and income tend to be bunched together around the mean (the standard bell-shaped curve), the earnings difference between the 45th and 55th percentile is rather small compared to the gap between the 10th and 20th or 80th and 90th percentiles. Whether this is a problem raises the issue of what to make of the transition matrices. What does it mean if only 50 percent of people stay in the same quintile? Is this a high or low number? Mobility has been termed high because this figure is not closer to 100 percent. Perhaps, if 75 percent had stayed in the same quintile, then mobility would have been considered low. But no analyses, including one-year transition matrices, have come up with a number this high. 6

7 Finally, it is important to emphasize that mobility is different from economic growth. Consider the following simple no-growth economy: everyone starts working at 25 years old and retires at 60; each age cohort is the same size; every worker gets a 2 percent raise each year; and each new entrant starts at the same salary (assume no inflation). Over time, the average earnings of all workers remain steady even though each individual worker gets a raise every year and doubles his/her earnings over 35 years. Any growth would permit individual circumstances to improve even faster. Furthermore, a bit more complexity can easily be added to permit within cohort inequality. Everyone could still receive 2 percent yearly raises but their starting positions at age 25 can differ. As long as the distribution and level of the starting positions remained the same, then a no-growth economy can have intra- and inter-cohort inequality, no changes in relative positions within cohorts, yet a lot of life-cycle mobility. The real issues of concern in determining long-term economic well-being are path (rising, steady or falling earnings over time) and level (long-term average earnings). A false dichotomy has been created between relative and absolute change: relative change is nothing more than people with absolute rates of change that differ from the average. It is only when everyone in the same cohort changes by the same amount over time that there is no change in relative position. In this work, individual earnings are tracked over extended periods of time minimally five years but preferably 15 years. Age is very important because the transition into the labor force after leaving school is usually associated with a big upward jump while the movement into partial retirement leads to losses; consequently, in this analysis, workers are never younger than 26 or older than 59. For each person, an average rate of change is computed by doing a simple time series regression analysis: the coefficient of the yearly change divided by the average of the entire period equals the yearly rate of change. Further, people can be sorted into rich and poor on the basis of their long-term average income or earnings. The data produced are individual growth rates and long-term average earnings. The distribution of each of these variables can be presented in relationship to some meaningful markers. In the data reported below, individual earnings trajectories are divided into five groups: decliners (losing more than 0.5 percent per year), stable earners (plus or minus 0.5 percent per year), small gainers (between 0.5 and 2.5 percent per year), gainers (2.5 to 5 percent per year), and large gainers (more than 5 percent per year). In terms of earnings levels, the following definitions (in 1999 dollars) are used: low earnings: less than $15,000 a year (the amount it would take to support a family of three at the poverty level); near low earnings: $15,000-$25,000 (many polls show that people think that $25,000 is the minimum amount necessary to support a family); middle earnings: $25,000-$50,000; middle to high earnings: $50,000-$75,000; and high earnings: greater than $75,000. Finally, in determining average earnings of individuals and groups, all earnings are capped at $200,000 ( top-coded ) to avoid having a few very high earners skew the results. 7

8 Data Results Finding #1: Persistence is very important. As Table 1 shows, missing one year, even over 15 years, exacts a heavy penalty during the years of employment. Even adjusting for education, workers who miss a single year earn 20 percent less each year that they are working relative to those who meet the minimal standard of having earnings greater than $3,000 in each of 15 years. Missing more years exacts a greater penalty. Shares Average Annual Earnings Years Out of the Labor Force Men Women Men Women None $51,372 $29,439 One $37,830 $23,870 Two or Three Four of More $31,878 $16, $28,374 $10,835 Table 1 Average Earnings by Labor Force Continuity, (in 1999 Dollars) This table also demonstrates that women s earnings are much lower than men s if long time periods are used instead of just comparing full-time, full-year workers in a single year. Women s experiences are shaped by their role as primary family caregivers, as evidenced by the fact that they are much more likely to have labor force interruptions (some of which last many years). Furthermore, many of the most highly-paid professional jobs require 50 or more working hours per week, and this is more difficult for many women to manage. So, women and economy have adjusted to accommodate women s family responsibility and their need to have lower hours and to have periods in which they leave the labor force. This has constrained women s choices and they have been relatively isolated in traditional women s jobs. It is important to note that although women s earnings are substantially less than men s, their living standards as measured by long-term average total family income are just slightly below that of men s. The real losers in this arrangement are women who do not have other breadwinners and must rely solely on their own earnings. Finding #2: Very few men who stay continuously in the labor force in their prime-working years average below $15,000 a year. A common way to determine labor market performance is to look at the distribution of earnings in a single year (the top seven rows in Table 2). Almost a third of workers earned below our low-income threshold of $15,000 a year and less than half earned greater than $25,000. For young and old workers, the incidence of low earnings was very high. But even among prime-age men, the prevalence of low earnings reached one in eight. 8

9 Less than $15,000- $15,000 $25,000 Single Year 1995 All 32% 20% Young (less than 29) Male Female Prime Age (30-59) Male Female Older (> than 60) Male Female Year Average Annual Prime-Age Adults All Male 5 13 Female No Interruptions Male 1 10 Female At least One Year Male Out of Labor Force Female Table 2 Percent with Low Earnings in 1995 and in However, the picture is much different if the reference period is expanded to 15 years and the focus is just on prime-age workers. This is especially true if the focus is on workers who have no years out of the labor force. Among these continuously-employed workers, only 1 percent of males and 17 percent of females earned less than $15,000. These figures are much lower than the single year data show. However, it should also be noted that among the group with the strongest labor force attachment, 92 percent of workers stuck in very low earnings were female. Another 10 percent of males and 27 percent of females earned between $15,000 and $25,000. This means that 89 percent of men with continuous labor force participation averaged over $25,000 a year over 15 years. Obviously education preparation plays a key role. The shares earning more than $25,000 by education level: For men: For men by ed level: No HS diploma 50% HS diploma/no college 84% Some college/no B.A. 93% B.A. degree 95%; Graduate Degree 98 No HS diploma 11% HS diploma/no college 37% Some college/no B.A. 55% B.A. degree 77% Graduate degree 87% The fact that fewer men average below $15,000 over 15 years in comparison to a single year demonstrates the prevalence of volatility. In other words, many workers have intermittent spells of low earners that are offset by their more successful years. Having at least one spell over 15 years due to unemployment, changing jobs, or low commissions is quite common affecting almost one-third of all prime-age male workers. This shows both how transient most low earnings experiences are and that workers in good times may be burdened by paying off debts incurred during their infrequent bad spell. 9

10 Finding #3: Stuck in Neutral: Many men do not have Increasing Real Earnings as they move through their careers. On average, men gained 21 percent over this period while women gained 51 percent (see Table 3). If the analysis is limited to those continuously employed, these figures rise to 28 and 51 percent, respectively; these figures translate into 1.7 and 3.3 percent gains per year. The difference between women and men is deceiving because women are frequently starting over after they have taken a full or partial break from the labor market. Thus, they experience strong gains from a low base. Table 3 Earnings Trajectories, 1981 to 1995 Males Continuously Employed Non-continuously Employed Females Continuously Employed Non-continuously Employed Growth No Decline Change Fast Slow Slow Fast Average Earnings Percent Change 1991 to $46, $50, $29, $21, $29, $12, In terms of individual rather than group experience, only 60 percent of men who were continuously employed were on positive earnings paths, 28 percent were on declining paths, and 12 percent had no change. In Table 4, we see that this affected men of all ages. Remarkably, despite an average earnings growth of 39 percent, only 66 percent of men who started at 26 to 31 years old and were continuously in the labor force for the next 15 years had positive earnings growth, another 10 percent had no change, and 23 percent actually had declining earnings paths. For men making the transition from mid- to late-careers, there was no average gain in real earnings over these 15 years. Among this group, 46 percent had their earnings rise over time, 17 percent were stable, and 37 percent were on declining real earnings trajectories. Not surprisingly, educational attainment played a great role in determining the level of earnings and a lesser role in determining path over these 15 years. Among prime-age men, with the exception of those with graduate degrees, between 27 and 32 percent of continuously-employed men had negative earnings growth depending on their exact level of education. For those with graduate degrees, only 11 percent had negatives earnings trajectories. In terms of gainers, the shares were similar (at about 52 percent) among men with no post-secondary education and those with some college. By contrast, 60 percent of men with bachelor s degrees and almost 80 percent with graduate degrees were gainers. 10

11 Table 4 Earnings Trajectories of the Continuously-Employed Men, 1981 to 1995, By Age and Educational Attainment Growth No Decline Average Fast Slow Change Slow Fast Earnings Starting Age Men $47, Educational Attainment Men High School Dropout High School Grad Some College Bachelor's Degree Graduate Degree Percent Change 1981 to $53, $55, $27, $40, $48, $60, $72, Table 5 presents the division of growth paths by overall average earnings over the entire 15 years, thus showing the interaction between path and level. As is evident, earnings gains among men is sensitive to average earnings levels. Except for the few men earning below $15,000, average earnings gains (the last column) were very low for men who averaged between $15,000 and $50,000. Among this group of workers, barely more than 50 percent had increasing earnings over these years. For men with 15-year average earnings between $50,000 and $75,000, the average growth was higher but just 55 percent were on individually rising earnings paths. It is only among men who averaged more than $75,000 that we find substantially higher income growth and 80 percent being on positive earnings trajectories. This implies that a fundamental cause of rising inequality among male earners is the ability of high earners (predominantly managers and professionals) to have large earnings gains over their careers. The flip side of this coin can be seen by looking at the share of workers on declining paths. These are the workers who have often been dislocated and forced to take much lower-paying jobs. The share of decliners was 37 percent of those in the $15,000-25,000 category and 33 percent in the $25,000-50,000 group; this figure declined to 23 percent for those earning $50,000-75,000 and only 11 percent among the top earners above $75,000. One might use this table to come up with some rough estimates about labor market successes and failures. To begin, all those with average earnings below $25,000 with the exception of 23 percent earning $15,000 to $25,000 with fast growth are stuck in low earnings. In the large group of male earners in the $25,000-50,000 range, those with earnings declines (one-third) are probably destined to face tough times in the future. Thus, combining these two groups leads to 24 percent of men with continuous labor force participation facing tough times. Conversely, successes might include all those averaging over $75,000, those in the $50,000-75,000 except for those who are on steep declining paths, and the fast growers in the $25,000-50,000; the total of these three groups is 40 percent. Obviously, these are somewhat arbitrary definitions, but this table provides us with raw material to make judgments about longterm performance. 11

12 For female workers, the division of decliners and gainers did not change greatly by overall level of earnings. The vast majority were gainers and relatively few were on declining trajectories. Nonetheless, the level of earnings does affect the share with declining paths. Table 5 Earnings Paths by Average Earnings Levels, Distribution Decline No Growth Change Fast Slow Slow Fast Average Earnings 1981 to 1995 Percent Change 1981 to 1995 Male All , <$15, , $15,-25, ,996 2 $25,-50, ,811 9 $50,-75, , >$75, , Females All , <$15, , $15,-25, , $25,-50, , $50,-75, , >$75, , Finding #4: Life Cycle Earnings Gains were much more common in earlier years. Since there is no longitudinal data available prior to 1967, we can only make an educated guess about what was happening then. To do this, we can follow overall changes of synthetic cohorts. This procedure compares groups of people by age over time. In other words, those who were 25 to 34 in 1960 were 35 to 44 in The match is not perfect because some people would have died or left the country, while some immigrants would have moved in. Also, since we are comparing earnings, some people may be in the labor market just one of the two comparison years. But these factors are relatively small since we are only interested to see if there have been significant changes over time. In Table 6, three comparisons are made. The second column follows the youngest group of prime-age workers starting at 25 to 34 years old and ending 10 years later when they are 35 to 44. The third column compares an older cohort: starting at 35 to 44 and ending at 45 to 54. The fourth column is a 20-year comparison of those who began at 25 to 34 and ended at 45 to 54. Two examples may help clarify how to read this table. The 22.8 in the top lefthand corner of the data means that the average earnings of those who were 25 to 34 in 1988 grew by 22.8 percent when this group was 35 to 44 in The 53.7 represents the percentage earnings growth of those who were 25 to 34 in 1947 compared to when they were 35 to 44 in While the data are available for all of the years, only selected years are used during the early years to make the table easier to read. 12

13 Percent Growth Previous 10 Years Previous 20 Years Final Ages Table 6 Average Earnings Growth of Male Synthetic Cohorts, Using male synthetic cohorts, we can see that the growth rates over 10 years were much higher through 1973: typically from 1947 through 1963, young workers who began at had a 50 percent real gain over the ensuing 10 years. By contrast in the 1980s, those who began at this age group could expect only a 20 percent increase over the ensuing 10 years. Among older workers before 1973, a 30 percent gain over 10 years was common. In the transition from the 1980s to the 1990s, these middle age workers did not have any real gains as they aged to Finally, in the fourth column, workers experienced almost 100 percent gains on average over 20 years after they started their careers in the late 1940s and early 1950s. If they started in the 1970s, these gains were just 20 percent (in line with the results reported here). Using the back of the envelope estimations, the early experience of 25- to 34- year-olds equaled an average growth rate of more than 4 percent per year. From the longitudinal study from , the standard deviation of the growth rate among men was about 2 percent. Thus, only about 2 percent of these men might be expected to have earnings losses of greater than 0.5 percent per year. Among the older group, their 10-year growth in the period from 1947 to 1973 translates into an annual growth rate of just under 3 percent per year. Again, using a 2 percent standard deviation, a rough estimate is that slightly less than 10 percent of this group had declining earnings paths prior to While this exercise most likely underestimates the number of people with bad experiences who have declining earnings paths, the differences based on these estimates are so large that there is strong indirect evidence that very few prime-age male workers were on individual negative earnings trajectories between 1947 and

14 Finding #5: What kind of occupation you hold affects your earnings greatly. In looking at job holding over multiple years, it is important to develop a limited number of categories that are internally consistent. In previous research, I showed that the official government occupation categories had certain inconsistency that grouped jobs that were very different into the same summary category. A new eight-way division was developed, and it was shown that it could be collapsed into three large categories: elite jobs of managers and professionals, good jobs of supervisors, skilled crafts, police, firefighters, technicians, and clericals, and less-skilled jobs of factory workers, laborers, sales clerks, and food and personal service workers. Although the data were collected in the more detailed eight-way division, it is presented in Table 7 in the more aggregated three-way division. Also, in tabulations not shown, persistence in occupations began quite early in people s careers. If they had not made their way into top tier elite jobs by the beginning of their 30s, very few workers made the transition to persistence in managerial and professional positions in later years Earnings Change Men $46,932 $42,586 $51, Elite job 12 or more years* $69,104 $56,672 $80, Table 7 Earnings Level by Long-term Occupation Holding Good job 12 or more years* $45,366 $42,951 $47, Elite or good job 14 or 15 $45,844 $44,709 $42,308-5 years* Less-skilled job 12 or more $33,353 $32,275 $35, years* Other continuouslyemployed $46,087 $40,590 $49, With at least one year out $29,695 $30,905 $24, of LF Women $21,649 $18,386 $27, Elite job 12 or more years* $40,951 $29,332 $49, Good job 12 or more years* $26,509 $20,981 $29, Elite or good job 14 or 15 years Less-skilled job 12 or more years* Other continuouslyemployed With at least one year out of LF * No Labor Force Interruptions $33,246 $23,297 $36, $18,248 $14,323 $18, $24,775 $17,825 $31, $12,835 $12,161 $17,

15 Findings In previous research, inequality was shown to be associated with the rising pay advantage of top managerial and professional jobs over pay in both good and less-skilled jobs. This finding is validated here as men in elite jobs had their earnings rise by 41 percent during these years, while men who persisted in good or less-skilled jobs had an earnings gain of approximately 10 percent. This large discrepancy in growth rates meant that in 1981 men in elite jobs began just a little ahead of men in good and less-skilled jobs, but increased their advantage dramatically. This was caused by 20 percent of persistently employed elite job holders doubling their incomes (greater than 5 percent a year) and another 20 having strong growth (at least 2.5 percent per year). By contrast, men in other jobs had few who doubled their earnings and only 20 percent gained more than 2.5 percent a year. Men with continuous employment but not persistence in a single tier had to have at least two years of employment in less-skilled jobs. Thus, this group consists of those moving up and down the occupation and earnings ladder with 12 percent being gainers and 6 percent being losers. Women workers were likely to be on increasing earnings paths independent of their job holding. Nonetheless, the increase was greatest the more that employment was in managerial and professional jobs. Finding #6: In terms of household income, men and women are much closer than the earnings indicate. To complete the long-run picture of how workers fared over these 15 years, it is useful to see how earnings translated into total household income. Table 8 tracks prime-age men and women in terms of the households they are part of during each of these years. The huge labor market gap is much smaller here because women are part of households with multiple earners. A big problem occurs when women do not have a male earner with whom to pool income. Again the effect of continuous employment is strong, especially among men. For men with any interruptions, the average household income over 15 years is more than 40 percent less than men with no interruptions. This effect is much smaller among women, but still there is a 20 percent effect. In terms of very low household income ($25,000 which some analysts would consider a more realistic poverty line), 10 percent of males and 16 percent of females fall into this category. But this figure is only 5 percent for males with continuous employment versus 38 percent for males without labor force continuity. Drawing lines is always difficult. But if we pick $50,000 a year as a demarcation, we find that 37 percent of men and 47 percent of women are in households with long-term incomes below this level. 15

16 All Continuously- With Employed Interruptions Men Average Income $68,996 $73,218 $43,562 Income levels Less than $25, $25,000-$37, $37,500-$50, $50,000-$100, Greater than $100, Women Average Income $61,610 $66,718 $51,360 Income levels Less than $25, $25,000-$37, $37,500-$50, $50,000-$100, Greater than $100, Table 8 Distribution and Average Family Income, Conclusion The research debate over inequality and low earnings has been heated because these findings have important policy implications. A multiyear analysis gives a different picture of economic performance than a single-year snapshot. Singleyear snapshots have been used as representative of long-term conditions because they were much more readily available. Further, it was thought that averages would adequately capture a group s performance because those with unusually bad years would be offset by those with unusually good years. The data that are presented here add a new perspective. On the one hand, the share of men who are long-term low earners is small, especially if they meet the minimum criteria of being in the labor force continuously over 15 years. This figure is a bit misleading because many men cycle through bad periods and may require several good years to pay off debts incurred. Women workers have a different labor market connection and are more likely to have interruptions, not work long hours per year (many top-paying jobs require at least 50 hours per week), and be stuck in low-paying occupations. Therefore, their long-term earnings trail those of men by over 40 percent even if they are continuously employed. In terms of mobility, some researchers have overestimated its impact among low earners. Surprisingly, as compared with the years between 1947 and 1973, a large number of male workers did not have the expected experience of rising earnings over 15 years of their careers. Workers with lower 15-year average earnings were less likely to have increasing earnings trajectories than workers with higher average earnings. Similarly, workers who were not persistently employed in managerial and professional jobs often did not have positive real earnings gains. The picture that emerges is that earnings gains among male workers are more concentrated in top managerial and professional jobs. To get these jobs, a bachelor s degree has become virtually a minimum requirement. For other workers, their situation has become more tenuous. Although they rarely average less than $25,000 a year, there is a sense that they are not keeping up and a 16

17 fear that they might have a bad experience. As I have documented in a series of papers with Barry Bluestone, families have responded by increasing the labor force hours of wives and other family members, but this has come with many hidden costs. So, these are bittersweet results. Most prime-age people make it at least into the bottom rungs of respectable middle class life. Yet, they are more squeezed not moving forward as they had expected, working harder to stay in place, feeling left out of the luxury goods that are available to their more affluent cohorts, and wondering what it will take for the children to succeed. References Akerlof, George, William Dickens, and George Perry, The Macroeconomics of Low Inflation, Brookings Papers on Economic Activity I, pp Armey, Richard, The Freedom Revolution: The New Republican House Majority Leader Tells Why Big Government Failed, Why Freedom Works, and How We Will Rebuild America, Washington, D.C.: Regnery Publications. Cox, Michael and Richard Alm, By Our Own Bootstraps: Economic Opportunity & the Dynamics of Income Distribution, Dallas: Federal Reserve Bank of Dallas., Myths of Rich and Poor: Why We re Better Off than We Think, New York: Basic Books. Gottschalk, Peter and Sheldon Danziger, Family Income Mobility How Much is There and Has it Changed? in James Auerbach and Richard S. Belous, Eds., The Inequality Paradox: Growth of Income Disparity, Washington, D.C.: National Planning Association. McMurrer, Daniel and Isabel Sawhill, Getting Ahead: Economic and Social Mobility in America, Washington, D.C.: Urban Institute Press. Mellow, Wesley and Hal Sider, Accuracy of Response in Labor Market Surveys: Evidence and Implications, Journal of Labor Economics, Vol. 1, No. 4, pp Rose, Stephen, On Shaky Ground: Rising Fears about Incomes and Earnings. Washington, DC: National Commission for Employment Policy. Sawhill, Isabel and Mark Condon, Is U.S. Income Inequality Really Growing? Sorting out the Fairness Question, Policy Bites, No. 13, Washington, D.C.: The Urban Institute. U.S. Department of Treasury, Household Income Mobility During the 1980s: A Statistical Assessment Based on Tax Return Data, Tax Notes, Vol. 55, No. 55, Special Supplement. 17