Let me tell you a story about Monica and Jaime.
Monica is an African American in her mid-30s. She is single, and has a bachelor’s degree from a middle-of-the-road state university. She was recently laid off from her job as an assistant branch manager of the local bank. Monica has spent several weeks looking for a new job, but has had no luck so far.
Jaime is a 21-year-old agricultural worker who came to the United States from El Salvador four years ago. After meeting the girl of his dreams, he got married and became a U.S. citizen. Just last week, he completed his high school education, thanks to night school classes. With his diploma in hand, Jaime will soon be promoted to foreman.
Neither Monica nor Jaime is real. But their situations are pretty common.
Like Monica, millions of Americans are out of work and wondering why they are having such hard times finding jobs. Like Jaime, many are doing everything they can to support themselves and their families, even if this means juggling multiple jobs or combining school with work.
There is no single reason why Monica doesn’t have a job or why Jaime does.
Many factors helped shape their places in the labor market, from who their parents were and how hard they studied, to their ages, their sexes, and their ethnic heritages. Disentangling these various factors is what economic analysis aims to do.
As I argued in “Uncovering the Labor Market Recovery,” the first installment of The Century Foundation’s Working Paper Series, the headline unemployment rate has serious shortcomings as a catch-all indicator of labor market health.
That said, the government’s focus on the unemployment rate means that we have a lot of detailed statistics about who the jobless really are—their age, sex, race, education, and much else. We know the places they live, the occupations they’ve worked, and even what their family lives look like.
These kinds of details matter for policymakers. After all, if most of the unemployed turned out to be college graduates, for example, then policies aimed at making it easier to go to college would do very little to put people back to work. (In fact, unemployment rates are lower for those with greater schooling).
Probing the data on unemployment yields four key insights:
- Race still matters, big time. African Americans are unemployed at twice the rate of whites—even when they are identical in dimensions other than skin color.
- Education is the surest path to securing employment. The unemployment rate among college grads is half that of high school dropouts.
- Sex, age, marital status, and citizenship also have meaningful impacts on unemployment, sometimes in surprising ways.
- Regression analysis is a powerful tool for understanding the labor market’s complexities, but it must be used carefully. Associations are not the same as causations.
This installment in the Working Paper Series begins with an overview of the dimensions of unemployment in the United States today. It then unpacks those dimensions, examining in detail a few of the demographic characteristics most strongly associated with unemployment: race, sex, education, age, and citizenship.
A decade and a half into the twenty-first century, an African American like Monica is nearly twice as likely as a white American to be unemployed, all else equal.
Similarly, a young person like Jaime is 53 percent more likely to be unemployed than an older worker. Women and the unmarried also face employment headwinds.
On the flip side, middle-aged, married white men with advanced degrees are secure relative to other groups: their chances of suffering unemployment are a fraction of those faced by most other Americans. Advantage begets advantage: success in the labor market is influenced by, and in turn influences, favorable endowments in other areas of life.
In fact, as Figure 1 shows, your race, your age, your sex, your education level, and even your marital status has considerable effect on your likelihood of being unemployed.
As much as we pay lip service to equality of opportunity, we are not a nation of equal chances.
As it is, the stark demographics of unemployment are a sobering reminder that labor market outcomes—of which unemployment is the worst—rarely come down to personal merit alone. Not all of us had parents who read to us a children. Not all of us grew up in leafy suburbs where our biggest worry was whether we’d make the travel soccer team. And, sad as it is to say, in some cases, no accumulation of qualifications can fully displace the deeply held prejudices some employers still harbor.
Nor can qualifications alone overcome the implicit biases embedded in our institutions and social structures—things like the the power of neighborhood or interpersonal networks—that hold some back while propelling others to the top. These floating inequalities—disadvantages that are passively propagated rather than deliberately maintained—can be the most challenging.
So how can we be sure that differences in unemployment rates that correlate with things like race, sex, and age are not just meaningless coincidences? To understand that, we need to do a little statistical analysis.
Did you know that the amount of money Americans spend on pets can predict the number of lawyers in California? It’s true. See Figure 2.
Okay, so you’re right to be skeptical. This is a chart from a website called “Spurious Correlations.” The site exists to drive home the lesson that two variables that produce similar graphs aren’t necessarily related.
When you hear economists say “correlation is not causation,” this is pretty much what they mean. (Actually, you’re more likely to hear economists giving a nerdier version such as “association doesn’t imply causality.” It means the same thing. It’s also why economists don’t get invited to many parties.)
The point is pretty easy to see when you’re plotting toys for Fido against job openings at McKenzie, Brackman, Chaney and Kuzak.
It’s a little harder to see when we start trying to figure out what’s going on with someone like Monica. What explains why she is unemployed? Her college degree? Her race? Her gender? Her age? Her marital status?
It turns out that none of these is the single cause, but all of them can have an impact.
Unemployment figures from the Bureau of Labor Statistics contain a multitude of demographic details. It’s a relatively straightforward matter to pull those out and examine each in turn. For example:
- unemployment declines with education;
- racial minorities are more likely to be unemployed than whites;
- young people have a harder time finding jobs than experienced workers; and
- men are more susceptible to economic downturns than women.
Patterns like these are interesting, and they can be informative. Simple x–y relationships—that is, correlations—are how we calibrate our understanding of the world. But as seductive as these links can be, one-dimensional analysis like this isn’t enough to tell us whether x causes y.
Think of the labor market as a big pot of stew, full of vegetables, starches, proteins, and spices. Even when the stew tastes okay, conscientious cooks know it’s possible to do better.
The question is, how?
Randomly stirring things up is unlikely to get us very far. Haphazard changes may well make things worse. Instead, to make real progress, we need to identify the contribution of each ingredient. But with all the flavors mixing together, that’s no small task.
Fortunately, such chunky recipes are the bread and butter of econometric analysis.
Economists carefully tinker with recipes in ways that allow them to understand the properties of each ingredient individually and how they interact with others. And the key to this tinkering—the trusted base in the economist’s cookbook, if you will—is something called “regression.”
Regressions allow economists to apply statistical theory to data and thus estimate relationships between cause and effect.
As is the case with stew, unemployment is complicated. Whether or not someone can find work depends on many factors, both personal (e.g., age, race, or education) and macro (e.g., economic conditions or the composition of local industries). The magic of regression is that it enables us to consider all of these factors simultaneously and parcel out the unique contribution of each, one at a time, as if “all else” was held constant.
Figure 3 presents the results of our regression analysis of the demographics of unemployment.
The data come from the Census Bureau’s Current Population Survey (CPS), for the the most recent twelve months (October 2013 to September 2014). The CPS is the same source the Bureau of Labor Statistics uses to calculate the monthly unemployment rate.
Like the BLS, we are interested in a subset of the population: the civilian labor force, age 16 years and older. The share of the people in this group who are looking for work but can’t find it are who the BLS define as the unemployed. In fact, in the analysis that follows, we are going to be a bit more strict in our definition of unemployment than even the BLS: we’re only going to consider people with prior work experience, which allows us to consider the influences of industry and occupation.
Our regression model (technically known as a “logit”) is designed to predict the changes in the probability of unemployment associated with each of the characteristics listed in Figure 3’s rows—that is, age, sex, race, citizenship, education, marital status, children, and class of worker—as well as industry, occupation, state of residence, and month.
Here’s how to interpret the numbers attached to the bars. They express the average percentage point change in unemployment associated with a given row’s characteristic, as compared for the baseline for that category (shown parenthetically), holding all other characteristics constant.
Taken together, the results suggest education, age, sex, marital status, citizenship, and class of worker all matter for unemployment—sometimes in quite large ways.
But one factor stands apart: race.
Blacks are 4.4 percentage points more likely to be unemployed than whites. And that is among people statistically similar in ways other than race.
It’s worth emphasizing that percentage points are not the same thing as percentages. For context, note that among whites, the overall unemployment rate during the last year averaged 4.7 percent. Relative to this, an increase of 4.4 percentage points is a massive amount: it implies a percentage increase in unemployment of 94 percent. That is, blacks are about twice as likely as whites to be unemployed.
Other minorities are also considerably more likely to experience unemployment than whites: Native Americans by 4.3 percentage points, Hispanics by 0.7 percentage points, Asians by 0.6 percentage points, and those of mixed race by 3.5 percentage points.
Perhaps surprising to some, race matters even more than education, which is traditionally seen as the ultimate gateway to upward mobility.
But education is still a big deal.
In fact, it’s the second most powerful predictor of unemployment. Compared to those with a bachelor’s degree, high school dropouts are 3.8 percentage points more likely to be unemployed. Similarly, those with at most a high school degree are 1.7 percentage points more likely to be unemployed, and those with some college, but not a degree, are 1.1 percentage points more likely.
Having an advanced degree appears to confer a modest advantage—but clearly avoiding unemployment is not the primary motivation for seeking a masters or Ph.D. (note that, in Figure 3, negative values indicate reduced unemployment—a good thing).
Age continues to play a role, too. Compared with workers age 45 to 54 years, typically considered peak employment years, both younger and older workers are more likely to be unemployed. Only those aged 35 to 44 years, are employed at comparable rates. The most disadvantaged are the very youngest workers (more on that below).
Women also appear to face an employment penalty. Despite generally being employed at higher rates, women are actually more likely to be unemployed than men, once we control for things like education and marriage, to the tune of 0.8 percentage points. So, while women outperform men along a number of important dimensions, like education, it seems that the labor market is still not gender-neutral.
Interestingly, the foreign born, both naturalized and not, are significantly more likely to be employed than natives, all else equal.
Perhaps even more interesting, there appears to be an employment bonus for being married. Holding aside sex, education, race, age, and the rest of our explanatory variables, married people are 2.7 percentage points less likely to be unemployed. This suggests there may be something inherently beneficial about marriage in terms of employment.
But the case of marriage also reminds us, even with the increased confidence that regression analysis gives us, that we still should remain somewhat circumspect in interpreting relationships. Marriage might be one factor where causality is a two-way street: it could be that employed people are more marriageable. Or perhaps our married variable is picking up characteristics omitted from our data, like favorable personality or intellect, both of which can positively affect employability as well as marriageability. Without additional assumptions or information, the results of a regression are still associations, not causations. (See the Explainer below, “What are the limits of regression analysis?”)
The type of job a person holds also impacts unemployment. The broadest classification of workers is by “class,” which basically describes the type of employer they have. Most people are “private wage and salary workers”—that is, people who work for a wage or a salary in the private sector. Government workers fall into a similar class. Most distinct are the self-employed and those who work in agriculture.
As the figure shows, being an agricultural worker isn’t easy—the seasonal patterns in farm work make steady jobs scarce and unemployment more likely. Indeed, many labor market analyses exclude agricultural workers entirely, because their situation is so unusual. But that doesn’t make unemployment any less painful.
By contrast, government workers appear somewhat better off than private sector workers. This bears out the popular impression that government work is more secure.
Finally, people seeking full-time work (or who have worked full-time in the past) are much more likely (by 3.5 percentage points) to be unemployed than those who consider themselves to be part-timers. This makes sense: most people who want to work want full-time jobs, so finding a full-time job is more competitive. From an employer’s perspective, hiring a full-timer is a more costly and riskier investment than enlisting part-time labor.
To summarize our findings so far, we’ve seen the marginal effects of various characteristics on unemployment. Marginal effects tell us how outcomes change as characteristics change. But we can also look at the regression results in more familiar terms, by summing up all the marginal effects into predicted unemployment rates for people with each type of characteristic.
Figure 1 does just that.
It gives the average predicted unemployment rate for each group we’ve considered (the rate predicted by our regression model), assuming that members of the particular group are otherwise average—that is, that they have the population-average level of all other attributes. (My Methodology Explainer (above) has a lot more about how all this works. If you’re interested in digging into all the details, go take a look.)
The results conform to the patterns we’ve been discussing. Minorities, young people, the less educated, women, and the unmarried fare the worst; whites, middle-aged adults, men, immigrants, and married people do the best.
Back to Monica. There are a lot of factors influencing her situation. But, odds are, the biggest one is something she can’t control: her race. Our analysis says that Monica is twice as likely to be unemployed as she would be if she were white. Full stop.
Despite social progress in recent decades, African Americans continue to suffer the ill effects of a legacy of discrimination. Consider:
- blacks are more likely than whites to grow up poor—or even if they are not, to grow up in neighborhoods of concentrated poverty, with their attendant disadvantages;
- blacks are less likely to succeed in school, and more likely to wind up involved with the criminal justice system; and
- even their health tends to be worse.
In short, blacks find their opportunities limited, sometimes by overt prejudice, and sometimes by the legacies of our racist past. In a society that prides itself on equality of opportunity, the sad fact remains that, in far too many cases, blacks face an uneven playing field. The same is often true of other historically disadvantaged groups, including women, gays, and racial minorities more generally. As we’ve seen, those disparities are borne out strikingly in the current labor market.
Like Figure 1, Figure 4 translates our regression results into predicted unemployment rates, assuming everyone is average in all ways besides the characteristic we’re investigating. Under these assumptions, black unemployment is predicted to be 7.8 percent, versus 4.0 percent for whites—a 96 percent increase (for a fuller discussion, see the Methodology Explainer).
Unfortunately, the present is no anomaly. As Figure 5 below shows, black unemployment regularly runs about double that of whites.
For twenty-two straight months—from December 2009 through September 2011—black unemployment averaged 16 percent. Blacks actually fared worse over that period than did high school dropouts.
It’s important to remember that we’re not necessarily saying that skin color alone causes unemployment.
What we are describing is a strong association between race and unemployment. Deliberative (or unconscious) discrimination might be part of it. But it is also likely that factors correlated with race that impact unemployment (but that aren’t included in our data) are also at play—things like whether or not someone was raised in a single-parent household or lives in a poor neighborhood.
For Jaime, education is a big part of the story. A high school diploma firms up his labor market prospects considerably, and his future will be even brighter if he continues to college.
Our analysis supports the well-known findings that, all else equal, the more education someone has, the better his chances are of having a good job, earning a good wage, and enjoying the good things life has to offer (see Figure 6). In a world full of known risks and unpredictable uncertainties, it is difficult to find a mechanism more effective than human capital investment at improving one’s future job prospects.
In the specific context of unemployment, the story is simple: the more education you have, the less likely you are to be unemployed.
In October 2014, the unemployment rate for the population of college graduates 25 years of age and older was 3.1 percent, compared with 4.8 percent among those with some college, 5.7 percent among high school graduates, and 7.9 percent among high school dropouts. As Figure 7 below makes clear, this pattern has held for years.
Most of the unemployment pain is concentrated at the bottom of the skills distribution.
According to our regression analysis, a high school dropout is about twice as likely to be unemployed than is a college graduate, and about 36 percent more likely to be out of work than someone who has completed high school. Starting, but not finishing, college offers some benefits—the unemployment rates for high school graduates and those with some college are similar—while completing college confers strong labor market protections.
Monica has a second big strike against her. Turns out, women fare worse than men in the labor market.
Historically, women have faced employment discrimination rivaling that of blacks. But women have made rapid progress during the past few decades. Indeed, looking just at raw historical figures, it seems as if women actually fare a bit better than men in many respects.
Prior to the recession, as Figure 8 shows, men (4.7 percent) and women (4.5 percent) had similar rates of unemployment, both as a whole and among prime-age workers (those 25 to 54 years of age).
Our regression analysis puts sex differences in starker terms.
As Figure 9 shows, once we control for factors such as race, marital status, and education (more women go to college than men), a woman is about 15 percent more likely to be unemployed than a similarly situated man.
Jaime isn't black or a woman, but he does have his own disadvantage: he’s young.
Our analysis shows that a worker aged between 16 and 24—Jaime’s cohort—is 53 percent more likely to be unemployed than is a worker who is between 35 and 54 (see Figure 10).
As Figure 11 below demonstrates, the unemployment rate is inversely related with age. The older someone is, the less likely she is to desire work but lack it.
The explanations for this pattern are mostly straightforward. With age comes experience, expertise, and seniority. Maturity generally brings stability, in all aspects of life, and work is no exception. Especially as workers reach the peaks of their careers, which for most industries occur between the mid-40s and the mid-60s, workers reap the rewards, in terms of pay, influence, and work hours, of the hard work put in over the years.
The unemployment rate for 16–24-year-olds stands out the most; it is often double that of other cohorts. During the Great Recession, the unemployment rate for that age group increased from 10.6 percent to 19.5 percent, which means that one in five youth seeking work couldn’t find it. It has since returned to 13 percent, which is still high by historic standards.
You might think this is obvious. After all, many 16–24-year-olds are students.
But school actually isn't to blame. That’s because youth unemployment rate, like the unemployment rate among all other age groups, refers only to those who are actively looking for work. In other words, youth who chose not to work because they’re going to school (or for any other reason) are not included in the unemployment figures. Nor are youth who want jobs but have given up looking (that is, those who’ve dropped out of the labor force). Consequently, because youth labor force participation is low, the share of the youth population who are employed is much lower than the unemployment rate suggests. (Also remember that our analysis excludes first-time job seekers, many of whom are young.)
For our purposes, the takeaway is that youth unemployment isn’t artificially inflated by students.
As it turns out, about three-quarters of youth unemployment is accounted for by youth who aren’t going to school. Most of these are so-called disconnected youth—young people who are neither studying nor working—have never attended college, and a sizeable share are high school dropouts. Helping them get back on track is a particularly pressing policy challenge.
So Jaime’s age is working against him, even as his newly-minted diploma is helping him get ahead. Fortunately, he also has two other factors working in his favor. (Or, to put things a bit more precisely, he possesses two characteristics associated with low rates of unemployment.)
First, he is not a native citizen.
Our analysis shows that immigrants actually fare better than native-born Americans when it comes to unemployment (see Figure 12). Naturalized citizens are 19 percent less likely than native citizens to be unemployed (controlling for the other factors we’ve discussed). The foreign born who are not yet citizens do even better: they’re 32 percent less likely to be unemployed than natives.
Immigrants have a reputation for being hard working, and the data seems to bear this out. Returning to our favorite theme, however, we must be wary not to recklessly attribute cause and effect. Immigrants are a self-selecting population: those motivated and persistent enough to seek and receive citizenship in a foreign country are also likely to be diligent in their job search, no matter what barriers they may face. Furthermore, those who are unable to retain gainful employment might return to their native countries.
Our data doesn’t allow us to distinguish between these possibilities. But what we can say is that Jaime, as an naturalized citizen working in America, has some mix of traits, related to his immigrant status, that bodes well for his employment prospects.
Jaime also has another plus. Actually, it's a +1: he’s married.
According to our analysis, as Figure 13 shows, those who are married (with otherwise average characteristics) have unemployment rates around 3.6 percent—significantly lower than the 5.9 percent rate among the unmarried.
Though marriage can certainly enrich one’s life, many people find it somewhat unexpected to see such a profound association with unemployment. But even after controlling for all other factors, being unmarried means you’re 65 percent more likely to be unemployed.
That’s not to say unemployed singles should spend their time looking for a spouse rather than a job. As we discussed above, a person who has strong credentials in the dating market is likely to carry those same positive character traits into the job market. On the other hand, an undependable person may have difficulty finding both a spouse and a job.
Nevertheless, it is still possible that marriage itself creates an employment bonus; it is easy to think of ways a stable, loving relationship—along with the desire to provide for another—can inculcate positive behaviors at work. Which explanation carries the most weight is a question for additional research.
Most of the time, when we talk about the unemployment rate, we act as if it is a shared status symbol—a simple, all-encompassing indicator of our national well-being. When it’s high, it’s obviously only a matter of time before the United States turns into the next Greece. When it’s low, it’s clear proof America is the greatest economic superpower the world has ever seen.
In some ways, this national obsession with the unemployment rate makes sense; to a first approximation, it can be a useful shorthand for the overall state of the economy. But as the foregoing analysis makes clear, the burdens of unemployment are anything but equally shared.
Understanding who the unemployed are—and who they are not—can provide us with valuable information about how our society is structured: in a truly egalitarian country, widely divergent demographic patterns in unemployment would not exist.
At the same time, remembering that association is not causation forces us to think systematically about what these relationships imply—why they exist, what they represent, and, perhaps most importantly, how we can remedy them.
What this installment of the Working Paper Series has shown is that when it comes to employment, inequities among different groups abound.
Nevertheless, there is hope. There are things under our control. Education works to improve employment outcomes, and its impact is large. Immigrants can, and do, succeed in the U.S. labor market. And marriage and family may help cultivate steadier careers.
The challenge is to ensure that access to advantages—be they education, citizenship, or social supports—are more evenly distributed.
And, as we work for equality of opportunity, we must also remember that luck has a lot to do with it. The difference between having a good job and having none at all often is determined by forces beyond our control. A good safety net protects those who draw a bad hand—and that’s exactly why we need make it stronger.
In other words, when we say all else equal, we must really mean it.
Mike Cassidy is a policy associate at The Century Foundation. He is a strong believer in the power of scientific analysis, and his research focuses on using economics to understand human behavior, especially at it relates to poverty, inequality, performance, and progress. A proud alum of Princeton’s Woodrow Wilson School, Mike holds a Master in Public Affairs with a concentration in economics and public policy. From 2007 to 2012, he worked at the New York City Office of Management and Budget, where he oversaw the city’s social service and criminal justice agencies. In his spare time, Mike is a semi-professional distance runner and competed in the 2012 U.S. Olympic Marathon Trials.
Joe Miller, Jason Renker, and Greg Anrig
Art Direction and Graphic Design
Project Manager and Developer