A lot of attention (myself included) has been recently put on the Top 1% income and wealth. However, there is also substantial inequality in the other 99% that is worth exploring. To get an idea, if we took all the Top 1% income growth between 1979 and 2012 and distributed it among the other 99%, each of us (I assume you also belong to the other 99%...) would earn around $7000. However, the increase in the earnings gap between a college-educated and a high-school educated household is four times that in the same period. Hence, here we will focus on wage inequality among the other 99%, but particularly between the bottom 10% and top 90%, so as to exclude the very extreme cases (which deserve a different attention). But first, Figure 1 shows how wages have changed between 1963 and 2005 by wage percentile. Here we see that generally there was a much bigger increase in wages among the top half than the bottom one.
Figure 1: Change in real wages by percentile, 1963-2005.
A common measure for overall inequality is the ratio of those at the 90 and 10 percentiles. A typical issue is that the population might be changing its structure, with more people getting educated or more work experience. As this happens, the typical person in either of these percentiles might be changing, hence changing our standard interpretation of increasing inequality. Another take on inequality is to look at between-groups inequality, where the typical comparison is those with a college degree and those with a High School degree. This tries to avoid the issues of other characteristics of the population changing as in the overall inequality measure. However, another alternative look at inequality is to look within-groups, hence evaluating how much variation there is among small groups (for example: college educated, 25-30 years old, male). This three measures of inequality are displayed in Figure 2, where we see that even though the three of them have increased over the long haul, they have done so at different paces and through different paths. Particularly the college premium follows a strange path, increasing in the 1960s, decreasing in the 1970s and increasing very fast since the 1980s. This suggests that a simple, unique explanation for the recent increase in inequality is not likely to work.
Figure 2: Three measures of Income Inequality.
But has inequality changed more among the top or the bottom? An easy way to look at this is to compare the 90 and 50 percentiles (upper-tail inequality) and, separately, the 50 and 10 percentiles (lower-tail inequality). Note this still excludes the very bottom and very top. Figure 3 shows that even though lower-tail inequality grew in the 1980s, it has not grown since then. On the other hand, upper-tail inequality has increased continuously.
Figure 3: Upper- and Lower-Tail Inequality.
So what is behind this monumental increase in inequality? Identifying the cause of this change is very hard or probably impossible, but what we can do better is identify the proximate causes, meaning what seems to be closely associated with this change, even if we do not understand what led to the first thing. And this is where the skills of economists Autor, Katz and Kearney comes into play. They argue that we cannot simply think of people as belonging to one of two groups - skilled and unskilled - where the top one is associated with higher education. Figure 4 shows that from 1979 to 2005 the wages of those with a post-college education grew by a lot more than those with college degree. Moreover, the difference between those with exactly college and high-school degrees slowed down significantly since the 1980s. And finally, the difference between those with high school degrees and those without one has flattened or even decreased since the mid-1990s. All this suggests that, since the 1990s we are in a situation where the income among the very high- and very low-skilled workers has increased relative to those in the middle. Income has polarized.
Figure 4: Changes in wages by Education.
What explains this? The main hypothesis is that computerization has changed the demand for job tasks and affected the demand for skills in such a way that explains this polarization of income. Computers are good at doing routine tasks which are codifiable, like bookkeeping, clerical work or repetitive production tasks. (If you have interacted with a call-center lately, you will probably know how computers have improved in voice recognition and seem to have taken over those tasks that require gathering the same information all the time). On the other hand, abstract tasks like those performed by "high-skills" managers or educated professionals are hard to automatize since they require cognitive and interpersonal skills and adaptability. Similarly, manual tasks used in many "low-skilled" jobs like security guards, cleaners and servers are hard to computerize and hence have not been affected much by the advance of computers. Figure 5 confirms this intuition that low-skill jobs (taking the average education of those performing such jobs) usually have manual tasks. On the other end, high-skill jobs are mainly filled with abstract tasks. However, routine tasks are concentrated between the 20th and 60th percentiles.
Figure 5: Task intensity by Occupational Skill.
The conclusion is that the change in wage inequality may be substantially explained by changes in the demand of skills, which has been lately polarized by the introduction of computers. As the demand for these jobs increased, so did their wages. But why haven't workers matched the increase in demand by educating themselves more? Well, most likely this change was very hard to predict and so not enough people found higher-education to be "worth it." However, recent trends in education attainment suggest that young people are catching up to this increased demand.
Based on an article by Autor, Katz and Kearney.
As I spend my vacations back home in Argentina, I am being asked by many people what do I think about the Argentinean economy? My answer to most of these people was that, on top of what the newspapers say, it is very hard to comment. Data is necessary to properly evaluate the state of an economy. And reliable data on my country's situation is hard to get.
A lot has been said about the inflation index in Argentina. The figure published by the INDEC, the national statistics agency, has been accused by almost every media source as unreliable. The government is accused of cherry picking the stores and goods it follows such that inflation is reported to be low. But every Argentinean you meet on the streets will tell you the index is a lie. They go to the supermarket or any store and they can "estimate" the index on their own heads. And things do not add up to what is published. MIT's billion dollar project includes an index where they follow a - albeit smaller - selection of goods online and report on them. This MIT inflation index is not perfect and probably would not exist if people trusted the official inflation. The orange line in Figure 1 shows their index, while the blue one is official inflation. Clearly, the official one is usually below half of the one built from online prices.
Figure 1: Argentina's official versus independent inflation indices.
Figure 2: Argentina's life cycle of income (monthly), 2012.
This will not come as a surprise to any Argentinean. They all know they cannot trust the official inflation index. However, I am always surprised to notice that do not comment on the reliability of other reports. It is relatively easy for everyone in the street to partially check the inflation index while going to the supermarket. But it is absolutely impossible for a regular person on the street to check if GDP has grown. If the government or the statistics agency is willing to lie on inflation - which we can all easily check - imagine what they can be doing with reports on things like GDP, unemployment, reserves or government spending/revenue - which are very hard to measure.
What can we do about it? Probably not much. But in order to try to reduce some of these issues I decided to look at the micro data available. The Encuesta Permanente de Hogares is a regular household survey (also held by the INDEC) where people are asked several questions on their education and economic status among others. My hope was that looking at something that is not directly published from this, maybe I can extract some more reliable information. I am not certain about this but here we go.
Looking at the life cycle of income the pattern found is surprinsingly similar to that of the US in the 1970s. (Are we really 40 years behind?) Using the EPH for 2012, the shape observed is very similar. The peak is observed around the age 40, a lot earlier than currently in the US, but similar to them in the 1970s. But the level of income is shockingly low, with an average monthly income at the peak age of below 4000 Argentinean pesos (between 300 and 500 dollars, depending on the dollar exchange rate used). Similar calculations for 1995 give a similar shape but an average monthly income at the peak of twice as much, around 900 dollars.
Going from the average, at the peak age, of say 500 dollars a month to the published value of GDP per capita of over 14 thousand dollars seems a bit complicated to me (12 months x 500 = 6000 dollars a year, which is a lot less than 14 thousand...). Remember that the average of 500 dollars was based on people who have jobs. Unemployed people were not included. The sample is small and results may not be very accurate due to that. But even then, you still need to add all the people who are not working but still count for the "per capita" part of those 14 thousand dollars. This is the issue with Argentinean data nowadays. We cannot trust them and is hard to find out if we researchers are making a mistake in our calculations or if things simply just do not add up.
Figure 1: Live First-Birth Rates by Age of Mothers
The 1960s were revolutionary times. As Bob Dylan - one of my favorite musicians and probably one of the most famous characters of that time - said, "there is nothing so stable as change". This was certainly true in the US at the time: The Civil Rights Movements, social unrest due to the Vietnam War, the invention of the microchip, antidiscrimination legislation, the women's movement. And the invention of Enovid, the first contraceptive pill. Yes, you read right. The contraceptive pill was a revolutionary element. And as such, it has also been studied by an economist (and by the way published in the Quarterly Journal of Economics, among the top 3 economics journals). Martha Bailey evaluated the effect the release of this little pill in 1960 had on female labor participation. Gary Becker had previously said that "the contraceptive revolution [...] has probably not been a major cause of the sharp drop in fertility". However, Bailey will show that even if fertility did not decrease because of the pill, it did delay it, allowing women to get more education and improve their labor outcomes.
Figure 1 shows trends in first-birth rates by age groups since 1940. A marked decline in childbearing among young women (focus on 20-24 years old) is seen since the pill was introduced. This lasted until 1976 when all unmarried minors were allowed obtain contraceptives under the law. Early access allowed women between 18 and 21 to get access to the pill and hence the largest decline is seen for those 18-19 years old. A first robustness check can be seen from those 15-17 years old. Since they are expected to be too young to benefit from the pill, we should and do observe no effect for them. This gives us confidence we are not just seeing a spurious result.
As the diffusion of the pill increases, the distribution of age at first-birth also changes. Figure 2 plots the fraction of women first giving birth by age groups and cohorts. Among women born before the 1940 who were too old to benefit from early access to the pill, around 62% report having children before age 22. For those born around 1955, this had dropped by 25%. Notice that both figures suggest that these effects were not due to preexisting trends. Also no changes are seen between 1955 and 1960, when all women would have already had access to the pill.
Figure 2: Distribution of Age at First-Birth, by Cohorts.
And where does the economics come in? Early access to the pill was reflected in female labor force participation. Before 1940 the increase in women's participation had been driven by married women over 30 years old, who returned after their children had grown. On the other hand, for those born in 1955 the "fertility dip" is not observed any more. Participation rates were 25% higher at age 25.
Figure 3: Labor Force Participation, by Age and Cohort.
But how can we disentangle the effect of the pill from all the other things going on the 1960s that I mentioned above? Here is were econometric tools come in. The expansion of the pill was different across states, which individually changed the legal rights of individuals ages 18 to 21. Indirectly, this effect empowered women to get early access to the pill, without parental consent.* This exogenous variation will allow Bailey to compare the effect of the pill on women's life cycle labor force participation. Just to fix ideas, the methodology is like taking two states that were previously equal. But one state decides to extend legal rights to younger individuals and the other does not. Consequently, only one state allows young women to get access to the pill. Then, the difference in the labor force participation of the women between the two states will be coming from the pill. More than two states and more controls are used to obtain the results, but the intuition of the technique is in the previous simple example.
A first thing to check is whether early access to the pill had an effect on fertility. Table 1 shows the baseline estimate (column 2) is that it reduced the probability of giving birth by age 22 by 14%. Interestingly, early access to abortion does not seem to drive the results (column 3). As expected, it did not reduce the number of children before 19, since women did not have legal access to the pill without parent consent before that age. Finally, as other people had reported, the pill did not reduce the number of children women had, suggesting it just delayed it.
Table 1: The effect of early legal access to the pill on fertility.
What effect did this have on labor outcomes? Bailey shows that early access to the pill increased labor force participation of women ages 26-30 by 7%, and also increased those of ages 31-35. They also seem to work more hours, hence getting closer to male labor outcome averages. For women under 25 years old, results suggest that the pill increased their enrollment in school. Changing career trajectories - resulting from delay in childbearing - was the primary mechanism this little pill increased female labor-force participation.
* Bailey goes into some detail to justify that this extension of rights was not related to states characteristics that could be directly related to the variables of interest. Most of the changes are suggested to have to do with discrepancy under federal law of being old enough to be drafted to the Vietnam war by age 18, but not being able to vote. At the state-level, legislation was extending rights to 18 year old men and women.
A few weeks ago I wrote about the life cycle of earnings, where Guvenen, Karahan, Ozkan and Song had used over 200 million tax data observations from the Social Security Administration (between 1978 and 2010) to see how income moved with age. With that amazing data they showed that mean income would peak around the age of 50, though for the median person income would peak earlier. Given their findings I decided to look (though with worse data) at how the life cycle of earnings has changed over time.
Using Census data from the US (available through IPUMS for any other data addicts reading this), I looked at average income by age for each decade. The caveat from using this information is that if there are some cohort effects (meaning earnings are changing differently for young than for older people within one decade) I will not capture this directly, possibly leading to some confusion in the analysis. Nevertheless, the patterns are quite striking.
Figure 1 shows that average labor income used to peak a lot earlier than it does nowadays. Back in the 1960s, it used to peak around the age of 35. Income was expected to start going down after 35. However, decade by decade, this peaking point has been increasing. By the 1990 the peak seemed closer to 45, and nowadays the peak can be as high as 50. Given the results found in Guvenen's research it might be that nowadays the median worker's labor experience is much more different from the mean worker than it used to be. But why?
Figure 1: Average Income by Age, over time.
Figure 2: Relative variance of Income by Age, over time.
Source: IPUMS Census USA. Family labor income by age of head, excluding people in school or with no income.
One possibility is the increase in the share of people going to school and looking for skill demanding jobs. Back in the 1960s, the share of young people who were high school graduates was around 53%. Another 13% had graduated from college as well, hence leaving a 34% of high school dropouts. Nowadays, there are only 10% high school dropouts, while the share of college graduates has increased to around 34%. I believe this might be pushing the peaking point. For example, an engineer or lawyer probably needs to go through some lower paying job training (or internship) and needs to try many different offices until it finds the one that suits him best. Hence, they start with a quite low pay but see a high increase over time. On the other hand, a construction worker's income will probably not change much over his life. Most companies will probably pay similarly, and his wage will not change as much over his life as it will for the lawyer/engineer.
This is consistent with what is found for the variance of income. Figure 2 shows variance relative to the variance at age 26, so that we can see how it moves over life. Another interesting pattern emerges here. It used to be that income differences were quite constant until the age of 40. However, since the 2000s differences seem to have started showing up earlier. A constant increase since the age of 25 is found nowadays.
Source: IPUMS Census USA. Variance of log family labor income by age of head, excluding people in school or with no income, relative to variance at age 26.
Once again, I believe this probably might have to do with education. Back in the 1960s, more than 80% of the population would start looking for jobs around the age of 18, leaving many years until the age of 25 - where my plots start - for them to find the appropriate job.* Moreover, these types of jobs probably did not experience much wage differentials between employers. On the other hand, nowadays, 34% of young people graduate to college (and even attempt go and fail to graduate), leaving them with less years to find a job. Moreover, once again their skills probably need more time to find the appropriate employer ("matching" issues in the economics jargon).** Hence, incomes are a lot more varied nowadays since earlier stages of life. And my income peak is getting farther and farther away...
* Plots start at age 25, so as to avoid having selection issues with people who go to college and start showing up in the sample after they graduate (say at age 22). For example, if plots started at age 18, the data until age 22 would include only people who did not go to college. Starting age 22 the pool of people would change a lot, as college graduates come in. The mean income might change significantly, but this would not be due to the life cycle of earnings of the workers, just because the pool of people in the data changed. Hence, this problem is reduced by starting the analysis at age 25.
** An interesting way to evaluate this would be to look at the same data but focusing only on college graduates. Maybe another week.
Given the recent events in Ferguson - where white policeman was not indicted for shooting a black young man - that lead to protests around the US to try to stop racial discrimination, I thought it was a (unfortunately) perfect moment to see what economics has to say about this. What is the status of inequality between black and white? Using research from the Urban Institute, Figure 1 suggests that white people have 6 times more wealth than blacks, having this gap increased almost threefold since 1983. So it seems the situation has not got better over time.
Figure 1: The wealth gap in the last three decades
Source: The Urban Institute.
Moreover, whites accumulate more wealth over their lives than black (or Hispanics) do. Focusing on those born between 1943 and 1950, Figure 2 shows that this wealth gap increases over the life cycle. In 1983, whites between 32 and 40 have an average family wealth of $184,000, rising to over a million by age 59 to 67. However, blacks wealth goes from $54,000 to only $161,000 between the same ages. So whites have three and a half more wealth than blacks when they are young, but over seven times more when they are old.
Figure 2: The life cycle of wealth by race
Source: The Urban Institute.
On top of this wealth gap, an over-simplistic look at the data suggests that blacks receive worse sentences and are more likely to be suspended in school. Finally, Figure 3 shows are twice more likely to be unemployed.
Figure 3: Unemployment rate by race.
Source: Bureau of Labor Statistics.
What is behind such big gaps? Econ 101 teaches us that a properly working market system should hire and pay people according to their value. Discrimination makes no sense in competitive environment. Suppose every employer is discriminating against blacks, hence providing them with a lower wage even though they are as productive as whites. This would allow any unbiased person to take over the market. She would be able to hire these discriminated people at a wage level in between the gap (i.e. between the black and white ones) and get a better profit than everyone else, possibly kicking all racist businessmen out of the market. This Econ 101 logic is definitely too simplistic, but should let us frame our thoughts to see what it is missing out.
A possible issue is that blacks and white differ in their characteristics, beyond their color. For example, white people could be more educated. Focusing on unemployment, the question then is: When faced with observably equivalent (i.e. education, experience, etc) black and white job applicants, do employers favor the white one? Evidence goes both ways. Some suggest they do not, claiming the black-white gap stems from supply factors: African-Americans lack many skills when entering the labor market, so they perform worse. Others suggest that employers do discriminate, either by prejudice ("Taste-biased" in economics jargon) or, more usually, by what economists call "statistical discrimination": race is used as a signal for unobservable characteristics. For example, if blacks tend to be raised in worse environments (which could lead to worse productivity), employers who care about productivity (but do not about race) and cannot observe it perfectly, would use race (or ZIP codes) as a signal for it. Hence, black people would be discriminated but not (directly) because of their color.
Data limitations make it difficult to test these views. Researchers posses far less data than employers do, so even if applicants appear similar to researchers they may not be to employers. Employers can observe social skills during interviews and assess the quality behind what is stated in the typical resume information. And any racial difference in labor outcomes could easily be attributed to that. That would be a highly unsatisfactory open ending to this post.
Fortunately, Bertrand and Mullainathan designed a field experiment to try to circumvent this problem. They sent close to 5,000 resumes to more than 1,300 help-wanted ads and measured the call-back for interview for each resume. Since race cannot be explicitly written in the resume, they manipulated the perception of race by (randomly) assigning names to those resumes. Half the names used are white-sounding (e.g. Emily Walsh) while the other half is black-sounding (e.g. Lakisha Washington). A side experiment showed that the names used are associated with their respective races by more than 90% of the people. They also experimented on changing the quality of the resumes, in order to see if call-backs for black are more responsive to quality than for white (like statistical discrimination might suggest). Approximately four resumes are sent to each ad: Black-High (quality), White-High, Black-Low, and White-Low. Even though this does not go further than the call-back stage (i.e. it does not go all the way to employment), this methodology guarantees that the information the researcher and employer have is the same.
Table 1 shows the callback rate for both groups: Whites have 50% more chances of being called back. A white person would need to send 10 resumes to receive one callback, while a black one would have to send 15. Using the data on quality of the resumes (Table 5 in the paper), the return to a white name is equivalent to as much as eight additional years of experience. Moreover, there seems to be no difference depending on the industry or occupation category of the job itself. They all show differences of this sort.
Table 1: Callback rates by age.
A possible issue with this strategy is that when employers read a name like Lakisha, they may assume more than just skin color. They could interpret that the applicant comes from a disadvantaged background. In such a case, signals of quality like experience or special skills should be more important for black applicants. Similarly, ZIP codes could be used to get an idea of the social background of the applicants. If we expected statistical discrimination to be behind the gap, we should expect black applicants callbacks to respond more to either of this. However, the study suggests they don't. Higher quality of resumes improves the callbacks for white applicants but not so much for black ones. And ZIP codes don't seem to matter much either. Finally, a way of looking at this directly is by examining the average social background (proxied by mother's education) for each name used. Table 2 shows the first names used, together with their callback rate and average mother high-school completion rate. The social background hypothesis would suggest higher callback rate for higher mother education. However, no such evidence is found.
Table 2: Social background and callback rate for each name used.
If statistical discrimination is not behind this, what is then? "Taste-based" discrimination where people consciously think worse about blacks seems contradictory to other studies in the literature. In a second paper the same authors (together with Chugh) suggest that a possible explanation is that we might be unintentionally discriminating. Using a tool popular in neuroscience and sociology, the Implicit Association Test (IAT), they suggest that people have unconsciously more difficulties in associating black persons with positive words. And this is found to be harder to control in environments with time-pressure or considerable ambiguity (like looking at job applicants).
What is the best way to improve on unconscious discrimination? Is making differences between skin colors, that go as far as avoiding any topic that refers to colors which are as obviously there as any other parts of our bodies, the correct way to improve our unconscious mind? If we are raised with these concerns of what is politically correct to say, we might be doomed to unconsciously make such an unfair and damaging difference between people's skin colors.
How do individual labor earnings evolve over the course of a person's life? If you have ever asked yourself "Should I expect my income to increase this year?" and "By how much?" this post might interest you. In a very elegant study, Guvenen, Karahan, Ozkan and Song have tried to answer these questions and more, using over 200 million tax data observations from Social Security Administration (between 1978 and 2010). If you ever thought tax data was not public, this (and my last post) might suggest that they are not. Don't worry. Only a few people are allowed to use this information, and even then they are not allowed to actually see the name of the person behind each income observation.
Looking at employed people between the ages of 25 and 60, they focus on how much earnings grow every year in average. A first look at the data is provided in Figure 1. Taking the average among all the population, yearly income peaks around 50 years old, with an increase as high as 127% from age 25.
Figure 1: Average (Log) Earnings by Age.
If you are past age 50 and have not seen such an increase, you might be wondering what's wrong with yourself. Before entering into such a depressing state of mind, please read a few more lines. This average income path hides a lot of variation across different people. More importantly, it is strongly influenced by the very top earners. Figure 2 shows that the median worker only shows a 38% increase in his earnings between age 25 and 55. It is the very top earners who influence the 127% number before. For example, the Top 1% shows 1500% increase in their earnings in that same period. More than 300 times the median increase in earnings...
Figure 2: Earnings Growth (25 to 55) by Lifetime Earnings.
Another interesting finding is that income does not peak at the same age for everyone. Even though the average person's income peaks around the age of 50, this is not the case for most people. Figure 3 shows that the median worker has almost no income growth between 35 and 45, and only the top 2% actually experience earnings growth after 45.* I hope these depressing findings for the median worker might help your self-confidence. The average numbers shown in Figure 1 are not the appropriate ones to question your life. (Figures 2 and 3 might be...)
Figure 3: Earnings Growth by Decade of Life.
Some other interesting findings in these article are that the dispersion of income growth (i.e. how much income growth differs across individuals) has a U-shape, decreasing with age up to when people are 50 years old where it spikes up again. Top earners are the exception once again, since their income dispersion grows every year of their lives.
How about the asymmetries? Is it more likely to be below or above the average increase in income? The data suggests that as people get older or richer, it is more likely to get negative shocks to income. And these seems to be due to there being more room to fall down (not less room to move up). The higher your income, the more you can lose (remember most people are not willing to pay to work, so you cannot have negative wages).
Finally, let me end with a happy note. Suppose you just saw your income go down. You might be worried that it will remain like this for a long time. The data suggests otherwise. If the decrease was very strong, it is most likely that the persistence will be very short (unless you were a very high earner). In less than a year you should see your income recover most of its previous value.
(Very Small Print Note: this does not mean you should just lay down and wait for this fact to bring your salary back to normal. No complaints are accepted if incomes do not go up.)
* Remember that if a distribution is such that there are a few outliers with extremely high income growth, we will observe an average growth much higher than the one the median worker has. Hence, focusing on the median worker might be more illustrative in these cases.
Based on an article by Guvenen, Karahan, Ozkan and Song.
How much do children's social and economic opportunities depend on parents' income and social status? This is a politically correct way of asking: How doomed are children from poor parents?. The answer is essential to analyze policies that try to make every kids chances more equal. As always, a first step is to analyze what the data has to say about this. Fortunately, Chetty, Hendren, Kline and Saez (economists at Harvard and Berkley) are currently doing some beautiful analysis on this matter. Since opportunities are hard to measure, they focus primarily on income (although they also study education, crime or pregnancy) differences.
Using tax-income data on 40 million children born between 1980 and 1982 and their parents, they are going to rank people according to their income level. Parents are going to be ranked in groups from 1 to 100, according to how they do income-wise relative to other parents. Similarly, children are going to be ranked according to their incomes when they are 30-32 years old relative to the other children. Then, they are going to focus on two measures of intergenerational mobility:
1) Relative Mobility: What are the outcomes of children from low-income families relative to those from high-income ones?
Example: If my parents income increases by one ranking point, how much is my income rank expected to increase?
(The problem with this measure is that higher mobility may be due to richer people doing worse, not poor ones doing better. Hence, the second measure might be more useful.)
2) Absolute Mobility: What are the outcomes of children from families of a given income level in absolute terms?
For example, what is the mean income of a child born from parents in the 25th percentile?
The chart below shows the national statistics of the rank-rank (relative mobility) relationship in 3 countries: Canada, Denmark and US. The slope in the US is 0.341, while the other two are half that much. This suggests that increasing one percentage point in parent rank, increases child mean rank by 0.341 percentage points. The fact that Canadian and Danish data suggest higher relative mobility should be taken with caution since this could be due to worse outcomes from the rich, rather than better ones from the poor. Interestingly, this strong correlation with parents income rank is also observed in children's college (attendance and quality) and teenage pregnancy, suggesting differences emerge well beyond the labor market. This is consistent with evidence from my previous post.
The previous chart suggests that the rank-rank relationship is highly linear. Hence, the authors are going to take advantage of this when analyzing the intergenerational mobility across different areas in the US. The question now is: Is mobility the same across the US? Or are some regions better for children to make the jump forward? Given the issues with relative mobility, we can now focus on absolute mobility: What is the mean income of a child born from parents in the 25th percentile? The heat map below shows that the Southeast shows the lowest mobility in the country, while the Great Plains, West Coast and Northeast display much higher mobility levels (the map should be read the map as darker is worse mobility). While in some regions children of parents in the 25th percentile tend to remain in the same percentile when they grow up, in other areas similar children do twice as better (in income rank terms). This pattern seems robust to controlling for children moving to other areas and cost of living or demographic reasons like marriage differences.
The obvious next question is why are regions' mobility so different from each other? Why children in some areas seem to be born with more opportunities than those in other ones? This question is not directly addressed by the authors, but they provide some correlations with local characteristics. Given econometric issues like selection and endogeneity (also explained in a previous post!), the following should NOT be interpreted as causes.* However, they show interesting descriptive information.
1) Race and Segregation: The higher the share of African-Americans, the lower the mobility observed. However, the data suggests that this holds true for the white people in those areas as well. Hence, it is not that black people tend to remain stagnant. Segregation in the area seems to be correlated with everyone's mobility. Particularly, segregation of poverty seems to be the strongest reason (isolation of rich people does not seem to be behind). Some potential reasons could be: successful role models are not present for the poorest children; worse public goods provision; or access to jobs might be harder in such areas.
2) Income: The average income level is not correlated with mobility (i.e. it is not that richer areas do better or worse). However, areas with higher income inequality show lower degrees of mobility. Interestingly, the inequality in the upper tail is not correlated with mobility. Hence, it is not about the existence of some extremely rich people. It is more about the size of the middle class. The bigger the middle class, the higher the mobility.
3) School Quality: Better schools are associated with higher mobility.
4) Social Capital: Social participation in elections, census or even religious events is positively correlated with mobility.
5) Family Structure/Stability: The higher the number of single parents, the lower the mobility. Once again, this effect extends to children who are born from parents who remained together, suggesting that the effect is not at the individual level but at the social environment one. Regions with more divorce somehow have lower mobility.
To summarize, parents income seems to be very important on children opportunities. However, there is substantial variation across different areas in the US. Some areas seem to fit much better than others the concept of "Land of Opportunity." A child raised in the Great Plains has much better chances of making a leap forward than one born in the Southeast. Segregation, inequality and family structure are highly correlated with mobility. Unfortunately, why remains a mystery.
* Families choose where they live and what institutions they support. So we can imagine that families that prefer to live in areas with better education systems or less income inequality are intrinsically different than those that prefer to live in the more segregated South of the US.
Many major economic and social problems such as crime, teenage pregnancy, dropping out of high school and adverse health conditions can be traced to low levels of skill and ability in society. The figures below show that (measurable) ability is highly correlated with having been in jail or being single with children. Economists have always been interested in the way typical goods (say agricultural or industrial ones) are produced. But if these skills and abilities are that important in life, shouldn't we focus on better understanding their production process?
Ever been in Jail by 30 years old, by ability (males)
Probability of being single with children (females)
Would it be simplifying but relatively fair and innocuous to summarize all our skills into one word like ability or intelligence? Or is it important for us to recognize that there is more than one skill? For example, the No Child Left Behind Act concentrates attention on cognitive skills (math, reading) through achievement test scores, not evaluating a range of other factors. However, noncognitive (personality, social, emotional) skills seem to be very important as well. They contribute to performance in society at large. Gaps in skills seem to be present early in the lives of children, being family environments very good predictors of them. The chart below shows that Children's cognitive skills gaps are present as early as 3 years old and are strongly related related to mother's education.
Mean Cognitive Score by Maternal Education
Can early intervention fix these differences? Economically speaking, are these skills better "produced" when children are one year old or can we later remediate them when they are older (through primary and high school)? These issues are essential when analyzing public policies to improve education or adult socioeconomic behavior. Current policies, like reducing pupil-teacher ratios, focus on later remediation. But shall we improve schools? Or is it better to focus in educating parents so that they can take better care of their children in the first three years of their lives?
Heckman has started a major project in order to be able to understand these important questions. But (good) economists like using data carefully in order to answer questions objectively. It's important to recognize that we don't have direct data on these skills. People don't carry a number with them saying "I have cognitive skills of level 2 and noncognitive ones of level 5". So Heckman and his coauthors are going to do some heroic econometric work to get around this.
They assume that these unobserved factors are related to children parents (through their also unobserved skills, income, education, etc) and their "investments" in their kids (reading them, taking them to museums, etc). But how can Heckman estimate the effects of things we don't observe? Basically they are going to assume that these (unobserved) skills are related to test results and later outcomes in life like education, crime, early pregnancy and many others. Taking into account how these multiple outcomes and investments correlate with each other, will allow them to estimate these two set of skills and give them the information they are after. Notice we need very large amounts of information on the same children and their parents at many periods of their lives. So where does the data come from? It may be hard to believe, but many countries have such data. For example, over 10 thousand American children (and their mothers) have been followed since they were born, which will let Heckman estimate what matters in child development and how we should distribute our efforts in order to improve it?*
Let me give you an idea of the amount and type of questions these families answered when they were interviewed, usually every two years. From these surveys, we know: children's gestation length weight at birth, memory for locations, picture vocabulary, standard test results (on reading and algebra), friendliness, sociability and behavior problems; whether their parents read them, how many books they have, how often the family eats together and whether they go to museums or concerts; their mother's arithmetic skills, self-esteem and their family income and savings. And many many more. A crazy large amount of information is collected in a regular manner from these same people. A by product of this study is that we can find out what seems to work best in improving children's skills. Interestingly, how often the mother reads to the child or if they eat together during the first year seems to be some of the most important factors. Similarly, once the children grow older (6+ years), going to museums or concerts seem to become very important for their development as well. (If you are thinking about applying this nowadays, you should take this with a grain of salt. Keep in mind these children grew up in the 1980s so they did not ask to go to Miley Cirus concerts. Music was probably better then.)
So what about the production process of these skills? Is it better to invest in the first years or we can achieve the same results by "fixing" children's bad initial years when they get older through better education systems? Heckman's results suggest we should focus on the first three years of their lives. Parental investment in these years has a much greater impact than later ones. Moreover, during these early years improving one set of skills seems to increase the quality of the other one. Skills beget skills.
And are the two skills equal? No. Children's cognitive skills tend to stabilize early in life (say around 6 years old) and are difficult to change later on. On the other hand, social skills seem to flourish when children are between 6 and 14 years old. In case you are wondering whether economics has gone mad, let me say that this seemingly crazy study suggests that what happens in these early years of life can explain over 50% of the years of education, criminality or teenage pregnancy. This is very relevant. Moreover, it suggests that if governments were interested in improving any of these outcomes, they should try to invest very early (before schooling years) in the disadvantaged. Possibly educating parents on the importance of reading to their children or taking them to social activities might help. How to approach this parental education is the next issue at hand.
* They are actually going to use only 2000 first-born white children in their estimates, in order to avoid issues related to my last week posts. They want children to be as similar as possible, in order to avoid capturing wrong effects in their estimates. They also allow for endogeneity and measurement error in their estimation process.
How much is education worth? I personally believe that education provides you much more than just a higher income in the future, but let's simplify and focus here on how much more money you would earn if you studied a few more years. This question has interested economists for a long time (at least since Mincer in the 1950s). However, this holy number that could potentially explain whether I am (economically speaking) wasting my time doing a PhD has proved very hard to estimate. But let's start at the beginning.
Around the end of the 1950s, Mincer proposed - after building a rather simple model of human lives - doing a regression of (log) income to years of education and years of work experience. If you did this you would basically find that one year of schooling increased your income by between 10 and 14%. However, recent analysis by Heckman (and his army of co-authors) shows that returns to education are not that easy to estimate. Particularly, the effect of each year of education may not be the same across levels (e.g. the year you finish high-school is worth a lot more than any of the previous) and that education might change your future income growth due to years of experience (e.g. if you finish high school each year of work may bring you a higher increase in income than if you did not finish high school). This is seen in Table 3a below for white men both in 1940 and 1990 (bottom row is Heckman's and top row is Mincer's). Each column refers to a different year of education: 10-12 is finishing High School and 14-16 is completing college.
Another interesting finding here is that this would suggest that finishing high school in the 1990s is worth a lot more than it used to be in the 1940s (50% vs 24% income increase). Are we learning more nowadays? Possibly. But possibly not as well. This could come from a well known issue to applied economists: selection bias.
Let me introduce this concept with a (hopefully) clearer example that comes from the health sector: the number of deaths and hospitals. For every 100 people hospitalized for diagnosis in the US more than 2 die every year. On the other hand, for every 100 people in the US only 0.8 die every year.* Hence, comparing the two pools of people, anyone ill might think: "Wow! If I go to the hospital I increase my chances of dying by more than 100%. Then, I should stay away from hospitals and try to get better on my own." But this clearly makes no sense (at least not if you have health insurance!). You are comparing a pool of people who are sick (hospitalized) with a healthy one (everyone else). If the first one stayed out of the hospital, we would expect that more of them would die. And the same logic applies to education.
Focusing on the 10-12 column, these regressions are (intuitively) comparing people who finished high school with those who did not. But are these people equal in all other terms besides having finished school? Most likely not. We can imagine that people who don't finish high school have had a worse childhood, come from worse neighborhoods and are generally raised in a more distressed environment ("sick" in the hospital example above). This would suggest that even if this people did go to school they might do differently (worse?) in the labor market later on. Similarly we can imagine that the people who did finish high school had families with a better economic background who could more easily provide job opportunities to their children, hence increasing their labor income independently of schooling choices. Basically, the two groups of people cannot be compared directly. Hence, the increase in the observed returns could be because the pool of people who don't finish High School nowadays is (relatively) worse than the one in the 1940s. Most people finish high school nowadays, while this was not the case 60 years ago. In other words, the selection bias could have gotten worse over time.
What economists might like to do to solve the enigma of schooling returns is to randomly assign people to different education levels. Someone would be flipping coins and deciding everyone's education. This way we would be able to make sure that all kinds of people are equally distributed across the different education levels. And so the income levels of the different groups could be easily compared. Fortunately, economists are not allowed to dictate people's lives that much. And the best solution so far has been to look for uncontrolled events that make (some) people more likely to go to school (but are not related to their wages in the labor market directly). And then we compare this group to some other (similar) people who were not affected by such an event. A nice example comes from Seth Zimmerman and his estimation of returns to college admission.
Zimmerman focuses on a large public university in Florida (FIU), which was particularly easy to get in when compared to other universities (kind of like a last-resort university. Apologies to any FIU students reading this!). This way he can be more confident that if someone was not admitted there, they would not be admitted by another school. But, how does he separate people randomly into the two groups (admitted versus not admitted)? His trick is to take people just around the GPA admissions threshold. Figure 4 shows that people right above it are 23% more likely to be admitted to this university than students just below it (and more likely to attend as well).
Assuming that people are not able to control their GPA at this particular university, this would provide him with people being "randomly" assigned to "admitted" and "not-admitted" groups.** Hence, we can now compare the income across the two groups worrying less about selection. Figure 8 suggests that being admitted to college (i.e. from being just above the threshold) increases your income by around 22%.
It is important to notice a few limitations of this kind of studies. These econometric techniques don't come for free. This number is the return to college admission only for people who, for various reasons, are near the threshold. And once again, this people might be very different than the ones who had no trouble being admitted. So the 22% rate should be understood as the return to this particular group of people and not for everyone else. Nevertheless, this number might be the relevant one if you are thinking about a policy that changes the requirements for admission. Such a policy would affect this particular group and not the general population. Moreover, another drawback is that this return does not consider "General Equilibrium" effects: If such a policy were applied in all the country we would expect to have lot more people graduating in the next few years, which might affect the wages of college graduates. Hence, the returns to education might change.
Economics research can (sometimes) be extremely difficult when compared to physics and other such sciences. In these sciences nature's rule is well defined. It may be hard to understand but it is there, and all the data you observe is from such a rule. In economics, data observed is driven from people's different lives, crazy personalities, complicated families and interestingly different regions of the world. And on top of these differences (which we can't observe in the data), individuals are making choices which manipulate the data we economists try to work with. Hence, understanding humans and outcomes related to them can be very complicated. Like Stephen Hawking once said,
"While physics and mathematics may tell us how the universe began, they are not much use in predicting human behavior because there are far too many equations to solve. I'm no better than anyone else at understanding what makes people tick, particularly women."
* I would like to have the number of deaths of people not hospitalized to make the comparison with education but I wasn't able to find that number easily.
** Note that GPAs are computed differently by various universities. So students would need to preview that they want to apply to this particular school and be able to control their results very precisely to affect their result around the threshold. People way above the threshold can be easily thought to be very different, but people just above and just below the threshold are probably quite similar.
With the recent constant appearance of Alibaba on the news, the increasing relevance of Chinese exports to the world is extremely clear. Low-income countries accounted for just 9% of US manufacturing imports in 1990. But by 2007, they had more than tripled its share. And who do you think was behind this? China accounted for as much as 89% of this increase.
In this period, China's transition to an open economy included a massive 150 million people migrating from rural to urban areas. Imagine reallocating around half of the United States population geographically, with a particular focus on manufacturing production. Add to this formula novel access to foreign technologies as well as capital and Chinese exports growth to seem reasonable. However, did this come to the expense of anyone? This is the main objective of this post. One group being threatened by Chinese takeover of manufactures is obviously the manufacturing workers in the rest of the world. As these goods are easily tradable, we could expect job losses in these sectors. The figure below shows that as Chinese increased its relevance in US imports, the share of the population working in the manufacturing sector in the US decreased by one third.
However, many things could explain this decline. For example, it could be that Americans themselves were getting more educated and moving to other sectors. Alternatively, the service sector could be becoming more productive in the US, offering higher wages and hence draining employees from the manufacturing industries. These (and many other alternatives) do not involve China's exports growth. Moreover, they could be causing the increase in Chinese exports themselves. (Imagine US decides to get out of the production of manufactures, leaving a lot of unsatisfied demand which leads the Chinese to produce more). Hence, in order to make sure we are capturing the correct effect, modern econometric techniques come to the rescue! Autor and Dorn (AER, 2013) basically exploit the differences in the exposure to import competition across cities in the US. For example, it would be expected that - if the leading cause comes from the Chinese side - an area where manufacturing employed 25% of the people to be more affected by Chinese exports than an area that only employs 10% in manufacturing. Particularly, they will differentiate areas by how specialized they are in each division within manufacturing, and how imports from each of these changed over time. And these differences will give us the information we are after.
Looking at wages, the effect found of imports from China is negative. An increase in the imports per worker of around three thousand dollars (which was the average change from 1990 to 2007), would explain a decline of around 2.25% (0.76 times 3). More interestingly, this effect is stronger among men and people without college education. It is important to remark that this can only be calculated for the employed. Hence, if we expected workers with lower ability and earnings to be more likely to lose their jobs after the Chinese expansion, the effect on wages above would be understated. And so wages would have fallen even more for the whole sector, it's just that the effect could be hidden by the increasing number of people losing their jobs.
And what if we divide the effect between sectors? I would have expected the wage effect to be stronger in the manufacturing sector itself. But well, the data seems to suggest the opposite: wages seems to have been unaffected in this sector. However, the manufacturing sector was particularly affected by a major reduction in employment (predicting a decline of 12% due to China's increase in exports).
So most of the effect on wages mentioned for the whole economy seems to come from the non-manufacturing sectors. How can this be possible? Well, (adaptive) story telling is a prerequisite for any upstanding economist. And here is the one that seems most appropriate given the results: the increase in imports from China led to firing of the lower skilled workers in the manufacturing sector but had no effect on their average wages (note this could still involve a decrease in the wages of the ones that remained employed). Having no new paychecks, these newly unemployed decided to reduce spending and so decreased their purchases of services that have to be provided locally (like a haircut or a dentist). This reduced this local sector's revenue. Moreover, the newly unemployed also fled to other sectors looking for jobs. Having lower revenues and seeing lots of people of willing to work for less, other sectors reduced their wages.
Based on an article by David Autor and David Dorn (AER, 2013).
Follow me on twitter at @diedaruich
News and posts for an active mind.