When the Equality of Opportunity Project released new data on mobility at the end of January the initial headlines focused on the authors’ overarching finding that mobility at the national level had basically stayed the same while inequality had risen over the past half century. Most of the media coverage, however, missed the nuances at local and regional levels that were discussed in the paper. So we decided to use the data to make some visuals that help pull those nuances out.
In this post you’ll find five maps that examine changes in local and regional mobility measured by income mobility, college mobility, and a composite mobility measure. The different presentations of the data demonstrate that intergenerational mobility over this period of time in the United States has changed substantially by region even if the national average remained stable.
As you’ll see, the first three maps show that by several measures of mobility the South and West experienced the highest gains in mobility over this period, while much of New England, the Rust Belt and upper Midwest saw declines in mobility. But when looking at the level of mobility, the South has remained among the lowest despite the improvements while most of the West started with fairly high mobility and has generally gotten better.
Let’s start with changes in composite economic mobility at the local level. Check out Map 1 below. (Note: Green areas saw an increase in mobility, red areas saw a decline, and the color intensity reflects by how many standard deviations mobility changed—how much mobility increased or decreased.
Source: Washington Center for Equitable Growth
To make these maps, we standardized the mobility changes for each measure. In technical language, we divided the difference between the initial measure and the weighted mean value by the weighted standard deviation. Thus, for each of the standardized variables, the average value is zero and the standard deviation is one.
The data in the map above are a composite measure of mobility that uses the income and college mobility data from the Chetty et al (2014) data. To get an idea about how economic mobility changed over time, we calculated the change in mobility across birth cohorts. For income mobility, this is simply the difference between the 1986 value of relative mobility and the 1980 value. Likewise, the change in college mobility was determined by taking the difference between the 1993 and 1984 values. The composite measure was constructed by summing the standardized income and college mobility changes for each commuting zone. It is important to note that because both the college and income measures include the years 1984-1986, those years are double-counted in the composite measure—though this does not appear to impact the results substantively.
To be clear, the years referenced in the data set are the years an individual was born. The data for 1987, for example, captured information on individuals born in 1987.
Now you may be asking if the trends look different for the different data sources. To help answer those questions, we have Map 2, which uses data on income mobility and Map 3, which uses data on college mobility.
Source: Washington Center for Equitable Growth
Source: Washington Center for Equitable Growth
Looking at Map 1, we noticed that this seemed to be very different from the places that had high mobility in the starting period (which can be seen in the first working paper by the Equality of Opportunity team). Map 4 has the income mobility from 1980 shaded by decile. The red portions in the South, Rust Belt, and Southwest regions of the United States indicate that those are parts of the country with the lowest absolute mobility. Those with dark green, mostly in the West, Midwest and New England, have the highest economic mobility. Map 5 has the income mobility from 1986 shaded by decile which looks similar but distinctly different.
Source: Washington Center for Equitable Growth
Source: Washington Center for Equitable Growth
The take away: Not only does intergenerational mobility vary quite a bit in the United States but the changes in mobility vary quite a bit by region as well. And while mobility in the United States may not have changed overall, mobility has changed for individual local labor markets. If we want to understand the roots of intergenerational mobility then our best shot seems to be understanding the differences at the local level.
Interested in more about the data? Here’s some more information.
I’ll first try to provide an idea of what the data look like. Table 1 has the information from the new data set for the Johnson City, TN commuting zone:
Table 1: Sample of the Mobility Data Set
Commuting Zone |
CZ Name |
Birth Cohort |
Income Age 26 Slope |
Income Age 26 Intercept |
College Age 19 Slope |
College Age 19 Intercept |
Number of Children |
100
|
Johnson City, TN |
1980
|
0.275
|
0.353
|
|
|
5,941
|
100
|
Johnson City, TN |
1981
|
0.29
|
0.348
|
|
|
6,202
|
100
|
Johnson City, TN |
1982
|
0.304
|
0.332
|
|
|
6,337
|
100
|
Johnson City, TN |
1983
|
0.31
|
0.335
|
|
|
6,108
|
100
|
Johnson City, TN |
1984
|
0.284
|
0.349
|
0.942
|
-0.014
|
6,240
|
100
|
Johnson City, TN |
1985
|
0.295
|
0.35
|
0.933
|
-0.012
|
6,433
|
100
|
Johnson City, TN |
1986
|
0.3
|
0.341
|
0.903
|
-0.018
|
6,341
|
100
|
Johnson City, TN |
1987
|
|
|
0.929
|
-0.028
|
6,355
|
100
|
Johnson City, TN |
1988
|
|
|
0.885
|
0.012
|
6,449
|
100
|
Johnson City, TN |
1989
|
|
|
0.854
|
0.009
|
6,725
|
100
|
Johnson City, TN |
1990
|
|
|
0.873
|
0.018
|
6,831
|
100
|
Johnson City, TN |
1991
|
|
|
0.838
|
0.025
|
6,735
|
100
|
Johnson City, TN |
1992
|
|
|
0.827
|
0.014
|
6,557
|
100
|
Johnson City, TN |
1993
|
|
|
0.831
|
0.02
|
6,261
|
The column titled “Income Age 26 Slope” is the difference in the average income rank at age 26 of someone from a high-income family compared to someone from a low-income family. This was calculated from 1980 to 1986 for each commuting zone and serves as one measure of economic mobility as described in Chetty et al. (2014). We used this “Income Age 26 Slope” data for the income mobility measure. The column titled “College Age 19 Slope” is the gap in the college attendance rate at 19 between people from high-income families compared to those from low-income families. In both cases, higher numbers indicate lower economic mobility. We used this data for the college mobility measure.
For the Johnson City commuting zone, the income mobility of 26-year-olds born between 1980 and 1986 decreased slightly because the gap in the outcomes between the children of low-income and high-income families grew. For people born between 1984 and 1993 in the Johnson City, Tennessee area, the college mobility increased slightly.
Table 2 has the mean, standard deviation, population-weighted mean (for these, the population is the cohort size), and population-weighted standard deviation for the changes in mobility for the income, college, and composite mobility measures. For each of these measures, a negative value indicates a smaller gap between people born into low-income and high-income and thus higher mobility.
Table 2: Summary Statistics for the Various Measures
|
Mean |
St. Dev. |
Weighted Mean |
Weighted St. Dev. |
Income |
0.0082 |
0.0546 |
0.0088 |
0.0568 |
College |
-0.0613 |
0.1059 |
-0.0583 |
0.1123 |
Composite |
-0.0536 |
0.1249 |
-0.0500 |
0.1310 |
For each mobility measure we plot a histogram that shows the number of commuting zones that display a particular range of mobility changes. Each of the histograms peaks near zero, which suggests the most common change in mobility is little or no change in mobility. (Recall that, based on the way the variables were constructed, a higher negative value means higher mobility over time.) For the change in income mobility there is little variation over time, but the change in college mobility has a much higher variation over time, as reflected in the higher range of values displayed on the horizontal axis. In part, this may be due to the longer time frame for the college variation (10 years) compared to income (7 years).
For the income mobility measure, the average among the commuting zones is basically zero but there is substantial variation in the trajectory over time (you can tell this from the breadth of the histograms or from the size of the standard deviation in Table 2). There is a slight increase in college mobility overall, which leads to a slight increase in the composite mobility. The spread for the college mobility is even higher than for income indicating a large distribution in trajectories between regions.
We tested the relationship between the change in the mobility and the starting mobility and found a significant “reversion to the mean,” which is to say that local mobility has been trending toward a national rate over time. Table 3 has the regression coefficients for the change in income mobility and college mobility as a function of the starting mobility (mobility in 1980 for income and 1984 for college). The negative coefficients indicate that the trend has been for a lower change in mobility for places with a higher initial mobility. Essentially, this means that the places that were good have been tending to get worse and the places that were bad have been tending to get better.
Table 3: Regression Results of Initial Income Mobility on the Change in Mobility
|
Coefficient |
P-Value |
Adjusted R^2 |
Income Mobility 1980-86 |
-0.36 |
<0.001 |
0.24 |
College mobility 1984-93 |
-0.55 |
<0.001 |
0.25 |
While this is statistical evidence that regional variations are becoming less important, these regional differences are still significantly stronger than temporal differences. We have only scratched the surface of these data and we hope to come out with some more analysis in the future. If you are interested in some of our other analysis of these economic mobility studies, you can read about it here: mobility vs. manufacturing employment, the Gatsby Curve, and mobility vs. growth. We hope that this analysis can further the discussion.