How to think about upward mobility (with eight indicators, two countries, and a thousand regions)
Miles Corak Graduate School of Public and International Affairs University of Ottawa, Ottawa Canada MilesCorak.com @MilesCorak
Presentation to the Conference on Social Mobility and Inequality in Israel Jerusalem June 15th , 2017
Three motivating pictures 1. Shania Twain is one of my favourite musicians
Three motivating pictures 2. Intergenerational mobility varies across countries
50
Italy
United Kingdom
Fraction of earnings inequality in the parent's generation passed on to the children's generation (%)
United States
40
France
Japan Germany
30
New Zealand Sweden Canada
Australia
20 Finland Norway Denmark
10 20
25
30
Higher Inequality (disposable income Gini in 1985)
35
Three motivating pictures 3. Intergenerational mobility varies within the US
The ‘American Dream’ means the same thing to Canadians as to Americans Being middle class
Americans Canadians
Becoming rich Owning your own business Getting married, having kids Enough income to afford a few luxuries Getting a college degree Owning a house Succeed regardless of background Being financially secure Children being better off financially than you Free to say or do what you want Being free to accomplish anything with hard work
0
20
40
60
80
Percent reporting eight or higher on a ten point scale
100
Major messages
1. We need to hold more than one thought in our minds to appreciate the meaning of upward mobility for public policy I I I
economic theory gives us some, but not complete, guidance in deciding upon appropriate metrics the intuitions and instincts of policy makers and advocates are helpful, but also incomplete a need to be agnostic, yet structured in our thinking
Major messages
1. We need to hold more than one thought in our minds to appreciate the meaning of upward mobility for public policy 2. Upward mobility as a “rags to riches� story has resonance I I I
but bottom to top mobility can’t be the whole story mobility as relative movement also implies a need to focus on intergenerational cycles of poverty, and of privilege mobility as absolute movement also implies a focus on income mobility
Major messages
1. We need to hold more than one thought in our minds to appreciate the meaning of upward mobility for public policy 2. Upward mobility as a “rags to riches� story has resonance 3. Promoting more upward mobility I I
is about more than just efficient cream-skimming of the most innately talented children of the least advantaged it involves raising the chances of escaping low income across the entire population of the relatively disadvantaged
Ms. Twain’s success embodies a whole set of important issues relevant to thinking about social mobility
1. The important counterfactual is ultimately not measurable I I
the development of capacities and opportunities that permit a child to become all that she or he can be at a broad level it is natural to think of this in three different ways I I I
relative to the previous generation relative to counterparts in the current generation relative to those living elsewhere
Ms. Twain’s success embodies a whole set of important issues relevant to thinking about social mobility
1. The important counterfactual is ultimately not measurable 2. Upward mobility is about directional movement I I
relative movement in ranks absolute movement in incomes
The Harvard economist Benjamin Friedman stresses a trade-off between relative and absolute mobility Most people . . . favor equal opportunity and fairness in the abstract sense that they would certainly support furthering these aims if doing so implied no risk, and no other cost, to themselves. But in practice, taking more steps to move a society toward greater fairness or more equal opportunity typically does impose risks, as well as costs, on at least some people. The importance of economic growth for this purpose is that rising incomes make people more willing to accept these risks and costs in the interest of what they take to be a better society for themselves as well as others. Benjamin Friedman (2006), The Moral Consequences of Economic Growth, page 95.
Three measures of intergenerational mobility we care about
1. incomes I
average incomes of children from different communities vary for at least three statistical reasons related to differences in absolute mobility, relative mobility, and average incomes of their parents lnYi,t = αj + βj lnYi,t−1 + εi,j βj Y¯t = e αj Y¯t−1
I
measurement and estimation must address some concerns to avoid bias
Canadian tax data for those born between 1963 and 1970 Province / Territory
Administrative Data Weighted Unweighted
1986 Census Total
Ratio weighted to Census
Newfoundland and Labrador Prince Edward Island Nova Scotia New Brunswick Quebec Ontario Manitoba Saskatchewan Alberta British Columbia Yukon Northwest Territories, Nunavut
84,050 16,750 112,900 91,500 796,650 1,057,550 122,150 122,500 284,550 304,250 2,950 7,150
59,000 12,400 79,350 67,600 531,000 796,800 91,650 81,600 184,250 206,200 1,700 3,600
87,000 18,100 121,700 98,600 842,950 1,191,750 139,500 132,700 319,550 344,850 2,950 8,200
0.97 0.93 0.93 0.93 0.95 0.89 0.88 0.92 0.89 0.88 1.00 0.87
Canada
3,002,950
2,115,150
3,307,900
0.91
Note: Numbers are rounded to the nearest 50.
Table 1: Weighted and unweighted sample sizes compared to the Census
Three measures of intergenerational mobility we care about
1. incomes 2. position I
the average rank in the national income distribution of children from different communities also depends upon absolute rank mobility and on relative rank mobility yi,t = aj + bj yi,t−1 + i,j
I
measurement issues raise even more concerns to avoid bias
Rank mobility at two points in the life cycle
Province / Territory
At 35 to 48 years of age Absolute Relative Expected (aj ) (bj ) Rank
At 31 and 32 years of age Absolute Relative Expected (aj ) (bj ) Rank
Newfoundland and Labrador Prince Edward Island Nova Scotia New Brunswick Quebec Ontario Manitoba Saskatchewan Alberta British Columbia Yukon Northwest Territories, Nunavut
35.3 35.1 32.6 31.6 36.7 41.0 31.2 41.5 44.4 39.6 36.3 34.1
0.273 0.245 0.251 0.280 0.249 0.225 0.325 0.226 0.206 0.184 0.248 0.281
40.8 40.0 37.6 37.2 41.7 45.5 37.7 46.0 48.5 43.3 41.3 39.7
33.2 35.3 32.0 31.1 36.9 43.4 29.9 37.7 41.1 39.9 38.5 31.4
0.277 0.239 0.249 0.286 0.240 0.215 0.320 0.236 0.203 0.185 0.176 0.283
38.7 40.1 37.0 36.8 41.7 47.7 36.3 42.4 45.2 43.6 42.0 37.1
Canada
38.3
0.242
43.1
38.4
0.240
43.2
Source: Least squares estimates using Statistics Canada, Intergenerational Income Data as described in text.
Table 2: Absolute and relative intergenerational rank mobility
Three measures of intergenerational mobility we care about 1. incomes 2. position 3. upward mobility, avoiding poverty I
I
moving up the income distribution may reflect a non linear process, and an interaction with the chances of being stuck in the bottom, and of falling out of the top transition probabilities, and particularly three specific quintile transition probabilities P1,5 = Pr{Yt ∈ top|Yt−1 ∈ bottom} P1,1 = Pr{Yt ∈ bottom|Yt−1 ∈ bottom} P5,5 = Pr{Yt ∈ top|Yt−1 ∈ top}
I
measurement and estimation must address non-classical errors
0.35
0.35
0.30
0.30
0.25
0.25
Rags to Riches
Rags to Riches
As an illustration consider the Chetty-Hendren data
0.20 0.15
0.20 0.15
0.10
0.10
0.05
0.05
0.1
0.2
0.3
0.4
0.5
(a) Cycle of Privilege
0.6
0.1
0.2
0.3
0.4
0.5
0.6
(b) Cycle of Poverty
Figure 4: Rags to Riches movement in over 700 Commuting Zones
Similar patterns in the Canadian data
0.20
Rags to Riches
Rags to Riches
0.20
0.15
0.10
0.15
0.10
0.05
0.05
0.00
0.00
0.1
0.2
0.3
0.4
0.5
(a) Cycle of Privilege
0.6
0.1
0.2
0.3
0.4
0.5
(b) Cycle of Poverty
Figure 5: Rags to riches movement in 266 Census Divisions
0.6
Probability of staying in the bottom quintile for men and women having bottom quintile parents 0.40 or higher 0.35 to 0.40 0.30 to 0.35
0.25 to 0.30 0.20 to 0.25 less than 0.20
Figure 6: The intergenerational cycle of bottom income
Probability of bottom quintile income for men
0.6
Chances of an intergenerational cycle of low income are higher for men
0.5
0.4
0.3
0.2 Chances of an intergenerational cycle of low income are higher for women
0.1 0.1
0.2
0.3
0.4
0.5
0.6
Probability of bottom quintile income for women
Figure 7: Intergenerational Cycles of low income by gender, as measured by bottom quintile probabilities
Probability of moving to the top quintile for men and women having bottom quintile parents 0.20 or more 0.15 to 0.20 0.10 to 0.15
0.05 to 0.10 0.025 to 0.05 less than 0.025
Figure 8: Rags to riches mobility
Clustering communities together by unsupervised machine learning Eight parameters of three alternative measures I I I
α, β, Y¯t−1 a, b P1,1 , P1,5 , P5,5
Bottom up agglomeration to cluster similar Census Divisions I I
single, complete, average linkage of clusters? pruning the tree, and choosing the number of clusters?
Complete Linkage
Average Linkage
Euclidean distance
Figure 9: Hierarchical Agglomorative Clustering yields Dendrograms sensitive to modelling choices
0
5
10
15
Complete linkage with Euclidean−Based Distance
2 clusters
Figure 10: Preferred Dendrogram illustrating a two-cluster grouping
Figure 11: The landscape of us and them
Census Division Clusters (determined by Hierarchical Agglomorative Clustering) 1 2 3 4 unclassified
Figure 12: The contours of mobility in a multi-cluster country
Figure 13: The Canada-United States border would not be chosen by a machine learning algorithm minimizing within-cluster variance of six indicators of intergenerational mobility
Figure 14: A four cluster mapping shows that some regions lie largely on either side of the Canada-United States border but that others are not confined to one country
Major messages
1. We need to hold more than one thought in our minds to appreciate the meaning of upward mobility for public policy 2. Upward mobility as a “rags to riches� story has resonance 3. Promoting more upward mobility in the United States I I
is about more than just efficient cream-skimming of the most innately talented children of the least advantaged it involves raising the chances of escaping low income across the entire population of the relatively disadvantaged