Mechanism Design and Marginal Distributions
- 1. Mechanism Design and Marginal Distributions Mechanism Design for Social Good 2018 workshop Robert Manduca - Harvard Sociology June 22, 2018
- 2. Marginal distributions and allocation processes • Social outcomes are determined by two types of input: • Marginal distributions – what is the set of possible outcomes available • Allocation processes – who gets assigned to which positions • Sociologists have historically been most interested in allocation processes: • Who lives in which neighborhoods? • Who gets hired for which jobs? • What determines where a child goes to school? 1
- 3. Today’s big social challenges will not be solved solely by better allocation of existing positions • Many of our most pressing social problems are not fully solvable through better allocation of existing social positions alone. In this talk I will give two examples from my research: • Upward income mobility • Racial economic equality • Then I will describe ways mechanism design might be harnessed to create a better marginal distribution of outcomes, rather than trying to more optimally allocate the inadequate set we have today 2
- 4. Allocation and distribution 1: declining upward income mobility • Upward mobility is central to American identity • “A better life for your children” • Also key for tolerance, fairness, democracy... (Friedman 2005; Mullainathan and Shafir 2013) • But absolute upward mobility rates have been falling for 50 years • How do we reverse this decline? q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q 0% 25% 50% 75% 100% 1940 1950 1960 1970 1980 Child's birth cohort Pct.ofchildrenearningmorethantheirparents Absolute upward mobility by birth cohort 3
- 5. Fairer allocation processes would not restore mass upward mo- bility • “Equality of opportunity” is typically defined as children’s economic positions have zero correlation with their parents’ • But perfect equality of opportunity would not increase aggregate upward mobility • Observed mean = 46.9% • Mean with perfect relative mobility = 46.4% 0% 25% 50% 75% 100% 0 25 50 75 100 Parent income percentile Pct.ofkidsearningmorethanparents Observed Perfect relative mobility Absolute upward mobility by parent income percentile, 1980 cohort, observed and counterfactual with perfect relative mobility 4
- 6. Widespread upward mobility requires a more equitable income distribution • If fair allocation is not enough, how do we restore upward mobility? • Returning to the income distribution of 1970 would reverse most of the mobility decline • Observed mean = 46.9% • Mean with 1970 income distribution = 78.3% 0% 25% 50% 75% 100% 0 25 50 75 100 Parent income percentile Pct.ofkidsearningmorethanparents Observed 1970 income distribution Absolute upward mobility by parent income percentile, 1980 cohort, observed and counterfactual with 1970 level of inequality 5
- 7. Fixing upward mobility requires changing the set of outcomes that are available, not better allocating the outcomes we have now 5
- 8. Allocation and distribution 2: Black-white family income dispar- ities • The family income gap between blacks and whites has not changed for the last 50 years • Most explanations for the lack of racial progress emphasize continued (and well documented) racial stratification: • Processes that sort whites into better jobs, better schools, etc. 0% 25% 50% 75% 100% 1970 1980 1990 2000 2010 Ratioofmedianblacktomedianwhitefamilyincome A: Ratio of median black to median white family income over time 6
- 9. US racial income stratification has decreased substantially since the 1960s • In 1968, the median black American was at the 25th percentile of US family income • In 2016 they were at the 35th percentile • The black-white gap rank has shrunk by 28% 0 10 20 30 40 50 1970 1980 1990 2000 2010 Nationalincomepercentile B: Median black income rank over time, family income 7
- 10. Reductions in racial stratification were negated by rising economy- wide income inequality • As the rank gap was closing, overall inequality was rising • 1968 income shares: • Richest 1%: 12.0% • Poorest 50%: 19.2% • 2014 income shares: • Richest 1%: 19.0% • Poorest 50%: 10.3% • These shifts reduced the payoff for African Americans climbing the income ladder 0% 25% 50% 75% 100% 1970 1980 1990 2000 2010 Incomerelativetonationalmean C: Ratio of income at 35th percentile (median for blacks in 2016) to national mean income over time, family income 8
- 11. Allocation processes have become less racially stratified, but that improvement was entirely undone by changes to the marginal distribution 8
- 12. Many social problems require marginal shifts • In both examples, improvements to allocation processes could not overcome worsening marginal distributions • Similar dynamics exist in many important issue areas: • Education • Health care • Housing • Mechanism design can help improve marginal distributions 9
- 13. Mechanism Design 4 Better Marginals • Our current marginal distributions result in part from poor social decision making. Across many domains, current policy differs markedly from majority opinion: • Support for single player health care: 63% • Support for marijuana legalization: 60% • Support for limiting political spending by individuals: 77% • Aggregating from individual preferences to social choices is hard, and leaves room for manipulation by well-organized or wealthy interests • Mechanism design can help! 10
- 14. MD4BM 1: Participatory budgeting 11
- 15. MD4BM 2: Voting systems 12
- 16. MD4BM 3: Easier group decision-making Getting large groups of people to reach consensus on a course of action is hard, and a major barrier to political organizing and cooperative ownership. Internet platforms with good mechanism design can help overcome this 13
- 17. MD4BM 4: Determining social preferences 14
- 18. What these approaches have in common • Focus is on making better social decisions so that we can get the set of options that we want, instead of trying to choose among the bad options that we have • This involves both: • Technical challenges – determining the optimal social choice given disparate preferences • Organizational challenges – overcoming coordination problems and distortionary power centers • Mechanism design can contribute to overcoming both sets of challenges 15
- 20. How marginal distributions shape mobility
- 21. How marginal distributions shape mobility