Income Inequality and Population Health Ichiro Kawachi, M.D., Ph.D. Professor of Social Epidemiology Harvard School of Public Health Sulzberger Distinguished.

Income Inequality and Population Health

Ichiro Kawachi, M.D., Ph.D.Professor of Social Epidemiology

Harvard School of Public Health

Sulzberger Distinguished Lecture Series

April 5, 2010

Pop QuizWho was highest income earner in USA in 2007-2008?

?Hint:

It wasn’t Bill Gates, Warren Buffett, or Oprah.

John Paulson

Occupation:

Hedge fund manager.

Income in 2007:

$3.7 billion

Income tax rate:

15 percent.

Pop Quiz

How tall is a $1 million stack of $100 bills?

43 inches(≈ 3.5 feet).

How about a $3.7 billion stack of $100 bills?

≈ 13,000 ft.

13,000 ft

…which is how much taller than

Washington Monument (555 ft.)?

13,000 ft =

x 23 taller than Washington Monument (555 ft.)

Source: Professor Dave Abler, Penn State University

(poorest to richest)

Correlation between U.S. state-level Gini and mortality.

Source: Lubotsky & Deaton (2003)

Correlation between MSA-level Gini and mortality

Source: Lubotsky & Deaton (2003)

Four theories about how income inequality can damage population health

1. Absolute income effect

2. Contextual effect of income inequality

3. Relative deprivation

4. Relative rank

Four theories how income inequality can damage population health

Theory Formulation

Absolute income effect hi = f (yi)

f’ > 0, f” < 0

Assumptions

• Association between income and health is causal (Case et al. Am Econ Rev 2002 ; Costello et al. JAMA 2003).

• Association is concave.

Health

Income

Rodgers, Population Studies 1979, 33:13-16.

• Simulation based on New Zealand Census-based cohort.• Each 10% reduction in Gini ≈ 4% reduction in total mortality.• Assuming full causality & no “leaky bucket”.

How large is the absolute income effect?

Theory Formulation

Absolute income hi = f (yi)

f’ > 0, f” < 0

Contextual effect of income inequality

hi = f (yi, Gini)

Four theories how income inequality can damage population health

Contextual effect

Life expectancy

Income

Low Gini society

High Gini society

Contextual effect

Life expectancy

Income

Contextual effect …implies that an individual with the same income, Y, will experience different levels of health depending upon the distribution of incomes in society.

Y

}Low Gini society

High Gini society

What’s the mechanism behind a “contextual” effect of income inequality?

• Exposure to pathologies of poverty - increased homicides, violence & infectious diseases.

• Exposure to affluence - Pecuniary spillover effects.

How do we empirically test for the presence of a contextual effect of income inequality?

• Examine the health status of people living in different communities with varying degrees of income inequality.

• Controlling for individual income, does living in an unequal place = worse health status?

• Accomplished through multi-level modeling.

Meta-analysis of Multi-level Studies of Income Inequality and Health.

(Naoki Kondo, Grace Sembajwe, Ichiro Kawachi, Subramanian, Rob van Dam – BMJ November 2009)

Search of PubMed, ISI Web of Science, and the National Bureau for Economic Research databases.

27 multi-level studies – 9 cohort, 18 cross-sectional. 60 million subjects. Random effects meta-analysis. Risks of total mortality per 0.05 unit increase in Gini.

Pooled Relative Risks from Cohort Studies

Is RR of 1.078 a big deal?

In air pollution studies, each 10 μg/m3

increase in PM2.5 is associated with 4-8% increase in risk of all-cause mortality (Pope et al. 2002).

This association prompted U.S. Environmental Protection Agency to set air pollution standards.

Meta-analysis of Cohort Studies

No. of studies

Pooled RR (95% CI)

Cohort studies overall 9 1.078 (1.059-1.098)

Fixed effect models 3 1.016 (0.987-1.046)

Null findings from fixed effects analyses.

• Cross-national1 and within-country2,3 analyses using first-differences and fixed effects approaches find no association between income inequality and population health.

• These analyses rely upon examining the impact of Δ income inequality on contemporaneous Δ population health,

1. Beckfield, JHSB 2004, 45, 231-48.2. Mellor & Milyo, JHPPL 2001, 26, 487-522.3. Kravdahl, Demographic Res 2008, 18: 205-32.

First differences/Fixed effects models

• Motivated by unobserved heterogeneity in data (controlling for time-invariant unobserved state characteristics).

• But ignores differences in intermediary variables (e.g. educational investment) that may reflect the causal effect of Gini in a prior time period.

• Time periods involved in testing change-on-change effects may not reflect biologically plausible lag times and induction periods.

“Methodological concerns regarding unobserved heterogeneity are, of course, entirely valid. The problem lies in taking recourse to models that rely on sources of identification that exclude mechanisms implied by substantive theory…”

Clarkwest (2008), SSM, 66, p. 1873

Income Inequality and Collective Action

• “Secession” of affluent from rest of society leads to - • Private provision of education, health services, security

services, rubbish collection...

• Pressure to cut taxes on rich → reduced social spending.• Lower quality of public schools & public health services

for everybody else.

Evidence from experimental economics

• Trust game involving 8 people per session.• Each subject given 10 tokens at the beginning of the

session.• A player can either:

(a) put token into private account, resulting in $1 of earnings for each token for that person; or (b) put token into a public account, which is then doubled & split equally between 8 players at the end of the game.

• A selfish player would walk away with $10 at end of the game.• But if players cooperate (e.g. contributed all 10 tokens to public

account), they could double their money at the end of the game.

Anderson, Lisa R., Jennifer M. Mellor and Jeffrey Milyo (2004) American Economics Review Papers and Proceedings, 94(2): 373-376.

Induced Income Inequality

• Induced by 3 different distributions of “show-up” payments given to players:

• Egalitarian: All 8 players receive $7.50.• Skewed (Unequal): 1 player gets $20, 4 get $7,

3 get $4.• Symmetric: 3 players get $10, 2 get $7.50, 3 get

$5.• Moreover, type of inequality either kept private

or public (via award ceremony).

Mean Contributions ($)

Private Inequality

Egalitarian 2.85

Symmetric 3.04

Skewed

(Unequal)

2.76

Source: Anderson, Mellor & Milyo, 2005

Mean Contributions ($)

Private Inequality

Public Inequality

All sessions

Egalitarian 2.85 3.17* 3.01*

Symmetric 3.04 2.22 2.63

Skewed (Unequal)

2.76 2.46 2.61

*P < .05Source: Anderson, Mellor & Milyo, 2005

“The results of this study provide novel support for recent claims that inequality has important “psychosocial” effects that reduce the tendency for cooperation in collective action problems.”

Anderson, Mellor & Milyo 2004

Theory Formulation

Absolute income hi = f (yi)

f’ > 0, f” < 0

Contextual effect hi = f (yi, Gini)

Relative income (relative deprivation)

hi = f (yi - yp)

Four theories how income inequality can damage population health

Theory of Relative Deprivation

“We can roughly say that a person is relatively deprived of X when (i) he does not have X, (ii) he sees some other persons as having X, (iii) he wants X, and (iv) he sees it as feasible that he should have X.”

W.G. Runciman, Relative Deprivation and Social Justice. London: Routledge & Kegan Paul, 1966, p.10.

Empirical tests of relative deprivation

• Cultural consensus - Mixed methods approach to establish community norms of material consumption, e.g. owning designer clothing, RAZR cellphones.

• Cultural consonance - extent to which individuals conform to the cultural model .

(Dressler et al. 1998; Sweet et al. 2010).

Source: E. Sweet AJPH 2011;101:260-4

Yitzhaki index

For person i with income yi, who is part of reference group with N people –

RDi = 1N ∑

j

(yj – yi) V yj > yi

Source: S. Yitzhaki, “Relative deprivation and the Gini coefficient”. Quarterly J Economics 1979;93(2):321-4.

Relative Deprivation, Poor Health Habits, and Mortality.

C.E. Eibner & W.N. Evans J Human Resources (2005), XL:592-619.

• National Health Interview Survey linked to National Death Index.

• RD calculated for 122,504 males aged 21-64.• Reference groups: state of residence, race,

age, education.• Adjusted for individual income.

The Model

Prob (dying during 5 year interval for individual i in reference group r) =

β0 + β1RDir + incomekir + δr + XirΓ + εir∑k=1

26

Independent income effect is captured by term incomekir; δr is a reference group fixed effect (e.g. state fixed effect); and Xir is a vector of dummy variables (age, race, education, martial status) that controls for individual characteristics.

Weighted Linear Probability Models, 5-Year Mortality Equations.

Males aged 21-64 years (104,247 observations, 2.44% died in 5 years)

State State and age

State, age, and race

State, age, race, and education

0.0041

(0.0023)

0.0120

(0.0016)

0.0106

(0.0015)

0.0069

(0.0013)

Coefficients and SEs for reference groups defined by -

Source: Eibner & Evans, 2005, Table 2.

Findings

• A 1.0 standard deviation increase in RD associated with 57% excess mortality in state/age model.

• Smoking.• Obesity.• Mental health services utilization*

Eibner, C. E., R. Sturm, et al. (2004). "Does relative deprivation predict the need for mental health services?" J Ment Health Policy Econ 7(4): 167-75.

Theory Formulation

Absolute income hi = f (yi)

f’ > 0, f” < 0

Income inequality hi = f (yi, Gini)

Relative income (relative deprivation)

hi = f (yi - yp)

Relative rank hi = f (yi, Ri)

Four theories how income inequality can damage population health

Discovery of the “Pecking Order”

Thorleif Schjelderup-Ebbe (1894-1982)

• Experimental induction of subordinate vs. dominant status (status construction theory, Ridgeway et al. Am Sociol Rev 1998).

• N=42 female subjects randomized in laboratory.• Cardiovascular reactivity measured during stress tasks.

606264666870727476

Baseline MeaningInsight

(Round 1)

MeaningInsight

(Round 2)

StressorTask

Recovery

Subordinate group (n=20)Dominant group (n=22)

100

105

110

115

120

125

Baseline MeaningInsight

(Round 1)

MeaningInsight

(Round 2)

StressorTask

Recovery

Systolic blood pressure.

DBP

TimeTime

SBP

Diastolic blood pressure.

Note: Age, race, and BMI were controlled in analyses.

*

* p < .01 for group x time interaction

Blood Pressure as a Function of Status Group

Policy Implications of Relative Rank

• If rank is what matters, this suggests a limited role for social policy in reducing health disparities, because:

a) Rank is zero sum – in order for someone to enjoy the benefit of high rank, someone else has to suffer subordinate rank.

b) Income transfers raise the incomes of the poor, but usually preserve rank.

Policies to lift the bottom

• Invest in early education.• Child care subsidy for working families.• Affordable health care.• Index minimum wage to cost of living…