Healthy, Wealthy, and Wise? Healthy, Wealthy, and Wise? The Causal - - PowerPoint PPT Presentation

SLIDE 1

Healthy, Wealthy, and Wise? Healthy, Wealthy, and Wise?

The Causal Pathways between The Causal Pathways between Health and Socioeconomic Status Health and Socioeconomic Status Daniel McFadden Daniel McFadden

University of California, Berkeley, and NBER University of California, Berkeley, and NBER June 2003 June 2003 Research sponsored by National Institute on Aging Research sponsored by National Institute on Aging Behavioral and Social Research Behavioral and Social Research Richard Richard Suzman Suzman, Director , Director

SLIDE 2

The links between health, wealth The links between health, wealth

Higher socio Higher socio-

• economic status (SES) is

economic status (SES) is associated with better health and associated with better health and longer life longer life The association is found in different The association is found in different eras, places, genders, and ages eras, places, genders, and ages

SLIDE 3

The association holds for: The association holds for:

– – A variety of health variables (most A variety of health variables (most illnesses, mortality, self illnesses, mortality, self-

• rated health

rated health status, psychological well status, psychological well-

• being, and

being, and biomarkers such as biomarkers such as allostatic allostatic load) load) – – Alternative measures of SES (wealth, Alternative measures of SES (wealth, education, occupation, income, level education, occupation, income, level

• f social integration).
• f social integration).

SLIDE 4

Prevalence Relative Risk Prevalence Relative Risk

Low Low vs vs High SES, AHEAD 1993 High SES, AHEAD 1993 2.69* 2.69* 3.09* 3.09* P/F SRHS P/F SRHS 1.63* 1.63* 1.31* 1.31* Arthritis Arthritis 4.27* 4.27* 4.67* 4.67* Smoker Smoker 1.58* 1.58* 1.21* 1.21* HBP HBP 5.33* 5.33* 2.78* 2.78* Depression Depression 1.64 1.64 5.25* 5.25* Diabetes Diabetes 1.78 1.78 0.86 0.86 Psychiatric Psychiatric 2.38* 2.38* 3.18* 3.18* Lung Lung 6.97* 6.97* 1.82* 1.82*

• Cog. Imp.
• Cog. Imp.

2.46* 2.46* 1.44 1.44 Stroke Stroke 2.31 2.31 1.68 1.68 Hip Hip Fract Fract. . 1.35* 1.35* 2.15* 2.15* Heart Heart 1.29 1.29 1.15 1.15 Incontinent Incontinent 1.03 1.03 0.52 0.52 Cancer Cancer M M F F Condition Condition M M F F Condition Condition

SLIDE 5

The association must be rooted in The association must be rooted in fundamental heterogeneities of nature fundamental heterogeneities of nature and nurture in human populations and nurture in human populations It is It is not not explained solely by: explained solely by:

Poverty Poverty Social discrimination Social discrimination Availability of medical technology Availability of medical technology Specific systems for delivery and Specific systems for delivery and financing of health care financing of health care

SLIDE 6

Sources: Sources:

• P. Adams, M.
• P. Adams, M. Hurd

Hurd, D. McFadden, A. Merrill, T. , D. McFadden, A. Merrill, T. Ribeiro Ribeiro (2003) "Healthy, Wealthy, and Wise? Tests for Direct (2003) "Healthy, Wealthy, and Wise? Tests for Direct Causal Paths between Health and Socioeconomic Causal Paths between Health and Socioeconomic Status," Status," Journal of Econometrics, 112, 3 Journal of Econometrics, 112, 3-

• 56

56. .

• ---------- (2003) “Addendum”, forthcoming in D.Wise

(2003) “Addendum”, forthcoming in D.Wise (ed) (ed) Research in the Economics of Aging Research in the Economics of Aging, NBER. , NBER. J.

• J. Adda

Adda, T. , T. Chandola Chandola, M. Marmot (2003) “Socio , M. Marmot (2003) “Socio-

• economic Status and Health: Causality and Pathways,”

economic Status and Health: Causality and Pathways,” Journal of Econometrics, 112, 57 Journal of Econometrics, 112, 57-

• 63.

63.

• A. Case, A. Deaton (2003) "Broken Down by Work and
• A. Case, A. Deaton (2003) "Broken Down by Work and

Sex: How Our Health Declines,“ Carefree Conference on Sex: How Our Health Declines,“ Carefree Conference on the Economics of Aging, Princeton/NBER working the Economics of Aging, Princeton/NBER working paper. paper.

• J. Smith (2003) “Consequences and Predictors of New
• J. Smith (2003) “Consequences and Predictors of New

Health Events,” Carefree Conference on the Economics Health Events,” Carefree Conference on the Economics

• f Aging, RAND/NBER working paper.
• f Aging, RAND/NBER working paper.

SLIDE 7

GENETICS TASTES/BEHAVIOR HEALTH HISTORY SES HISTORY HEALTH EVENTS SES EVENTS

SLIDE 8

HEALTH HISTORY ONSET OF NEW HEALTH CONDITION DIAGNOSIS TREATMENT REGIMEN DEATH CONTROL CURE BEHAVIOR SES HISTORY MEDICAL, DRUG, CAREGIVER COSTS, LOST INCOME NEW SES

SLIDE 9

Policy Implications Policy Implications

Research interest in the health Research interest in the health-

• wealth nexus is

wealth nexus is fueled by the potential policy implications of fueled by the potential policy implications of alternative causal paths. alternative causal paths. If the poor are less healthy because of specific If the poor are less healthy because of specific failures in the health care delivery system, failures in the health care delivery system, targeted changes in that system could have a targeted changes in that system could have a significant net social benefit. significant net social benefit. If the poor are less healthy because of behavior If the poor are less healthy because of behavior that increases risk, delays diagnosis, or reduces that increases risk, delays diagnosis, or reduces the effectiveness of treatment, then policies that the effectiveness of treatment, then policies that increase information and encourage protective increase information and encourage protective behavior, or increase preventative interventions, behavior, or increase preventative interventions, may be beneficial. may be beneficial. If the less healthy are poor due to work disability If the less healthy are poor due to work disability and medical costs, then health and disability and medical costs, then health and disability insurance need to be strengthened. insurance need to be strengthened.

SLIDE 10

Ht-1 Ht St-1 St

SLIDE 11

Ht-1 Ht St-1 St

SLIDE 12

Ht-1 Ht St-1 St Ct-1 Ct

SLIDE 13

Ht-1 Ht St-1 St Ct-1 Ct

SLIDE 14

A Markov model of Health/SES dynamics A Markov model of Health/SES dynamics The evolution of socioeconomic states S The evolution of socioeconomic states St

t

and health states H and health states Ht

t is modeled as a first

is modeled as a first-

• rder Markov process,
• rder Markov process,

f f(H (Ht

t,S

,St

t |H

|Ht

t-

• 1

1,S

,St

t-

• 1

1) =

) = f fH

H(H

(Ht

t|H

|Ht

t-

• 1

1,S

,St

t-

• 1

1)

)f fS

S(S

(St

t |H

|Ht

t,H

,Ht

t-

• 1

1,S

,St

t-

• 1

1),

),

the conditional distribution of H the conditional distribution of Ht

t,S

,St

t given

given H Ht

t-

• 1

1,S

,St

t-

• 1

1.

. The model The model f f(H (Ht

t,S

,St

t |H

|Ht

t-

• 1

1,S

,St

t-

• 1

1) is

) is valid valid for a for a history H history Ht

t-

• 1

1,S

,St

t-

• 1

1 if it is the true conditional

if it is the true conditional distribution of H distribution of Ht

t,S

,St

t given this history.

given this history.

SLIDE 15

f f is a is a structural structural or

• r causal

causal model for H model for Ht

t,S

,St

t

relative relative to a family of histories if it has the to a family of histories if it has the invariance invariance property that it is valid for each property that it is valid for each history in the family. history in the family. Operationally, invariance means that Operationally, invariance means that within specified history and treatment within specified history and treatment domains domains, , f f has the has the transferability transferability property property that it is valid in different populations that it is valid in different populations where the marginal distribution of H where the marginal distribution of Ht

t-

• 1

1,S

,St

t-

• 1

1

changes, and the changes, and the predictability predictability or

• r

invariance under treatments invariance under treatments property that property that it remains valid following policy it remains valid following policy interventions that alter the marginal interventions that alter the marginal distribution of H distribution of Ht

t-

• 1

1,S

,St

t-

• 1

1 .

.

SLIDE 16

Restricted history and treatment domains Restricted history and treatment domains identify identify families families of structural models

• f structural models.

. S is S is conditionally non conditionally non-

• causal

causal for H if for H if f f(H (Ht

t|H

|Ht

t-

• 1

1) is a valid model; i.e., given H

) is a valid model; i.e., given Ht

t-

• 1

1,

, knowledge of S knowledge of St

t-

• 1

1 is

is not needed not needed to to achieve the invariance properties of a achieve the invariance properties of a causal model. [Granger non causal model. [Granger non-

• causality]

causality] Conversely, if Conversely, if f f(H (Ht

t|H

|Ht

t-

• 1

1,S

,St

t-

• 1

1)

) … … f f(H (Ht

t|H

|Ht

t-

• 1

1), then

), then knowing S knowing St

t-

• 1

1 contributes to the

contributes to the predictability predictability of H

• f Ht
• t. A direct causal link is

. A direct causal link is possible, but not proved unless common possible, but not proved unless common factors can be ruled out. factors can be ruled out.

SLIDE 17

We array H components in a We array H components in a Wold Wold causal causal chain based on the etiology of health chain based on the etiology of health

• events. Either one or both conditional
• events. Either one or both conditional

non non-

• causality of S for H and conditional

causality of S for H and conditional non non-

• causality of H given S may hold. If

causality of H given S may hold. If either holds, then H and S can be arrayed either holds, then H and S can be arrayed in a (block) causal chain. If both hold, H in a (block) causal chain. If both hold, H and S are conditionally independent. and S are conditionally independent. However, if critical invariance tests for a However, if critical invariance tests for a valid model fail, then non valid model fail, then non-

• causality tests

causality tests are inconclusive. are inconclusive.

SLIDE 18

Data: The AHEAD Panel Data: The AHEAD Panel

7447 elderly Americans, aged 70+ in 7447 elderly Americans, aged 70+ in 1993, including spouses, followed 1993, including spouses, followed through waves in 1995 and 1998. (A 2001 through waves in 1995 and 1998. (A 2001 wave is now released.) wave is now released.) This population is retired, so that health This population is retired, so that health problems have little impact on earnings. problems have little impact on earnings. Relatively homogeneous, comprehensive Relatively homogeneous, comprehensive health care at limited out health care at limited out-

• of
• f-
• pocket cost

pocket cost to the individual is provided by Medicare. to the individual is provided by Medicare.

SLIDE 19

Tests for Invariance and Non Tests for Invariance and Non-

• Causality

Causality

SES to health incidence, SES to health incidence, significance levels significance levels

0.990 0.990 0.534 0.534 0.393 0.393 0.007 0.007 HBP HBP 0.025 0.025 0.110 0.110 0.234 0.234 0.189 0.189 Diabetes Diabetes 0.010 0.010 0.343 0.343 0.689 0.689 0.552 0.552 Lung Lung 0.364 0.364 0.652 0.652 0.376 0.376 0.221 0.221 Mortality Mortality 0.059 0.059 0.657 0.657 0.204 0.204 0.384 0.384 Stroke Stroke 0.243 0.243 0.398 0.398 0.690 0.690 0.085 0.085 Heart Heart 0.166 0.166 0.313 0.313 0.129 0.129 0.857 0.857 Cancer (no Cancer (no prev prev.) .) M M F F M M F F Non Non-

• Causality

Causality Test Test Invariance Invariance Test Test Condition Condition

SLIDE 20

Tests for Invariance and Non Tests for Invariance and Non-

• Causality

Causality

SES to health incidence, SES to health incidence, significance levels significance levels

0.020 0.020 0.001 0.001 0.934 0.934 0.376 0.376 P/F SRHS P/F SRHS 0.065 0.065 0.011 0.011 0.944 0.944 0.211 0.211 Depression Depression 0.065 0.065 0.004 0.004 0.295 0.295 0.127 0.127 Psychiatric Psychiatric 0.026 0.026 0.002 0.002 0.288 0.288 0.005 0.005 Cognitive Imp. Cognitive Imp. 0.430 0.430 0.159 0.159 0.126 0.126 0.491 0.491 Hip Fracture Hip Fracture 0.463 0.463 0.163 0.163 0.351 0.351 0.781 0.781 Incontinent Incontinent 0.395 0.395 0.085 0.085 0.071 0.071 0.046 0.046 Arthritis Arthritis M M F F M M F F Non Non-

• Causality

Causality Test Test Invariance Invariance Test Test Condition Condition

SLIDE 21

Incidence Relative Risk Incidence Relative Risk

Low Low vs vs High SES, AHEAD 1993 High SES, AHEAD 1993-

• 98

98 1.55* 1.55* 1.48* 1.48* P/F SRHS P/F SRHS 1.35 1.35 1.12 1.12 HBP HBP 3.50* 3.50* 2.66* 2.66* Depression Depression 0.67 0.67 1.44 1.44 Diabetes Diabetes 6.29* 6.29* 2.54* 2.54* Psychiatric Psychiatric 1.98 1.98 3.06* 3.06* Lung Lung 1.69* 1.69* 1.28 1.28

• Cog. Imp.
• Cog. Imp.

0.86 0.86 1.45 1.45 Mortality Mortality 4.77* 4.77* 3.06* 3.06* Hip Fracture Hip Fracture 1.07 1.07 1.37 1.37 Stroke Stroke 0.90 0.90 1.18 1.18 Incontinent Incontinent 0.98 0.98 1.03 1.03 Heart Heart 1.84* 1.84* 0.97 0.97 Arthritis Arthritis 1.37 1.37 0.76 0.76 Cancer Cancer M M F F Condition Condition M M F F Condition Condition

SLIDE 22

Test Results Test Results -

• 1

1

Cross Cross-

• wave invariance holds for most

wave invariance holds for most conditions conditions Conditional independence in the assumed Conditional independence in the assumed causal chain is supported, with important causal chain is supported, with important exceptions: mortality, ADL, IADL, exceptions: mortality, ADL, IADL, cognitive impairment, accidents cognitive impairment, accidents Non Non-

• causality is

causality is accepted accepted for most acute for most acute conditions and mortality conditions and mortality Either common behavioral factors or Either common behavioral factors or direct causality lead to rejection of non direct causality lead to rejection of non-

• causality for mental conditions and SRHS

causality for mental conditions and SRHS

SLIDE 23

Test Results Test Results -

• 2

2

The absence of an SES gradient for The absence of an SES gradient for incidence in the 70+ population gives no incidence in the 70+ population gives no evidence on the issue in working evidence on the issue in working-

• age

age populations. populations. Case and Deaton find interactions Case and Deaton find interactions through the working years in which through the working years in which increased exposure to manual work is increased exposure to manual work is associated with more rapidly declining associated with more rapidly declining health. health.

SLIDE 24

Links from Health to Wealth Links from Health to Wealth

Tests for direct causal links from health Tests for direct causal links from health to wealth innovations are inconclusive to wealth innovations are inconclusive due to invariance failures arising from due to invariance failures arising from wealth measurement problems in wave 1. wealth measurement problems in wave 1. Granger non Granger non-

• causality tests fail for liquid

causality tests fail for liquid wealth, and in some cases also for non wealth, and in some cases also for non-

• liquid wealth, for intact couples and

liquid wealth, for intact couples and singles. singles.

SLIDE 25

AHEAD (US), Whitehall (GB), ULF (Sweden) AHEAD (US), Whitehall (GB), ULF (Sweden) SES Non SES Non-

• Causality Test

Causality Test Source: Source: J.

• J. Adda

Adda, T. , T. Chandola Chandola, M. Marmot (2003) , M. Marmot (2003) 0.22 0.22 0.11 0.11 0.12 0.12 0.47 0.47 0.30 0.30 M M -

• ULF

ULF 0.00 0.00 0.42 0.42 0.06 0.06

• 0.00

0.00 M M -

• Whitehall

Whitehall 0.02 0.02 0.99 0.99 0.03 0.03 0.36 0.36 0.04 0.04 M M -

• AHEAD

AHEAD 0.00 0.00 0.00 0.00 0.94 0.94 0.04 0.04 0.18 0.18 F F -

• ULF

ULF 0.00 0.00 0.42 0.42 0.10 0.10

• 0.18

0.18 F F -

• Whitehall

Whitehall 0.00 0.00 0.53 0.53 0.11 0.11 0.65 0.65 0.73 0.73 F F -

• AHEAD

AHEAD P/F P/F SRHS SRHS HBP HBP Diabet Diabet Mort Mort Heart Heart -

• no

no prev prev No causality No causality

• sig. level
• sig. level

SLIDE 26

AHEAD (US), Whitehall (GB), ULF (Sweden) AHEAD (US), Whitehall (GB), ULF (Sweden) Relative Risk, Low Relative Risk, Low vs vs High SES High SES Source: Source: J.

• J. Adda

Adda, T. , T. Chandola Chandola, M. Marmot (2003) , M. Marmot (2003) 5.0 5.0 3.5 3.5 3.5 3.5 4.1 4.1 3.5 3.5 M M -

• ULF

ULF 2.73 2.73 1.13 1.13 4.24 4.24

• 1.90

1.90 M M -

• Whitehall

Whitehall 1.55 1.55 1.35 1.35 0.67 0.67 0.86 0.86 0.77 0.77 M M -

• AHEAD

AHEAD 5.1 5.1 3.8 3.8 4.7 4.7 7.5 7.5 5.8 5.8 F F -

• ULF

ULF 4.13 4.13 1.18 1.18 4.05 4.05

• 1.67

1.67 F F -

• Whitehall

Whitehall 1.48 1.48 1.12 1.12 1.44 1.44 1.45 1.45 1.22 1.22 F F -

• AHEAD

AHEAD P/F P/F SRHS SRHS HBP HBP Diabet Diabet Mort Mort Heart Heart -

• no

no prev prev No causality No causality

• sig. level
• sig. level

SLIDE 27

Net Cumulative OOP Medical Costs Net Cumulative OOP Medical Costs

HRS, incidence in wave 2 HRS, incidence in wave 2

$18,492 $18,492 $307 $307 No insurance No insurance $18,035 $18,035 $23 $23 All (age 51 All (age 51-

• 61)

61) Mild Condition Mild Condition $91,195 $91,195 $2,080 $2,080 No insurance No insurance $40,201 $40,201 $3,496 $3,496 All (age 51 All (age 51-

• 61)

61) Severe Condition Severe Condition 95 95th

th

Percentile Percentile Median Median Source: Source:

• J. Smith (2003)
• J. Smith (2003)

SLIDE 28

Cumulative Effect Cumulative Effect

• f New Health Events
• f New Health Events

$6,532 $6,532 $10,376 $10,376 AHEAD Sample AHEAD Sample Income Loss + Income Loss + Medical Cost Medical Cost $10,792 $10,792 $44,164 $44,164 HRS Sample HRS Sample Income Loss + Income Loss + Medical Cost Medical Cost Minor Health Minor Health Event, Wave 2 Event, Wave 2 Major Health Major Health Event, Wave 2 Event, Wave 2 Source: Source:

• J. Smith (2003)
• J. Smith (2003)

SLIDE 29

Survival Probabilities from Age 70 Survival Probabilities from Age 70

Female and Male, Baseline and No Diabetes Female and Male, Baseline and No Diabetes

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 75 80 85 90 95 F Base F No Diab M Base M No Diab

SLIDE 30

Conclusions Conclusions -

• 1

1

In the U.S. retired, Medicare In the U.S. retired, Medicare-

• eligible

eligible population, the evidence is against strong population, the evidence is against strong direct causal links from SES to incidence direct causal links from SES to incidence

• f most new health conditions, or to
• f most new health conditions, or to

mortality. mortality. Self Self-

• reported health status and mental

reported health status and mental conditions show an SES gradient, with conditions show an SES gradient, with the cross the cross-

• country evidence favoring

country evidence favoring individual behavior as the source, rather individual behavior as the source, rather than deficiencies in delivery of mental than deficiencies in delivery of mental health services. health services.

SLIDE 31

Conclusions Conclusions -

• 2

2

There is inconclusive evidence for direct There is inconclusive evidence for direct causal links from health to wealth causal links from health to wealth changes within the AHEAD panel, but changes within the AHEAD panel, but

• ccupational exposure, income loss and
• ccupational exposure, income loss and

medical costs suggest strong links in medical costs suggest strong links in working working-

• age populations.

age populations. Even though relative risks are mostly Even though relative risks are mostly near one, SES gradients operating from near one, SES gradients operating from age 70 to the end of life have an age 70 to the end of life have an economically significant cumulative effect economically significant cumulative effect

• n incidence and mortality.
• n incidence and mortality.