EPUNet Conference – BCN 06EPUNet Conference – BCN 06

“The causal effect of socioeconomic characteristics in health limitations

across Europe: a longitudinal analysis using the European Community

Household Panel”

   Cristina Hernández Quevedo (DERS, University of York)

Andrew M. Jones (DERS, University of York)Nigel Rice (CHE, University of York)

OBJECTIVES OF THE STUDYOBJECTIVES OF THE STUDY

OBJECTIVES

– Investigate causal effect of SE characteristics in health limitations within and between MS of EU-15

Interested in whether and to what extent, SE characteristics as education, income and job status affect health limitations and how this varies across time and countries included in the ECHP-UDB

– Analyse dynamics of SE gradient in two binary indicators of health limitations across EU-15 by exploiting longitudinal nature of ECHP (8 waves)

LITERATURE REVIEW (I)LITERATURE REVIEW (I)

Several studies on the causal effect of SEC in health– But:

Not been adequately addressed (Ettner, 1996) Poorly understood (Deaton & Paxson, 1998) Degree of confusion due to use of occupational class as

proxy for income and failure of taking into account reverse causality (Benzeval et al., 2000)

Issue of interest for public health policy [Ettner, 1996; Frijters et al, 2003]

Limited scope of most previous literature, that focuses on cross-sectional data

[Frijters et al, 2003; Benzeval & Judge, 2001]

LITERATURE REVIEW (II)LITERATURE REVIEW (II)

Panel data provides additional information on dynamics of individual health and income and its impact on inequalities on these periods (Contoyannis, Jones and Rice, 2004)

Useful information for public health policies, if policymakers are interested in the lifetime history of the individual (Williams & Cookson, 2000)

SAMPLE – ECHPSAMPLE – ECHP

8 waves of data (1994 – 2001)

Adults (16+)

Countries: B, DK, EL, E, F, Irl, I, NL, P (8 waves)

Balanced sample– Only includes individuals from the first wave

who were interviewed in each subsequent wave

DATA – VARIABLES (I)DATA – VARIABLES (I)

HEALTH LIMITATIONS VARIABLE

– PH003A. “Are you hampered in your daily activities by any physical or mental health problem, illness or disability?” [HAMP]

1. “Yes, severely” 2. “Yes, to some extent” 3. “No”– 2 binary measures of health problems:

HAMP1. Indicator of any limitation HAMP2. Indicator of severe limitation

DATA – VARIABLES (II)DATA – VARIABLES (II)

EXPLANATORY VARIABLES– Income measure: disposable household income per equivalent adult– Marital status: married, separated, divorced, widowed, never married– Education: primary, secondary, tertiary– Household Size– Number of children: aged 0 – 4, 5 – 11, 12 – 18– Age groups, men/women: 16 – 25 (men), 26 – 35, 36 – 45, 46 – 55, 56 – 65,

66 – 75, 76 – 85, +86– Job Status: employed, self-employed, unemployed, retired, housework,

inactive– Time dummies

DESCRIPTIVE ANALYSIS (I)DESCRIPTIVE ANALYSIS (I)

.065.159

.7760

.2.4

.6.8

De

nsity

1 2 3PH003A

Germany

.0538.16

.7862

0.2

.4.6

.8D

ens

ity

1 2 3PH003A

Denmark

.0722.159

.7687

0.2

.4.6

.8D

ens

ity

1 2 3PH003A

Netherlands

.0455 .1026

.8519

0.2

.4.6

.8D

ens

ity

1 2 3PH003A

Belgium

.0461.1632

.7907

0.2

.4.6

.8D

ens

ity

1 2 3PH003A

Luxembourg

.095 .1294

.7756

0.2

.4.6

.8D

ens

ity

1 2 3PH003A

France

.0673.1848

.7479

0.2

.4.6

.8D

ens

ity

1 2 3PH003A

UK

.0338.1281

.838

0.2

.4.6

.8D

ens

ity

1 2 3PH003A

Ireland

.0429 .0831

.874

0.2

.4.6

.8D

ens

ity

1 2 3PH003A

Italy

.0737 .0994

.8269

0.2

.4.6

.8D

ens

ity

1 2 3PH003A

Greece

.0567 .1177

.8256

0.2

.4.6

.8D

ens

ity

1 2 3PH003A

Spain

.1031 .1525

.7444

0.2

.4.6

.8D

ens

ity

1 2 3PH003A

Portugal

.0517.1355

.8128

0.2

.4.6

.8D

ens

ity

1 2 3PH003A

Austria

.0762.2055

.7183

0.2

.4.6

.8D

ens

ity

1 2 3PH003A

Finland

DESCRIPTIVE ANALYSIS (II)DESCRIPTIVE ANALYSIS (II)

Distribution of HAMP1 across Member States

05

1015202530

Italy

Belgium

Irelan

d

Greec

e

Spain

Austri

a

Luxe

mbo

urg

Denm

ark

Germ

any

Franc

e

The N

ethe

rland

sUK

Portu

gal

Finlan

d

Countries

Per

cen

tag

e

DESCRIPTIVE ANALYSIS (III)DESCRIPTIVE ANALYSIS (III)

Distribution of HAMP2 across Member States

02468

1012

Irelan

dIta

ly

Belgium

Luxe

mbo

urg

Austri

a

Denm

ark

Spain

Ger

man

yUK

The N

ethe

rland

s

Gre

ece

Finlan

d

Franc

e

Portu

gal

Countries

Per

cen

tag

e

METHODS (I)METHODS (I)

Dynamic latent variable specification for binary choice model

Hence,

* ,1it it it i ith x h

*1, if 0

0, otherwiseit it

it

h h

h

POOLED & RE PROBITPOOLED & RE PROBIT

POOLED PROBIT– It does not take into account that the panel dataset

contains repeated observations– The estimates are consistent

Model is estimated using a misspecified likelihood function

– We allow for robust standard errors

RANDOM EFFECTS– Both components of error term (ηi, εit) are normally

distributed – Both independent of x’s strong exogeneity assumption,

PP more robust but less efficient

REP MODEL (I)REP MODEL (I)

Different approaches to relax assumption

– Mundlak (1978) Relationship as linear regression of mean value of

explanatory variables, averaged over t for a given i

where ξi is iid

– Chamberlain (1984) Relationship as a linear regression of x’s in all waves

where ξi|xi ~ N(0, σ2

η)

,i i ix

, ,1 1 ...i i Ti T ix x

REP MODEL (II)REP MODEL (II) Wooldridge (2005). J. Appl. Econ.

– Approach to deal with correlated individual effects and initial conditions problem in dynamic, nonlinear unobserved RE probit model

– 2 problematic factors: Starting point of survey not the beginning of process Individuals inherit different unobserved & t-invariant

characteristics endogeneity bias in dynamic models with covariance structures not diagonal

– W (2005) models distribution of unobserved effect conditional on initial value and any strictly exogenous explanatory variables

COMPLEMENTARY LOG-LOGCOMPLEMENTARY LOG-LOG

Less used F(.) is the cdf of the extreme value distribution Asymmetric around zero Used when one of the outcomes is rare Probability (p=Pr[h=1|x])

Marginal effect

( ' ) 1 exp( exp( ' ))C x x

/ exp( exp( ' )) exp ( ' )j jp x x p x

ESTIMATION STRATEGY (I)ESTIMATION STRATEGY (I)

Dynamic panel probit and complementary log-log specifications on balanced sample for HAMP1 and HAMP2

Include previous health limitations: capture state dependence and reduce bias due to reverse causality

Specification of binary latent variable

* ,1it it it io i ith x h h

ESTIMATION STRATEGY (II)ESTIMATION STRATEGY (II)

Apply Wooldridge’s (2005) approach to deal with initial conditions problem by including initial value of health limitations hio

To allow for possibility that observed regressors may be correlated with individual effect parameterize individual effect

i io io ix h

ESTIMATION STRATEGY (III)ESTIMATION STRATEGY (III)

Final specification

xit

– Education, household size, number of children by age, age-sex groups

xo

– Log income, job status xit-1

– Martial status, log income, job status

1it it it io io i ith h x h x

AIC, BIC & Reset Test – HAMP1AIC, BIC & Reset Test – HAMP1AIC BIC Reset Test

PPM 1197.16 12332.17 1.94 (.1638)REM 11430.54 11796.33 6.38 (.0115)CLL 12005.84 12363.85 21.72 (.000)PPM 11826.28 12202.81 14.40 (.000)REM 21265.93 21650.82 3.01 (.0826)CLL 22680.03 23056.56 16.34 (.000)PPM 10453.37 10818.8 6.46 (.011)REM 4699.404 5072.779 11.39 (.0007)CLL 10579.77 10945.2 71.99 (.000)PPM 32187.85 32581.67 4.31 (.038)REM 30479.35 30881.92 1.24 (.265)CLL 32456.53 32850.35 132.69 (.000)PPM 10681.61 11043.23 13.70 (.000)REM 10258.49 10627.98 13.26 (.0003)CLL 10816.73 11178.35 89.53 (.000)PPM 28445.79 28863.29 .010 (.0748)REM 26696.04 27122.62 9.88 (.0017)CLL 28690.34 29107.84 136.55 (.000)PPM 26826.2 27226.25 32.13 (.000)REM 25719.52 26128.27 20.57 (.000)CLL 27070 27470.05 127.45 (.000)PPM 31665.4 32073.21 157.97 (.000)REM 30182.55 30599.23 115.65 (.000)CLL 32270.23 32678.04 572.28 (.000)PPM 34644.8 35050.55 34.37 (.000)REM 33055.27 33469.83 45.64 (.000)CLL 35085.77 35491.51 277.91 (.000)

P

Irl

I

EL

E

DK

NL

B

F

AIC, BIC & Reset Test – HAMP2AIC, BIC & Reset Test – HAMP2AIC BIC Reset Test

PPM 4687.831 5045.839 16.20 (.000)REM 4489.575 4855.366 12.14 (.0005)CLL 4776.912 5134.92 57.65 (.000)PPM 11826.28 12202.81 14.40 (.000)REM 11263.23 11648.12 4.14 (.0418)CLL 11953.23 12329.76 68.12 (.000)PPM 4941.878 5298.021 17.09 (.000)REM 4699.404 5072.779 11.39 (.0007)CLL 5049.503 5405.646 70.90 (.000)PPM 19136.92 19530.74 30.77 (.000)REM 18087.17 18489.74 22.78 (.000)CLL 19412.01 19805.83 147.30 (.000)PPM 3810.951 4172.574 5.53 (.0187)REM 3655.647 4025.131 5.40 (.0201)CLL 3867.635 4229.258 34.56 (.000)PPM 13591.67 14009.17 51.11 (.000)REM 12915.62 13342.2 38.31 (.000)CLL 13892.29 14309.79 199.74 (.000)PPM 16351.66 16751.72 17.62 (.000)REM 15828.11 16236.87 13.97 (.000)CLL 16498.67 16898.72 89.96 (.000)PPM 16730.75 17138.56 99.65 (.000)REM 16036.45 16453.12 78.52 (.000)CLL 16996.33 17404.13 257.91 (.000)PPM 22234.27 22640.01 70.42 (.000)REM 21439.23 21853.8 63.97 (.000)CLL 22582.71 22988.46 265.02 (.000)

DK

NL

B

F

Irl

I

EL

E

P

Mg.Eff. PPM – HAMP1Mg.Eff. PPM – HAMP1

DK NL B F Irl I EL E P

hamp1_lag 0.466* .471* .399* .451* .412* 0.365* .394* .258* .506*primary -0.03* -.050* -.026* -.047* -.012 -.003 -.020* -.022* -.010secondary -0.008 -.025* -.009 -.020* -.004 -.005** -.007 -.006 .002ln_inc_lag -0.004 -.015* .004 -.016* -.016* -.00005 -.006* -.011* -.026*selfemploy_lag 0.0001 -.031** -.007 -.006 -.008 .0005 -.004 -.013 .010unemployed_lag 0.046* .033 .036* .040* .066* .006 .036* .042* .053*retired_lag 0.1* .014** .022* .037* .019 .014* .042* .057* .089*housework_lag -0.02 .016** .035* .067* -.010 .005 .027* .048* .056*inactive_lag 0.131* .044* .183* .003 .261* .079* .173* .159* .131*

Mg.Eff. PPM – HAMP2Mg.Eff. PPM – HAMP2

DK NL B F Irl I EL E Phamp2_lag .276* .334* .227* .344* .209* 0.242* .257* .122* .360*primary -.011* -.014* -.011* -.016* -.004** -.002* -.009* -.007* -.001secondary -.003 -.003 .001 -.006** -.002 -.002* -.004** -.001 .00001ln_inc_lag -.004** -.007* .0001 -.005* -.002 -.001** -.004* -0.005* -.011*selfemploy_lag .005 -.007 -.001 .003 -.002 .002 .001 .0004 -.004unemployed_lag .029* .018* .025* .023* .012* .006* .022* .0174* .035*retired_lag .046* -.004 .015* .012* 0.012** .007* .041* 0.03* .053*housework_lag -.004 .005 .017* .043* .002 .006* .032* .022* .029*inactive_lag .043* .014* .043* .065* .076* .032* .154* .086* .096*

CONCLUSIONSCONCLUSIONS

Our contribution– Present a dynamic approach taking account

the 8 waves available of the ECHP – UDB– Focus on health limitations– Job status included in our analysis as

explanatory variables

Provisional conclusions– Probit model adequate for our sample

Specification of model should be refined– Considerable persistence in health limitations