EPUNet 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 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) 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) 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 – 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) 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) 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 (II)

DESCRIPTIVE ANALYSIS (III)

METHODS (I) Dynamic latent variable specification for binary choice model Hence,

POOLED & 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 (η i, ε it ) are normally distributed –Both independent of x’s  strong exogeneity assumption, PP more robust but less efficient

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 ξ i is iid – Chamberlain (1984) Relationship as a linear regression of x’s in all waves where ξ i |x i ~ N(0, σ 2 η )

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-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

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

ESTIMATION STRATEGY (II) Apply Wooldridge’s (2005) approach to deal with initial conditions problem by including initial value of health limitations h io To allow for possibility that observed regressors may be correlated with individual effect  parameterize individual effect

ESTIMATION STRATEGY (III) Final specification x it –Education, household size, number of children by age, age-sex groups x o –Log income, job status x it-1 –Martial status, log income, job status

AIC, BIC & Reset Test – HAMP1

AIC, BIC & Reset Test – HAMP2

Mg.Eff. PPM – HAMP1

Mg.Eff. PPM – HAMP2

CONCLUSIONS 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