Socio-economic influences on self-rated health trajectories: evidence from four OECD countries

1. Socio-economic influences on self-rated health trajectories: evidence from four OECD countries Amanda Sacker Peggy McDonough Diana Worts Mplus Users Meeting Meeting 8th June 2009

2. Background • The life course and the welfare state • Societies with weak social safety nets have worse population health than those with strong supports • Problem: based almost exclusively on aggregate, cross-sectional measurements of health • Comparative individual health dynamics and their social patterning

3. Aims of study • Describe average national trajectories of self-rated health over a 7-year period • Identify socio-economic determinants of cross-sectional and longitudinal health • Compare cross-national patterns

4. Welfare state typologies • Esping Andersen (1990) • “three worlds of welfare” • decommodification, social stratification, private-public mix • liberal, conservative, social-democratic • Castles & Mitchell (1993) • 2 X 2 cross-classification • aggregate welfare expenditure, benefit equality • liberal, conservative, non-right hegemony, radical

5. Castles & Mitchell’s typology

6. The data • Four panel surveys • US Panel Study of Income Dynamics • British Household Panel Survey • German Socio-Economic Panel Survey • Danish panel from the European Community Household Panel Survey • Respondents of working age throughout the follow-up period • Covariates measured in 1994 • Health reported 1995-2001

7. Methods • Latent growth curves • Linear growth curves by elapsed time controlling for age/age squared at baseline • Covariate effects on intercept and slope • Graphing health over time • Aging vectors • Synthetic cohort trajectories • Estimate health trajectories • Compound effect of covariates

8. The latent growth curve model . . . . . . . . . . . . . . . SRH 1995 SRH 1996 SRH 2001 0 1 6 1 1 1 Slope Intercept Age Gender Ethnicity Marital status Employ status Occup class Educ Income

9. The latent growth curve model

10. Mplus syntax VARIABLE: NAMES ARE (omitted) USEVARIABLES ARE srhth95 srhth96 srhth97 srhth98 srhth00 srhth01 rout94 missoc94 minority single94 excoup94 unempl94 outlf94 meded94 lowed94 female agecen agesq logposp94; MISSING ARE ALL (-999); WEIGHT IS XRWGT94; STRAT IS strata; CLUSTER IS psu; CATEGORICAL ARE srhth95 srhth96 srhth97 srhth98 srhth00 srhth01; CENTERING IS GRANDMEAN (rout94 missoc94 minority single94 excoup94 unempl94 outlf94 meded94 lowed94 female logposp94); DEFINE: agecen = agecen/10; agesq = agecen^2; ANALYSIS: TYPE IS COMPLEX; MODEL: i s | srhth95@0 srhth96@.1 srhth97@.2 srhth98@.3 srhth00@.5 srhth01@.6; i s ON rout94 - logposp94;

11. Self-rated health • US “Would you say [your/his/her] health in general is excellent, very good, good, fair, or poor?” • Britain “Please think back over the last 12 months about how your health has been. Compared to people of your own age, would you say that your health has on the whole been excellent, good, fair, poor or very poor?” • Germany “How would you describe your current health, very good, good, satisfactory, poor, bad?” • Denmark “How is your health in general, very good, good, fair, bad, very bad?”

13. Aging vector graphs

14. Stata syntax ** data file with one record for each synthetic birth cohort ** age centred on 40 years in 10 year units ** estimates imported from Mplus growth curve model ** calculate self-rated health in 1995 and 2001 for each birth cohort gen srh_1 = -i_mean - (i_age*agecen + i_agesq*agecen^2) gen srh_2 = -i_mean - (i_age*agecen + i_agesq*agecen^2) /// - 0.6*(s_mean + s_age*agecen + s_agesq*agecen^2) ** figure with arrows from health in 1995 to health in 2001 for those aged 25-57 in 1994 gen age_1 = age gen age_2 = age+6 graph twoway (pcarrow srh_1 age_1 srh_2 age_2), /// xtitle(Age) ytitle("Predicted SRH Z-score", size(large)) /// xtick(25(5)65) xlabel(25(5)65, labsize(large)) xtitle(,size(large)) /// ytick(-1(.5).1) ylabel(-1(.5)1,angle(0) labsize(large)) /// title("a) United States", size(vlarge)) legend(label(1 "1995 to 2001")) /// saving("US aging vector.gph", replace)

15. Socio-demographic covariates • Age • Gender • Ethnicity • Marital status • Education • Occupational class • Employment status • Income

16. Intercept regressed on covariates * p < 0.05 ** p < 0.005 *** p < 0.0005

17. Slope regressed on covariates * p < 0.05 ** p < 0.005 *** p < 0.0005

18. Intercept regressed on covariates:interactions with age * p < 0.10 ** p < 0.05 *** p < 0.005

19. Age by ethnicityeffects on baseline health

20. Age by employment status effects on baseline health

21. Aggregate effects • Average ideal type • Mean values for all covariates • Advantaged ideal type • Male, majority ethnic group, cohabiting, tertiary educated, employed, non-routine occupational class, above median income • Disadvantaged ideal type • Female, minority ethnic group, no longer living with partner, lower secondary education, unemployed, routine occupational class, below median income

22. Growth curves by levels of advantage

23. Substantive conclusions • Socio-economic influences on trajectories of self-rated health broadly consistent with welfare typologies • Results suggest • Health of minority groups may be more affected by aggregate welfare expenditure • Work and health may be more affected by benefit equality

24. Methodological conclusions • Latent variable modelling cannot overcome differences in question wording and response labelling • Cannot make between-country comparisons about mean levels of health and rate of change • Ageing vectors useful for identifying cohort effects

25. Discussion • Causality? • Lack of relationship between covariates and changes in health • Common finding • Violations of measurement invariance • How far can anchoring vignettes help? • Other adjustment procedures?