Self-reported health assessments in the 2002 World Health Survey: how do they correlate with education?

Abstract

Objective

To assess the value of self-rated health assessments by examining the association between education and self-rated poor health.

Methods

We used the globally representative population-based sample from the 2002 World Health Survey, composed of 219 713 men and women aged 25 and over in 69 countries, to examine the association between education and self-rated poor health. In a binary regression model with a logit link function, we used self-rated poor health as the binary dependent variable, and age, sex and education as the independent variables.

Findings

Globally, there was an inverse association between years of schooling and self-rated poor health (odds ratio, OR: 0.929; 95% confidence interval, CI: 0.926–0.933). Compared with the individuals in the highest quintile of years of schooling, those in the lowest quintile were twice as likely to report poor health (OR: 2.292; 95% CI: 2.165–2.426). We found a dose–response relationship between quintiles of years of schooling and the ORs for reporting poor health. This association was consistent among men and women; low-, middle- and high-income countries; and regions.

Conclusion

Our findings suggest that self-reports of health may be useful for epidemiological investigations within countries, even in low-income settings.

Introduction

There are doubts about the validity of using self-reports of health for assessing population health, particularly in disadvantaged populations. Since self-assessment of health is directly contingent on social experience, it has been argued that disadvantaged groups will fail to perceive and report the presence of illness or health deficits, which may result in misleading assessments of population health.¹ This bias, referred to as “reporting heterogeneity”, has been demonstrated using hypothetical scenarios – formally referred to as vignettes – that make it possible to compare self-reports from respondents with different socioeconomic and other personal characteristics.² The bias has also been demonstrated by the finding that advantaged populations tend to report higher levels of poor health than disadvantaged populations.¹

In spite of reporting heterogeneity, a recent meta-analysis of 40 studies found a strong, statistically significant positive association between education and health, such that individuals with higher education reported better health status.³ We updated the literature review from the meta-analysis to include over 60 publications. Although most of these studies showed a positive association between self-rated health and education, they varied widely in terms of sample size, the specification of the self-rated health and education measures, the choice of the covariate set and the modelling strategy. Also, few studies^4,5 have focused on low- and middle-income countries, perhaps due to doubts about self-rated health assessments.^1,2,6 Thus, there is a need for a study of any association between self-rated health and education in a dataset that is globally representative and designed for making comparisons among as well as within countries.

An association between education and self-rated health would not in itself show whether self-rated health is a precise or valid means of assessing population health. For example, one study has suggested that people in Sweden tend to overrate, and those in Germany to underrate, their health status.⁷ However, in both Germany and Sweden, socioeconomic status was found to have a statistically significant positive association with health.⁸ Also, even if self-reports of health are not valid for comparing aggregate health among countries, they may still be valid for a within-country comparison.^2,9 In this study, we test the hypothesis that self-reports of health are valid by examining the association between self-rated health and years of schooling in 69 countries.

Methods

Data source

The data for this study came from the 2002 World Health Survey, which was conducted in 69 countries, across all continents, between January and December 2002.^10–12 The target population for the World Health Survey was all adults aged 18 years or over who were living in private households. The survey did not cover populations on military reservations, in group quarters or in living arrangements other than private households.

Sampling plan

In 10 countries, respondents were selected through a single-stage random sample; in the remaining 59 countries, a multistage stratified sample survey was conducted.^10–12 Populations were stratified by province in each country, and again by county in 58 countries. Sampling units were selected based on a probability proportional to population size, followed by a random selection of households. In most of these countries, enumeration areas (geographical areas canvassed by one representative) and households were used as additional units for stratification. Anyone considered to be a member of the household (i.e. “someone who usually stays in the household, sleeps and shares meals, who has that address as primary place of residence or who spends more than six months living there”), aged 18 years and over was eligible to be a respondent in the survey.^10–12 Within households, respondents were selected using Kish tables, which gave each eligible respondent an equal probability of being selected.

Population and sample size

From the total study population (n = 275 996),^10–12 we selected men and women aged 25 years or over (n = 228 993), because younger respondents were less likely to have completed their ultimate educational level, which was the primary marker of social disadvantage in this study. From this sample, information was missing on a total of 9280 respondents – 3206 on self-rated health, 7328 on years of schooling and 687 on covariates (Appendix A, available at: http://www.hsph.harvard.edu/faculty/venkata-sankaranarayanan/files/Bull-WHO-2009-Web-Appendices.doc; note: for many respondents, information was missing on multiple variables). Thus, the final analytical sample size was 219 713 respondents from 69 countries.

Self-rating of health

Self-rated health was assessed by asking respondents: “In general, how would you rate your health today” with the possible choices being “very good” (1), “good” (2), “moderate” (3), “bad” (4) or “very bad” (5).¹³ This scale is similar to the five-point Likert scale of self-rated health, which is a robust predictor of mortality and correlates strongly with other objective health indicators, especially in developed countries.^14,15 We analysed self-rated health as a dichotomous measure of self-rated poor health – “very good”, “good” or “moderate” were coded as “0”, and “bad” or “very bad” as “1”.

Education

The World Health Survey collected data on respondents’ educational attainment in two ways.¹³ Respondents were asked to report the highest level of education completed with the following options: “no formal schooling”, “less than primary school”, “primary school completed”, “secondary school completed”, “high school (or equivalent) completed”, “college/pre-university/university completed” and “postgraduate degree completed”. Respondents were also asked to report the number of years of schooling they had completed, including higher education. We used reported years of schooling as an indicator of educational attainment, mainly to overcome the issue of incomparability among countries on the categorical measure of educational attainment. For 22 182 respondents, the number of years of schooling was coded as “missing” in the dataset, even though 16 573 of these respondents had selected “no formal schooling” as a response to the categorical question on educational attainment. We therefore coded these respondents as having zero years of schooling in the analytical dataset. We specified years of schooling as a continuous measure and also separately specified quintiles based on years of schooling, using country-specific distribution of years of schooling.

Covariates

Age in years and sex were included as covariates in the study (Table 1).

Table 1. Data on self-reported assessment of health and on schooling for 69 countries, organized by income level and region^a.

Country	No. of survey respondents	No. (after deleting missing data)	Percentage with bad/very bad health	Mean self-rated health score	Mean age, in years	Percentage female	Median schooling, in years
Total	228 993	219 713	9.8	1.333	45.372	56.15	6
Europe and central Asia
High-income countries
Austria	943	902	4.3	0.936	47.975	62.75	10
Belgium	883	794	6.1	1.092	48.618	55.92	14
Denmark	959	956	4.5	0.938	52.194	53.03	11
Finland	944	935	7.1	1.402	54.966	55.19	11
France	890	768	4.5	1.074	46.936	58.46	14
Germany	1 147	1 090	8.5	1.309	53.401	60.28	10
Greece	916	914	8.9	1.138	53.820	50.11	9
Ireland	867	521	4.8	0.789	47.942	60.08	13
Israel	1 079	1 059	5.9	0.956	48.345	57.98	14
Italy	908	891	6.4	1.313	51.210	58.25	11
Luxembourg	620	601	5.2	1.120	48.296	50.58	12
Netherlands	825	757	4.7	1.147	51.407	72.52	12
Norway	884	866	6.7	0.989	50.760	50.58	12
Slovenia	514	507	11.7	1.466	50.998	54.04	12
Spain	5 960	5 827	10.2	1.354	54.758	59.19	8
Sweden	908	893	15.5	1.309	53.732	58.90	12
United Kingdom	1 072	1 047	9.1	1.196	53.944	63.13	11
Upper middle-income countries
Croatia	932	929	18.7	1.573	54.097	59.96	12
Czech Republic	828	811	12.8	1.499	51.083	55.86	12
Estonia	928	924	15.9	1.780	52.308	63.96	12
Hungary	1 262	1 254	14.9	1.643	53.034	59.25	11
Latvia	764	745	24.2	1.976	54.486	68.32	11
Slovakia	1 950	1 324	7.5	1.323	42.821	69.79	13
Lower middle-income countries
Bosnia and Herzegovina	917	917	16.1	1.420	50.097	58.34	11
Kazakhstan	4 111	4 102	6.0	1.561	43.275	65.77	13
Russian Federation	4 070	3 900	20.4	1.901	53.577	64.69	12
Turkey	9 681	9 664	10.4	1.502	45.277	56.33	5
Ukraine	2 522	2 475	27.1	2.061	50.689	65.29	12
Low-income countries
Georgia	2 441	2 436	24.3	1.916	52.228	57.88	11
Middle East and north Africa
High-income countries
United Arab Emirates	984	962	2.5	0.759	40.366	47.71	13
Lower middle-income countries
Morocco	4 184	4 163	27.0	1.879	44.883	58.32	0
Tunisia	4 214	4 057	10.8	1.324	45.786	55.11	6
South Asia
Lower middle-income countries
Sri Lanka	5 642	5 098	5.2	1.111	44.717	54.30	8
Low-income countries
Bangladesh	4 528	4 499	19.7	1.757	42.616	52.12	2
India	8 140	7 560	17.7	1.477	43.064	52.34	2
Nepal	6 979	6 970	10.9	1.386	43.280	56.30	0
Pakistan	5 031	4 884	5.6	1.118	41.646	45.70	0
East Asia and Pacific
High-income countries
Australia	1 621	1 619	2.5	0.804	49.991	58.62	12
Upper middle-income countries
Malaysia	5 250	5 133	3.8	1.082	44.174	57.08	9
Lower middle-income countries
China	3 674	3 665	8.4	1.279	47.230	51.35	7
Philippines	8 381	8 355	4.0	1.398	42.550	54.53	9
Low-income countries
Lao People’s Democratic Republic	4 060	4 053	4.2	0.931	41.767	52.73	3
Myanmar	4 996	4 996	2.9	1.074	44.578	57.31	5
Viet Nam	2 983	2 954	7.7	1.526	43.382	55.48	7
Sub-Saharan Africa
Upper middle-income countries
Mauritius	3 385	3 362	15.0	1.356	45.184	52.71	7
Lower middle-income countries
Namibia	3 288	3 014	7.2	1.027	42.863	59.85	7
South Africa	1 877	1 864	8.5	1.085	41.833	53.17	9
Swaziland	2 403	1 496	48.9	2.280	43.800	57.69	7
Low-income countries
Burkina Faso	3 608	3 545	8.2	1.215	41.504	51.03	0
Chad	3 640	3 379	14.9	1.439	42.172	53.18	0
Comoros	1 411	1 407	18.4	1.563	47.535	57.00	0
Congo	1 939	911	14.8	1.399	42.052	53.57	5
Côte d’Ivoire	2 403	1 827	11.1	1.347	41.007	43.90	0
Ethiopia	3 775	3 750	6.5	0.972	41.953	51.01	0
Ghana	3 304	3 194	8.2	1.097	45.084	55.57	4
Kenya	3 449	3 339	9.5	1.266	42.653	58.16	8
Malawi	3 766	3 647	5.8	0.819	42.337	56.73	5
Mali	3 291	2 642	8.5	1.154	46.544	44.13	0
Mauritania	3 051	2 778	5.0	1.187	43.419	62.85	0
Senegal	2 581	1 981	9.8	1.368	43.383	49.92	0
Zambia	2 848	2 799	8.4	1.104	41.208	53.48	7
Zimbabwe	3 024	2 966	13.8	1.548	42.999	64.70	7
Latin America and the Caribbean
Upper middle-income countries
Mexico	32 130	32 129	6.5	1.280	45.136	57.47	6
Uruguay	2 689	2 674	2.5	1.024	48.692	51.80	10
Lower middle-income countries
Brazil	4 209	4 167	10.5	1.509	45.613	56.83	5
Dominican Republic	3 758	3 733	10.4	1.440	45.785	53.28	6
Ecuador	3 873	3 526	10.6	1.471	44.930	55.53	6
Guatemala	3 836	3 749	12.0	1.495	44.686	61.48	2
Paraguay	4 063	4 057	3.2	1.059	44.939	54.57	6

^a Based on The World Bank classification.
Data source: World Health Survey 2002.

Statistical analysis

We modelled the log odds of reporting poor health using binary regression with a logit link function and robust error variance, given as:

where the quantity π_i/(1– π_i) is the odds that self-rated poor health for individual i = 1, 0 otherwise; β₀ represents the log odds of reporting poor health for the reference category (intercept); and ΒX represents the change in the log odds of reporting poor health for a one unit change in a vector of independent variables (age, sex and education). Where appropriate, the coefficients and standard errors took account of the multistage cluster survey sampling design. Models were fitted using SPSS v 15.0 for Windows (SPSS Inc., Chicago, IL, United States of America). Statistical precision was ascertained using two-tailed Wald tests and the results are presented with 95% confidence intervals (CIs). The logits were exponentiated to odds ratios (ORs) for interpretative reasons.¹⁶ All analyses were adjusted for age and sex; at the global level, they were based on The World Bank income classification of countries and The World Bank geographical regions.¹⁷ The analyses in the pooled global and regional sample included fixed effects for countries, achieved by including an indicator variable for each country. All analyses were repeated separately for men and women.

Ethical review

The 2002 World Health Survey was conducted under the scientific and administrative supervision of the WHO, and there was an independent ethics review of the World Health Survey protocol. Interviewers obtained informed consent for the survey, in writing, from the respondents.¹⁸ The study was reviewed by the Harvard School of Public Health, whose Institutional Review Board judged the study as exempt from full review because it was based on an anonymous, public use dataset with no identifiable information on the survey participants.

Results

The global prevalence of self-rated poor health was 9.8%; the mean age in the sample was 45.3 years; 56.2% of the respondents were female and the median schooling across all countries was 6 years (Table 1). There was considerable variation between countries on the prevalence of self-rated poor health and education. The percentage of respondents self-reporting poor health was highest in Swaziland (48.9%) and lowest in Australia, the United Arab Emirates and Uruguay (2.5%). Median years of schooling was highest in Belgium, France and Israel (14 years) and lowest in Burkina Faso, Chad, the Comoros, Côte d’Ivoire, Ethiopia, Mali, Mauritania, Morocco, Nepal, Pakistan and Senegal (0 years). At the country level, there was no correlation between the percentage of the population self-reporting poor health and the median years of schooling (r = −0.091; P = 0.459). However, at the individual level, there was a statistically significant negative correlation between the years of schooling and self-rated poor health (r = –0.143; P < 0.0001). At the country level, there was also no correlation between the percentage of the population reporting poor health and life expectancy in 2002 (r = –0.198; P = 0.103).

In pooled models adjusted for age and sex, there was an inverse association between years of schooling and self-rated poor health (OR: 0.929; 95% CI: 0.926–0.933) (Table 2). A similar relationship was observed for men and women in the age-adjusted pooled sample of all countries (Table 2). Compared to individuals in the highest quintile of years of schooling, those in the lowest quintile were twice as likely to report poor health (OR: 2.292; 95% CI: 2.165–2.426). There was also a dose–response relationship, in both men and women, between quintiles of years of schooling and the ORs for self-reporting poor health (Fig. 1).

Table 2. ORs reflecting the change in the odds of self-reporting poor health with every one-year increase in schooling, in 69 countries grouped by income level and region^a.

	All			Men			Women
	ORb	95% CI		ORb	95% CI		ORb	95% CI
All countries	0.929	0.926–0.933		0.933	0.927–0.939		0.927	0.922–0.932
National income
High	0.910	0.897–0.923		0.905	0.884–0.927		0.913	0.897–0.930
Upper middle	0.923	0.914–0.932		0.925	0.911–0.939		0.922	0.911–0.934
Lower middle	0.920	0.914–0.926		0.921	0.911–0.931		0.920	0.912–0.927
Low	0.948	0.942–0.955		0.951	0.942–0.960		0.945	0.936–0.954
Region
Europe and central Asia	0.903	0.895–0.910		0.907	0.894–0.921		0.903	0.893–0.912
Middle East and north Africa	0.942	0.929–0.956		0.939	0.920–0.959		0.944	0.926–0.963
South Asia	0.939	0.930–0.949		0.944	0.932–0.957		0.933	0.919–0.948
East Asia and Pacific	0.925	0.911–0.939		0.931	0.909–0.953		0.919	0.901–0.938
Sub-Saharan Africa	0.949	0.941–0.957		0.948	0.936–0.960		0.949	0.938–0.960
Latin America and the Caribbean	0.928	0.921–0.936		0.926	0.913–0.939		0.930	0.920–0.939

CI, confidence interval; OR, odds ratio.
^a Based on The World Bank classification.
^b Adjusted for age and country fixed effects.
Data source: World Health Survey 2002.

Fig. 1.

Self-reporting of poor health in each schooling quintile, in a pooled sample of 69 countries from the 2002 World Health Survey

CI, confidence interval.

^a Adjusted for age and country fixed effects.

The inverse association between years of schooling and self-rated poor health was observed in countries of all income levels (Table 2). Comparing countries, the ORs for reporting poor health for every one-year increase in schooling ranged from 0.910 (95% CI: 0.897–0.923) for high-income countries to 0.948 (95% CI: 0.942–0.955) for low-income countries. Again, a similar association (between years of schooling and the odds of self-reporting poor health) was observed for men and women (Table 2). Comparing regions, the ORs representing the change in the odds of reporting poor health for every one-year increase in schooling ranged from 0.903 (95% CI: 0.895–0.910) for countries in Europe and central Asia to 0.949 (95% CI: 0.941–0.957) for countries in sub-Saharan Africa. The strength of the association was similar for men and women in all regions (Table 2).

The inverse association between years of schooling and self-rated poor health was observed in all countries, even though the relationship did not always attain conventional levels of statistical significance (Fig. 2). Countries where the ORs for reporting poor health for a one-year increase in schooling was greater than 0.959 (i.e. a relatively small effect) were Australia, Bosnia and Herzegovina, Burkina Faso, Chad, the Congo, Côte d’Ivoire, Ethiopia, Ghana, Malawi, Morocco, Myanmar, Nepal, Senegal, Sweden and Zambia. This pattern was replicated for men and women (Appendix B, available at: http://www.hsph.harvard.edu/faculty/venkata-sankaranarayanan/files/Bull-WHO-2009-Web-Appendices.doc).

Fig. 2.

Self-reporting of poor health for every one-year increase in years of schooling, in a pooled sample of 69 countries from the 2002 World Health Survey

^a Adjusted for age and sex.

The country-specific association between quintiles of years of schooling and self-rated poor health was also as expected. With the exception of Burkina Faso and Chad, in countries in the lowest quintile of years of schooling people were consistently more likely to self-report poor health than in those in the highest quintile. Country-specific results showing the OR for self-reporting poor health by quintiles of years of schooling, stratified by men and women, are presented in Appendix C (available at: http://www.hsph.harvard.edu/faculty/venkata-sankaranarayanan/files/Bull-WHO-2009-Web-Appendices.doc).

To test sensitivity, we repeated the key analysis (Table 2, Fig. 1, Fig. 2) without dichotomizing self-rated health but instead using the entire 5-year scale item in an ordered multinomial regression. Patterns of association between self-rated health and years of schooling remained the same, and mirrored the results for the binary regression models (Appendix D, available at: http://www.hsph.harvard.edu/faculty/venkata-sankaranarayanan/files/Bull-WHO-2009-Web-Appendices.doc).

Discussion

Analysis of a globally representative and comparable, disaggregated dataset from 69 countries showed that adults (both men and women) with lower levels of education were consistently more likely to self-report poor health than those with higher levels of education. This finding was not dependent on a country’s level of economic development or on regional geography.

Within each country, we found little reporting heterogeneity (or bias), in that disadvantaged individuals did not appear to underreport poor health when compared to advantaged individuals. Nevertheless, it is possible that disadvantaged individuals underestimate the extent of their poor health. Furthermore, the level of reporting heterogeneity by level of education may differ among countries, making it difficult to compare the magnitude of the association between education and self-reporting of poor health among countries. Thus, as a validation of the self-rated health measure, our findings must be interpreted with caution.

In spite of these limitations, our results suggest that the magnitude of underestimation of poor health by those with low education is not so large as to be misleading. A more thorough test would be to examine the predictive capability of self-rated poor health for objective outcomes such as mortality. Indeed, self-rated poor health is a robust predictor of mortality in the context of industrialized countries.^14,15,19 Evidence from Bangladesh⁵ and Indonesia²⁰ suggests that these associations might also be true in developing countries. However, variations in self-rated health may not mirror variations in mortality, because the former captures more than objective physical health; for example, it also incorporates important dimensions of mental health.¹⁸

We deliberately restricted the number of covariates for this analysis to facilitate comparisons within and among countries. For example, besides age and sex, there are likely to be differences in how individuals perceive their health, depending on whether they live in urban or rural locations that have different levels of health awareness and expectations. We chose not to include distinctions between urban and rural areas among countries because of the considerable variation in definitions of these terms. Nevertheless, the consistency of our findings across multiple geographical regions and different levels of economic development suggests that findings would probably be similar across urban and rural areas within a country. We also used only one factor – low level of education – as a chronic measure of social disadvantage. Previous research suggests that education is less likely than income, wealth or occupation to be a consequence of adult health.²¹ Also, education is likely to be a determinant of other socioeconomic markers such as income, wealth and occupation. Had we used other markers of an individual’s socioeconomic status, it would have been difficult to make straightforward comparisons among multiple countries.

This study fills a critical gap by providing baseline global assessments of the association between education and self-rated health. The 2002 World Health Survey is the most recent international survey purposefully designed to obtain comparable data on health and related determinants across countries in all world regions. Therefore, we could not examine whether the pattern for 2002 was reproduced in more recent years. However, associations between education and health have been shown to be pervasive and to change slowly. Therefore, the main finding of our paper is unlikely to have changed substantially since the 2002 World Health Survey. As more recent and comparable data become available, future studies should examine whether the association continues.

Although self-reports of health may not always accurately capture variations in absolute health across countries, doubts about the use of self-reported health measures to study health disparities within countries, especially developing countries, should be reappraised. The ease, speed and economy of collecting self-reports of health (even with a single item global question such as the one used here) make such collection attractive for rapid appraisals. Also, collecting self-reports of health will make it easier to empirically assess epidemiologic associations between various exposures and health, especially in countries where objective health data are lacking and where subjective health data have been viewed with considerable scepticism. ■