Comparing subjective and objective measures of health: Evidence from hypertension for the income/health gradient

https://doi.org/10.1016/j.jhealeco.2009.02.010Get rights and content

Abstract

Economists rely heavily on self-reported measures of health to examine the relationship between income and health. We directly compare survey responses of a self-reported measure of health that is commonly used in nationally representative surveys with objective measures of the same health condition. We focus on hypertension. We find no evidence of an income/health gradient using self-reported hypertension but a sizeable gradient when using objectively measured hypertension. We also find that the probability of false negative reporting is significantly income graded. Our results suggest that using commonly available self-reported chronic health measures might underestimate true income-related inequalities in health.

Introduction

The relationship between income and health, and any underlying causal mechanisms, is a hotly debated topic (see, for example, Deaton and Paxson, 1998, Smith, 1999, Chase, 2002, Jones et al., 2008). In examining this topic, economists have relied heavily on self-reported measures of general health status and, to a lesser extent, self-reported chronic health conditions1. Recent studies using self-reported measures of general health status and panel data have found only a small role for increased income leading to improved health (see, for example, Adams et al., 2003, Meer et al., 2003, Contoyannis et al., 2004, Frijters et al., 2005a, Lindahl, 2005, Jones et al., 2008). However, if such self-reported measures suffer from reporting error, this has implications for the robustness of the findings; particularly if the reporting error varies systematically with the same characteristics that are most widely used to assess whether inequalities in health exist.2 The extent to which there is reporting error in self-reported health is therefore an important issue (see, for example, Butler et al., 1987, Mackenbach et al., 1996, Lindeboom and van Doorslaer, 2004, Jones et al., 2008).

Reporting heterogeneity can occur for a number of reasons. Bago d’Uva et al. (in press) argue that reporting heterogeneity is associated with individuals’ concept of what ‘health’ means, their expectations of their own health, their use of healthcare, and their comprehension of the actual survey questions asked.3 Etile and Milcent (2006) find evidence that reporting heterogeneity is associated with income. Lindeboom and van Doorslaer (2004) find that it is associated with age and gender though not with income or education.4

One method of tackling this potential measurement problem is to ‘purge’ self-reported general health measures of reporting error using other measures of health typically available in survey data that are still self-reported, but which are seen to be more objective. Most commonly, these other measures of health are chronic health measures, such as specific medical conditions and functional limitations. The purging method replaces the actual reported values of self-assessed health with predicted values of self-reported general health derived from models which estimate self-reported health as a function of the self-reported chronic health measures. Demographic variables are often also included as well with the aim of purging further subjectivity and endogeneity in individuals’ self-reports. Examples of papers that use this statistical approach in a range of contexts are Bound et al. (1999), Disney et al. (2006), Hagan et al. (2006), Zucchelli et al. (2007).5

However, it is not clear that these more objective health measures are not also subject to reporting error. Baker et al. (2004) matched a wide range of self-reported chronic health conditions to records of public healthcare usage in Canada and found clear evidence that such conditions are subject to a large amount of systematic reporting error, leading to large attenuation biases when these are used as explanatory variables. They also found that false negative reporting, where individuals had used medical services but not reported a matched health condition in the survey, was around 50% for most chronic conditions examined. The purging approach will also be problematic if the demographic variables that are used to form the purged predictions (e.g. age, education and region of residence) are themselves directly related to the measurement error. Mackenbach et al. (1996), for example, found more underreporting of self-reported chronic conditions (lung disease, heart disease and diabetes) among less educated persons. Butler et al. (1997) found that reporting error in self-reported arthritis differs by employment status and income.

In this paper, we contribute to the literature by examining whether there are income-related differences in subjective and objective reports of a single condition, which has two important features. The first is that it is a condition that is a major public health concern; the second is that individuals may be unaware they have this condition. We investigate the differences between self-reported and actual health by matching self-reports of hypertension (or high blood pressure), to precise clinical measures of that same condition. Hypertension is one of the most prevalent chronic health conditions in Western countries and is a major risk factor for cardiovascular disease. Moreover, hypertension is often called the ‘Silent Killer’, because it is typically asymptomatic at moderate and even highly evaluated levels of blood pressure. The asymptomatic nature of hypertension means that individuals may simply not be aware that they have such a condition.

Lack of awareness may arise for several reasons: individuals may not recognise their condition, they may give incorrect information to clinicians, they may misinterpret or misremember medical advice or they may be given incorrect information by clinicians. If lack of awareness is random then this simply adds noise to estimates using subjective measures. If, on the other hand, it is socially graded and, more specifically, varies systematically with income, then use of subjective measures in analyses of the relationship between health and income will provide distorted estimates of the income/health gradient.

In this paper we use detailed data drawn from the Health Survey for England (HSE), which contains information from both survey respondents and an interview by a trained nurse, to examine how subjective and objective measures of hypertension vary with income and other individual attributes. Unlike Baker et al. (2004) our data do not contain possible errors arising from the need to match separately collected public healthcare records with survey data.6 Our clinical measures of the condition are taken from individuals at the same time (within 2 weeks) as they are asked questions about their health. The limitation of our study is that we are only able to focus on hypertension, as it is the only condition where we can exactly match self-reported with objective measures.

Using the matched information from individual questionnaires and nurse interviews, we investigate if there are differences in the estimated relationship between socioeconomic characteristics, with a particular focus on income, and the probability of having hypertension using self-reported compared to objective measures. We then examine whether there are systematic reporting differences by socioeconomic characteristics, by estimating a model of false negative reporting. Reporting differences in hypertension arise because individuals are unaware of their true health status. This situation is different to that described, for example, in Lindeboom and van Doorslaer (2004), in which reporting differences in general health arise because individuals interpret the severity of known health conditions differently.7 In addition to standard demographic controls, we are also able to control for several factors which are likely to be correlates of awareness, namely the use of hypertensive medication, the severity of the disease, individual lifestyle choices and the availability and use of healthcare.

The asymptomatic nature of hypertension means that our study will obtain estimates of reporting error, which we define as a systematic difference between the subjective and objective measure, towards the upper-bound. This is because individuals are most likely to be unaware of a condition which is asymptomatic and hence will not report it when asked to do so in social or health surveys. This will result in a large number of false negative reports. This contrasts with other conditions where the individual is likely to have much higher levels of awareness, for example problems with eyesight, hearing or musculoskeletal pain. However, the asymptomatic nature of hypertension at moderate and even advanced levels is shared by many of the most prevalent chronic health conditions including diabetes, cardiovascular disease and many types of cancer. Thus our findings might also be applicable to a relatively large class of health conditions.

Section snippets

Key facts about hypertension

Hypertension or elevated blood pressure is an highly prevalent health condition, with the worldwide prevalence estimated to be as high as 1 billion individuals. In the US alone, it is estimated that some 50 million adults currently have high blood pressure (BP) that warrants some form of medical treatment (JNC7, 2004). While hypertension can occur in childhood and adolescence, the condition is mostly an adult concern, partly arising from the need for the heart to pump harder as the body ages.

A

The health survey for England

Our data source is the Health Survey for England (HSE), which was commissioned by the Department of Health, and carried out by the Joint Health Surveys Unit of Social and Community Planning Research, and the Department of Epidemiology and Public Health at University College London. Beginning in 1992, the HSE is an annual survey and is designed to monitor trends in the nation's health. The unit of survey in the HSE is the household, and information is collected from both adults and children.

The estimated income/health gradient

In this section we start by examining whether the lack of income gradient in self-reported chronic hypertension and the large gradient in measured hypertension, as well as the clear regional differentials, seen in the raw data are robust to controls for demographics, education, genetic predisposition, measures of location, employment, health and lifestyle.

We begin by estimating probit and regression models that control for basic demographic characteristics and year of survey. We then extend the

Conclusions

The use of self-reported general health status measures is widespread in the examination of the relationship between income and health. As a measure it captures what an individuals ‘feels’ about their own health at a given point in time, which may be important information for predicting individuals’ behaviour and decisions. However, such a measure may be less appropriate in the context of identifying health inequalities that should be of concern to policy-makers. This is because self-reported

Acknowledgements

We would like to thank two anonymous referees, Maarten Lindeboom, Eddy van Doorslaer, Karen Ireland and seminar participants at the Universities of Bristol, Bergen, Melbourne, South Australia, Western Australia, and Wollongong for their helpful comments.

References (39)

  • E. Van Doorslaer et al.

    Income-related inequalities in health: some international comparisons

    Journal of Health Economics

    (1997)
  • E. Van Doorslaer et al.

    Does inequality in self-assessed health predict inequality in survival by income? Evidence from Swedish data

    Social Science and Medicine

    (2003)
  • Bago d’Uva, T., van Doorslaer, E., Lindeboom, M., O’Donnell, in press. Does reporting heterogeneity bias the...
  • M. Baker et al.

    What do self-reported objective measures of health measure?

    Journal of Human Resources

    (2004)
  • J. Banks et al.

    Disease and disadvantage in the United States and England

    Journal of the American Medical Association

    (2006)
  • J. Bound

    Self-reported versus objective measures of health in retirement models

    Journal of Human Resources

    (1991)
  • J. Butler et al.

    Measurement error in self-reported health variables

    Review of Economics and Statistics

    (1987)
  • V. Burt et al.

    Prevalence of hypertension in the US adult population: results from the third National Health and Nutrition Examination Survey 1988–1991

    Hypertension

    (1995)
  • D. Card et al.

    The measurement of medicaid coverage in the SIPP: Evidence from a comparison of matched records

    Journal of Business & Statistics

    (2004)
  • Cited by (202)

    • CEO health

      2023, Leadership Quarterly
    View all citing articles on Scopus
    View full text