The Correlation Between Education and Health: The Role of Differential Reporting Error

One of the most robust findings in health economics is that the better educated tend to be in better health.  The model of health capital (Grossman, 1972) predicts that those with more schooling will demand more health, which has been widely confirmed (e.g. Grossman and Kaestner, 1997; Cutler and Lleras-Muney, 2010).  This positive correlation is often attributed to allocative efficiency (the better educated choosing a different set of health inputs or behaviors) and/or productive efficiency (the better educated producing more health from the same inputs). 

This study examines a novel hypothesis: that this relationship may to some extent be due to differences in reporting error by education levels.  For example, there may be a greater social desirability bias among the better educated: they may be more embarrassed to report engaging in risky health behaviors, leading them to underreport smoking habits, sexually transmitted infections and underestimate their weight.  On the other hand, the better educated may be more accurate in reporting their health behaviors because they are more knowledgeable about their health or are better able to understand the survey questions.  Thus, differential reporting error across education groups may lead to either an overestimate or an underestimate in educational gradients in health.

We examine data from the pooled National Health and Nutrition Examination Survey (NHANES) Continuous for 1999-2010.  These data include both self-reports and objective verification for an extensive set of health behaviors and conditions, including current smoking status, pregnancy, sexually transmitted infections, weight, high blood pressure, cholesterol and diabetes.  We examine how respondents in different education categories vary in terms of: 1) accuracy of self-report, regardless of direction; and 2) accuracy of self-report that accounts for direction of error, for instance the extent to which weight was over-reported or under-reported.  We estimate limited dependent variable models of these outcomes on education, controlling for various demographic characteristics (age, sex, race, marital status) and year fixed effects. 

We find that the better educated report risky health behaviors and their consequences more accurately than those of less education. This suggests that the better educated are either more knowledgeable about their own health or are more willing or able to respond accurately to surveys.  In contrast, we do not find much evidence that the better educated are more prone to social desirability bias, even for relatively embarrassing conditions such as sexually transmitted infections or stigmatized behaviors such as smoking.  The implications of this study are that using self-reported data on health and health behaviors actually leads to an underestimate of the true gradient in health across education groups.  This underscores the importance of surveys collecting objective measures of health and health behaviors, and in general reducing the influence of reporting error.