As a demographer, sociologist, and social epidemiologist, I pursue several research questions.   They all center on select important issues in population health.  Some questions are fascinating because of deep substantive and policy implications — in particular, how is educational attainment related to adult health in North America?  Other questions captivate me because of methodological paradoxes or quandaries.  Here are examples of both kinds of questions.


My main research area concerns the relationship between educational attainment and health of U.S. adults.

A few years ago, I discovered unexpected anomalies in the health gradient for two groups:  recipients of the GED diploma and adults with some college but no degree.  The first group is assumed to be equivalent to regular high school graduates and thus to have equivalent health.   The second group is assumed to have better health than HS graduates as a function of their additional education.  However, the data I examined did not support these expectations.  Instead, they showed that GED recipients had worse health and college dropouts no better health than high school graduates.

The first results were published in 2012 and 2013 in Social Science Quarterly, American Journal of Public Health, and Social Science and Medicine.   These studies documented the anomalies at the GED level and the subbaccalaureate level.  Since then, I focused on understanding the why these anomalies exist.

For instance, these recent studies used self-reported data.   What if the observed patters are due to systematic reporting differences across groups rather than differences in the actual underlying health status?  With Vicki Johnson-Lawrence,  we addressed this question by using biomarker measures available in NHANES.  We found that the anomaly was smaller but still visible even when we eliminated potential self-report bias. The paper was published in 2016 in Social Science and Medicine-Population Health.

I am currently working on additional studies aiming to understand these anomalies, and I will post more information as we firm up our findings.  I welcome your emails if you have questions or want to exchange ideas about this topic.


Another line of research I have been pursuing for nearly a decade pertains to the obesity paradox.  The obesity paradox is a phenomenon wherein high body weight is associated with poor health during the life course but lower mortality at older ages. Many researchers, including me, believe that at least a part of this phenomenon is due to our inability to model weight changes adequately.  In particular, we posit that at older ages lower body weight is a marker of high frailty, not healthy low body fat percentage as among younger people.

In a paper published in The Gerontologist a few years ago, my colleagues and I applied a joint generalized growth mixture-proportional hazards survival model and found that weight gain among older adults has little detrimental effects on mortality while weight loss is a powerful predictor of mortality.

Recently,  working with brilliant collaborators Snehalata Huzurbazar and Mark Greenwood, we used a powerful nonparametric approach to model BMI trajectories.  In a paper published in Journal of Aging and Health, we applied hierarchical clustering of body weight functions estimated via the PACE algorithm to determine typical weight trajectories and their health correlates.  This method is completely novel – its components have recently been developed in statistical journals but it has never been applied to any substantive questions, whether in population health or elsewhere.

We found three types of weight trajectories: stable overweight, obese declining, and overweight gaining.   Losing weight among older adults was associated with substantially worse health outcomes than either being overweight or even further gaining weight.   Looking forward, I plan to expand this flexible functional approach to understanding aspects of body weight trajectories such as socioeconomic and health correlates.