At a Glance
Methodological Advances in Mortality Modeling and Population Aging
Roland Rau, Marcus Ebeling, Torsten Sauer; in Collaboration with Anders Ahlbom, Karin Modig (both: Karolinska Institutet, Stockholm, Sweden), Carl P. Schmertmann (Florida State University, Tallahassee, USA)
The development of new methods to estimate mortality and population aging remains a core activity for demographers and statisticians. The following research questions are currently being addressed in our group:
How can we estimate mortality for small populations? There are many ages in small populations with very few or even zero ages. Thus, noise might dominate the signal, i.e., due to large random fluctuations it might be difficult to detect the underlying mortality. Based on the TOPALS approach by de Beer, the group researchers developed a new Bayesian model to obtain probabilistic estimates of mortality. We presented an application to life expectancy in German counties at various scientific conferences and in Germany's Federal Ministry of the Interior. Using Swedish data on the municipal level, the group is extending the limits of the model.
Since demographers are not yet as invested in the methods of machine learning as researchers from other disciplines, we look into the following questions: Can machine learning methods provide new insights for demographic analysis? We currently investigate whether random survival forests are a more useful tool to predict death than conventional methods. Our presentation of preliminary results received a poster award at the PAA 2019 conference and at a conference of young demographers.
Epidemiologists and demographers often work on the same subject: the health and mortality of individuals. One of the key strengths of how demographers can contribute to the debate is the development and application of decomposition methods. The group developed a new method to decompose lifetime risk into changes in incidence risk and increasing longevity.
Demographers often express their results in life expectancy, whereas epidemiologists use relative risks. Our research addresses the question of whether and how we can translate between the two measurements to connect the two disciplines even further.
Aging, Mortality and Longevity, Statistics and Mathematics
BMC Public Health 20:1523, 1–8. (2020)
PLoS One 13:4, e0195307–e0195307. (2018)