Gary King

King was the inaugural speaker for the 2005-06 Dean's Distinguished Lecture and spoke on September 29 in Snyder Auditorium. He was introduced by Dean Barry Bloom. The talk has been archived at http://www.hsph.harvard.edu/ddl/.

King focused on demography and forecasting mortality rates, starting out by describing weaknesses in current tools. The most common method-called deterministic projections-extends today's data into the future. So, for example, if Portuguese people were dying in car accidents at a certain rate in the 1950s, scientists could project how many would die in 2020. While working well with the right kind of data, the method fails to account for real-world influences, known in statistical language as covariates. In the Portuguese example, a deterministic projection would not have the flexibility to adjust for more cars on the road, improved roadways, or the introduction of seatbelts, noted King.

Another way of looking at data is through mortality age profiles, which are based on the observation that death rates tend to drop from birth until about age five and then increase steadily over the life span. Yet, this relationship between death rates and age differs considerably for different causes of death, noted King. So the methods that specify any one particular mathematical form for age profiles do not represent the data well and forecast poorly. They also exclude covariates and ignore what is important about mortality age profiles, such as the fact that adjacent age groups have similar death rates for almost all causes of death.

The methods developed by King and his colleagues avoid these and many other pitfalls. Information is elicited about mortality, which experts know about, rather than mathematical quantities such as regression coefficients, which are never observed. They include covariates, allow them to be different in each country or age group, and offer a variety of ways to combine data. They allow for ignorance, said King, because researchers may not know exactly why some patterns change. And they work even if the real-world meaning of variables is not the same in every country, so that GDP, for example, can be understood through DVD purchases in the U.S. or through goat purchases in another country.

King has co-authored a book manuscript on demographic forecasting with scientist Federico Girosi. Information about their new methods, manuscript, and easy-to-use software are available at http://gking.harvard.edu/.