Academic journal article Demographic Research

Does Selection of Mortality Model Make a Difference in Projecting Population Ageing?

Academic journal article Demographic Research

Does Selection of Mortality Model Make a Difference in Projecting Population Ageing?

Article excerpt

1. Introduction

Projecting mortality is a crucial first step in studying the prospects of population ageing and its consequences. As life expectancy increases and population-ageing speeds up worldwide (Lutz et al. 2008), a considerable effort is being made to expand the methodology of mortality projection (Booth and Tickle 2008; Ediev 2011; Lee and Carter 1992; Mayhew and Smith 2013; Pollard 1987; Raftery et al. 2012; Stoeldraijer et al. 2013). However, authors rarely pay attention to the importance of choosing mortality models and their implications for assessing future population size and composition, and consequently indicators of population ageing.

Even though many scientists have demonstrated crucial differences between mortality scenarios in terms of life expectancy at birth (often a scenario variable in population projections) and predicted death rates (Bell 1997; Benjamin and Soliman 1993; Cairns et al. 2011; Janssen and Kunst 2007; Pollard 1987; Shang et al. 2011; Stoeldraijer et al. 2013) the impact on projected population ageing is rarely studied. An infinite number of age-specific mortality patterns - with potentially different consequences for population ageing - may produce the same trajectory of life expectancy at birth.

A note is due here on existing approaches to projecting age-specific mortality. Often, projections rely on a single input parameter, typically life expectancy at birth, to describe future mortality scenarios, and then derive details of mortality by age and sex using a proper model. Partly this is done because of convenience in describing future scenarios. Another reason for applying the top-down approach comes from the observations that (linear) trends in life expectancy provide better fit compared to models for age-specific (log) mortality (Lee 2003; White 2002; see also Oeppen and Vaupel 2002 on a related matter). A widespread approach is extrapolating the agespecific trends in mortality rates despite its mentioned limitation. The Lee-Carter model is one of the best-known extrapolation methods. It relies on a singular-value decomposition (SVD) of age-specific log-mortality rates by age-time (different options exist for sex and regional trends) in order to determine the general time trend and agetime interactions (Lee and Carter 1992). The model is convenient for producing stochastic mortality forecasts, although it is also widely used in deterministic projections. A particular limitation of the method is its potential to generate implausible (non-monotone at old age) age patterns of future mortality, but this drawback may be mitigated by either using the life expectancy produced by the model as an input for another model (the approach was once adopted by the U.S. Census Bureau, although currently the Bureau is back to the life expectancy extrapolation method), or by applying adjustments to the model parameters to avoid implausible age patterns (Ediev 2007; a similar model of 'robust rotation' is used by the UN team to improve the model performance at old age, Sevcikova et al. 2015). Direct linear extrapolation of agespecific log-mortality rates (Ediev 2008) is similar to the Lee-Carter method in dealing with disaggregated mortality, yet it differs in producing age-specific time trends based on data periods of different duration at different ages (which would not be possible to combine with the SVD used in the Lee-Carter method), and in using simpler computational procedures. It is also supplemented by a special adjustment procedure to avoid implausible age patterns in the projected mortality.

The target mortality approach, which assumes convergence of the age pattern of the death rates to a specified target, is somewhat similar to the extrapolative methods in assuming age-specific trends that are not produced using statistical procedures applied to the past data but rather are imposed by assumption. The third domain comprises parametric models used to describe and project the age profile of mortality rates. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.