Demographic Projections & Forecasts Abstracts

Strand organiser: Professor Ludi Simpson, University of Manchester

Forecasting the components of population change - Tuesday 9 September 9.00am

Evaluating extensions to coherent mortality forecasting models
Syazreen Shair1, Sachi Purcal1, Nick Parr2, 1 Department of Applied Finance & Actuarial Studies, Faculty of Business and Economics, Macquarie University, 2 Department of Marketing and Management, Macquarie University

This paper evaluates the forecast accuracy of two recently published coherent mortality models; the Poisson common factor and the product-ratio functional. Both models aim for long term non-divergent mortality forecasts of sub-populations. These coherent models plus independent models (including Lee-Carter) are applied to sex-specific mortality data for Australia and sex-specific and ethnicity-specific data for Malaysia. The out-of-sample age-specific death rate errors from each model are compared and closely examined across groups. Results show that the mortality forecasts of both coherent models are not consistently more accurate than independent models. However, the sex ratios of sub-populations remain constant and are not increasingly different from each other. Although the product-ratio functional model out-performs the Poisson common factor model against the overall log death rates errors, none is consistently more accurate than the other for life expectancy forecasts. Our data reports that coherent models reduce the life expectancy mean forecast error for females significantly with the gap between forecast and actual is as small as 14 days.


Stochastic Forecasting Methods for the Total Fertility Rate in All Countries
Guy J. Abel1,2, Warren Sanderson3,4,5, Sergei Scherbov1,4
1 Wittgenstein Centre (IIASA, VID/ÖAW, WU), 2 Vienna Institute of Demography/ Austrian Academy of Sciences, 3 Wittgenstein Centre for Demography and Global Human Capital (IIASA, VID/ÖAW, WU), 4 International Institute for Applied Systems Analysis, Austria, 5 Stony Brook University, USA

We present a range of projections models to provide country specific stochastic projections of the total fertility rate (TFR) for all countries. Building off modelling concepts current used in the Bayesian hierarchical model of the United Nations (Alkema et. al. 2011, Raftery et. al. 2014), our models are divided into stages for fertility decline and post decline and in selected models include an hierarchical model structure. However, all our models have far fewer parameters to model the decline in TFR. In addition, we experiment with alternative definitions for the start of post fertility decline. Throughout we use regular functions in standard statistical software to quickly estimate linear and hierarchical (multi-level) model parameters. Comparisons of the accuracy and the calibration of uncertainty between all models are made through various validation statistics for forecasts based on truncated time series. We find the accuracy of our median simulated projections and their uncertainty to be similar and in some cases better than the Bayesian model. Further, we find that the definition of the start of the decline phase plays an important role for medium and long term projections. We propose four alternative definitions that would allow model parameters for forecasting post decline TFR to be based on a far more post decline observations, and from a wider range of post decline countries, than in the current Bayesian hierarchical model of the United Nations.


Shift and Compression of Mortality at Old Age: A Conservative Scenario
Dalkhat M. Ediev, IIASA’s World Population Program and the Vienna Institute of Demography of the Austrian Academy of Sciences (Wittgenstein Centre)

Formal relations are used to demonstrate inability of the relational Brass mortality model to keep up with declining mortality at old age. In order to adjust the model, a descriptive study is undertaken of mortality shifts at old age. To this end, ages X(M) at given levels of the mortality rate are studied. When arranged as functions of the life expectancy at birth, those ages show increasing steepness of the trend. This pattern is explained by approaching, as mortality declines, to upper limits of period mortality compression. In order to take this changing pattern into account, we obtain empirical lower-bound limits to X(M). Our models of lower-bound limits may be useful both in examining tendencies in period mortality shift and compression and as a starting point in adjusting the mortality projection models at older age. For the Brass model (standard: 1970; projection period: 2000+), the adjustment brings down the prediciton errors by about half at ages above 80 years. Our results may also be useful in the continuing discussion of prospects for further mortality decline.


Multistate & other projection models - Tuesday 9 September 11.00am

Using the 2011 Census to fix ethnic group estimates and components for the prior decade
Philip Rees, Nikolas Lomax, University of Leeds

One of the key functions of the decadal census is to serve as a check on roll-forward estimates of local populations, their components of change and attributes from the previous census. The Office for National Statistics, for example, has revised the 2002 to 2011 series of local authority mid-year population estimates, allowing for and distributing the difference between the roll-forward estimates for mid-year 2011 and the 2011 Census based estimates for mid-year. In this paper, these revisions are extended in several ways. First, we show how a demographic accounting framework and iterative proportional fitting can be used to adjust populations and components by age from 2001-2 to 2010-11. Second, the adjustments are extended from the ONS territory of England and Wales to the whole of the UK. Third, we use and adjust recently estimated consistent and harmonised inter-district migration in the process. Fourth, we apply this methodology to ethnic group populations. ONS has recently evaluated its roll forward methodology for local Population Estimates for Ethnic Groups (PEEG); a Leeds team has similarly evaluated its forward projection from 2001 to 2011 of local ethnic group populations for England and Wales. The table sums local population estimates to a national total and reveals differences which need resolution. The paper will report progress with this research agenda, made possible by the publication of population and migration data from the 2011 census.This analysis is an essential preparation for revising a set of projections of the ethnic populations of local areas in England, based on the 2001 Census, the results of which are reported via the website


Forecasting population and income distribution (Flanders: 2011 – 2031)
Ingrid Schockaert, SVR-Flemish Government and Interface Demography-VUB, Rembert Deblander, ECRU-UCL and CES-KuLeuven, André Decoster, CES-Kuleuven, Patrick Deboosere, Interface Demography-VUB

The aim of the paper is to forecast income inequality due to population change in Flanders between 2011 and 2031. Using census and register data, we make two multi-state population projections (Lipro-projections): one projection by age, sex and household position and one by age, sex and educational attainment. These projections are used to reweigh EU-SILC 2008 up to 2031. The reweighted data are calibrated to obtain household weights congruent with the population composition of both projections. We first present the application of the Lipro-household projection. We discuss (1) data issues, (2) the problem of consistency and (3) scenario setting of household formation dynamics. Next, we present the reweighting and calibration of the EU-SILC data. Thirdly, we demostrate that, when adapting wages, income from investment and pensions to capture a real yearly 1% economic growth, a clear increase of inequality between 2008 and 2031. Without such uprating of economic variables however, demographic change alone leads to a decrease in inequality and poverty. Finally, using quantile regression we assess how change in several population covariates such as age, household composition and educational level is related to change in the income distribution. We show that (1) different covariates affect different parts of the income distribution, (2) different covariates have opposite effects on inequality, (3) the impact of each covariate can be decomposed into a compositional and a return effect. The compositional effect refers to the effect of change in the prevalence of the covariate; the return effect refers to the change over time in the covariate-income relationship.


A simple projection of ethnic diversity
Ludi Simpson, University of Manchester

The population share within a local authority area changes approximately linearly for each group. A crude projection can be made by adding the previous decade's change in share to the most recent estimate. This crude approach does not provide an age structure, and will be much improved when a cohort component model uses each group's age structure to more sensitively provide future scenarios of local ethnic composition. However, it gives a useful indication of the magnitude of changes in diversity that are likely over the next 20 years. This approach is first evaluated by projecting 8 groups to 2011 from their change in each local authority 1991-2001. Increased immigration 2001-11 means that the projection underestimates the diversity of 2011. For this reason, the projection to 2021 and 2031 of 13 groups from change during 2001-11 is likely to overestimate future diversity, if immigration is less than it was during 2001-11. The projection is modified to prevent any group exceeding 100% or falling below 0% (5% for White British), and is scaled so that the shares in a local authority sum to 100%. The shares in each local authority district are used to calculate summary indicators of diversity for 2001 (actual), 2011 (actual), 2021 (projected) and 2031 (projected), including the Reciprocal Diversity Index, plurality, and the numbers of groups with more than 2% and 10% of the population. The results show increasing diversity on average, but that few local authorities are likely to exceed Newham's currently high diversity. The projected diversity will take many different forms, including areas where 'Other' is the largest group, but it does not include mono-ethnic majorities other than White British. The focus of attention may no longer be on the fate of 'White British' relative to any other group, but on a future which is not about groups as such, but about diversity itself.


Demographic projections for sub-national planning - Wednesday 10 September 9.00am

A sub-regional housing market model for England with endogenous migration and household formation: its role in assessing the adequacy of planned new housing Glen Bramley, Chris Leishman, David Watkins, Heriot-Watt University

Traditional extrapolative projections of population and households remain popular as a basis for planning housing provision, although the processes of migration and household formation are arguably influenced by, as well as influencing, economic and housing market conditions at sub-national scale. This paper describes an alternative approach which treats these processes as endogenous within models that seek to forecast key housing market outcomes at sub-regional level across England. This includes specific models to predict local migration flows and household formation rates conditional on economic conditions and planned land release. The impact of changes in the latter on the former are illustrated for different groups of sub-regions, also drawing out the significance of spatial interactions. Implications for policy and practice in the context of the contemporary English planning system are drawn out.


Incorporating recent trends in household formation into household projections for Scotland
Esther Roughsedge, Debbie Amabile, Household estimates and projections branch, Demography, National Records of Scotland

We are seeing substantial changes in the ways people are living. Household sizes have been falling for decades, but more recently this has slowed. There has been an increase in the number of young adults living with their parents over past ten years or so. These trends affect the overall number of households, and the types of household people live in. Household projections need to reflect trends in household formation as accurately as possible, as they are crucial for assessing housing need and demand, for use in the Planning system. In the past, household projections in Scotland were based on projections of household figures from the 1991 and 2001 Censuses, combined with the population projections. The 2011 Census results gave us the opportunity to include the impact of more recent changes in household formation in our projections. Following analysis of 2011 Census data, and consultation with our users, we concluded that projecting figures from three censuses would give a better reflection of the latest trends without losing sight of the longer term pattern. The method we have used combines two sets of projections – one from the 1991 and 2001 Censuses, and the other from the 2001 and 2011 Censuses. For the first time, we have used survey data to decide how much weight to place on each projection. This produces figures which should more accurately reflect the latest trends in household types. We will describe trends in household formation, the changes we made to our projections, and what the results show.


The urban and regional planning challenges of demographic change in Australia Jeremy Reynolds, Demographic Research, Department of Transport Planning and Local Infrastructure, Victoria

Australia is facing a threefold planning challenge created by its changing demography. First, the national population is projected to increase over the next forty years by double the amount of the previous forty years. Secondly, most of this increase is projected to be in a handful of cities. An already urban nation is therefore becoming more urban. Thirdly, reconstructing cities and regions will have to occur against a backdrop of population ageing and growing Government expenditure on health, aged care and pensions. A former Commonwealth Treasury Secretary described future population growth as the greatest challenge facing Australian since federation (in 1901). He added that it would require leadership in urban development. This presentation will focus on the planning challenges faced by the Australian state of Victoria by its changing demography. The State Government regularly prepares projections for use by its land use planning, infrastructure and service delivery agencies. There are constant tensions between demographic trends / projections and government policies / practices. Governments would like to ease growth pressures on Melbourne and ‘divert’ growth to non-metropolitan regions but this has proven difficult to bring about. Similarly, urban renewal challenges a planning system and service delivery models that have been oriented to greenfield development. The presentation will draw on the successful and unsuccessful experiences of preparing and using population projections in Government.


LDS: A Small Area Population Projection Model Using Ecological And Contextual Parameters
Guillaume Marois, Alain Bélanger, Institut national de la recherche scientifique

A common difficulty faced when projecting small area populations has to do with the implementation of regional feedback factors which highly influence local population growth. This problem is particularly important when simulating events that have a strong geographic component, such as internal migration (moves), and destination choice of external migrants. This paper is part of a larger research program that aims at developing a dynamic time-based microsimulation projection model of the Montreal metropolitan area population and its 79 municipalities, by age, sex, and language spoken most ofen at home. This model, called LDS (Local Demographic Simulations), proposes an innovative treatment of migrations by taking into consideration future changes in local conditions when determining individual’s mobility and destination choice. First, internal movers are randomly identified based on each simulated individual characteristics. Then, based on five contextual variables which are important predictors of residential choice, a utility function, stratified according to the life cycle of each individual, is estimated to parameterize the microsimulation module on location-choice. These determinants are (1) the presence of a highway, (2) the distance from the downtown area, (3) the current size of the population, (4) the number of new housings and (5) the proportion of francophones. External migrants are also added to the model each year and their place of residence is allocated following a similar function. Parameters are estimated using conditional logistic regressions. The municipality of destination is determined probabilistically according to these utility functions. The time horizon of the projection is from 2006 to 2031. Results of the model are validated and calibrated using a pre-simulation for the 2006-2011 period. This small area population projection model reveals some trends that could challenge public policies. One of them is the continuous urban sprawl implied mostly by the urban development plans used as assumptions.

Email: guillaume_marois@ucs.inrs.c