How accurate are population forecasts?

Population issues have become highly political in Australia, dominating headlines whenever new data is released.  In 2019, the Australian Government responded to the growing policy pressures by establishing the Centre for Population.  One of its aims is to produce more accurate population forecasts.  As someone with a background in producing population forecasts, this claim intrigued me.  Population forecasts are not predictions, they are the outcome of a set of assumptions.  In this blog I will reflect on the issues and challenges associated with population forecasting, and what options might be available in the future.

How are population forecasts produced?

The key to understanding population forecasts lies in the transparency and robustness of the assumptions, as well as good base data.  Demographers do not sit at their desks with a crystal ball.  The numbers produced are the outcome of a set of assumptions about the components of demographic change (natural increase and net migration), as well as dwelling construction and household formation.  Will fertility rates increase?  Will life expectancy continue to increase?  What impact does government policy have on migration levels?  At smaller geographic levels, dwelling assumptions and local knowledge are important in determining where people will live.  To paraphrase an old film - if they do not build it, they will not come.

Migration, particularly from overseas, is the most volatile component of demographic change.  You only have to look at the fluctuations in the level of net overseas migration over time to appreciate this.  This volatility makes it very difficult to make accurate assumptions beyond the short term - and even then the rapidity of change in a global economy can change migration patterns almost overnight.  Demographers compensate by making assumptions about long term trends rather than trying to anticipate peaks and troughs, but it remains a huge challenge to producing accurate forecasts.

In contrast, changes in birth and deaths rates occur over a longer period of time as they respond to different drivers.  In the past, fertility rates have been influenced by access to contraception and the increasing participation of women in higher education and the labour force.  Improvements in health and water sanitation have typically influenced mortality and is best illustrated by increasing life expectancy.  

Migration trends in Western Australia

Migration levels tend to respond to wider economic conditions.  An example of this is the mining boom in Western Australia, where the higher rates of growth in the late 2000s and early 2010s were heavily influenced by higher levels of net migration from overseas and interstate.  Quite simply, people moved to where the high paying mining jobs were created.  Once the job market contracted around 2014, so too did population growth. Once again, migration played an important role here, as people moved out of Western Australia.

These trends are clearly shown in the chart below.  The gain from net overseas migration to Western Australia was in excess of 20,000 pa from 2006 to 2013 - a level only achieved once before (1988) in the last 40 years.  At the same time, there was a net gain of population through interstate migration.  However, since 2014 the population gain through migration has declined sharply.  Western Australia has been losing population interstate since 2014, and the level of net overseas migration is less than one-third of the peak figure of 49,970 recorded in 2012.  However, the downward trend appears to have reversed, as the numbers recorded in 2018 were slightly higher than in 2016 and 2017.  

On the other hand, population change through natural increase (births minus deaths) has been relatively stable since 1982.  The level of natural increase has shown a slight upward trend since the mid-2000s, mainly due to an increase in the number of births while deaths stayed relatively stable.  However this trend has moderated slightly since 2016 and possibly relates to the higher levels of migration out of Western Australia.

How accurate are population forecasts?

Surprisingly, there has been little in the way of published research examining the accuracy of population forecasts.  Tom Wilson has been active in this space, producing several papers in recent years - click here for an example.  Overall, forecasts have been shown to be less accurate in the long term, and smaller populations also increase the possibility of greater errors.

Wilson has also been instrumental in the development of the "shelf life" concept of a population forecast.  This relates to its useability and level of required accuracy.  In my experience, short term forecasting (up to five years) is relatively straightforward, because most of the known elements are already apparent, eg building approvals, shovel ready projects, incorporation of the latest population data.  The medium term (10-15 years) can somewhat be guided by assumptions around births and deaths and household formation, as these don't change dramatically in a short space of time.  However, interstate and overseas migration levels become harder to determine due to their volatility.  This becomes more challenging in the long term, as do trends around fertility, mortality, housing construction and household formation.  For example, many forecasts produced in the 1980s and 1990s did not anticipate the slight increase in the fertility rate that occurred in the mid-2000s, resulting in an underestimate of the number of young children.

So the question remains - are population forecasts wrong, or are the drivers of change moving so quickly that demographers find it challenging to keep up with evolving trends?  There are a vast array of practitioners who rely on good population forecasts for planning purposes.  Long term forecasts are a vital input into the feasibility of major infrastructure projects.  What is the way forward?

Some potential solutions might include:

• more regular forecasts.  Some of the fastest growing areas in Australia need frequent updates to the data otherwise existing forecasts become increasingly unusable.  The Victorian State Government plans to produce Victoria in Future on an annual basis in order to satisfy the demands from urban fringe councils, water authorities, and other users.  Frequent updates have the advantage of incorporating the latest data as it becomes available from the ABS as well as satisfying the demand for more timely data.
• the use of big data.  As big data becomes more accessible, the quality improves and coding becomes more sophisticated, it raises the potential to use new data sources as the evidence base for assumptions.  Big data has already been used to estimate university day-time populations.  Who knows - demographers may one day be able to produce real-time population forecasts!  
• better data.  The ABS is continually making improvements to the accuracy of their population estimates.  For example, the change to the component method for calculating ERPs is a major step forward as it considers demographic events in a more practical way.  The real test will be after the 2021 Census is conducted and the ERPs are rebased and matched to the previous estimates - watch this space.  
• scenario forecasting.  Population forecasts based on a scenario are especially useful for smaller areas such as LGAs.  For example, a forecast that considers the development or rezoning of a controversial site.  At the macro level, forecasts that include a zero net gain through overseas migration would be an interesting exercise.
• production of high, medium and low series.  I have a love-hate relationship with these.  On the one hand, they allow users to see the outcome of different assumptions made, but they also allow those same users to "cherry-pick" the series that best suits their needs.  They are also confusing to many users as they require a better understanding of the relationship between the components of population change.  On the other hand, they can be useful for providing a range of population outcomes that can be used for planning purposes.

In the end, demographers can only make assumptions based on what data is available to them at the time forecasts are produced.  The key is not to treat the forecasts as a given, rather consider them as a possible future based on the assumptions made.  Will the Centre for Population produce more accurate forecasts?  Time will tell, but it's an ambitious goal.

Summary

Population forecasts are a tricky beast.  They can be confronting and controversial, but they are critical for a vast range of planning purposes.  The key to understanding population forecasts is the transparency and robustness of the assumptions.  Demographers can only prepare assumptions based on the data and knowledge available at the time they're produced.  The biggest challenge is accounting for the volatility of migration over time, which has been demonstrated here using Western Australia as an example.