How often should population projections be updated?

Different producers of population projections publish updated projections at quite different frequencies. For example, the Australian Bureau of Statistics publishes projections every 5 years, while the United Nations Population Division releases a new set of World Population Prospects every 2 years. The Centre for Population produces projection updates every year in its annual Population Statement (as well as separate projections for the Budget). But how often should population projections be updated, and specifically how often should State and local area population projections in Australia be updated? It’s not an easy question to answer.

How often projections should be revised depends on several factors, including:

• user needs,

• demographic data updates,

• major changes in demographic trends or policy, and

• user perceptions of frequent or infrequent projection revisions.

How often they are revised in practice depends on considerations such as:

• the budget and resources available to produce the projections,

• internal organisational needs and the value placed on projections,

• the complexity and speed of the projections data release approval process, and

• political considerations.

User needs vary widely. Users who take population projections as inputs to their own models of service demand which are run annually would probably appreciate annual updates. Users who need population numbers might not need revisions as often as those who make use of projected population growth or births, deaths and migration. Over the short-run, relative forecast errors in population numbers are usually smaller than errors in projected components. Users who are interested in broad-brush long-run population futures will probably not need annual updates. Users who focus on childhood age populations will need projection updates more often than those focusing on older age populations because the former tends to experience high forecast errors and the latter lower errors. Not surprisingly, a recent survey of projection users revealed a range of views on preferred projection update frequency.

It’s generally a good idea to produce a new set of projections when a major update to population data is released. The most important of these is the re-basing of Estimated Resident Populations (ERPs) following the five-yearly census. The re-based ERP for the census year is based on the new census count (with allowances for undercount, people temporarily overseas on census night, and a small timing shift from the census date to mid-year). ERPs for the previous five years back to the last census are also re-based to correct for ERPs which ‘drifted off course’ during the intercensal period. This is mostly due to approximations in the migration estimates. In recent years, preliminary re-based national and State/Territory ERPs have been released in the middle of the year following a census, while preliminary re-based sub-state ERPs have been published a little later in August or September. One year later, final re-based ERPs become available.

Major changes in demographic trends or policy (especially those related to immigration and land-use) which result in demographic outcomes which deviate markedly from existing projection assumptions also indicate a need to revise projections. In recent years, the decline in the fertility rate has been greater than most demographers expected. Overseas migration has proved stubbornly difficult to forecast. Net overseas migration (NOM) numbers plunged when the national border was closed during COVID, and then briefly surged to record highs when the border re-opened. And given the current political debate about lowering NOM, a re-think of NOM projection assumptions may be required if a new more restrictive migration policy comes into effect, and when there is a change of government and policy at a future federal election.

User perceptions of projections are also important. For users to retain confidence in projections, they need to seem plausible and, over the short-run, look like they are heading in the right direction given the latest published demographic data. But if they are revised annually (or more often), there is the risk of giving the impression that regular ‘corrections’ are being made to fix problems with previous sets of projections. Or that the projection incorporates an over-reaction to the published data for the most recent year (recency bias). However, if projections are revised too infrequently, there is also a risk that they will appear of date if the discrepancy between projections and actual demographic outcomes becomes substantial. I think it is also generally sensible not to release a set of projections in a census year shortly after the census has been undertaken. It could be misinterpreted as having included the latest data based on the census. It won’t do, because that data isn’t available until the following year.

So how often should State and local population projections in Australia be updated? Every 5 years is too infrequent for many users. My suggestion is 3 revisions in each 5-year intercensal period. It is largely based on the timing of data releases, and it involves producing:

• a new set of projections when preliminary re-based sub-state ERPs by age and sex are released in August or September in the year after a census,

• a minor revision to those projections a year later when final rebased sub-state ERPs by age and sex are released (with updated populations and possibly fine-tuned short-term assumptions, but without changes to long-run assumptions), and

• a new set of projections two years later to account for possible trend and/or policy changes.

This approach could be useful most of the time. However, if major changes to demographic trends or policies occur, then it could be sensible to consider deviating from this plan and prepare a revised set of projections.