The case of the missing 294,000 missing Australians

When the Australian Bureau of Statistics recently published its Preliminary Rebased 2011 Estimated Resident Population (ERP), much of the media attention was on the 294,000-person difference in the ABS’s earlier estimates of the population (based on a 2006 census base) and the revised number based on the results of the 2011 census. This difference between the previous estimate and the new rebased number is known as the intercensal error.

However, this does not mean that the growth between 2006 and 2011 was 294,000 persons fewer than the ABS originally indicated. A change by the ABS to a more accurate methodology in undertaking its post-census survey to clarify unrecorded census data suggests that the method used in the previous census overstated the population by about 247,000 persons at the 2006 census. As a result, the actual magnitude of the intercensal error between 2006 and 2011 is likely to be much lower than that suggested by the 2006 and 2011 census counts.

For forecasters such as BIS Shrapnel, who use population growth data in estimating household formation and consequently the underlying demand for new dwellings, an accurate measure of the population is important in determining whether there is an underlying undersupply or oversupply of dwellings. This in turn drives our forecasts for new dwelling construction and prices.

At face value, the downward revision to the June 2011 population would imply much lower growth between 2006 and 2011, leading to lower demand for dwellings and therefore a lower undersupply (or higher oversupply) of dwellings. However, there is more to the numbers than meets the eye, and a description can be found in the technical notes section of the ABS’s Australian Demographic Statistics publication (Cat 3101.0).

In calculating the ERP, the ABS starts with the census count at the person’s location on census night and then places them back to their usual address. The ABS also undertakes a post-enumeration survey (PES) to determine the level at which population was missed or counted more than once. Typically, more people are missed than double counted, so this is called the undercount and is added to the population enumerated on census night.

The technical notes highlight that due to a more effective method (automated data linking or ADL) in conducting the 2011 census PES, the net undercount was 374,500 persons in the 2011 census. This compares with an undercount of 549,600 persons at the 2006 census, suggesting that the previous method overstated the undercount in the 2006 census, and by extension overstated the ERP in 2006.

A statistical impact study by the ABS took a sample of PES records of the 2011 census and processed them using the 2006 method in order to compare the impact on the undercount using the 2011 ADL method. While there were some other changes between the 2006 and 2011 census that also made a minor difference, the result suggested that the undercount measured by the 2011 ADL method was around 247,000 persons (subject to the standard error of the sample) below that using the 2006 method.

This would indicate that the impact of the new ADL method accounts for around 247,000 persons of the 294,000-person intercensal error, and that the actual intercensal error assuming a consistent PES methodology is in the order of 47,000 persons. Consequently, rather than being out by 294,000 persons in the estimate of population growth between 2006 and 2011, the error is only 47,000 persons, with the other 247,000 being accounted for by a lower actual figure in 2006 due to the overestimation of the census undercount.