City of Tea Tree Gully
The information presented in the tables in profile.id is based on detailed tables produced by the Australian Bureau of Statistics at the Local Government Area level, and at the Statistical Area Level 1 (SA1) level for suburbs and small areas in 2016 and 2011 (Census Collection District (CD) for prior Census years).
The Australian Bureau of Statistics (ABS) adjusts information it provides in tables to preserve the confidentiality of individuals. All cells are slightly adjusted to prevent any identification of personal details. Methodologies for doing this changed between 2001 and 2006, and then changed again from 2011 to 2016.
Data tables released prior to the 2006 Census had small numbers (values of 1 or 2) randomly adjusted to either 0 or 3 by the ABS. As tables are randomly adjusted independently of each other, totals differ slightly across tables with the same population. The effect of randomisation is increased with the aggregation of CDs into suburbs and localities.
For the 2006 and 2011 Census, a different method called “perturbation with additivity” was used. All figures included within any table may be randomly adjusted by a small amount. These adjustments result in small introduced random errors. This method was introduced, so that not only could individuals not be directly identified in the data, but “differencing” could not be employed to derive individual characteristics. These random adjustments were balanced out across the table, so the totals were consistent across the categories. Ie. A random decrease in one category was matched by a random increase in another, to balance out the table.
For the 2016 Census, the ABS changed this method again. 2016 Census data use “perturbation without additivity”. Adjustments to all cells in a table are possible, but small cells are randomly “suppressed” or changed to zero. This has the effect of always reducing the table total when derived by adding up rows and columns. Whereas prior Censuses had table totals which could be slightly higher or lower than the actual total due to random adjustment, the 2016 Census table total will always be equal to or lower than the actual unadjusted total, due to this suppression. This has a significant effect on tables with a large number of categories, such as Birthplace and Language, and geographic areas built from smaller geographic areas, such as small areas on this site, which are built from SA1 units.
While the ABS now publish table totals separately at the bottom of each table, which are closer to the “true” population, these can’t be used on profile.id, as the percentages derived from them will add to less than 100, and could not be compared to earlier years.
The effect of this on the profile.id site is that summing the population of small areas will give a number somewhat lower than the total population for the Local Government Area. How much lower depends on the complexity of the tables involved. For most topics, the difference is only 1-2% on average, however, for some areas, and for some of the more complex topics, the difference could be as much as 10%.
Please note that this issue occurs ONLY with 2016 Census data. Differences between table total populations and small areas are negligible in previous Census years.
Although the information value of the table as a whole is not impaired by permutation, care should be taken when interpreting very small numbers, since randomisation will affect the relative size of small numbers far more than larger numbers. The effect of the randomisation methodology also ensures that values of 1 and 2 do not appear in tables.
No reliance should be placed on small cells as they are impacted by adjustment, respondent and processing errors. Rather than adding up totals from small areas to derive LGA totals, please use the LGA total provided on the site. Small area totals may also differ compared to downloading the equivalent dataset directly from the ABS website, due to geographic differences, and differences in the level at which permutation is applied.
This level of compromise should not impact on decision makers making effective resource allocation and planning decisions.