SAMPLE DESIGN

In the PES, private dwellings (houses, flats, etc.) and Discrete Communities are separately identified and sampled. In total, about 43,000 dwellings are expected to respond to the PES. This represents an increase of 20% on the size of the 2011 PES sample. The sample includes approximately 600 dwellings selected from 33 Discrete Communities in NSW, Qld, SA, WA and NT.

Table 1 shows the expected number of fully responding households for each state and territory. It is important to note that, due to sample loss and other non-response, the number of dwellings selected in the sample will be greater than shown below. Based on the results of other ABS surveys, the rate of sample loss is expected to vary between different components of the sample, which is taken into account.

Table 1: Expected number of fully responding households
	Total
	no.
New South Wales	9 400
Victoria	8 200
Queensland	7 900
South Australia	4 700
Western Australia	5 200
Tasmania	2 500
Northern Territory	3 400
Australian Capital Territory	1 700
Australia	43 000

This sample size is expected to be sufficient to provide estimates of the net overcount or net undercount rate with the following standard errors:

• persons at Australia level - standard error to be less than 0.2 percentage points;
• persons at state/territory level - standard error to be less than 0.6 percentage points, for state/territories larger than 500,000 people;
• persons at state/territory level - standard error to be less than 3,000 persons for states/territories smaller than 500,000 people*;
• Indigenous persons at Australia level - standard error to be less than 2.0 percentage points;
• males and females at Australia level - standard error less than 0.3 percentage points;
• persons at Australia level by 5 year age brackets - standard error less than 0.6 percentage points.

* NT and ACT had populations below 500,000 as at December 2014.

GENERAL POPULATION SAMPLE

For all general population areas other than Discrete Communities, private dwellings are selected through a stratified, multistage cluster design from the private dwelling framework of the ABS Population Survey Master Sample. The geographic area used for selection is comprised of mesh blocks from the Australian Statistical Geographical Standard (ASGS). Areas are grouped into strata based on locality, population density, remoteness, projected population growth and socio-economic characteristics. Each private dwelling within a stratum has the same probability of selection.

The 2011 PES used larger sampling fraction for areas with larger proportions of households with Aboriginal and Torres Strait Islander persons in residence, based on data from the 2006 Census. The aim of this was to improve estimates of Indigenous overcount and undercount. However, a review of the 2011 PES sample design showed that most of the improvement of Aboriginal and Torres Strait Islander estimates was achieved by increases to the overall sample size.

DISCRETE COMMUNITIES SAMPLE

For the purposes of sampling, Discrete Communities are grouped into 'sets' comprising main communities and their associated outstations. The selection of main communities is undertaken with probability proportional to the size of the set. The aim is to select a representative sample while also considering cost constraints, reasonable interviewer workloads and expected sample size.

For communities selected in the sample, selection of dwellings within the community is performed using systematic equal probability sampling.

A selection of outstations associated with each selected main community is also included in the sample. Each outstation has an equal chance of selection and, once selected, all dwellings within the outstation are enumerated. Occasionally, an outstation is not enumerated if it is difficult to access from the main community, resulting in a small loss of coverage.

The 2016 PES has selected a sample of Discrete Communities similar in size to that of 2011 PES in Qld, SA, WA and NT, with the addition of a small sample in NSW. For each state, the sampling fraction for Discrete Communities is 2 to 3 times the sampling fraction for the rest of the state.