Chapter 6 - Sampling
6.1 This chapter describes the sampling methods and approach to updating the sample in the WPI.
6.2 A two-stage sampling methodology is used to generate a sample of employee jobs for the WPI. The first stage selects a sample of businesses whilst in the second stage a sample of employee jobs are selected from within these businesses.
6.3 In the first stage of sampling, approximately 3,000 private and public sector businesses are selected from the ABS Business Register. These businesses are selected by stratifying the target population of businesses by state/territory, sector (private/public), industry division and business size and selecting a random sample from each strata. For a number of complex organisations further sampling is undertaken to simplify reporting arrangements.
6.4 In the second stage of sampling, businesses selected in the first stage are asked to select a sample of jobs from their payrolls using instructions provided by the ABS. The number of selections depends on the total number of jobs in the business, but is limited to a maximum of ten jobs per business. Approximately 18,000 jobs are selected.
6.5 The WPI uses probability sampling at both stages. Probability sampling means that each unit (i.e. business or job) has an equal chance of selection. Other price indexes produced by the ABS (including the Consumer Price Index and the Producer Price Index) are based on purposive samples which use selections based on knowledge or judgement of the index.
6.6 For effective and efficient ongoing index construction, it is important that a high proportion of the initial sample of businesses is retained in subsequent quarters, and that the same jobs within those businesses stay in the sample where possible. However, to ensure its continued relevance, the survey sample is refreshed annually. This ensures that the sample remains representative of the Australian labour market. Refreshing the sample also allows the ABS to control the length of time that businesses are included in the sample. By controlling the length of time businesses are included in the survey the ABS can ensure that the reporting burden placed on businesses is kept to a minimum.
6.7 The sample selection methodology used to refresh the sample of businesses in the survey is referred to as the 'Rotating Panel Methodology'. In this methodology, the survey population is initially broken into five equal population panels. The sample is refreshed from within only one population panel for each annual update. All new businesses commencing operations during the course of a year are included in the next population panel to be refreshed, and therefore have a chance of being selected at the next annual update, ensuring the index reflects changes to the population of businesses.
6.8 Between each annual refresh of the business sample, a small number of jobs will be lost from the survey sample because of the closure of some businesses. In addition, some jobs in continuing businesses will be replaced in the sample because of restructuring and other job changes. These small sample losses do not affect the index as the jobs lost from the sample are represented by other similar jobs in the sample through adjustments to the sampling weights.
6.9 In addition to the closure of businesses, if a job ceases or changes dramatically, it will become impractical to follow the same job over time. When this occurs the provider is asked to select another job to take its place using instructions provided. This ensures that the number of selected jobs remains representative of the Australian labour market.
Calculation of sampling weights
6.10 Sampling weights are calculated for each contributing job. Jobs in the WPI are assigned sample weights according to the number of similar units they represent within a strata. The procedure for assigning a sample weight to a job takes place at the business and job level. Total sample weight is determined by multiplying business and job weights together; the role of sample weight in contributing to aggregate data is described in Chapter 10. The following paragraphs describe how job and business weights are calculated.
6.11 Business (or unit) level weights are calculated at the first stage of sampling. They represent the number of similar businesses a unit (business) represents in a particular strata. These weights are calculated annually and applied in the December quarter, by dividing the number of businesses in a particular strata on the survey frame by the number of businesses in that strata in the survey sample.
6.12 A second stage of sampling arises with the occasional sub sampling of reporting splits within businesses. These splits are created to ease the burden of reporting for providers. Weights for this stage of sampling will depend on the method of sampling used.
6.13 The third and final stage of sample weighting occurs at the job level. These weights are calculated by dividing the agreed number of jobs reported on by a particular unit by the total number of persons employed by that business. As such, job weights describe the number of jobs in a business represented by a particular job. Total employment figures are updated in the September quarter of each year, and job weights are updated based on this in the December quarter.
6.14 The first, second and third stage weights are multiplied together to produce a final sample weight. This sample weight indicates how many jobs a sampled job represents in the Australian economy.