2013 Post Enumeration Survey

Respondents in the PES are individually matched to census forms by comparing the information given by PES respondents with the information given on census forms. Measures of undercount and overcount are estimated from the results obtained in the matching procedure.

Two major changes from the 2006 PES are the increased sample size (15,000 dwellings compared to 11,000) and the introduction of automatic matching of records to supplement clerical matching between PES and census. The net undercount estimates produced from the PES are subject to both sampling and non-sampling errors. The enhanced sample size aimed to reduce the sampling error of coverage estimates, especially for population sub-groups, and the introduction of automatic matching significantly improved the quality of the estimated coverage. Reduced non-sampling error was also expected due to improvements in the quality of the matching, enumeration and interviewing practices compared with 2006.

Matching was done to determine if a PES respondent was counted in the census at the address at which they were interviewed in the PES or at each address where a census form may have been completed for them (this is referred to as the "search address"). This was achieved by comparing the information given by PES respondents with the information given on census forms. If an address given was different to the PES address, searching was carried out to locate the address.

The estimation methodology for the 2013 PES includes several improvements over that used in 2006, 2001 and 1996. The main change was to include a dwelling weighting step to adjust for dwelling-level non-response in the PES. This enables us to take into account the extra information we have about census attributes of dwellings found in the census.

A matching impact study (MIS) was undertaken on a subsample of PES records. This study evaluated the new method of automated electronic matching against the manual-only method used previously, in order to measure the effect of the change in method. The MIS verified that the 2013 PES results improved the quality of matching. The overall impact is a reduction in the net census undercount from the new method because we were able to locate more people who gave vague or inaccurate census-night addresses in the PES. In the MIS, the national net undercount rate was estimated to be 3.9 percent (+/- 0.6 percent), compared with 2.4 percent (+/- 0.5 percent) using the new method.

The sampling process was complex. The geographical framework of New Zealand consists of 46,637 meshblocks (a meshblock in urban areas is usually a residential area containing about 40 dwellings surrounded by streets; in rural areas a meshblock covers a much wider area because dwellings are sparsely spread). For sampling purposes , these meshblocks are aggregated into 21,813 primary sampling units (PSUs). To improve the sampling efficiency these PSUs are stratified into 127 groups (or strata) based on region, urban/rural mix, ethnic population, and other socioeconomic variables (income, employment status, population aged 65 years and over).

The major change for the 2013 PES is that the sample design was based on the General Social Survey (GSS). In 1996, 2001 and 2006 the PES was based on the sample design of Statistics NZ's Household Labour Force survey. The GSS used a more up-to-date Household Survey Frame, designed to make targeting of ethnic groups more effective.

The 2013 PES sample comprised 1,299 PSUs containing almost 15,000 dwellings (or about 0.9 percent of total permanent private dwellings in New Zealand).

Across the 127 strata, 1,199 PSUs were randomly selected for the GSS. All GSS PSUs were selected for the PES. Balancing of cost and accuracy requirements led to the decision to enumerate and sample an extra 100 'new' PSUs, together with oversampling of dwellings within a selection of PSUs. The 100 new PSUs were selected by randomly sampling 5 PSUs from each of 20 Household Survey Frame strata deemed to be the most 'difficult' to enumerate. This 'difficulty' was measured by creating a stratum-level difficulty index (DI).

Oversampling of dwellings within a PSU is achieved by selecting more than one panel of a selection of the 1,299 PSUs (including the 100 'new' PSUs) . Approximately a quarter (365) of the PSUs in the PES sample have part of a second panel selected and a small number (37) have a full second panel selected. The 365 PSUs with extra part panels were selected from the top twenty ' difficult ' strata on the 2008 Household Survey Frame. The 37 PSUs with two panels were targeted due to their high selection weight.

New Zealand. Outputs in two series:

1. Northern North Island, southern North Island, South Island, New Zealand
2. Auckland region, Wellington region, rest of North Island, Canterbury region, rest of South Island.