A3 ACNielsen area sampling frame

In nearly all cases, house-to-house interviewing by ACNielsen has required national coverage. This includes many large-scale descriptive surveys undertaken for a range of government and quasi-government agencies, SOEs and universities on a range of topics with social and economic policy implications, as well as the industry-commissioned media use surveys (newspaper and magazine readership, television viewing, and radio listening) which ACNielsen has conducted for many years.

From all of these types of organisations, there has been the requirement for statistically reliable data. This led ACNielsen to some years ago develop and standardise a reliable and replicable national sampling frame that could be linked closely with official population statistics. This was meant for two reasons:

• To project the survey data up to total population (i.e., report data in terms of thousands and tens and hundreds of thousands of persons) rather than reporting simply in percentage terms.

• To be able to break survey data down to relatively small and concise geographical areas.

ACNielsen’s national sampling frame – the "Nielsen Area Unit" ("NAU") – uses geographical units that are intermediate in size between the two smallest types of unit defined by Statistics NZ.

The smallest of Statistics NZ geographical units are "Meshblocks". (These are used for Census data collection purposes, and are defined so that they can be amalgamated into most other "official" territorial geographical constructs.[44] They can also be defined in terms of a density of population continuum, which consists of Main Urban Areas, Secondary Urban Areas, Other Urban Areas, and Rural Areas). Statistics NZ has split the country into approximately 38,000 Meshblocks, which contain on average around 100 residents living in about 35 dwellings. This is too small a unit for survey research purposes, because random selection procedures could mean that a particular Meshblock might be selected a number of times and be "worked out" in a relatively short time, especially where the sample frame is in continuous use for many surveys.

The next smallest Statistics NZ geographical construct is called the Area Unit. There are approximately 1,850 of these. They contain on average an amalgamation of around 20 Meshblocks, and so usually cover around 2,000 residents living in about 700 dwellings. However, 1,700 Area Units is insufficient to fulfil ACNielsen’s practical requirement to be able to have a large number of areas in use at any one time.

The Nielsen Area Unit (NAU) frame combines on average around seven Meshblocks, with an average population of around 700 living in around 230 dwellings. These units have been created in such a way that they will always conform, in combination, with Statistics NZ Area Unit boundaries. They can therefore be used as building blocks for all of the territorial geographical and density of population constructs described above.

All NAU areas have been mapped using Statistics NZ Meshblock maps that show individual property boundaries. The first time an NAU is selected for a survey, a particular property is designated as the interviewer’s "start point", the first dwelling to be called upon. This original start point is selected by using a grid pattern overlay template and a random number selection procedure that produces grid coordinates. Interviewers record all houses called upon on a sheet attached to the inside of the map folder, and the next time the map is used for a survey the start point is the house adjacent to the last house at which contact was made on the previous survey.

Multi-Stage Sampling

Multi-Stage Sampling meets the practical need to impose a staged, sequential selection procedure on the process that interviewers use in recruiting the sample. It is needed as there is no comprehensive list of all New Zealanders that can be used to select a simple random sample from.

Providing a staged, sequential and systematic guideline serves two purposes. In addition to giving interviewers a practical framework for selecting respondents, it also preserves the central tenet of probability sampling necessary to allow calculation of sampling error margins. That is, it ensures all members of the population retain a known chance of being included in the sample.

The ACNielsen procedure has three stages. It imposes random or quasi-random guidelines for:

(a) selecting area units within stratification grid cells (strata)

(b) selecting streets and dwellings within area units, and

(c) selecting respondents within dwellings.

Clustered Sampling

Clustered sampling is where a number of dwellings are selected to be sampled in a patterned way around a single "start point", where a start point is defined as a numbered house on a named street, or in rural areas as a systematically-defined dwelling on a known or named road. The purpose of cluster interviewing is to reduce cost through efficiency gains in interviewers’ travel time and mileage.

Within ACNielsen, clusters are generally defined in terms of the number of interviews, for example five, to be obtained around each start point. This, technically, is a quota, because it requires interviewers to keep calling on and replacing dwellings until they have obtained their quota of five interviews. This means an indeterminate (and possibly large) number of addresses are called upon, which can of course reduce overall response rate, and thereby increase non-response bias.

For the 2006 NZCASS, as with the previous crime survey, a cluster approach was employed providing interviewers with a number of dwelling addresses to be approached - a ‘fixed number of addresses’ approach. The cluster was completed when a "final outcome" was achieved for each address. In most cases a final outcome was an interview, a refusal, or a non-contact – after the specified number of calls back to the address has been achieved. (The number of callbacks is dealt with elsewhere.) The ‘fixed number of addresses’ clustering approach differs from the alternative of ‘quota clusters’, which centres on achieving a fixed number of interviews. The latter approach was used in the 1996 survey. It attracted some criticism in the ABS methodological review insofar as replacement households might differ in key victimisation respects from households that had refused an interview. The ‘fixed number of addresses’ approach means that an indeterminate number of interviews are achieved in each cluster, but it also gives the possibility of a better overall response rate. This is because it reduces the impact on the overall response rate of low-response area units, or of interviewers who are perhaps less persuasive in obtaining respondent cooperation.

Footnote

44 These include Area Units, local authority Wards, Territorial Local Authorities (Cities and Districts), and Regional Council Regions.