Weighting is the process of adjusting the survey results so they represent the total population, rather than just the 6,943 respondents interviewed in 2014. This page explains the weighting used for the New Zealand Crime and Safety Survey (NZCASS).
A weight is a number that represents how many units the survey unit represents. For example, one respondent may represent 500 other people in New Zealand.
The way the NZCASS sample is designed means that different respondents have different chances of being selected and so their responses cannot be treated equally.
For example, NZCASS incorporates a Māori booster sample, which gives Māori a higher chance of being selected for the survey. If this was not adjusted for, the overall survey results would be biased towards the outcomes that are correlated with being Māori.
Weighting the data means people (or households or incidents) that are under-represented in the survey are given higher weight when calculating results, while people over-represented are given lower weight. The weighting process makes adjustments for non-response and to better align the data to independent population data.
There are three types of weights in NZCASS:
For each type of weight, there are 100 ‘replicate weights’, which we use to estimate sampling error. Using replicate weights is a well-established method to estimate the sampling error for complex surveys, such as NZCASS. We calculate replicate weights for a sub-sample of respondents using the same method as the main weight. See . for a description of this
We used a similar weighting process for the 2014 NZCASS as we used for the 2009 NZCASS. We had to make some changes to allow for changes in the sample design (see and we updated the population benchmark figures. )
All NZCASS analysis is weighted. We use different weights for different purposes in NZCASS analysis:
In some instances, household offences have been weighted using the individual (person) weights. For example, 24% of adults had been victimised once or more in 2013 by any offence. Note ‘any offence’ is made up of offences against a person (like assault) and against a household (like burglary). In this case, it’s more appropriate to use the smaller weight (the person weights). Person weighting this estimate means that the 24% of adults (rather than households) were victimised by either a personal crime or lived in a household that was a victim of a household crime.
Some estimates in NZCASS use mixed units. For example, NZCASS asks about fear of crime. We ask individuals about their personal perception and consider this ‘person unit’ information. Some demographic questions (such as household composition) relate to households and are ‘household unit’ information. When we look at fear of crime by household composition, we are mixing both person units and household units. For these types of mixed estimates, the commentary or footnotes describes whether the data represents adults (people), households or incidents.
The following section summarises the technical detail of how each set of weights were calculated. You can find more detail about how we calculated weights in the
To calculate weights for each household, first households were given an initial weight, which was the inverse probability of the household being selected in the survey. For example, if the household had a 1 in 300 chance of being selected in the survey, then the household is given an initial weight of 300.
These initial weights were then adjusted for non-response through a model that included:
The weights were then calibrated to align with independent population data (known as ‘benchmarks’) for the level of urbanisation. This process ensures the survey estimates match independent population data (Statistics NZ) for the estimated number of households in each urbanisation category, as at 30 June 2014.
We calculated person weights using a similar process to the household weights, with initial weights adjusted for non-response and then aligned with population benchmarks. However, the initial weight also took into account the selection of one person from the household and used population benchmarks of age, sex and ethnicity rather than urbanisation.
Incident weights were derived from person weights by dividing them by the selection probability for that incident. For example, if 10 incidents were reported for the first self-completion victim form, then the selection probability of that incident is 1 in 10.
This page was last updated: