7 Classifications and coding

Classifications and standards

Using standards and classifications increases the value of survey data, as outlined in the Official Statistics System Statement of Principles (Statisphere 2006: 8). A number of official classifications and standards were used in the 2006 NZCASS, including:

• Statistical Standard for Ethnicity 2005

• Standard Classification of Ethnicity (2 digit, with one addition; see below)

• Statistical Standard for Occupation 2002

• New Zealand Standard Classification of Occupations 1995 (4 digit)

• New Zealand Socio-economic Index

• Statistical Standard for Marital Status 1999 (using a combined legal and social marital status classification)

• New Zealand Index of Deprivation – NZDep 2001 (not official, but independent and widely used)

• The MOSAIC geodemographic classification (not official, but independent and widely used)

• Statistical Standard for Usual Residence 1999

• New Zealand Standard Areas Classification (for Territorial Authorities, used to define broad regions, and Urban Areas)

• All of these except MOSAIC (PMP Micromarketing, 2006) and NZDep (Salmond and Crampton, 2002) are documented by Statistics New Zealand on their website (www.stats.govt.nz).

Coding

Information given in relation to the main income earner’s occupation had to be coded for socio-economic classification. For the 2006 survey, socio-economic classification was done using the New Zealand Socio-economic Index (or NZSEI). This was a two-stage process. First, ACNielsen coders translated the open-ended responses to questions about occupation and main tasks in the occupation (questions Q157-Q160) into the New Zealand Standard Classification of Occupations 1990 (Statistics New Zealand, 2001), following the Statistical Standard for Occupation 2002. Second, these codes were converted into NZSEI scores using a concordance available from Statistics New Zealand. NZSEI scores range from 10 to 90, and were collapsed into six groups for tabulation purposes.

Other coding completed by ACNielsen was:

• Backcoding from ‘Other Specify’ comments into the existing code frames (i.e. correcting interviewer errors in code selection). Given that no new codes were created, we did not create extra coding documentation for this. Instead, coders relied on the existing code frame and interviewer instructions. (Only a few cases were unclear based on these documents; they were referred to researchers for decisions.)

• Coding of ethnicity responses other than the main groups presented (as in the Census) into level 2 of the Statistical Standard for Ethnicity (2005). In addition, a separate code was assigned to responses of New Zealander and/or Kiwi, even though the official standard only includes this code at level 4. These were combined with European responses in the Key Findings and Topic Reports.

In all cases, open-ended responses were typed directly into the laptop, and subsequently coded by coding staff in the office rather than by interviewers in the field.

Classifying incidents

The classification of offences was a separate exercise done after the end of fieldwork. In essence, incidents needed to meet legal definitions of criminal behaviour to be counted. The coding principles were agreed with New Zealand Police. Full details of the offence coding procedures are given by Cameron and Mayhew (2006); some highlights relevant for analysis and interpretation are reviewed below.

Classification of offences is based on the responses in the Victim Forms and Self-Completions, including a short description in the respondent’s own words. It also draws to some extent on which screener question was answered affirmatively.

There were some changes made to the coding exercise for the 2006 survey compared to the two previous ones. The main ones were:

• Greater reliance on information from the Victim Form. In 1996 and 2001, the offence code decision was made principally on which screener question the offence came from, and the offence description given by the victim in the Victim Form. In 2006, a substantial amount of other data from the Victim Form itself was inspected. The number of offences that could not be coded in 2006 was smaller than in the two previous surveys where many incidents were not coded because of doubts about their ‘relevance’ (see above), or which offence code they should attract. In 2001, 22% of incidents for which a Victim Form was completed were not coded either because they were ‘out of scope’, or an offence code could not be allocated.[19] In 2006, the figure was 15%.

• Double offence codes. In New Zealand, an incident of crime can give rise to more than one offence being coded by the police if two dissimilar or unrelated offences occurred at the same time. (If one offence, say damage to a car, is simply part of another, say, theft from a car, or the means of committing it, then only the major offence is coded.) There was discussion with New Zealand Police on the issue of double coding at the start of the coding exercise. The result of this was that more crime incidents from the 2006 survey were given two offence codes (6%) than in 2001 (1%). The increase appeared to relate to burglaries; if combinations of burglary with vehicle offences are excluded, the 2006 proportion also drops to 1%. Estimated risks of victimisation take account of both offences, except for combinations of burglary with bicycle theft or vehicle offences, which are treated as burglaries only. Thus, most of the second codes were discounted in the calculation of victimisation rates from the 2006 survey, resulting in little difference from 2001.

In order to make better comparisons between the results from the 2006 and 2001 surveys, two approaches were taken with respect to coding 2006 offences. They were coded first on the basis of the best information in the Victim Form. They were also coded as it was judged they would have been coded using the procedures used in the 2001 survey. As would be expected, more 2006 incidents were deemed ‘out of scope’ when 2001 coding procedures were followed.[20] This had the most noticeable effect with regard to incidents reported in the Self-Completion components. The comparison of rates of victimisation in 2005 compared to those in 2000 (from the 2001 survey) in Chapter 2 of the Key Findings report draws on the ‘2001-like’ coding approach.

Table 7.1 shows the 2006 survey offence codes alongside the corresponding codes from the 1996 and 2001 surveys. It also shows whether they attract the personal (P) or household (H) weight, which depends on the nature of the offence. A few changes to the weights between 2001 and 2006 should be noted. Offences excluded from the survey count are also shown.

Table 7.1 Offence codes

Footnotes

19 The figure in 1996 was 9%, although this may not be comparable due to the very different incident selection framework used in that survey, and the large number of victim forms required by the design that were missing in the 1996 survey.

20 22% of 2006 incidents were coded as out of scope using 2001 procedures, as against 15% under 2006 coding procedures. The most marked effect was in relation to incidents from the Self-Completions. Using 2001 coding procedures on the 2006 survey data, 24% of these incidents were not coded, whereas it was a much lower 7% under 2006 procedures. In the 2001 survey, just over 30% of incidents from the first and second Self-Completion components were not coded.

21 The legal definition of burglary changed between the 2001 and 2006 surveys. It now covers thefts from enclosed yards, which may have been included under Code 13 before, and does not require forced entry. Thus, a new code has been added to handle the new definition. The old code has been retained to allow comparisons with previous surveys.

22 Bicycle theft was treated as a personal offence in 1996 and 2001, so (in theory) thefts of bicycles belonging to other household members were excluded. Also many bicycles were stolen from outside the house, and should in principle have been given Code 13, but were coded as bicycle theft. In 2006, bicycle theft is to be treated as a household offence.

23 In previous surveys, Code 20 covered wilful damage in general. This included damage to both household and personal property in 2001. Damage to motor vehicles was also included in Code 20 in 1996.

24 ‘Not relevant’ codes were only added to the 1996 data during analysis of 2001 survey data. Prior to that, these incidents were simply omitted from the dataset.