Stats NZ has a new website.

For new releases go to

www.stats.govt.nz

As we transition to our new site, you'll still find some Stats NZ information here on this archive site.

  • Share this page to Facebook
  • Share this page to Twitter
  • Share this page to Google+
Results

The results are divided into four categories:

  1. consistency with the standard for ethnicity
  2. coverage and missing data
  3. comparison of aggregate counts
  4. comparison of individual-level records between the census and administrative sources.

Consistency with the standard for ethnicity

The census collection of ethnicity is consistent with the statistical standards described in section 3. Most administrative sources also capture the same concept of ethnicity as cultural affiliation, and aim for self-identification where possible. Most, but not all, sources record multiple responses. On the whole, up to three ethnic groups are recorded, the minimum requirement of the standard.

The 2005 standard classification appears to be used by most administrative data sources, with all the larger sources capturing responses between level 4 and level 2. Some administrative sources use prioritised ethnicity for reporting – a practice that is not consistent with the standard.

The question asked for ethnicity differs widely across administrative data sources, and often differs within each administrative source depending on the mode of collection or the form used.

All collections, both census and administrative, rely on the respondents’ understanding of the concept of ethnicity. This may depend on the context, and may vary over time.

Because a person may change the ethnicity or ethnicities they identify with over time, the time reference of a source collection is important. The census measures ethnicity at a single point in time (census day), and official population estimates also measure ethnicity at a given reference date. In contrast, many administrative sources reflect information collected from individuals at different points in time. Birth registrations record ethnicity for a single event in a person’s life, while other sources may capture ethnicity at multiple times depending on contact with the agency. What is recorded may represent the latest value, or an agency may accumulate ethnicities over time. Administrative systems do not always report the date at which ethnicity was collected.

These differences in the timing of data collection introduce conceptual differences between the measurement of ethnicity used currently in the census and official population estimates, and any measures of ethnicity produced from administrative sources.

Table 1

 Key elements of ethnicity in the statistical standard as applied in the census and for administrative sources in the IDI
Consistency with statistical standard

 Census

 DIA births (from 1998)

 ACC

 Ministry of Education

 Ministry of Health

 MSD

 Claims

 Schools

 Tertiary

 NHI

 Benefits

Self-identified cultural affiliation?

 Y

 Y

 Y

 Y

 Y

 Y

 Y

Is multiple response recorded?

 Y

 Y

 Y

 Y

 Y

 Y

 N

Statistics NZ 2005 ethnic classification?

 Y

 Y

 Y

 Y

 Y

 Y

 N

What level of classification is used?

 Level 4

 Level 4

 Level 4

 Level 2

 Level 3

 Level 2

 Level 2 equivalent

Is the question consistent?

 Y

 Y

 N

 N

 N

 Y

 N

Key:
Y = consistent with statistical standard
N = inconsistent with statistical standard

Coverage and missing data

This section examines the availability of ethnic data in each administrative source and for all sources combined, by age and sex, and by ethnic group.

The census includes all age groups. The administrative sources in the IDI currently provide information about different age groups depending on the source. Individuals of all ages access the New Zealand health care system and have the potential to be captured in the Ministry of Health data. The MSD benefits source has low coverage, because only a small proportion of adults receive working-age benefits. The education system provides information about children from five years of age. Births data covers all children born in New Zealand since 1995, although we restrict our analysis to records since the digitisation in 1998.

Table 2 shows the coverage of each source, with the base population being the total linked Census-IDI population. The low coverage rates for education and births data are partly because data is only available for recent years.

Table 2

Coverage of ethnicity information in each main administrative source (linked Census-IDI data)
Data source

 % of linked IDI-Census population with ethnicity information

Birth registrations

 18

Ministry of Education (schools)

 27

Ministry of Education (tertiary)

 44

Ministry of Social Development

 32

ACC (claims)

 47

Health (NHI)

 98

Overall, 99 percent of individuals in the linked Census-IDI data have at least one ethnic code recorded in the IDI personal details table. However, the proportion of individuals with ethnicity information recorded varies with age. Figure 3 shows the percentage of individuals in the linked Census-IDI data who have no ethnicity recorded in any administrative source, by age and sex.

Figure 3
Graph, Percent missing ethnicity in combined administrative sources, by age and sex (linked Census-IDI data).

Some ethnicity information is available for more than 97 percent of individuals in the linked Census-IDI at every year of age. There are very few children missing ethnicity data because nearly all the children in the IDI spine come from recent birth records, which have very low rates of non-response. Education records are also good quality for recent years. From around age 19 to 35 there is an increase in missing ethnicity. There is a gradual increase in missing ethnicity from age 50 up for linked females, which is not seen in linked males.

Table 3 shows the number and percentage of respondents for each major ethnic group from the census that have no ethnicity information in the IDI.

Table 3

 Percent missing ethnicity in combined administrative sources, by ethnic group, (linked Census-IDI data)
 Ethnic group  Responses in the census, but missing in the IDI

 Number

 Percent

 European

 20,400

 0.69

 Māori

 2,800

 0.50

 Pacific peoples

 2,400

 0.85

 Asian

 12,100

 2.65

 MELAA

 1,200

 2.61

 Note: MELAA = Middle Eastern/Latin American/African

Overall, a very high proportion of individuals across all ethnic groups have ethnicity information from some administrative data in the IDI. The level of missing ethnicity data varies somewhat across ethnic groups. Māori, European, and Pacific peoples have the lowest rates of missing ethnicity, all under 1 percent. The Asian and MELAA (Middle Eastern/Latin American/African) groups have higher levels of missing data, at around 4 percent.

Comparing aggregate counts

We first compare total responses output for level 1 ethnic groups in the census and for each administrative source. In table 4 we show the ratio of total responses in the administrative source to the census. We only use records from the particular administrative source that have been linked to the census. A ratio close to 1 shows that similar results could be expected between census and the administrative source when estimating ethnicity for the same population group.

Table 4

 Comparison of total response counts between census and each administrative source
 Ethnic group  Administrative source to census ratio for total responses to ethnic group

 Births

 Health

 Tertiary

 Schools

 MSD

 ACC

 European

 0.98

 0.90

 0.93

 0.91

 0.90

 0.90

 Māori

 0.99

 0.79

 0.92

 0.89

 0.93

 0.74

 Pacific peoples

 1.01

 0.91

 0.98

 0.89

 0.91

 0.87

 Asian

 1.04

 0.89

 0.88

 0.92

 0.89

 0.79

 MELAA

 1.03

 1.07

 0.65

 1.42

 0.18

 0.74

 Other

 27.27

 3.67

 73.39

 14.06

 212.33

 126.56

 Note: MELAA = Middle Eastern/Latin American/African

In most cases ratios are less than one, indicating fewer people with each ethnic group in the administrative sources than in the linked census records. In all datasets, the ‘Other’ group is much more prevalent in the administrative data. This is most likely due to inconsistencies in coding. The ratios for MELAA ethnicities are also poor in several data sources, which again may indicate a problem with coding.

Birth registrations have ratios closest to 1, indicating a good agreement at an aggregate level. The other datasets all have at least one ethnicity that is problematic, though there does not appear to be a consistent pattern across ethnic groups.

Comparing individual-level responses

Because the 2013 Census records have been linked to the IDI, we can compare census responses to those recorded in administrative sources for each individual. This analysis uses an individual’s ethnicity in the census as a benchmark. What we are analysing is disagreements between census and administrative responses, which can generally be referred to as measurement errors.

In some cases the disagreements may not result from mistakes or misreporting. For example, an individual may not have identified as Māori when they filled out the forms for educational enrolment in 2002, but by the time of the 2013 Census, had come to identify as Māori.

Comparing ethnic group total responses

We first compare ethnicity using total responses for level 1 ethnic groups. We want to know, for each source and each ethnic group, how closely the ethnicity responses agree with census responses.

Table 5 shows one example of an administrative source compared with the census. Any individual with an ethnicity response in both sources must have either a ‘yes’ or a ‘no’ response in each source. If the various errors and conceptual difficulties described above did not exist, we would expect everyone to have either a ‘yes’ in both sources or a ‘no’ in both sources.

Table 5

 Comparison of Māori ethnic response in census and Ministry of Health data
   Census Māori ethnicity

 Yes

 No

 Total

 Number  %  Number  %  Number  %
 Health Māori ethnicity  Yes  411,800  11  28,900  1  440,700  12
 No  149,400  4  3,203,200  84  3,352,600  88
 Total    561,200  15  3,232,100  85  3,793,300  100

Table 5 shows that while 95 percent of people have the same responses in the two datasets, around 5 percent have different responses. This difference in classification has a greater impact on the smaller ethnic group. The under-reporting of Māori in health data compared with the census seen in the aggregate comparisons in table 4 is largely because over one-quarter of people with Māori ethnicity according to the census do not have Māori ethnicity in the Ministry of Health data. A smaller group do have Māori ethnicity in the health data, but not in the census. We can carry out the same type of analysis for each administrative source and ethnic group.

Each panel in figure 4 shows the percentage agreement between a source in the IDI and the census responses linked to that source.

The percentages in the left-hand panels are calculated as:

Equation for calculation of percentages in left-hand panels of figure 4.

And on the right-hand side:

Equation for calculation of percentages in right-hand panels of figure 4.

These two measures together provide agreement rates for administrative records compared with the census. A low percentage in the left-hand panel indicates a source that has failed to identify members of an ethnic group according to the census. A low percentage in the right-hand panel indicates a source in which a large proportion of people are incorrectly flagged as belonging to a certain ethnic group. The appendix includes tables for all combinations of source agency by ethnic group.

European and Asian ethnic groups show higher consistency than other level 1 ethnic groups. There is also a general trend for the percentages on the right-hand side to be higher than those on the left, typically over 90 percent. In other words, people with a given ethnic group in the IDI are likely to have it in the census as well. On the left-hand side, some datasets such as ACC and Health assign Māori ethnicity to less than 80 percent of census Māori.

One notable feature of the results is that the ‘Other’ ethnic group has very low rates of agreement. Only a very small number of the people flagged in the administrative data as having ‘Other’ ethnicity have ‘Other’ ethnicity in the census. There are further problems in certain datasets, such as ACC where almost everyone with MELAA ethnicity also has ‘Other’ ethnicity, making the MELAA ethnic profile results inconsistent with census. Because of these problems we have excluded the ‘Other’ ethnic group from the remaining analysis in this paper.

Figure 4

Graph, Agreement rates for individuals using total response for each administrative source and ethnic group.

Comparing combinations of responses

The results in the previous section considered total responses to each ethnicity independently. As people may belong to two or more ethnic groups, we now compare different combinations of ethnicities for a given individual. We use the term ‘ethnic profile’ to refer to these combinations of responses, for example ‘European and Māori’ or ‘Asian only’. These profiles are mutually exclusive (so an individual can only be in one of the categories).

Figure 5 shows the percentage of people who have the same ethnic profile as recorded in the census, for each administrative source. The denominator used to calculate the percentages is the number of people who have each ethnic profile in the census. Only census people linked to the dataset under consideration are counted in this figure. People with missing ethnicity data are also excluded from the counts.

The main feature of figure 5 is the much higher agreement rates for single ethnic profiles, than for those with combinations of two or more ethnicities. This is true for all data sources except birth registrations. Lower rates of people with multiple ethnicities in administrative sources appears to be the main reason for under-reporting of ethnic groups compared with the census.

Some of this disagreement may be because results of older coding practices are still included in the IDI data used here. Birth registrations show the highest consistency across all profiles, which is probably a result of the standardisation and quality controls introduced since the late 1990s.

The different ways of comparing ethnicity data for individuals shown in the tables above represent a spectrum of strictness of comparison. Asking an administrative source to provide an exact ethnic profile (figure 5) for an individual is a more difficult test than asking how many people have the same level 1 ethnic group as the census (figure 4).

We also need to remember that ethnicity in administrative data is reported by people at different points in time, compared with the single reference date of the census. We would expect those who strongly consider themselves to belong to only one ethnic group to be more consistent across sources than those with multiple ethnicities whose responses may be more affected by differences in time or context.

Figure 5 

Graph, Percent agreement with census ethnic profiles, by administrative source.

  • Share this page to Facebook
  • Share this page to Twitter
  • Share this page to Google+
Top
  • Share this page to Facebook
  • Share this page to Twitter
  • Share this page to Google+