Quality information for 2018 Iwi Affiliation Estimated Counts

In April 2019 Stats NZ announced that it would not release official statistical counts of iwi from 2018 Census due to the number of records missing iwi affiliation data (iwi data) and the lack of alternative government data sources to fill the gaps.

Since then, Stats NZ and iwi data technicians from the Data Iwi Leaders Group (Data ILG) have looked at methods to reduce the impact of missing iwi data. This work has led to the release of the 2018 iwi estimates on Te Whata and on the Stats NZ website at Iwi affiliation (estimated counts): 2018

Strengths and limitations of 2018 iwi affiliation estimated counts provides an overview of the data and guidance for iwi data users about how to use the estimates.

This document provides more detailed information about the quality of the estimates to help users make an informed decision on whether the estimates will be fit for their purpose.

In this document we discuss:

• background information on how iwi data is collected in the census

• an overview of the methods used to reduce the impact of missing iwi data

• a summary of issues that affect the quality of data across all iwi

• a list of issues that may affect the usefulness of the data for specific iwi

• guidelines for using the data.

Key messages

To reduce the impact of the high level of missing iwi data in the 2018 Census we used a range of statistical methods to produce the 2018 iwi estimates. This included sourcing data from the 2013 Census and using iwi data from the parents of children.

For many purposes, the 2018 iwi estimates provide a more relevant and up-to-date picture than continued use of 2013 Census iwi data. These estimates are not official 2018 Census counts and the quality ranges from moderate to poor depending on how the data is used.

We have not provided a single 'overall’ quality rating for the 2018 iwi estimates as the quality differs based on the type of use. The table below describes how the data can be used and where caution is required.

Background information on iwi affiliation collected in the 2018 Census

When completing the 2018 Census questionnaire, respondents who indicated that they are of Māori descent or that they don’t know if they are of Māori descent were asked if they know the name(s) of their iwi. If they answered yes, they were then asked to specify up to five iwi and the region using free text fields. Respondents could specify ‘Don’t know’ as a valid response to the iwi question.

Stats NZ release iwi statistics based on the Iwi and iwi-related groups statistical classification V1.0.0. As part of census processing, text responses from respondents are mapped to a standard list of common responses, known as synonyms. These synonyms are then mapped to the iwi classification.

There are several differences between the 2013 and 2018 Census questionnaires. Two significant changes to the online questionnaire in 2018 were:

• only one name and one region textbox were initially displayed, and a respondent had to click a button labelled ‘add’ to add additional iwi; in the 2013 Census five sets of textboxes were presented upfront

• as-you-type functionality was added to help respondents provide valid, detailed responses; this functionality brought up common synonyms which matched the text entered by the respondent as they typed. The estimated iwi counts represent iwi for the usually resident Māori descent population in the 2018 Census. This is consistent with the population who was included in the 2013 Census iwi counts.

In the census context, respondents of Māori descent report their iwi with no requirement for detailed whakapapa information. This is different to many iwi registers which require a higher level of external validation to register.

Methods used to reduce the impact of people missing iwi data

Iwi data was missing for 29 percent of the 2018 Census Māori descent population. Without adjustments the resulting counts would not be fit for use as they would be too low and also would under-represent groups more likely to be missed by the census.

To improve the quality of the iwi estimates we used three steps:

1. Iwi data was sourced, if possible, from either the 2013 Census or from the iwi provided by the parents of children aged less than 15 years for people missing iwi data.

2. Additional iwi data was sourced from the 2013 Census or from the iwi provided by the parents of children aged less than 15 years for individuals where we had evidence that the design of the 2018 Census online form may have caused iwi to be under-reported.

3. A set of weights were developed to ‘rate-up’ the people we have iwi data for so that they represent the 2018 Census usually resident Māori descent population.

Iwi data collected in other administrative sources was not considered for this project due to the poorer quality of that data. If a respondent indicated that they did not know their iwi, we respected this as a valid response. A range of data sources were also used in the 2018 Census processing to reduce the level of missing data for the different characteristics (census variables) used in this release (for example, personal income).

This is the first time we have used this experimental methodology, which was developed in collaboration with Data ILG. We recognise that the use of alternative data sources and weighting are a significant change to the standard methods Stats NZ apply to census data but, without these adjustments, the iwi data from the 2018 Census would remain unusable. There is continuing work to address issues relating to Stats NZ’s social licence to use alternative data sources for census and wider statistical use purposes as identified by the 2018 External Data Quality Panel. Implementing the recommendations of the 2018 External Data Quality Panel provides updates on work Stats NZ is doing to address the recommendations.

In producing these estimates we have assumed:

• iwi affiliation for a person has not changed between 2013 and 2018

• children under 15 years will have the same iwi as their parents, based on genealogical connection

• the model used to generate the weights captures the remaining non-responding population.

There are many reasons a response in the 2013 Census may differ from a response in the 2018 Census. These include changes in the questionnaire and collection mode, changes in the classification and coding practices, as well as changes in the iwi that an individual identifies with or chooses to report in the census. To understand this, we looked at people who responded and provided iwi affiliation data in both the 2013 and 2018 Census. The chance of a person reporting an iwi in 2018 that they reported in 2013 was 78 percent. This rate was also assessed for each iwi. The rates for most fully coded iwi lie between 56 and 83 percent, with larger iwi tending to have higher consistency.

For children aged less than 15 years who were still missing iwi data we looked to use the iwi affiliations reported by their parents, if this was reported. The union of the iwi reported by both parents was used (for example, if one parent reported Ngāpuhi and the other parent reported Raukawa and Ngāti Hei, then the iwi listed for the child would be Ngāpuhi, Raukawa and Ngāti Hei).

There are a range of reasons why the iwi reported for a child in the census may not be the same as the iwi reported for their parents. First, in the census we are not able to distinguish between biological, whāngai, adopted, and stepchildren. Second, adults may complete the forms of children less thoroughly than their own form, or they may not know all of the iwi that the other parent affiliates with. To understand how often the iwi reported for a parent matched the iwi reported for a child we looked at 2018 Census records where a valid iwi response was provided for children and their parent(s). We found the iwi reported for a child

• exactly matched all the iwi of at least one parent half of the time

• matched or was a subset of the iwi reported for at least one parent 95% of the time.

Methodology for the 2018 iwi affiliation estimated counts contains further details on the methods we used to produce these estimates.

Quality of the 2018 iwi estimates

Data quality assurance for 2018 Census outlines the quality rating scale and quality assurance framework applied to the 2018 Census products to help assess and communicate the quality of different variables.

This framework uses three quality metrics that contribute to an overall quality rating:

• data sources and coverage

• consistency and coherence

• data quality.

Each metric is given a rating (for example, poor, moderate, or high) with the lowest rating used to describe the overall data quality of the data.

In this section we describe issues affecting the quality of the iwi estimates using these three metrics. In the following section we provide information on quality issues which may affect the quality of data specific for each iwi. We have not generated an overall quality rating as the quality of the iwi estimates depends on the type of estimate being produced, and the quality framework was not designed for a methodology that used weighting to adjust for missing data.

Data sources and coverage

The table below provides the breakdown of data sources across the total number of iwi affiliations. Respondents affiliating with more than one iwi are represented once for each iwi that they affiliate with.

We did not calculate an overall ‘data sources and coverage’ rating, as was done for the 2018 Census data quality framework, due to the use of the weighting adjustment. The data sources table above shows that 65 percent of the iwi affiliation estimate data came from 2018 Census responses, 19 percent from alternative data sources, and the remaining 16 percent of missing data has been represented through the weighting.

Iwi affiliation (estimated counts): 2018 data sources and quality by iwi provides information on the data sources used for each iwi. For most iwi the contribution of data from each data source ranged between:

• 58 and 69 percent for 2018 Census responses

• 14 and 22 percent for 2013 Census responses

• 2 and 4 percent for parental iwi data

• 14 and 17 percent for weighting.

Impact: The use of additional data sources and weighting is designed to reduce the risk of estimates differing from the true population. The most significant impact of high levels of missing data is it reduces the ability to estimate counts for groups with a high level of confidence. This is discussed in more detail in the section ’Guidelines for using the data’.

Consistency and coherence

To assess consistency and coherence, we compared the 2018 iwi estimates with expectations based on the 2013 Census counts. These expectations accounted for the change in the size of the Māori descent population between the 2013 Census and the 2018 Census, as well as changes to the iwi classification. For these comparisons, the 2018 iwi estimates are generally consistent with expectations. For this reason, the quality rating for consistency and coherence is Moderate.

However, there is a break in the time series due to large changes between the 2013 Census and 2018 Census.

• We improved the estimate of the 2018 Census Māori descent population in 2018 by making use of administrative enumeration. A similar approach was not used in the 2013 Census. As such, Māori descent counts increased by 30 percent, moving from 668,721 in the 2013 Census to 869,850 in the 2018 Census. Information by Variable: Māori Descent provides more information on the changes in the Māori descent population.

• Changes were made to the online collection of iwi data in the 2018 Census (see Background information above) and there was a large increase in the proportion of respondents completing the form online.

• In 2017 there was a substantial review of the iwi classification with 32 iwi being added to the classification. For more information on the change in classification see Iwi - New Zealand Standard Classification V6.0 to Iwi and iwi-related groups statistical classification V1.0

• In preparation for the 2018 Census some changes were made to the synonyms used to code responses to specific iwi.

Impact: For most iwi, the 2018 iwi estimates are between 20 and 40 percent larger than the 2013 Census counts. Because of these changes, caution is advised when comparing 2018 estimated iwi population counts and characteristics to previous census years. Any changes may be due to changes in the underlying population rather than changes in the outcome being compared (for example, education levels). The data used to produce the 2018 iwi estimates is available in Stats NZ’s Integrated data infrastructure (IDI). Stats NZ and Data ILG will be looking to produce a range of time series information using this information to improve iwi data users’ ability to understand changes that are occurring over time. An example of this is the Job Seeker Support series on Te Whata.

Data quality

To enable high-quality coding, Stats NZ developed a list of over 5,000 common text responses, known as synonyms. This list has been built up over multiple census collections and further refined in preparation for the 2018 Census. Iwi text responses are mapped to this list before being mapped to the iwi classification.

The increased use of the online census questionnaire and the as-you-type functionality in the 2018 Census has allowed for higher accuracy of coding and a larger volume of auto-coding compared with previous census collections. This can be seen in a reduction in the number of responses coded to iwi within the categories ‘Confederations and Waka, iwi not named’ and ‘Iwi named, region not known’.

There are quality issues affecting several aspects of the iwi data. This includes situations where there is not a clear choice for how to code a response based on the text responses and concerns for some iwi about how the as-you-type functionality performed. For these reasons data quality has been rated as moderate.

Impact: The ability to accurately code the responses to iwi listed in the classification can have a large impact on the resulting counts. Where we have concerns with coding and classification these have been noted in the iwi-specific quality section below.

Iwi-specific quality

For each iwi, we have provided information on the contribution of different data sources to the overall estimate, how consistent the 2018 estimate is with the 2013 Census, as well as information on any classification or coding changes. Iwi affiliation (estimated counts): 2018 data sources and quality by iwi contains this quality information for individual iwi. This information should be considered when using this data to inform decisions.

Below we have identified key issues by iwi that users should be aware of.

• Raukawa The classification name for 0303 Raukawa (Waikato) was changed from 0303 Ngāti Raukawa (Waikato) as part of the 2017 iwi classification review. 0303 Raukawa (Waikato) is around 4,000 people lower than expectations based on 2013 Census counts. There have been no significant changes to the coding for 0303 Raukawa (Waikato) since 2013 Census, so this does not account for the lower count. If respondents typed 'Ngāti Raukawa' when completing the 2018 Census form they were presented with the suggestion of 'Ngāti Raukawa (Horowhenua/Manawatū)' but not 'Raukawa (Waikato)'. This may have caused some people who affiliate with 0303 Raukawa (Waikato) to be classified under 1004 Ngāti Raukawa (Horowhenua/Manawatū) rather than 0303 Raukawa (Waikato). Comparisons between 2018 Census and 2013 Census proportions should be looked at with extra caution, as changes may be due to the change in who is now counted in 0303 Raukawa (Waikato).

• Ngāti Whātua The classification for Ngāti Whātua has changed as part of the 2017 iwi classification change. 0114 Ngāti Whātua o Ōrākei and 0113 Ngāti Whātua o Kaipara were included as part of 0109 Ngāti Whātua in 2013 Census, but they have now been listed separately. 0109 Ngāti Whātua was updated to 0109 Ngāti Whātua (not Ōrākei or Kaipara). The classification change has caused a decrease in counts in 0109 Ngāti Whātua (not Ōrākei or Kaipara) in 2018 Census compared to 0109 Ngāti Whātua in 2013 Census as some individuals are now counted in 0114 Ngāti Whātua o Ōrākei and 0113 Ngāti Whātua o Kaipara. The classification change meant that it was not possible to use data from 2013 Census for people missing iwi affiliation data in 2018 Census. A different estimation approach was used which increased the contribution made to the iwi count via the weighting. Comparisons between 2018 Census and 2013 Census proportions should be looked at with extra caution, as changes may be due to the change in who is now counted in 0109 Ngāti Whātua (not Ōrākei or Kaipara).

• Te Arawa The size of Te Arawa iwi grouping is in line with expectations based on 2013 Census. However, the partially coded category ‘2002 Te Arawa - iwi not named’ has shown a substantial increase compared to 2013 Census in both the online and paper responses. Subsequently, counts for more specifically coded iwi within the Te Arawa group have not increased as much as expected compared with the 2013 Census counts. Because of the size of ‘2002 Te Arawa - iwi not named’ within the Te Arawa group it is likely to be important to include this group in any analysis. The as-you-type functionality will continue to be improved as part of the development of the 2023 Census with the aim of facilitating more detailed coding.

• Ngāti Kahungunu The introduction of as-you-type suggestions in the online questionnaire supported a greater number of users to select specific Ngāti Kahungunu responses in 2018 Census. This has resulted in a large decrease in the residual category 2112 Ngāti Kahungunu, region not known. There have been large increases in the counts of 0708 Ngāti Kahungunu ki Tamatea, 0709 Ngāti Kahungunu ki Tamakinui a Rua, 0703 Ngāti Kahungunu ki Heretaunga and to a lesser extent, 0702 Ngāti Kahungunu ki Te Wairoa. 2112 Ngāti Kahungunu, region not known remains a large residual category and is likely to be important in analysis. Ngāti Kahungunu - iwi not named remains a large residual category with over 18,000 responses and is likely to be important in many analyses. As part of reviewing the synonyms used to code responses to specific iwi, the response “Ngāti Kahungunu - Hawkes Bay” was moved into “2112 Ngāti Kahungunu - region not known”. In previous census coding it had been coded to 0702 Ngāti Kahungunu ki Te Wairoa. This change affected the coding of around 1,900 records.

• 0105 Ngāpuhi ki Whaingaroa-Ngāti Kahu ki Whaingaroa Population estimates have increased substantially, moving from 2,050 in 2013 Census to 4,740 in 2018. No coding or questionnaire changes have been identified that may be contributing to the large increase. Comparisons between 2018 Census and 2013 Census proportions should be looked at with extra caution, as changes may be due to the change in who is now counted in 0105 Ngāpuhi ki Whaingaroa-Ngāti Kahu ki Whaingaroa.

• 0304 Waikato 2018 Census included approximately 17,000 responses where 2001 Tainui, iwi not named was specified as the iwi and Waikato as the region. These responses have been coded to 0304 Waikato to be consistent with the coding approach used in the 2013 Census.

• 0111 Te Uri-o-Hau Population estimates are lower than expectations based on 2013 Census, with 2018 Census estimates showing only a small increase compared to 2013 Census counts. No coding or questionnaire changes have been identified during census evaluation.

• 0802 Ngāti Maru (Taranaki) Population estimates are substantially higher than expectations based on 2013 Census. No coding or questionnaire changes have been identified that may be contributing to the large increase. Comparisons between 2018 Census and 2013 Census proportions should be looked at with extra caution, as changes may be due to the change in who is now counted in 0802 Ngāti Maru (Taranaki).

• 1006 Te Atiawa ki Whakarongotai Population estimates are lower than expected, with 2018 Census counts showing only a small increase compared to 2013 Census counts. No coding or questionnaire changes have been identified during census evaluation.

• 1111 Ngāti Toarangatira (Te Waipounamu/South Island), 1005 Ngāti Toarangatira (Te Whanganui-a-Tara/Wellington) and 2108 Ngāti Toa – region not known A large number of changes were made to the synonyms used to code responses to these iwi.

o 19 synonyms were added to 1111 Ngāti Toarangatira (Te Waipounamu/South Island).

o 8 synonyms were added to 1005 Ngāti Toarangatira (Te Whanganui-a-Tara/Wellington).

o 5 synonyms were moved from 1005 Ngāti Toarangatira (Te Whanganui-a-Tara/Wellington) to 2108 Ngāti Toa - region not known.

These changes are expected to improve the accuracy of coding. Population estimates for 1111 Ngāti Toarangatira (Te Waipounamu/South Island) are substantially higher than expectations based on the 2013 Census, moving from 321 in the 2013 Census to 909 in the 2018 Census. Comparisons between 2018 Census and 2013 Census proportions should be looked at with extra caution as changes may be due to the changes in who is counted in each iwi.

• 1112 Waitaha (Te Waipounamu/South Island) Population estimates are substantially higher than expectations based on 2013 Census, moving from 1,041 in the 2013 Census to 1,704 in the 2018 Census. No coding or questionnaire changes have been identified that may be contributing to the large increase. Comparisons between 2018 Census and 2013 Census proportions should be looked at with extra caution, as changes may be due to the change in who is now counted in 1112 Waitaha (Te Waipounamu/South Island).

Iwi added to the classification in 2017

For iwi that were added to the classification in 2017 we do not have a baseline count to compare the 2018 estimates against. This increases the risk for these iwi that there may be coding or classification issues which we could not identify.

For most iwi added to the classification in 2017, we were able to use 2013 Census data by recoding the 2013 Census text responses to the new classification. For some iwi, however, this approach was not effective and would have underestimated their size. This occurred when the synonyms associated with that iwi were not seen in 2013 Census responses (or were seen very infrequently). For these iwi, we omitted the step that looked to use 2013 Census data and increased the contribution made to the iwi total via the weighting step.

These iwi are:

• 0109 Ngāti Whātua (not Ōrākei or Kaipara)

• 0113 Ngāti Whātua o Kaipara

• 0114 Ngāti Whātua o Ōrākei

• 0117 Te Paatu

• 0713 Maungaharuru Tangitū

• 0716 Te Hika o Pāpāuma

• 0905 Ngāti Whitikaupeka (Rangitīkei)

• 0906 Ngāi Te Ohuake (Rangitīkei)

• 0907 Ngāti Tamakōpiri (Rangitīkei)

• 0910 Tamahaki (Ruapehu, Waimarino)

Using a different methodology does not mean that the quality for these iwi will be lower, but it does mean that the contribution from weighting will be higher.

Guidelines for using the data

For many purposes, the 2018 iwi estimates provide a more relevant and up-to-date picture than continued use of 2013 Census iwi data. These iwi estimates are not official 2018 Census counts and the quality of these estimates ranges from moderate to poor depending on how the data is used.

When using this data you should be aware of the following:

• There has been a large increase in the Māori descent population and the majority of iwi since the 2013 Census. This makes it difficult to look at changes over time. Using proportions rather than totals may help with this. However, proportions should also be used with caution as any changes in the attributes could be due to changes in who was counted rather than changes affecting individuals.

• There is variability in the data sources for iwi and some iwi have specific quality concerns. We have noted significant issues above, and in Iwi affiliation (estimated counts): 2018 data sources and quality by iwi

• The reliability of the iwi estimates reduces as the size of the estimate gets smaller. Guidelines are provided below to help users understand if the estimate will be fit for their purpose.

• The characteristics (such as highest qualification) published as part of this release were chosen for their higher levels of quality compared with other census variables. However, the level of missing data for these characteristics is still higher than that reported for the New Zealand population as a whole in Data quality ratings for 2018 Census variables. We have reported the contribution from different data sources and weighting when producing iwi estimates for each characteristic below.

Quality guidelines when looking at small groups

The reliability of the 2018 iwi estimates reduces as the size of the estimate gets smaller. This is because when we are producing an estimate for a small group we are less confident that the assumptions we make about the people with missing iwi data will hold true than when we are looking at a large group.

Below are some guidelines on the uncertainty in the estimates due to non-response. These are guidelines only and sometimes the true count will be outside these guidelines. The intention in providing these is to help inform users of how much uncertainty there is around an estimate, not to provide a definitive range that the true value lies within.

• If you have an estimate of 20, the uncertainty in the estimate due to non-response will be around +/- 10 (+/- 50 percent). This means that if everyone had responded to the 2018 Census, we would expect the true count to normally be somewhere between 10 and 30.

• If you have an estimate of 100, the uncertainty in the estimate due to non-response will be around +/- 20 (+/- 20 percent).

• If you have an estimate of 500, the uncertainty in the estimate due to non-response will be around +/- 50 (+/- 10 percent).

Quality guidelines for iwi characteristics

Estimates of iwi characteristics (for example, counts of religious affiliation by iwi) are lower quality than we would expect from a census due to the high level of non-response. This is particularly the case when estimating the size of very small groups (for example, less than 100 people). The guidelines above can be used to understand how accurate an estimate for different sized groups is likely to be.

The level of data that is either missing or sourced from alternative data sources will also vary for each characteristic. This can influence whether the data will be fit for purpose. The tables below show the contribution of data sources for each characteristic in this release, and the contribution from weighting.

For example, when producing iwi affiliation by age, 75 percent of the age characteristic is sourced from the 2018 Census, 9 percent is sourced from administrative data sources, and 16 percent of the estimate is produced by applying weighting to people in the previous categories. In contrast, administrative data sources were not available to derive ‘Status in employment’ so a higher level of statistical imputation was used.

As highest qualification is derived from two input variables (highest secondary school qualification and post-school qualification level), we have combined the categories ‘2013 Census data’ and ‘Administrative’ data into a single column which indicates if either one or both of the input variables were sourced from these alternative sources. This is displayed in the table below.

For some characteristics, such as household income, there can be high levels of missing data and imputation occurring. Users should check the data sources and level of missing data if requesting additional characteristics to ensure they are fit for their purpose.

Further information

Data quality ratings for 2018 Census variables and the External Data Quality Panel: Assessment of variables provides further information about the quality of different variables in the 2018 Census.

Stats NZ Store House provides samples for both the 2018 individual and dwelling census paper forms. has samples for both the individual and dwelling census paper forms.

Iwi affiliation – Information by variable (2018) has a detailed summary of the differences between the 2013 Census and 2018 Census questionnaires.