"Family Ethnicity": Knitting a Jumper Using Two Woolly Concepts

Paul Callister¹
Victoria University of Wellington

Robert Didham
Statistics New Zealand

Jamie Newell
MERA

Deborah Potter
Statistics New Zealand

Abstract

While ethnicity, as collected in surveys in New Zealand, is a personal attribute not a group measure, there is some demand from the policy community and researchers for measures of family ethnicity. Yet both ethnicity and family are "woolly" concepts. The paper explores the uses made of ethnic family measures in research and policy making in New Zealand and, based on census data, explores a range of possible classification systems. The diversity of individual ethnic affiliations within New Zealand families leads us to suggest that measures of family ethnicity that incorporate the responses of all individuals are likely to be more suitable for informing research and policy than those that lead to an artificial simplification of ethnic responses.

Background

The concept of an ethnic family is commonly used in everyday conversation and in the media. Similarly, policy analysts sometimes talk about ethnic families in contexts such as the incidence of poverty or the adequacy of retirement savings. Yet the 2004 Review of the Measurement of Ethnicity (RME), undertaken by Statistics New Zealand, argued that ethnicity was a personal attribute that could not be ascribed to a group. The review noted that the high rate of intermarriage between ethnic groups in New Zealand, as well as the significant number of individuals who record dual or multiple ethnicities, creates major challenges in assigning an ethnic group to a household or family, both conceptually and practically. Although some submissions to that review and a subsequent review of family statistics said that family and household ethnicity output would be of interest, there was an awareness of the problems of associating an individual response variable to a family or household.

Given both the interest in family ethnicity, but also the challenges in measuring it, Statistics New Zealand recognised that further exploration of the concept would be useful and therefore provided research funding. Based on a mixture of theoretical considerations, empirical investigation and consultation, four main issues were explored:

• What are the uses made of ethnic family measures in research and policy making in New Zealand?
• Are these common enough uses to warrant standard measures, and if so, what type of measure should be used?
• Alternatively, should a range of measures be developed to suit a variety of uses?
• What sorts of results are obtained using different methods of classifying the ethnicity of families applied to census data?²

As indicated in the title, family ethnicity brings together two concepts, "family" and "ethnicity", which are, to some greater or lesser extent, woolly (that is, they are difficult to define). We therefore begin with a brief discussion of what these two concepts are measuring. The paper then provides some examples of how family ethnicity is used in research and policy making within New Zealand. This is followed by examples of how family ethnicity could be measured.³

Overall, our research supports the earlier Statistics New Zealand view that family ethnicity is primarily a personal attribute that cannot be easily attributed to a group. However, the consultation process showed that there is some demand from within the policy and research community for ways of classifying family ethnicity. Part of this demand comes from having population agencies and a need to report ethnic outcomes within this context. We also found a range of family ethnicity measures in current use without methodological critique. In discussion it was recognised that there was a need for some evaluation of measurement strategies. Given this demand, and based on our statistical exploration, we set out the family ethnicity measures we consider most useful.

What is ethnicity measuring?

There has been a long, vigorous and unresolved debate about the measurement of the ethnicity of individuals in New Zealand. This includes papers recently published in this journal (e.g. Callister 2004, Kukutai 2003). As a result of its latest review of ethnicity statistics, Statistics New Zealand (2004) listed a number of factors that may contribute to, or influence, a person’s ethnicity. As they note, many of these are interrelated. This list is:

• name⁴
• ancestry
• culture
• where a person lives and the social context
• race
• country of birth and/or nationality
• citizenship
• religion and language.

This list suggests that a quite diverse set of influences will be guiding individual responses to surveys. For example, some people may be strongly influenced by their ancestry but have little connection with the culture that may be associated with such ancestry. Others may emphasise more strongly their cultural links and/or the country they migrated from. One of the issues that has received much attention in recent years, and which emphasises the "woolly" nature of ethnicity, is how to analyse data where people record more than one ethnic group. Another, especially post the 2006 Census, is how to deal with "New Zealander" responses. In his 2005 paper Understanding and Working with Ethnicity Data,Didham sets out recommendations in relation to both these issues. In terms of output data, in relation to individuals Statistics New Zealand has recommended no longer using ethnic prioritisation but instead using total counts or, where appropriate, main single and multiple combinations of ethnicities. In relation to the response "New Zealander", past practice has been to place these responses in the New Zealand European category, which sits within the level 1 category "European". However, these responses are now placed in a new "Other" level 1 group in 2006 Census output. In the 2006 Census, "New Zealander" was the third-largest ethnic group. While this level of response is not yet being seen in administrative surveys, discussions with the policy community suggest "New Zealander" responses may be becoming more common in such data sets.

Despite the discussions as to what ethnicity may be measuring, our consultation suggests there remains considerable uncertainty among data users as to what information the ethnicity variables collected in official statistics are providing. Although ethnicity is supposed to be primarily a self-defined cultural measure, many people see ancestry as having the strongest influence on ethnicity. There also remains a variety of opinions about recognition and handling of New Zealander responses, with some suggesting these responses should simply be re-classified as Europeans, whereas others see this as a valid response signalling an important cultural change occurring in New Zealand. It also seemed that, in part, the uses made of ethnicity data by researchers and the policy community are also helping shape concepts of ethnicity. Uncertainty about what individual ethnicity responses are measuring needs to be kept in mind when considering family ethnicity measures where these uncertainties become manifold.

What is a family?

Although not an issue that is directly focused on in this study, the definition, then measurement and classification of families is an evolving area. The "woolly" nature of families in the 21st century is acknowledged in the definition of them in the Families Commission Act 2003. Recognising diversity the Act states "family includes a group of people related by marriage, [civil union,] blood, or adoption, an extended family, 2 or more persons living together as a family, and a whānau or other culturally recognised family group". Challenges in defining then measuring families include the recognition that families extend beyond household boundaries, or that households may contain more than one, not infrequently interrelated, family. In addition, individuals within a family may conceptualise their family in different ways from other family members. These dynamics will also affect the way in which individuals report their ethnicities and consequently the measurement of ethnicity within families. Moreover, data for family members are often obtained by proxy from one family member.

In the early stages of our research, Statistics New Zealand was undertaking a Review of Official Family Statistics, and in relation to family ethnicity this review recommended further research. Also informing our research was a Statistics New Zealand Review of Culture and Identity Statistics, a Ministry of Health scoping project to investigate options for measuring whānau (Ministry of Health 2006), and a Families Commission project to consider whānau measurement (Walker 2006).

Use of family ethnicity by the policy and research community and the wider public

In New Zealand, terms such as "Māori family", "Pacific household" or "Somali family" are commonly used by the general public, are reported in the media and can be found in research literature. Government policy statements at times focus on family or household ethnicity rather than on individuals and their ethnic affiliations. Often this focus is on actual or hoped-for outcomes for families rather than on the mechanisms for targeting. As an example, in relation to the Working for Families policy, the Government noted that "Budget 2004 is good news for Māori families".⁵ Terms such as "Pacific Island families" and "Pacific families" have also been used in the research arena. For example, these terms are used when reporting on research regarding Pacific children and their families (Butler et al. 2003, Paterson et al. 2006).⁶

Although most services are targeted at individuals, a variety of government agencies and non-government organisations potentially have an interest in ethnicity, and its associated attributes, within a family or household context. For example, having information on the dynamics of ethnicities within families and households may help a better understanding of language retention and transmission. Research indicates that in 2006 Māori language (Te Reo) was spoken by 24,072 people who do not record Māori ethnicity, but that most speakers do record Māori ethnicity. Knowing the relationship between ethnicity, and gender, of parents and the ability to speak Te Reo may assist in determining the potential for in-home initiatives to support the acquisition of the language by children relative to initiatives outside of the home.

However, the family as an ethnic unit may be of interest. For example, the Ministry of Social Development has a strong interest in evaluating outcomes under the Working for Families package. This programme is targeted at families rather than individuals and, as such, is based on family income and patterns of paid work. The Government has expressed an interest in determining outcomes for Māori and Pacific families.⁷ The Ministry of Social Development notes that in terms of take-up and then outcome, some policies may, often through data limitations, focus first on one family or household member, but with an ultimate aim of understanding whole family or household dynamics. In the Working for Families example, an initial interest in the level of take-up by the primary caregivers is driven by MSD’s ability to measure take-up by primary caregivers (or applicants for Working for Families Tax Credits – WFFTC), but the interest in ourcomes has always been more widely focused.

The Ministry of Social Development is interested in the living standards of individuals (particularly children) and families of different ethnic groups. Similarly, the Retirement Commission is concerned about saving and wealth patterns and has attempted to think about how these might vary by individuals and couples, and the influence of ethnicity for individuals and within couples (Scobie et al. 2005).

Another possible reason for examining family or household ethnicity is to understand remittances by migrant groups and their New Zealand-born members back to members of the wider family in the country of origin. For example, remittances by Pacific people are very important, but there is some debate as to whether they will decline among third and fourth generation Pacific people in New Zealand (de Raad and Walton 2007). It is possible that marriage outside of a Pacific group may weaken the propensity to remit, so full information on family ethnicity could be important when investigating these transfers. On the other hand, it is also possible that it may not be family ethnicity as such that is an influence, but the ethnicity of a key influencer within families and households.

Health-related targeting may also be assisted by a better understanding of ethnicity within a family setting. Immunisation targeting is one example. It is useful for health policy analysts to have a clear picture of children’s families in order to assess how closely ethnicity can be connected to immunisation take-up and, in turn, tailor programmes to improve take-up levels. But there can be some complexity in relation to both monitoring and tailoring. In terms of targeting services, health promoters will generally focus on the mother, yet a programme aimed at Māori mothers may not reach a significant group of Māori children. As Howard and Didham (2005) have shown, many partnered mothers of Māori children record only a European ethnic group and, increasingly, there is a group of partnered mothers who record a Pacific ethnicity but have Māori children.⁸ One in five children who have Māori recorded as one of their ethnicities in single-parent households have a residential parent who is not Māori (Statistics New Zealand 2001).

Use of early childhood education and care (ECE) services can potentially also be thought about in a family context. For example, how many Māori families use kōhanga reo as their main childcare provider? The New Zealand Childcare Survey⁹ carried out in 1998 included a focus on ethnicity. The researchers were careful not to use terms such as "Māori families", but there was an interest in childcare use by both children and their parents, and both the ethnicity of parents and children was considered (although not together). The ECE work undertaken by J. Robertson (personal communication, 2007) suggests that the ethnicity of parents might be important in determining if a child affiliating with both Māori and European ethnicities goes to kōhanga reo. Since mothers seem to be primarily responsible for selecting an ECE provider, it is possible that a Māori mother might be more inclined to choose kōhanga reo compared to a New Zealand European mother (with a Māori partner).

Many of these examples show a need to understand the complexity of ethnicity within family or household settings. This does not require the whole family or households to be assigned an ethnic classification, but instead requires an understanding of the interactions of individuals within the family or household. However, some research and policy work provides stronger incentives for developing whole-family or household measures.

Types of data collected, and some of the wider methodological challenges in determining family ethnicity

Historical studies and current research practices identified in our research indicate a range of challenges when considering ethnicity in relation to families. Some relate back to the collection and reporting of ethnicity for individuals, while some are more specific to classifying ethnic groups for families.

Although researchers outside government policy agencies rely heavily on official data collections, including the population census and the Household Labour Force Survey, researchers and policy analysts within government have access to a much wider set of administrative data. Many of these data are not collected primarily for research purposes, but are increasingly seen as a valuable tool for research. Some outside researchers are also starting to have some access to administrative data sets.

Some important data sets, such as those held by the Inland Revenue Department (IRD) and external migration data, do not include ethnicity data.¹⁰ Other data sets, such as Ministry of Education data, do record ethnicity but, while the Ministry recognises the importance of family settings in educational outcomes, family data are not collected. Although it does not have key demographic data, IRD data are increasingly used as a vital component of linked data sets. Examples include the Linked Employer–Employee Data and student loans data. In some cases the ethnicity data can be supplied from another data set that is being linked to the IRD data, such as Ministry of Education data, but in other circumstances, such as employment data, ethnicity is not collected. In the US, attempts have been made to predict the ethnicity of individuals in taxation data sets based on other characteristics such as age, income and residential address (Perez 2006).¹¹

A set of administrative data that does contain the ethnicity of all household members is the Housing New Zealand tenancy data. In this data set a primary tenant is assigned, who then records details of other people living there. Currently, ethnicities collected in these data are prioritised, with only one response per person being retained. However, the methodology is to be updated to reflect the current ethnicity standard for official statistics.

Proxy ethnic data are a consideration for researchers using ethnic family measures. Although it is sometimes clear that the data have been supplied by another person (such as in surveys where only one person in a household is interviewed but supplies data on other people in the household), at other times there is an implicit assumption that the surveys were self-completed when in all likelihood only one person in a household completed all individual forms. An example is the population census. However, in some other surveys, the ethnicity of only one family member may be collected. This raises a similar challenge to that discussed by Perez. Can family ethnicity be predicted based on the ethnicity (and other characteristics) of just one individual?

At times the use of proxies may allow a better understanding of families, at least from the perspective of one member, in a situation when it is not possible to ask all members directly. An example would be a European sole mother who had a Māori child and was interviewed in a survey. If only her ethnicity were collected it would likely be difficult to predict the ethnicity of the other members of the family, but if the mother were asked the ethnicity of all family members then this could provide a richer data source in the circumstances.

Another challenge that affects the analysis of family ethnicity is that in small surveys the sample size does not support complex ethnic analyses of either individuals or couples, let alone the family as a whole.

Finally, when considering family ethnic classifications, there is the challenge of creating "outcome" measures for families that contain more than one adult (or child). Simple measures such as family income can be created if the income of all family members is asked or if one person can estimate this income. Creating a measure such as family education levels, labour market outcomes (except perhaps total hours worked) or languages spoken is problematic because each member’s attributes cannot be meaningfully grouped. In the past, often the education and/or income level of the male partner was considered a good proxy for the socioeconomic status of the family. Ideally, data are needed on each member, but how to use these data is problematic. Options include somehow combining responses (for example, to give an "average" level of education or use of Te Reo in the family), or somehow considering all the data. This problem parallels that of creating a single ethnic measure for families or households.

The methods tested

In our research we tested a number of methods. However, in this summary paper we report only on total counts; prioritised ethnicity; fractional ethnicity; main, single and multiple counts; and, for families with children, the ethnicity of the children versus that of the parents. Most of the empirical analysis was based on 2001 Census data, but we utilised some 2006 data primarily to understand the effect of an increase in “New Zealander” responses and a change in the way they are classified. Given the variety of family types, our data analysis was based primarily on couples with dependent children. We also undertook most of the analysis at level 1 of ethnicity.

Total Counts

Total counts are now a standard output method when considering individuals. This method is also easily applied to families. In the total count approach all ethnicities of all members of the family, if they are available, are considered. Whenever an ethnicity is found within an individual in a family, this is counted. However, each ethnicity is only counted once. This means, for example, if there are family members who record Māori as their only ethnicity, or if they record it as one of their ethnicities, then this becomes a Māori family. Equally, particularly relevant to 2006 Census data, if anyone records New Zealander, then the family is also counted as a New Zealander family. As our 2006 data showed, when using total counts and focusing on couples with children, New Zealander families are a very important ethnic family. Using this method there were nearly as many New Zealander families as Māori families, and considerably more than Asian or Pacific families.

At level 1 at least, total counts are well suited to small surveys, because all families who record a particular ethnicity are counted, thus increasing the size of particular subpopulations. This is in fact the way in which families have been generally categorised in output in recent years, whereby a family has been considered to be of a particular ethnicity if at least one person within the family has reported that as one of their ethnicities.

There are, however, some problems with the total count solution. First, the family total counts sum to more than the number of families, since multi-ethnic people or family members with different ethnicities get counted in all the groups to which they belong. This may at times be confusing. But more important, multiple ethnicity remains hidden to the observer in total count data, and as such the method can disguise diversity within a particular ethnic population

Another problem is that using total counts makes some statistical analysis techniques difficult due to the overlap of counts of individuals, families and households. But it can be argued that this is less a problem of the total counts and more a problem of the statistical techniques. The New Zealand population increasingly reports multiple membership, in areas such as income sources or types of work, so the models are not reflecting real life. This suggests developing better statistical analytical techniques is necessary.

At an end users workshop held in June 2007, there was a high level of support for using total counts as a main method if family ethnicity is to be measured.

Ethnic Prioritisation

As already discussed, this was a system previously promoted by Statistics New Zealand when multiple ethnic responses were less commonly reported in official statistics. However, it has been a relatively short-lived method. Under this system, a Māori response had priority coding, followed by Pacific peoples, then Asian, other ethnic groups besides European and, finally, European. This prioritisation system meant that, for example, if a person recorded himself or herself as belonging to both the Māori and Samoan ethnic groups, they were classified as belonging just to the Māori ethnic group. Under the system of prioritisation described, the count of Māori is the same as the count based on total count data, but all other groups are reduced in number.

There were both advantages and disadvantages to prioritisation. The one major advantage was that ethnic counts equal counts of the total population. However, this advantage was greatly outweighed by the disadvantages. The disadvantages were that (1) there is no underlying logic to the order of prioritisation, (2) it is not ethnically neutral (that is, it elevates one ethnic group over another), (3) it undermines how people might otherwise choose to identify themselves in a particular context, and (4) it biases population estimates. When prioritisation of ethnic responses was first introduced, multiple reporting of ethnicity was relatively uncommon, and so prioritisation of the responses had little impact on the resulting statistics. However, evidence has increasingly shown that prioritisation was problematic in that it did not reflect the New Zealand population.

Although some within the research and policy community (mainly in the health domain) still use it, Statistics New Zealand no longer recommends the system for individuals (Statistics New Zealand 2004). However, given that it is still used by some researchers, it was useful to consider the effect of using this classification system.

Research has already shown that for individuals, prioritisation results in a major loss in counts of Pacific people, particularly young Pacific people. This loss is also dramatic when families are considered. For example, when 2001 data were used for couples with children under the prioritisation system, there is a loss of 26% of “Pacific families” but also 22% of “European families”. The “lost” families in this system of prioritisation are classified as “Māori families”. When sole parents are considered, the “losses” from all the ethnic groups other than Māori are even higher. In particular, while 22% of “European” couples with children are lost, 34% of “European” sole-parent families are lost. This primarily reflects the significant number of sole-parent families where there is a European mother and a Māori child.

At the end users workshop held in June 2007, there was little support for using ethnic prioritisation if family ethnicity is to be measured. However, there was also recognition that many historical data sets only have prioritised data available for individuals.

Fractional Ethnicity

In relation to individuals, Gould (2001, 2002) has proposed the use of the fractional ethnicity model. This has, for a number of reasons, always been a contentious idea (e.g. Jackson 2003). In this method the number of times each ethnicity is claimed is counted. However, unlike total responses options, the response of each individual would be given equal weight, with a total value of one for a person’s ethnicities. This would be achieved by adding to each ethnicity a coefficient equal to the reciprocal of the number of affiliations claimed. Thus, an individual respondent ticking only Māori would be coded (1/1) Māori; but a respondent ticking both the Māori and the New Zealand European options would be (1/2) Māori plus (1/2) New Zealand European. The total of the responses would then equal the total population. In a family setting this would simply be extended to the number of people in the family. Potentially, this gives researchers the chance to determine whether a family is “more” or “less” of a particular ethnicity, such as being “more European”. This could be seen as a very crude scale of “cultural strength” if ethnicity is seen as a reasonable indicator of culture.¹²

At the end users workshop held in June 2007, there was virtually no support among the policy community for using fractional ethnicity if family ethnicity is to be measured. However, some academics saw that, if used with care, it may be useful as an exploratory tool for family ethnic analysis.

Main Single and Multiple Counts

When samples are large enough, all main single and combination ethnic groups recorded in a family can be counted. This system can reflect the complexity of ethnicity within families. In addition, there is no double counting of families. The system also allows results to be converted into a prioritised or total response equivalent for comparison with other historical data, for example. However, this system creates many possible combinations for individuals, and this often becomes more complicated in families and households as more people are added. For small surveys the sample size would not support the identification of many combinations, although targeted aggregations are possible.

When considering the ethnic groups of members of a family, no weighting is put on how often an ethnicity is recorded. For example, in a couple without children, if one partner puts Māori only and the other puts European only, then this couple will be classified as Māori/European. A couple with one partner who records Māori only and the other partner records Māori and European would also be classified as a Māori/European couple.

In our exploratory work we found that there were quite significant differences in outcomes between various single and combination groups. This is illustrated in Table 1, where average yearly income for couples with dependent children is calculated. Taking the example of Pacific families, in the total count system these families have the lowest average income at $45,676. But when main single and combination ethnic responses are explored, it is shown that it is families where everyone only records a Pacific ethnicity (Pacific only) that bring the average down. While smaller in number, the Pacific/European couples have average incomes nearly $20,000 higher.

Table 1  Average Yearly Family Income from all Sources for Couples with Dependent Children, 2001, Ranked by Size of Income

Source: Census of Population and Dwellings, Statistics New Zealand

At the end users workshop, there was some support for using single and multiple ethnic counts if family ethnicity is to be measured. However, it was also pointed out that many factors would be behind the type of result shown in Table 1. Such factors include whether the people were New Zealand or overseas born, or the age of the parents and number of children. When all the other factors are taken into account, it may be that a complex ethnicity classification tells us less than initially thought. It was also recognised that many data sets would not support this level of analysis.

Base the Ethnicity of the Family only on the Ethnicity of the Children or the Parents

Just as the ethnicity of families in the past has been based on the ethnicity of the adult “occupier” if this person was a member of the family, for those families with children the ethnicity of the family can be defined by the child or children. This system has the advantage that when issues such as child poverty are being examined, then child ethnicity can be focused on. However, there remain a number of methodological challenges. These include:

• If there is more than one child in the family, which child is used, or could there be some system to determine the ethnicity of an “average” child?
• Are adult children considered?
• If the child records more than one ethnic group, how are these multiple ethnicities to be handled?

A system based on the ethnicity of children has a number of drawbacks. In some data sets, such as those used to monitor Working for Families, child ethnicity is not recorded. For health targeting, such as child immunisations, the ethnicity of the parent most likely to be making a decision about immunisation would not be known. Also, outcome populations of interest – such as children – do not necessarily share the ethnicity of their parent/s; as already indicated, a significant proportion of children who record Māori as one of their ethnicities do not have a mother who records Māori ethnicity.

It is also possible to base the ethnicity of a family on one or more adults. This method is especially suited to families with no children. In some surveys, ethnicity data information will only be available for the one adult who responds to the survey. For couples, when there are strong gender differences in intermarriage rates (for example, among the Asian ethnic group) there could be differences in ethnic family counts if it is only the women or, alternatively, the men who are considered in the analysis.

In addition, when using the ethnicity of either the child or parent(s), methods for handling multiple ethnicities still need to be adopted.

In the end users meeting, although it was accepted that knowing the ethnicity of all family members provides optimal information, it was acknowledged that classifications based on either children or parents can be the only practical solution if a family ethnicity measure is required based on particular data sets.

Conclusion

Our research has supported an earlier Statistics New Zealand view that family ethnicity is primarily a personal attribute that cannot be easily assigned to a group. However, the research also showed that there is some demand from within the policy and research community for ways of classifying family ethnicity because this concept is embedded in the discourse within which they work. Part of this demand comes from having population agencies and a need to report ethnic outcomes within this context.

The creation of a possible standard measure, or measures, of family ethnicity was not an aim in itself of this project. The measures that were explored in this research were primarily aimed at meeting specific information needs or to compare outcomes using a variety of measures either used in the past, still in common usage, or suggested as alternative approaches. The research showed that, in the past, a variety of methods have been used to define family ethnicity and that this is likely to continue. Reasons for the diverse methods used include the characteristics of the data sources being used, but also the range of reasons for wanting to understand ethnicity within a family context. However, if we were to be seeking some "gold standard" for both data collection and output, the diversity of ethnic affiliations within New Zealand families leads us to suggest that, overall, measures of family ethnicity that incorporate the responses of all individuals are likely to be the most suitable for informing research and policy.

The importance of incorporating all available information on all individuals within the family is centrally important to a full understanding of family form and function. However, we are also aware of the fact that in many cases this level of detail is neither collected nor, indeed, collectible when the operational requirements of the collection preclude contact with more than one member of the family, or the purpose of the collection does not fit with the inclusion of suitable questions. This is especially pertinent to some administrative data sets, such as migration information, where it is simply not practical to add ethnic questions to an already crowded form which is used in an environment where the efficient collection of other information paramount to the purpose of the collection must take precedence. In other administrative data sets information is not updated and will not reflect ethnic mobility.

In the case of administrative sources where there is contact with only one family member, the use of this information to define generalised characteristics of the family is highly problematic and may be sufficiently misleading to render the information unsuitable as a source of family ethnicity, depending on how the data were recorded and what other associated information is available. However, the pressure to derive family ethnicity from these types of data remains strong.

It is also clear that in many contexts the use of family ethnicity measures by the research and policy community is not backed by adequate documentation of how the measures are being defined. As has now become standard in health research based around individual ethnicity, when family ethnicity is being used it is important that researchers and policy analysts clearly set out how their measure (or measures) is being derived.

As always, the research process raised a number of questions. Some were technical, such as how to use particular statistical techniques that are suited to single responses when people report multiple responses to questions. But more importantly, the subtitle of this paper emphasises that family ethnicity is an essentially fictional concept that not only relies on one increasingly problematic concept – "a family" – but adds into this discussion another idea, ethnicity, the woolliness of which is becoming increasing apparent. A fundamental question raised in the research process is what the ethnicity data collected for individuals is actually telling us. In part this question had already been prompted by the emergence of the significant New Zealander-type responses. But we also do not clearly understand a range of other issues, including the significance of single versus multiple responses, why some people report Māori ancestry but not ethnicity, how stable some responses are over time, and how ethnicity is transferred to the next generation within a family setting. More investigation, including qualitative research, is needed on the evolving meaning of ethnicity, both in an individual and in a family context in New Zealand.

References

Butler, S., M. Williams, C. Tukuitonga and J. Paterson (2003) "Problems with damp and cold housing among Pacific families in New Zealand" New Zealand Medical Journal, 116(1177).

Callister, P. (2004) "Ethnicity measures, intermarriage and social policy" Social Policy Journal of New Zealand,23:109–140.

De Raad, J.P. and M. Walton (2007) "Pacific people in the New Zealand economy: Understanding linkages and trends" background paper prepared for the Pasifika Project Conference, Institute of Policy Studies, Victoria University of Wellington, 8 February.

Didham, R. (2005) Understanding and Working with Ethnicity Data: A Technical Paper, Statistics New Zealand, Wellington.

Gould, J.D. (2001) Ethnic Bias in the Census, submission by John Gould to the Review of Ethnicity Statistics.

Gould, J.D. (2002) "Ethnic shares in the 2001 population" New Zealand Population Review,28(1):147–154.
Howard, S. and R. Didham (2005) Ethnic Intermarriage and Ethnic Transference amongst the Maori Population, Statistics New Zealand, Wellington.

Jackson, M. (2003) "The part-Maori syndrome" Mana, 52(June–July):62.

Johnston, R., M.F. Poulsen and J. Forrest (2003) "The ethnic geography of New Zealand: A decade of growth and change, 1991–2001" Asia and Pacific Viewpoint, 44:109–130.

Kukutai, T. (2004) "The problem of defining an ethnic group for public policy: Who is Māori and why does it matter?" Social Policy Journal of >New Zealand, 23: 86–108.

Ministry of Health (2006) Whakatātaka Tuarua: Māori Health Action Plan 2006–2011: Discussion Document, Wellington, Ministry of Health.

Paterson, J., C. Tukuitonga, M. Abbott, M. Feehan, P. Silva, T. Percival, et al. (2006) "Pacific Islands families: First Two Years of Life Study – design and methodology" New Zealand Medical Journal, 119(1228).

Perez, A.D. (2006) "Choosing a race: The effects of proxy reporting on the identification of mixed race children" paper presented at the Annual Meeting of the Population Association of America, March 30 – April 1, Los Angeles.

Scobie, G., J. Gibson and T. Le (2005) Household Wealth in New Zealand, Institute of Policy Studies, Wellington.

Statistics New Zealand (2001) Tamariki and Rangatahi Maori Regional Reports

Statistics New Zealand (2004) Report of the Review of the Measurement of Ethnicity, Statistics New Zealand, Wellington.

Walker, T. (2006) Whānau is Whānau, Blue Skies Report No. 8/06, Families Commission, Wellington.

Footnotes

¹ Acknowledgements
This report was commissioned by Official Statistics Research, through Statistics New Zealand. The opinions, findings, recommendations and conclusions expressed in this report are those of the author(s), do not necessarily represent Statistics New Zealand and should not be reported as those of Statistics New Zealand. The department takes no responsibility for any omissions or errors in the information contained here.
Correspondence paul.callister@vuw.ac.nz

² Although this paper focuses on family ethnicity, most of the issues canvassed are relevant to the measurement of household ethnicity.

³ A technical working paper provides the detailed analysis on which this summary paper is based. This paper is available from the authors.

⁴ Statistics New Zealand (2004:7) notes that a "name" is "a common proper name that collectively describes a group of individuals and authenticates the characteristics and the history of its members".

⁵ http://www.scoop.co.nz/stories/PA0405/S00641.htm

⁶ The study started with the collection of data from a cohort of infants born during 2000, predominantly involving infants of Samoan, Tongan, Cook Island Māori and Niuean ethnicity.

⁷ Personal communication with Ministry of Social Development.

⁸ There is also a group who record more than one ethnicity; for example, Māori

and Pacific.

⁹ www.stats.govt.nz/datasets/social-themes/childcare.htm

¹⁰ Inland Revenue, in its role of administrator of Child Suppo

rt, WFFTC and Paid Parental Leave, holds some information about family units and their dynamics. In addition, information is held on "spouse" or "associated person" relationships as declared in many tax returns.

¹¹ Predicting by residential address can be more powerful when there is a high level of geographic ethnic segregation in a society. However, New Zealand has a lower level of such segregation than the US (Johnston et al. 2003).

¹² It is open to debate how much information ethnicity gives about cultural strength.