The census cannot hope to capture the complexity of Indigenous principles of kinship and household structure. The attempt to do so in the 2001 Census led to the collection of incoherent and uninterpretable data (see in particular the case studies by Martin and Morphy). The further step of 'classifying' this data into the family types recognised in ABS definitions, which do not coincide with the family types found in many, if not most, Indigenous communities, is a completely spurious exercise, and any analysis which takes these data as a basis must allow for this. It should be noted that the complex familial structures of Indigenous societies are one of their most enduring aspects, persisting in communities in 'settled' Australia as well as in remote, 'traditionally-oriented' communities (see Smith 2000b).

The designers of the census need to step back from the questions on household structure, and decide precisely what information they wish to elicit. Is it information primarily about family structure, or about the age distribution, gender composition, and dependency structures of households? ^[23] If it is decided that the latter data are the most important, one possibility which would sit more comfortably with the Indigenous facts would be to add a new type of household to the ABS list—the extended family household. This definition would apply to large households in which everyone is related to everyone else, and would therefore conflate the 'family' with the usual residents of the dwelling, or household. It would not attempt to distinguish any putative 'couple families' or 'one-parent' families among the residents of the dwelling, and the post-enumeration categorisation would not attempt to break extended family households into such smaller family units.

This solution is grounded in Indigenous reality, in that it recognises the incommensurability of Indigenous and mainstream principles of household formation (as defined currently by the ABS), while still allowing the dwelling to function as a unit of analysis and measurement across the board. It does not address the issue of linked households, which are such a prevalent feature of Indigenous community life, but that problem seems insoluble given the dwelling-based framework of the census enumeration.

If place were defined to refer unambiguously to the community in question rather than to an address within the community, this would eliminate the need to provide numerous instances of 'addresses', identified by Morphy as both unduly time-consuming and culturally inappropriate. It would also circumvent the problem, highlighted by Martin, of identifying place of residence over time in communities where there is both a high level of intra-community mobility and a high turnover of housing stock.

The set of questions on ancestry and origin should be framed from the Indigenous perspective, not from that of the settler culture. If all that is required is an 'Indigenous identifier' (as the case in most Australian census enumerations), then only one question needs to be asked, and that is: 'Are you of Aboriginal or Torres Strait Islander origin?'.- 100 - The questions on overseas parentage are almost certainly superfluous in the vast majority of cases. If more detailed information on ethnicity is sought in future censuses, then the supplementary question should be framed in terms of non-Indigeneity, and the prompts should include 'non-Indigenous Australian', the main Anglo-Celtic categories, Chinese and Afghan, all of which are more likely ancestries for Indigenous Australians than some of the ones suggested in the prompt to the question on the 2001 SIPF.

In many cases, neither the interviewers nor the interviewees will have a high level of formal education. In whichever way the questions on education are framed, the Indigenous interviewers need focused training on the precise meanings of the questions, and on how to interpret the likely answers of the respondents. This applies particularly to the question of further education, where interviewers need clear guidance on what counts as an accredited course, and what is meant by 'full name' and 'field of study'. They should also be helped to identify what counts as a 'place of study'.

The vast majority of Indigenous people in the north and centre of Australia live in areas where there is very little employment, in the mainstream sense of the word. 'Looking for work', in this context, is a pointless activity for most people. In the 2001 Census, the series of questions on the SIPF on employment and looking for work were constructed on the premise that labour market conditions in these regions are comparable to those in other parts of Australia. CDEP schemes were treated as if their sole purpose is to provide 'real' jobs, and all CDEP participants were deemed to have a job. The design of the questions rested, therefore, on two fictions: firstly that there is a local labour market in which employment can be sought, and secondly that all CDEP participants are in paid employment.

This is another example of the dilemma faced in trying to reconcile local Indigenous reality with comparability across the board. For many reasons, comparability may be considered a desirable outcome, and one which outweighs other considerations. That is the first question to be settled. If the principle of comparability is to be preserved, some thought needs to be given as to how the level, or degree of fiction can be reduced, both in the forms of the questions and in the likely responses.

If the quantifiable population characteristics of Indigenous Australians are to emerge clearly from census data, the questions on the Indigenous form need to be as culturally neutral as possible, in order to minimise misunderstanding on the part of the Indigenous interviewers and respondents. If a person fails to understand the meaning of a question, they are unlikely to provide the kind of answer that is sought. Care must be taken to avoid the 'naturalisation' of Anglo-Celtic cultural categories and assumptions, as happened with the 'nuclear family' and with the 'ethnic' division of the Australian population in the- 101 - questions on ancestry and origin. We would further suggest that, given the increasing ethnic diversity of the Australian population, this issue may not be unique to Indigenous people.

That being said, the case studies by Martin and Morphy show clearly that certain kinds of questions, however they are phrased, particularly those that address core elements of contemporary Indigenous identity, are likely to elicit answers that are symbolic rather than technical (in Martin's terminology). The responses to the question on religion are a case in point. The design of the question itself played a part, in making the equation between 'Traditional Beliefs' and Christian denominations, and in failing to signal clearly that more than one answer was possible. It is interesting that the responses in both communities were symbolic, but they were different. In Aurukun, there was perhaps a symbolic identification of 'Traditional Beliefs' with a private Wik domain, not considered appropriate for exposure in a non-Indigenous context, whereas in the outstation community (but not elsewhere in the region, it must be noted) 'Traditional Beliefs' were embraced as a public symbol of Aboriginality. The question on the ability to speak English elicited a symbolic answer from Wik respondents, because of the connection that they make between lack of English and 'being myall', but in the Northern Territory outstation, where the ability to speak English is viewed merely as a useful skill, the answers were technical in nature.

Some questions are inherently more likely than others to elicit symbolic replies. Such responses, which are indicative of underlying beliefs and attitudes rather than constituting objective answers to factual questions, are not amenable to quantitative analysis.