The accuracy and reliability of census data is dependent on success in these and other staged procedures—in the wording of census questions, in the design of census questionnaires, in field procedures for administering questionnaires, in the level of respondent acceptance of the census, in the nature of the responses to the questions, and in the efficacy of the processes for translating census information to appropriate and meaningful data categories. For each of these procedures, issues and nuances arise that may potentially compromise data content and quality in the final analysis. For a variety of social, cultural, economic, political, administrative, and geographic reasons, this potential is heightened in the enumeration of Indigenous Australians.

The criticism that official statistics inadequately represent Indigenous numbers usually makes reference to an undercount (Commonwealth of Australia 1992), but may also refer to an overcount, as in the case of Tasmania, for example (ATSIC Tasmania 2000: 3). Whatever claims are made about census coverage, there is no escaping the fact that the official ABS estimate of a 7 per cent Indigenous census undercount in 1996 was substantially higher than the figure of 1.5 per cent calculated for the rest of the population (ABS 1998: 28), and this was true again in 2001 (6.5% compared to 2.2%; ABS 2002: 4). While this undercount is compensated for in ABS Indigenous population estimates at the Statistical Local Area level, the ABS cannot make adjustments at the more local level of Indigenous Locations where data error is most starkly exposed. At the 2001 Census, a total of 62 884 Indigenous people, or 17.8 per cent of the total, were counted using the SIPF: the potential implications of any undercount (or overcount) that might arise from these special procedures are not insignificant.

There appears to be little doubt that difficulties in accurately enumerating some remote Aboriginal populations have persisted. The ABS acknowledges that enumeration procedures failed in particular localities in 1996, such as in the Anangu Pitjantjatjara Lands (Ross 1999: 62–4), and wider problems are suggested by often erratic community and regional trends in intercensal population change that may reflect census or respondent error from one enumeration to another (Taylor 1997). Specific insight into the possible nature of such errors is provided by a case-study example of the 1996 enumeration in Kakadu National Park (Taylor 1999). While it is true that variation in counts from one census to another may simply reflect the changing de facto population on census night, the Kakadu study found similar variation in place of usual residence (de jure) figures which one would expect to display greater stability over time.

Only one known study has approached a rigorous validation of remote Indigenous census counts (Martin & Taylor 1995, 1996). This concluded that there were systematic methodological flaws in the manner in which household membership was constructed using the Dwelling Check Lists and SIHFs. The result was an underenumeration, particularly of the young, the more mobile and the more socially marginal. Using ethnographic techniques to derive a separate and simultaneous count, this study estimated a census undercount of 17 per cent in the Cape York Peninsula community of Aurukun. While there is no basis for asserting that discrepancies of a similar magnitude exist elsewhere, the under-representation of such cohorts in official census counts is something - 10 - that has long been noted by researchers (Gray & Tesfaghiorghis 1993: 84) and is routinely acknowledged by the ABS (ABS 1993: 16-17, 1998: 28; Benham & Howe 1994: 3), although no suggestions are put forward about why this might occur.