The standard census procedure in Australia is to count those who sleep in each dwelling on census night, asking them whether they are usual residents or visitors and also asking them whether there are other absent usual residents of the dwelling. In the IES this is still, officially, the approach. However, in the Northern Territory, where two of our case studies were located, there is some tendency to move towards a 'usual residents' basis of counting. In this approach, all usual residents of a dwelling are counted and visitors to a dwelling are not counted if they have some other usual residence where it is judged they will be counted. We say some 'tendency', since this only occurred in one of our two Northern Territory case studies.

The counting in Morphy's outstation case study was, in our judgment, essentially done using the standard procedure of counting those in situ and asking them about absent usual residents. This procedure carries some potential for double counting, if absent usual residents are also counted elsewhere as visitors. But the in situ principle has the advantage of counting people where they are found, or 'encountered', and of relying on the people themselves for information about their demographic and socioeconomic characteristics. The reasons that the Northern Territory branch of the ABS seems tempted to abandon this methodology in favour of a usual residents count would appear to relate to the time extension involved in the IES and to Indigenous people's high levels of residential mobility. Because the IES is interview-based, it takes a week or weeks, rather than a day or two, and is often occurring in different places at different times. The Northern Territory administration of the ABS fears that through this time extension and with people's mobility, counting in situ will miss people, as both they and the census collectors move around. While under-counting is a possible outcome of time extension and mobility, we do not believe that switching to a usual residents basis of counting solves this problem. We also believe that such a strategy introduces problems of its own, relating potentially to both under-counting and to data quality.

In the Alice Springs town camps, for example, where the usual residents method of counting was used, a number of people who were present in the town camps at the time of the census were not counted, on the grounds that that they were only visitors and would be counted elsewhere as usual residents. But the point has been made in all the case studies that Aboriginal people in northern and central Australia often move between several dwellings within one or more than one community location and so it is not clear that all such visitors will in fact be counted elsewhere as usual residents of a particular dwelling. Indeed, as the case studies and earlier work by Martin and Taylor (1995) show, people can be missed even when their mobility is only within one community between several dwellings. The idea of counting usual residents is too much of a gamble and it also means that more census interviewees are supplying information about third parties, which introduces greater data quality concerns. The methodology used in the Alice Springs town camps probably led to a greater under-count than that used in the Northern Territory outstation case study. While the number of usual residents of the town camps was probably reasonably well captured, the number of visitors in the town camps at the time of the census was not well captured at all. It is quite possible that not counting these people in the town camps contributed to under-counts in the larger central Australian region, as they were possibly not counted elsewhere.

The Aurukun case study in far north Queensland, along with the outstation case study in the Northern Territory, shows how counting those present, plus asking them about absent usual residents, can work as a census methodology in Aboriginal communities. As we have suggested, this methodology will, if anything, err towards double counting; however, this would seem the lesser of two evils and can at least potentially be eliminated by informed cross-checking between areas. No such cross-checking is even potentially possible when visitors encountered in particular places are not counted, on the assumption that they will counted elsewhere as usual residents.^[22]

The other way to guard against both double-counting and under-counting using the standard census methodology is to specify more clearly the treatment of absent usual residents. Presently the training package specifies that people who are 'away fishing, hunting, on sorry business, etc.' should have a SIPF filled out for them and hence be counted, as it is judged that they are unlikely to be counted elsewhere. It also specifies that if they are 'away in a city, town or other community' they do not need a SIPF because it is judged likely they will be counted elsewhere. This seems to us too generous an assumption. Two of our case studies identified the issue of usual residents away at sports carnivals in Aboriginal communities. In both cases there was no attempt made to count those at the sports carnival, and indeed at Yuendumu there was a deliberate choice made to move the count away from the carnival. So under the standard census counting methodology these people need to be put clearly in the category of absent usual residents who need to be counted at their usual residence, which in the two-form structure means having a SIPF completed in their absence. Sports carnivals constitute one very important case in which the circumstances of absent usual residents from Indigenous communities need to be more closely specified, but probably not the only one.

The standard ABS approach to the issue of who to count, advocated by both central office and all State and Territory administrations within the ABS except the Northern Territory, needs in our view to be adhered to. It is not perfect, but it is the best available resolution of the issue, particularly if the circumstances in which to count absent usual residents are well elaborated. To keep error due to time extension to a minimum, the best solution is simply to keep time extension of the count itself to a minimum. This may be achieved, in part, by streamlining the interview procedures, which is an issue relevant to both of our next two sections.