This study found substantial differences in identification of children between the survey data and the WA Register of Births and the WA Midwives’ Notification System. As the identifications in the survey were all made during face-to-face interviewing observing a standard protocol and using the standard question, these figures suggest that identifying Aboriginal and/or Torres Strait Islander births by using the Indigenous status of the mother on the WA Midwives’ Notification System would substantially under-identify Aboriginal and/or Torres Strait Islander births, both by missing cases where the father was identified as being of Aboriginal and/or Torres Strait Islander origin and the mother was not, and cases where the mother’s identification differed between the WA Midwives’ Notification System and the survey. Only 77% of the WAACHS children would have been identified as Aboriginal and/or Torres Strait Islander births using the mother’s reported status on the WA Midwives’ Notification System. Use of this field is the standard practice for deriving statistics on Aboriginal and/or Torres Strait Islander births in WA.
The process described in this paper for evaluating the consistency between identification of Indigenous status between a survey and an administrative data source can potentially be used for other demographic items collected in administrative data. For instance, socio-demographic indicators such as educational attainment, employment and occupation are collected in many administrative data sets. However, they are often not directly relevant to the administrative process, and are often of poor quality with substantial amounts of missing data. With the rapid expansion of the use of record linkage methodologies for undertaking research, validating the quality of fields such as these, with the potential to adjust analyses for observable patterns in inconsistent and under-recording has the potential to increase the quality of research undertaken using administrative data.
Other methods have been suggested for improving the quality of Indigenous identification in administrative data sets. The use of probabilistic record linkage to combine information from multiple data sets allows for the development of algorithmic approaches for filling in missing data on one data set with information from other data sets
], or calculating a best practice indicator maximising available information across multiple data sets. Several studies have identified that under-identification of Aboriginal and Torres Strait Islander peoples in administrative data sets is non-randomly distributed, and methods to improve Indigenous identification in administrative data can result in both changes of counts of records, and changes in averages and other statistics derived from these records
Identification of Aboriginal and/or Torres Strait Islander births directly from the WA Register of Births and the WA Midwives’ Notification System information in WA is not straight-forward. Relying on administrative data alone identifies substantially fewer Aboriginal and/or Torres Strait Islander births than is likely to be the case if parents were asked to directly identify the status of the child at the time of the birth. This is due to the significant proportion of missing Indigenous status for parents as recorded on the birth registration forms, and that the Midwives’ Notification Forms only record ethnicity of the mother, and even among the mothers not completely. Neither the WA Register of Births nor the WA Midwives’ Notification System is designed to identify the Indigenous status of infants. Both systems set out to record the status of one or both parents. However, it is common practice in the analysis of data from the WA Register of Births and the WA Midwives’ Notification System to derive the Indigenous status of the baby from the Indigenous status of the mother. This linkage study has shown that this practice will identify up to 25% fewer births as being of Aboriginal and/or Torres Strait Islander origin than otherwise might be the case.
However, the differences in identification status between the two systems are not randomly distributed. Children who are identified as being of Aboriginal and/or Torres Strait Islander origin consistently in both data sources were more likely to live in a regional, rural or remote area, more likely to live in areas of relative socio-economic disadvantage and were more likely to have worse outcomes on a range of measures of wellbeing. As a result, increasing the level of identification of Aboriginal and/or Torres Strait Islander births in the administrative birth data results in lower estimates of the proportion of low birth weight and low gestational age babies.
Applying the modelling approach used here results in lower overall prevalence estimates for poor neonatal outcomes, but the overall pattern among the time series is essentially preserved. This suggests that this approach could be useful in producing indicators designed to measure progress in closing gaps in Indigenous disadvantage. An alternative approach that has been used in other studies
] is to link the administrative data source of interest to other administrative data sources which also include Indigenous status, and then use an algorithm to derive Indigenous status from combined data. This method has been used successfully, particularly in relation to improving the quality of mortality data
]. It could potentially be used for birth data as well, if the data were being analysed sufficiently retrospectively for there to be sufficient time to have elapsed that the children would have had contact with other services where Indigenous status were collected.
The fact that children with consistent identification across both data sources have on average worse outcomes on all wellbeing measures used in this study reflects the fact that there is no single gap in wellbeing between Aboriginal and Torres Strait Islander peoples and non-Indigenous peoples. There are gradients in wellbeing outcomes among Aboriginal and Torres Strait Islander populations, as there are among non-Indigenous populations, by remoteness and by many other factors
]. While the Closing the Gap
indicators are useful aggregate measures to assess progress in meeting national goals, it is important that progress is not seen to be made solely by improving the way in which Aboriginal and Torres Strait Islander people are identified in source data sets, particularly if those improvements are driven by the inclusion of Aboriginal and/or Torres Strait Islander persons with clearly different socio-demographic profiles.
Where there are gradients in outcomes among Aboriginal and Torres Strait Islander people, studies should continue to be undertaken to measure and report on these differences. Progress in meeting national Closing the Gap targets will not necessarily mean that all Aboriginal and Torres Strait Islander people benefit equally from the improvements that are being made.
Indigenous status is determined by self-identification. Apart from differences in how or if the information is asked in a standard way, people may choose to identify differently in different situations, reflecting differing levels of cultural safety and appropriateness, and the perceived benefits or consequences of identifying as being of Aboriginal and/or Torres Strait Islander origin in each situation
Standardising the process for collecting demographic data in administrative data collections can be difficult if data collection is part of the routine activity of an organisation and that activity itself varies in context between different operational centres. Smaller district and regional hospitals have different case loads and case mixes compared with facilities in the metropolitan area where there are several large facilities dedicated to the delivery of newborns. Smaller centres, particularly in remote areas, may also employ staff in differing roles, including community liaison roles which may result in greater likelihood of expectant mothers being known to staff. The proportion of the population that identifies as being of Aboriginal and/or Torres Strait Islander origin is larger in more remote areas, which may also affect the perceived relevance of asking for this information in administrative contexts. In settings where the number of mothers who identify as being of Aboriginal and/or Torres Strait Islander origin is very low, there may be less motivation for staff to take care with the collection of this data. The physical appearance of individuals could also affect administrative data collection, with staff possibly making inferences based on appearance, especially if they are uncomfortable asking the question.
The approach used here to validate the identification of Indigenous status using survey data is useful in quantifying overall trends and patterns in differences in identification. There was substantial under-identification of Aboriginal and/or Torres Strait Islander births just using information from the WA Register of Births and the WA Midwives’ Notification System, and this was particularly so for infants with a non-Indigenous mother and an Aboriginal and/or Torres Strait Islander father, and was also more common in less isolated areas. While this is useful information in understanding and interpreting data derived directly from administrative data sources, the method of estimating probability of identification in survey conditions is not specific enough to be able to accurately identify individual records where identification would have been different between survey and administrative sources. Thus the adjustments to the time series derived from the administrative sources based on using these probability weights should be considered as an aggregate adjustment to the time series. The technique is not sufficient in and of itself to identify all births for children who would be expected to subsequently identify as being of Aboriginal and/or Torres Strait Islander if a survey was conducted. Thus this approach is not appropriate if the end goal of a project is to analyse associations between Indigenous status and other outcomes at an individual level, such as via regression modelling using birth weight or gestational age as an outcome variable.
This study has some limitations. Patterns of Indigenous identification may change over time. Several studies investigating the quality of ascertainment of Indigenous status have reported that practices have improved over time in several registers
]. However, no information is available as to how practices may have changed over time in respect of the WA Register of Births and the WA Midwives’ Notification System. If patterns change over time, since the WAACHS data was collected, the model used in this approach would be unable to reflect those changes.
Participation in the WAACHS was obviously limited to children who were alive at the time of the survey. As low birth weight and low gestational age babies may be at more risk of premature death it is possible that the exclusion of any chance of a second ascertainment of the Indigenous status of children who died prior to the survey could affect the results of our model. While premature mortality is significantly higher among Aboriginal and/or Torres Strait Islander babies and children
], the small proportion of children who die at a young age suggests this is unlikely to have a major impact on the model fitted to these data.
The explanatory variables used in the models fitted to the data in this study were limited to those variables that are available on the administrative data sources, and thus available for use in calculating probability of identification in a survey setting. There may well be other factors that are relevant to describing the differences between those children who are identified consistently between the two sources and those who are not, which could not be captured in these data.