IMPACTS OF SAMPLE SIZE ON ESTIMATE VARIABILITY
The smaller a population is in proportion to the total population, the higher the relative standard errors. Further disaggregation of data for these small populations produces even higher standard errors proportionately.
As an example, in SDAC 2015, 18.3% of the total population had a disability, including 5.8% who had a profound core-activity limitation (includes total profound population living in households and establishments). Of this profound population, there were 3,600 people aged 15-24, living in households, who were employed. The 95% confidence interval for this estimate is 3,600 ± 2,900 i.e. we can be 95% sure that in 2015 the actual number of employed people aged 15-24 with a profound core-activity limitation and living in households, lies between 700 and 6,500.
It should be noted however, that there are stronger estimates for some of the more common measurements. For example, in the scenario above, if the age range were to be widened to include 15-64 year olds, the estimate would be 20,700 people with profound disability who were employed. The confidence interval for this estimate is 20,700 ± 8000 i.e. we can 95% confident that in 2015 the actual number of employed people aged 15-64 with a profound core-activity limitation and living in households, lies between 12,700 and 28,700.
Although the analysis presented in this paper does not generally rely on such detailed classification of data items that the resulting confidence in the estimates is low, the variability of some of the measures compared is relatively high. For further information on the potential effects of sample survey methodology, please refer to: ‘Understanding Statistics’ on the ABS website.