Detailed data sources and methods
Data sources
The analyses presented in this report are based on data collected in 3 national cross-sectional surveys conducted by the ABS—the 1995 National Nutrition Survey, the 2007–08 National Health Survey and the 2017–18 National Health Survey.
These data sources were chosen because they provided nationally representative measured height and weight data over an extended period of time. Information on the surveys, including data quality statements, are available on the ABS website.
The 2007–08 National Health Survey and the 2017–18 National Health Survey were based on nationally representative samples that included only residents of private dwellings, and excluded residents of non-private dwellings such as hospitals, nursing homes, hotels, motels, boarding schools, and prisons.
The 1995 National Nutrition Survey was based on a subsample of participants from the 1995 National Health Survey. The subsample was designed to provide national estimates by fine age groups and sex. The 1995 National Nutrition Survey also included only residents of private dwellings.
The sample size in each survey varied, with about:
- 13,800 people surveyed in the 1995 National Nutrition Survey
- 20,800 people surveyed in the 2007–08 National Health Survey
- 21,000 people surveyed in the 2017–18 National Health Survey.
The 1995 National Nutrition Survey and the 2017–18 National Health Survey included measured height and weight data for people aged 2 and over, while the 2007–08 National Health Survey included these data for people aged 5 and over.
Each survey included collection of measured height, weight, and waist circumference by trained interviewers. The 1995 National Nutrition Survey used scales that could weigh a maximum weight of 140 kg. The 2007–08 National Health Survey used scales that could weigh a maximum weight of 150 kg. The 2017–18 National Health Survey used scales that could weigh a maximum weight of 200 kg.
The response rates for physical measures varied between surveys. The ABS imputed BMI for those people for whom BMI was not measured in the 2017–18 National Health Survey. In this method, participants with a missing response were given the response of similar participants. For adults, the similarity of participants was based on age group, sex, part of state, self-perceived body mass, level of exercise, whether or not a participant had high cholesterol as a long-term health condition, and self-reported BMI category (calculated from self-reported height and weight) (ABS 2019). For 2–14 year olds, the similarity was based on age group, sex, self-reported BMI and part of state, while for 15–17 year olds, level of exercise and self-perceived body mass (only if a person answered for themselves) were also used. There was no imputation of BMI in the 1995 National Nutrition Survey or the 2007–08 National Health Survey.
For each survey, the ABS allocated a person weight to each participant, corresponding to how many people in the population they represented. Estimates based on the person weights can be used to infer results for the in-scope population. Note that these person weights are separate to the body weight measurements that are used in the calculation of BMI.
Methods
Birth cohort analysis
This report did not track the same individuals over time. Rather, birth cohorts were constructed using cross-sectional survey data representing the Australian population at various time points. This approach treated, for example, survey participants aged 25–34 in 2007–08, and survey participants aged 35–44 in 2017–18 as representative of the same group of people (those born in 1973–1982) as they aged 10 years between the surveys.
Year of birth was approximated by subtracting age at survey from survey year. For the 2007–08 National Health Survey and the 2017–18 National Health Survey, interviews were conducted in 2 calendar years (for example, interviews for the 2007–08 National Health Survey were conducted from August 2007 to June 2008). Details of which year an individual was interviewed in were not available in the data sets. For these surveys, this analysis assigned survey year as 1995 for the National Nutrition Survey, 2007 for the 2007–08 National Health Survey and 2017 for the 2017–18 National Health Survey.
Records were then grouped into cohorts based on approximated year of birth, using 10-year spans. The width of the spans was chosen to ensure that there was no overlap of birth cohorts at the time points compared.
The statistical significance of any difference in prevalence (percentage) estimates between people in each birth cohort at each age, and also within a birth cohort as people aged, was assessed using 95% confidence intervals. The statistical significance of the difference in median BMI and other BMI percentile estimates between people in each birth cohort at each age, and also within a birth cohort as people aged, was also assessed using 95% confidence intervals. For further details, see ‘Significance testing’ on this page.
Crude prevalence estimates
Crude prevalence estimates are presented as percentages in this report. Crude prevalence, as a percentage, is defined as the number of people with a particular characteristic, divided by the number of people in the population of interest, multiplied by 100.
In calculating crude prevalence estimates, those people for whom the information of interest (for example, BMI or waist circumference) was not available were excluded from the denominator. For 2017–18, imputed data were used for those people for whom the information of interest had not been measured.
All crude prevalence estimates in this report are weighted estimates that use person weights allocated to each survey participant by the ABS.
Median and other percentile estimates of BMI
The median is defined as the midpoint of a list of observations that have been ranked from smallest to largest. The median BMI, for example, is the BMI value at which half the population has a BMI higher than that value and half have a BMI lower than that value. This is also known as the 50th percentile.
A percentile measure is the value below which a given percentage of values in the population will fall. For example, the 10th percentile of BMI is the BMI score for which 10% of people have a BMI that is equal to or lower than this. The percentiles used in this report are the 10th percentile and 90th percentile.
In calculating these estimates of BMI percentiles, those people for whom BMI was not available were excluded from the denominator. For 2017–18, imputed data were used for those people for whom BMI had not been measured.
All estimates of percentiles of BMI in this report are weighted estimates that use person weights allocated to each survey participant by the ABS.
Standard error, relative standard error, margin of error and confidence intervals
For all survey data, the jack-knife replication method was used to derive the standard error (SE) for each estimate, using replicate weights provided by the ABS.
The relative standard error (RSE) of an estimate is a measure of the error likely to have occurred due to sampling. The RSEs of the estimates were calculated using the SEs:
The margin of error (MoE) at the 95% confidence level for each estimate was calculated using 1.96 as the critical value:
The MoE was then used to calculate the 95% confidence interval (CI) around each estimate:
Significance testing
Variation or difference in observed values or rates may be due to a number of causes including, among other things, actual differences in the study’s populations and sampling error. A statistical test of significance indicates how incompatible the observed data are with a specified statistical model. To assess whether differences between estimates are incompatible with a null hypothesis that the survey estimates are normally distributed and that there is no difference between the groups being compared, 95% CIs were used.
A difference between estimates was considered statistically significant if the 95% CIs around the estimates did not overlap. Where there was an overlap between 95% CIs, a 95% CI for the difference between estimates was calculated. To do this, the SE of the difference was approximated by:
The 95% CI for the difference between estimates was then calculated as:
If the 95% CI for the difference between estimates included 0, then the difference was not statistically significant. If it excluded 0, then the difference was considered to be statistically significant.
References
ABS (Australian Bureau of Statistics) 2019. National Health Survey: users’ guide, 2017–18. ABS cat. no. 4363.0. Canberra: ABS.