Apparent Consumption of Selected Foodstuffs, Australia methodology

This is not the latest release View the latest release
Reference period
2018-19 financial year
Released
26/06/2020

Explanatory notes

Introduction

1 The Apparent Consumption of Selected Foodstuffs, Australia, 2018–19 (cat. no. 4316.0) is intended to provide information on trends in food consumption, nutritional adequacy of the food supply and impacts of changes to food supply. The data is complementary to the more detailed information available from national nutrition surveys, but with the advantage of being an ongoing (annual) collection that can be produced in a relatively timely and cost-effective manner. This helps to address a critical data gap relating to food and nutrient consumption required to inform policy makers, regulators, health researchers and the food industry.

2 This is the first release of the Apparent Consumption of Selected Foodstuffs (ACSF) collection since the previous collection ceased in 2000 (1998–99 reference year) due to data quality concerns around fitness for purpose of previous data sources used for estimating the availability and consumption of certain foodstuffs. A review at that time found the costs and complexity of addressing the data quality concerns to be too prohibitive to resolve within available funding and competing statistical priorities.

3 The relatively recent innovation of using transactions 'scanner' data to improve the collection efficiency and data quality of measuring consumer price change for the quarterly Consumer Price Index (CPI) offers the opportunity to extract further value from this information through the development of a new ACSF collection. The development of these estimates was supported and funded by the Department of Health to help inform Australian food nutrition policy and research. The use of sales data to measure apparent consumption is a key difference from the previous ABS Apparent Consumption of Foodstuffs collection which used the traditional approach based on production plus imports minus exports.

4 The apparent consumption of selected foodstuff (ACSF) estimates in this publication are comprised of two components:

  1. Directly calculated measures of food quantities available for consumption from aggregated sales data provided by the major retailers using Scanner Data (SD), and
  2. Indirectly estimated measures of the food quantities available for consumption that are not captured by the major retailers in the SD which are based on household expenditure data.
     

Data sources

Scanner data

5 The primary data source for this publication are the aggregated scanner data (SD) provided to the ABS from major supermarkets. The major supermarkets provide aggregated data based on information compiled from barcode scanning at the point of sale. The aggregate information comprises the Stock Keeping Unit (SKU) of the sales item, a description of the item, the geographical region and weekly aggregate amounts of the quantity and value of the sales. The major supermarkets that provide data to the ABS account for an estimated 82% of the Food Retail sector.

2015–16 Household Expenditure Survey

6 The 2015–16 Household Expenditure Survey (HES) is used to help estimate and impute the value of product sales which are not captured by the major supermarkets in the SD (that is, the remaining 18% of foodstuffs available for consumption in the Food Retail sector). For example, food purchases made at convenience stores, butchers, fish shops, bakeries, delis and vegetable markets. Expenditure data from the HES are used to estimate the ratio of expenditure in stores not represented by SD to expenditure from the supermarkets in scope of SD (for a given product category in a given geographical area). No adjustment is made to the 2015–16 HES data to align to the 2018–19 ACSF reference period.

Scope

7 The scope for this publication is all food and non-alcoholic beverages purchased for consumption in Australia from the Food Retail sector. This includes all food and non-alcoholic products assigned to the food and non-alcoholic beverage expenditure classes (ECs) used to compile the CPI with the exception of Restaurant meals and Take away and fast foods. See CPI's 17th Series Expenditure Weights for the complete list of ECs.

8 The 2015–16 HES collects expenditure from households representing about 97% of people in Australia. In expenditure terms, the 2015–16 HES estimated an average weekly expenditure of $237 per household on Food and non-alcoholic beverages. Of this, 34% ($80) was estimated to be on Meals out and fast foods, leaving 66% ($157 in 2015–16) of total food expenditure as in-scope in the ACSF.

9 All in-scope foods are coded to the AUSNUT 2011–13 database which was developed for the 2011–13 Australian Health Survey. Assignment of AUSNUT codes provides the ACSF with an established food classification and ability to group AUSNUT codes into useful food groups for analysis. In addition, AUSNUT 2011–13 contains comprehensive nutrient data for each food and non-alcoholic beverage product and links directly to other established metadata including information about the amounts (serves or grams) of Australian Dietary Guidelines (ADG) Food Groups. This alignment means that conceptually the data can be used for similar food and nutrient analyses as a national nutrition survey. There are however important differences which should be considered due to the differing nature of foods as purchased (scanner data) and food as eaten (nutrition surveys).

10 It should be noted that estimates of 'apparent consumption' of food and non-alcoholic beverage products are derived using information related to sales (that is, food available for consumption from purchases made from the Food Retail sector). Therefore, they are at best a partial representation of the apparent consumption due to scope limitations (see ‘Out of scope’ below). Furthermore, ‘apparent consumption’ methodologies typically don’t adjust for waste (see below) or changes in household inventories i.e. all food and non-alcoholic beverages available for apparent consumption (purchased) in a particular year is assumed to have been consumed in that year.

Out of scope

11 While SD (when used with the HES) can represent the majority of the food available for consumption in Australia, there are clear gaps where either SD sales or 2015–16 HES do not adequately capture representative food and non-alcoholic beverage amounts. These are:

  • Fast foods, cafe and restaurant meals
  • Meals provided by institutions that source food from the non-retail sector
  • Home grown or produced food
  • Wild harvested/hunted bush food or seafood.
     

12 The exclusions in the HES sample comprise of:

  • Very Remote areas
  • Residents of non-private dwellings (e.g. hotels, boarding schools, nursing homes and prisons)
  • Households containing members of non-Australian defence forces or diplomatic personnel of overseas governments stationed in Australia
  • The exclusions associated with the HES survey should have minimal impact on the indirect estimated measures compiled to represent the food quantities available for consumption from the non-major supermarkets not captured in the SD provided to the ABS.
     

Alcohol

13 Comprehensive alcohol data are not currently available from scanner data. Although the annual Apparent Consumption of Alcohol (ACA) collection provides national annual estimates for beer, wine and spirits, the ACA data are not incorporated in the ACSF because of differences in scope and coverage. The alcohol estimates from the ACA represent all alcohol available for consumption in Australia, while the ACSF is limited to just the foods available for consumption in the Food Retail sector (less Restaurant meals and Take away and fast foods). Including ACA estimates in the ACSF would present an over-estimate of the quantity of alcoholic beverages available as well as the relative contribution of alcohol to macronutrient energy. A further misalignment of the data sources results from the different reference periods with the latest ACA data for 2017–18, 12 months prior to the ACSF reference period.

Waste

14 The estimates presented in this publication are not adjusted for possible product wastage. The most recent estimate of total organic food wastage in Australia was around 3.1 million tonnes for 2016–17. This is the equivalent of around 350 gram per person per day (or 2.4 kg per week). While it may be expected that all perishable goods will be susceptible to a degree of waste, the waste estimates cannot provide further breakdowns, for instance how much is vegetable and fruits and how much is baked goods, dairy produce, various meat products etc. Given most discretionary foods are packaged (and have a longer shelf life) they may be less likely to be disposed of than fresh produce. Such an imbalance in waste suggests greater care should be taken in interpreting the apparent consumption of non-discretionary foods and therefore the relative balance between discretionary and non-discretionary foods. In addition different fresh foods may have different edible portions (e.g. grapes have a higher edible portion than pineapple) and no adjustment has been made to subtract the inedible portion which may lead to further over-estimation of certain fresh products.

Geographical coverage

15 The SD provides aggregate data representing all regions of Australia serviced by major supermarket chains. Estimates in this publication are presented at the national level. While state and sub-state estimates are feasible, they are not included in this release.

Reference period

16 The first release is for the financial year 2018–19. Because this is the first release of the new ACSF series covering just one year, trend information will only become available with subsequent releases.

Methods

AUSNUT coding

17 The process of AUSNUT coding involves assigning each food item SKU to an AUSNUT code and a size (weight or volume) based on the product description. AUSNUT coding has been performed on around 95% of the total value of SD foods in the 2018–19 period, equivalent to over 100k unique SKUs.

18 As noted above, the AUSNUT 2011–13 classification was designed to code foods as consumed, whereas many foods in the SD require preparation. Although the great variety and detail of AUSNUT focuses on foods in their prepared forms, AUSNUT also includes codes and nutrient data for the unprepared varieties (e.g. uncooked pasta, flour, jelly crystals, uncooked cuts of meat) and these were applied as appropriate. The general limitation of coding is the availability of specific and accurate codes to handle the variety of unprepared foods and the changing nature of the food supply. For example, where manufacturers have reformulated their products to reduce sugar or sodium content, the available AUSNUT code used may not reflect the most up-to-date nutrient profile.

Coding weighting adjustment

19 Weighting has been applied to adjust for the food items (value and quantity) that were unable to be coded. The weight (or coding ratio) is a value which indicates the number of products in the Food Retail sector represented by the coded SD product. Each coded SD food item is assigned a weight, that is applied at the EC, week, Greater Capital City Statistical Area (GCCSA) and supermarket chain level. This level is used so reliable denominators can be calculated. This means that each food item within an EC/week/GCCSA/retailer will have the same weight, even though at a finer product category level there may be differences in coding coverage.

20 Therefore, the two principles and one assumption in this adjustment are:

  1. The adjustment is important to overcome the bias that would result without it (because coding ratios have changed over time and vary between product groups)
  2. Adjustments are relatively minor because coding rates are relatively high (~95%) and therefore minimising introduction of new bias when removing target bias
  3. The coding ratio of each EC is representative of the foods within that EC.
     

Missing ECs

21 Around 3% of the value of foods in the SD did not have an assigned EC. These products were grouped in their own EC and the same coding weighting adjustment was applied as described above. This group of ECs is more heterogenous, than the identified EC groupings, so the coding ratio of this group may not be fully representative of the different foods within it.

Estimation

22 As previously mentioned, there are two components to the estimates of apparent consumption of foodstuffs presented in this publication. The components are:

  1. Directly estimated component. This is simply the quantity of each item multiplied by the size (g or mL) of the item in the SD.
  2. Indirectly estimated component. This uses 2015–16 HES data to derive an expenditure ratio of food items from the non-major supermarkets to the major supermarkets to estimate the food quantities available for consumption that are not captured in the SD. This expenditure ratio is then applied to a grams per dollar value to derive the estimated quantity (weight or volume) of that food and non-alcoholic beverages available for consumption from the non-major supermarkets.
     

23 The adjustment for the indirectly estimated component may be summarised as a four-step process:

  1. Map 2015–16 HES food and non-alcoholic beverage expenditure codes to AUSNUT 2011–13
  2. Estimate the Food Retail sector expenditure ratio from non-major supermarkets to the major supermarkets by each food group by region from the data reported in the 2015–16 HES. This step provides the expenditure ratio for a defined food group for each GCCSA
     

This is expressed below where i = a given food group in a given GCCSA geographical area, MS represents major supermarkets and NMS represents the food retailers from non-major supermarkets.

\(\large{Coverage \ weight _{i}=\frac{\sum w_{i} \times E x p_{i}(N M S)}{\sum w_{i} \times E x p_{i}(M S)}}\)

  1. Derive amounts (grams or mLs) per dollar for each product based on the known prices per product size in the SD
  2. Apply the amounts per dollar (from step 3) to the expenditure ratios (from step 2) and multiply by the observed sales values in the SD to derive the imputed amounts in grams or mLs of food and non-alcoholic beverages available for consumption from non-major supermarkets.
     

Limitations of method

24 Three main limitations have been identified which will impact the validity of the expenditure ratios. These are:

  1. Loss of fine product detail in mapping of HES expenditure codes (HEC) to AUSNUT codes
  2. Representation of identified stores in the subset of HES data available to estimate proportion of expenditure from the major and non-major supermarkets (and that it represents a period 3 years prior to the ACSF reference period)
  3. Estimation of the amount (grams or mLs) per dollar - of food represented in the SD.
     

Mapping of HES expenditure codes to AUSNUT codes

25 The aim of this step is to define common groups of products that represent broad types of food and non-alcoholic beverages groups in terms of their propensity to be bought from the major supermarkets or other retail outlets.

26 Alignment of 2015–16 HES food expenditures with the SD was performed by creating a correspondence which maps food and non-alcoholic beverages from both the HES (124 relevant codes) and AUSNUT (5,740 codes) to common food groups. After collapsing a number of HES codes which were not similarly split in AUSNUT, each classification was mapped to a common set of 63 defined food groups (see the 2011–13 AUSNUT to 2015–16 HEC correspondence table available from the Data downloads section).

Representation of identified stores in the subset of HES

27 The HES was used to provide estimates of the expenditure ratios (by food group and geography) of the in-scope supermarkets. However, the receipt information identifying the store at which food purchases were made in the 2015–16 HES was not an essential survey input data item, and only captured incidentally (where the respondent retained the receipt to assist in recording expenditures). While almost three-quarters (74%) of relevant food and non-alcoholic beverages expenditure had a store receipt which identified the store, this subset is likely to over represent the transactions where the respondent purchased multiple grocery items. This led to a moderate correlation between a food item having a receipt and that food being purchased at one of the major supermarkets in-scope of the SD. With no further adjustment made, this association may indicate some bias favouring expenditures from the major supermarkets. Such a bias may tend to underestimate the expenditure ratios for foods coming from from specialty shops (e.g. butchers, bakeries, vegetable markets or seafood shops).

Deriving a gram or mL amount of food or beverage per dollar

28 Translating the expenditure ratios to amounts (grams or mLs) of food per dollar was performed by:

  1. Deriving a median amount (grams/mLs) per dollar at the AUSNUT code level, based on all contributing SKUs to each AUSNUT code
  2. Applying that gram/mL per dollar amount to the expenditure ratio to estimate the imputed amount of each food. In practice, foods being imputed are aggregates of AUSNUT coded foods, so the imputation uses weighted aggregates of the applicable AUSNUT code medians.
     

29 The major limitation of this method is the assumption that amounts per dollar from major supermarkets are likely to represent the amounts for other retailers (e.g. convenience stores and specialty stores). It is more likely that for many products the major supermarkets provide greater amounts per dollar on equivalent items.

30 By using the median, it assumes the midpoint value is representative of the aggregate amount (g or mLs) per dollar across all products in the defined food group. Analysis of the SD shows some amounts (g or mLs) per dollar distributions are bi-modal, with product items concentrated around small sized products (lower g or mL per dollar) and bulk products (higher g or mL per dollar). Therefore, if the expenditure in non-major supermarkets includes less of the bulk packages then the imputed amount (gram or mLs) per dollar may be an overestimate.

Per capita estimates

31 Daily per capita estimates are derived by dividing the annual total by 365 days and dividing by the Estimated Resident Population (ERP). Data representing the 2018–19 period uses the ERP for December 2018. No adjustments have been made to the ERP for food and non-alcoholic beverages that are typically consumed by selected age groups, for example baby food.

Measuring Australian Dietary Guidelines food groups

32 Because many food products are mixtures of food groups, measuring the total amounts of each of the food groups utilises an Australian Dietary Guidelines (ADG) database which measures the amounts of each food group in each AUSNUT food. The ADG database used was originally developed by FSANZ for the 2011–12 NNPAS. For more information, see Assessing the 2011–13 AHS against the Australian Dietary Guidelines - Classification System and Database Development Explanatory notes, available from https//www.foodstandards.gov.au/science/monitoringnutrients/australianhealthsurveyandaustraliandietaryguidelines/Pages/default/aspx.

Reliability of estimates

33 The estimates presented in this publication are subject to error which can be broadly categorised as either sampling error or non-sampling error. For more information refer to the Technical Note.

Technical note

Reliability of estimates

1 The apparent consumption of selected foodstuff (ACSF) estimates in this publication are comprised of two components:

  1. Directly calculated measures of food and non-alcoholic beverage quantities available for consumption from aggregated sales data provided by the major retailers using Scanner Data (SD), and
  2. Indirectly estimated measures of the food and non-alcoholic beverage quantities available for consumption that are not captured in the SD which are based on household expenditure data.
     

2 These two components are subject to two types of error, non-sampling and sampling error.

Non-sampling error

3 Non-sampling error can occur in any collection, whether the estimates are derived from a sample, administration data or from a complete collection such as a census. Sources of non-sampling error include non-response, errors in reporting by respondents or recording of answers by interviewers, and errors in coding and processing the data.

4 Non-sampling errors are difficult to quantify in any collection. However, every effort is made to reduce non-sampling error to a minimum by careful design and testing of the questionnaire, training of interviewers and data entry staff, and extensive editing and quality control procedures at all stages of data processing. Further information on methods adopted to reduce the level and impact of non-response is given in the Explanatory Notes of the 2015–16 Household Expenditure Survey (HES).

Sampling error

5 The directly calculated component of the ACSF estimates is based on administrative data source and by nature does not have an associated sampling error.

6 The indirectly estimated component of the ACSF estimates is derived from the 2015–16 HES to estimate the food and non-alcoholic beverage quantities available for consumption by the non-major retailers. As the HES is a household survey, the data used to estimate the food quantities available for consumption by the non-major retailers are based on a sample and are subject to sampling variability. The HES estimates may therefore differ from the figures that would have been produced if expenditure information had been collected for all households.

7 One measure of the sampling uncertainty is given by the standard error estimate (SE), which indicates the extent to which a sample estimate might have varied compared to the population parameter because only a sample of dwellings were included. There are about two chances in three that the sample estimate will differ by less than one SE from the population parameter that would have been obtained if all dwellings had been enumerated, and about 19 chances in 20 that the difference will be less than two SEs.

8 The SEs for estimated amounts (volumes and weights) have been calculated using a delete-a-group jackknife method which take account of the survey design. As the proportions in this publication have been calculated after the processing period of the 2015–16 HES collection, the delete-a-group jackknife method can not be used for calculating SEs of proportions. However, from https://www.stat.cmu.edu/~hseltman/files/ratio.pdf, assuming the covariance term is zero, we can approximate the SE by the following formula:

\(\large{S E\left(\frac{X}{Y}\right)=\frac{X}{Y} \sqrt{\frac{V a r(X)}{X^{2}}+\frac{V a r(Y)}{Y^{2}}}}\)

9 For this publication, relative standard errors (RSE) has been provided for estimated amounts (volumes and weights) and Margins of Error (MoE) have been provided for proportions and are available from the Data downloads section. These measures of sampling error are typically small as the indirectly estimated measures contribute a small component of the overall ACSF estimate. The RSE is obtained by expressing the SE as a percentage of the estimate, while the MoE, which describes the distance from the population value that the sample estimate is likely to be within, is calculated as 1.96 multiplied by the SE. The RSE and MOE can be calculated by the following formula:

\(\large{RSE}(X)=\frac{S E(X)}{X}, {MoE}(X)=1.96 \times {SE}(X)\)

10 The MOEs in this publication are calculated at the 95% confidence interval. This can easily be converted to a 90% confidence level by multiplying the MOE by: 1.645/1.96 (for example, see Reliability of Estimates in National Health Survey: First Results, 2017–18) or to a 99% confidence level by multiplying by a factor of: 2.576/1.96.

11 A confidence interval expresses the sampling error as a range in which the population value is expected to lie at a given level of confidence. The confidence interval can easily be constructed from the MOE of the same level of confidence by taking the estimate plus or minus the MOE of the estimate.

Glossary

Show all

Quality declaration

Institutional environment

Relevance

Timeliness

Accuracy

Coherence

Interpretability

Accessibility

Back to top of the page