|Page tools: Print Page Print All|
FOOD AND MEASURE CODING
To allow for the coding of foods and measures, and the calculation of nutrients, FSANZ developed a food and measures database. This database contains 5,644 foods and 15,847 measures. Each food within the food database has a name, associated food description, inclusions, exclusions and an 8-digit code. The measures are either densities for the foods (including multiple densities for some foods e.g. chopped, grated or piece) or a gram weight for food specific measures. Food specific measures were developed by FSANZ based on available food sizes (for processed foods or packaged foods such as a bottle of soft drink) or average available food sizes (e.g. small, medium or large banana). More information on the development of these databases is available from FSANZ.
The 8-digit food codes are grouped into broader food groups based on groupings used in 1995 NNS. These Major, Sub-major and Minor groups, along with dietary supplements, form the AHS food classification. This classification was developed by FSANZ in conjunction with the food and measures database to allow for analysis of food, beverage and supplement consumption. The AHS food classification is available as an Excel spreadsheet from the Downloads section of this publication.
For example, a banana smoothie would be coded to ‘Smoothie, cows milk, all flavours, added banana’ and grouped under Milk products and dishes (major), Flavoured milks and milkshakes (sub-major), and Milk based fruit drinks (minor).
Information on the comparability of this classification with the 1995 NNS classification is available in the Interpretation of Results section.
FOOD AND MEASURE CODING
Foods and beverages can be consumed in different ways:
A set of metadata was developed by FSANZ and the ABS to automatically code commonly consumed foods. Given the numerous ways a respondent may report any given food, a set of pathways through the AMPM were mapped out only for common and straightforward foods.
427 different pathways through the instrument were mapped for 282 different foods. 93,409 foods were auto-coded, resulting in 27.47% of foods auto-coded.
Certain combinations of foods were split (disaggregated) into their component parts to simplify coding. The food groups that were disaggregated include:
All interviews from a household were coded together to allow for efficiencies in coding the same foods consumed by multiple members of the household, or the same foods (such as milk brands or margarine types) on both interview days.
Reported foods were matched with a description of food (taking into account specified inclusions and exclusions) from the food database, and associated code. Foods were coded according to food type, cooking methods, ingredients, types of fats used, fat and sugar content and differences in fortification.
Once the food code was selected, measures were used to code the portion size. General measures, either from the food model booklet or common measures (such as teaspoon or mouthful) were converted to a gram amount by multiplying the volume of the measure by the food density. Food specific measures had a gram weight calculated based on the amount.
If a food or portion size was unable to be matched to an existing food in the food database, a “query” was raised. Throughout the coding process these queries were investigated and resolved by the ABS. Where the ABS was unable to resolve a query, it was referred to FSANZ. These queried foods and measures were either matched to an existing food or measure, or new codes were created. New codes were only created where the nutrient make-up was different enough to warrant the creation of a new food within the database, otherwise an existing code description or inclusions would be updated to assist coders and reduce the need for further queries.
QUALITY ASSURANCE FOR CODING
Coders underwent a rigorous training program on the collection instrument and coding process. On commencing food coding, coders were subject to a 100% dual coding quality control process. This process allowed for each coder to have a certain percent of households coded independently and then adjudicated by a supervisor. Once coders reached a 90% accuracy level, the proportion of households dual coded gradually reduced to no lower than 50%. Adjudication allowed each food and measure coding result from two different coders to be compared and for information on any errors to be relayed back to the coders.
1 Raper N, Perloff B, Ingwersen L, Steinfeldt L, Anand J 2004, ‘An overview of USDA’s dietary intake data system’ Journal of Food Composition and Analysis 17(3-4): 545-555, Available from <http://www.sciencedirect.com/science/article/pii/S0889157504000511>. Back
Back to top