DATA QUALITY Although care was taken to ensure that the nutrition results of the 2011-12 NNPAS are as accurate as possible, there are certain factors which may affect the reliability of the results and for which no adequate adjustments can be made. Factors specific to the nutrition results of the 2011-12 NNPAS are discussed below, and should be considered in conjunction with the Users' Guide's general Data quality and interpretation of results chapters.

QUALITY ASSURANCE

Throughout the coding and processing of the 2011-12 NNPAS, efforts were made to remove error from the data, and to ensure that the highest quality estimates were produced. This process required consideration of how clearly incorrect the data was, and in many cases unusual records were left unchanged as there was insufficient information to warrant change. Changes were only made to data when it was clear that it was erroneous in order to avoid introducing bias into the results. Given dietary intake data is self-reported, other sources of bias may be present. Please see Under-reporting in Nutrition Surveys for further information.


Coding Quality Assurance

In addition to the quality assurance processes outlined in Food and Measure Coding, further quality assurance conducted throughout the coding process.

• Errors identified during coding from collected data were corrected where possible, with decisions documented and reviewed.
• Field interviewers were able to attach remarks to allow for clarification or correction of the data entered and these remarks were reviewed and addressed.
• Through continual validation of coded files, FSANZ identified ad-hoc corrections to the data, ensuring consistency, and taking into account food codes created post coding and query resolution.

Imputation


As the coding and quality assurance process took over 12 months to complete for 380,000 intake records the final quality of the data was unable to be determined in time for the initial sample weighting process. This has led to a situation where there were 240 respondents with sufficient survey (i.e. demographic and physical activity) data being included in the weighting specification and published in other AHS publications, but intake data appeared to be corrupted during collection and thus there was no usable intake data. These 24-hour dietary recall records had full intakes (including supplements ) imputed to allow for analysis of both foods and nutrients. Donor records were identified by finding the best match on a set of variables known to influence intakes (Age group, Sex, State, Marital status, Employment Status, Country of Birth, BMI and Season). Estimates were produced including and excluding the imputed records to ensure no errors were introduced.

Food Validation

Targeted validation of both food and measure coding was conducted to identify both ad-hoc and systemic errors in the data. With approximately 350,000 foods in the final coded file, targeted validation of foods was an effective way to reduce potential large errors in aggregate data in an efficient manner. Focus was placed on foods known to have collection or coding issues, and foods likely to contribute proportionally large nutrients to a diet. These included:

• powdered mixes (e.g. hot chocolate, gravy) were known to contain errors in reporting. Prepared mixes are often reported as the dry equivalent, which overestimates the nutrient content (particularly for fortified products).
• collecting and coding the correct type of milk was problematic, with insufficient detail provided in many cases on the fat content.
• sandwiches were disaggregated to aid in coding, and this created issues with the correct recording of spreads and fillings. Targeted validation of sandwiches was conducted to resolve these issues.
• additions to foods (i.e. foods added when serving a dish) were often misreported. Many of these foods are included in the coded main food, and the additions were deleted to ensure they were not double counted.
• foods reported as the “Same as” a previous meal or another member of the household were not able to be addressed in a satisfactory way during coding, and were validated individually to ensure accurate data.
• other foods with known errors in reporting or coding included that were targeted for validation included: soft drinks with ice, flavoured and herbal teas, cordial and pasta.

Frequency counts of common food codes and food paths through the AMPM instrument were run and analysed to ensure that systemic issues with coding the most common foods were not present.


Portion Validation

As with targeted food validation, targeted validation of portions was an effective way to reduce potential large errors in aggregate data in an efficient manner. Targeted portion validation included:

• common reporting errors (e.g. 15 bananas reported instead of 1.5) were identified and corrected. Other systemic issues with pizza slice sizes, tea measures (allowing for the addition of milk without the mug overflowing), meat densities (i.e. density with and without bone, or diced or sliced meat) and espresso volumes (reports of full mugs of espresso) were all investigated and corrections made where required.
• large volumes of liquids for individual foods and for a person’s total daily intake were investigated.
• large gram weights for individual foods were assessed based on how large an outlier they were.

Nutrient Validation

Once nutrient values were applied to the coded foods, total amount of nutrients per reported portion of food were calculated. These were then aggregated up to provide a figure for consumption of the selected nutrients per person per day (where 2 days of data were available).

Data was validated at the person level, and compared to expected daily intakes for each of the 44 nutrients, for each age and sex group for both days of data (where available). Both ends of the distribution were investigated, with any high or low values interrogated. Expected levels of consumption were calculated based on NRVs for each nutrient, where available.

Alcohol intake was validated against the 2011-12 NHS and Apparent Consumption of Alcohol, Australia 2011-12, to investigate consistency of results. The validation showed that the proportion of persons consuming alcohol on the day before the interview (by sex and age), was consistent with NHS data. The data also supports the long term trend shown in Apparent Consumption in that beer consumption has fallen and wine consumption has increased between 1995 and 2011-12. There are some differences in data about alcohol beverage intakes, due to the difference in collection methods between the NNPAS and NHS.

Population group aggregate levels of nutrients were compared to 1995 NNS data, and where possible to other relevant ABS data (e.g. 2009-10 Household Income and Expenditure Survey data).


Back to top

Previous Page

Next Page