FILE STRUCTURE


EFFECTS OF SAMPLING

The NNS was conducted using a multi-stage areas sample to select private dwellings and a sub-sample of participants in those dwellings was then selected. As the survey was conducted on only a sample of all households and members of those households, it is important to take account of the method of sample selection when deriving estimates from the unit record file. This is particularly important as a person's chance of selection in the survey varied, depending on the State/Territory and region in which they lived. If these chances of selection are not accounted for, by use of appropriate weights, the results will be biased.


USE OF WEIGHTS

There are six weights on the NNS data file:

• main survey weight (IFIQWT), which contains a weight for each person in the sample;
• FFQ weight (FFQWT), which contains a weight for each person aged 12 years and over who completed a usable FFQ;
• SF36 weight (IFIQSFWT), which contains a weight for each person who completed the SF36 component of the NHS;
• SF36 weight for FFQ respondents (FFQSFWT), which contains a weight for each person who completed the SF36 questionnaire and the FFQ;
• non-SF36 weight (IFIQNSFWT), which contains a weight for each person who completed the non-SF36 components of the NHS; and
• non-SF36 weight for FFQ respondents (FFQNSFWT), which contains a weight for each person who completed the non-SF36 items and the FFQ.

These weights take account of the person's probability of selection in the sample from their region, with adjustments for under-enumeration at the age, sex, part of State level and other factors which affected response.

The estimates conform to independent estimation of the Australian population for the third quarter of 1995. Specifically, the estimates conform to Australian age by sex estimates and Australian State/Territory by part of State estimates. The estimation procedure used response information collected in the course of the survey to counter known biases in target variables resulting from partial response. This information, in the form of models, was used to adjust data for differential response by class, and also to specify weighting classes for applying benchmarks. Target variables for which adjustments were made included household size, income, age, State, marital status and employment status.

If estimates of population sub-groups are to be derived from the CURF it is essential that they are calculated by adding the weights of the person in each category, not just by counting the number of people falling into each category. If each person were to be counted only once then no account would be taken of the fact that a person's chance of selection in the survey varied from region to region and the resulting estimates may be seriously biased.

Since all estimates from the survey are based on a sample, they are subject to sampling and non-sampling error (for further details see Chapter 9 of National Nutrition Survey: Users' Guide, 1995 (cat. no. 4801.0)).


NNS DATA FILE

Main survey weight

The main survey weight (IFIQWT) should be used when analysing any data on the NNS data file except for the data items in the FFQ group.

FFQ weight

The FFQ weight (FFQWT) should only be used in conjunction with FFQ items. An FFQ weight has been calculated for those people aged 12 years and over who completed a usable FFQ because of the voluntary nature of the FFQ. People aged 12 years and over were asked whether they wished to complete the FFQ, and a questionnaire was left with those who agreed. The main NNS weight should be used for all other items on the NNS data file.

Impact of different weights on estimates

As a result of the method of estimating the 'weights', estimates for the same items can vary depending on the weight used. The following table gives population estimates of usual type of diet using IFIQWT and FFQWT. Usual type of diet was collected from all survey participants. The table shows that the total is the same in each case but that the estimate of the number of people within each category varies depending upon which weight is applied.

If the purpose of the analysis is to estimate the number of people on different types of diets or cross-tabulate it against non-FFQ data, then IFIQWT should be used. It is based on the entire survey sample, rather than a sub-sample, and therefore provides more reliable estimates. However, if usual diet is being cross-tabulated against an FFQ item, then FFQWT should be used.

Persons aged 12 years and over: Usual type of diet, Australia, 1995 ('000)

USUAL TYPE OF DIET		IFIQWT	FFQWT
No special diet		10 037	10 016
Special diet
	Vegetarian	529	541
	Weight reduction	711	711
	Diabetic	277	262
	Fat modified	1 923	1 935
	Other	1 492	1 510
Total(a)		14 983	14 983
(a) Total includes not stated responses.

NHS DATA FILE

Users who merge the NNS data file with the NHS data file should take care that they use the appropriate weights. Analysis on NHS items which were not sub-sampled should use IFIQWT and FFQWT as appropriate. There are also four additional weights that have been calculated to take into account sub-sampling in the NHS for respondents to the NNS.

In the NHS, adults who were selected to answer the General Health and Well-being Form (SF-36) were sequenced around questions relating to educational qualifications gained since leaving school, alcohol consumption and health insurance. In addition, females aged 18 years and over from the SF-36 stream were not asked to complete the Women's Health Supplementary Form.

The file therefore contains four additional 'weights' to be used with different population groups. The four additional weights are:

• IFIQSFWT— this should be applied when cross-classifying non-FFQ items with SF-36 items.
• IFIQNSFWT— this should be used when cross-classifying non-FFQ items with education qualifications, alcohol consumption, health insurance and women's health.
• FFQSFWT— this should be used when cross-classifying FFQ items with SF-36 items.
• IFIQNSFWT— this should be used when cross-classifying FFQ items with education qualifications etc.

Those items which are mutually exclusive (e.g. SF36 and educational qualifications) cannot be cross classified and therefore none of the weights are appropriate.

Those users who have separately purchased a copy of the NHS CURF should use the weights on the NNS data file for any weighted analysis of NNS participants. The weights on the NHS CURF have been calculated for the entire NHS sample, not the NNS sample.

STANDARD ERRORS

Since the estimates from the NNS are based on information obtained from a sample they are subject to sampling variability. That is, the estimates may differ from figures that would have been produced if all dwellings had been included in the survey. One measure of the likely difference is given by the standard error (SE), which indicates the extent to which an estimate might have varied by chance because only a sample of dwellings was included.

The Users' Guide contains detailed information on SEs. Standard errors can only be applied to estimates which use population 'weights'.

Specific standard errors have been calculated for person estimates of the whole NNS sample, person estimates of the FFQ sample and estimates of mean food and nutrient intake for the whole NNS sample.

Approximate standard errors for SF36 and non-SF36 items can be obtained by:

• multiplying the whole sample SE by 1.6, if analysing SF36/non-SF36 data against any non-FFQ item; or
• multiplying the FFQ SE by 1.5, if analysing SF36/non-SF36 data against any FFQ item.

For example, if SF36 data is cross-tabulated against Body Mass Index (not part of the FFQ) and an estimate of 200,000 persons is produced, then:

SE [estimate of 200,000 from the whole survey] =16,950 (from p. 146 of the Users' Guide)

Therefore, SE [200,000, SF36 by BMI] = 1.6 * 16,950
= 27,120