Methods

Survey sample

In 2008, the target population for the New South Wales Population Health Survey was all residents living in households with private telephones. The target sample comprised approximately 1,500 people in each of the 8 area health services (a total sample of 12,000).

The sampling frame was developed as follows. Records from the Australia on Disk electronic white pages (phone book) were geo-coded using MapInfo mapping software.[1,2] The geo-coded telephone numbers were assigned to statistical local areas and area health services. The proportion of numbers for each telephone prefix by area health service was calculated. All prefixes were expanded with suffixes ranging from 0000 to 9999. The resulting list was then matched back to the electronic phone book. All numbers that matched numbers in the electronic phone book were flagged and the number was assigned to the relevant geo-coded area health service. Unlisted numbers were assigned to the area health service containing the greatest proportion of numbers with that prefix. Numbers were then filtered to eliminate contiguous unused blocks of greater than 10 numbers. The remaining numbers were then checked against the business numbers in the electronic phone book to eliminate business numbers. Finally, numbers were randomly sorted.

Households were contacted using random digit dialling. One person from the household was randomly selected for inclusion in the survey.

Interviews

In 2008, interviews were carried out continuously between February and December. Selected households with addresses in the electronic phone book were sent a letter describing the aims and methods of the survey 2 weeks prior to initial attempts at telephone contact. An 1800 freecall contact number was provided for potential respondents to verify the authenticity of the survey and to ask any questions regarding the survey. Trained interviewers at the NSW Health Survey Program CATI facility carried out interviews. Up to 7 calls were made to establish initial contact with a household, and 5 calls were made in order to contact a selected respondent.

Call outcomes and response rates

In total, 12,485 interviews were conducted, with at least 1,450 interviews in each area health service and 10,296 with adults aged 16 years or over. The overall response rate was 63.4 per cent (completed interviews divided by completed interviews and refusals).

Data analysis

For analysis, the survey sample was weighted to adjust for differences in the probabilities of selection among subjects. These differences were due to the varying number of people living in each household, the number of residential telephone connections for the household, and the varying sampling fraction in each health area. Post-stratification weights were used to reduce the effect of differing non-response rates among males and females and different age groups on the survey estimates. These weights were adjusted for differences between the age and sex structure of the survey sample and the Australian Bureau of Statistics 2006 mid-year population estimates (excluding residents of institutions) for each area health service. Further information on the weighting process is provided elsewhere.[3,4]

Call and interview data were manipulated and analysed using SAS version 8.02.[5] The SURVEYFREQ procedure in SAS was used to analyse the data and calculate point estimates and 95 per cent confidence intervals for the estimates. The SURVEYFREQ procedure calculates standard errors adjusted for the design effect factor or DEFF (the variance for a non-random sample divided by the variance for a simple random sample). It uses the Taylor expansion method to estimate sampling errors of estimators based on the stratified random sample.[5]

For a pairwise comparison of subgroup estimates, the p-value for a two-tailed test was calculated using the normal distribution probability function PROBNORM in SAS, assuming approximate normal distribution of each individual subgroup estimates with the estimated standard errors, and approximate normal distribution for the estimated difference.[5]

References

Australia on Disk [software]. Sydney: Australia on Disk, 2000.
MapInfo [software]. Troy, NY: MapInfo Corporation, 1997.
Barr M, Baker D, Gorringe M, and Fritsche L. NSW Population Health Survey: Description of Methods. Sydney: Centre for Epidemiology and Research, NSW Department of Health, 2008. Available online at www.health.nsw.gov.au/resources/publichealth/surveys/health_survey_method.asp (accessed 27 February 2009).
Steel D. NSW Population Health Survey: Review of the Weighting Procedures. Sydney: Centre for Epidemiology and Research, NSW Department of Health, 2006. Available online at www.health.nsw.gov.au/pubs/2006/review_weighting.html (accessed 27 February 2009).
SAS Institute. The SAS System for Windows version 8.2. Cary, NC: SAS Institute Inc., 2001.

Tables

Source:	New South Wales Population Health Survey 2008 (HOIST). Centre for Epidemiology and Research, NSW Department of Health.
Print version:	Although this page can be printed directly from your web browser, a higher quality version is available as a PDF file that can be printed or viewed on screen.
Produced by:	Centre for Epidemiology and Research, Population Health Division, NSW Department of Health.
Last updated on:	1 March 2009