Joseph Daffy
Joseph Daffy is Data Science Team Leader at the Social Research Centre. Joseph specialises in R-Shiny, TWINS address cleaning software, automation and finding resourceful and efficient coding solutions. Joseph leads sample processing and production for A-BS projects across the business – a key component of this role is refinement of existing methods and approaches to maximise the value of this sampling frame.
Joseph has a Bachelor of Commerce (Actuarial Studies) and Graduate Diploma of Science (Statistics), both at the University of Melbourne.
Sessions
Contemporary Australian social research—particularly for specific localities—faces a variety of challenges. The rapid decline in landline use and decline in telephone survey production rates means that random digit dialling (RDD) telephone surveys are impractical for all but the largest states. Although nonprobability online access panels offer rapid and inexpensive access to sample, error is greater than probability samples and regional quotas may be hard or even impossible to meet. Face-to-face surveys have become more challenging as a result of the COVID-19 pandemic, with the Australian Bureau of Statistics increasingly transitioning away from this mode.
Address-based sampling (A-BS) was developed in the U.S. as a response to the challenge of declining landline coverage. Although it was eclipsed in popularity by dual-frame RDD as a response to this challenge, it has proved to be enduring in the longer-term in the face of declining telephone survey production rates. In this paper, we address considerations in the conduct of A-BS surveys in Australia. A-BS surveys are largely insulated from the decline in telephone survey production by the use of mail as a primary means of contact, typically in sequential multi-mode design (push-to-web then hard copy).
The Geo-coded National Address File (G-NAF) is used as the sampling frame for A-BS surveys fielded by the Social Research Centre (SRC). The G-NAF is the authoritative source of information on physical addresses in Australia. It combines address information from multiple official sources, including state and territory land agencies, the electoral roll and Australia Post in one dataset. It is managed by Geoscape Australia (previously the Public Sector Mapping Authority) and accessible under open data terms through data.gov.au. We provide an overview of how we draw samples from the G-NAF, including:
• Reconciling records from multiple sources.
• Validating the address is specific enough to be delivered to a single residence (as opposed to a block of apartments).
• Appending Australia Post Delivery Point Identifiers (uniquely identifies a physical point to which Australia Post delivers mail).
• Washing non-residential addresses (factory, car space, office, shed, etc) and deduplicating addresses.
• Statistical sampling to achieve an even distribution within strata.
• Using expected stratum-level response rates to estimate the final achieve sample.
We also describe field methods. The SRC has used two different protocols with differing levels of response maximisation based around engagement and reminder activities.
Finally, we address strengths and weaknesses of A-BS using the G-NAF. Strengths include higher response rates than are possible under a telephone design, likely greater longevity than telephone surveys and ability to target small areas. Weaknesses include high cost (particularly for designs that seek to maximise response rates), over-representation of older adults, the more highly educated and women, extended time in field and difficulties implementing within-household selection procedures.
Recording link: https://acspri-org-au.zoom.us/rec/share/sFdBbpep0OoacPzm634BiKe4bVY9lbHVDFWPYTVAhPA0z5bW5jP-TMNaiTTN1hdG.b2WvY5pK2mJq3Prh?startTime=1669172420000