Telephone survey methods: implications of the increasing mobile‐only population for public health research

Telephone surveys are a standard research method in public health, yet telephone technologies are changing and public health researchers need to be aware that this affects recruitment, response and representativeness of surveys. In particular, the almost universal ownership of landline telephones is declining through ‘landline substitution’ and there is an increasing ‘mobile-only population’ who have given up – or never take on – a landline. Telstra reports this trend to be accelerating at “disturbing speed”. 2 There are few data on Australian mobile phone ownership and distribution. Limited data suggest that 14% of adults are 'mobileonly', although in the US 25% of homes now have no landline. This trend has significant implications for research. For example, one health promotion study gave results from a random telephone survey but did not clarify whether the conclusions applied to the 'landlineonly population', or whether 'mobile-only' adults were included. Importantly, US data show that the 'mobile-only population' differs significantly from the 'landline-only' on socioeconomic characteristics, with the largest 'mobile-only' groups being adults living with non-related adults (63% 'mobile-only'), adults aged 2529 years (49% 'mobile-only'), and adults in rental properties (43% 'mobile-only', compared with 14% of homeowners). Mobile-only' adults are also more likely to have financial barriers to health care and have no usual medical service. Our qualitative Australian research showed approximately one in three adults from disadvantaged backgrounds were 'mobile-only', while other South Australian analysis finds 'mobile-only' households more predominant among young people, the unemployed and low income households, rural residents, smokers and asthmatics. On this basis, a study to increase the use of a NSW smoking Quitline, which used random recruitment from the telephone directory, probably missed the mobile-only population despite its higher proportion of smokers. Our study also showed that landline and mobile phones differ in ways that affect research. For example, participants deemed a landline phone a household object, but a mobile phone a personal object on which they disliked accepting calls or texts from unknown sources. Furthermore, people may easily find a private place to talk on a landline but could be anywhere when a mobile call is received, resulting in increased refusals to answer. This reflects the lower response rates and higher refusal rates in US national surveys using sampled mobile numbers. Although researchers must be experiencing these challenges, only a few have published on alternative methods. One Australian study used a dual-frame method, recruiting landline users via the White Pages and 'mobile-only' users from phone use indicated in previous surveys. The US national Behavioral Risk Factor Surveillance System was expanded from landline-based random digit dialling to a dual-frame survey, which helped obtain valid, reliable, and representative data, although “biases resulting from exclusion of mobile-only adults from the landline-based survey were found for 9 out of the 16 health indicators”. Furthermore, the 'mobile-also population', who can afford both landline and mobile, have a greater chance of selection if numbers are combined into one sampling frame. Landline survey data can be weighted by population characteristics or with proportional quota sampling, but data may be unrepresentative if they exclude 'mobile-only' groups with completely different characteristics. Incentives can also boost mobile response rates, yet until call costs to mobiles decrease in Australia, researching an increasing 'mobile-only' population could be prohibitively expensive. Research shows calling mobiles can cost at least double that of calling landlines, although for some groups it can make contact more effective. I believe three key changes are required: First, we need national and state data to monitor the size and characteristics of phone populations (individual and household). A new question in Australia’s current face-to-face National Health Survey will at least identify households without landlines. Second, we need more well-designed trials and publication of alternative phone methods. Third, we need research to clearly detail methods, including the types of phone numbers sampled and whether the method under-represents or misses particular groups. This is especially important since health is distributed unevenly across the socioeconomic gradient and unevenly in relation to phone ownership. Since telephone-based surveys are likely to remain an important tool in public health research, researchers must stay aware of technology trends and their differential population distribution, and consider their impact when designing and reporting research.