서베이앤헬스폴리시리서치센터

Kim, S. W., Lee, S, K., Hong, S. J., and Park, S. H. (2009). “List-assisted RDD sampling in Korea: testing the feasibility of national surveys,” Paper presented at the 62th annual conference of the World Association for Public Opinion Research, Lausanne.

Research Background and Summary

In Korea, telephone interviewing has been widely used for social surveys and public opinion polls. By the mid-2000s we mainly used printed telephone directories as official frames, because of a set of regulations designed to protect private information. Recently, a commercial firm began providing computerized telephone directories, and by developing them into national databases of telephone numbers, survey organizations have used them as list frames.

Like other countries such as the USA and the UK, Korea has had statistical concerns about incomplete telephone coverage due to the exclusion of unlisted numbers. Kim et al. (2008) presented that when using telephone directories or databases of telephone numbers as frames, telephone coverage of households in Korea may be slightly lower than 65.6%. It was clear that most surveys would suffer from significant coverage bias.  

Since the mid-1960s, a number of RDD sampling methods for coverage of unlisted numbers have developed in the USA. In addition to the early RDD methods, there could be many alternatives to telephone directory sampling in Korea. The Mitofsky-Waksberg two-stage sampling method may be one of them, but there was growing dissatisfaction with the approach in the past, because of operational and statistical limitations. With the development in the early 1990s of a national database, list-assisted RDD designs have been broadly adopted in the USA. Nicolaas and Lynn (2002) first examined list-assisted RDD in the UK and concluded that it was a viable sampling method.

In Korea, there was much recognition of a serious coverage problem, as noted above, but survey researchers did not react quickly, and little attention has been paid to the area of RDD sampling. Fortunately, Kang et al. (2008) first introduced a form of the early RDD sampling proposed by Cooper (1964), and it encouraged to develop other efficient alternatives. In these situations, we recognized that judging from the experiences in the USA and the UK, list-assisted RDD might be an attractive approach to overcome the current undercoverage problem in Korea.

Of course, there were barriers to the development of list-assisted RDD methods. Korea has a telephone numbering system quite different from the USA or the UK. For example, the length of landline phone numbers is not standardized to 10 digits, and the first three digits in mobile phone numbers do not represent specific geographic areas. Also, telephone companies are haphazard in the assignment of numbers within the blocks of 10000 telephone numbers. In addition, although random sampling is required to compare the findings with other countries, most survey organizations in Korea have used quota sampling for the selection of respondents within households, and accordingly, we had little experience in random selection within a household.

Based on the professional advice of survey sampling statisticians in the USA, we collected a variety of materials for exploring the possibility of list-assisted RDD sampling for Korean telephone surveys. We trained the interviewers for a within-household selection and the CATI system was specially designed for quality control. In 2008, we finally completed a national telephone survey, which targeted on the population of about 16 million households, testing the feasibility of list-assisted RDD designs using 100-banks.

In this paper, we first describe the details of list-assisted RDD sampling designs applied in Korea. Second, we compare the distributions of 100-banks by the number of listed telephone numbers between Korea and the USA. Third, we present the residential hit rates to see the overcoverage problem in RDD sampling. And then we show the distribution of the RDD sample by both the type of telephone numbers and demographic characteristics. Finally, we advocate further research to improve the undercoverage problem due to mobile-only households increased in recent years.