Is Simple Random Sampling Better than Quota Sampling? An Analysis Based on the Sampling Methods of Three Surveys in South Korea

This paper considers whether random sampling always produces more accurate survey results in the case of South Korea. We compare information from the 2010 census to the demographic variables of three public opinion surveys from South Korea: Gallup Korea's Omnibus Survey (Survey A) is conducted every two months by Gallup Korea; the annual Social Survey (Survey B) is conducted by Statistics Korea (KOSTAT); the Korean General Social Survey (KGSS or Survey C) is conducted annually by the Survey Research Center (SRC) at Sungkyunkwan University (SKKU). Survey A uses quota sampling after randomly selecting the neighborhood and initial addresses; Survey B uses random sampling, but allows replacements in some situations; Survey C uses simple random sampling. Data from more than one year was used for each survey. Our analysis suggests that Survey B is the most representative in most respects, and, in some respects, Survey A may be more representative than Survey C. Data from Survey C was the least stable in terms of representativeness by geographical area and age. Single-person households were underrepresented in both Surveys A and C, but the problem was more severe in Survey A. Four-person households and married persons were both over-represented in Survey A. Less educated people were under-represented in both Survey A and Survey C. There were differences in income level between Survey A and Survey C, but income data was not available for Survey B or the census, so it is difficult to ascertain which survey was more representative in this case.