Response Reliability in a Longitudinal Survey in Thailand

The 2 rounds of the National Longitudinal Study in Thailand (conducted in 1969 and 1972) provide an unusual opportunity to explore response reliability in a large-scale survey in a developing country. The 2 surveys each covered a national cross-section of the Thai population. The responses of the same individuals for each of the 2 rounds were compared for consistency of response; although the 2 surveys were not identical a number of questions were repeated. For both samples there is a considerable range in reliability. In general reliability was much higher for straightforward questions on age, number of household members, and the like. These questions had reliability ratios of 65 or above. Attitudinal variables had reliability ratios of about 21. A comparison of multiple codings for the 3 items concerning current age, age at 1st marriage, and educational attainment shows that respondents' reliability was 94-100%. This indicates that some of the inconsistencies may have been in data processing coding. Especially disturbing is the low reliability on the number of children ever born. Nonrandom consistency on the question of ideal family size ranged from 19% in rural women to 29% in male household heads. However, the variations did not differ by a response of more than 1 or 2 children. The woman who considered 3 ideal might say 4 or 2 but she did not switch to 1 or 6. Educational levels show little relationship to reliability. Overall, the Thai survey compares well with several leading U.S. surveys. However, the low reliability at the individual level for attitudinal questions should be a matter of concern for those using survey results.