Abstract

In the theory of total survey error (TSE), the deviation between a value observed in a survey and the true theoretical value that characterizes the population is modeled. In this study, a new aspect of describing the precision of values found in surveys is discussed: how much the values observed in a survey differ from the values that could be obtained in a replication of the survey. The first finding of the study is that the total difference between the responses obtained in two replications of a survey can be decomposed into nonresponse uncertainty (NU) and measurement uncertainty (MU). The former is related to the fact that the completed samples of the surveys are usually different, while the latter is related to the same individuals giving different responses in the replications. The magnitudes of NU and MU are presented for factual and attitude ordinal scale variables in two replications of the European Social Survey. The second finding is that the magnitude of MU is especially relevant: respondents seemed to be inconsistent overall in their answers at the individual level even in the case of factual variables where real change is logically not possible. The phenomenon of MU can be attributed to regression to the mean (RTM). Although RTM is well known for normal distributions, we show that this phenomenon also occurs for finite ordinal-scale variables. The study presents two different models of MU, which show that it can be well described with the Shannon's theory of entropy.