Accounting for the Effects of Data Collection Method

Application to the International Tobacco Control Netherlands Survey

Mary E. Thompson  By the same author

      Y. Celia Huang  By the same author

      Christian BOUDREAU  By the same author

      Geoffrey T. Fong  By the same author

      Bas Van Den Putte  By the same author

      Gera E. Nagelhout  By the same author

      Marc C. Willemsen  By the same author


Mixed mode surveys are becoming increasingly common. This has led to calls for tests of the differences in response patterns between survey modes. In this article, we present an analysis of mode effects, using data from Wave 1 of the ITC Netherlands Survey, conducted by web (CAWI) and telephone (CATI). For many of the questions, the web and telephone samples differed in the distribution of response options. This was found to be partly attributable to selection effects, since the web and telephone respondents were recruited in different ways, and the web and telephone samples differed on demographic characteristics. Another source of difference in the response option distribution was “administrative” in origin, having to do with the tendency of respondents to process the options differently depending on survey mode. This article illustrates an approach to modelling in a mixed mode survey that takes into account both selection and administration mode effects. The model is also embedded in an analysis of reactions to labels on cigarette packages using ITC data from the Netherlands, Germany, France, and the United Kingdom.