Latent trait and latent class models in survey analysis: Case studies in public perceptions of biotechnology.

In latent variable models the existence of one or more unobserved (latent) variables is posited to explain the associations between a set of observed (manifest) variables. These models are useful for analysing attitudinal survey data, where multiple items are used to capture complex constructs such as attitudes, which cannot be directly observed. In such research they are most commonly applied in the form of factor analyses based on linear regression models. However, these are inappropriate when observed items are categorical, which is often the case with attitudinal surveys. Latent trait and latent class models, based on logistic models, are then more suitable. In this thesis I demonstrate how they can be employed to address common challenges in attitudinal survey research. The case study data illustrating these challenges are from the Eurobarometer survey on public perceptions of biotechnology, fielded in 2002 in fifteen European countries. Using these data I investigate the viability of cross-nationally comparable measures of three central constructs in studies of public perceptions of biotechnology: attitudes towards applications of biotechnology, knowledge of biology and genetics, and engagement with science and with biotechnology. The analyses aim to capture these complex constructs, taking account of 'don't know' responses by including them as categories of nominal observed items, and exploring the comparability of measures of these constructs cross-nationally by assessing the similarity of measurement models between countries. The results of these analyses are informative in three ways: substantively, adding to our knowledge of people's representations of biotechnology; methodologically, increasing our understanding of how the survey items function; and practically, informing future questionnaire design. I also formulate a taxonomy of issues and choices in attitudinal survey research as a conceptual framework through which to discuss more broadly the potential value of latent trait and latent class models in survey research in social psychology.