Assessing latent dimensionality in psychological research

Research in cognitive science and neuroscience has the shared goal of understanding how cognitive and neural representations and processes mediate the observed relationships between stimuli and responses in different experimental paradigms. The almost ubiquitous basis for inferences about the number of processes or latent dimensions involved, is the observation of a dissociation; an interaction due to an unequal or opposite effect of one independent variable on the levels of another independent variable. However, it has been clearly shown that dissociations do not provide strong evidence for the need of an extra dimension. In this thesis, which is a collection of published and submitted papers, we describe an extension of the dissociation methodology – state-trace analysis (Bamber, 1979) – that does provide a rigorous basis for this inference. In the first section, an informal introduction to state-trace analysis is provided. We also develop Bayesian methods suitable for quantifying state-trace evidence in favour of a one-dimensional or multi-dimensional explanation, as well as for refining state-trace experiments. In the second section of this thesis, an application of state-trace analysis is presented that examines the question of whether human face recognition is special in the sense that faces can be encoded in terms of a dimension or dimensions additional to those available to most other objects. Over a series of experiments, and using the new methods developed in Section One, we confirm that the encoding of unfamiliar faces is special and discuss the need to extend this type of analysis to other psychological phenomena.