Diffusion versus linear ballistic accumulation: different models for response time with different conclusions about psychological mechanisms?

Two similar classes of evidence-accumulation model have dominated theorizing about rapid binary choice: diffusion models and racing accumulator pairs. Donkin, Brown, Heathcote, and Wagenmakers (2011) examined mimicry between the Ratcliff diffusion (RD; Ratcliff & Smith, 2004) and the linear ballistic accumulator (LBA; Brown & Heathcote, 2008), the 2 least similar models from each class that provide a comprehensive account of a set benchmark phenomena in rapid binary choice. Where conditions differed only in the rate of evidence accumulation (the most common case in past research), simulations showed the models supported equivalent psychological inferences. In contrast, differences in 2 other parameters of key psychological interest, response caution (the amount of information required for a decision), and nondecision time, traded-off when fitting 1 model to data simulated from the other, implying the potential for divergent inferences about latent cognitive processes. However, Donkin, Brown, Heathcote, and Wagenmakers did not find such inconsistencies between fits of the RD and LBA models in a survey of data sets from paradigms using a range of experimental manipulations. We examined a further data set, collected by Dutilh, Vandekerckhove, Tuerlinckx, and Wagenmakers (2009), which used a manipulation not surveyed by Donkin, Brown, Heathcote, and Wagenmakers's practice. Dutilh et al.'s RD model fits indicated that practice had large effects on all three types of parameters. We show that in this case the LBA provides a different and simpler account of practice effects. Implications for evidence accumulation modelling are discussed.