The role of learning in sensory-motor modality switching

Sensory-motor modality switching represents a specific case of task switching. That is, the tasks that have to be switched are defined by their different sensory-motor modality mappings and not by the performed cognitive operation (e.g., semantic categorization). More precisely, participants switch between different sensory-motor modality mappings while performing cognitive operations that are fixed. Thus, the resulting switch costs are also termed as modality switch costs.

To further draw inferences about this null-effects, we calculated the Bayes factor (BF) that is a ratio that indicates how much more likely given data fit under one model than under another model. We conducted a Bayesian three-way repeated measures ANOVA with JASP, a freely available statistical program (Love et al., 2015; Rouder, Speckman, Sun, Morey, & Iverson, 2009). At this point, it has to be mentioned that Bayesian analyses always require the selection of prior probability distribution for the model parameters, always implicating a certain amount of subjectivity or prior knowledge about the effect that is to be studied. For the present analyses we used the default prior scales of JASP, which assumes a Cauchy distribution of possible effect sizes (cf. Rouder et al., 2009)–even though this slightly favors the null model because it assumes relatively large effect sizes for the alternative model. To draw inferences about the interaction modality compatibility × stimulus set, we compared the model including this interaction and the three main effects to the main effects-only model. The resulting BF₀₁ = 3.5 indicates that the present data are about three times more likely under the main-effects-only model than under the interaction model. As regards the interaction transition × modality compatibility × stimulus set, the comparison of the full model to the model without the three-way interaction revealed the Bayes factor BF₀₁ = 3.1, indicating that the data are more likely under the model without the three-way interaction than under the full model.

In order to evaluate the evidence of this numerical trend, which shows a reversed pattern compared to RT switch costs, we calculated the Bayes factor regarding the interaction modality transition × modality compatibility in the large stimulus set in a similar manner as described in Footnote 2. Specifically, we compared the model including the interaction modality transition × modality compatibility to the main-effects-only model. The resulting Bayes factor BF₁₀ = 0.8 neither supports the alternative nor the null model, indicating that the error data of the large stimulus set are not very conclusive regarding this interaction.

In the practice blocks a consistent or variable concept-to-modality mapping was not possible because the sensory-motor modality mapping was fixed (e.g., a consistent auditory presentation of the stimuli in the auditory-vocal practice block).

Similar to Experiment 1 (see footnote 2), we calculated the Bayes factors to further draw inferences about these null-effects. As regards the interaction modality compatibility × concept-to-modality mapping, we compared the model including this interaction to a model without this interaction. The resulting Bayes factor BF₀₁ = 5.0 indicates that the data were more likely under the model without the interaction than under the model including the interaction. A similar procedure with regard to the interaction transition × concept-to-modality revealed the Bayes factor BF₀₁ = 5.2, indicating that the data are about five times as likely under the model without the interaction than under the interaction model. As regards the interaction transition × modality compatibility × concept-to-modality mapping, we compared the full model to the model without the three-way interaction. The analysis revealed the Bayes factor BF₀₁ = 4.0, indicating that the data are about four times as likely under the model without the three-way interaction than under the full model.