Execution-based and verbal code-based stimulus–response associations: proportion manipulations reveal conflict adaptation processes in item-specific priming

For the sake of simplicity, we will for now speak of a cognitive control process without implying either of the theoretical accounts for proportion congruency effects. In the “General discussion” section we will discuss in detail which theory best accounts for our findings.

This must not be confused with assessments of item-specific proportion congruency effects where stimuli are presented multiple times and the proportion of, for instance, congruent versus incongruent trials for each stimulus is manipulated (e.g., stimulus A 75% congruent, stimulus B 25% congruent). In the present experiments, each stimulus was only probed once. The item-specific manipulation we refer to is only related to whether the S–C and S–A mapping for a stimulus repeated or switched between its single prime and probe instance. As each stimulus was only associated with either a single item-specific switch or repetition in S–C and S–A mapping, the respective S–C/S–A switch proportions associated with a single stimulus are always either 0% or 100%. Framed differently, each individual stimulus was contingency unbiased. Repetitions/switches in S–C and S–A mapping occurred for the first and only time during the single probe instance of each stimulus. Thus, item-specific learning of the frequency of switches in S–C and S–A mapping could not occur.

Please note that the prime and probe instances of specific stimuli never occurred on two consecutive trials. Our S–C switch effects, therefore, differ from task switching effects that are observed from trial N-1 and trial N (e.g., Kiesel et al. 2010, for a review). In the present item-specific paradigm, task switching effects are observed independent from the item-specific S–C and S–A switch effects (see Pfeuffer et al. 2017).

CR = classification repetition, CS = classification switch; AR = action repetition, AS = action switch.

Please note that memory recall performance is not informative about whether S-R switch proportion affected S-R encoding and/or retrieval processes, as probe trials occurred in between prime and memory recall trials. Probe S-R repetitions should have supported recall, whereas probe S-R switches should have obstructed it. Thus, it cannot be determined whether differences in recall performance are to be attributed to differences in prime encoding or the respective probe S-R switch condition. Memory recall data can therefore only be used to assess whether participants attended to verbal codes.

Bayesian analyses We observed interactions between classification and proportion switch group, but we did not observe three-way interactions between prime type, classification, and proportion switch group or prime type, action, and proportion switch group. This null effect is theoretically relevant as it might suggest that execution-based and verbal code-based S–C and S–A associations did not differ in terms of how they were influenced by processes of cognitive control. To ascertain the theoretically relevant null effects of these three-way interactions, we conducted additional Bayesian Repeated Measures ANOVAs with default prior scales using JASP (version 0.8.0.0, Love et al. 2015; see Rouder et al. 2009, 2017, for information on Bayesian statistics) on RTs and error rates. To compute the Bayes factors for these three-way interactions, we excluded the influence of the respective main effects or two-way interactions of the factors included in the three-way interaction. We were thus able to assess the evidence in favor of the null hypotheses that the interactions of classification and proportion switch group and action and proportion switch group were not modulated by prime type. The Bayes factor (\({\text{BF}}_{01}\)) indexes how strongly the data is in favor of the null hypothesis. As influences of other effects are already excluded, the reported Bayes factors can directly be interpreted as the likelihood of null effects for the respective three-way interactions. Bayes factors between 1 and 3 are considered anecdotal evidence for the null hypothesis (see Jarosz and Wiley 2014). Bayes factors between 3 and 10 are considered substantial evidence for the null hypothesis and Bayes factors between 10 and 30 are considered strong evidence in favor of the null hypothesis.

As an alternative to the assumed modulation of the rate of evidence accumulation of the instance retrieval route, one might suggest that instead the response threshold could have been affected. An overall modulation of the response threshold for both response options (though selectively for actions and classifications) with a decreased response threshold when repetitions were frequent and an increased response threshold when switches were frequent, could not account for the observed pattern of results. Responses on S–C/S–A repetition trials would be speeded when S–C/S–A repetitions are frequent. However, responses on S–C/S–A switch trials would be further slowed down when S–C/S–A switches are frequent. Thus, we would not have observed reduced S–C/S–A switch effects for frequent switches as compared to frequent repetitions in the respective mappings. One would have to assume that participants selectively and item-specifically decreased the response threshold for the action/classification response given in the prime when repetitions were frequent and selectively and item-specifically decreased the response threshold for the opposite action/classification when switches were frequent. As the respective response thresholds that would need to be decreased depend on the action/classification associated with a stimulus during its prime trial and thus change item-specifically, it is hard to fathom how such a mechanism would work and a much more complex model would have to be assumed to accommodate it.

Alternatively, it might also be that only either frequent repetitions or frequent switches of S–R mapping affect the rate of evidence accumulation.

Note that for the present model, it is irrelevant whether the rate of evidence accumulation of the instance retrieval route (i.e., item-specific S–R associations) is influenced by cognitive control processes at encoding or retrieval.

We additionally included RT Ntile as a predictor in the regression, as the overall RT level might also have affected the size of the observed S–C and S–A effects. At least for S–C switch effects, percentile analyses of data we previously collected within the same paradigm typically showed an increase of S–C switch effects with percentile.