Figure 1
Illustration of recency-based retrieval of response sequences.
Note. The green solid arrows illustrate the path that leads to a retrieval of a matching response: The “middle” response in trial 7 retrieves the last occurrence of the same response (trial 2). The response that was bound to the middle response in this sequence (i.e., a “right” response executed in trial 3) is then retrieved and influences responding in the following trial of the series (trial 8). In this case, a matching response was retrieved and reactivated (trial 8 also needs a “right” response), leading to facilitation. The red dotted arrows illustrate a situation in which a mismatching response is retrieved: Trial 8 will retrieve the last matching sequence starting with the same response (trials 5–6), and will reactivate the second element of the response sequence (trial 6) for the next upcoming trial (trial 9). In this case, a “left” response is retrieved from trial 6, but in trial 9 a “middle” response is to be given. In general, retrieval of matching (mismatching) responses is much more likely for high (low) frequency (HF/LF) trials, because the retrieved sequence is most likely a high frequency sequence.
Figure 2
Transition probabilities between responses in the present study.
Note. Response-response contingencies were created by biased transition probabilities between three spatial positions at which a dot could appear on the screen, indicating which key had to be pressed on a keyboard (left [B], middle [N], right [M]).
Figure 3
Contingency awareness assessment at the end of the experiment for a sample sequence.
Table 1
Results of a stepwise multi-level regression analysis predicting RT based on contingencies for the current and previous sequence (CL and PCL, step 1), episodic retrieval (ER, step 2), contingency awareness (CA) and its interactions with CL and ER (step 3).
| PREDICTOR | MODEL 1 | MODEL 2 | MODEL 3 |
|---|---|---|---|
| Intercept | 418*** [403.1, 432.0] | 418*** [403.3, 431.9] | 419*** [404.1, 432.9] |
| CL (hc vs. lc) | –65*** [–67.6, –63.2] | –32*** [–34.4, –29.5] | –33*** [–35.9, –30.7] |
| PCL (hc vs. lc) | –17*** [–19.5, –15.3] | –23*** [–25.2, –21.2] | –23*** [–25.1, –21.1] |
| CL × PCL | –31*** [–36.0, –25.3] | –51*** [–56.6, –46.1] | –52*** [–56.9, –46.4] |
| ER (matching vs mismatching) | –56*** [–57.7, –53.5] | –56*** [–57.8, –53.6] | |
| CA (correct vs. incorrect) | –9*** [–10.9, –6.8] | ||
| CL × CA | –24*** [–29.5, –18.9] | ||
| ER × CA | 4 [–0.4, 8.8] | ||
| BIC | 352091 | 349518 | 349371 |
| ∆ BIC | – | –2573 | –147 |
[i] Note. *p < 0.05, **p < 0.01, ***p < 0.001. CL/PCL, contingency in the current or preceding response sequence; hc/lc: responses following the previous response with high/low probability. The interaction of CL and PCL indicates differences between reversal and non-reversal sequences. ER, episodic retrieval; matching/mismatching second response during the last occurrence of a sequence that started with the same response; CA, contingency awareness; correct/incorrect: identification of the likelihood of occurrence for the current response sequence. BIC, Bayesian information criterion. We implemented a person specific intercept to control for individual differences in RTs. All other variables were implemented on a trial level. Values in brackets indicate the 95% confidence interval (lower and upper limit) for each regression weight. Regression weights (ß) reflect the difference in milliseconds between the conditions that define a contrast.
Table A1
Number of experimental trials for each combination of the experimental design across participants.
| CONTINGENCY | ||
|---|---|---|
| Last sequence with same first response | High | Low |
| Same second response | 22,746 | 1,486 |
| Different second response | 5,738 | 5,669 |
[i] Note. Each trial was coded as (a) high or low contingency depending on whether it was preceded by its likely or unlikely predecessor, and (b) as retrieving the same or a different response from the last sequence that had started with the same initial response as the current one.
Table A2
Results of a stepwise multi-level regression analysis predicting accuracy (1 = correct response, 0 = erroneous response) based on contingencies of the current or immediately preceding sequence (CL and PCL, step 1), episodic retrieval (ER, step 2), contingency awareness (CA) and its interactions with CL and ER (step 3).
| PREDICTOR | MODEL 1 | MODEL 2 | MODEL 3 |
|---|---|---|---|
| Intercept | .963*** [.955, .971] | .963*** [.955, .972] | .963*** [.955, .972] |
| CL (hc vs. lc) | .092*** [.087, .097] | .065*** [.059, .071] | .064*** [.058, .070] |
| PCL (hc vs. lc) | -.003 [-.008, .002] | .002 [-.003, .007] | .002 [-.003, .007] |
| CL × PCL | .033*** [.021, .046] | .048*** [.036, .061] | .049*** [.036, .062] |
| ER (matching vs mismatching) | .044*** [.038, .049] | .044*** [.038, .049] | |
| CA (correct vs. incorrect) | .001 [-.004, .007] | ||
| CL × CA | .008 [-.004, .021] | ||
| ER × CA | -.021*** [-.033, -.010] |
[i] Note. *p < 0.05, **p < 0.01, ***p < 0.001. CL/PCL, contingency in the current and preceding response sequence; hc/lc: responses following the previous response with high/low probability. ER, episodic retrieval; matching/mismatching second response during the last occurrence of a sequence that started with the same response; CA, contingency awareness; correct/incorrect: identification of the likelihood of occurrence for the current response sequence. We implemented a person specific intercept to control for individual differences in accuracy rates. All other variables were implemented on a trial level. Values in brackets indicate the 95% confidence interval (lower and upper limit) for each regression weight. Regression weights (ß) reflect the difference in %correct responses between the conditions that define a contrast.