Underlying Sources of Response-Response Contingency Learning

We examined contingency learning of response sequences (response-response contingency learning, RR-CL) in a serial reaction time task (SRTT), in which keys were pressed corresponding to the spatial position of dots appearing successively on a screen. First-order contingencies between pairs of successive responses were introduced into the task by biasing the transition probabilities between the spatial positions of subsequently appearing dots, rendering specific response sequences more likely than others. A pre-registered study (n = 40) revealed evidence for robust RR-CL effects, indicated by shorter (longer) response times for responses that were preceded by a likely (unlikely) preceding response. Part of this RR-CL effect was due to episodic retrieval of the most recent response sequence that started with the same response as the current sequence. Yet, a robust genuine RR-CL effect remained after controlling for recency-based episodic retrieval. This residual RR-CL effect was dependent on contingency awareness, with stronger residual RR-CL effects for sequences that were correctly identified as high or low in frequency, indicating that learning of response sequence contingencies partly reflects propositional knowledge. A reliable residual RR-CL effect, however, was obtained also in the absence of contingency awareness, indicating that another part of the learning of response sequence contingencies operates automatically and outside of awareness.