Figure
2. Effect of agency and learning block on %Correct.a. %Correct mean and standard error across all participants
(N=23) in the agent condition (solid line ) and observer condition
(dotted line ) across learning blocks. Asterisks mark learning
blocks in which significant differences between conditions were found in
post-hoc tests. b. Difference between the agent and observer
conditions for each participant across learning blocks (black
dots ), as well as the median and interquartile range across
participants (boxplots ). The difference between the agent and
observer conditions is inversely correlated with learning block. An
exponential curve was fitted to the data (blue solid line ) to
illustrate the decreasing effect of agency with the progress of
learning. The effect of agency decreases rapidly during the first three
learning blocks and then remains constantly low in the remaining four
learning blocks.
Electrophysiological
results
For each effect and interaction studied, we present here two
complementary approaches: a data-driven analysis using cluster-based
permutation tests, and an ERP component driven analysis using ANOVAs to
assess effects on targeted ERP components.
Acquisition sounds
Learning progress was reflected in ERPs as an attenuation of the P3a
component. The cluster-based analysis comparing early and late
acquisition sounds (late – early learning stage) revealed a negative
cluster with a fronto-central distribution (T = 4659.8, p <
.01; 220 ms to 400 ms), encompassing the P2 and P3a components (figure
3a). The targeted-component ANOVA yielded a significant main effect of
learning stage on the amplitudes of the P3a component at Fz
[F(1,22) = 14.436, p < .001,
ηp2 = 0.40], reflecting more
negative values with increased learning stage.
Regarding the effect of agency, we found more positive ERPs in parietal
electrodes in the agent compared to the observer condition (figure 3b).
The cluster-based permutation test comparing the agent and observer
conditions detected a significant positive cluster with an
occipito-parietal distribution (T = 10900, p < .01; 60 ms to
400 ms). The cluster temporally encompasses the P2 and P3 components,
revealing overall more positive amplitudes in the agent condition. The
targeted-component ANOVA detected a significant main effect of agency on
the P3b component at Pz [F = 25.706, p < .0001,
ηp2 = 0.54]. The occipito-parietal
distribution of the effect led us to suspect that the observed
differences are due to motor differences related to the control of the
sound stimuli. To explore this possibility, we conducted further
analyses, which can be found in the supplementary materials.
In order to study possible interactions between agency and learning
stage using cluster-based permutation tests, we subtracted the early
learning stage from the late learning stage in both the agent and
observer condition and then ran a cluster-based analysis (late minus
early learning stage in agent condition versus late minus early learning
stage in observer condition). Studying this interaction, possible
confounding factors due to eye movement differences (see discussion in
supplementary materials) were eliminated from the data. However, this
analysis yielded no significant results. We also tested for interaction
effects in the targeted-component ANOVAs, but this also did not produce
any significant interaction effects. All in all, no interactions between
the factors agency and learning stage were detected.