To test the potential feature binding contributions to effect monitoring, we designed an experiment in which we varied the response dimensions in Task 1 and Task 2 (horizontal vs. vertical) so that R2 feature repetitions were either possible (when relying on the same spatial dimension) or impossible (when relying on different spatial dimensions). In Task 1, participants produced effects that were either spatially compatible or incompatible, and Task 2 required them to categorize colors.
If both tasks rely on the same response dimension (both horizontal, as in Table 1, or both vertical), a compatible action effect in Task 1 would always lead to a full repetition or a full alternation in Task 2, and an incompatible action effect in Task 1 would always lead to a partial feature alternation (and, hence, slowdown) in Task 2. But if both tasks relied on different response dimensions, in no case would R2 repeat any of the spatial features that were bound in Task 1, leading to full alternations in both compatible and incompatible trials. Therefore, if a slowdown in Task 2 after an incompatible action effect is only found with dimensional overlap between the two tasks, but strongly reduced or even absent without dimensional overlap, then these incompatibility costs mainly constitute feature binding limitations. However, if a Task 2 slowdown after an incompatible effect is found irrespective of the dimensional overlap of the two tasks, then feature binding only plays a minor role in monitoring (spatially compatible and incompatible) effects.
Forty-eight participants were recruited (mean age 25.6 years, SD 4.4) and received monetary compensation. This sample size was chosen because it was the minimum number of participants required for counterbalancing, and it should provide high statistical power (>.99) for the critical propagation of the compatibility effect onto Task 2, assuming similar effect sizes to those previously found (Wirth, Janczyk, & Kunde, 2018a, Experiment 1). All participants reported normal vision and were naïve concerning the hypotheses of the experiment. All participants provided written informed consent prior to the experiment. One participant was removed from the final sample due to performing below chance level and was replaced.
Apparatus and stimuli
All stimuli were presented on a 22-in. screen against a black background (see Fig. 1). S1 were pictures of puzzle pieces with a connector at the left, right, top, bottom, or at two opposing sides (left and right, or top and bottom), presented centrally on the screen. If the puzzle piece included only one connector, a puzzle piece had to be produced that fitted the connector (forced choice). If two connectors were presented, participants could choose freely at which side they wanted to add a piece (free choice). In free-choice trials, participants were encouraged to choose their option spontaneously while maintaining an approximately equal ratio between all options. Participants responded with the left hand on the “S”, “D”, “E”, or “X” key of a standard QWERTZ keyboard (R1). These keypresses produced puzzle pieces of the screen as an effect (E1).
For Task 2, participants had to categorize a splash according to its color (S2). The color splash was presented centrally within S1 and required a response with the right hand on the “2”, “4”, “6”, or “8” key on the number pad (R2).
The trial procedure is illustrated in Fig. 1. A fixation cross marked the beginning of a trial. After 500 ms, S1 was presented centrally on the screen and required the production of E1 via pressing a key with the left hand (R1). The produced puzzle piece appeared immediately after R1 and stayed on-screen until the trial ended. If, after a maximum of 2,000 ms, no R1 key was pressed, this counted as an R1 omission and no E1 was displayed.
After an interval of 50 ms after E1 onset (or after 2,050 ms after S1 onset in case of R1 omissions), S2 was displayed centrally and called for R2. Again, if no key was pressed within 2,000 ms, this counted as an R2 omission. The two tasks were always presented in that order, and there was no temporal overlap between the tasks (except for the effect monitoring process). In case of any errors or omissions, written feedback was presented at the end of the trial (e.g., “Puzzle task: Too slow!”, or “Color task: Error!”, or both) for 500 ms in red. If both tasks were completed correctly, the next trial, indicated by a fixation cross, started immediately.
Two factors were experimentally manipulated. First, the response dimensions for both tasks were varied between blocks (R1-R2 combinations, see Fig. 2). In half of the blocks, Task 1 required only horizontal R1s, so that only puzzle pieces with left and right connectors were presented. In the other half of the blocks, only vertical R1s were required, presenting only top or bottom connectors (“rows” in Fig. 2). Orthogonally to the R1 dimension, the response requirements in Task 2 were also varied. In half of the blocks, Task 2 also only required horizontal R2s, so that only blue and orange splashes were displayed. In the other half of the blocks, only vertical R2s were required, presenting only red and green splashes (“columns” in Fig. 2). The combination of both variations resulted in blocks that either used the same response dimension in both tasks (dimensional overlap conditions: horizontal R1 & R2; vertical R1 & R2), or different response dimensions for both tasks (no dimensional overlap conditions: vertical R1 & horizontal R2; horizontal R1 & vertical R2). Participants were informed before each block which response dimension would be relevant for both tasks during the upcoming block.
Second, the spatial compatibility between response and effect in Task 1 was varied between experimental sessions (R1-E1 compatibility). The R1-E1 mapping could either be spatially compatible, so that the relative location of R1 would match the location of the E1 that it would produce (R1→E1 combinations: left→left; right→right; top→top; bottom→bottom; see Fig. 1 for an example), but it could also be incompatible, with responses producing effects on the opposite side (left→right; right→left; top→bottom; bottom→top). R1-E1 compatibility was constant for a whole session, and participants would be reminded before each block about the current compatibility.
Participants completed two sessions that were scheduled on different days, with 20 blocks each, five blocks per R1-R2 combination (see Fig. 2). One session employed a compatible R1-E1 mapping throughout, the other employed the incompatible R1-E1 mapping. The order of the sessions (compatible first vs. incompatible first), as well as the order of the R1-R2 combinations within these sessions (24 possible sequences), was counterbalanced across participants. Each block consisted of 80 trials, with 40 forced-choice trials and 40 free-choice trials in randomized order. In forced-choice trials, each combination of S1 (left vs. right connector; or top vs. bottom connector; depending on the R1 dimension of the current block) and S2 (blue vs. orange splash; or red vs. green splash; depending on the R2 dimension of the current block) was presented ten times (2 × 2 × 10), and in free-choice trials, the free-choice stimulus (left and right connectors; or top and bottom connectors; depending on the R1 dimension of the current block) was paired with each S2 20 times (1 × 2 × 20).
In free-choice trials, all four response options in Task 1 were chosen with an equal ratio (left key: 24.9%, right key 25.0%, top key: 24.5%, bottom key 25.2%, all ts<1, response omissions: 0.4%). For RT analyses, we excluded trials with errors (Task 1: 6.5%, Task 2: 8.1%) and omissions (Task 1: 0.4%, Task 2: 0.3%). The remaining trials were screened for outliers and we removed trials in which RTs for any task deviated more than 2.5 standard deviations from the corresponding cell mean, computed separately for each participant and experimental condition (4.3%). Overall, 15.7% of the trials were removed.
The remaining data was then analyzed separately depending on whether Task 1 was forced or free choice. RTs were analyzed with a 2 × 2 ANOVA with Task 1 R1-E1 compatibility (compatible vs. incompatible) and type of R1-R2 dimensional overlap (overlap vs. no overlap) as within-subjects factors (see Fig. 3). Error rates and omissions were analyzed accordingly. As the interaction of compatibility and dimensional overlap in RT2s are the crucial result here, these interactions will be followed up via Bayes factor analysis.
In Task 1, we observed faster responses for the compatible R1-E1 mapping (452 ms) compared to the incompatible mapping (545 ms), F(1,47)=91.84, p<.001, ηp2=.66. Further, there were faster responses in Task 1 when response dimensions differed between tasks (490 ms) than when they overlapped (507 ms), F(1,47)=7.63, p=.008, ηp2=.14. There was no interaction between the two factors, F<1.
The compatibility manipulation in Task 1 propagated to Task 2, F(1,47)=7.59, p=.008, ηp2=.14, with faster responses after compatible effects (443 ms) compared to after incompatible effects (468 ms). Similarly, there were faster responses in Task 2 when response dimensions differed between tasks (449 ms) than when they overlapped (463 ms), F(1,47)=9.51, p=.003, ηp2=.17. There was no interaction between the two factors, F<1, BF01=5.60, indicating that compatibility effects were of similar size both with, t(47)=2.76, p=.008, d=0.40, ∆=18 ms, and without dimensional overlap, t(47)=3.18, p=.003, d=0.46, ∆=20 ms.
Errors were more prominent in Task 1 with the incompatible mapping (6.8%) compared to the compatible mapping (3.7%), F(1,47)=27.14, p<.001, ηp2=.37. The main effect of dimensional overlap was not significant, F<1. An interaction, F(1,47)=4.20, p=.046, ηp2=.08, indicated more errors with the incompatible mapping relative to the compatible mapping in the dimensional overlap condition (∆=3.4%), but not without dimensional overlap (∆=2.7%). Errors in Task 2 did not show any significant effects, Fs≤2.16, ps≥.148.
Omission rates in Task 1 did not show any main effects, Fs≤1, ps≥.326, but they were more prominent with the incompatible mapping relative to the compatible mapping in the dimensional overlap condition (∆=0.2%), but not without dimensional overlap (∆=0.0%), F(1,47)=4.12, p=.048, ηp2=.08. Omission rates in Task 2 did not show any significant effects, Fs≤3.57, ps≥.065.
In Task 1, we observed faster responses for the compatible mapping (389 ms) compared to the incompatible mapping (433 ms), F(1,47)=32.75, p<.001, ηp2=.41. Neither the main effect of dimensional overlap, F(1,47)=1.60, p=.213, ηp2=.03, nor the interaction, F(1,47)=2.27, p=.139, ηp2=.05, reached significance.
Importantly, the compatibility manipulation in Task 1 propagated to Task 2, F(1,47)=11.47, p=.001, ηp2=.20, with faster responses after compatible effects (436 ms) compared to trials after incompatible effects (465 ms). Further, there were faster responses in Task 2 when response dimensions differed between tasks (444 ms) than when they overlapped (457 ms), F(1,47)=11.30, p=.002, ηp2=.19.Footnote 3 There was no interaction between the two factors, F<1, BF01=6.32, indicating that compatibility effects were of similar size both with, t(47)=3.34, p=.002, d=0.48, ∆=24 ms, and without dimensional overlap, t(47)=3.61, p=.001, d=0.52, ∆=23 ms.
Errors revealed neither main effect nor interaction in both Task 1, Fs≤2.84, ps≥.099, and Task 2, Fs<1.
Omission rates did not show any significant effects in either Task 1, Fs≤3.23, ps≥.079, or Task 2, Fs≤2.07, ps≥.157.
In Experiment 1, we tested whether feature binding processes contribute to the monitoring of spatial action effects. Task 1 required participants to produce a foreseeably compatible or incompatible effect via keypress, and Task 2 had them categorize color splashes. Further, the response dimensions were manipulated, so that both tasks relied on the same spatial dimension (both vertical or both horizontal), allowing for partial feature alternation costs in Task 2, or they relied on different response dimensions (one vertical, one horizontal), which eliminates any feature overlap to Task 2.
As expected, we found that the production of an incompatible effect takes longer and is more prone to errors than the production of a compatible effect. As these effects are not present at the time that the response is given, this influence on response production must be based on anticipatory codes of the effects (e.g., Kunde, 2001, 2003; Kunde, Schmidts, Wirth, & Herbort, 2017; Pfister & Kunde, 2013; Pfister, Janczyk, Wirth, Dignath, & Kunde, 2014; Wirth, Pfister, Janczyk, & Kunde, 2015). Also, replicating our previous results, we showed that with a dimensional overlap between the two tasks, responses after incompatible effects took longer to complete than responses after compatible effects (Wirth, Janczyk, & Kunde, 2018a; Wirth, Steinhauser, Janczyk, Steinhauser, & Kunde, 2018b). This slowdown after incompatible compared to after compatible action effects denotes increased costs of monitoring, as Task 2 itself did not employ any compatibility manipulation, and the still ongoing monitoring is the only cognitive process that could propagate to Task 2. This was not only true for forced-choice trials, but crucially also for free-choice trials that are free from potential influences of S-R compatibility.
Up until now, we have explained these response costs after incompatible trials exclusively as an indication for longer effect monitoring processes, interfering with the processing of the second task (e.g., Kunde, Wirth, & Janczyk, 2018); however, in this setup, these costs might also constitute partial alternation costs (see Table 1). Therefore, the crucial test to show costs for monitoring self-produced effects is when feature binding cannot contribute to the effect monitoring process, which is the case if the response dimensions of the two tasks do not overlap. In this condition we still find response costs after incompatible effects relative to after compatible effects, demonstrating that feature binding at best has a minimal influence on the effect monitoring process. That said, we do find an overall benefit for Task 2 if response dimensions differed between tasks in RTs, so our setup is in principle able to detect feature binding influences, but the limitations of binding events into event files seem to be independent of the limitations that occur when monitoring self-produced action effects.
Please note that participants used different hands for the two tasks, so there was a hand switch in each trial. This may have lowered our chances to find feature binding contributions in the first place, as participants did not repeat the identical responses (e.g., with their left index finger), but rather just the abstract spatial code of the response (e.g., “right”). We intentionally designed our experiment this way to disentangle feature binding contributions from motor affordances; however, this way we may somewhat underestimate the main effect of feature binding.
What is also noteworthy is that S2 was always presented 50 ms after R1 was given, so in fact, participants also produced the S2 onset on screen, so that S2 might also be construed as an action effect and might be integrated into the event file of Task 1. This might be the case, and we have no hard evidence against this, but several points make this unlikely. First, only the time of S2 onset was linked to R1, whereas the identity of S2 was completely unpredictable, making it less likely to be perceived as action-contingent against the perfectly controllable puzzle piece (but see Experiment 2 for unpredictable action effects). Second, we have previously found that monitoring costs also arise with a variable interval between the two tasks (Wirth, Janczyk, & Kunde, 2018a, Experiment 4). This does not directly speak against the idea that S2 is also integrated in the Task 1 event file, but it shows that the fixed time interval cannot drive the data pattern found in Task 2. Finally, even if S2 were integrated in the Task 1 event file as a secondary effect, S2 does not have an inherent spatial direction (as R1, E1, and R2), but only varies in color, so that we think it is unlikely that S2 could interfere with any of the results that we presented here.
To sum up, we found that monitoring costs in Task 2 due to an incompatible effect in Task 1 seems to be rather free from specific feature binding contributions. In the following two experiments, we will further elaborate on these results by (a) checking whether they hold in situations in which the spatial compatibility of R1 and E1 are less task relevant (Experiment 2), and (b) by dissecting what kind of feature overlap in incompatible trials (R1-R2 vs. E1-R2) mainly drives the slowdown in partial alternations relative to full repetitions/alternations (Experiment 3).