In our first experiment, we investigated whether observers complete missing information immediately while perceiving an event. We asked participants to either detect the critical moment of ball contact or ball release directly during event perception or to give their response after they finished watching the whole event. If completion occurs quickly, we should observe comparable completion effects both when giving responses directly during an event and when responses are delayed until after an event.
Method
Participants
Sixty-four students of the University of Tübingen participated in the experiment in exchange for course credit. We determined the sample size using the following rule. In a previous experiment (Brockhoff et al., 2016), we had observed an effect size of d = 1.47 for the completion effect with delayed responses (dependent-measure sensitivity). In order to account for the fact that the effect size might be lower for immediate responses, we collected the data of 64 participants, which is large enough to detect effect sizes of d = .71 with a power of .8.
Apparatus and stimuli
We extracted short video clips from the video coverage of a soccer match between the Young Boys Bern and the Grasshoppers Zürich that took place on 23 March 2014 as stimulus material. We created the video clips according to the following rules (see Fig. 1). Each video clip consisted of two parts combined by a filmic cut. The first part of each clip was extracted from the footage of the lead camera focusing on the player in ball possession and depicting this player including some of its surrounding (duration: 1.4 to 15 s). The second part of each clip was extracted from the footage of the high camera showing a larger part of the soccer field (duration: 1.2 to 6.3 s). The end of the first part of each clip depicted a clear action of a player toward the ball: 14 × kick-off, 5 × corner kick, 13 × throw-in, 8 × free kick. In the complete conditions, the moment of ball contact (e.g., player hitting the ball for kick-off) or ball release (e.g., ball just released from the player’s hand at throw-ins) occurred in the third last frame of the first part of the clip (presentation rate: 25 frames per second). In the removed conditions, we deleted the last four frames (160 ms) of the first part of the clips, thus resulting in the critical moment of ball contact or ball release not being visible anymore. In half of the stimuli, the second part of each clip depicted a causal continuation of the first part of each clip, such as a ball flying. In the other half of stimuli, the second part of each clip depicted a non-causal continuation, such as a player injury or the players getting ready for a free kick. In order to increase the readability of this article, we will refer to the critical moment of ball interaction as contact moment in the remainder of this article, irrespective whether it actually was a ball contact, such as a kick, or a ball release, such as a throw-in. Note that we ensured that the critical moment of ball contact was never visible following the cut.
Participants were placed at an unrestricted viewing distance of approximately 60 cm to the display. We presented the video clips on a gray background using PsychoPy (Peirce, 2007, 2009). The size of the video clips was 31.5° of visual angle horizontally and 18.1° of visual angle vertically.
Procedure and design
At the beginning of the experiment, participants gave informed consent. Thereafter, they performed two versions of the experiment with order of the experimental versions balanced across participants. In the immediate-response version of the experiment, we instructed participants to press the spacebar immediately after detecting the critical moment of ball contact. We analyzed the first response following the contact moment. In the delayed-response version of the experiment, we instructed participants to indicate whether they had seen the critical moment of ball contact after each trial, using a rating scale ranging from 1 (sure no) to 6 (sure yes). Our instruction explaining the task to the participants contained drawings of a player performing a throw-in and just releasing the ball as well as of a player kicking a ball and just touching the ball with the toe of the shoe in order to make sure that participants understood what we meant by the critical moment of ball contact. Each experimental version consisted of 40 different clips (trials) with trial order randomized for each participant. The association of clips to conditions was balanced across participants, that is, each clip occurred equally often within each condition across all participant. Participants were allowed to take self-paced breaks between trials.
Each participant saw only one causality condition. Further, we manipulated the presence of the contact moment within subjects. This resulted in a 2 (causality: causal, non-causal; between) × 2 (contact: present, absent; within) design for both experimental versions (immediate response, delayed response). For each experimental version, there were 20 repetitions per condition for each participant. There were no practice trials.
Results
We calculated the dependent-measure sensitivity (d’) as an indicator of contact-detection performance and we calculated the dependent-measure response criterion (c) as an indicator of response bias for both experimental versions. Because d’ and c are not defined for hit rates and false-alarm rates of 1.0 or 0.0, we replaced these values by half a trial incorrect or half a trial correct, respectively.
Immediate response
Results obtained in the immediate-response experimental version are depicted in Fig. 2. We observed a lower contact-detection performance for participants watching the video clips that presented a causal continuation compared to participants watching the video clips that had a non-causal continuation, t(62) = − 2.77, p = .007. Furthermore, participants’ response bias was more liberal (more contact responses) when viewing causal continuations than non-causal continuations, t(50.17) = − 2.66, p = .011 (Welch’s unequal variances t test). We further investigated this result pattern with a mixed analysis of variance (ANOVA) containing the factors contact (present, absent; within) and causality (causal, non-causal; between) and the dependent-measure proportion of contact responses. There was a significant interaction of contact and causality, F(1, 62) = 7.23, p = .009, ηp2 = .10. We used follow-up Welch’s unequal variances t tests to further investigate this interaction. This revealed that the reduced contact-detection performance was the result of an event-completion effect, that is, the presence of a causal continuation instead of a non-causal continuation resulted in an increased false-alarm rate, t(50.00) = 3.35, p = .002, but in no change of the hit rate, t(54.51) = 0.08, p = .935. Thus, participants showed a higher tendency to falsely report having seen the contact moment although it was not present if they saw a causal continuation instead of a non-causal continuation. The other effects of the ANOVA were as follows. There was a significant main effect of contact, F(1, 62) = 180.63, p < .001, ηp2 = .74, and a significant main effect of causality, F(1, 62) = 9.67, p = .003, ηp2 = .13.
We analyzed response times of hits using a Welch’s unequal variances t test. Participants response times in the causal condition were slower than response times in the non-causal condition, t(34.28) = 2.60, p = .014. This pattern of results goes in line with the reduced contact-detection performance in the causal condition. Interestingly, the mean response times in the causal condition (662 ms) indicate that event completion occurs quickly.
Delayed response
Results obtained in the delayed-response experimental version are depicted in Fig. 3. Based on participants’ responses to the rating scale, we obtained measures of contact-present responses, contact-absent responses as well as a measure of confidence of response. We treated responses of one to three as contact-absent responses and responses of four to six as contact-present responses. Confidence was 0 for responses three and four, 0.5 for responses two and five, and 1.0 for responses one and six.
We observed a lower contact-detection performance and a more liberal response criterion in the causal than non-causal condition, t(62) = − 3.57, p = .001 and t(62) = − 2.30, p = .025, respectively. A mixed ANOVA containing the factors contact (present, absent; within) and causality (causal, non-causal; between) and the dependent-measure proportion of contact responses revealed a significant interaction of contact and causality, F(1, 62) = 11.04, p = .002, ηp2 = .15. This was caused by an increased false-alarm rate in the causal compared with the non-causal condition, t(62) = 3.44, p = .001. There was no significant difference in hit rates across the causality conditions, t(62) = − 1.21, p = .231. Thus, we also observed an event-completion effect in the delayed-response experimental version. The other effects of the ANOVA were as follows. There was a significant main effect of contact, F(1, 62) = 294.77, p < .001, ηp2 = .83, and a significant main effect of causality, F(1, 62) = 5.69, p = .020, ηp2 = .08.
We analyzed the confidence of responses using a mixed ANOVA containing the factors contact (present, absent; within) and causality (causal, non-causal; between). There was a significant main effect of causality, F(1, 62) = 9.42, p = .003, ηp2 = .13, that is, participants were more confident in their responses in the non-causal condition in which they also showed a higher contact-detection performance. The main effect of contact was also significant, F(1, 62) = 6.37, p = .014, ηp2 = .09, that is, participants were more confident in their responses if the contact moment was present than if the contact moment was absent. Importantly, however, the interaction of causality and contact was not significant, F(1, 62) = 0.46, p = .502, ηp2 = .01. Thus, we did not find any evidence of participants being less confident in the condition without contact moment and with causal continuation. That is, although participants’ tendency to false alarm selectively increased in this condition, they were not selectively less confident in their responses.
Comparison: Immediate response vs. delayed response
We ran a comparison across the two experimental versions in order to investigate whether event completion differs between participants responding immediately after the contact moment and participants responding at the end of each clip. First, we compared contact-detection performance across experimental versions using a mixed ANOVA containing the factors causality (causal, non-causal; between) and experiment (immediate, delayed; within) and the dependent-measure sensitivity (d’). There was a significant main effect of causality, F(1, 62) = 13.44, p = .001, ηp2 = .18, indicating a reduced contact-detection performance in the causal condition as compared with the non-causal condition. The main effect of experiment was also significant, F(1, 62) = 5.80, p = .019, ηp2 = .09, indicating a higher contact-detection performance in the delayed-response experimental version than the immediate-response experimental version. Importantly, however, the interaction of causality and experiment was not significant, F(1, 62) = 0.26, p = .613, ηp2 < .01. Thus, the effect of causality on contact-detection performance was not influenced by the experimental version.
Second, we compared response bias across experimental versions using a mixed ANOVA containing the factors causality (causal, non-causal; between) and experiment (immediate, delayed; within) and the dependent-measure response criterion (c). There was a significant main effect of causality, F(1, 62) = 10.31, p = .002, ηp2 = .14, indicating a more liberal response bias in the causal condition as compared with the non-causal condition. The main effect of experiment was also significant, F(1, 62) = 10.95, p = .002, ηp2 = .15, indicating a more liberal response bias in the immediate-response experimental version than the delayed-response experimental version. Importantly, however, the interaction of causality and experiment was again not significant, F(1, 62) = 0.14, p = .706, ηp2 < .01. Thus, the effect of causality on response bias was also not influenced by the experimental version.
We also investigated the influence of experimental version on the hit rates and false-alarm rates using a mixed ANOVA containing the factors causality (causal, non-causal; between), contact (present, absent; within) and experiment (immediate, delayed; within) and the dependent-measure proportion of contact responses. Again, the event-completion effect did not differ across experimental version as indicated by a non-significant three-way interaction of causality, contact and experiment, F(1, 62) = 0.15, p = .699, ηp2 < .01. The other effects of the ANOVA were as follows. The significant main effects of causality, F(1, 62) = 13.34, p = .001, ηp2 = .18, and contact, F(1, 62) = 309.62, p < .001, ηp2 = .83, as well as the significant interaction of causality and contact, F(1, 62) = 11.95, p = .001, ηp2 = .16, correspond to the findings that we reported above in the respective individual ANOVAs of the immediate-response experimental version and delayed-response experimental version. Further, there was a significant main effect of experiment, F(1, 62) = 8.43, p = .005, ηp2 = .12, indicating a slightly higher proportion of contact responses in the immediate-response experimental version than delayed-response experimental version. The interaction of contact and experiment was also significant, F(1, 62) = 6.13, p = .016, ηp2 = .09, corresponding to the higher contact-detection performance in the delayed-response experimental version than in the immediate-response experimental version. The interaction of causality and experiment was not significant, F(1, 62) = 0.59, p = .447, ηp2 = .01.