4.2 The Comparison of Behavioural Measures across Studies revealed consistent
4.2.1 The Incremental Sequence Condition best induced Visual Learning
The pilot study was conducted to investigate the differences in the learning progress while manipulating the sequence of stimuli at the sample position. Two different sequences were tested: One was a randomized (RS condition) and the other an incremental (IS condition) sequence of illumination changes at the sample position. We assume that the IS condition in face stimuli in the sample position more slowly reveals facial features, because the stimuli presented one after the other are less varying, and, thus, the unveiling of new features by new illumination situations is delayed. We assume that one consequence of this is that learning is delayed.
Speed and accuracy are two critical components to describe behaviour and brain responses (Heitz, 2014). Therefore, we assumed that we will observe a positive learning progress by an increase in hit rates (HR) and by a decrease in reaction time (RT) across sequential positions for the match responses. Further, we also assumed that the response rate (RR) for match responses will increase across sequential positions (in turn, the RR for non-match must decrease. The response rate refers to the given responses and was also considered, because the instruction to the participants was to responses as truthfully as possible based on their perception. Thus, an increase in RR becomes a viable measure of learning, but it also makes it necessary to investigate a response bias. The last aspect was only possible to investigate with the MEG and fMRI studies, because they had a higher false alarm rate due to a greater number of trials.
Overall, three observations from the pilot study were crucial: First, for the IS condition a positive learning curve could be verified by a significant main effect of the factor
Sequence position for HR and RR for the match response condition. Post-hoc analyses
revealed an increase in HR and RR by a significant second and third simple effect. Second, this increase is due to a low performance at the beginning for HR of match responses compared HR of non-match responses in the IS condition. It is also a low performance of HR when comparing it to the HR of match and non-match responses of the RS condition. Third, no main effects of the factor Sequence position for RT were observed. However, post-hoc analyses revealed a decrease in RT for the RS condition for the condition overall response. Overall responses are the responses not separated in match and non-match responses. In the following, we discuss these three points in more depth.
A positive learning curve was verified by a significant increase in HR and RR for the match condition in the tested IS condition. This was not observed for non-match responses in the IS condition and not for any of the response types of interest in the RS condition. The HR and RR for non-match responses referred to correctly identified non-target faces. We prevented people from not learning non-target face identities by using four different face identities as non-target faces, but we still used three stimuli of each, which left a little chance for learning non-target identities. Because the HR were high from the beginning, we tentatively interpret that non-target faces were not learned. Instead, we think that the high HR for match responses from the beginning onward could be explained by an
exclusion strategy. This means that the participants were able to identify face stimuli as non-target better than the target. We propose this was possible from the beginning on, because few pictures of the target face identity (TFID) were probably sufficient to extract traits of the TFID to identify non-target faces in face stimuli than to identify the TFID. These traits could refer to secondary (or global) identity features or also to holistic- configural identity features (Farah et al., 1998; Gauthier and Tarr, 2002; Tanaka and Farah, 1993). These are, for example, gender, shape of the head or haircut, but also the relative distances between eyes, nose and mouth. We suggest that this information about the TFID was sufficient to identify non-target faces from the beginning on by exclusion criteria.
However, also the high HR of match responses of the RS condition indicate that the identification of the TFID could also be high from the beginning on. This observation confirmed our assumption that the IS condition slowly revealed visual information of the TFID (see Chapter 2.1.5.1 Processing Model Based on the Experimental Design) and in contrast to that, the RS condition quickly reveals visual information by more varying illumination directions at the beginning. Participants already gain sufficient visual information, given the first six random samples, and, thus, were able to extract more facial features from the beginning on than in the IS condition. Consequently, it was possible for the participants to correctly identify the TFID in a test image with a higher chance in the RS condition then in the IS condition.
The hypothesized RT decrease for HR and RR for match responses across the sequential positions during the learning procedure was not met. Here, we argue that the missing RT effect was most likely due to the instruction we gave to the participants. In awareness of the speed-accuracy trade-off (Heitz, 2014; Knight and Kantowitz, 1974, 1976), we instructed participants to focus on accuracy rather than on speed to ensure responses that correspond to their true perception. Further, the instruction also included that in case of double responses only the first (which we determine the intuitive) answer would be considered. This could be interpreted as a penalty for quick responses. As a consequence of the above facts, the pilot study showed that the incremental condition is best suited to proceed with further measurements in which additional neurophysiological data are collected in parallel.