B.1.6.1.2. Cueing effects

The relational vector account should also be evaluated in relation to the endogenous effects observed in Experiments 2 and 3. Again, the similarity effects in priming were assessed, in the relevant conditions (large- and small- relevant combined): They appeared significant in the block-wise experiment (Exp. 3), but not in the trial-wise experiment (Exp. 2)⁸. On the one hand, in the Experiment 2, the trial-wise cueing mode precluded any relational priming effect to be expressed in the performance. Moreover, no clue of such effects was evidenced in the analysis of priming. On the other hand, the block-wise cueing mode of Experiment 3 was expressly used to assess the inter-trial priming effects. In this experiment, priming effects seemed to occur, and to be influenced by the prime-target similarity. This was supported by both the pattern of RT results (relative to Exp. 2) and by the supplementary analysis of priming. However, these similarity effects in priming were consistent with both the relational and feature-based accounts.

Finally, according to Becker's hypothesis, a relational (rather than feature-based) target template could have been used in the relevant conditions of Experiments 2 and 3. There was no clear evidence for this, in the present results, however. Furthermore, the comparison of the random and large-relevant conditions of Experiment 2 argued against this hypothesis. Indeed, only the large-medium RT difference increased with cueing, not the medium-small difference. This strongly suggested that only the "large" square was favoured by cueing, not all the "larger" squares (which would include the medium one). Thus, the use of a relational target template seemed unsupported in Experiment 2. In this respect, no difference would be expected depending on the cueing mode, that is between Experiments 2 and 3. So the size effects observed between irrelevant items in the relevant conditions of the Experiment 3 were most likely attributable to similarity effects occurring through perceptual priming.

In sum, it seemed that the relational vector hypothesis was not strongly supported in the present experiments. Indeed, it could not account for most of the results (e.g. salience effects in random conditions, relevance effects in Exp. 2). The present data did not contradict those of Becker (2010), though, since the tasks differed in at least one important way. Indeed, in Becker's experiments, the target was defined as the singleton, and no precise feature of the target was known. This procedure precluded the use of feature-based guidance of attention. On the contrary, this mode of search was possible in the relevant conditions of the present experiments.

One might suggest that feature-based and relational attention guidance were actually two distinct modes of attentional guidance. It is also possible, as incidentally suggested by Becker (2010), that the "relational guidance effects" were actually feature-based. Actually, the evidence brought by Becker (2008, 2010) could be criticized in some ways. This author tried to show that a relational account could account for the inter-trial effects usually attributed to perceptual priming (Maljkovitch & Nakayama, 1994). First, she rejected this latter hypothesis on the basis of a lack of significance (in the "same-relation" condition of Exp. 5, Becker, 2008, and of Exp. 1 & 2, Becker, 2010). This sounded a bit imprudent, particularly given the weak number of participants by experiment (n = 6), and given the low p-value in at least one experiment (Exp. 1 of Becker, 2010: F(1, 5) = 3.9, p = .105). More experiments, fulfilling the "good-effort criterion" suggested by Frick (1995), would be needed to reject this feature-based perceptual priming hypothesis definitely. Second, Becker (2010) considered the "filter shift" hypothesis, and argued that this latter could made the results of her Experiment 3 consistent with the feature-based account, but could not, on the other hand, account for the results of her Experiment 2. In this experiment, three conditions were compared. In the "switch" and "reversal" conditions, target-distractor relation could vary between consecutive trials. For instance, the target could be larger than distractors in trial n-1 and smaller than distractors in trial n. In the "same relation" condition, the target could be small among medium distractors or large among extra-large distractors. She evidenced priming effects only in the two first conditions and concluded that priming effects could depend on target-distractor relations. Following the reasoning presented by the author, it seemed that the "filter shift" hypothesis was indeed consistent with the results all the experiments in Becker (2010). Precisely, in the second experiment, the target could be small among medium, or large among extra-large. Following the "filter shift" hypothesis, participants might have used the small target as a (not relational) template. This would have been optimal for the small target among medium distractors. But this would have been quite efficient for the other condition too. Indeed, the "small" channel would respond stronger for the large target than for the extra-large distractors. Hence, no change of the attentional set should occur between these two target's conditions, contrary to what should happen in the "switch" and "reversal" conditions of the same experiment. It ensues that what was attributed to a "relational vector template" might actually be due to an uncertainty about which feature-based template using in the trial at hand. While a unique set could be used in the "same relation" conditions, as argued, at least two feature-based templates were necessary in the other conditions. Indeed, following the filter switch hypothesis, no template could favour at the same time a small target among medium and an extra-large target among large distractors.

In summary, first, the relational vector could not account for all the results reported here (e.g. salience effects in random conditions, relevance effects in Exp. 2), but might perhaps be involved in some others (e.g. the priming and cueing effects in the block-wise experiment). Second, the influence of salience as well as the feature based account of cueing effects were supported in the present results. Finally, the relational vector account seemed to rest on some not completely secure grounds. Therefore, further research seem necessary to decide whether and how the two sets of data could be reconciled, either by subsuming the Becker's data under a feature-based account, or by assuming two modes of attentional guidance.