B.1.3.2.1. Vincentization

The fact that RTs were shorter for the medium item than for the small one in the heterogeneous condition suggested that salience effects lasted long during the visual search. In order to further evaluate the time-course of salience effects, vincentized RT distributions were computed (Ratcliff, 1979). For the analysis of the RT vincentiles, two goals and one drawback dictated the procedure. First, we aimed at showing that salience effects were due to genuine processing speed variations. Indeed, one might suggest that mean RT differences occurred only because larger items were processed more often at first, and not because they were processed faster. Alternatively, salience might also have influenced "post-selective" processes, such as disengagement, for instance. To discard these interpretations, the effects had to be evidenced for the shortest RTs. Second, to further support the claim of a long life expectancy of salience, we aimed at testing the salience effects for the longest RTs. Finally, due to multiple comparisons (n = 12), the family-wise (or experiment-wise) error rate was high. However, the use of a Bonferroni-type procedure would lead to severely decrease the power of each comparison, thus increasing the Type-I error rate (i.e. the risk of not rejecting a wrong null hypothesis). Consequently, we used uncorrected t-tests, but considered them significant only when they were significant for the two consecutive quintiles (Q1 and Q2 for shortest RTs, and Q4 and Q5 for longest RTs; see Mozolic, Hugenschmidt, Peiffer, & Laurienti, 2008, for a similar procedure). Results were summarized in Tableau 1. Globally, these results (see Figure 2, top panel, and Tableau 1) confirmed the mean RT results for both the shortest and longest RTs. Indeed, no difference emerged at any point of the RT distribution for the homogeneous condition, while differences between small and medium and between medium and large were all significant at the short and long RT distribution extremities.

Tableau 1: Tests de Student entre tailles de cibles contiguës, pour les différents quintiles des distributions de TR studentisées. Les comparaisons sont considérées significatives (en gras) lorsque les valeurs de p sont inférieures au seuil alpha = .05, ou proches de l'être, pour deux quintiles consécutifs (Q1 et Q2, ou Q3 et Q5).