The phenomenon of ‘Change Blindness’ ostensibly provides strong evidence that changes made to unattended items are not consciously detected (see Simons & Levin, 1997; Simons, 2000 for reviews). Subjects consistently fail to detect relatively large changes across a number of different methodologies.
Importantly, all such experiments (e.g.. Levin & Simons, 1997; Simons & Levin, 1998; 0 ’Regan, Deubel, Clark & Rensink, 2000; Rensink, O’Regan & Clark,
2000) typically measure only explicit detection of change. That is, subjects are asked
to respond deliberately to indicate whether they consciously detect a change in the stimuli. As discussed earlier in this chapter, an explicit question o f this type may not
reveal whether unattended changes can be extracted implicitly by the visual system.
Some recent research has focused on the issue o f whether undetected changes may be processed implicitly, and the experiments in the present thesis will pursue this further.
Femandez-Duque & Thornton (2000; Thornton & Femandez-Duque, 2000) claim that subjects, who declare themselves unaware o f any change, can show some evidence o f correct implicit localisation and identification o f changed items. In their first studies (Femandez-Duque & Thornton, 2000), observers were presented with a display of between 8 and 16 vertically and horizontally arranged rectangles, one of which changed in orientation over the course of two briefly presented successive displays (see Figure 1.6 for schematic examples of these studies). Subjects were told to attempt to identify where this change occurred. Their knowledge of the change was
assessed with a two-alternative forced choice test, in which two rectangles, in their original positions, were presented alone. One of these rectangles was the changed item; and the other was the item diagonally opposite to the changed rectangle’s position. Subjects were required to select which rectangle they thought was the one that had changed. Importantly, subjects were subsequently asked to indicate whether they had been ‘aware or unaware’ of the change. Fernandez-Duque & Thornton (2000) demonstrated that correct judgement of which item had changed could be somewhat above chance, even in trials where subjects indicated retrospectively that they had not been aware of the change. This was also supported by the fact that, despite signalling that they were unaware of the change, reaction time for correct selections of the changed item was significantly faster than for incorrect selections.
250 msec
250 msec
250 msec
B.
Figure 1.6. Schematic example of the general design used by F ernandez-D uque & T h o rn to n
(2000) and T h o rn to n & Fernandez-D uque (2000).
A. The changing displays and their timing. The red circle highlights the change, this circle was
not present in the actual experiments.
B. An example of the response-display a ppearing immediately a fter the presentation of the experimental displays in Fernandez-D uque & T ho rn to n (2000). Subjects make a forced choice decision regarding which of the two rectangles they believe changed.
C. An example response-display from Thornton & Fernandez-D uque (2000), again ap p earin g
immediately after the second experim ental display. Subjects here make a speeded decision regarding w hether the item is vertical o r horizontal.
The authors interpreted this to signify that subjects are able to form some representation of the changed location without awareness. They suggested that, in contrast to previous Change Blindness research, processing of change (albeit implicit) may be possible outside the focus of attention.
It should be noted that there is a possible flaw with this study. The main result relies on the fact that forced-choice accuracy can be above chance even when subjects
are reluctant to acknowledge awareness retrospectively. However, forced-choice
tasks in general usually reveal better performance than judgements of confidence; this is expected according to standard signal detection theory considerations. The second criticism concerns the possible role of criterion in the subjects’ own retrospective judgement of whether they were aware of the change they had just reported on, or were just guessing. There is likely to be some degree o f variation across subjects according to the point at which they declare that they are aware of a change. It is difficult to control for this fact, yet it could have the important effect o f some ‘aware’ trials being incorrectly coded as ‘unaware’ trials, thereby biasing the results to detect more ‘implicit’ processing than is actually taking place.
The authors also investigated whether implicit change detection might be a result of visual attention being redeployed to the location o f the changed item. To this end, further experiments required observers to make a speeded identity discrimination for a probe item appearing at the location of the change, or in the diagonally opposite position. Past studies have consistently shown probe-judgements of this type to be faster at attended locations (e.g., Posner, 1980; McCormick, 1997). Results suggested that there was no redeployment of attention to the site of the change, as there was no reliable facilitation of reaction time for identity discriminations at the locus of the change. The authors suggested that this confirmed implicit change detection occurred
in the absence of visual attention. These results raise a further criticism as this particular conclusion is based on a null-result. The apparent lack o f attentional allocation may simply be due to their particular method being unsuitable of insufficiently sensitive to demonstrate this allocation.
Thornton & Femandez-Duque (2000) investigated whether information regarding the identitv of the changed item (in addition to its location) can also be extracted implicitly. The method for these studies was similar to that discussed previously, except that rather than making a forced-choice decision between two items, subjects were now required to indicate as quickly as possible whether a probed item was horizontal or vertical. As before the probed item was either that which had
changed or the item diagonally opposite. Additionally, any item probed in the
position opposite the change could have the same identity to that o f the changed rectangle in the second experimental display, or the opposite identity. Thus, the effect
of two manipulations was assessed; probe validity and probe congruency. The probe
item was valid if it appeared at the changed location and invalid if it appeared in the alternative position. Congruency refers to whether the probed item’s orientation was congruent with the final orientation of the changed rectangle. In the valid-probe condition, the item was always congruent, as it was the changed item itself that was being probed. However, when the alternative position was probed (an invalid probe) the orientation could be identical (congruent) or the alternative orientation (incongruent) to the changed rectangle. Both of these manipulations were examined under conditions where subjects indicated retrospectively that they had been aware of a change and also when they indicated they were unaware of any change.
Results indicated that responses were speeded for items at the location of a change, as subjects were faster to indicate whether the item was vertical or horizontal
in the valid-probe condition. This suggests attentional allocation to the site of the change (in apparent contradiction of the results from the previous study by the same authors!). Additionally, there was some indication that subjects might possess some implicit knowledge o f the “nature” of the changing item. When responding to an invalidly cued item (i.e. away from the location of change), responses were facilitated if the item was congruent with the changed item in terms o f orientation. That is, if the cued item was identical to the changed item, responses were more accurate. Note that this was the case even for changes that subjects said retrospectively that they had not seen. Thornton & Femandez-Duque (2000) concluded that this provides evidence of more detailed information than simply the location of a change being detected in the absence o f awareness.
However, these results raise another criticism regarding Femandez-Duque and Thomton’s work. The issue of attentional allocation becomes confused over the course of their two papers. The first paper (Femandez-Duque and Thomton, 2000) putatively demonstrates that attention is not allocated to the changed region, thereby suggesting that any implicit processing occurs in the absence o f attention. However, within the second paper using the same basic paradigm, attention now appears to be drawn to the changing area, as validly-cued items are responded to faster than those that are invalidly-cued. The lack of attention allocation at the changing area in the first paper was based on a null-effect, so their method may not have been powerful enough to reveal attentional allocation. This criticism is supported by the fact that it appears attention was drawn to the changed location within the second paper. This being the case, detection of the change location even in the first paper could be based on attentional processing, rather than the changes being unattended. Additionally, this could have implications for their claims that the “nature of the unattended change” is
implicitly processed (Thomton & Femandez-Duque, 2000). If attention is allocated to the changed location, the orientation of the item at that position may presumably also be attended. This could result in priming of congment orientations at the invalid cued position, without any tmly unattended processing taking place.
Additionally, a further issue concems the displays themselves. The first
experimental display generally consisted of an equal number of vertical and horizontal rectangles; hence a change in the array results in the second display containing an unequal number of stimuli types (see again Figure 1.6). This is the case for most experiments in the first paper (Femandez-Duque & Thomton, 2000) and for all of the experiments carried out in the second paper (Thomton & Femandez-Duque, 2000). The effect of the unequal number is that the orientation of the changed item also becomes the most common orientation in the second display. This could prime responses to the changed orientation, without necessitating the need to invoke implicit change detection across the two displays. For example, if the subject is presented with a second display containing 5 vertical and 3 horizontal rectangles, they are then probed with a vertical item. It seems clear that priming to this vertical item could take place from within just the second display, without encoding and comparison of the
first and second display. Thomton & Femendez-Duque (2000) unconvincingly
attempt to dodge this criticism by claiming that an unreported control experiment somehow ruled this out.
Smilek, Eastwood & Merikle (2000) also attempted to assess whether there may be implicit processing of change. In their task subjects searched for a change occurring in a flickering visual-search image. The stimuli were created such that a change involved the manipulation of a varying number of features. Smilek et al discovered that search slopes for detecting changes involving a smaller number of
features were steeper than those for changes to a larger number o f features. That is, subjects were much slower to detect smaller changes, and increasing set-size
exacerbated this. The authors concluded from this that unattended changes were
being processed, arguing that this guided attention towards the change at a differential rate according to the magnitude of the change. They believed that if this were not the case, search for change should be uniformly slow and serial regardless o f the number of feature changes to the target item.
However, this study and its conclusions seem problematic. Smilek et al assert that their results demonstrate accumulation o f unattended processing of changes ‘guiding’ focused attention. However, it is conceivable that the results actually show something more straightforward. The fact that larger changes were detected more efficiently than smaller changes does not necessarily require explanation based on any prior implicit processing of the unattended changes. Rather, simply by being larger, these changes are easier to detect. They are more likely to be spotted as the subject searches the scene. It seems unnecessary to suggest that this is as a result of attention being drawn to them through prior implicit processing, especially since it is well known that easier discriminations generally lend to flatter set-size functions. Moreover, note that the search task used by Smilek et al. (2000) may involve ‘diffuse attention’ as discussed and criticised by Mack and Rock (1998).
Williams & Simons (2000) reported further evidence of putative implicit change detection. They asked subjects to identify whether a change occurred to a
novel object that travelled across the screen. While moving, it passed behind an
occluder. When the item emerged it was identical to before, or changed by a varying number of features. Despite explicit detection of change being poor (as expected from previous Change Blindness work), analysis of reaction time when subjects
claimed nothing had changed suggested that there may be some implicit detection of change. Subjects were consistently slower to make a ‘same’ response when the item had actually changed as opposed to when there really had been no change. Furthermore, there was a greater slowing of these responses when the novel object
was changed by a greater number of features. Williams & Simons’ (2000)
demonstration is suggestive that subjects may have extracted some information that the display had changed.