Change Detection: Paying Attention To Detail

Changes made during a brief visual interruption sometimes go undetected, even when the object undergoing the change is at the center of the observer's interest and spatial attention (Simons & Levin, 1998). This study examined two potentially important attentional variables in change blindness: spatial distribution, manipulated via set size, and detail level, varied by having the change at either the global or local level of a compound letter. Experiment 1 revealed that both types of change were equally detectable in a single item, but that global change was detected more readily when attention was distributed among several items. Variation of target level probability in Experiment 2 showed further that observers could flexibly set the detail level in monitoring both single and multiple items. Sensitivity to change therefore depends not only on the spatial focus of attention; it depends critically on the match between the detail level of the change and the level-readiness of the observer. 1. General Introduction There has been a recent explosion of interest within the psychophysical community in the role of attention in perception (e.g., change blindness, inattentional blindness, repetition blindness, the attentional blink, masking by object substitution, amnesic visual search). A central theme in this research is that perception of the visual world is not as rich as our subjective experience gives us to believe. In the present study, we examined the role of an attentional variable, detail level, which has not yet been studied systematically in this context. We believe that its consideration is at least as important as other factors that have been studied and that it may hold the key to some puzzles that currently beset the relationship between attention and perception. Much credit for the current interest in attention goes to Mack and Rock (1998), who reported a series of experiments in which they cleverly distinguished between perception with and without expectation. What they demonstrated was that observers were often unaware of otherwise salient visual stimuli if these stimuli were presented unexpectedly along with expected stimuli on which observers performed a difficult perceptual judgment. Even some unexpected stimuli presented to the fovea went undetected in these tasks, prompting the authors to coin the apt term "inattentional blindness." Research involving a related phenomenon, "change blindness," took this idea one step further in reporting that some changes to a scene go undetected even when the object undergoing the change is the focus of attention (Simons & Levin, 1998). On the surface, this finding is disturbing, since it implies that even focused attention on an object does not guarantee accurate perception. It is one thing to learn that unexpected and unattended objects are not always seen (Mack & Rock, 1998). But is even our subjective experience of seeing a fully attended object only an illusion? To help understand the claim that change to an attended object can go undetected, we review the background research briefly. Interest in change blindness began with reports that observers did not notice image changes made during a saccade in the inspection of a picture (e.g., a switch in hats worn by two gentlemen), although these same changes were easily detected when they occurred during a fixation (e.g., Grimes, 1996, McConkie & Currie, 1996). Other reports indicated that similar results could be obtained if the changes occurred during a brief visual interruption in the scene (Rensink, O'Regan & Clark, 1997), if the changes occurred during a 'cut' in a movie sequence of real-world actions (Levin & Simons, 1997; Simons, 1996), and even if they occurred during a real-world conversation between an unwitting participant and an actor. In this case it was the actor who exchanged places with another actor when a door being carried by other actors briefly interrupted the conversation (Simons & Levin, 1998). In was clear from the outset that attention to the relevant portion of a scene is a necessary component of successful change detection. Changes are detected more readily when they occur to objects that are of interest to the observer (Rensink, 1999; Rensink et al, 1997), and when they occur in locations that have had attention drawn to them by a salient cue such as a local visual transient or a unique color (Rensink et al, 1997; Scholl, 1999). However, attention to an object is also not sufficient to prevent change blindness. For example, in the study in which real-world participants failed to notice the change in an actor, there was every indication that participants were fixating the actor both before and after the change, and that the actor was the visual focus of interest in the scene (Simons & Levin, 1998). Similarly, in studies of movie cuts, the change involved a single moving actor in the scene approaching a telephone (Levin & Simons, 1997). Findings such as these indicate that focused attention does not guarantee detection of changes to the attended object. Or do they? We began our thinking on this issue by differentiating three different aspects of spatial attention, following Coren, Ward and Enns (1999; Chapter 15). Attention to a scene or an object can be understood in terms of: the locus of the attention (where in the visual field is the center of attention?), the extent or distribution of attention (how widely is attention spread over space?), and detail level (is attention set for the 'forest' or the 'trees'?). From this perspective, it is clear that studies of change blindness to date have manipulated the locus (e.g., Scholl, 1999) and distribution of attention (e.g., Rensink, 1999; Smilek, Eastwood & Merikle, in press). However, none have systematically varied the detail level. Before describing such a study, we will briefly summarize some of the past research on global-local perception in order to provide a context for the present study. The debate over whether visual analysis begins with the details (the 'trees') or with the larger configuration (the 'forest') has a long history (see review by Kimchi, 1992). Most modern studies of this issue rely on the logic outlined by Navon (1977), who presented observers with compound letters (a large or 'global' letter made up of many smaller or 'local' letters). The letters could be the same at each level (consistent) or different (inconsistent). Response time (RT) was recorded for the detection of a specified target letter that could occur at either the local or global level. The main findings were first, that RT to global targets was generally faster than to the local targets and second, that RT to local targets were slowed down more by a conflicting global letter than was RT to a global letter made up of inconsistent local letters. These patterns were interpreted as evidence for a more rapid analysis of the 'forest', or global precedence. Numerous studies since Navon's (1977) have examined various stimulus factors (Amirkhiabani & Lovegrove, 1996; Grice, Canham, & Boroughs, 1983; Hughes, Layton, Baird, & Lester, 1984; Kimchi, 1992; Kinchla & Wolfe, 1979; Martin, 1979; Navon & Norman, 1983) and observer expectations (Boer & Keuss, 1982; Lamb & Robertson, 1989; Miller, 1981; Robertson, Egly, Lamb & Kerth, 1993; Ward, 1982; 1983), leading to a large and confusing set of results. While there are certainly some conditions that favor global precedence (e.g., small local letters densely arrayed), others favor the local level (large local letters sparsely arrayed). Most notably, the attentional set of the observer (e.g., which level is expected) is often a more important predictor of the RT pattern than are the specific stimulus parameters that are used (Ward, 1982; 1983). Accordingly, the research focus in this area seems now to have shifted from the question of "which level is analyzed first?" to the more useful question of how sensory factors of stimulus registration (bottom-up) interact with internal mechanisms of selection and decision making (top-down) to bias perception toward one or the other level in a given situation. In what follows, we report two experiments examining the role of detail level in successful change detection. Experiment 1 found equally successful detection of global and local changes under focused attention, but superior global change detection when attention was spatially distributed. This pattern of results raised a number of questions, including (a) whether the results for focused attention were subject to ceiling effects, (b) whether the results for focused attention implied rich representations of attended objects or merely a probability matching strategy, and (c) whether the global detection advantage under distributed attention was primarily a sensory or attentional effect. These questions were addressed by the design of Experiment 2, where the probability of change at each level was manipulated. The results showed that change detection depends critically on the expectancy of the observer, for both focused and distributed attention conditions.