Download Divided Attention and more Study notes Literature in PDF only on Docsity!
Chapter 3
Divided Attention
Contents
3.1 Definitions and Domain............................. 33 3.2 The Visual Search Paradigm.......................... 34 3.3 The Dual-Task Paradigm............................ 44 3.4 Theoretical Accounts of Divided Attention................. 48 3.5 The Locus of Processing Dependencies................... 52 3.6 Visual Search Revisited: Simultaneous-Sequential Paradigm...... 53 3.7 Dual Tasks Revisited.............................. 62 3.8 The Generality of Divided Attention Effects................ 63 3.9 Chapter Summary................................ 63
3.1 Definitions and Domain
In the last chapter we introduced selective attention, which concerns the use of one source of infor- mation rather than another. We now turn to divided attention, which concerns the use of multiple sources of information rather than a single source. Divided attention in vision is fundamentally about the dependence versus the independence of visual processing across stimuli. As with selec- tive attention, we take pains to distinguish between phenomena, or effects of divided attention, and theoretical concepts that are used to explain effects of divided attention in terms of the relevant internal processes.
Consider first phenomena. Effects of divided attention are behavioral consequences (such as impaired performance) of manipulating the number of relevant stimuli. We can again use reading as an example. If we showed you a display of this page for a third of a second or so, and asked you to read the word that is printed in bold in this paragraph, you might be able to do that fairly well. However, if we asked you to read all of the words on the page in that period of time, you would probably be much less successful, and depending on which other words you read during
33
34 CHAPTER 3. DIVIDED ATTENTION
the brief period, you may or may not have gotten to the word in bold. In this example, reading performance changes as the number of to-be-read words increases from one to many, thus reflecting a dependence of reading one word on reading other words. This is an effect of divided attention across words.
Our focus in this example is to make clear it is an attentional effect rather than a non-attentional sensory effect. We did this by varying the number of relevant stimuli rather than the total number of stimuli. The total number of stimuli was actually the same across the two situations. To be clear, a difference in the ability to read the word “bold” when it is the only word on the page versus when it is one word of many on a full page of text should not be taken as definitive evidence of an effect of divided attention. This is because there are multiple sensory differences, unrelated to attention, that could lead to a difference in reading ability across those two situations. Divided attention effects are effects of the number of relevant stimuli, and must be distinguished from stimulus effects that depend on stimuli whether they are task relevant or not.
Now consider some corresponding theoretical terms used to refer to the nature of internal pro- cesses that give rise to divided attention effects. An extreme example, is serial processing. Some aspect of visual processing, say lexical access, might be capable of handling only one stimulus at a time. Such a theoretical statement predicts that any task that requires lexical access, such as reading, will show divided attention effects; performance will depend on the number of task-relevant stimuli. A contrasting extreme to serial processes would be those that proceed simultaneously (i.e., in parallel ) for multiple stimuli and are unaffected by the number of to-be-processed stimuli. For some situations, a set of completely independent parallel process is expected to suffer no divided attention effects. Besides serial versus parallel processing, there are other types of dependencies that can lead to divided attention effects. We will expand on these possibilities in a later section of this chapter.
In this early chapter, we introduce two basic paradigms that have been used to study divided attention: visual search and dual task. We first ask whether there is evidence from these paradigms that divided attention effects occur, starting with simple stimuli and tasks as we did with selective attention. We then offer an initial theoretical analysis, asking what aspect of processing is the source of the divided attention effects. As was the case for selective attention in the last chapter, this initial consideration of divided attention is limited to a consideration of simple stimuli and tasks and focuses on divided attention across space. The purpose is to introduce the paradigms, recognize some of their strengths and weaknesses, and understand the kinds of theoretical questions that can and cannot be addressed by each. A more fully developed treatment of divided attention is offered in Chapter xxx after we have considered a richer set of stimuli and tasks.
3.2 The Visual Search Paradigm
We first consider visual search which is a paradigm in which observers search a spatially distributed display of stimuli for a target stimulus. For example, you might have to search for a stimulus that is red among other-colored stimuli and respond “yes” when a target is present and “no” when it is absent. Although target presence versus absence is a common task, there are other ways of asking observers to report the outcome of their visual search. They might, for example, be asked
36 CHAPTER 3. DIVIDED ATTENTION
Set Size 6 Set Size 48
Figure 3.2: An illustration of two set-size conditions in Motter & Simoni (2008): Set-Size 6 on the left and Set-Size 48 on the right. The task was to detect a tilted-to-the-left, red “T” shape among a variety of distractors. The target is present in both of these examples in the upper right.
3.2.1 Example 1: Visual search with and without eye movements
This experiment was conducted in two ways. First, the observers were allowed to move their eyes freely to find the target. This is typical of many early visual search experiments beginning with Neisser (xxx) and Atkinson et al. (xxx). The results are shown in the top two panels of Figure 3.3. On the top left, correct response time is plotted as a function of set size and on the top right the percent errors is plotted by set size. Both functions are further subdivided by the presence or absence of the target. For both measures, performance declines with set size. The response time increases for both target present and target absent conditions. The increase is about twice as large for target absent compared to target present. The larger effect for target absent trials is from about 900 ms to 3000 ms. One account of such an effect is that groups of items are processed in sequence and one terminates the trial when one finds the target in the target present conditions but must go through the entire sequence for target absent. Indeed the eye movements allowed in this condition may contribute to the appearance of such serial processing. Regarding errors, there are relatively few errors on the target absent trials (false alarms) and they were not reported broken down by set size. In contrast, errors on the target present trials (misses) were more common and increased with set size from around 1% to 10%. In sum, there were effects of set size on both response time and errors but average errors remained relatively low (average misses of about 5%).
The second way the experiment was conducted was to require the observers to maintain central fixation and not move their eyes. To enforce this, eye position was monitored and trials rejected in which there was an eye movement. After a bit of practice, observers could do this and less than 1% of trials were excluded because of eye movements. Conditions with enforced fixation are not common in the literature.
3.2. THE VISUAL SEARCH PARADIGM 37
Response Time (ms)
Set Size
Target Absent
Target Present
Eye Movements Allowed
No Eye Movements
Response Time (ms)
Set Size
Target Absent
Target Present
Target Present Target Absent
Percent Errors
Set Size
Target Present Target Absent
Percent Errors
Set Size
Figure 3.3: Results from Motter & Simoni (2008). The top two panels are for conditions that allow eye movements and the bottom two panels are for conditions without eye movements. On the left side, the mean correct response time is plotted as a function of set size. On the right side, the mean percent errors is plotted as a function of set size. Both are further broken down by whether the target is present or absent. Both with and without eye movements, response time and errors increase with increasing set size. Preventing eye movements has little effect on correct response time but causes a large increase in errors especially with the larger set sizes. We plot accuracy in response time experiments in terms of percent errors rather than percent correct because there are usually few errors and so that increasing values reflect poorer performance in both measures.
3.2. THE VISUAL SEARCH PARADIGM 39
Figure 3.4: An illustration of crowding. Fixate the central cross and compare the visibility of the solitary Landolt C on the left with the visibility of the crowded Landolt C on the right.
the combined effects of eccentricity and the fact that crowding effects largely disappear when the relevant stimulus is directly viewed.
A key feature of this demonstration of crowding is that the relevant stimulus in the two conditions are identical: a single Landolt C. Instead, crowding is the result of a manipulation of the number of task-irrelevant stimuli. That the mere presence of the additional irrelevant stimuli can have a large impact on performance means that one has to control for the effects of crowding in order to interpret set-size effects as effects of divided attention.
The effect of crowding can be quite substantial and can occur for quite widely spaced peripheral stimuli. As a rule of thumb, crowding occurs for spacings that are half of the eccentricity of the relevant stimulus and smaller (Bouma, 1970). Thus, at a 10 ◦^ eccentricity, stimuli within 5 ◦ cause crowding. Clearly, increasing set-size as illustrated in Figure 3.2 will cause more crowding. Indeed, Motter and Simoni (2008) argue that the most if not all of the set-size effect in their experiment are due to a combination of eccentricity, crowding and the eye movements to overcome these effects.
In summary, although set-size effects in visual search have the potential for revealing effects of divided attention, differences in set size are often confounded with stimulus differences that can lead to sensory effects. We now present two examples of visual search experiments that have gone some way toward dealing with the confounds of eccentricity and stimulus density, and ask whether the remaining set-size effects are due to dividing attention across stimuli at multiple locations.
3.2.3 Example 2: Brief displays and minimal crowding
The task in this example (Carrasco, Mclean, Katz and Frieder, 1998, Experiment 2) was to detect the presence of a vertical red line among tilted red lines. The target was present on half the trials and absent on the others, and observers reported “present” or “absent” by making corresponding key presses as quickly as possible following the presentation of the display. Both response time and response accuracy were measured as a function of set size. In the actual experiment, set size varied from 2 to 32. For purposes here, we consider only set sizes 2, 4, 6, and 8, because these conditions
40 CHAPTER 3. DIVIDED ATTENTION
Figure 3.5: An illustration of two set-size conditions in Experiment 2 of Carrasco et al. (1994): Set Size 2 on the left and Set Size 8 on the right. The task was to detect a vertical target among tilted distractors. The vertical target is present in both of these examples.
allowed for stimulus spacings at which visual crowding should have been minimal (Bouma, 1970). The stimuli were distributed over a 6-by-6 array of possible positions with jitter. Thus the same positions (and eccentricities) were sampled for all set sizes. Figure 3.5 provides an illustration of the smallest and largest set sizes that we are considering here (2 and 8).
The conditions that we have described so far are typical of most visual search experiments in the literature. What is unusual about this experiment, however, is that the search displays were presented for only 100 ms, thereby preventing eye movements within the display. Thus, between limiting our consideration to conditions in which crowding should have been minimal and allowing for only a single fixation, this experiment presents an opportunity for asking whether there are set size effects in visual search above and beyond sensory effects. Set-size effects under these conditions can be interpreted as divided attention effects with more confidence.
Indeed, as shown in Figure 3.6 there were set-size effects in this experiment. The left panel shows mean correct response time a function of set size. The right panel shows mean percent error as a function of set size. As set size increased, both response time and errors increased. We can be confident that differences in eye position and eccentricity did not contribute to these set-size effects because no eye movements could be made given the brief displays, and stimulus positions were sampled from the same set of possible positions for all set sizes. In addition, crowding probably did not play a large role for these small set sizes given the large spacing between stimuli. In summary, these set-size effects appear to be examples of an effect divided attention across multiple spatial locations.
3.2.4 Example 3: Relevant set-size
The task for our third example (Palmer, 1994) was to detect the presence of a higher-contrast disk (30%, 40%, or 50% contrast) among distractors of a fixed lower contrast (20% contrast). Observers
42 CHAPTER 3. DIVIDED ATTENTION
Figure 3.7: An illustration of the two display-set-size conditions in Experiment 2 of Palmer (1994): Set-Size 2 on the left and Set-Size 8 on the right. The task was to detect a contrast increment target which is present in both of these examples. The relative contrasts are exaggerated to make them easily visible.
that in mind, we present a final example that is a follow-up to last experiment described. In this example, all possible stimulus influences on the measured set-size effect were eliminated by having the search displays across the different set sizes be physically identical. In this case, any set-size effect that occurs must be an effect of divided attention because there were no stimulus differences of any kind. Moreover, because this final example relies on a cueing procedure, it highlights the interdependence between divided and selective attention.
The strategy used here, which we will refer to as the relevant-set-size strategy, was to provide cues prior to the search display that indicated, in this case, either 2 locations or 8 locations that were relevant to the task. As illustrated in Figure 3.9, an initial display was presented that contained both a fixation cross and additional crosses in the periphery that indicated the location of every stimulus that would appear in the search display. The cues further indicated the relevance of each location with regard to the search task by being either black or white. In the illustrated example, the black crosses indicated relevant locations and the white crosses indicated irrelevant locations. In some experiments, this mapping has been reversed for some observers ensuring an arbitrary relation between cue-color and stimulus. A Relevant Set Size 2 example is illustrated on the left with 2 black cues and 6 white cues. A Relevant Set Size 8 example is illustrated on the right with 8 black cues. After the cue, a search display containing 8 discs was presented. Observers knew that targets would only appear in relevant locations. Thus, the relevant set size was 2 in one case and 8 in the other, even though the number of discs that were physically present in the search display was 8 in all cases. The critical question was would performance depend on relevant set-size?
The results are shown in Figure 3.10. The open squares and dashed line function show mean percent correct as a function of relevant set-size. Despite the fact that the physical displays for these two conditions were identical, there was clearly an effect of dividing attention across 2 versus 8 relevant spatial locations. The solid circles and solid line function are a replotting of the data from the display set-size experiment (see Figure 3.8). The relevant-set-size effect is essentially indistin- guishable from the display-set-size effect, and that the absolute performance in the corresponding conditions was also well matched.
3.2. THE VISUAL SEARCH PARADIGM 43
Percent Correct
Set Size
Figure 3.8: Results from Palmer (1994). The mean percent correct of 4 observers is plotted as a function of the display set size. Performance declines with increasing set size. The error bars are standard error of the mean.
The relevant-set-size effect that occurred in this experiment is clearly attentional. The use of identical stimuli makes that certain; there was no possibility of stimulus influences on the set-size effect. The relevant-set-size effect, therefore, is a clear example of a divided attention effect, and this is the most important point for purposes of this introductory chapter on divided attention.
Nonetheless, another important aspect of these experiments is that they provide a case for converging measures of the set-size effect. Consider first the weakness of the relevant-set-size experiment. Observers must be able to use the cues to allow them to search among stimuli at the cued locations and not at uncued locations. If observers either cannot or do not use the cues, then the relevant-set-size effect will be an underestimate of the effect of divided attention. Furthermore, if the selection mechanism is not as effective as actually changing the displayed stimuli physically, then the relevant-set-size effect will be an underestimate of the divided attention effect. Next consider the weakness of the display-set-size experiment. Any contributions of sensory effects would cause the display set-size experiment to overestimate divided attention effect. The fact that the results from the the display-set-size and relevant-set-size experiment were nearly identical is consistent with a particularly simple interpretation of these issues. For relevant-set-size, the observers used the cues reliably, and the selection mechanism was as effective as if the displays were actually changed. For display-set-size, the contributions of crowding and other sensory effects were small relative to the attentional effects. Of course, these conclusions are specific to the conditions of these experiments. But the fact that such conditions are achievable provides an example of isolating attentional effects from effects of the stimulus alone.
The experiments described here to introduce divided attention are success stories in finding converging measures of divided attention effects. They should, however, be considered exceptions in the larger world of visual search because set-size manipulations are often confounded with stimulus differences. If one does not control eye fixation, eccentricity and crowding, the measured effect of
3.3. THE DUAL-TASK PARADIGM 45
Display Set Size Relevant Set Size
Percent Correct
Set Size
Figure 3.10: Results from Palmer (1994). The mean percent correct of 4 observers is plotted as a function of the display set size for both display set size and relevant set size. Performance declines with increasing set size in a similar fashion for both manipulations. This is consistent with both effects being attentional.
examples of clear effects of divided attention in order to set the stage for developing a theoretical understanding of attentional processes. With that in mind, we start by describing an experiment in which the the two tasks were simple and identical (i.e., two luminance detection tasks), and performance was measured in terms of response accuracy. In dual tasks, response time introduces significant complications because dual-tasks (usually) require two separate responses. By speeding responses and measuring how long they take, contributions from the response processes themselves can contribute to differences between the dual- and single-task conditions. This is a topic of intense study in the larger world of attention (e.g., the psychological refractory period, Welford, xxx; Pashler, xxx). Our focus, however, is visual attention, and so we will limit our discussion to dual tasks in which observers are allowed as much time to make their responses as they need and performance is measured in terms of accuracy rather than time. Finally, we focus on tasks in which the stimuli are presented briefly and simultaneously. This prevents eye movements and the potential sensory contributions that we’ve discussed before. Indeed, most dual-task experiments are similar to the relevant-set-size experiment in that identical stimuli are used in all conditions. We will call experiments that satisfy these constraints dual search tasks to distinguish them for dual tasks more generally.
The procedure for our example dual-search experiment (Bonnel, Stein and Bertucci, 1992) is illustrated in Figure 3.11. Observers were asked to detect luminance increments that could appear on either the left or right side of a display, which consisted of two small LEDs on either side of a video camera lens (used to record eye position). Observers fixated the center of the lens of the camera which put the LEDs 7◦^ from fixation. Cues at the beginning of each trial indicated whether observers were to monitor only a single location (single-task condition) or monitor both locations (dual-task condition). Following the cue, observers indicated that they were ready by pressing a footswitch. After a short delay, a small luminance increment was (or was not) presented briefly (
46 CHAPTER 3. DIVIDED ATTENTION
Dual-Task Condition
Blank Interval (150 ms)
Luminance Increment (20 ms)
Blank Interval (till response)
Time
Precue (500 ms)
Single-Task Condition
Blank Interval (until “ready” response)
Figure 3.11: An illustration of the dual-task procedure used by Bonnel et al. (1992). The left column shows the dual-task condition and the right column shows the single-task condition. The displays are identical except for the precue that identifies the condition (two horizontal lines for the dual task and left or right vertical lines for the single task).
48 CHAPTER 3. DIVIDED ATTENTION
different for congruent and incongruent trials, then it would imply a dependency of processing across the two stimulus locations. In other words, an effect of congruency is another kind of attention effect. For this experiment, there was a hint of a congruency effect in the dual-task condition, but it was not statistically reliable (82% versus 67% correct for congruent and incongruent trials, respectively). There was not even a hint of a congruency effect for the single-task trials (78% correct in both conditions). We mention these congruency effects because they will show up in future chapters on both selective and divided attention. In particular, this is the kind of analysis that motivates the concept of interactive processing (e.g. crosstalk) in the next section.
In summary, for contrast detection with dual search tasks, there is no evidence of an effect of dividing attention across two locations. The lack of divided attention effects in this case seems to contrast with the divided attention effects found in the visual search experiments described in the previous sections. But before pursuing such interpretations, one must put some theory on the table.
3.4 Theoretical Accounts of Divided Attention
We now turn to theoretical accounts of divided attention. What aspects of internal processing can lead to dependence of processing across stimuli, and thereby effects of divided attention? To begin, we introduce three different kinds of processing dependence. Various combinations of these dependencies can lead to a variety of models, some of which make distinctive predictions. In this introductory chapter, we describe three theoretical distinctions and consider four generic models.
3.4.1 Unlimited versus limited capacity processing
The first theoretical distinction regarding potential processing dependencies is between unlimited and limited (processing) capacity. This distinction, like all of the processing dependencies we consider, can be thought of as involving a kind of independence property. Is the processing of an individual stimulus independent of the number of relevant stimuli? To make this concrete, consider the Bonnel and colleague’s experiment with two lights that was described above. Does the perception of a given light depend on whether one must judge that light alone or must judge both lights?
The term “capacity” derives from considering perceptual processing as a communication channel (Broadbent, 1958). The idea is that if additional stimuli do not impact the quality of information that is transmitted per unit time about each stimulus, then that processing has unlimited capacity. Unlimited capacity does not imply perfect processing. “Unlimited” simply refers to the usual quality of processing being unchanged by having to process additional stimuli (independence). In contrast, if processing has limited capacity then the quality of the information for a given stimulus declines as increasing numbers of stimuli are processed (dependence). The idea is that the outcome of a given process is either limited or not by how many stimuli must be processed.
Unlimited capacity is one extreme of the capacity distinction. The other extreme is a specific version of limited-capacity processing that we refer to as fixed-capacity processing, and it is worth
3.4. THEORETICAL ACCOUNTS OF DIVIDED ATTENTION 49
considering separately. For fixed-capacity processing, only a fixed total amount of information can be transmitted per unit time. As a consequence, the amount of information about any individual stimulus will be limited directly by the number of stimuli that must be processed. Fixed-capacity models imply an extreme dependence of processing and as a consequence they make specific pre- dictions regarding divided attention effects that can be useful in testing among alternative models. We will consider some of these in a later section of this chapter.
3.4.2 Parallel versus nonparallel processing
The second theoretical distinction regarding potential processing dependencies is between parallel and nonparallel processing. With parallel processing, the timecourse of the processing of any one stimulus is independent of the number of relevant stimuli. In contrast, nonparallel processing implies the timecourse of processing any one stimulus depends on the presence of other relevant stimuli. Consider again the Bonnel two-light example. Parallel processing implies the timecourse of processing one of the lights is unaffected by the relevance of the other light. The processing of each light has independent and identical timecourses.
The best known example of a nonparallel model is the standard serial model. In this model, information from each stimulus is processed one at a time in sequence. Eye movements provide a concrete example of a serial process. To directly view two lights, you have to move your eyes to view each light one at a time in sequence.
3.4.3 Noninteractive versus interactive processing
The third theoretical distinction regarding potential processing dependencies concerns the interac- tive processing of individual stimuli on individual trials. If channels of processing are noninteractive, then the processing of one stimulus is unaffected by the specific value of other stimuli that are being processed at the same time. If channels of processing are interactive, then the value of a given stim- ulus affects the processing that occurs for another stimulus. Consider the Bonnel example again. A example of interactive processing is to have the processing of one light affected by the value of the other light. For such a case, congruent lights have an advantage compared in incongurent lights.
The best known example of interactive processing is what we call the standard crosstalk model (e.g. Ernst, Palmer & Boynton, 2012; Navon & Miller, xxx). In this model, the stimuli are processed in parallel and without general dependencies on the number of stimuli (limited capacity). But, there are dependencies among the specific stimuli being processed. Specifically, there is some degree of pooling across the different stimuli.
3.4.4 An illustration of the three dependencies
These three dependencies are illustrated in Figure 3.13. The cube represents all combinations of the three dependencies. Each axis represents a different dependency. On the bottom are labels
3.4. THEORETICAL ACCOUNTS OF DIVIDED ATTENTION 51
for the dependencies in timecourse and in capacity. On the right, is the label for the dependency on interactive processes. This cube is more than a 2-by-2-by-2. One value of each dependency represents independence. There is only one way to be independent while there are many ways to have a dependency. Hence the cube has a relatively small volume to represent the independent side of each possible dependency.
3.4.5 Four example models
There are many different ways in which properties of these three potential sources of processing dependency can be combined to form specific process models. For purposes of illustration, we briefly introduce four different models and illustrate them in our figure of possible dependencies. The figure uses a cube to illustrate the range of all possible models. We use black to denote the cube and its labels; and, we use color to denote the four specific models.
The first and simplest model is the standard parallel model. As the name implies, this model assumes parallel processing. In addition, the modifier “standard” is used to indicate the further as- sumptions of unlimited-capacity and noninteractive processing. This is the simplest possible model within the context of the three potential sources of processing dependency described above. Each of the three properties – parallel, unlimited capacity, noninteractive stimulus-specific processing – implies independence of processing. This unique model is shown in blue.
The second model is the fixed-capacity, parallel model. Like the standard parallel model, this model assumes parallel processing. However, it also assumes fixed-capacity processing, which im- plies a particular processing dependence. This model and the other models with dependencies are shown in red.
The third model is the standard serial model. It assumes serial processing, which implies a specific dependence in the timecourse of processing. The modifier “standard” is used to indicate the further assumptions of limited capacity and noninteractive processing.
Finally, the fourth model is the standard crosstalk model. This parallel model is built around a dependency in stimulus-specific processing. The term “standard” refers to the assumption of par- allel processing and independence of any general effect of the number of relevant stimuli (unlimited capacity).
These four examples are intended as generic process models that can be applied to specific task contexts in order to build theories of divided attention. Such theories must elaborate how the model applies to a given task and stimulus set. Detailed examples are given in Chapter xxx on quantitative models.
3.4.6 Comment on terminology
The three dependencies of capacity, timecourse and interactive processing each have a history. The terminology of capacity has long been used in theories of attention from Broadbent (1958), to Kahneman (1973) to Pashler (1998). We maintain their usage in our treatment. But, in the
52 CHAPTER 3. DIVIDED ATTENTION
literature more generally, one must take care exactly how the term is used. Some authors (e.g. Townsend & Ashby, 1983) use it in a slightly different way to contrast capacity with dependencies in the timecourse of processing.
We depart from convention in contrasting parallel and nonparallel processing rather than parallel and serial processing. Our purpose is to distinguish the larger idea of independent time courses (parallel processing) from the specific example of nonparallel processing that is the standard serial model. This hopefully will make more clear the parallels between the three distinctions.
Finally, we differ from recent attention texts in considering dependencies due to interactive processing on par with dependencies due to capacity limits and nonparallel processing. We were drawn to this by the mounting examples of interactive processing we found in the literature. For this case, the terminology in the literature varies quite widely. Interactive processing (xxx) refers to any possible dependence between stimuli in separate channels. Crosstalk (xxx) refers to the partial “pooling” of stimulus information across channels. Stochastic dependence (xxx) refers to the trial-by-trial “noise” correlation of otherwise identical stimuli. For purposes of introduction, the concepts of interactive processing, crosstalk and stochastic dependence can be considered essentially interchangeable.
We now turn to one of the major challenges to building theories of divided attention. Effects of divided attention (e.g., set-size effects or dual-task effects) are measured using tasks that un- fold through multiple processes. Some of the processes might function independently for multiple stimuli, while others might have dependencies. Moreover, some of the processes may fall into the broader domain of interest (e.g., vision), whereas others that are necessary to do that task, nonetheless fall outside of domain of theoretical interest (e.g., response execution). One challenge, therefore, is ascertaining what component processes gave rise to the effect of interest. In the next section, we consider a first-pass through visual processing similar to what we did for selection in the last chapter. Specifically, we ask whether observed effects of divided attention reflect processing dependencies in perception, in decision, or perhaps in both.
3.5 The Locus of Processing Dependencies
Building toward a general theory of divided attention, we seek to ask which visual processes possess dependencies of one kind or another, as assessed by evidence of divided attention effects. This is the divided attention version of the locus question that we raised for selective attention in the last chapter. In that case, we asked about the locus of selection. Here we ask about the locus of processing dependencies. For selective attention, we began by contrasting the selective-perception hypotheses with the selective-decision hypotheses. For divided attention, we contrast dependent- perception hypotheses with dependent-decision hypotheses.
3.5.1 Dependent perception
The most common interpretation of set-size and dual-task effects is in terms of dependent per- ception. Specifically, they are interpreted as evidence that one or more perceptual process has
