Pe~ceptualand Motor Skills, 1979, 49, 75-84. @ Perceptual and Motor Skills 1979
STRATEGIES FOR ATTENTION ALLOCATION IN MULTIPLE INPUT, MULTIPLE RESPONSE TASKS SID J. SCHNEIDER Franklin Delano Roosevelt Veterans Administration Medical Center Montrose, New York Summary.-9 tasks involvink multiple inputs and multiple responses were administered. T h e tasks involved monitoring, shadowing, and ignoring, i n all combinations, monaural and dichotic messages. Miss rates were compared to determine the effect of adding additional demands to different tasks. Results suggested that, when a demand was added to a task, attention was allocated away primarily from those demands already receiving the greatest amounts of attention, to accommodate the new demand. W h e n the new demand drew attention toward a channel, those demands already receiving the least amounts of attention gained the largest amounts. There were several exceptions to these trends. suggesting that the total amount of effort available for allocation was apparently an important variable. The implication of the results for research in attention was discussed.
In many psychological research tasks, subjects are asked to see or hear more than one input at once, or initiate more than one response at once. When the demands of such situations exceed the individual's capaciry, it is necessary for him to favor some inputs or responses over others, to perform well on at least one activity. The process by which individuals emphasize certain inputs, mental operations or responses at the expense of other ones has been called the allocation of attention (e.g., Kahneman, 1973) and has been the subject of much research. One of the important findings has been that attention can be allocated in a very large number of ways. For instance, one can choose one input over another on the basis of its physical characteristics (pitch, volume, brightness, location and so on) or on the basis of its content, e.g., its meaning (Broadbent, 1977; Johnston & Heinz, 1978). Also, while attention is being allocated among inputs it must also be allocated among responses which comPete for attention. The allocation of attention must be highly flexible and capable of adapting to added demands. The present study was to uncover the strategy subjects employ to allocate attention in certain attention allocation tasks. Kahneman ( 1973) proposed that the amount of effort or attention available for allocation at any time is variable and dependent in part upon the demands present at the time. The attention available could be allocated among more than one input or more than one response at once, according to Kahneman. Moreover, he proposed that a feedback loop operated, whereby the attention allocated to an input can change so that when an input requires a great deal of mental processing, effort can be reallocated accordingly. If rwo inputs were
received simultaneously, the feedback loop would ensure that the input requiring the most detailed processing received more attention than the other at the perceptual level. If even this increased attention were insufficient to meet the demands of the situation, the amount of total effort available might increase, so thac more could be allocated. Some authors (e.g., Mowbray, 1964; Ostry, Moray, & Marks, 1976; Moray, 1970) take issue with Kahneman's idea that attention can be allocated among more than one item at once and propose instead a "single-channel" model. They maintain thac inputs or activities can at best take turns on the single channel for processing. Both Kahneman's idea and the single-channel idea agree, however, that the resources available at any time are limited and must be distributed according to some strategy. When a subject is presented more than one input at once and asked to respond to each input, the subject will either have to divide his effort according to some priorities (Kahneman) or give one input more time on the single channel than the other (Moray and others). Whichever model better describes the situation, the end result would be superior performance in the responses to the favored input. For example, suppose a subject must perform a target detection task for one input, while other inputs are present and other tasks required. The subject's accuracy on the target detection task will depend upon several factors, including: the total capacity available, the emphasis given the input at the perceptual level, and the amount of effort devoted to the response. If the competing inputs or responses require a great deal of attention themselves, the subject has to devise some strategy or "economy" to allocate attention. Like any economy, there must be some rules and priorities governing it for optimal performance. In the present study, subjects received a host of different tasks, with a varying number o f inputs received and responses required. Some tasks required selective attention, i.e., listening to one channel while ignoring the other; some required divided attention, i.e., listening to two channels simultaneously; and some required undzvided attention, i.e., attending to one channel, the only one present. "Channel" is defined here as a spoken message to one ear. Such inputs apparently do not fuse with inputs to the opposite ear and do function as distinguishable channels (Puleo & Pastore, 1978; Ostry, Moray, & Marks, 1976). Thus, in the present study, sometimes an input had to be selected on the basis of one of its physical characteristics: which ear received it. A varying number of responses were required of a subject. Sometimes, the subject had to ignore the input; sometimes he had to detect targets in the input, i.e., "monitor" it, and sometimes he had to shadow it. Sometimes he had to both shadow and monitor it. The instructions called for the best possible performance on all tasks present, without one taking precedence.
STRATEGIES FOR ATTENTION
77
In this way, subjects often received two messages and had to respond in one or more ways to at least one message. The subjects had to hear non-ignored messages, and then words in the messages had to be recognized. The messages had to be processed, for shadowing or signal detection (monitoring). Finally, a response had to be issued-in the form of speaking or pressing a button. Most likely, competition occurred at the level of mental processing (Posner & Boies, 1971; Treisman & Fearnley, 1971; Iwasaki, 1976; Shaffer, 1975). The actual responses, i.e., speaking or pressing a button, required little attention (Posner & Keele, 1969), as did the memory retrieval involved in recognizing the words in the messages (Keele, 1972) so little competition was likely there. Either mental effort (e.g., Kahneman, 1973) or time on a single processing channel (e.g., Moray, 1970) had to be allocated among inputs, according to their requirements for processing. What were the rules governing this allocation? There were a number of ways subjects could apportion attentional resources. One way to discover what strategies subjects actually used is to compare the subjects' accuracy on two different tasks. For example, the subjects made a certain number of shadowing errors when they were asked to shadow and monitor the message to one ear while ignoring the other channel. This number of shadowing errors changed when the task became, shadow one channel while monitoring both. By comparing the two shadowing error rates, one could find the effect on the shadowing error rate of adding a monitoring task to the message to the ear opposite the one receiving the shadowed message. Presumably, adding a monitoring task will draw attention away from the remaining tasks. The channel which must now be monitored might receive additional attention, so that its targets can be detected. As the attention to this channel increases, the attention to the other channel must decrease, so performance for that channel should be impaired. Now consider the case in which che same demand of monitoring the opposite ear is added to the task of shadowing one ear and ignoring the other. The new task is shadowing one ear while monitoring the other. Just as in the former case, in which the demand of monitoring the opposite ear was added to the task of both shadowing and monitoring one ear while ignoring the other, adding the demand of monitoring the opposite ear should impair performance on the shadowing task in the task of shadowing one ear while ignoring the other. Will the degree of impairment be the same or will the degree of impairment vary, depending on the n a y e of the initial tasks? There are several possibilities. First, adding an additional task might have the same effect regardless of the effort required by the initial tasks. The possibility does not seem likely; the task of reading this article, for example, would probably be allocated different amounts of effort when it was added to the task of doing difficult mental calculations which require great mental effort, than when it was added to the
78
S. J. SCHNEIDER
task of simply tapping a pencil, which requires little efforc. An alternative possibility is that the amount of impairment caused by an additional task is proportional to the effort required by che initial tasks. Perhaps, the greater the amount of effort required by the initial tasks, the less the impairment brought on by the additional task. In this case, highly demanding tasks would lose less effort to an added demand than would less demanding tasks. This strategy would conserve effort for the more highly demanding initial tasks. However, still another possibility is that the greater the amount of effort required by the initial tasks, the greater the impairment brought on by adding the additional task. This strategy would preserve a relative equality in the allocation of effort among activities. Those inputs or responses which are already receiving large amounts of effort would be cut back so that new demands could be accommodated. This strategy would enable an individual always to allocace a certain minimal amount of efforc to every activity present, regardless of the presence of additional demands. Another possibility is that, when an additional demand is added to the tasks occupying an individual, he makes more effort available to his tasks so that he can sacrifice as little effort as possible from his original tasks. Nonetheless, the total effort available to him may remain insufficient to handle all he has to do, and he would have to resorc to one of the allocation stracegies. The present study was to determine which of these allocation stracegies subjects use when facing multiple inputs and tasks. METHOD Sz~bjectsand Procedare
Twenty-one employees of the Franklin Delano Roosevelt Veterans Administration Medical Center, Montrose, New York volunteered as subjects. There were 15 males and 6 females, all right-handed, with no known hearing impairment. Subjects were tested individually. They wore stereophonic headphones, held a momentary contact switch in each hand, and sat before a microphone. Each heard a tape of 13 1-min. tasks, each preceded by instructions. The inscructions informed the subject whether to ignore a message, shadow it, or monitor it, that is, press the button in the ipsilateral hand when a word ending with the letter s occurred. Each message used in the study was a I-min., 100word passage from Grimm's Fairy Tales, concaining 10 words ending with J separated from each other by at least two other words. Each message was composed of complece, coherent sentences. Messages were randomly assigned to tasks. The first four tasks were practice trials, consisting of the task, shadow one ear while the other ear receives silence (written "shadow/silence"), monitor
STRATEGIES FOR AT'I'ENTION
79
one ear while ignoring the message to the other ear ("monitor/ignore"), and monitor each ear while shadowing one ("shadow monitor/monitor"). The nine experimental trials included three tasks with only one message (shadow/silence; monitor/silence; and shadow monitor/silence), and six tasks with dichotic messages (shadow/ignore; shadow monitor/ignore; shadow/ monitor; shadow rnonitor/monitor; monitor/monitor; and monitor/ignore). These nine tasks were arranged in two counterbalanced orders, and about half of the subjects received each order. Shadowed messages were always received by the right ear while monitored messages were received by the left ear whenever possible.
Scoring Each subject's shadowing protocol was tape recorded. For a word to be scored correct, it had to be pronounced exactly correctly in the appropriate order. If the word was omitted, mispronounced, spoken with a stutter, or repeated needlessly, an error was scored. The subjects' accents were taken into account. If the subject inserted a word which did not appear in the message, an error was scored. If a word was omitted but replaced by a different word, only one error was scored. The number of errors of all kinds was added for a total error score for each shadowed message. So that monitoring tasks could be scored, each message a subject heard was simultaneously rerecorded. When a subject pressed a button in his left hand, a tone was recorded over the left message, and similarly for the right hand. If the subject pressed the button in the appropriate hand within two words after the occurrence of a word ending with the letter J, a hit was scored. If the subject pressed the button any other time, a false alarm was scored. Since each message contained 10 words ending with J, misses equaled 10 minus the number of hits. RESULTS Moizitoring Errors The mean number of misses and false alarms which occurred in each task are shown in Table 1. The results indicate that different tasks resulted in different rates of misses ( F s , l o o = 32.98, p < .001), and misses plus false alarms (l;8,100 = 21.75, p < ,001). t tests performed to find differences between the means (Kirk, 1968, p. 145) showed that fewer monitoring errors occurred in the shadow monitor/silence condition than in the monitor/silence condition ( p < .05 both for misses and for misses plus false alarms). Also, the shadow tnonitor/ignore condition resulted in fewer monitoring errors than did the monitor/ignore condition ( p < .05 both variables). In these cases, the number of monitoring errors committed in a message both monitored and shadowed was lower than it was in a message only monitored.
80
S. J. SCHNEIDER TABLE 1
MONITORING ERRORS Task
R
Monitor Silence Ignore Monitor R
Misses M
Monitor L
3.14 3.43 4.29 3.14 1.35 2.09 1.82 1.42 Misses Plus False Alarms M 5.19 3.43 3.81 4.86 1.90 1.47 2.18 2.48 SD Note.-Entries are the number of misses or There were 10 targets/message.
SD
Shadow Monitor
5.43 2.53
Shadow, Monitor Silence Ignore Mon- Mon-
1.90 1.76
1.95 2.06
4.62 1.94
7.48 1.29
5.81 8.14 2.76 2.71 6.81 2.29 1.62 2.62 2.62 2.26 misses plus false alarms in each message.
Student's t tests further indicated that fewer monitoring errors occurred in the shadow monitor/ignore condition than occurred on the shadowed message in the shadow monitor/monitor task, which in turn caused fewer monitoring errors than the nonshadowed message in the shadow monitor/monitor task ( p < .05 each time for both variables). As for the monitor/moniror task, there were more errors on the right side than the left, which an a posteriori Tukey test showed to be significant ( f < .05). Inspection of the data suggests that three subjects strongly favored their left ear on the task and were largely responsible for this effect. Shadowing Errors The number of shadowing errors committed in each task is shown in Table 2, and again there was a significant treatment affect (F5,100 = 33.12, p < ,001). t tests showed that the number of errors committed in the shadow/ monitor condition was greater than the number for the shadow monitor/monitor condition ( p .05). The number of errors committed in the shadow/ silence, shadow/ignore, and the shadow monitor/silence did not differ among themselves.
STRATEGIES FOR ATTENTION ALLOCATION IN MULTIPLE INPUT, MULTIPLE RESPONSE TASKS SID J. SCHNEIDER Franklin Delano Roosevelt Veterans Administration Medical Center Montrose, New York Summary.-9 tasks involvink multiple inputs and multiple responses were administered. T h e tasks involved monitoring, shadowing, and ignoring, i n all combinations, monaural and dichotic messages. Miss rates were compared to determine the effect of adding additional demands to different tasks. Results suggested that, when a demand was added to a task, attention was allocated away primarily from those demands already receiving the greatest amounts of attention, to accommodate the new demand. W h e n the new demand drew attention toward a channel, those demands already receiving the least amounts of attention gained the largest amounts. There were several exceptions to these trends. suggesting that the total amount of effort available for allocation was apparently an important variable. The implication of the results for research in attention was discussed.
In many psychological research tasks, subjects are asked to see or hear more than one input at once, or initiate more than one response at once. When the demands of such situations exceed the individual's capaciry, it is necessary for him to favor some inputs or responses over others, to perform well on at least one activity. The process by which individuals emphasize certain inputs, mental operations or responses at the expense of other ones has been called the allocation of attention (e.g., Kahneman, 1973) and has been the subject of much research. One of the important findings has been that attention can be allocated in a very large number of ways. For instance, one can choose one input over another on the basis of its physical characteristics (pitch, volume, brightness, location and so on) or on the basis of its content, e.g., its meaning (Broadbent, 1977; Johnston & Heinz, 1978). Also, while attention is being allocated among inputs it must also be allocated among responses which comPete for attention. The allocation of attention must be highly flexible and capable of adapting to added demands. The present study was to uncover the strategy subjects employ to allocate attention in certain attention allocation tasks. Kahneman ( 1973) proposed that the amount of effort or attention available for allocation at any time is variable and dependent in part upon the demands present at the time. The attention available could be allocated among more than one input or more than one response at once, according to Kahneman. Moreover, he proposed that a feedback loop operated, whereby the attention allocated to an input can change so that when an input requires a great deal of mental processing, effort can be reallocated accordingly. If rwo inputs were
received simultaneously, the feedback loop would ensure that the input requiring the most detailed processing received more attention than the other at the perceptual level. If even this increased attention were insufficient to meet the demands of the situation, the amount of total effort available might increase, so thac more could be allocated. Some authors (e.g., Mowbray, 1964; Ostry, Moray, & Marks, 1976; Moray, 1970) take issue with Kahneman's idea that attention can be allocated among more than one item at once and propose instead a "single-channel" model. They maintain thac inputs or activities can at best take turns on the single channel for processing. Both Kahneman's idea and the single-channel idea agree, however, that the resources available at any time are limited and must be distributed according to some strategy. When a subject is presented more than one input at once and asked to respond to each input, the subject will either have to divide his effort according to some priorities (Kahneman) or give one input more time on the single channel than the other (Moray and others). Whichever model better describes the situation, the end result would be superior performance in the responses to the favored input. For example, suppose a subject must perform a target detection task for one input, while other inputs are present and other tasks required. The subject's accuracy on the target detection task will depend upon several factors, including: the total capacity available, the emphasis given the input at the perceptual level, and the amount of effort devoted to the response. If the competing inputs or responses require a great deal of attention themselves, the subject has to devise some strategy or "economy" to allocate attention. Like any economy, there must be some rules and priorities governing it for optimal performance. In the present study, subjects received a host of different tasks, with a varying number o f inputs received and responses required. Some tasks required selective attention, i.e., listening to one channel while ignoring the other; some required divided attention, i.e., listening to two channels simultaneously; and some required undzvided attention, i.e., attending to one channel, the only one present. "Channel" is defined here as a spoken message to one ear. Such inputs apparently do not fuse with inputs to the opposite ear and do function as distinguishable channels (Puleo & Pastore, 1978; Ostry, Moray, & Marks, 1976). Thus, in the present study, sometimes an input had to be selected on the basis of one of its physical characteristics: which ear received it. A varying number of responses were required of a subject. Sometimes, the subject had to ignore the input; sometimes he had to detect targets in the input, i.e., "monitor" it, and sometimes he had to shadow it. Sometimes he had to both shadow and monitor it. The instructions called for the best possible performance on all tasks present, without one taking precedence.
STRATEGIES FOR ATTENTION
77
In this way, subjects often received two messages and had to respond in one or more ways to at least one message. The subjects had to hear non-ignored messages, and then words in the messages had to be recognized. The messages had to be processed, for shadowing or signal detection (monitoring). Finally, a response had to be issued-in the form of speaking or pressing a button. Most likely, competition occurred at the level of mental processing (Posner & Boies, 1971; Treisman & Fearnley, 1971; Iwasaki, 1976; Shaffer, 1975). The actual responses, i.e., speaking or pressing a button, required little attention (Posner & Keele, 1969), as did the memory retrieval involved in recognizing the words in the messages (Keele, 1972) so little competition was likely there. Either mental effort (e.g., Kahneman, 1973) or time on a single processing channel (e.g., Moray, 1970) had to be allocated among inputs, according to their requirements for processing. What were the rules governing this allocation? There were a number of ways subjects could apportion attentional resources. One way to discover what strategies subjects actually used is to compare the subjects' accuracy on two different tasks. For example, the subjects made a certain number of shadowing errors when they were asked to shadow and monitor the message to one ear while ignoring the other channel. This number of shadowing errors changed when the task became, shadow one channel while monitoring both. By comparing the two shadowing error rates, one could find the effect on the shadowing error rate of adding a monitoring task to the message to the ear opposite the one receiving the shadowed message. Presumably, adding a monitoring task will draw attention away from the remaining tasks. The channel which must now be monitored might receive additional attention, so that its targets can be detected. As the attention to this channel increases, the attention to the other channel must decrease, so performance for that channel should be impaired. Now consider the case in which che same demand of monitoring the opposite ear is added to the task of shadowing one ear and ignoring the other. The new task is shadowing one ear while monitoring the other. Just as in the former case, in which the demand of monitoring the opposite ear was added to the task of both shadowing and monitoring one ear while ignoring the other, adding the demand of monitoring the opposite ear should impair performance on the shadowing task in the task of shadowing one ear while ignoring the other. Will the degree of impairment be the same or will the degree of impairment vary, depending on the n a y e of the initial tasks? There are several possibilities. First, adding an additional task might have the same effect regardless of the effort required by the initial tasks. The possibility does not seem likely; the task of reading this article, for example, would probably be allocated different amounts of effort when it was added to the task of doing difficult mental calculations which require great mental effort, than when it was added to the
78
S. J. SCHNEIDER
task of simply tapping a pencil, which requires little efforc. An alternative possibility is that the amount of impairment caused by an additional task is proportional to the effort required by che initial tasks. Perhaps, the greater the amount of effort required by the initial tasks, the less the impairment brought on by the additional task. In this case, highly demanding tasks would lose less effort to an added demand than would less demanding tasks. This strategy would conserve effort for the more highly demanding initial tasks. However, still another possibility is that the greater the amount of effort required by the initial tasks, the greater the impairment brought on by adding the additional task. This strategy would preserve a relative equality in the allocation of effort among activities. Those inputs or responses which are already receiving large amounts of effort would be cut back so that new demands could be accommodated. This strategy would enable an individual always to allocace a certain minimal amount of efforc to every activity present, regardless of the presence of additional demands. Another possibility is that, when an additional demand is added to the tasks occupying an individual, he makes more effort available to his tasks so that he can sacrifice as little effort as possible from his original tasks. Nonetheless, the total effort available to him may remain insufficient to handle all he has to do, and he would have to resorc to one of the allocation stracegies. The present study was to determine which of these allocation stracegies subjects use when facing multiple inputs and tasks. METHOD Sz~bjectsand Procedare
Twenty-one employees of the Franklin Delano Roosevelt Veterans Administration Medical Center, Montrose, New York volunteered as subjects. There were 15 males and 6 females, all right-handed, with no known hearing impairment. Subjects were tested individually. They wore stereophonic headphones, held a momentary contact switch in each hand, and sat before a microphone. Each heard a tape of 13 1-min. tasks, each preceded by instructions. The inscructions informed the subject whether to ignore a message, shadow it, or monitor it, that is, press the button in the ipsilateral hand when a word ending with the letter s occurred. Each message used in the study was a I-min., 100word passage from Grimm's Fairy Tales, concaining 10 words ending with J separated from each other by at least two other words. Each message was composed of complece, coherent sentences. Messages were randomly assigned to tasks. The first four tasks were practice trials, consisting of the task, shadow one ear while the other ear receives silence (written "shadow/silence"), monitor
STRATEGIES FOR AT'I'ENTION
79
one ear while ignoring the message to the other ear ("monitor/ignore"), and monitor each ear while shadowing one ("shadow monitor/monitor"). The nine experimental trials included three tasks with only one message (shadow/silence; monitor/silence; and shadow monitor/silence), and six tasks with dichotic messages (shadow/ignore; shadow monitor/ignore; shadow/ monitor; shadow rnonitor/monitor; monitor/monitor; and monitor/ignore). These nine tasks were arranged in two counterbalanced orders, and about half of the subjects received each order. Shadowed messages were always received by the right ear while monitored messages were received by the left ear whenever possible.
Scoring Each subject's shadowing protocol was tape recorded. For a word to be scored correct, it had to be pronounced exactly correctly in the appropriate order. If the word was omitted, mispronounced, spoken with a stutter, or repeated needlessly, an error was scored. The subjects' accents were taken into account. If the subject inserted a word which did not appear in the message, an error was scored. If a word was omitted but replaced by a different word, only one error was scored. The number of errors of all kinds was added for a total error score for each shadowed message. So that monitoring tasks could be scored, each message a subject heard was simultaneously rerecorded. When a subject pressed a button in his left hand, a tone was recorded over the left message, and similarly for the right hand. If the subject pressed the button in the appropriate hand within two words after the occurrence of a word ending with the letter J, a hit was scored. If the subject pressed the button any other time, a false alarm was scored. Since each message contained 10 words ending with J, misses equaled 10 minus the number of hits. RESULTS Moizitoring Errors The mean number of misses and false alarms which occurred in each task are shown in Table 1. The results indicate that different tasks resulted in different rates of misses ( F s , l o o = 32.98, p < .001), and misses plus false alarms (l;8,100 = 21.75, p < ,001). t tests performed to find differences between the means (Kirk, 1968, p. 145) showed that fewer monitoring errors occurred in the shadow monitor/silence condition than in the monitor/silence condition ( p < .05 both for misses and for misses plus false alarms). Also, the shadow tnonitor/ignore condition resulted in fewer monitoring errors than did the monitor/ignore condition ( p < .05 both variables). In these cases, the number of monitoring errors committed in a message both monitored and shadowed was lower than it was in a message only monitored.
80
S. J. SCHNEIDER TABLE 1
MONITORING ERRORS Task
R
Monitor Silence Ignore Monitor R
Misses M
Monitor L
3.14 3.43 4.29 3.14 1.35 2.09 1.82 1.42 Misses Plus False Alarms M 5.19 3.43 3.81 4.86 1.90 1.47 2.18 2.48 SD Note.-Entries are the number of misses or There were 10 targets/message.
SD
Shadow Monitor
5.43 2.53
Shadow, Monitor Silence Ignore Mon- Mon-
1.90 1.76
1.95 2.06
4.62 1.94
7.48 1.29
5.81 8.14 2.76 2.71 6.81 2.29 1.62 2.62 2.62 2.26 misses plus false alarms in each message.
Student's t tests further indicated that fewer monitoring errors occurred in the shadow monitor/ignore condition than occurred on the shadowed message in the shadow monitor/monitor task, which in turn caused fewer monitoring errors than the nonshadowed message in the shadow monitor/monitor task ( p < .05 each time for both variables). As for the monitor/moniror task, there were more errors on the right side than the left, which an a posteriori Tukey test showed to be significant ( f < .05). Inspection of the data suggests that three subjects strongly favored their left ear on the task and were largely responsible for this effect. Shadowing Errors The number of shadowing errors committed in each task is shown in Table 2, and again there was a significant treatment affect (F5,100 = 33.12, p < ,001). t tests showed that the number of errors committed in the shadow/ monitor condition was greater than the number for the shadow monitor/monitor condition ( p .05). The number of errors committed in the shadow/ silence, shadow/ignore, and the shadow monitor/silence did not differ among themselves.
