Psychol Res (1994) 56:83-98
Psychobgical Research PsychologischeForschung © Springer-Verlag 1994
Attentional zooming and the global-dominance phenomenon: Effects of level-specific cueing and abrupt visual onset Thomas H. Stoffer Institut fiJrPsychologie,Allgemeineund ExperimentellePsychologie,Ludwig-Maximilians-Universitfit,Leopoldstrasse 13, D-80802 Mtinchen, Germany (Also at the Max-Planck-Institutfiir PsychologischeForschung, Mtinchen) Received April 22, 1993/AcceptedJuly 22, 1993
Summary. In four experiments level-specific cueing of hierarchically structured stimuli was used to test the hypothesis that valid cues can reduce the global-dominance phenomenon. Compound stimuli (Experiments 1 and 2) or simple geometric forms (Experiments 3 and 4) were presented with different SOAs after a valid, an invalid, or a neutral level-specific cue (cue validity 80%). Costs for invalid cues and benefits for valid cues were produced in all experiments. However, a reduction of the global-local RT difference to about zero was achieved only after the reduction of the abrupt visual onset accompanying stimulus presentation in Experiments 2 (with compound stimuli) and 4 (with simple geometric forms). In addition, there was no longer the typical asymmetric-interference pattern (i.e., features of the global level interfere with local identification, but not vice versa) that was one of Navon's (1977) main arguments for assuming a perceptual precedence. It is concluded that the RT that is longer for local than for global identifications is produced by the time needed to refocus visual attention intentionally from the global level, which is focussed at first unintentionally, to the local level.
Introduction In the last fifteen years a great deal of research has been devoted to experimental tests of different functional explanations of a phenomenon termed global-processing dominance (Ward, 1983). This phenomenon was first observed by Navon (1977) who used hierarchically structured compound stimuli consisting of a large letter (global level) made up mosaic-like from identical small letters (local level). The subject's task was to report a letter at one level and to ignore the other level. Navon's experiments produced two patterns of results that demonstrate a processing dominance of global-stimulus information: first, a global latency advantage, i.e., the mean reaction time (RT) for reporting a letter at the global level was much shorter than
that for a letter at the local level; and second, an asymmetrical-interference pattern, i.e., RT for reporting the global level was not affected by interference from irrelevant letters at the local level, but RT to the local level increased greatly when both levels contained different letters. Two classes of functional explanations for the observed pattern of results were proposed. The first account (the global-precedence hypothesis), advanced by Navon (1977), holds that early visual processes involved in feature analysis are the primary functional causes of the global-dominance phenomenon. An alternative account, first proposed by Miller (1981), proceeds from the assumption that under certain conditions visual attention is first allocated to the global level. As the continuous publications on this subject illustrate (e.g., Greaney & MacRae, 1992; Kimchi, 1992; Lamb & Robertson, 1990; Lovegrove, Lemkuhle, Baro, & Garzia, 1991; Luna, Meriono, & Marcos-Ruiz, 1990; Navon, 1991; Paquet, 1992; Robertson & Lamb, 1991), the issue of whether global dominance is caused by perceptual or attentional processes, or even by both, is still an unsettled matter. The central thesis of the present study is that - everything else being equal - attentional zooming (see, e.g., Eriksen & St James, 1986; Eriksen & Webb, 1989; Eriksen & Yeh, 1985) of the local level is in most experimental conditions counteracted by an abrupt visual onset. This onset causes attention to be captured by that part of the visual field in which the onset is produced, this being the area occupied by the global level. Because attentional zooming from the global to the local level needs time, mean RT to local aspects of a stimulus should be longer than to global aspects. Therefore, the elimination of abrupt onsets and the use of level-specific cues should reduce the RT difference between conditions with global and local reports to about zero.
Navon's global-precedence hypothesis The principal empirical basis for Navon's global-precedence hypothesis is the twofold outcome of his experi-
84 ments, i.e., the global latency advantage and the asymmetrical-interference pattern, replicated in a number of studies using different procedures and stimulus materials (e. g., Antes & Mann, 1984; Luna, Merino, & Marcos-Ruiz, 1990; Navon, 1981, 1983; Navon & Norman, 1983; Wandmacher & Arend, 1985). It states that recognition of global features is faster than recognition of local features (Navon, 1977). This hypothesis is functionally substantiated by the assumption that in the course of feature analysis the visual system makes global features available before local features. When the global level is to be reported, local features should not interfere with global analysis because local features should not yet be available for most of the time needed by global analysis; but when the local level is to be reported, global features should interfere with local analysis because global features should already be available when local analysis starts. The global-precedence hypothesis accounts for the observed phenomena by reference to early visual processes. Broadbent (1977) and Navon (1977) recognized a relationship between the absolute size of the information at a certain level and the distribution of spatial frequencies corresponding to that level. The frequency spectrum corresponding to the global level contains more of the lowfrequency end of the spatial-frequency distribution, whereas the one corresponding to the local level contains more of the high-frequency end. In early visual analysis, the rate of processing decreases with increasing spatial frequency (e.g., Breitmeyer, 1975). This means that the global information may be processed by fast-acting spatialfrequency channels tuned to low frequencies and the local information by slow-acting channels tuned to high frequencies. The functional role of the composition of the spatialfrequency spectrum of the stimulus levels has been demonstrated in a number of experiments. Shulman, Sullivan, Gish, and Sakoda (1986) adapted their subjects to different spatial frequencies and assessed the effect on performance in a subsequent identification task. The results showed that lower adapting frequencies affected the global task more than the local one, while higher adapting frequencies affected the local task more than the global task. Shulman and Wilson (1987) corroborated this finding, showing that the detection of probe gratings composed of low spatial frequencies was facilitated by performance of a task that involved global processing. Conversely, the detection of gratings composed of high frequencies was facilitated by the performance of a local processing task. Blurring the stimulus, which is equivalent to a low-pass filtering of the stimulus, primarily raises the RTs to the local level (Lovegrove et al., 1991). The fact that global dominance is always reported if the stimulus pattern is exposed peripherally with spatial uncertainty, but that there is no dominance, or even a reversal to local dominance, when it is exposed centrally or with spatial certainty (Grice, Canham, & Boroughs, 1983; Lamb & Robertson, 1988; Luna et al., 1990; Stoffer, 1991), can be interpreted in the context of different sensitivity in the retina to high and low spatial frequencies as a function of retinal eccentricity (e. g., Fukoda& Stone, 1974). Hughes, Fendrich, and Reuter-Lorenz (1990) demonstrated that low spatial frequencies are ne-
cessary for the occurrence of global dominance. Altogether, these experimental results suggest that early perceptual processes are indeed involved in some way or another in the formation of the global-dominance phenomenon. Navon's (1977, Experiment 3) explicit conclusion from his experiments is that global dominance is inevitable. Navon (1981) qualified his precedence hypothesis by suggesting that - everything else being equal - global structure is available in the percept earlier than are local features. This is nothing but a modified inevitability assumption conditioned by the comparability of the global and local forms with respect to discriminability. At present, there is a growing body of data that questions the inevitability assumption. The data demonstrate the dependence of the global-dominance phenomenon on factors that affect either stimulus structure or the conditions of stimulus exposure. No dominance at all, not even local dominance, has been found in several experiments. For compound letters of equal size with many, rather than few, local elements, Martin (1979) found global dominance when many elements were used and local dominance when few elements were used. Kimchi and Palmer (1982) varied the number of locai elements and their sizes. Their results showed that in patterns composed of a few relatively large elements, the elements can be perceived as individual parts of the global structure. This produces local dominance. However, when many quite small elements are used, global dominance is observed. Hoffman (1980) selectively distorted the figural quality of the global or local level. This manipulation produced global dominance when the local level was distorted, and local dominance when the global level was distorted. Kinchla and Wolfe (1979) demonstrated global dominance with stimuli of less than 7 ° of visual angle, no dominance with 8° of visual angle, and local dominance with more than 8 ° of visual angle. Similar results have been presented by Lamb and Robertson (1989, 1990). Stoffer (1991) and Greaney and MacRae (1992) used three-level compound letters to test the generalizability of Navon's hypothesis of a global-to-local processing sequence. Contrary to expectations derived from Navon's hypothesis, the medium level was the one with the longest response latency in both studies. Stoffer (1991) compared the identification of two- and three-level compound letters. His results showed that the same stimulus was recognized faster when it was the global level of a two-level letter rather than the medium level of a three-level letter. Similarly, symmetrical or reversed asymmetrical-interference patterns, which were not compatible with Navon's precedence hypothesis, were observed in some studies (e. g., Antes & Mann, 1984; Hoffman, 1980; Miller, 1981; Stoffer, 1991). Likewise, factors that affect conditions of stimulus exposure can prevent global dominance. We have already mentioned the variation in size of the global-dominance effect as a function of retinal eccentricity (Grice et al., 1983; Lamb & Robertson, 1988; Luna et al., 1990; Stoffer, 1991). Another result that highlights this point was reported by Paquet and Merikle (1984). They demonstrated that the interference pattern is affected by exposure duration. When the compound stimulus is presented for only 10 ms, the typical asymmetrical-interference pattern was observed. At
85 longer exposure durations (40 ms and 100 ms), the interference pattern was symmetrical. Altogether, the experimental demonstrations of size modulations of the globaldominance effect and the fact that it can be reversed into local dominance invalidate the inevitability assumption of Navon (1977). The plausibility of an explanation that makes exclusive use of perceptual processes may be challenged as well, even in those cases in which a global-dominance effect is observed. In an analysis of RT distributions from an experiment in which subjects had to attend to both global and local information, Miller (1981) failed to find any clues that indicated a different time course in the processing of global and local information. He therefore argued for an attentional explanation. On the assumption that information about all levels is available at around the same time, any kind of processing dominance could result from a difference in the facility of directing attention to the different levels of a stimulus pattern. An analysis of speed-accuracy trade-off functions by Boer and Keuss (1982) revealed no differences in the time course of global and local features. This corroborates Miller's (1981) conclusion that global dominance must be a an attentional effect. If all the reviewed data are taken into account, a perceptual-precedence explanation of the global-dominance phenomenon no longer seems appropriate. While it would be premature to argue for a specific alternative explanation, it seems safe to conclude that the global-dominance phenomenon is based on postperceptual processing (Boer & Keuss, 1982; Hoffman, 1980; Miller, 1981; Wandmacher & Arend, 1985) - that is, it reflects which part of the available information is utilized first by the subject during the task, rather than a specific order of its initial perceptual analysis.
Global dominance as an attentional phenomenon
The fact that attention can be allocated selectively at a level of a hierarchically structured object was first proposed by Eriksen and Rohrbaugh (1970). This feature of visual attention may be described by use of the zoom-lens metaphor (Eriksen & St James, 19'86; Eriksen & Webb, 1989; Eriksen & Yeh, 1985): like a variable zoom lens, visual attention can be allocated to different-sized areas in the visual field. A change in the level attended to can be conceived of as a redistribution of resources over an area of the visual field that is larger in proportion to the height of the level in the hierarchy (Castiello & Umilt~, 1990; Eriksen & St James, 1986). Some experiments have more or less directly addressed the question of whether visual attention is involved in the formation of the global-dominance phenomenon. Lamb and Robertson (1988, 1990) suggest that the size of the attended area is of functional significance in identifying hierarchically structured stimuli. Lamb and Robertson (1988) showed that the RT to the local level of foveally presented stimuli was longer when they occurred randomly and were intermixed with peripherally presented stimuli than in the condition in which all stimuli were presented in the center of the screen. They explained these results by assuming that in the condition with central-stimulus presentation a
smaller focus of attention could be maintained, entailing a benefit for local-stimulus analysis. An explicit test of an attentional explanation of the global-dominance phenomenon must include an experimental manipulation that affects attentional zooming. One attempt to do this was an experiment by Kinchla, SolisMacias, and Hoffman (1983). They manipulated the relative probability of a target letter at one level of a compound letter. Their results show that the RTs for a level become shorter with an increase in target probability. If the target probability is 1.0 for a level, RT is the same for the global and local levels. These results were replicated by Robertson, Lamb, and Knight (1988). This relatior~ship implies a trade-off between the identification speeds for the global and local levels: information from one level is utilized faster only at the cost of slower utilization of information from the other level. As a consequence, if the probability of a target is high for the global level, and therefore low for the local level, a global-dominance phenomenon occurs; if it is high for the local level, a localdominance phenomenon will result. In the experimental conditions introduced by Navon (1977), the subject's attention is always directed to the same level in a block of trials. This is equivalent to a condition with probability 1.0 for a target at a specific level. On the basis of the results of Kinchla et al. (1983), one would expect that no dominance should result whenever attention is consistently directed at a level within a block of trials. But as a rule, this is not so (e.g., Navon, 1977, 1981; Wandmacher & Arend, 1985). As a rule, variation by instruction of the location on which attention is selectively focussed between blocks of trials does not produce strong effects of advance knowledge about the location of the imperative stimulus on processing (e. g., Grindley & Townsend, 1968; Mertens, 1956; Shiffrin & Gardner, 1972). Ward (1982) used double identification to control the level at which attention is directed when the imperative stimulus is presented. In a block of trials, subjects were instructed to report the first of a pair of compound letters at one level-immediately after they had responded, another compound letter was presented, which again had to be reported at one level according to instruction. It was assumed that the allocation of attention to a specific level of the first letter would be maintained until the second letter was reported. The pattern of the results of several experiments by Ward (1982) can be summarized by what he called the level-readiness effect. It appeared that processing at a certain level is faster if the processing directly precedent was directed at the same level. Nevertheless, global dominance was still observed in these experiments. However, when Kinchla et al.'s (1983) method with different target probabilities for the two levels and Ward's (1982) doubleidentification task were combined, global dominance was no longer observed (Ward, 1985). Apparently, very effective means of controlling attentional zooming are needed to prevent a primary focussing of the global level. Thus, it can be concluded that attentional manipulations affect the relative speed for the use of global and local information in action control. We suggest that the globaldominance phenomenon may result from the focussing of attention at the global level first - which may capture at-
86 tention contrary to the subject's intention. Only after attention is refocussed at the local level, can the utilization of the local information for response selection start. The RT difference between global- and local-directed conditions can, at least in part, be a consequence of the extra time needed to refocus attention in such a condition. The hypothesis tested in the following experiments states that if this extra time can be saved by successfully directing the attentional focus at the level to be reported, the RT difference between global and local reports should be reduced to about zero.
H H H i"- H H H !"q H H H H H
F F F ~ F:: F:i = F= F F
F F F -FFFFFF F F
H -q H -q H ~- H H H HHHHHHH H FH
F F ~= :.7F -
Experiment 1 An effective means of controlling spatial focussing of visual attention is Posner's cost-benefit analysis of spatial cueing (e.g., Posner, 1980; Posner & Snyder, 1975). In these experiments, RTs are compared between conditions with valid cues (that are informative about the location of the imperative stimulus) and conditions with neutral cues (that are noninformative). The RT difference between these conditions defines the benefits of spatial cueing. In addition, a small proportion of invalid cues is used to establish the effects of inadequate spatial focussing of attention. The RT difference between conditions with invalid and neutral cues defines the costs. Significant costs and benefits of spatial cueing in terms of RTs have been found in detection tasks (e. g., Posner, Nissen, & Ogden, 1978; Posner, Snyder, & Davidson, 1980) as well as in identification tasks (e.g., Rock & Gutman, 1981; Tsal & Kolbet, 1985). The time course for shifting attention shows that benefits usually rise up to about 300 ms or 400 ms SOA between cue and imperative stimulus and then drop off again (e.g., Remington & Pierce, 1984; Tsal, 1983). After that, increasing costs can be observed for the processing of stimuli at the cued location (e.g., Maylor, 1985; Maylor & Hockey, 1985). If these results can be generalized to attentional zooming, the temporal parameters of attentional focussing are probably the functional cause of the ineffectiveness of instructions to control attentional zooming before stimulus presentation. The purpose of Experiment 1 was to test the hypothesis that the global-local RT difference should approach zero for conditions with valid level-specific cues presented well before the compound letters. In general, an RT difference of about zero should exist between any conditions associated with the same number of steps taken in the focussing of attention. Thus, the addition of one step for a global report by presentation of an invalid local cue prior to the stimulus should result in about the same RT as for a local report after valid cueing, assuming that inhibition of the prepared local focussing represents an additional focussing step. With a local cue and the presumption of global dominance, two focussing steps are involved: for a global report, these steps consist of focussing the global level first and then inhibiting the prepared focussing of the cued local level; for the local level, they consist of focussing the global level first and then refocussing to attend the cued local level. This hypothesis will be tested in Experiment 1, with a combination
[]
Fig. 1. Stimuli used in Experiment 1 (top: the two consistent stimuli; middle: the two inconsistent stimuli; bottom, from left to right: global cue, neutral cue, and local cue)
of Posner's cost-benefit analysis and Navon's interference task with level-specific cues.
Method Subjects. Eight students from the University of Bielefeld, Germany,
were paid to participate in five sessions run on five consecutive days. Their age ranged from 19 to 23 years. All subjects were right-handed and had uncorrected normal vision. Apparatus and stimuli. Stimuli were presented as green pixels on a
dark background on the CRT screen of an Apple IIe microcomputer (with fast P31 phosphorus) that also controlled the presentation of the stimuli and registered the responses. Warning and instruction tones were generated by the internal loudspeaker of the computer. Subjects responded by pressing one of the two special keys to the left and right of the keyboard's space bar. Figure 1 shows the stimuli, together with the cues. The size of the local elements, when viewed from a distance of 1.65 m, was 0.18° of visual angle horizontally and 0.20° vertically; that of the global stimuli was 1.9° horizontally and 2.1 ° vertically. The global and local cues were slightly larger than the letter at the corresponding level (plus 0.1°). The neutral cue was vertically and horizontally half the size of the global cue. Stimulus luminance was 3.7 cd/m 2, the screenbackground luminance was 0.12 cd/m 2. The room was dimly lit (29 cd/ m2 measured at the wall directly behind the CRT screen). Conditions. The experimental design comprised five orthogonal factors: (1) level to be reported (global or local); (2) stimulus consistency (consistent or inconsistent, see Figure 1); (3) cue validity (of the informative cues 80% were valid and 20% invalid, while 33% of all cues were neutral); (4) SOA between cue and instruction tone (SOA,:
87
120
100 ms or 700 ms; this tone was to inform the subject which level to report); and (5) SOA between instruction tone and imperative stimulus (SOA2:100 ms or 700 ms).
100
Procedure. A trial started with the presentation of a fixation cross in the center of the CRT screen for 1,000 ms; 500 ms after its onset, a 1,000-Hz warning signal sounded for 100 ms. Cue presentation started with the offset of the fixation cross and ended 40 ms before the imperative stimulus was exposed; 100 ms or 700 ms (SO&) after the onset of the cue, the instruction tone, specifying the level to be reported, was presented (global level; 2,500 Hz; local level: 250 Hz). The instruction tone, presented for 100 ms, was followed by the imperative stimulus with SOA 100 ms or 700 ms (SOA~). After offset of the stimulus, a dark field was presented until the subjects responded. Error feedback was provided as text on the screen. The display was then erased, and the next trial started 3 - 4 s later. Subjects were told that they had to determine which letter (H or F) had been displayed on the level designated by the instruction tone. It was stressed that the instruction tone, and not the cue, indicated which level to report. They were further told that global and local cues would provide important preparatory information: 80% of the informative cues would indicate the level to be reported. They were also told that it would be to their advantage to attend to the cued level immediately, so as to increase their response speed and accuracy. Subjects responded by pressing the left key with their left index finger whenever the letter F was presented and the right key with their right index finger on presentation of the letter H. They were encouraged to respond as quickly as possible, but still to keep the error rate low. At the end of each block of trials, feedback on the error percentage was given. It was announced that subjects with errors of more than 5% in the first session would have to be excluded from the experiment. Two subjects did not reach this criterion. Each experimental condition was conducted 50 times with neutral, 80 times with valid, and 20 times with invalid cues. Within one block of trials, either informative (valid or invalid, randomized within a block) cues or neutral cues were used. The SOAs were blocked. All the other factors were randomized. Each stimulus alternative appeared 50% of the time at each level. The experiment was conducted in five sessions on consecutive days, the first one being considered a practice session. Each session was divided into five blocks of 128 trials each, starting with eight practice trials; one block for each of the four combinations of SOAs with informative cues and one additional block with neutral cues, with SOAs blocked within this block of trials. The order of blocks (as well as subblocks with different SOAs within the block with neutral cues) was randomly selected for each subject. Each block lasted 12 rain and was followed by a break of about 5 rain.
Results and discussion M e a n correct RTs per c o n d i t i o n and subject w e r e calculated and e n t e r e d into a f i v e - w a y analysis o f v a r i a n c e ( A N O V A ) for r e p e a t e d m e a s u r e m e n t s on all factors. RTs o f less than 100 ms and o f m o r e than 2,000 ms w e r e c o n s i d e r e d errors (<0.5%). T h e o v e r a l l m e a n error rate was 1.9%. A n error analysis was also c o n d u c t e d , to e x c l u d e the possibility o f a s p e e d - a c c u r a c y trade-off. B e c a u s e its results m i r r o r e d those o f the R T data, it will not be p r e s e n t e d in detail.
The efficiency of level-specific cues. A test o f the central hypothesis p r e s u p p o s e s a d e q u a t e e f f i c i e n c y o f the l e v e l specific cues. T h e significant m a i n effect o f C u e Validity, F(2,14) = 77.8, p <.001, and Scheff~ tests r e v e a l e d that this c o n d i t i o n was m e t o v e r a l l by invalid cues (mean: 596 ms) p r o d u c i n g significant costs (mean: 57 ms; p <.001), but not by valid cues (520 ms) p r o d u c i n g no significant benefits
~- SOA2=100msvalid cue 4- SOA2=700msvalid cue ~ :OA:_--1700msinvalid cue
/ /
,-, 8 o E ,7cn
.'~_
6o
¢.. m -o 40-
%
c
-a_
+ "-" 20¢J} o O 0-
m
-20
I 100
I
SOA1 (ms)
700
Fig. 2. Mean costs and benefits of level-specific cueing as a function of SOA, (between cue and instruction tone), SOA~ (between instruction tone and imperative stimulus), and cue validity in Experiment 1
(19 ms). H o w e v e r , the significant interaction o f the factors Cue Validity, S O A , and SOA2, F(2,14) = 13.3, p <.01 (see F i g u r e 2), r e v e a l e d a significant benefit w h e n b o t h S O A s w e r e 700 ms (28 ms, F(2,14) = 3.4, p <.10), simple m a i n effect o f C u e Validity and Scheff~ test). This is the condition in w h i c h the m a x i m u m benefit should be e x p e c t e d , b e c a u s e there was e n o u g h t i m e to focus the l e v e l specified by the cue b e f o r e the instruction t o n e s i g n a l e d the cue validity. T h e r e was also e n o u g h t i m e to e x e c u t e the prepared f o c u s s i n g after the cue turned out to be valid. In contrast to the c o n d i t i o n of S O A 2 = 700 ms, only a nonsignificant benefit (about 14 ms) c o u l d be o b s e r v e d w h e n SOA2 was 100 ms. T h e m a i n cause o f the interaction b e c o m i n g significant was the difference b e t w e e n the costs as a f u n c t i o n o f SOA1 w h e n SOA2 was 100 ms and w h e n it was 700 ms (see F i g u r e 2). W h e n S O A 2 was 100 ms, costs w e r e v e r y m u c h higher, w i t h SOA~ o f 700 ms c o m p a r e d to 100 ms; w h e n SOA2 was 700 ms, costs w e r e l o w e r w h e n S O & was 700 ms c o m p a r e d to 100 ms. This pattern o f results can be e x p l a i n e d if one assumes that o n l y in c o n d i t i o n s with a r e l a t i v e l y l o n g SOA~ l e v e l - s p e c i f i c cues can be u s e d e f f e c t i v e l y to prepare attentional focussing. W h e n the cue turns out to be invalid and there is not e n o u g h t i m e to alter t h e p r e p a r e d f o c u s s i n g (in c o n d i t i o n s w h e n SOA2 = 100 ms), strong costs result b e c a u s e the w r o n g l e v e l will be f o -
88 cussed first and an additional focussing step has to be executed. But when SOA2 = 700 ms, there is enough time to correct the prepared focussing of the wrong level, at least in part.
The effects of cues on the global-local R T difference. If valid cues were effective in directing attention successfully to the level to be reported, the global-local RT difference should be about zero. For neutral cues, the tendency to attend to the global level first should produce longer RTs for global than for local reports. For invalid cues, the global-local RT difference should again be zero, but the means should be much higher than for valid cues. Figure 3 shows the mean RTs for conditions with global and local reports and different cue validities. If the hypothesis were valid, the interaction of the factors Level to be reported and Cue Validity should be significant. But this is not the case. The global-local RT difference is about the same in all cue conditions. The main effect of the Level to be reported is significant (global: 523 ms, local: 579 ms, F(1,7) = 55.9, p <.001) showing a global-dominance effect of 56 ms. It was also predicted that when global dominance is produced by focussing the global level first, regardless of the level to be reported, then adding one additional focussing step to a global report should produce about the same RT as a local report, since in the latter situation two focussing steps are involved any way. The mean RT for global report after presentation of an invalid cue was 571 ms, the mean RT for local report following a valid cue was 555 ms (see Figure 3). This difference is not significant, t = .84, df = 7, p >.25. In this case, level-specific cueing effectively reduced the global-local RT difference to about zero. This result cannot be accounted for in terms of Navon's perceptual-precedence explanation. If one attempts an explanation that rests on attentional mechanisms alone, the reason for the relative ineffectiveness of valid cues for directing attention to the intended level can be conceived of only as an involuntary focussing of the global level. Introspective reports from our subjects suggest that attention is captured by the abrupt visual onset of the bright imperative stimulus on the dark screen. Subjects frequently said that they just could not prevent themselves from attending first to the whole area on which the stimulus appeared. Only afterwards did they intentionally refocus their attention at the local level when this level had to be reported. There is empirical evidence that the abrupt appearance of an object may capture attention. Jonides (1981) demonstrated that a peripheral cue with abrupt onset initiated a shift of attention to its location and that this exogenous (unintentional) orienting was not affected by increased processing demands, whereas endogenous (intentional) orienting was affected• He also demonstrated that subjects who were instructed to ignore the peripheral cue were unable to do so, but had no problem in ignoring a central cue, which would otherwise initiate endogenous orienting. This result shows that attention shifts that follow an onset stimulus are not under voluntary control• The function of abrupt visual onsets in capturing attention has been documented in several experiments (e. g.,
Cue Validity 630
valid -+-
neutral
~-
invalid
580
g /
,+
$1 / / / 11
530 -
480
j
I
global
local
Level to be Reported
Fig. 3. Mean reaction time (RT) as a function of the level to be reported and cue validity in Experiment 1
Miller, 1989; Remington, Johnston, & Yantis, 1992; Theeuwes, 1991; Yantis & Jonides, 1984, 1990). Todd and Van Gelder (1979) developed a no-onset procedure consisting of a stimulus presented with additional line segments before the imperative stimulus. The stimulus was then presented with the removel of the additional line segments. This was compared with a condition that had normal onset. The results of this experiment, as well as those of Krumhansl (1982), showed a definite advantage in detection speed for stimuli presented with abrupt onsets• Using this no-onset procedure, Yantis and Jonides (1984) demonstrated that the effect of the number of letters in a circular display on RT in a visual-search task was greatly reduced when the target was presented with an abrupt onset, compared to a no-onset presentation. It is probable the attention can be focussed directly on the target when it is displayed with abrupt onset instead of there being a serial search through the display when the target is presented without onset. There are also results that demonstrate the ability of subjects to prevent their attention from being allocated involuntarily to a stimulus (e.g., Folk, Remington, & Johnston, 1992; Warner, Juola, & Koshino, 1990; Yantis & Jonides, 1990), but this ability is contingent on very specific conditions (e. g., a high degree of practice or the similarity of properties of cue and stimulus) not met in our experiments.
89 Applied to processing a level of a compound letter, the hypothesis follows that in principle an abrupt onset should be able to divert attention from the local level, even if attention at the local level was prepared a good deal before the stimulus onset. Thus, valid level-specific cueing is bound to be somewhat unsuccessful in such a situation. It seems that only when the onset is reduced, or even eliminated, is there a better chance of successfully testing the hypothesis that the global-local RT difference reflects the time needed to refocus attention from the global to the local level.
The effects of stimulus consistency. While Navon's (1977) data show no interference from global features, our experiment produced interference from global and local features. There is a main effect of Stimulus Consistency (consistent: 497 ms, inconsistent: 605 ms, F(1,7) = 22.6, p <.01) and a significant interaction of the Level to be reported and Stimulus Consistency, F(1,7) = 10.1, p <.05. The interference is 39 ms higher with local than with global report. Regardless of cue validity and level to be reported, subjects obviously cannot disregard the features at the other levels. Even with global report following a valid cue, there is substantial interference from local features (see Figure 4). This can be interpreted quite easily without the assumption that the local level was focussed. Local elements constitute the texture of the global object. Focussing the global level is therefore somewhat like the right setting of the focus of attention to discriminate between textures (Kimchi & Palmer, 1982). The letters F and H form discriminable textures: with Fs, the texture is dominated by horizontal lines, and with Hs, by vertical elements. The detection of these local features can interfere with global reports, even without focussing at the local level. The interaction of Level to be reported, Stimulus Consistency, and Cue Validity, F(2,14) = 6.1, p <.05, revealed that the higher interference observed in the case of local reports mainly originated from conditions with invalid cues (see Figure 4). Brief intentional focussing of the global level obviously leads to recognition of the global letter, and for that reason strong interference results after attentional zooming to the local level. Local recognitions that are preceded by an unintentional focussing of the global level due to an abrupt onset lead to a somewhat smaller amount of interference. This difference suggests functional consequences of the intention of attending to the irrelevant level. One difference between intentionally and unintentionally initiated attentional zooming could be in the speed of attention disengagement (see Posner, Walker, Friedrich, & Rafal, 1984), but this issue deserves further investigation.
Experiment 2 The aim of Experiment 2 was to examine further the hypothesis that the global-dominance phenomenon is at least partly due to an unintentional focussing of attention at the global level of a stimulus object caused by an abrupt visual onset. For this purpose we replicated Experiment 1 using a
730-
global, valid cue
-~ local, valid cue
global, neutral cue
-~- local, neutral cue
~'- global, invalid cue
-v- local, invalid cue
680
630 "
E
m
Fn" 5 8 0 -
530 ~ -
480 -
430
I consistent Stimulus
I inconsistent Consistency
Fig. 4. Mean reaction time (RT) as a function of stimulus consistency, level to be reported (global, local), and cue validity in Experiment 1
variant of Todd and Van Gelder's (1979) no-onset procedure. Elimination of the onset should prevent unintentional focussing of the global level. As a result, the efficiency of level-specific cueing was expected to be greater in Experiment 2 than in Experiment 1, and therefore the globallocal RT difference should approach zero.
Method Because Experiment 2 replicated Experiment 1 in most respects, we shall mention only the differences here.
Subjects. Eleven paid subjects participated in this experiment. Their age ranged from 19 to 28 years. All were right-handed and had normal or corrected-to-normal vision.
Stimuli. The elimination of abrupt visual onsets was achieved by presentation of a superposition stimulus together with the cue. This stimulus (for an example with global cue see Figure 5) was constructed by a superposition of all line segments of the stimulus alternatives at both levels, forming an A-like figure as local elements and global object. The imperative stimulus was presented by the removal of the line segments not needed to form a stimulus. For presentation of local elements, five pixels were removed for each of the 24 local elements. For presentation of the global level, five local elements of the 24 pixels were removed. The global and local aspects of the superposition stimulus change at the same time. As each aspect changes by the removal
90 i:::::i i:::::ii:::::i !:::::i i:::::i i:::::i !:::::i i:::::i i:::::i
U::::i
i:::::i
Fig. 5. Superposition stimulus to eliminate abrupt visual onsets together with a frame serving as a level-specific cue as used in Experiment 2 (example of a global cue)
of 120 pixels, there is no level-specific bias in this manipulation. The frame that formed the cue was presented simultaneously with the superposition stimulus. In order to exclude unnecessary offsets before the presentation of the imperative stimulus, the cue remained on the screen together with the imperative stimulus.
Procedure. The events in a single trial did not differ from Experiment 1, except that the interval between cue and imperative stimulus was changed from 40 ms to 0 ms to prevent an onset with the exposition of the imperative stimulus.
Results and discussion Mean RTs per condition and subject were calculated and entered into a five-way ANOVA for repeated measurements on all factors. Again, RTs of less than 100 ms and more than 2,000 ms were considered errors (<0.5%). The overall mean error rate was 2.7%. The error analysis again mirrored the results based on RT data.
The efficiency of level-specific cues. The main effect of Cue Validity was significant, F(2,20) = 37.6, p <.001. The mean RT for valid cues was 620 ms, for invalid cues 731 ms, and for neutral cues 659 ms. So mean benefits for valid cues amounted to 39 ms, mean costs for invalid cues amounted to 72 ms (each difference is significant: Scheff6test, p <.01). In comparison with Experiment 1, cue efficiency was much higher, as was revealed by an ANOVA with Experiment 1 vs. Experiment 2 as an additional factor. (There was an interaction between Experiment and Cue Validity, F(2,18) = 5.47, p <.05.) Moreover - and most important for the evaluation of our central hypothesis there were substantial benefits for valid cues, suggesting that the setting of the attentional zoom initialiated after cue presentation was no longer counteracted by focussing the global level first, regardless of the level to be reported. Compared with Experiment 1, cue efficiency was not modified much by the SOA variation. There was an SOA2 main effect, F(1,10) = 93.7, p <.001 (see Figure 7), revealing a decrease in mean RT when SOA 2 was longer. This latter effect is probably partly due to nonspecific alerting (e. g., Colgate, Hoffman, & Eriksen, 1973; Posner & Boies, 1971), but it must have something to do with level-specific attentional focussing as well. This time interval can be optimally used for preparing and/or executing attentional zooming, because at its start the subject knows the validity
of the cue. The time available during the SOA2 interval is probably used to execute the level-specific refocussing of attention prepared during the SOA1 interval. The fact that there was an SOA l effect in Experiment 1, but none in Experiment 2, can be interpreted along this line. Preparing attentional refocussing during the SOA~ carries the risk of preparing the focussing of the wrong level in 20% of the trials in a block with informative cues. It might have been a strategy on the part of the subjects to postpone any preparatory refocussing until the instruction tone revealed the validity of the cue. This should not produce any SOA1 effect, but only an SOA 2 effect. However, in Experiment 1 there was an SOA 1 effect. Obviously the subjects here must have used the SOA~ interval for preparing to refocus attention. They probably did this in order to use all the time they had to try to prevent the onset-induced focussing of the global level that counteracted their setting of the focus when they intended to focus the local level. As there was no such onset in Experiment 2, subjects could successfully employ the strategy of waiting until the instruction tone had revealed the validity of the cue, and act accordingly.
The effects of cues on the global-local R T difference. The difference between the mean RT for global (658 ms) and local reports (682 ms) was not significant, F(1,10) = .93, p >.25. A direct comparison of Experiments 1 and 2 revealed a significant interaction of the factors Experiment (1 or 2) and Level to be reported, F(1,18) = 5.64, p <.05 (see Figure 6). There was a large global-local RT difference in Experiment 1, but the no-onset procedure obviously prevented a global-dominance effect in Experiment 2. This was probably achieved by the successful prevention of an unintentional focussing of the global level. If this really occurred, valid as well as invalid cues should have been of equal efficiency in directing attention at both global and local levels. The interaction of the factors Level to be reported and Cue Validity was far from significant, F(2,20) = 1.0, p >.25, and there was no significant higherorder interaction involving these factors. The comparison of Experiments 1 and 2 revealed a significant shorter mean RT in Experiment 1, F(1,18) = 9.31, p <.01 (see Figure 6). This replicates Todd and Van Gelder's (1979) finding that RT is shorter in conditions with normal onsets than in conditions with a no-onset procedure. This effect is presumbly due to the difficulty of detecting the beginning of stimulus presentation when the superposition stimulus precedes the imperative stimulus. Why a nonsignificant global-local RT difference of 24 ms is left can be understood if we look at the significant interaction of the factors SOA z and Level to be reported, F(1,10) = 16.9, p <.01 (see Figure 7). There is a globallocal RT difference of 42 ms when SOA 2 is 100 ms, but of only 6 ms when SOA2 is 700 ms. The SOA2 can be used to set the focus for a report of the level designated by the instruction tone, regardless of the cue presented. When SOA 2 is 100 ms, this interval may be too short to accomplish adequate focussing, but When SOA 2 is 700 ms, there is plenty of time. So the latter condition is the one most adequate for preparation and/or execution of attentional zooming at the level to be reported, and therefore the mean
91 1
4- Experiment 1 4- Experiment 2J
Level to be Reported 800 -
700
0
+ global 4- local
750
•
650
~, 700 @ E iv-
600
650-
550 600
500
I
!
global
local
Level to be Reported
550
1 O0
700 SOA2 (ms)
Fig. 6. Mean reaction time (RT) as a function of level to be reported in Experiments 1 and 2
Fig. 7. Mean reaction time (RT) as a function of SOA2 and level to be reported in Experiment 2
RT as well as the global-local RT difference should be minimal in this condition. However, the global-local difference when SOA 2 is 100 ms can only mean that attentional refocussing starts with a global setting. It is reasonable to assume that the superposition stimulus that eliminates the onset of the imperative stimulus produces an onset itself when it is presented. This onset automatically allocates attention to the global level of this stimulus. As we have already explained, refocussing attention does not start until the instruction tone has sounded• When the SOA2 is only 100 ms, this is not enough to complete adequate focussing of the local level before the imperative stimulus is presented. That is the reason why at least a small advantage for global processing should be observed with a short SOA 2.
attention can be effectively allocated at the level to be reported before stimulus presentation. Moreover, interference' should be less in Experiment 2 than in Experiment 1. The significant interaction of the factors Experiment (1 or 2) and Stimulus Consistency is consistent with this hypothesis, F(1,18) = 6.37, p <.05 (see Figure 8). Interference should be less in conditions of adequate level-specific focussing for reporting a level (following valid cues), but should increase when attention is temporarily focussed at the level not to be reported (following invalid cues). The significant interaction of the factors Stimulus Consistency and Cue Validity, F(2,20) = 5.61, p <.05 (see Figure 9), supports this expectation. Taken together, there are some results of Experiment 2 that are different from those of Experiment 1: the symmetrical-interference pattern, a smaller amount of interference, and a reduced interference in conditions with valid global and local cues. These are more evidence of the successful experimental manipulation of level-specific attentional focussing. Experiments 1 and 2 reveal that the time needed to refocus attention from the global to the local level is one of the major factors in the occurrence of a global-dominance phenomenon.
The effects of stimulus consistency. The main effect of Stimulus Consistency was significant, F(1,10) = 18.4, p <.01, revealing a mean RT of 643 ms with consistent stimuli, but 696 ms with inconsistent stimuli. Contrary to Experiment 1, there was no significant interaction of the factors Level to be reported and Stimulus Consistency, F(1,10) = 1.92, p >. 10, i. e., interference was symmetrical in Experiment 2. This result has to be expected whenever
92 730
l
÷ Experiment 1 * - Experiment 2 I
Cue Validity -.- valid -~ neutral •
invalid
790680
740 630
%-
g
E
v
t-re
690 580 -
s.s
640
530 -
t
J 480
1
consistent
T
inconsistent
Stimulus Consistency
590 consistent
inconsistent
Stimulus Consistency
Fig. 8. Mean reaction time (RT) as a function of stimulus consistency for Experiments 1 and 2
Fig. 9. Mean reaction time (RT) as a function of stimulus consistency and cue validity in Experiment 2
Experiment 3
(1985) replicated Pomerantz's result (see also Boer & Keuss, 1982)• In these stimuli, the form of the local elements constitutes the form at the global level. In the compound stimuli, however, local features are structurally irrelevant to the global form. It cannot be ruled out that this structural relationship may be the reason why processing of the global level does not require processing of local features (Wandmacher & Arend, 1985)• For this reason, Pomerantz (1983) believes that one should not use compound stimuli at all in the context of experiments on top-down and bottom-up processing• This raises the question of whether the results of our experiments can be generalized to the type of stimulus material used by Pomerantz (1983) and Wandmacher and Arend (1985). A replication of these results seems to be possible only if the strong organizational effects that form perceptual objects at different levels in its structural hierarchy can be overcome by attentional zooming so that the component parts of the objects may be accessed selectively at subordinate levels. There is an additional argument that suggests a replication of Experiment 2 with the stimulus material of Pomerantz (1983). One property of compound stimuli is the great difference in size between the global and local levels, usually ranging from about 5:1 to 10:1. Pomerantz (1983) argues that in some studies the global form may have been more easily discriminable than the form of
If level-specific attentional focussing is indeed the critical factor in the formation of the global-dominance phenomenon, selection of the target level should be strongly determined by configurational properties such as are described by Gestalt laws of perceptual organization (Wertheimer, 1923). Neisser (1967) describes configurational properties as the outcome of preattentive processing, defining perceptual objects as possible candidates for selection. Similar positions are held by those who favor an object-based view of attentional selection (e.g., Duncan, 1984; Kahneman & Henik, 1981; Yantis, 1992)• Martin (1979) found global dominance when the local elements were closely packed to form a coherent global object, whereas local dominance was found when the local elements were so far apart that each one could be seen as a separate local object. Pomerantz (1983) demonstrated that the processing of a global feature of triangles or arrows is faster than the processing of a local feature of the component line segments, although the local features had about the same spatial extent as the global features. He attributed this result to the Gestalt property of closure that makes the global figure perceptually more salient than the local level. In several variations of the task, Wandmacher and Arend
93 Orientation of global form left right
a
local
-"- SOA2=100ms valid cue -SOA2=700rns valid cue
Jef'
"\,\
\,\
'\\\\\,
+ SOA2=100ms invalid cue
/ / / /
-+ SOA2 = 700 100
,_==-=
g S
Fig. 10. Imperative stimuli used in Experiment 3 together with their response assignment (left vs. fight) for global and local analysis
~- 50 m 'o c..-.~'~
the local elements because of their size. As a consequence, it m a y just be harder to ignore the global form than the local elements. The fact that absolute size m a y indeed be a critical variable for the formation of global dominance was demonstrated b y Kinchla and Wolfe (1979). The triangles and arrows used as stimuli by Pomerantz (1983) have the interesting property that their global and local features have about the same spatial extent so that discriminability differences can be excluded. Nevertheless, only global dominance has been observed with this kind o f stimulus material (Pomerantz, 1983; W a n d m a c h e r & Arend, 1985). Experiment 3 was conducted to test the generalizability o f the results of Experiment 1 with the stimulus material of Pomerantz (1983) without the elimination of stimulus onset. Experiment 4 again uses a no-onset procedure, but this time with the simple geometrical forms used b y Pomerantz.
O
/-
..........
+
0
-50
1 100
700 SOA1 (ms)
Fig. 11. Mean cost and benefits of cueing as a function of cue validity, SOA1, and SOA2
Procedure. The procedure was essentially the same as in Experiment 1. Method
The mapping of stimulus orientation and response side was compatible. The subjects responded by pressing the left key whenever a leftoriented stimulus was presented and the right key to a fight-oriented stimulus.
In many respects Experiment 3 is a replication of Experiment 1, so we shall mention only methodological differences here.
Results and discussion Subjects. Eight students were paid to participate in four sessions run on four consecutive days. Their age ranged from 19 to 23 years. All subjects were right-handed and had normal or corrected-to-normal vision.
Stimuli. The set of imperative stimuli is presented in Figure 10. The size of the stimuli was 1.8° of visual angle horizontally and vertically. The subjects' task was to classify the orientation of the diagonal line segment as left or right when the instruction tone signaled local analysis and to classify the orientation of the stimulus form as left or right when the instruction tone signaled global analysis (for the response assignment see Figure 10). The size cues used in Experiments 1 and 2 could not be used in this experiment. Instead, we used symbolic cues: a display of two differently oriented triangles and arrows within a frame of the size of an imperative stimulus, instructing the subject to attend to the global level; and a display of two differently oriented diagonal lines within a dotted frame, signaling the subject to attend to the local level. As a neutral cue we used the dotted frame with a small dotted circle in its center.
Means for correct RTs were entered into a five-way A N O V A for repeated measures on all factors. Again, RTs of less than 100 ms and of more than 2;000 ms were considered to be errors (<0.5%). The overall mean error rate was 2.4%. The results of the A N O V A on errors mirrored those of the RT data.
The efficiency of level-specific cues. The main effect of Cue Validity was significant, F(1,7) = 59.8, p <.001. Scheff~ tests revealed that all paired comparisons were significant with at l e a s t p <.05: There were significant benefits (26 ms) of valid cues (mean: 496 ms) and costs (43 ms) produced by invalid cues (mean: 565 ms). The modulation of costs and benefits by S O A are revealed by the significant interaction o f cue validity, SOA~, and SOA2, F(2,14) = 26.1, p <.001 (see Figure 11). The figure shows that costs and benefits are larger when SOA~ is 700 ms than when it is
94 Cue Validity valid - -
neutral
invalid
~
61o
X
Fig. 13. Superposition stimulus used in Experiment 3 (on the left) and example of an imperative stimulus with irrelevant line segments dotted (on the right)
~,560E t-
/
in the case of simple geometric stimuli with no size differences between global and local features of form. As in Experiment 1, the cues were apparently not effective enough to reduce it to almost zero.
/ / / / / / /
510
460
I
I
global
local
Level to be Reported Fig. 12. Mean reaction time (RT) as a function of cue validity and level to be reported
100 ms, but only when SOA2 = 100 ms. If SOA 2 is 700 ms, there is enough time to refocus attention with reference to the instruction tone, thus considerably reducing the effects of cueing. In addition, the main effects of SOA1, F(1,7) = 23.3, p <.01, and of SOA2, F(1,7) = 103,8, p <.001, are significant, showing that the interval between cue and instruction tone, as well as the interval between instruction tone and imperative stimulus, can be used to prepare levelspecific attentional focussing. This replicates the equivalent interaction of Experiment 1 (see Figure 2).
The effects of stimulus consistency. The main effect of Stimulus Consistency was significant, F(1,7) = 14.3, p <.01, showing a mean interference effect of 87 ms (consistent: 484 ms; inconsistent: 571 ms). The asymmetric-interference pattem observed in Experiment 1 was replicated here: there is interference from local features when the global level is reported (66 ms), but a great deal more interference from global features when the local level is reported (107 ms). On the whole, these data replicate the results of Experiment 1. They demonstrate level-specific cueing effects that are sufficient to reduce the global-local RT difference substantially, but not effective enough to reduce it to about zero. The only difference from Experiment 1 is that the valid cues of Experiment 3 did not effectively reduce the interference in inconsistent-stimulus conditions. Experiment 4 replicates Experiment 2 in most respects, using the same stimuli as in Experiment 3. Its purpose was to test the hypothesis that the global-local RT difference would be reduced to about zero if the manipulation of attentional focussing were made more efficient. This was again done by a reduction of the abrupt visual onset produced by the imperative stimulus.
Experiment 4
Method The effects of cues on the global-local RT difference. The main effect of the Level to be reported is significant (global: 494 ms, local: 561 ms, F(1,7) = 82,3, p <.001). This replicates the equivalent effect found in Experiment 1. There is also a significant interaction between the factors Level to be reported and Cue Validity, F(2,14) = 4.17, p <.05 (see Figure 12). It reveals that the global-local RT difference is 33 ms less in conditions with a valid cue than in conditions with invalid cues. In addition, when conditions with the same number of steps for the refocussing of attention are compared (global report preceded by an invalid cue: 523 ms; local report preceded by an valid cue: 521 ms), the global-local RT difference is almost zero. Altogether, these results confirm that level-specific cueing is effective in reducing the global-local RT difference, even
Because Experiment 4 was largely a replication of Experiment 2 (but with different stimuli) and of Experiment 3 (but with a no-onset procedure), we shall mention only the differences here.
Subjects. Seven students were paid to participate in four sessions run on four consecutive days. They were aged between 19 and 34 years. All subjects were right-handed and had normal or corrected-to-normal vision.
Stimuli. The elimination of abrupt visual onsets was again achieved by presentation of a superposition stimulus prior to the imperative stimulus (see left part of Figure 13). The imperative stimulus (an example is presented on the right of Figure 13) was then presented by the removal of every second pixel in the additional line segments. Subjects were instructed to attend to the continuous line segments as a figure.
95
Stimulus Consistency ÷ consistent -*- inconsistent
720
Cue Validity
690
+ valid ~" neutral -~- invalid
640
670
~620
~590
I'-" n"
a-
J s, s,
570
f F
i,
540
520
470
[
I
i
100
300
700
SOA
(ms)
490t
J
consistent
inconsistent
Stimulus Consistency
Fig. 14. Mean reaction time (RT) as a function of stimulus consistency and SOA in Experiment 3
Fig. 15. Mean reaction time (RT) as a function of stimulus consistency and cue validity in Experiment 3
We did not remove all the pixels because we wanted to minimize the offset effect• We used verbal cues: the German word forfigure (Figur) instructing the subject to attend to the global level, the German word for line (Linie) telling the subject to attend to the local level, and XXXXX as a neutral cue. Cues were presented 2.6° below the center of the superposition stimulus•
Scheff6 tests). Costs and benefits were about 20 ms higher here than in Experiment 3 without the no-onset procedure.
Procedure• The verbal cues were presented t,500 ms before the imperative stimulus. The duration of the stimulus was 500 ms. With the offset of the cue the superposition stimulus was exposed for 1,000 ms, followed by the imperative stimulus, which was presented for 100 ms• With an SOA of 100 ms, 300 ms, or 700 ms before the onset of the imperative stimulus, the instruction tone sounded for 100 ms. Further procedural details were the same as in the previous experiments.
Results and discussion Mean correct RTs were entered into a four-way ANOVA for repeated measurements on all factors. The mean error rate was 3.4%. The error analysis mirrored the results based on RT data.
The efficiency of level-specific cues. The main effect of Cue Validity was significant, F(2,6) = 67.6, p <.001. The mean RT for neutral cues was 557 ms, and there was a significant benefit for valid cues (46 ms) and a cost for invalid cues (65 ms) (differences significant, with p <.01 according to
The effects of cues on the global-local RT difference. As in Experiment 2, the global-local RT difference was about zero (global: 561 ms; local: 565, F(1,6)
The effects of stimulus consistency. The main effect of Stimulus Consistency was small, but significant, F(1,6) -- 6.33, p <.05, with a mean RT of 543 ms for consistent stimuli and of 584 ms for inconsistent stimuli• The interaction of the factors Level to be reported and Stimulus Consistency was far from significant. This means that the interference was symmetrical with respect to levels. The main effect of S O A w a s also significant, F(2,6) = 250.2, p <.001, with a mean RT of 655 ms at SOA = 100 ms, 546 ms at SOA = 300 ms, and 489 ms at SOA = 700 ms. The fact that this latter effect was in part due to a progressively improved focussing of attention with longer SOAs is documented by the SOA x Consistency ineraction, F(2,6) = 25.4, p <.01, showing a reduction in the mean RTs with longer SOAs that is larger for inconsistent than for consistent stimuli (see Figure 14).
96 If the manipulation of attentional zooming by levelspecific cueing was effective, the factor Stimulus Consistency should interact with the factor Cue Validity, showing no effect of stimulus consistency with valid cues, but a strong effect with invalid cues. As can be seen from Figure 15, this significant interaction shows exactly that kind of pattern, F(2,6) = 23.7, p <.01. This result is consistent with the equivalent interaction of these factors in Experiment 2 (see Figure 9). Altogether, the results of Experiment 4 replicate those of Experiment 2 in almost every respect, i.e., cue efficiency was high, there was no global-dominance effect, and the interference pattern was symmetrical with respect to the level to be reported. This was probably due to adequate level-specific attentional focussing following valid cues, made possible by the elimination of the abrupt stimulus onsets that otherwise would have unintentionally attracted attention at the global level with local reports.
General discussion The major purpose of the present study was to test the hypothesis that the global-dominance phenomenon evidenced by a substantial global-local RT difference and by an asymmetrical-interference pattern is caused by an unintentional focussing of attention at the global level, even in those cases in which a local report is intended. In Experiment 1 we tried to demonstrate that effective level-specific cueing eliminates the global-dominance phenomenon. The data showed that neither the global-local RT difference nor the asymmetrical-interference pattern disappeared in conditions with valid cues. It was demonstrated, however, that the RT difference was about zero between conditions with an equal number of hypothetical steps to be taken in the course of refocussing attention at the level to be reported. The data also revealed that in principle level-specific cueing might be an effective means of manipulating attentional focussing if the effectiveness of valid local cues had not been reduced in Experiment 1 by some factor we failed to control. This probably led m the capturing of attention by the global level against the subject's intentions, counteracting the preparation and/or execution of a local setting of the focus. As a likely candidate for unintentional capture, the abrupt visual onset produced by the imperative stimulus was eliminated in Experiment 2. This modification led, first, to a significant improvement in the effectiveness of level-specific cueing, and second, to a reduced global-local RT difference to nearly zero in conditions with optimal properties for the most efficient cueing; and third, there was a nullification of the asymmetry of interference. Experiment 4, which employed different stimulus materials and different cues, replicated the results of Experiment 2 in almost every respect. It was concluded that the globaldominance phenomenon is produced by the functional property of visual attention to be drawn first to the global level by abrupt visual onsets before other levels of the structural hierarchy can be focussed. Kinchla and Wolfe (1979), using a wide range of stimulus sizes, predicted that the level of a compound letter subtending about 2 ° of visual angle should have an optimal
size that leads to its privileged processing, with subsequent processing of both lower and higher levels. The global level of the compound stimuli used in Experiment 2 subtended about 2 ° of visual angle, but nevertheless a global-dominance phenomenon could be prevented by level-specific cueing. Therefore, in most situations, the extent of a level in the visual field alone does not determine whether a global-dominance phenomenon is observed. This conclusion is corroborated by the results of our Experiment 4. Here the spatial extent of global and local features was about the same, and again subtended about 2 ° of visual angle. Using the same stimulus material, Pomerantz (1983) as well as Wandmacher and Arend (1985) always produced a global-dominance phenomenon. But when the abrupt visual onset was eliminated in our Experiment 4, thus ensuring adequate level-specific attentional focussing, the global-dominance effect observed in Experiment 3 was no longer present. The results of our experiments establish the abrupt visual onset as a likely and very powerful factor that is even able temporarily to override the subject's intention of attending to a certain level. Yantis and Jonides (1990) question their earlier conclusion (Yantis & Jonides, 1984) that abrupt onsets always capture attention automatically. They demonstrated that highly focussed attention, induced by completely valid cues, was resistant to attentional capture by abrupt onsets. However, with cues of lower validity and a more diffuse attention allocation, abrupt visual onsets did produce a strong tendency to divert attention. So if the subject has a strong intention of attending to a certain object and if other stimulation within the visual field is of no importance to the task (100%-validity condition), the power of abrupt onsets to attract attention is minimal. If we generalize their results and apply them to our experiments, the 80% validity of the cues in our experiments probably produced an equivalent to what Yantis and Jonides (1990) called a more diffuse attention allocation. It therefore seems legitimate in our case to speak of an automatic capture of attention by the global level. Moreover, there are data from Remington et al. (1992) showing that even the intention of ignoring objects exposed with abrupt onset does not always prevent attention from being involuntarily drawn to that object. Kinchla et al. (1983) as well as Ward (1985) were able to reduce global dominance even in conditions without elimination of the abrupt visual onset. How is this possible? Does this not contradict our results? The literature on the ability of abrupt onset to draw attention unintentionally to an object shows that this is not a strongly automatic process in the sense that there is no way of preventing unintentional capture of attention by onsets (Folk et al., 1992; Theeuwes, 1991; Warner et al., 1990; Yantis & Jonides, 1990). With some practice (Warner et al., 1990), and especially when the onset is outside of the focus (Theeuwes, 1991), attentional capture can be prevented. However, the visual system is usually prepared to give priority to onsets when the focus of attention is set relatively wide (Yantis & Jonides, 1990). Priority can be set quite low when other task demands require different actions that have high priority (Yantis & Jonides, 1990). This setting of priorities can be achieved without a reduction of the onsets, but onset reduction
97 probably makes it quite easy to give priority to the level the subject intends to focus.
Conclusions It has been shown that level-specific cueing makes the global-dominance phenomenon disappear only b y the elimination of the abrupt visual onset that accompanies the presentation of the imperative stimulus. The global-local RT difference becomes zero and interference from features of the irrelevant level is substantially reduced when there is enough time for the subject to preset the focus of attention before the stimulus appears. We have also shown that this conclusion not only holds for compound stimuli, but is generalizable to other kinds of hierarchically formed stimuli, e. g., simple geometric forms. Thus, one of the factors that contributes most powerfully to the formation of the global-dominance phenomenon is the abrupt visual onset that unintentionally captures attention and allocates it to the global level first. There m a y be many other factors that draw attention in a level-specific manner. Whatever they m a y be, the functional principle is probably always the same: attention is captured by a certain level first (usually the global). It is concluded that the longer RT for local reports that is usually observed is produced by the time needed to refocus visual attention from an unintentional focussing at the global level to an intentional focussing at the local level. On the basis o f the present experiments one can conclude that the global-dominance phenomenon is essentially an attentional phenomenon.
References Antes, J. R., & Mann, S. W. (1984). Global-local precedence in picture processing. Psychological Research, 46, 247-259. Boer, L. C., & Keuss, R J. G. (1982). Global precedence as a postperceptual effect: An analysis of speed-accuracy tradeoff functions. Perception & Psychophysics, 31,358-366. Breitmeyer, B. G. (1975). Simple reaction time as a measure of the temporal response properties of transient and sustained channels. Vision Research, 15, 1411-1412. Broadbent, D. E. (1977). The hidden preattentive processes. American Psychologist, 32, 109-118. Castiello, U., & Umilt?L C. (1990). Size of the attentional focus and efficiency of processing. Acta Psychologica, 73, 195-209. Colgate, R., Hoffman, J. E., & Eriksen, C. W. (1973). Selective encoding from multi-element visual displays. Perception & Psychophysics, 14, 217-224. Duncan, J. (1984). Selective attention and the organization of visual information. Journal of Experimental Psychology: General, 113, 501-517. Eriksen, C. W., & Rohrbaugh, J. W. (1970). Some factors determining efficiency of selective attention. American Journal of Psychology, 83, 330-342. Eriksen, C. W., & St James, J. D. (1986). Visual attention within and around the field of focal attention: A zoom lens model. Perception & Psychophysics, 40, 225-240. Eriksen, C. W., & Webb, J. M. (1989). Shifting of attentional focus within and about a visual display. Perception & Psychophysics, 45, 175-183. Eriksen, C. W., & Yeh, Y. Y. (1985). Allocation of attention in the visual field. Journal of Experimental Psychology: Human Perception and Performance, 11, 583-597.
Folk, C. L., Remington, R. W., & Johnston, R. W. (1992). Involuntary covert orienting is contingent on attentional control settings. Journal of Experimental Psychology: Human Perception and Performance, 18, 1030-1044. Fukoda, Y., & Stone, J. (1974). Retinal distribution and central projection of Y-, X-, and W-cells of the cat's retina. Journal of Neurophysiology, 37, 749-772. Greaney, J., & MacRae, A. W. (1992). The order of visual processing: Top-down, bottom-up, middle-out, or none of these? Bulletin of the Psychonomic Society, 30, 255-257. Grice, G. R., Canham, L., & Boroughs, J. M. (1983). Forest before trees? It depends where you look. Perception & Psychophysics, 33, 121 - 128. Grindley, C. G., & Townsend, V. (1968). Voluntary attention in peripheral vision and its effects on acuity and differential thresholds. Quarterly Journal of Experimental Psychology, 20, 11-19. Hoffman, J. E. (1980). Interaction between global and local levels of a form. Journal of Experimental Psychology: Human Perception and Performance, 6, 222-234. Huges, H. C., Fendrich, R., & Reuter-Lorenz, R A. (1990). Global versus local processing in the absence of low spatial frequencies. Journal of Cognitive Neuroscience, 2, 272-282. Hughes, H. C., Layton, W. M., Baird, J. C., & Lester, L. S. (1984). Global precedence in visual pattern recognition. Perception & Psychophysics, 35, 361-371. Jonides, J. (1981). Voluntary vs. automatic control over the mind's eye's movement. In J. B. Long & A. D. Baddeley (Eds.), Attention and performance IX (pp. 187-203). Hillsdale, NJ: Erlbanm. Kahneman, D., & Henik, A. (1981). Perceptual organization and attention. In M. Kubovy & J. R. Pomerantz (Eds.), Perceptual organization (pp. 181-211). Hillsdale, NJ: Erlbaum. Kimchi, R. (1992). Primacy of wholistic processing and global/local paradigma: A critical review. Psychological Bulletin, 112, 24-38. Kimchi, R. & Palmer, S. E. (1982). Form and texture in hierarchically constructed patterns. Journal of Experimental Psychology: Human Perception and Performance, 8, 521-535. Kinchla, R. A., & Wolfe, J. M. (1979). The order of visual processing. "Top-down," "bottom-up," or "middle-out." Perception & Psychophysics, 25, 225-231. Kinchla, R. A., Solis-Macias, V., & Hoffman, J. (1983). Attending to different levels of structure in a visual image. Perception & Psychophysics, 33, 1-10. Krumhansl, C. L. (1982). Abrupt changes in visual stimulation enhance processing of form and location information. Perception & Psychophysics, 32, 511-523. Lamb, M. R., & Robertson, L. C. (1988). The processing of hierarchical stimuli: Effects of retinal locus, locational uncertainty, and stimulus identity. Perception & Psychophysics, 44, 172-181. Lamb, M. R., & Robertson, L. C. (1989). Do response time advantage and interference reflect the order of processing of globaland local-level information? Perception and Psychophysics, 46, 254-258. Lamb, M. R., & Robertson, L. C. (1990). The effect of visual angle on global and local reaction time depends on the set of visual angles presented. Perception & Psychophysics, 47, 489-496. Lovegrove, W. J., Lehmkuhle, S., Baro, J. A., & Garzia, R. (1991). The effects of uniform field flicker and blurring on the global precedence effect. Bulletin of the Psychonomic Society, 29, 289-291. Luna, D., Merino, J. M., & Marcos-Ruiz, R. (1990). Processing dominance of global and local information in visual patterns. Acta Psychologica, 73, 131 - 143. Martin, M. (1979). Local and global processing: The role of sparsity. Memory & Cognition, 7, 476-484. Maylor, E. A. (1985). Facilitatory and inhibitory components of orienting in visual space. In M. I. Posner & O. S. M. Matin (Eds.), Attention and perjbrmance XI (pp. 189-204). Hillsdale, NJ: Erlbaum. Maylor, E. A., & Hockey, R. (1985). Inhibitory components of externally controlled covert orienting in visual space. Journal of Experimental Psychology: Human Perception and Performance, 11,777-787.
98 Mertens, J. J. (1956). Influence of knowledge of target location upon the probability of observations of peripherally observable test flashes. Journal of the Optical Society of America, 46, 1069-1070. Miller, J. (1981). Global precedence in attention and decision. Journal
of Experimental Psychology: Human Perception and Performance, 7, 1161-1174. Miller, J. (1989). The control of attention by abrupt visual onsets and offsets. Perception & Psychophysics, 45, 567-571. Navon, D. (1977). Forest before trees: The precedence of global features in visual perception. Cognitive Psychology, 9, 353-383. Navon, D. (1981). The forest revisited: More on global precedence. Psychological Research, 43, 1-32. Navon, D. (1983). How many trees does it take to make a forest? Perception, 12, 239-254. Navon, D. (1991). Testing a queue hypothesis for the processing of global and local information. Journal of Experimental Psychology: General, 120, 173-189. Navon, D., & Norman, J. (1983). Does global precedence really depend on visual angle? Journal of Experimental Psychology: Human Perception and Performance, 9, 955-965. Neisser, U. (1967). Cognitive psychology. New York: Appleton-Century-Crofts. Paquet, L. (1992). Global and local processing in nonattended objects: A failure to induce local processing dominance. Journal of Ex-
perimental Psychology: Human Perception and Performance, 18, 512-529. Paquet, L., & Merikle, E M. (1984). Global precedence: The effect of exposure duration. Canadian Journal of Psychology, 38, 45-53. Pomerantz, J. R. (1983). Global and local precedence: Selective attention in form and motion perception. Journal of Experimental Psychology: General, 112, 516-540. Posner, M. I. (1980). Orienting of attention. Quarterly Journal of Experimental Psychology, 32, 3-25. Posner, M. I., & Boles, S. (1971). Components of attention. Psychological Review, 78, 391-408. Posner, M. I., Nissen, M. J., & Ogden, W. C. (1978). Attended and unattended processing modes: The role of set for spatial location. In H. L. Pick & E. Saltzman (Eds.), Modes of perceiving and processing information (pp. 137-157). Hillsdale, NJ: Erlbaum. Posner, M. I., & Snyder, C. R. R. (1975). Facilitation and inhibition in the processing of signals. In P. M. A. Rabbitt & S. Dornid (Eds.), Attention and performance V (pp. 669-682). London: Academic Press. Posner, M. I., Snyder, C. R. R., & Davidson, B. J. (1980). Attention and the detection of signals. Journal of Experimental Psychology: General, 109, 160-174. Posner, M. I., Walker, J. A., Friedrich, E J., & Rafal, R. D. (1984). Effects of parietal injury on covert orienting of attention. Journal of Neuroscience, 4, 1863-1874. Remington, R. W., Johnston, J. C., & Yantis, S. (1992). Involuntary attentional capture by abrupt onsets. Perception & Psychophysics, 51, 279-290. Remington, R. W., & Pierce, L. (1984). Moving attention: Evidence for time-invariant shifts of visual selective attention. Perception & Psychophysics, 35, 393-399. Robertson, L. C., & Lamb, M. R. (1991). Neuropsychological contributions to theories of part/whole organisation. Cognitive Psychology, 23, 299-330.
Robertson, L. C., Lamb, M. R., & Knight, R. T. (1988). Effects of lesions of temporal-parietal junction on perceptual and attentional processing in humans. Journal of Neuroscience, 8, 3757-3769. Rock, I., & Gutman, D. (1981). The effect of inattention on form perception. Journal of Experimental Psychology: Human Perception and Performance, 7, 275-285. Shiffrin, R. M., & Gardner, G. T. (1972). Visual processing capacity and attentional control. Journal of Experimental Psychology, 93, 73 - 82. Shulman, G. L., Sullivan, M. A., Gish, K., & Sakoda, W. J. (1986). The role of spatial-frequency channels in the perception of local and global structure. Perception, 15, 259-273. Shulman, G. L., & Wilson, J. (1987). Spatial frequency and selective attention to local and global information. Perception, 16, 89-101. Stoffer, T. H. (1991). Verarbeitung hierarchischer Reizmuster mit drei Ebenen: Ein Test der Pr~izedenzhypothese von Navon. Zeitschrifl fiir Experimentelle und Angewandte Psychologie, 38, 113-148. Sudevan, E, & Taylor, D. A. (1987). The cueing and priming of cognitive operations. Journal of Experimental Psychology: Human Perception and Performance, 13, 89-103. Theeuwes, J. (1991). Exogenous and endogenous control of attention: The effect of visual onsets and offsets. Perception & Psychophysics, 49, 83-90. Todd, J. T., & Van Gelder, E (1979). Implications of a transient-sustained dichotomy for the measurement of human performance.
Journal of Experimental Psychology: Human Perception and Performance, 5, 625-638. Tsal, Y. (1983). Movements of attention across the visual field. Journal of Experimental Psychology: Human Perception and Performance, 9, 523-530. Tsal, Y., & Kolbet, L. (1985). Disambiguating ambiguous figures by selective attention. Quarterly Journal of Experimental Psychology, 37A, 25-37. Wandmacher, J., & Arend, U. (1985). Superiority of global features in classification and matching. Psychological Research, 47, 143 - 157. Ward, M. L. (1982). Determinants of attention to local and global features of visual forms. Journal of Experimental Psychology: Human Perception and Performance, 8, 562-581. Ward, L. M. (1983). On processing dominance: Comment on Pomerantz. Journal of Experimental Psychology: General, 112, 541 - 546. Ward, L. M. (1985). Covered focussing of the attentional gaze. Canadian Journal of Psychology, 39, 546-563. Warner, C. B., Juola, J. E, & Koshino, H. (1990). Voluntary allocation versus automatic capture of visual attention. Perception & Psychophysics, 48, 243-251. Wertheimer, M. (1923). Untersuchnngen zur Lehre vonder Gestalt II. Psychologische Forschung, 4, 301-350. Yantis, S. (1992). Multielement visual tracking: Attention and perceptual organization. Cognitive Psychology, 24, 295-340. Yantis, S., & Jonides, J. (1984). Abrupt visual onsets and selective attention: Evidence from visual search. Journal of Experimental Psychology: Human Perception and Performance, 10, 601-621. Yantis, S., & Jonides, J. (1990). Abrupt visual onsets and selective attention: Voluntary versus automatic allocation. Journal of Ex-
perimental Psychology: Human Perception and Performance, 16, 121-134.