A central conflict in attentional theory has been that between the notions of early selection and those of late selection. The theory of early selection holds that attention begins to influence processing in its initial stages, before a stimulus has been completely evaluated. With late selection, on the contrary, attended and unattended stimuli are preprocessed to equal degrees and then compete for attentional postprocessing. Assuming that the relatively early potentials whose amplitudes are modulated by spatial attention are markers of sensory processing, electrophysiological studies show that spatial attention operates by early selection. Amplitudes of the early potentials P1 and N1 are enhanced in response to stimuli that occur at attended locations [Eason 1981, Rugg & al. 1987], irrespective of the relevance of the specific features of the stimulus [Heinze & al. 1990]. This effect occurs both in the case of sustained attention and in the case of trial-by-trial shifts of attention [Eimer 1994b]. Furthermore, spatial selection seems to operate hierarchically in advance of feature selection. In a task that required a selective response to stimuli having a particular conjunction of colour and location, Hillyard and Munte [1984] found that when stimulus locations were widely separated, the broad negativity associated with selection for colour appeared only in response to stimuli at the attended location, whereas when stimulus locations were close to each other, this selection negativity co-occurred with, and predominated over, ERP signs of spatial selection. ERPs are a particularly useful piece of evidence in the debate on early versus late selection because of their temporal resolution which distinguishes such early augmentations of sensory signals from the other possible mechanism, a lowering of thresholds for behavioural response [Mangun & Hillyard 1987].
ERPs have yielded important information on the manner in which spatial attention operates statically, but what about the changes that occur when attention is in the process of shifting from one location to another? In the first study to address such attentional dynamics, Harter & al. [1989] found a sequence of slow, lateralised potentials associated with centrally cued shifts of spatial attention toward one or the other visual hemifield. This sequence began 180ms after stimulus presentation with an occipitoparietal negativity contralateral to the hemifield toward which attention was being directed. This effect was greater in response to leftward shifts (that is, over the right hemisphere). An occipitoparietal positivity followed at about 400ms and was more pronounced in response to rightward shifts (that is, over the left hemisphere). From about 600ms onward, this positivity was approximately equal in magnitude over both hemispheres. The subjects of this first study were children aged six to nine years, but the major results were later replicated in adults [Harter & Anllo-Vento 1991], albeit with a reduction in the amplitude of the late positivity. Wright & al. [1995] compared the effects of cues that gave information about the probable location of an impending target to those of cues that gave no locative information. They found that the contingent negative variation (CNV), a frontocentral negativity associated with expectation of a stimulus, becomes enhanced by 500ms following the appearance of the cue. It seems from these results that the full electrophysiological effect of attentional switching becomes apparent between 400ms and 600ms following an instruction to shift attention, during the period of the late, lateralised, occipitoparietal positivity and the enhanced CNV at frontocentral sites.
This 400ms-600ms latency is a problem, though, since some degree of behavioural response facilitation is apparent with cue-target intervals as brief as 200ms. Anllo-Vento [1995] measured enhancements in amplitudes and reductions in latencies of P1, N1, and P3 components in response to validly cued versus invalidly cued targets, at various cue-target intervals. She found significant effects with a 600ms interval, but not with a 200ms interval. She concluded that this analysis could not provide much information on the time courses of shifts in visual spatial attention, because at intervals as short as 200ms, electrophysiological data does not reflect the observed attentional facilitation of behavioural responses. She speculated that the absence of a positive finding may have been due to neural refractoriness. Nevertheless, clearly something is going on in the brain to produce the observed behaviour. By what sorts of observations might this early attentional processing be made to manifest itself electrophysiologically?
Eimer [1993] found, using centrally presented cues, that attentional facilitation operates in two stages during the interval between cue and target. Only one of these stages is subject to voluntary control. An early readiness potential appears in the hemisphere contralateral to a spatially specific cue, about 100ms following cue onset, and later shifts to the hemisphere contralateral to the cued response. This early, lateralised response to the location of the cue persists even when the cue calls for an endogenous shift of attention to the opposite hemifield, or when the cue gives no information about the location of the impending target. This suggests an involuntary, stimulus-driven process of spatial orienting that cannot be completely suppressed under voluntary control. Further evidence is the existence of an Nd wave evoked by validly peripherally cued targets, even in a context in which the probabilities of valid and invalid cues are equal [Eimer 1994a]. Like the aforementioned readiness potential in response to the cue, this Nd in response to the target was biphasic. It comprised an early (130-180ms), parietal negativity and a later (210-340ms), frontal negativity. This early phase is of especial interest both because of its locus over the parietal lobe, which is known to have a key role in spatial attention, and because it is present uniquely during cued-attention paradigms, not during sustained attention [Eimer 1996]. This target-evoked negativity may constitute a second iteration of the the cue-evoked negativity decribed by Harter [1989] during shifts of spatial attention. Comparisons between valid, invalid, and neutral cueing conditions reveal that this early phase reflects a cost of invalid cueing rather than a benefit of valid cueing [Eimer 1996]. It seems, then, that in the context of trial-by-trial cueing, the cue by itself does not suffice to draw attention completely to the cued location, so that the brain response even to a validly cued target reflects some amount of this cost of invalid attentional allocation.
With the notable exceptions of these studies, previous ERP studies of spatial attention have compared ERPs in blocks of stimuli in which attention is focussed in one location to ERPs in blocks of stimuli in which attention is focussed in another. Within each block, there is no change in the allocation of attention. Thus the effects found in such studies are characteristic only of statically allocated attention, and do not necessarily describe the characteristics of attention that is in the process of being shifted from one location to another.
The canonical experimental paradigm for behavioural measures of attentional dynamics is that pioneered by Posner [1980], in which a visual target is preceded by a spatial cue. Attentional function is indexed by the speed with which the subject responds to the target. Normal subjects show a validity effect in responding to cued targets, that is, at short cue-to-target latencies they respond faster when the cue directs attention to the correct location than when the cue directs attention to an opposite location. This effect occurs no matter whether the cue is a spatial signal, such as a flash in the periphery of the visual field, or a symbolic indicator, such as an arrow in the centre of the visual field. On the basis of lesion studies [Posner & al. 1984], Posner & al. [1987] have divided the process of shifting attention into three components: disengagement of attention from some prior focus, motion of attention from that focus to a new focus, and engagement of attention at the new focus. Posner identifies motion with the superior colliculus, engagement with the pulvinar nucleus of the thalamus, and, most significantly for this discussion, disengagement with the parietal lobes. The validity of this analysis is open to dispute, however, since in the invalidly cued trials upon which it depends, the physical signal to move and to engage attention is identical with the signal to disengage attention, and the motion and reengagement operations would therefore seem to be occurring after the fact.
Several previous experiments have used behavioural measures similar to those of the Posner paradigm to explore shifting of attention that is repetitive. One of these [Akshoomoff & Courchesne 1992; Courchesne & al. 1994] measured the execution time of attentional shifts between auditory and visual channels. Subjects were presented with common, nontarget stimuli and rare, target stimuli in each channel. In the auditory channel, the nontargets were low tones and the targets were high tones. In the visual channel, the nontargets were green squares and the targets were red squares. On detecting a target in the attended channel, subjects had to shift their attention to the other channel as quickly as possible and to ignore the previously attended targets while responding to the newly attended targets. Each block of stimuli incorporated many such shifts back and forth between audition and vision. A similar experiment studied intramodality shifts of attention between colour and form [Akshoomoff & Courchesne 1994]. Behavioural measures achieve good temporal resolution, but such measures specify only the output of the final, visible stage of a biological system, and so the observed effects often admit many competing functional explanations. In the case of reaction-time measures of attention, increased latency may represent a more sensory deficit that causes perception to be delayed, or it may represent a more motoric deficit that causes the threshold for behavioural response to be raised.
The combination of the functional specificity of ERPs with the attentional dynamics that arise in the attentional-shift paradigm would seem to permit a finer-grained analysis of attentional processing. However, such a combination is not straightforward. In order to generate the very short cue-to-target intervals and repetitive shifts that are necessary in order to avoid a ceiling effect, cues to shift attention would have to occur very rapidly in sequence. This would cause EEG recording epochs to overlap. This corruption of averages by overlapping epochs is why, although EEG has been recorded during the shift experiments mentioned above, the only useful features were the very large P300 and P700 components. These components have long latencies, so although they can corroborate the behavioural evidence that an attended target has been detected and an attentional shift has occurred, they do not add much information about the time course or functional basis of attentional shifts. They can supply only retrospective information.
One solution to this problem is to bite the bullet--to accept, and, in fact, to embrace a large overlap of EEG epochs. The visual system, like any electrical system that resists changes in potential and in current, can be driven by periodic input and resonates more or less well depending on the frequency of such input. The steady-state visual evoked potential (SSVEP) is the continuously oscillatory brain potential associated with a stimulus that repeats at a constant frequency [Regan 1977]. SSVEP transfer functions tend to peak in the alpha band [Fedotchev & al. 1990]. At frequencies of about 6s-1 and above, most of the output power is at the fundamental frequency, while at lower input frequencies, the second harmonic dominates [Previc & al. 1987]. To a first approximation, the SSVEP can be viewed as the superposition of successive single-trial evoked potentials to the individual stimuli that comprise the stimulus train. At such rapid stimulus presentation rates, however, responses to successive stimuli may influence each other, and therefore the resulting SSVEP may not be simply additive. Perturbations of the SSVEP can be used as indices of attentional state. In a visual vigilance task, for instance, the occipital SSVEP was attenuated during anticipation of a target, and the right frontal SSVEP was attenuated during detection of a target [Silberstein & al. 1990]. During the Wisconsin Card Sorting Test, the SSVEP was attenuated prefrontally and temporoparietally when the sort criterion was changed [Silberstein & al. 1995]. Presumably these attenuations indicate the recruitment of these cortical regions for processing tasks that compete with the oscillations of the SSVEP.
In addition to its resonance effect at the input frequency and at its harmonics, periodic visual stimulation influences the background EEG. At non-harmonic frequencies, EEG power decreases during resonance of the visual system [Mast & Victor 1991]. Thus the effect of periodic visual stimulation can be conceptualised at least partly as a transfer of power from a broad frequency band to narrow harmonics. This suggests the recruitment of cortical neural systems for the processing of the phasic visual stimulation, at the expense of the processing of other stimuli.
Assuming that the same generators contribute to single-trial ERPs and to the SSVEP, the SSVEP should show a pattern of modulation by spatial attention analogous to that of single-trial ERPs. That is, a shift away from one flashing stimulus in one visual hemifield, and toward another flashing stimulus in the opposite hemifield, should evoke a decrease in SSVEP amplitude in the hemisphere contralateral to the previously attended hemifield, and an increase contralateral to the newly attended hemifield. The enhancement of P1 has a diffuse distribution, which has a maximum at the focus of attention but which decreases as a spatial gradient surrounding that focus [Eason 1981; Mangun & Hillyard 1987, 1988], and covaries with behavioural measures such as detectability and reaction time [Mangun & Hillyard 1990]. Motion of this spatial gradient from one location to another might create a temporal gradient of P1 enhancement as the new focus of attention is gradually covered by the attentional spotlight. At the same time that P1 builds, other components may diminish. N1, in particular, seems to correspond more with the transient reorienting of attention than with static attentional priming, since its enhancement occurs only at the beginnings of sequences of stimuli occurring in an attended location [Luck & al. 1990]. Current source density mapping of P1 and N1 reinforces this hypothesis: whereas P1 is maximal over occipital cortex, N1 has a parieto-occipital maximum suggestive of involvement with the dorsal stream of visual processing [Mangun & al. 1993].
The repetitive nature of the SSVEP would make it possible to observe not only this modulation itself but also the time course over which it operates. Might the relatively early, negative component described by previous investigators be a manifestation of the attentional switching process itself? If so, one would expect a modulation of the SSVEP response to occur with a time course similar to this early negativity. Previous studies have shown us what happens to the cue-evoked response when a demand is made to invoke the switching process, and what happens to the target-evoked response after this switching process completes. They have not, however, shown us what happens to the target-evoked response while the switching process is still active.
A related question concerns whether or not there are differential effects at the SSVEP frequency and its harmonics versus at other, unrelated frequencies. Such a differential effect would suggest augmentation of the signals produced by task-relevant stimuli at the expense of signals associated with competing brain processes. The present experiment addresses these questions: is there a modulation, what is its time course, and what is its frequency specificity.