According to Lavie (1994) the parallel processing that occurs in the Stroop task is attributable to the fact that the task is characterized by an especially low level of perceptual load as well as poor physical distinction between the relevant and irrelevant stimuli. Lavie claims that early selection requires the perceptual load of the task to be sufficiently high to exceed the upper limit of available resources in addition to a clear physical distinction between relevant and irrelevant information. Therefore, the Stroop task does not allow the failure of the early selection process to be attributed to either the poor physical distinction or the low perceptual load.
Another criticism of the task is that it does not shed any light on unattended processing as due to overlapping of the irrelevant and relevant stimuli you have to attend the words to see the colour and vice versa. This means that neither of the stimuli are left unattended. This was overcome in experiments by Gatti and Egeth (1978) and Kahneman and Henik (1981) who separated the word and colour into two objects. Gatti and Egeth (1978) presented subjects with a central, relevant, coloured patch, vertically flanked by two identical irrelevant colour names. The words appeared at different distances from the colour patch. The findings showed that it took significantly less time to name the colour of the patch when the colour names were compatible with it rather than when they were incompatible. This effect became weaker as the distance between the words and the patch increased, however, it was never eliminated. It was argued that the Stroop paradigm employed by Gatti and Egeth (1978) confounds spacing with visual acuity and therefore is inadequate in studying the effects of spatial separation on distractor processing. Nontheless, it still provides evidence in favour of late selection.
Kahneman and Henik (1981) presented two remote colour words with one in a square and the other in a circle. Subjects were asked to name the colour of the word in a specified shape. A greater Stroop effect was found from incongruent colour words appearing within the specified shape with less effect for incongruent colour words appearing in the irrelevant shape. Heijden, Hagennar and Bloem (1984) used a similar procedure to Kahneman and Henik (1981) and found that cueing the relevant location eliminated any inference from the incompatible irrelevant word when the target itself was an incompatible colour name. However, interefernce resumed when the relevant word was neutral. It was posited that the irrelevant word was processed under both conditions, but that interference effects of this distant and uncued object could be masked by the more substantial interference found when the incongruent information was present in the target object itself. Distractor processing would be expected in this case as display size is small and so perceptual load is low, thus, allowing for late selection (Lavie, 1994).
Eriksen and Eriksen (1974) developed a variation of the Stroop task where subjects are presented with a row of letters comprising of one central target letter flanked by identical distractor letters on either side. The subjects’ task is to respond to the central letter as quick as possible by moving a lever in a particular direction. It was found that when two letters (H and K) were assigned to the same lever response, a target H is responded to more slowly in displays such as SSHSS, where the lever responses are incompatible, compared to KKHKK displays where the lever responses are compatible. This effect is known as the flanker compatibility effect (FCE). The Erikens’ model assumes that two or more items can complete processing at the same time and begin to evoke their responses. However, only one response can be elicited and, therefore, interference arises between the distractor and the target consequently giving rise to the need for a decision mechanism which determines which of the two responses actually corresponds to the target. In this way the data shows clear evidence for response level interference thus supporting the late view. The fact that there were only ever two different types of letters which were placed close together means that according to Lavie(1994) late selection would have been expected as low perceptual load and poor physical distinction lead to the failure of early selection. The FCE has also been found in a number of cueing tasks where a bar-marker is presented, before display onset, to direct the subject’s attention to the relevant location (Eriksen & St. James, 1986).
The results of Eriksen and Eriksen (1974) were completely contradicted by the studies of Bjork and Murray (1977) who found evidence for feature specific interference and thus supporting early selection. Their feature specific theory of interference effects is an extension of Estes’ theory (1974), which assumed that there are separate feature detectors for separate features and so the strongest inhibition on a specific feature detector comes from another detector of the same feature. This implies that the best inhibitor for a target letter is another example of the same letter. For example the letter B would be most difficult to detect if the flanker is another B. According to Eriksen’s experiments, as B and B have the same response there should be no interference between them. However, Bjork and Murray argue that in this case interference is due to feature specific interference between signal and noise elements in a display, which results in competition at an early level of feature extraction. This theory is very contradictory to the late view which proposes that if both Bs were concurrently active there would be facilitation of the response to B rather than inhibition.
Evidence against the featural level of extraction is provided by experiments by Styles and Allport (1986) who have found that it was the identity of distractors that caused differential interference. In their experiment subjects were told to write down the identity of the central letter from a display as well as any other letters they saw including their position. The target was surrounded by four distractors which could be either: hatchmarks, letter like flankers, digit flankers, letter from a non-target set and letters from the target set. If interference was at the level of feature extraction then any letter features should have caused equal interference, however, this was not the case. A hierarchy of interference effects was found and in particular the most interference was found when the distractors were from the target set with most errors being mislocations. Clearly it is the identities of letters, rather than their featural components that interact with response to the target. These data provide evidence for the identification of unattended letters prior to selection and so support a late-selection account of visual attention.
In the experiments discussed so far interference effects have been taken as evidence of whether processing of visual information has taken place or not. For example, Eriksen and Eriksen (1974) assume that, as incompatible distractors generally did not slow target response, those distractors were not processed. Driver and Tipper (1989) however, use a different index of processing to show that this assumption may not be valid. This measure of processing is known as negative priming and it has been used extensively by Tipper and his colleagues to explore the level of processing achieved by unattended stimuli and mechanisms of selective visual attention.
Negative priming is thought be one of the strongest sources of evidence for the late selection position and the concept of negative priming is expressed by the fact that ignored distractors are perceived but that responses towards them inhibited. In an experiment by Driver and Tipper (1989) they set out to confirm that the absence of flanker-interference effects cannot be taken as evidence of the absence of semantic processing of the distractor. In their experiment, subjects were required to report the number of red letters and to ignore irrelevant black digits which were either compatible or incompatible with the number of red items in the display. However, was always the same repeated digit. In the prime trial the ignored black digits did not produce any flanker interference in the reporting of the number of red items. On the probe display the subjects task was again to report the number of red letters but to this time ignore the black letters. It was found that the latency to report the number of red items in the probe display was slower when the black digits in the prime display were digits corresponding to the response now required to the probe. For example, if three red letters were presented in the probe trial, the subjects were slower to subjects were slower to say “three” if they had previously ignored the digit “3” in the prime trial compared to if they had ignored the digit “2”. A study by Fox (1995) also found that ignored distractors which produce no within-trial flanker interference in the prime task, still produce negative priming in the probe task, indicating the distractors are processed at least to the level of identity. The consequences of this data are that if the demonstration of no interference is not equivalent to demonstrating no processing, then any interpretation of experiments that rely on this assumption may be flawed.
Overall, although the negative priming phenomenon appears to provide a sensitive measure for detecting traces of irrelevant processing under certain conditions it is important to note that this model does not necessarily provide unequivocal support for the concept of late selection. Negative priming may be partially due to retroactive priming caused by the target processing in the subsequent trial. This infers that a relatively “raw” trace of the distractor in the prime trial (n) may be primed by the higher level of processing of the target in the probe trial (n+1).
Although mainly evidence in support of the late selection has been presented here there is a vast number of studies which support both sides of the early-late debate. The variety of factors associated with varying experimental conditions may gives rise to variation in the underlying perceptual processes. This fact has been addressed by Lavie (1995) who claims that “perceptual load plays a causal role in determining the efficiency of selective attention”. It was claimed that perception has a limited capacity but proceeds automatically on all stimuli relevant as well as irrelevant within capacity. High memory load tasks relate to early selection where there is limited processing of unattended stimuli as capacity is exceeded. However, if attentional demand of the task is low, irrelevant distractors are processed as there is still some vacant attentional capacity.
This argument may indeed account for the discrepancy between the two theories and offers a resolution in the debate over whether selective attention is an early or late process. However, there are experiments, which do not comply with this theory and therefore, Lavie claims that in some cases it must be assumed that in certain conditions feature level interference will occur. Lavie’s work offers a good compromise between the two approaches by integrating the (early-selection) assumption that perception is limited process with the (late selection) assumption that perception is an automatic process.