Event-related potentials to Stroop and reverse Stroop stimuli

Abstract

In the Stroop task, the latency of response to a colour is either faster or slower in the presence of a congruent or incongruent colour–word (J. Exp. Psychol. 18 (1935) 643). Debate remains as to whether this effect occurs during early stimulus processing or late response competition. The present study examined the task using reaction time (RT) and event-related potentials to determine temporal differences in this processing. The ‘reverse Stroop’ effect (where colour interferes with processing of a colour–word) which is much less well established, was also examined. Standard Stroop interference was found as well as reverse Stroop interference. A late lateralised negativity at frontal sites was greater for Incongruent trials and also for the word–response (reverse Stroop) task, and was interpreted as semantic selection and word-rechecking effects. Late positive component latency effects generally mirrored the speed of processing of the different conditions found in RT data. Stroop effects were also found in early temporal N100 and parietal P100 components, which differentiated Congruent from Incongruent trials in the reverse Stroop but not the standard Stroop, and were interpreted as early perception of physical mismatch between the colour and word. It was concluded that Stroop stimuli are processed in parallel in a network of brain areas rather than a particular structure and that Stroop interference arises at the output stage.

Introduction

The Stroop colour–word test is a widely known and robust measure of selective attention and interference. Typically the test requires a response to the colour of word stimuli: when a colour–word is printed in an incongruent colour (e.g. RED printed in blue) response is longer, compared to response to a non-colour (neutral) word printed in the same colour. The difference between response times to the Incongruent and Neutral conditions is known as interference. Conversely, when both colour–word and colour are congruent (e.g. RED printed in red) responses are faster than neutral stimuli, and are known as facilitation. This phenomenon, where words interfere with the processing of colour is known as the ‘Stroop effect’; a converse and less robust situation where colour interferes with the processing of words is termed the ‘reverse Stroop effect’.

Despite more than 60 years of research into this phenomenon, the nature of Stroop interference remains unexplained. Researchers have debated the locus of interference in the Stroop task: whether it occurs early at stimulus processing stages, or later at the output stage as response competition. The ‘late selection’ response competition theories have dominated the Stroop literature. According to the Relative Speed-of-Processing theory, for example, interference arises because processing of colour and word stimuli occur in parallel but at different speeds (e.g. Morton and Chambers, 1973). This processing converges on a response channel which will only accept one piece of information (colour or colour–word) at a time; the word information is processed faster and reaches the output stage before the slower colour information, thus interfering with the correct colour response (MacLeod, 1991). The related, Automaticity theory also proposes word processing to be faster, because word reading is a more practiced and automatic skill and requires few attentional or processing resources (e.g. Posner and Snyder, 1975). However, manipulating stimulus onset asynchrony (presenting the slower colour information before the word; Glaser and Glaser 1982) does not ‘speed’ colour processing or reduce the interference effect. Similarly giving extensive practice at the task and at colour responding (beginning with Stroop (1935), on the assumption that the word is processed faster due to greater training at word reading) reduces but does not eliminate the interference effect. If this output stage is the locus of Stroop interference, it is unlikely that response competition occurs because of the relative speed-of-processing of the word and colour elements.

Other models place the locus of interference early in processing. One example is the Parallel Distributed Processing model (e.g. Cohen et al., 1990) which comprehensively outlines the nature of information processing prior to output. This model allows processing of word and colour to occur in parallel, along pathways that focus on strength rather than speed of processing. Attention acts to strengthen the processing in a particular pathway, and the word may automatically be stronger due to greater practice at word reading. The strength of the pathway and its information processing capabilities influences the likelihood that it will produce the response (MacLeod, 1991). The limited-capacity response channel constraint of for example the Relative Speed-of-Processing theory is overcome, as there can be as many output units/response channels as there are possible responses. Response competition is however accommodated by this theory: two output units (for the colour and colour–word) can be activated at the same time, but the unit which is most strongly activated (or attended) produces a response.

The idea that interference occurs earlier in processing has also been supported by the interpretations of several recent behavioural studies, all of which point to a decision making stage after early feature analysis but before the response stage (Naish, 1985, Head and Pedoe, 1990, Koch and Brown, 1994). Initial event-related potential (ERP) evidence, however, has presented conflicting results. An attention study by Johnston and Venables (1982) examined early ERPs and concluded that Stroop interference was a breakdown in selective attention that occurred very soon after stimulus presentation. When subjects were attending (and responding) to words, word probes produced a larger P85 component compared to colour probes, and vice versa when subjects were attending to the colour of the stimuli (Johnston and Venables, 1982). They concluded that it is necessary in an interference task to focus attention on the task-specific feature of the stimulus, and so probes related to the attended feature produced the greater P85 ‘attention effect’. However, this early effect was produced by the probe stimuli and not by the Stroop stimuli themselves.

Another Stroop-like study reporting early interference effects in ERPs, examined processing of colours and words separately with coloured word stimuli (Aine and Harter, 1984a, Aine and Harter, 1984b). Attention was directed to a particular word or colour (target stimulus) with the participants required to press a button when the stimulus contained the target. Results indicated that Stroop interference occurred primarily in the left hemisphere as an increased amplitude for ‘Stroop’ (incongruent) stimuli at 225 ms, and that the interference occurred at some encoding stage of processing prior to response (Aine and Harter, 1984a, Aine and Harter, 1984b). The attend-colour and attend-word conditions were also able to distinguish between word- and colour-selection effects. Colour-related ERP modulations occurred consistently earlier than the word-related effects, again questioning the relative Speed-of-Processing view that Stroop interference occurs because the word is processed faster and thus interferes with the required colour response. However, ‘interference effects’ in the study occurred earlier than ‘word selection’ effects, implying that the words interfered with colour processing before the words were processed (Aine and Harter, 1984b). Also, the behavioural reaction time (RT) measures of the ‘Stroop’ and ‘non-Stroop’ stimuli were not significantly different, suggesting that the task did not elicit standard Stroop (or reverse Stroop) interference. Although, the results of this study are generally accepted (e.g. MacLeod, 1991), they are not deemed measures of the Stroop task.

Conversely, Duncan-Johnson and Kopell, 1980, Duncan-Johnson and Kopell, 1981 examined P300 latency; reasoning that if Stroop interference was due to slower semantic processing and stimulus identification in the Incongruent condition then P300 latency for that condition would be delayed relative to the Congruent/Neutral. Finding no difference in the latency of this waveform across the three conditions, Duncan-Johnson and Kopell (1981) concluded that Stroop interference arose primarily from response competition between the colour and word at the output stage. These findings were recently replicated by Ilan and Polich (1999), who also found no relationship between RT and P300 latency, taken to indicate that the Stroop effect occurs after the stimulus has been evaluated.

The theory that the Stroop interference effect is a late, response competition-related phenomenon has been supported by several recent ERP studies, which have reported that the Stroop task produces modulations of later components of the waveform. While not specifically addressing the early vs. late issue, Rebai et al. (1997) found a 350–450-ms N400 wave elicited at the midline frontal and central sites for coloured word stimuli. The N400, first labeled by Kutas and Hillyard (1980), is accepted to indicate processing of ‘mismatch’, for example a sentence ending in a semantically incongruent word or other deviation from the prevailing stimulus context. Rebai et al. (1997) proposed that the Stroop task was an example of anomalous stimuli, as the subject must evoke a response (the colour) that is incongruent with the stimulus context (the word).

However, Rebai et al. (1997) examined blocks of mixed stimuli (congruent and incongruent), as well as ‘homogenous’ blocks (trials either all Congruent or all Incongruent). In the homogenous congruent blocks, it is more likely that the subjects would simply read the words and rely solely on the word information, instead of responding to the colour, since the colour and word always matched. Therefore, the authors cannot know what their subjects were actually doing in this condition. Comparison of the P300/N400 epochs for these and neutral stimuli were not significantly different. Also, while the homogenous incongruent and neutral stimuli were significantly different, there was no difference between the mixed incongruent and neutral stimuli. The Incongruent and Congruent trials, either from homogenous or mixed blocks, were not compared statistically. Comparison of these interfering and facilitating conditions is crucial for any interpretations made regarding the nature of the Stroop effect, particularly given PET studies such as those by Bench et al. (1993) which failed to find a difference between the Congruent and Incongruent conditions. Further, RT data were not reported, so the pattern of responding in the conditions and whether a ‘Stroop effect’ was elicited is not known.

Liotti et al. (2000) found a similar negativity in analyses of Congruent and Incongruent condition difference waveforms, at a latency of 350–500 ms (maximal at 410 ms latency) over midline sites. The authors, however, claimed that this was not an N400, based on the differential distribution of the effect for vocal and manual response conditions in their results, and interpreted it as evidence of conflict processing and resolution in the anterior cingulate (Liotti et al., 2000). A 500–800-ms positivity was also found in the left temperoparietal areas, interpreted by the authors as evidence of additional word meaning analysis (apparently after resolution of the colour/word response conflict).

West and Alain (1999) also examined the time course of the Stroop effect with ERPs and reported a negative modulation over the bilateral fronto-central region (maximal at Cz), which was more negative for Incongruent trials. This slow wave, however, extended from 500 to 1000 ms; later than the negative modulation at this site reported by Rebai et al. (1997) and Liotti et al. (2000). A second modulation, a bilateral positivity peaking at 500 ms over the fronto-polar region (maximal at F12), was attenuated for Incongruent trials relative to the other conditions. West and Alain (1999) interpreted these two effects as evidence of conflict detection in the lateral prefrontal areas, and resolution in the anterior cingulate, in the interfering condition.

Two further modulations were reported in the left temperoparietal areas and interpreted as evidence for a ‘meaning-based conceptual level system’ (West and Alain, 1999, p. 157). These were similar in latency and site to the ‘additional word processing’ activity reported by Liotti et al. (2000). West and Alain (1999) proposed that these modulations reflected word and colour pathway processing respectively: e.g. the second (positive) effect indicated additional ‘colour pathway’ processing occurring to guide the appropriate response, after colour/word incompatibility had been detected. The network of brain regions activated was taken by West and Alain (1999) as evidence for the dual-pathway models of Stroop interference (e.g. Cohen et al. (1990) Parallel Distributed Processing theory already described).

These frontal and later temperoparietal modulations have thus far been the most promising ERP indices of Stroop interference. They are late components occurring near or at response latency. However, these effects and the processing they reflect still need to be clarified—the frontal modulation in particular has variously been reported as an N400 (which failed to distinguish between Congruent and Incongruent trials), an ‘early’ negativity but not an N400, and later slow wave negativity. While neuroimaging evidence to date suggests that Stroop interference is manifest late in processing, behavioural tasks continue to point to a ‘decision making stage’ prior to this output stage (e.g. Koch and Brown, 1994). Further, the most prominent model (Parallel Distributed Processing theory) concentrates on the strength of ‘processing units’ and automaticity/practice associated with colour naming vs. word reading.

While relatively little progress has been made towards understanding the Stroop effect, even less has been made in the converse ‘reverse Stroop’ effect, which many researchers argue does not even exist in the original task. The ‘reverse Stroop’ effect refers to the effect of colour on word processing. Stroop's original (1935) study included a word-reading condition, and reported that presenting the word in an incongruent colour had no effect on response times. Only extended practice at colour naming increased the amount of word-reading interference, supporting Stroop's (1935) conclusion that the standard interference was due to greater practice at word reading relative to colour naming. The effect however was transient. MacLeod's (1991) review reported that the reverse Stroop effect was unreliable and was typically only produced after considerable alterations to the original task. For example, Dyer and Severance (1972) degraded the word image to decrease word readability and increase the interference from the colour; Palef and Olson (1975) used a task presenting the words ‘above’ and ‘below’ either above/below a fixation point to create semantic interference.

Modern Stroop studies typically do not include a reverse Stroop condition. A word-reading condition is sometimes included, but as per West and Alain (1999), the words are not coloured and not a true ‘reverse Stroop’ situation as there is no conflict between the word and the ink colour. This condition was included by West and Alain (1999) to distinguish trials where word information could serve to guide response, and those trials where only the colour element could guide the appropriate response. As mentioned, Aine and Harter, 1984a, Aine and Harter, 1984b manipulated attention by making subjects respond to either the colour or the word of congruent and incongruent stimuli; this was reasoned by the authors to allow separation of word- and colour-selection effects. Colour processing was reported to occur before word processing effects, maximal in the occipital area; word processing was maximal at occipital sites but interpreted as reflecting the activity of parietal association areas (only four occipital and central electrodes were used).

Conversely, Rebai et al.'s (1997) ERP study examined the ‘reverse Stroop’ with the inclusion of a condition that required reading of congruently and incongruently coloured words. While not stating an explicit reason for including a word-reading condition, they reported no interference in this condition behaviourally and an absence of the N400 effect that was seen for the colour-naming condition in ERP results.

There are several reasons to include the reverse Stroop condition in a ‘standard’ Stroop task. Aine and Harter, 1984a, Aine and Harter, 1984b reasoned that using the two conditions would isolate ‘word selection’ from ‘colour selection’ effects. In a task as complex and little-understood as the Stroop, undoubtedly these two elements should be differentiated from processing unique to the interference, and would thus shed light on the effect. If, as several researchers claim, no interference is produced in this version of the task, then that makes it an ideal control condition for isolating the ‘automatic’ effects of word reading from the task-relevant effects of colour-naming, and interference between the two.

Further, the dual-processing models (e.g. Cohen et al.'s (1990) Parallel Distributed Processing model) postulate separate processing pathways for each element of the stimuli (colour and word in the Stroop case; see also evidence from Michie et al. (1999), regarding separate pathways for colour and pattern processing, which showed a colour-processing ERP modulation occurring even when subjects’ task was to respond to the pattern). Therefore in the Stroop both colour and word are processed, regardless of the task-appropriate response or the response actually made. The response produced depends on the strength of processing in the appropriate pathway. Also, the clinical version of the task typically includes both colour naming and word reading (of congruently and incongruently coloured words). Therefore research should implement a word-reading condition to clarify both Stroop and reverse Stroop interference, as well as further understand the contributions of colour and word processing to brain activity and to that interference.

An investigation and comparison of Stroop and reverse Stroop interference, with the only difference being in the attentional instruction (attend/respond to the colour or the word), would then contribute much to current understanding of this complex phenomenon. The primary aim of this experiment was to elucidate where in the chain of neural processing interference (and facilitation) occurs. For example, the Relative Speed-of-Processing theory argues that the interference arises due to the word being processed faster. The Parallel Distributed Processing model also predicts and demonstrates faster word reading compared to colour naming. If word reading is faster and does not interfere with the slower colour naming, RTs to words in the attend-word task would be equal regardless of their ink colour (conflicting or otherwise). That is, no interference would be apparent. According to the Parallel Distributed Processing model also, word reading is not affected by colour naming—the presence of colour (incongruent or otherwise) has no effect on word reading times. RTs for the word-reading task would also be faster than RTs for the colour-naming task.

ERPs are ideal for examining the neural processing corresponding to these two tasks, as they are temporally precise and allow a direct comparison of the time course of processing for the two tasks. As mentioned, most convincing ERP evidence on the locus of the Stroop effect point to relatively late modulations of the waveform near or at response latency. Behavioural studies and prominent theories of the effect however continue to point to a pre-response ‘decision making’ stage. This study attempted to validate one of those views by examining the entire waveform and time course of processing from stimulus presentation to response. Established early components as well as the later components of recent experimental focus were examined. For example the much-studied parietal P300 is a well-established index of evaluation and decision making (Regan, 1989) and will be examined for differential modulation in the Incongruent condition.

A further aim was to elucidate how processing of the congruent and incongruent stimuli differ. Several studies using less temporally-precise methods (such as fMRI and PET) have failed to distinguish between the interfering Incongruent condition and the facilitating Congruent condition in neural processing, leading to the interpretation that Congruent and Incongruent conditions are processed much the same but to slightly different degrees (e.g. Bench et al., 1993). Such effects which are comparable for the Incongruent and Congruent conditions do not, however, clarify anything about Stroop interference. As mentioned, one recent ERP study has failed to distinguish between the Congruent and Incongruent conditions in the frontal N400 component; two others have reported a difference between Incongruent and Congruent conditions only for later frontal components, well after stimulus identification and classification should have occurred (Liotti et al., 2000, Rebai et al., 1997, West and Alain, 1999). While Stroop interference then classically involves an Incongruent–Neutral comparison, the main comparison of interest here was between the Incongruent and Congruent conditions.

The experiment was comprised of two tasks examining adults’ performance on a Stroop task, measuring concurrent RT and scalp-recorded ERPs. All stimuli were presented centrally and comprised a colour–word or colour–unrelated word (Neutral condition) within a coloured rectangle. In Task 1 participants responded to the colour of the stimulus, while in Task 2 responded to the word.

Following Liotti et al. (2000), it was hypothesised that interference effects (higher amplitude for the Incongruent relative to Congruent/Neutral conditions) would be seen in frontal activation due to attentional modulation and conflict resolution/response competition. Later activity in the left hemisphere parietal areas may reflect further evaluation and processing of the semantic nature of the stimuli, especially in Task 2 where the word is the task-relevant aspect of the stimulus and should be processed without interference from the colour. Following West and Alain (1999), it is anticipated that left temporal modulation (‘colour pathway’ processing) will be absent or attenuated for Task 2 relative to Task 1.