Studies of cognitive control focus on how particular cognitive processes are engaged and inhibited in the service of goal-directed behavior. To address this in the laboratory, individuals are presented with stimuli consisting of two dimensions and asked to report the value of a single relevant dimension. Typically, response times (RTs) are slower when the stimulus dimensions are incongruent than when they are congruent, suggesting difficulty in inhibiting processing of the irrelevant dimension. This difference in RTs is referred to as the congruency effect; large congruency effects are commonly associated with lower levels of control, and small congruency effects with greater levels of control (Botvinick, Braver, Barch, Carter, & Cohen, 2001; Verguts & Notebaert, 2008).

Conflict monitoring (Botvinick et al., 2001) refers to a feedback mechanism that detects conflict, defined as the activation of multiple responses, and signals the need for increased selectivity. This increased selectivity is observed as the decrease in the congruency effect following trials producing conflict (typically incongruent trials) as compared to trials absent of conflict (typically congruent trials) (Gratton, Coles, & Donchin, 1992). This sequential modulation of the congruency effect is generally referred to as conflict adaptation. Here, we use the more theoretically neutral term sequential congruency effect (SCE) to separate the effect from the theoretical interpretation.

The way in which this signal is thought to increase selectivity has evolved considerably since it was originally proposed. At present, at least three general statements about the SCE are well (if not unanimously) supported by the available evidence. First, the SCE occurs in the absence of overlap between the trial N and N + 1 stimulus features (Kerns et al., 2004; Notebaert, Gevers, Verbruggen, & Liefooghe, 2006). This suggests a change in selectivity that is not driven by simple feature repetition. Second, the SCE occurs within rather than across tasks. Thus, if trials N and N + 1 involve different tasks, no SCE is observed (Fernandez-Duque & Knight, 2008; Funes, Lupiáñez, & Humphreys, 2010). However, one can observe an SCE between trials N and N + 2 if trial N + 1 is drawn from another task (Funes et al., 2010). Third, within a task, the SCE is specific to the dimension that produces conflict. A task that combines Simon and flanker stimuli shows modulation of the Simon effect on trial N + 1 in response to Simon conflict on trial N, and modulation of the flanker effect on trial N + 1 in response to flanker conflict on trial N. However, Simon conflict does not modulate the flanker effect, or vice versa (Akçay & Hazeltine, 2011). Combined, these results appear consistent with adjustments that are separable from specific stimulus features but that modulate processing for specific stimulus dimensions.

Because conflict occurs when a stimulus dimension incorrectly predicts a response, conflict monitoring is a mechanism that would allow the system to weight stimulus dimensions by how accurately each dimension predicts the correct response (cf. Melara & Algom, 2003).

Item specific proportion congruence

Complicating the preceding picture are item-level manipulations (see Bugg & Crump, 2012, for a review). In the Stroop task, one can construct stimulus lists such that particular words are more likely to occur in an incongruent condition and other words are more likely to occur in a congruent condition, while maintaining equal proportions of incongruent and congruent trials. This ensures equal probabilities that prior trials will be congruent or incongruent. The item specific proportion congruence (ISPC) effect (Jacoby, Lindsay, & Hessels, 2003) refers to the finding that words with a high proportion of incongruent occurrences are associated with smaller congruency effects than are those with a high proportion of congruent occurrences.

Two accounts have been advanced to explain the ISPC effect. One extends the conflict-monitoring mechanism, suggesting a form of item-level control whereby individual words and colors become associated with unique control settings on the basis of probability of conflict (Blais, Robidoux, Risko, & Besner, 2007; Bugg, Jacoby, & Toth, 2008). In contrast, a contingency account suggests that individuals learn to use word information to predict the response (Schmidt, 2013; Schmidt & Besner, 2008). As described, item-level control modulates the influence of word information on the basis of the specific word’s prior informativeness about a congruent response. Because this adjustment is based on the contingent relationship between the word, the color, and the response, item-level control will have the effect of encoding individual word–response contingencies.

Here, we do not try to distinguish between the item-level control and contingency accounts. Instead, our question is how variation in the informativeness of the word dimension might relate to the more general SCE. Note that in item-level manipulations, the variability in congruency of individual words means that a dimension-level weighting of word information is less accurate than it would be for lists with words containing equal proportions of congruent trials. If the SCE reflects a mechanism that helps to set the appropriate weighting of stimulus dimensions during processing, the presence of item-level variability may discourage such sequential adjustments, and hence reduce or eliminate the SCE.

Experiments 1A and 1B

In Experiments 1A and 1B, we simply asked whether the SCE could be observed in the context of an item-level manipulation. We report results from two different experiments that were conducted at different times with different participants and with slightly different list structures. Because one possible outcome was the lack of an SCE, the ability to replicate such a finding was important.

Method

Participants

The participants were recruited from the Georgia Institute of Technology undergraduate population and received course credit. Their ages ranged from 18 to 23 years. Experiment 1A consisted of 24 participants (mean age = 20.1), and Experiment 1B consisted of 24 participants (mean age = 19.8). One participant in Experiment 1A was removed due to an error rate above 40 %.

Materials and stimuli

In all of the experiments, we used E-Prime 2.0 software (Psychology Software Tools, Pittsburgh, PA) to control the display of stimuli and record RTs. Stimuli were displayed on a 14-in. color (VGA) monitor. A microphone connected to a Psychology Software Tools Serial Response Box measured voice onset times. Words were presented against a gray background with each letter subtending 0.17 deg of visual angle.

Experiment 1A

Six color words (“black,” “blue,” “green,” “red,” “white,” and “yellow”) were displayed in one of the six corresponding colors. Stimulus lists were constructed by creating two color–word sets. For a given participant, three of the six colors/words were presented in the mostly congruent condition, and the remaining three colors/words were presented in the mostly incongruent condition. Within the mostly congruent condition, a given word appeared as congruent on 60 trials and as incongruent on 20 trials. Within those 20 incongruent trials, each word was paired on ten of the trials with each of the remaining members of the set. The same distribution was used for all members of the set. In the mostly incongruent condition, each word appeared in a congruent trial 20 times and in an incongruent trial 60 times. Within those 60 incongruent trials, each word was consistently paired with one member from the set. The same distribution was used for each member of the set. Participants performed five blocks of 96 trials, for a total of 480 trials. Stimulus lists were counterbalanced across participants, such that each color/word appeared an equal number of times as a mostly congruent and mostly incongruent item.

Experiment 1B

Four color words (“blue,” “green,” “red,” and “yellow”) were displayed in one of the four corresponding colors. Stimulus lists were constructed using color–word pairs. This meant that a given word would appear in its matching color and in only one nonmatching color. For a given participant, one of the color–word pairs was presented in the mostly congruent condition, and the remaining pair was presented in the mostly incongruent condition. Within the mostly congruent condition, a given word appeared as congruent on 30 trials and as incongruent on ten trials. In the mostly incongruent condition, each word appeared as congruent on ten trials and as incongruent on 30 trials. This resulted in a total of 160 trials.Footnote 1 The stimulus lists were counterbalanced as in Experiment 1A.

Procedure

All participants were instructed to ignore the word and to name the color in which the word was displayed as quickly as possible while maintaining a high degree of accuracy. The following sequence of events occurred on each trial: (a) three fixation crosses (“+++”) were presented in the center of the screen for 500 ms, (b) a blank screen appeared for 200 ms, (c) the stimulus was presented and remained on the screen until a vocal response was detected, and (d) the screen cleared for the 500-ms intertrial interval. Participants performed 20 practice trials consisting of equal numbers of congruent and incongruent trials. Participants were tested individually while seated next to an experimenter who coded incorrect responses and voice key errors. The entire experimental session lasted 1 h.

Results

An alpha level of .05 was used for all of the reported results. Prior to all analyses, incorrect responses, voice key errors, and RTs greater than 2,500 or less than 200 ms were excluded. The trimming procedure resulted in the removal of 4.7 % (Exp. 1A) and 4.1 % (Exp. 1B) of all trials. The range in accuracy for the cells in the analyses reported below was between 94.5 % and 99 % across the two data sets. We do not report statistical analyses of error rates, both because the rates were low and because few error responses were made in the congruent condition.

ISPC analysis

To demonstrate the typical ISPC effect, the data were analyzed in separate 2 (Item Type: mostly congruent, mostly incongruent) × 2 (Current Trial Congruency: congruent, incongruent) repeated measures analyses of variance (ANOVAs). As can be seen in Table 1, each experiment shows a main effect of current trial congruency—1A, F(1, 22) = 139.97, η p 2 = .864, and 1B, F(1, 23) = 43.12, η p 2 = .949—and both data sets show the ISPC effect as reflected in the modulation of the Current Trial Congruency × Item Type interaction—1A, F(1, 22) = 42.20, η p 2 = .657, and 1B, F(1, 23) = 6.12, η p 2 = .210.

Table 1 Mean response times (in milliseconds) and 95 % within-subjects confidence intervals (in parentheses) for trial condition as a function of item type across the four item manipulations
Full size table

SCE analysis

For this analysis, we also excluded trials in which the color dimension or word dimension overlapped on the previous trial, removing all color or word repetitions (Kerns et al., 2004; Mayr, Awh, & Laurey, 2003). In addition, for Experiment 1A we excluded trials on which item type repeated. These criteria resulted in the removal of 40.9 % (Exp. 1A) and 48.9 % (Exp. 1B) of all trials. Across the experiments, each participant contributed a minimum of 18 trials to each cell in the analysis.

Data from each experiment were analyzed in separate 2 (Previous Trial Congruency: congruent, incongruent) × 2 (Current Trial Congruency: congruent, incongruent) repeated measures ANOVAs. As can be seen in Fig. 1, each data set showed a main effect of current trial congruency: 1A, F(1, 22) = 67.05, η p 2 = .753, and 1B, F(1, 23) = 44.73, η p 2 = .660. The main effect of previous trial congruency was significant only in Experiment 1A: F(1, 22) = 5.19, η p 2 = .191. Importantly, we observed no modulation of the Current Trial Congruency × Previous Trial Congruency interaction, Fs < 1. Therefore, across two ISPC manipulations, we found no evidence for the SCE.Footnote 2

Fig. 1
figure 1

Response times as a function of previous trial condition and current trial condition across Experiments 1A and 1B. Error bars represent 95 % within-subjects confidence intervals. Error proportions are shown in parentheses next to the corresponding point

Full size image

Discussion

The results of Experiment 1 are consistent with the suggestion that when words are differentially informative of a congruent response, the SCE is absent. Note that prior studies have shown the absence of the SCE across tasks or stimulus properties. Here, the distinction between words was not based on predefined task sets (Funes et al., 2010) or differences in perceptual or response features (Hazeltine, Lightman, Schwarb, & Schumacher, 2011). Instead, the distinction between words was established within the context of the experiment.

We acknowledge that our conclusions from Experiments 1A and 1B were based on a null finding. Experiment 2 was designed to test more directly whether list context determines the presence of the SCE. Here, participants performed blocks of trials in which words are differentially informative of a congruent response (item blocks) and blocks in which words are equally informative of a congruent response (balanced blocks). We have suggested that the SCE reflects the sequential updating of an estimate of the informativeness of the word dimension. Moreover, consistency encourages individuals to track such informativeness at the level of the word dimension, whereas item variability discourages such attempts to track informativeness at the dimension level. If so, balanced blocks should demonstrate the SCE, whereas item blocks should not.

Experiment 2

Method

Participants

A group of 48 participants were drawn from the same pool as in Experiment 1, and they ranged in age from 18 to 22 years (M = 19.7).

Materials and stimuli

Four color words (“blue,” “black,” “green,” and “red”) were displayed in one of the four corresponding colors. As in Experiment 1B, stimulus lists were constructed using color–word pairs. This meant that a given word would appear in its matching color and in only one nonmatching color.

The balanced blocks consisted of each word appearing in its matching color on ten trials and in a nonmatching color on ten trials. In item blocks, for mostly congruent items, each word appeared in its matching color on 15 trials and in a nonmatching color on five trials. For mostly incongruent items, each word appeared in its matching color on five trials and in a nonmatching color on 15 trials. The word–color pairs remained consistent during both the item and balanced blocks. Participants performed four blocks of 80 trials, for a total of 320 trials.

Procedure

Half of the participants performed two item blocks followed by two balanced blocks, and half of the participants performed two balanced blocks followed by two item blocks. Participants were not informed of any differences between the blocks, and they performed 16 practice trials consisting of equal proportions of congruent and incongruent trials. The instructions, timing, and response coding were identical to those aspects reported in Experiments 1A and 1B.

Results

An alpha level of .05 was used for all of the reported results. Data were excluded using the same criteria as in Experiment 1. This resulted in the removal of 5.2 % of all trials. Accuracy ranged between 93.5 % and 99 % for all of the cells reported below. Due to this low error rate, an analysis of errors is not reported.

ISPC analysis

Word–color pairs were consistent through the experiment. Therefore, the pairs presented together in balanced blocks were also presented together in item blocks. In this analysis, we simply showed that performance on these pairs was different as a function of block type. The data were analyzed in a 2 (Block Order: balanced first, item first) × 2 (Block Type: balanced, item) × 2 (Item Type: mostly congruent, mostly incongruent) × 2 (Current Trial Congruency: congruent, incongruent) repeated measures ANOVA, with Block Order as a between-subjects factor.

Consistent with Experiment 1, we observed a main effect of current trial congruency, F(1, 46) = 222.99, η p 2 = .823, which was modulated by item type, F(1, 46) = 38.22, η p 2 = .455. Furthermore, we observed that item type also modulated the block type effect, F(1, 46) = 6.89, η p 2 = .130, suggesting that performance was different on pairs in the balanced block than on pairs in the item block. More importantly, we observed the three-way Block Type × Item Type × Current Trial Congruency interaction, F(1, 46) = 19.18, η p 2 = .294, driven by an ISPC effect in the item (77 ms) but not the balanced (5 ms) blocks. Finally, we observed a Block Order × Item Type × Current Trial Congruency interaction, F(1, 46) = 13.69, η p 2 = .229. This was driven by a larger ISPC effect when item blocks were presented first (89 ms) than when they were presented last (48 ms) (see Table 1). Block Order did not interact with any other factors. These data suggest that item effects are dependent on the local list context, and when that list context changes to equal proportions of congruent trials for all items, the ISPC effect is eliminated.

SCE analysis

As in Experiment 1B, we removed trials on which the color dimension or word dimension overlapped on the previous trial. In total, 37.4 % of trials were excluded. As is shown in Fig. 2, regardless of order, balanced blocks showed the SCE, but the item blocks did not. The data were analyzed in a 2 (Block Order: balanced first, item first) × 2 (Block Type: item, balanced) × 2 (Previous Trial Congruency: congruent, incongruent) × 2 (Current Trial Congruency: congruent, incongruent) repeated measures ANOVA with block order as a between-subjects factor. We observed main effects of previous trial congruency, F(1, 46) = 39.77, η p 2 = .464; current trial congruency, F(1, 46) = 170.36, η p 2 = .787; and block order, F(1, 46) = 4.05, η p 2 = .081. Block Order did not interact with any other factor. In addition, the effect of current trial congruency was modulated by previous trial congruency, F(1, 46) = 13.65, η p 2 = .229. More importantly, a three-way Block Type × Previous Trial Congruency × Current Trial Congruency interaction was observed, F(1, 46) = 7.29, η p 2 = .137, consistent with the observation in Fig. 2 that the SCE effect is present for balanced but absent for the item blocks.

Fig. 2
figure 2

Response times as a function of previous trial condition and current trial condition for Experiment 2. Error bars represent 95 % within-subjects confidence intervals. Error proportions are shown in parentheses next to the corresponding point

Full size image

Separate analyses of the balanced and item blocks showed a Previous Trial Congruency × Current Trial Congruency interaction for balanced, F(1, 47) = 15.59, η p 2 = .249, but not for item, F(1, 47) = .024, η p 2 = .000, blocks. This was driven by an SCE (98-ms vs. 64-ms congruency effect for prior congruent and incongruent trials, respectively) for balanced blocks that was absent (108-ms vs. 110-ms) for item blocks.

Item-specific sequential adjustments

Participants at the beginning of an experiment are generally not informed about the presence of item-level manipulations, but Experiment 2 suggests that the ISPC effect can emerge relatively rapidly, given the appropriate list context. This suggests that individuals continually encode information about how individual words are predictive of a congruent response. When all words are similarly predictive, the SCE emerges, reflecting a process of tracking the informativeness of stimulus dimensions, while the tracking of individual word (and color) relations with responses continues in parallel. In this account, there continued to be item-level encoding of information in balanced lists, but the design of such lists did not allow for separately demonstrating such effects (e.g., Blais & Bunge, 2010).

An example of such item sensitivity would be the finding that if a given word (or color) appeared in the incongruent condition, then on the next occurrence of that word (or color), the congruency effect would be reduced relative to when the previous occurrence was congruent. This is essentially the description of the SCE, but tied to specific words or colors and spanning multiple intervening trials. This sequential effect must be present in item blocks for the demonstration of the ISPC effect. We asked whether that relation would also be present in the balanced blocks. We examined data from the 24 participants who performed balanced blocks first. We chose this portion of the data in order to eliminate the possibility of any residual effect of items that had previously been presented in an item context. These data were analyzed using a 2 (Previous Word Congruency: congruent, incongruent) × 2 (Current Word Congruency: congruent, incongruent) repeated measures ANOVA. As can be seen in Fig. 3, in addition to the main effects of current and previous word congruency—F(1, 23) = 134.98, η p 2 = .854, and F(1, 23) = 8.39, η p 2 = 0.258, respectively—current word congruency was modulated by previous word congruency, F(1, 23) = 7.99, η p 2 = .267, mirroring the pattern observed in the conventional SCE. Although space precludes reporting further analyses, a similar item-specific sequential adjustment was also present when the analysis was repeated with colors rather than words.

Fig. 3
figure 3

Response times as a function of the condition in which the current word previously occurred within balanced blocks, for participants who performed the balanced blocks first in Experiment 2. Error bars represent 95 % within-subjects confidence intervals. Error proportions are shown in parentheses next to the corresponding point

Full size image

These data suggest that although the SCE is present in balanced blocks and absent in item blocks, the tracking of information at the item level may be present regardless of list context.

General discussion

Across three data sets, the SCE was absent when words were differentially predictive of a congruent response, but present when words are equally predictive of such a response. Experiment 2 clearly indicated that the local list context is what determines the presence or absence of the SCE. When all words are equally predictive of a congruent response, the SCE is present, reflecting the sequential updating of weights assigned to the word dimension as a whole. If the local list context manipulates the proportion congruent at the individual-word level, the SCE is absent, reflecting the lack of an appropriate dimension-wide weighting of word information.

We argue that the contribution of these results does not depend on whether the account for the ISPC effect emphasizes an item-specific control mechanism (e.g., Blais et al., 2007) or the learning of contingencies between individual words and their associated responses (e.g., Schmidt & Besner, 2008). Given that the congruency effect for an individual word or color is influenced by the prior occurrence of that word or color, as we observed in the analysis of item-specific sequential adjustments, item properties are continually encoded regardless of the local list context (e.g., Hommel, 2004). However, if the SCE reflects individuals’ attempt to arrive at a more general, dimension-level weighting of word and color information, this seems very consistent with a control-like function. In this case, the goal is to modulate the contribution of distinct processing streams, and such a weighting applies across specific values (e.g., words) on a dimension (e.g., Cohen, Dunbar, & McClelland, 1990). Dimension-level weighting does not require the detailed representation of individual word or color histories, nor does it require the rapid modification of such control settings on the basis of these individual histories.

Previous work (e.g., Schmidt & De Houwer, 2011) has revealed the SCE when contingency biases are present, but not when contingency biases are absent. Note that both the item and balanced blocks in the present experiments were contingency-biased. However, the nature of the contingency varied. In item blocks, for example, the conditional probability p(“blue” response | “blue” word) could differ from p(“red” response | “red” word). In contrast, in balanced blocks, these two contingencies were equal. Our argument is that the sequential modulation in balanced blocks reflected the sequential updating/tracking of this contingency. The heterogeneity of this contingency information in item blocks discouraged such tracking, because only the item-specific contingency was useful for predicting the appropriate response.

To the extent that the SCE reflects the operation of control processes, these processes engage when the local context is conducive to learning dimension-level contingencies, and they disengage in the absence of consistent dimension-level contingencies. This emphasizes a close connection between control processes and the learning that these processes enable.