Distributional analyses reveal the individual differences in congruency sequence effect

Dandan Tang; Xuefei Chen; Hong Li; Yi Lei

doi:10.1371/journal.pone.0272621

. 2022 Aug 22;17(8):e0272621. doi: 10.1371/journal.pone.0272621

Distributional analyses reveal the individual differences in congruency sequence effect

Dandan Tang ^1,^*, Xuefei Chen ², Hong Li ³, Yi Lei ³

Editor: Ludovic Ferrand⁴

PMCID: PMC9394795 PMID: 35994475

Abstract

As a sequential modulation of conflict, congruency sequence effect indexes a conflict-induced performance improvement, which is observed as reduced congruency effects for trials after the incongruent trials than for trials after the congruent trials. Although congruency sequence effect has been investigated widely in healthy humans, the studies of distributional characteristics across prototypical congruency tasks are scarce. To investigate this issue, the present study adopts the between-subjects design to carry out three experiments, where subjects were separately informed to perform the Stroop, word Flanker, and letter Flanker tasks. The results showed that congruency sequence effect occurred in the congruent and incongruent trials in the Stroop and word Flanker tasks, respectively, and absented in the letter Flanker task, which is interpreted as the differences in the nature and difficulty of the tasks. The distributional properties of congruency sequence effect did not significantly differ from the Gaussian distribution in the Stroop and word Flanker tasks, but not in the letter Flanker task, suggesting the inter-individual variability of congruency sequence effect depends on the nature of tasks. Importantly, the delta plot analyses showed pronouncedly increased congruency sequence effect over the slowest percentile bines in both the Stroop and word Flanker tasks, verifying the activation suppression hypothesis. Altogether, the present study enriches the literature on the distributional characteristics of congruency sequence effect.

Introduction

The incongruence relative to congruence between target and distractor in the congruency tasks [1], e.g., the Stroop task [2], Flanker task [3], and Simon task [4], usually results in longer response times (RTs) and higher error rates in human. Such incongruence activates the cognitive control system to recruit the mechanisms of cognitive control to improve performance [5–9]. As a typically sequential modulation of performance, congruency sequence effect, which is thought to encompass two basic cognitive processes: conflict monitoring and control implementation [7, 8, 10], denotes such a conflict-induced performance improvement, which is observed as reduced congruency effects in the current trials after an incongruent trial as compared with a congruent trial in the congruency tasks [11–19]. The congruency effects are usually defined as the performance differences between the incongruent and congruent trials.

For example, in the Stroop task, stimuli comprise words with congruent or incongruent semantics with the printed color (e.g., BLUE is printed in blue font for the congruent condition, or RED is printed in blue font for the incongruent condition). Participants are informed to respond to the printed color and to ignore the semantics of the words. And in the letter Flanker task, stimuli consist of strings of horizontal-arranged letters with a central target letter compatible or incomparable with the flanker letters. Participants are informed to respond to the target letter and to ignore the flanker letters. The congruency effects can be indexed by longer RTs and/or higher error rates in the incongruent (I) relative to congruent (C) trial, which are calculated by incongruent minus congruent (I-C) in RTs or error rates. The congruency sequence effect can be quantified by the following formula: congruency sequence effect = RTs_{[(cI–cC)–(iI–iC)]} [20, 21], where cC is a congruent trial followed by a congruent trial, cI is a congruent trial followed by an incongruent trial, iC is an incongruent trial followed by a congruent trial, and iI is an incongruent trial followed by an incongruent trial. Accordingly, congruency sequence effect is revealed by smaller congruency effects in the current trials preceded by an incongruent relative to congruent trial, i.e., RTs_(iI–iC) < RTs_(cI–cC) [11, 14, 21–23].

The conflict monitoring theory [5, 10] proposes a top-down attentional control explanation of congruency sequence effect, which supposes that congruency sequence effect stems from a conflict-driven continuously adjustment in cognitive control [8, 14, 24–27]. This assumption has been supported by the neurophysiological data. That is, a conflict monitoring system (anterior cingulate cortex, ACC) on the medial surface of the frontal lobes of the brain monitors for the occurrence of conflict [5, 28–31]. Subsequently, the ACC exerts an influence on the dorsolateral prefrontal cortex (DLPFC), which triggers strong cognitive control processes and results in a behavioral adjustment [32–35].

Recently, individual differences in congruency sequence effect in patients with Parkinson’s disease [36, 37] and traumatic brain injury [38] or in healthy human have been studied [39–41]. In these studies, the behavioral data revealed comparable size of the congruency sequence effect between the experimental and control groups across the Stroop and Flanker tasks. Specifically, Wang et al. evaluated the regional homogeneity (ReHo) of resting-state functional magnetic resonance imaging (RS-fMRI) signals to explore the neural basis of individual differences in congruency sequence effect in the letter Flanker task [42]. They found significant positive correlations between the size of congruency sequence effect and the left DLPFC and left ventrolateral prefrontal cortex (VLPFC) ReHo values, suggesting that participants exhibiting greater ReHo in these regions tended to display a greater degree of congruency sequence effect. Importantly, they found that the ReHo values in the DLPFC predicted behavioral performance related to congruency sequence effect. Then, Wang et al. employed a psychophysiological interaction (PPI) analysis within the Stroop task to examine the functional connectivity mechanisms underlying the individual differences in congruency sequence effect [43]. The results found that high congruency sequence effect participants showed higher intra-CEN (central executive network) connectivity and lower intra-SN (salience network) connectivity; while low congruency sequence effect participants showed higher intra-SN connectivity and lower intra-CEN connectivity. Another study had reported that different activations in the prefrontal regions reflected the individual differences in congruency sequence effect [44]. Therefore, these studies reveal the underlying neural mechanisms of individual differences related to congruency sequence effect.

However, congruency sequence effect is traditionally obtained from the across-trial average of RTs_{[(cI–cC)–(iI–iC)]}, resulting in (1) the inter-trial differences in congruency sequence effect are not detectable, and (2) the true effects are counteracted. In this case, the analysis of congruency sequence effect in single-trial level can provide the information on the inter-trial variability of congruency sequence effect and can help us to observe the true effects when assessing the inter-individual differences of congruency sequence effect. Usually, the delta plotanalysis can be used to investigate the inter-trial variability of the congruency effects and to reveal the inter-individual differences in the congruency effects [45–48]. However, it is unclear how congruency sequence effect varies with the RTs in the delta plot analysis. In addition, the Gaussian distribution model may demonstrate typical inter-individual differences in a data set. However, it is unclear if the distributional properties of congruency sequence effect significantly differ from Gaussian distribution across tasks.

Thus, the present study is designed to investigate the distributional characteristics of congruency sequence effect in healthy human across the Stroop [2], word Flanker [49, 50], and letter Flanker tasks [3] in three Experiments, and to reveal both the inter-trial and the inter-individual variability of congruency sequence effect. In Experiment 1, 122 heathy human participants were informed to perform the typical Stroop task [2]. In Experiment 2, 120 heathy human participants were informed to perform the word Flanker task [49, 50]. In Experiment 3, 124 heathy human participants were informed to perform the letter Flanker task [3]. In all three experiments, the proportion of the congruent and incongruent trials was equal; the confounding of both the bottom-up repetition effects [51, 52] and negative priming effects [53] was excluded. In this way, we expected to obtain a pure congruency sequence effect. Meanwhile, the delta plots [54] were adopted to examine how congruency sequence effect evolves with the RT distributions. We also would fit the Gaussian distribution to the RTs_{[(cI–cC)–(iI–iC)]} data to reveal the distributional patterns of congruency sequence effect. Although the variability of individual differences in congruency sequence effect had been observed in healthy human across the Stroop and Flanker tasks [39–43], finding the inter-trial and inter-individual variability of the congruency sequence effect across these congruency tasks in a study may provide evidence for the generality and robustness of the observations.

It is notable that in the Stroop task, the word reading and color naming are processed in the different processing pathway and the word reading shows a processing advantage over the color naming [55, 56] according to the parallel distributed processing model [57–59]. On the contrary, in the word Flanker and letter Flanker tasks, both target and flankers are processed in the same pathway and, therefore, the two dimensions always overlap with each other and the flanker dimension has no processing advantage over the target dimension according to the dimensional overlap theory [60, 61]. Additionally, the stimuli consist of Chinese characters and letters in the word Flanker and letter Flanker tasks, respectively. In this case, the nature of tasks and the processes of stimuli are different in all three tasks. Thus, we forecast that the pattern of congruency sequence effect will vary across tasks. In the Stroop task, congruency sequence effect will be determined by the congruent condition according to the previous reports [21, 22]; in the word Flanker and letter Flanker tasks, the patterns of congruency sequence effect may differ from that in the Stroop task due to the differences in the nature of task. That is, congruency sequence effect may be determined by the incongruent condition or may be absent in the word Flanker and letter Flanker tasks. In addition, we expect the Gaussian model will well fit the RT distributional characteristics of congruency sequence effect. According to the activation suppression hypothesis [48], we predict that congruency sequence effect increases over the slowest percentile bines. Thus, the present study findings will provide substantial evidence verifying both the inter-trial and the inter-individual variability of the congruency sequence effect for the first time.

Experiment 1

Method

Participants

122 healthy right-handed volunteers (45 women), between 18 and 26 years old (19.99±1.19, mean±SD, similarly hereinafter), took part in the experiment. All volunteers reported normal or corrected-to-normal vision and normal color perception. All volunteers provided written informed consent and were paid for their participation. The Zunyi Normal College Human Ethics Committee (Zunyi, China) approved the experimental procedure, which were in accordance with the standards of the Declaration of Helsinki. In addition, the volunteers were blind to the experimental design.

Apparatus, stimuli, and task

The stimuli were presented in white against a black background using E-Prime 2.0 software (Psychology Software Tools, Inc., Learning Research and Development Center, University of Pittsburgh) on a 17-in Lenovel computer monitor. Responses were registered using a standard QWERTY keyboard. Participants sat approximately 60 cm away from the screen.

The stimuli consisted of four words (RED, GREEN, YELLOW, and BLUE; in Chinese with Song Ti font, size 28) that were printed in a red, green, yellow, or blue font. The RGB values for the stimulus colors were 255, 0, 0 (red); 0, 255, 0 (green); 255, 255, 0 (yellow); and 0, 0, 255 (blue), respectively. For the congruent condition, the color and word matched (e.g., the word “GREEN” was in green font); for the incongruent condition, the color and the word didn’t match (e.g., the word “RED” was in green font).

Participants were instructed to response to the printed color of word as quickly and accurately as possible, and to ignore the semantics of word. They were instructed to press the “D” key using the left middle finger if the color of the word was red, the “F” key using the left forefinger if the color of the word was green, the “J” key using the right forefinger if the color of the word was yellow, and the “K” key using the right middle finger if the color of the word was blue. The rules of stimulus-response button mapping were counterbalanced across the participants.

Procedure and design

The experimental procedure is illustrated in Fig 1A. In each trial, the stimuli were presented as follows: (1) a white fixation ‘+’ for 500 ms; (2) a blank interval for 300–500 ms (the interval varied randomly); (3) a colored word until a key was pressed or for 1.5 s if there was no response registered; and (4) a blank interval for 800–1200 ms (interval varied randomly). Presentation order of the trials was pseudo-randomized by using experimental manipulation, which excluded the confounding of bottom-up repetition effects [51, 52] including the stimulus-stimulus, stimulus-response, and response-response repetitions. And only the trial sequences with the stimulus-stimulus, stimulus-response, and response-response changes were remained.

Participants performed a block of sixty-four practice trials prior to the completion of the formal experiment, which acquainted them with the experimental stimulus-response mapping. In the formal experiment, 576 trials were separated into four blocks of 144 trials each. In each experimental block, each of the congruent and incongruent conditions contained 72 trials. To assess congruency sequence effect, the stimuli were separated into the cC, cI, iC, and iI conditions according to the congruency (‘c’ was congruent, ‘i’ was incongruent) of the previous trial and that of the current trial. Each of the cC, cI, iC, and iI conditions contained 36 trials for each experimental block. There was a 2-minute break between blocks to minimize the effect of fatigue.

Results

The RTs and error rates were analyzed using the IBM SPSS Statistics 20.0 software. For the analysis of the RTs, all incorrect trials (5.86% of all trials), post-error trials (5.23% of all trials), and RT outliers (±3 SDs, 0.76% of all trials) were excluded. The general linear model (GLM) was used to compare the mean RTs for each condition. The two-way repeated-measures analysis of variance (ANOVA) was conducted for the mean RTs with the following variables: the congruency (congruent, incongruent) in the previous trials and the congruency (congruent, incongruent) in the current trials, to reveal the essence of congruency sequence effect. The significance level (p-value) was corrected using a Bonferroni method to control for the problem of multiple comparisons. Fig 2A illustrates the mean RTs in the current trials. Table 1 (the left column) shows the two-way repeated-measures ANOVA results for the mean RTs.

Fig 2 — Panels **A, B, & C** illustrate the mean RTs in the current trial, which was calculated according to the congruency (congruent, incongruent) in the previous and current trials in Experiments 1, 2, & 3, respectively. Panel A shows a significant congruency sequence effect in the current congruent trials in Experiment 1. Panel B shows a significant congruency sequence effect in the current incongruent trials in Experiment 2. Notably, the RTs_(iI–iC) was significantly lower than the RTs_(cI–cC) in Experiments 1 and 2. Panel C shows a non-significant congruency sequence effect in Experiment 3. For each condition, error bars represent ±SEM across the participants. NB. ‘c’ and ‘C’ are congruent condition, ‘i’ and ‘I’ are incongruent condition, RTs is response times.

Table 1. Test statistics for the mean RTs in Experiments 1, 2, & 3.

The main effect of congruency in the previous trials (congruent, incongruent)
Experiment 1			Experiment 2			Experiment 3
F(1,121)	p	η _p ²	F(1,119)	p	η _p ²	F(1,123)	p	η _p ²
17.80	< .001	.128	7.90	.006	.062	<1	.484	.004
The main effect of congruency in the current trials (congruent, incongruent)
Experiment 1			Experiment 2			Experiment 3
F(1,121)	p	η _p ²	F(1,119)	p	η _p ²	F(1,123)	p	η _p ²
1088.86	< .001	.900	780.31	< .001	.868	236.14	< .001	.659
The interaction between the two factors
Experiment 1			Experiment 2			Experiment 3
F(1,121)	p	η _p ²	F(1,119)	p	η _p ²	F(1,123)	p	η _p ²
4.57	.035	.036	8.27	.005	.065	< 1	.960	< .001
Post hoc test (Bonferroni-corrected)
Experiment 1				Experiment 2
cC < iC, p < .001				iI < cI, p < .001
cC < cI, p < .001				cC < cI, p < .001
iC < iI, p < .001				iC < iI, p < .001

Open in a new tab

Typically, the repeated-measures ANOVAs are based on the average of each condition for each participant. In this case, the inter-trial variability is completely blinded to the estimation of the significance of congruency sequence effect. Thus, to reveal the inter-trial variability of congruency sequence effect and to test the random factors, we used the linear mixed models to compare the RTs on the single-trial level of each condition of each participant, which included the congruency (congruent, incongruent) in both the previous and current trials as the fixed effects, and the gender (female, male) and age (continuous variable) as the random factors. The Bonferroni method was used to correct the significance level (p value) when carrying out the multiple comparisons procedure. The results are showed in Table 2 (the left column). It is notable that the results of the single-trial RTs are in accord with those of the mean RTs.

Table 2. Test statistics for the RTs on the single-trial level of each condition of each participant in Experiments 1, 2, & 3.

The main effect of congruency in the previous trials (congruent, incongruent)
Experiment 1		Experiment 2		Experiment 3
F(1,64024.45)	P	F(1, 62403.19)	p	F(1, 62446.01)	p
19.37	< .001	5.35	.021	<1	.619
The main effect of congruency in the current trials (congruent, incongruent)
Experiment 1		Experiment 2		Experiment 3
F(1, 64024.49)	P	F(1, 62403.23)	p	F(1, 62446.00)	p
4902.81	< .001	1285.49	< .001	335.78	< .001
The interaction between the two factors
Experiment 1		Experiment 2		Experiment 3
F(1, 64024.14)	P	F(1, 62403.13)	p	F(1, 62818.89)	p
4.75	.029	7.01	.008	< 1	.878
Pair-wise comparisons (Bonferroni-corrected)
Experiment 1			Experiment 2
cC < iC, p < .001			iI < cI, p < .001
cC < cI, p < .001			cC < cI, p < .001
iC < iI, p < .001			iC < iI, p < .001

Open in a new tab

To demonstrate the reduced congruency effects in the current trial preceded by the incongruent compared with congruent trials, we calculated the differences of mean RTs in the current trials according to the congruency (c, i) in the previous trials, i.e., RTs_(iI–iC) and RTs_(cI–cC). And then, the RTs_(iI–iC) and RTs_(cI–cC) were compared within individuals using the paired-samples T test (2-tailed, Bonferroni correction, similarly hereinafter). The key observation revealed that RTs_(iI–iC) (100.01±36.90) was significantly smaller than RTs_(cI–cC) (106.50±39.85), t(121) = -2.14, p = .035, Cohen’s d = .176 [62].

The size of congruency sequence effect was calculated according to the following formula (similarly hereinafter) [20, 21], congruency sequence effect = RTs_{[(cI–cC)–(iI–iC)]}. To test the distributional properties of congruency sequence effect across participants, we performed the normality test by using the Shapiro-Wilk Test for RTs_{[(cI–cC)–(iI–iC)]}. The results did not show significant difference from Gaussian distribution, p = .827 (Fig 3A). To estimate the reliability of congruency sequence effect, we performed a reliability analysis by calculating the split-half reliability coefficient for the size of congruency sequence effect. We firstly divided all trials into the odd and even trials, respectively. And then we calculated the size of congruency sequence effect in the odd and even trials, respectively. Lastly, the Guttman split-half coefficient was calculated. The results revealed that the split-half reliability coefficient was 0.68. To estimate the time course of congruency sequence effect, we conducted the RTs distribution analysis by using delta plots [45, 47, 48, 54, 63]. The results are showed in Fig 4A.

Fig 3 — Panels **A, B, & C** illustrate the frequency distribution of congruency sequence effect in Experiments 1, 2, & 3, respectively. The size of congruency sequence effect was calculated by the RTs_{[(cI–cC)–(iI–iC)]}. In each Experiment, the sizes of congruency sequence effect were divided into twenty-five bins according to the Gaussian distribution model.

Fig 4 — Panels **A, B, & C** illustrate the time courses of congruency sequence effect in Experiments 1, 2, & 3, respectively. The congruency sequence effect, RTs_(cI–cC), and RTs_(iI–iC) were calculated over five quantiles in RTs of each participant and condition. Panel A depicts the time course of congruency sequence effect in the Stroop task. Noteworthy is that congruency sequence effect, RTs(_cI–cC), and RTs_(iI–iC) increased across all five quantiles. Panel B depicts the time course of congruency sequence effect in the word Flanker task. The negative slope between the first two quantiles indicated the decreased congruency sequence effect over the fastest bins; but the positive slope in the last three quantiles indicated the increased congruency sequence effect over the slowest bins. The positive slope between the first two quantiles indicated the increased congruency effects in RTs_(cI–cC) and RTs_(iI–iC) over the fastest bins; but the negative slope between the last two quantiles indicated the decreased congruency effects in RTs_(cI–cC) and RTs_(iI–iC) over the slowest bins. Panel C depicts the time course of congruency sequence effect in the letter Flanker task. The congruency sequence effect decreased with the increasing mean RTs, which was most pronounced between the last two quantiles. The RTs_(cI–cC) increased between the first two quantiles, but decreased in the last three quantiles. The RTs_(iI–iC) increased between the first two quantiles, but kept stable in the last three quantiles. NB. ‘cC’ and ‘iC’ are congruent condition preceded by congruent and incongruent conditions, respectively; ‘cI’ and ‘iI’ are incongruent condition preceded by congruent and incongruent conditions, respectively; RTs is response times; and CSE is congruency sequence effect calculated by the RTs_{[(cI–cC)–(iI–iC)]}.

The generalized estimating equations were used to measure the error rates in the cC, cI, iC, and iI conditions with the following variables: the congruency (congruent, incongruent) in the previous trials, and the congruency (congruent, incongruent) in the current trials, of which age and gender were used as the covariates, which avoided the bias of spurious interaction in the estimation of the error rates [64]. The Bonferroni method was used when adjustment for multiple comparisons. The tests of model effects showed (1) significant main effect in the current trials, Wald Chi-Square = 205.30, p < .001; and (2) significant interaction effect between the congruency in the current trials and that in the previous trials, Wald Chi-Square = 4.43, p = .035. Pairwise comparison results showed significant lower error rates (1) in the cC relative to both cI and iI trials; and (2) in the iC relative to both cI and iI trials, each p < .001 (Bonferroni-corrected). The paired-samples T test showed that the congruency effects in the current trials preceded by the incongruent trials (iI–iC = 0.03±0.03) were significantly smaller than those preceded by the congruent trials (cI–cC = 0.04±0.04), t(121) = -2.10, p = .038, Cohen’s d = .244 [62]. Fig 5A illustrates the error rates in Experiment 1.

Fig 5 — Panels **A, B, & C** illustrate the error rates in the current trials, which were calculated according to the congruency (congruent, incongruent) in the previous and current trials in Experiments 1, 2, & 3, respectively. As shown in Panel A, the congruency effects in the current trials preceded by the congruent trials (cI–cC) were significantly larger than those preceded by the incongruent trials (iI–iC), which is accordant with the results of RTs. Panels **B & C** show comparable congruency effects in the current trials preceded by both the congruent and incongruent trials. For each condition, error bars represent ±SEM across the participants. NB. ‘c’ and ‘C’ are congruent condition and ‘i’ and ‘I’ are incongruent condition.

Discussion

The present experiment found significant congruency sequence effect in the typical Stroop task. The results revealed typical pattern of congruency sequence effect in the current congruent condition, where the RTs were significantly shorter for the cC compared with iC condition in both the across-trial average level and single-trial level of each condition and participant (Tables 1 and 2). Importantly, the RTs_(iI–iC) were significantly smaller than the RTs_(cI–cC). These results replicated and extended several previous observations, i.e., congruency sequence effect was determined by the congruent condition in the current trials [21, 22, 65, 66]. Additionally, the differences of error rates in (iI–iC) were significantly smaller than those in (cI–cC), which are in accord with the results of RTs.

In the typical Stroop task, the target and distractor stimuli are processed in two separated pathway: the word reading and color naming pathways. The word reading is automatically processed while the color naming is controllably processed. The controlled process is related to task demand, and is affected by the automatic process. In this case, the word reading has a processing advantage over the color naming [55, 56] according to the parallel distributed processing model [57–59]. Hence, the smaller congruency effects preceded by the incongruent compared to congruent trials may be derived from that the processing advantage of word reading over the color naming is more enhanced in the cC relative to iC condition.

The distribution curve of congruency sequence effect did not significantly differ from a normal distribution (Fig 3A), which verified the inter-individual differences in congruency sequence effect. The Guttman split-half coefficient was 0.68, which suggested the moderate reliability of congruency sequence effect was [67]. Although other random factor may contribute to the inter-individual variability of congruency sequence effect, the Gaussian distribution of congruency sequence effect reflected some meaningful inter-individual variability. The congruency sequence effect increased over the slowest percentile bines in the delta plots (Fig 4A), suggesting the effective selective inhibition processes in these slowest trials and confirming the activation suppression hypothesis [48]. Thus, these results demonstrated the inter-trial and inter-individual variability of congruency sequence effect in the typical Stroop task.

Nevertheless, we also wonder if the pattern of variability of individual differences will be observed in other congruency tasks, where both the target and distractor are processed within the same pathway. Therefore, Experiment 2 adopted the word Flanker task [49, 50] to address the question and to expand the findings of Experiment 1. We used the experimental procedure and design that were same as those used in Experiment 1. In the word Flanker task, both target and flankers are processed in the same pathway and the target processing dimension always overlaps with the flanker processing dimension. The flanker processing dimension has no processing advantage over the target processing dimension according to the dimensional overlap theory [60, 61]. In this case, both the nature of task and processes of stimuli are different between the Stroop and word Flanker tasks. Thus, we anticipate the pattern of congruency sequence effect will differ from that in Experiment 1, and will be embodied in the current incongruent trials. However, we expect the other results will be consistent with those in Experiment 1 due to the same experimental design between two Experiments.