The final number of participants who met the inclusion criteria was 19 CWS (mean age 7.58 years; SD = 1 year; range = 6.08–9.17) and 19 CWNS (mean age 7.58 years; SD = 1.08 years; range = 6.08–9.33). The two groups were matched for gender (18 males and 1 female) and age (±3 months); the between-group age comparison was t (36) = 0.11, p = 0.91. Specific inclusion criteria for CWS were (a) a diagnosis of developmental stuttering by a speech-language pathologist, (b) a severity equivalent of at least “mild” on the Stuttering Severity Instrument-4 (SSI-4; Riley, 2009), and (c) no history of speech and/or language therapy other than for stuttering. The inclusion criteria for CWNS were (a) no parental concern regarding stuttering, (b) a severity equivalent of less than “mild” on the SSI-4, and (c) no history of speech and/or language therapy. Seven CWS (five males) and three CWNS (all males) were excluded from the study because they did not meet at least one of the inclusion criteria. Additional eligibility criteria for participating in the study (collected using parental questionnaires and interviews with the teachers) were the following: (a) monolingual Greek speakers with no other languages being spoken in their homes, (b) no known or questionnaire-reported psychological, neurological, or developmental problems, except for stuttering in the CWS group, and (c) normal hearing and normal or corrected-to-normal vision.

The parental educational level was determined based on the sum of the scores of the highest educational level of each of the two parents (primary education = 1, high school = 2, 1-or 2-year vocational college certificate = 3, university degree = 4), as in a previous publication (Eggers et al., 2018). According to Rindermann and Baumeister (2015), parental educational behavior is more closely related to a child’s cognitive development than any other factor that determines socioeconomic status. A Mann–Whitney test (scores were not normally distributed) revealed no differences between the CWS (mean = 6.47, range = 3–8) and the CWNS (mean = 7.26, range = 4–8), U = 133.50, p = 0.17.

The data collection took place in Cyprus and was conducted by the first author, an ASHA certified speech-language pathologist, and EU-certified fluency specialist. Only children attending the mainstream school system with no teacher-or SLP-reported speech, language, and/or learning difficulties were included. None of the participants had received any speech-language therapy, other than for stuttering for the CWS. There are no standardized language tests for school-aged Greek-speaking children. Therefore, we administered the Bus Story Test, which is widely used in its unstandardized form in studies with Greek-speaking populations (e.g., Theodorou et al., 2016). We compared the two groups on 15 measures such as information [CWS: mean = 39, SD = 8.41 and CWNS: mean = 43, SD = 9.52, t(36) = −1.37, p = 0.18], subordinate clauses [CWS: mean = 9.05, SD = 4.02 and CWNS: mean = 8.53, SD = 3.96, t(36) = 0.41, p = 0.69], mean length of utterance (CWS: mean = 6.10, SD = 0.95 and CWNS: mean = 6.20, SD = 1.75, U = 163.50, p = 0.62), T-units (CWS: mean = 20.74, SD = 3.83 and CWNS: mean = 22.63, SD = 6.09, U = 151.00, p = 0.40), total number of words (CWS: mean = 125.05, SD = 27.02) and CWNS: mean = 133.11, SD = 23.79, t(36) = −1.09, p = 0.28, and no significant differences occurred between the two groups.

Stuttering severity was assessed through the SSI-4 for which we collected four speech samples from each participant on two different days to obtain a reliable representation of stuttering severity. Scores on the SSI-4 were calculated based on a total sample of minimum of 300 words. The following were considered “stuttering-like disfluencies”: (a) part-word repetitions (e.g., b-but), (b) single-syllable word repetitions (e.g., and and), (c) dysrhythmic phonation (specifically prolongation, e.g., “mmmmy” “cooookie”), blocks (e.g., “#table”), and broken words (e.g., “o#pen”; Ambrose and Yairi, 1999). Seven CWS were classified as mild, nine as moderate, two as severe, and one as very severe. CWS had an average of 6.36 (SD = 3.00) stuttering-like disfluencies per 100 syllables, while CWNS had an average of 0.83 (SD = 0.42).

Participants’ hearing was screened using bilateral screening tone-audiometry at 500, 1000, 2000, and 4,000 Hz (Amplivox 240, United Kingdom) with signals presented at 20 dB. All participants passed the hearing screening. Hand preference was determined using parental questionnaires. Two CWS and two CWNS were reported to be left-handed, while all other participants were right-handed.

To preclude cognitive group differences, the vocabulary and block design subtests of the Wechsler Intelligence Scale for Children-Third Edition (Wechsler, 1997) were administered. This test consists of seven performance and six verbal subtests. Vocabulary (verbal) and block design (performance) subtests were chosen because previous studies have shown that the scores of these two subtests correlate well with the overall score of the test (Groth-Marnat, 1997). For the vocabulary subtest, participants are required to explain the meaning of single words, while in the block-design subtest, participants are required to rebuild, as quickly as possible, a geometrical pattern with the use of four to nine two-colored cubes. The mean vocabulary standard scores were 11.80 for the CWS and 14 for the CWNS. The mean block-design standard scores were 12 for the CWS and 12.89 for the CWNS. No significant group differences were found in the vocabulary subtest (U = 114.50, p = 0.053) nor in the block design subtest (U = 148, p = 0.35), although four CWS scored below the standard score 10 (5,8,8,6) in the WISC-3 test (Vocabulary subtest) that was also administered.

Two tasks from the ANT (de Sonneville, 2009) were used: the baseline speed task and Response Organization Objects (ROO). The ANT is a neuropsychological battery that consists of 38 tasks. It is designed to evaluate different EF processes by measuring the participants’ speed and accuracy of manual responses in different tasks. Over the last years, researchers have used ANT tasks to evaluate EF processes related to developmental stuttering (Eggers et al., 2013; Piispala et al., 2016; Eggers and Jansson-Verkasalo, 2017) as well as in relation to other disorders (Stoit et al., 2013; Sterken et al., 2016; Schuiringa et al., 2017). Several studies have shown that the ANT has good validity (de Sonneville, 1999, 2005; Polderman et al., 2007; Oerlemans et al., 2016) and test–retest reliability (Huijbregts et al., 2002; Oerlemans et al., 2016).

All participants were volunteers. They were recruited through an open call to participate in the study, sent to all private and public schools under the jurisdiction of the Cyprus Ministry of Education, Culture, Sports, and Youth; the open call was also provided to all registered speech-language pathologists on the island. The participants were tested in rooms where sounds and other distractions were minimal. Stimuli were presented on an LG laptop R510 with a 15.4. screen, and the laptop was placed on a table in front of a plain wall, approximately 18 in from the participant. Participants were seated facing the screen. Data collection required two sessions of 35to45minutes each, and tests were always administered in the same order. All procedures were approved by the Cyprus National Bioethics Committee and the Center of Educational Research and Evaluation of the Cyprus Pedagogical Institute. Participation consent forms were collected for all participants.

Prior to testing the hypotheses, a Mann–Whitney test was used to determine whether there were any between-group response time differences in the baseline speed task that would have confounded the results of the ROO task. This test was selected over a t-test because the response times of the two groups were not normally distributed.

To compare the efficiency of complex response inhibition between the groups, two pairs of mixed analyses of covariance (ANCOVA) were used, with response times and error percentages as their dependent variable. In both analyses, group (CWS vs. CWNS) was set as the between-subjects variable, block (compatible vs. incompatible) as the within-subject repeated-measures variable, and age as a covariate.

To compare the efficiency of cognitive flexibility between the two groups, two pairs of mixed ANCOVAs with the same dependent variables—response time and error percentage—were conducted. In all the analyses, group was set as the between-subjects independent variable and age as a covariate. Block was set as the within-subjects repeated-measures variable with three levels. For the comparisons between the compatible and mixed block (compatible part), the levels were (1) compatible block, (2) compatible trials of the mixed block without set-shifting, and (3) compatible trials of the mixed block with set-shifting. For the comparisons between the incompatible and mixed block (incompatible part), the levels were (1) incompatible block, (2) incompatible trials of the mixed block without set-shifting, and (3) incompatible trials of the mixed block with set-shifting. The analyses were repeated with the Vocabulary subscores of the WISC-3 as a covariate, but because results were comparable with the results of the analyses with age as a covariate, they are not reported.

For each of the above analyses, the normality assumption was checked by subjecting the residuals to a Shapiro–Wilk test; if the residual distribution for any dependent variable was not normal, appropriate data transformation was performed based on the observed shape of the variable’s distribution. For all analyses, the residuals for error percentage deviated from normality; the residuals for response time were normally distributed except for the residuals of the incompatible block. Skewed data were transformed by a power transformation. After transformation, all residuals followed a normal distribution for all response times (except for the incompatible block in the analysis comparing the compatible with the incompatible block). For the error percentages, residuals did not follow a normal distribution. Therefore, the results for the latter should be treated with caution, although ANOVA is known to be robust to violations of normality (Field, 2009).

To detect possible outliers, the distributions of the dependent variables (response times and error percentages) were checked with the criterion of ±3 standard deviations from the mean (Howell, 1998). Outlier detection was also performed on mean response times (cf. Eggers and Jansson-Verkasalo, 2017). No outliers were detected in any of the analyses.

The effects of the age covariate were investigated with correlation tests between age and change in response times (hereafter performance-cost) from the compatible block to the incompatible or the mixed block, or from the incompatible block to the mixed block. The correlations were performed using Spearman’s correlation analysis when the distribution of the variables deviated from normality or Pearson’s correlation analysis when the variables followed a normal distribution.

Based on our hypotheses, the main effects and interactions in all ANCOVAs were planned. Correlations with age were explored by two-tailed post hoc tests. Additionally, the relation between stuttering severity and performance-cost was investigated with Kendall rank correlation analysis. In these cases, Bonferroni correction was applied for multiple comparisons. The significance level for all analyses was α = 0.05. Data analysis was conducted using SPSS (Statistical Package for the Social Sciences – Version 25 for Windows, IBM, Corp., Armonk, NY, United States).

For the comparison of response times between the compatible and incompatible block, the between-subjects factor of group was not significant, F(1, 35) = 0.34, p = 0.57, partial η² = 0.01. However, the block × group interaction was significant, F(1, 35) = 4.53, p = 0.04, partial η² = 0.11. Further investigation revealed that, while both groups were slower in the incompatible than in the compatible block, CWS slowed down significantly more than CWNS, t(24.09) = −2.08, p = 0.049, Cohen’s d = −0.67, something that indicates higher performance-cost for the CWS (see Figure 2A). The effect of age was also significant, F(1, 35) = 10.75, p < 0.001, partial η² = 0.23. A two-tailed Spearman’s correlation analysis indicated that the change in response times from the compatible to the incompatible block was negatively correlated with age for the CWS group, ρ = −0.49, p < 0.005, but not for the CWNS group, ρ = −0.24, p = 0.30. Nevertheless, there was a substantial overlap in the 95% confidence intervals between CWS [−0.772, −0.046] and CWNS [−0.626, 0.240].

For the comparison of error percentages between the compatible and the incompatible block, the between-subjects factor of group was not significant, F(1, 35) = 2.50, p = 0.12, partial η² = 0.07. The block × group interaction was not significant, F(1, 35) < 0.001, p = 1.00, partial η² < 0.001 (Figure 2B). The effect of age was not significant, F(1, 35) = 0.32, p = 0.58, partial η² = 0.01.

For the comparison of response times between the compatible and the mixed block (compatible part), the between-subjects effect for the group factor was not significant, F(1, 35) = 0.60, p = 0.44, partial η² = 0.02. However, the block × group interaction was significant, F(1.54, 54.02) = 5.16, p = 0.01, partial η² = 0.13. Further investigation revealed that, while both groups were slower in the mixed block (compatible part) than in the compatible block, CWS slowed down significantly more than CWNS, something that indicates higher performance-cost for the CWS (Figure 2C). Higher performance-cost for the CWS compared to the CWNS was present both when there was no set-shifting (i.e., a compatible trial preceded a compatible trial), t(21.71) = −2.23, p = 0.04, Cohen’s d = −0.08, and when there was set-shifting (i.e., an incompatible trial preceded a compatible trial), t(23.71) = −2.44, p = 0.02, Cohen’s d = 0.47. The effect of age was significant F(1, 35) = 13.19, p < 0.001, partial η² = 0.27. Two-tailed Pearson’s correlation analyses indicated a significant negative correlation between age and performance-cost from the compatible to the mixed block (compatible part) for the CWS group, r = −0.676, p < 0.005, when there was no set-shifting, and r = −0.623, p = 0.01, when there was set-shifting. For the CWNS group, r = −0.232, p = 0.68, when there was no set-shifting, and ρ = −0.077, p = 1.00, when there was set-shifting. The 95% confidence intervals of the two groups showed substantial overlap both for the no set-shifting, ρ = −.077 (CWS: [−0.87, −0.32]; CWNS: [−0.621, −0.25]) and the set-shifting trials (CWS: [−0.84, −0.24]; CWNS: [−0.51, −0.39]).

For the comparison of response times between the incompatible and the mixed block (incompatible part), the between-subjects effect for the group factor was significant, F(1, 35) = 6.23, p = 0.02, partial η² = 0.15, with the CWS being slower than CWNS. The block × group interaction was significant, F(2, 70) = 3.69, p = 0.03, partial η² = 0.10. This interaction was investigated with a simple main effects analysis that indicated that CWS were slower than CWNS in the mixed block (compatible part) for both the no set-shifting (p = 0.01) and the set-shifting trials (p = 0.02); in the incompatible block, the two groups did not differ in speed, p = 0.14 (Figure 2E). In addition, while both groups were slower in the mixed block (incompatible part) than in the incompatible block, CWS slowed down significantly more than CWNS (i.e., there was higher performance-cost for the CWS group) when there was no set-shifting t(24.18) = −2.18, p = 0.04, Cohen’s d = −0.71; the two groups had comparable performance-cost when there was set-shifting, t(25.87) = −1.96, p = 0.06, Cohen’s d = −0.64. The effect of age was significant, F(1, 35) = 14.16, p < 0.001, partial η² = 0.29. Two-tailed Pearson’s correlation analyses indicated significant negative correlation between age and performance-cost from the incompatible to the mixed block (incompatible part) only for the CWS group when there was set-shifting, r = −0.549, p = 0.03, but not when there was no set-shifting, r = −0.418, p = 0.15. The 95% confidence intervals of the two groups showed substantial overlap both for the no set-shifting (CWS: [−0.73, −0.05]; CWNS: [−0.58, 0.30]), and the set-shifting trials (CWS: [−0.80, −0.13]; CWNS: [−0.76, 0.03]).

For the comparison of error percentages between the compatible and the mixed block (compatible part), no significant differences between the groups were observed, F(1, 35) = 0.31, p = 0.58, partial η² = 0.01. The interaction between block and group was not significant, F(2, 70) = 1.73, p = 0.19, partial η² = 0.05 (Figure 2D). The effect of age was not significant, F = (1, 35) = 0.41, p = 0.52, partial η² = 0.01. The two groups had comparable performance-cost both when an incompatible trial preceded a compatible trial, t(36) = 1.46, p = 0.15, and when a compatible trial preceded the compatible trial, U = 163.50, p = 0.61.

For the comparison of error percentages between the incompatible and the mixed block (incompatible part), no significant differences between the groups were observed, F(1, 35) = 1.42, p = 0.24, partial η² = 0.04. The interaction between block and group was not significant, F(2, 70) = 0.17, p = 0.85, partial η² < 0.005 (Figure 2F). The effect of age was not significant, F = (1, 35) = 0.10, p = 0.76, partial η² < 0.005. The two groups had comparable performance-cost both when a compatible trial preceded the incompatible trials, t(36) = −0.17, p = 0.86, and when an incompatible trial preceded the incompatible trials, U = 157.50, p = 0.50.

The relationship between severity of stuttering and the performance cost in response times for the analyses of complex response inhibition, and of cognitive flexibility was investigated with Kendall rank correlation analyses, which revealed non-significant results (p values between 0.24 and 1.00).

Our first hypothesis was that CWS would be less efficient than CWNS in task conditions that require complex response inhibition as documented by higher response times and higher error percentages. Indeed, there were significant differences between the two groups as a function of task (significant block × group interaction), with CWS slowing down more than CWNS in the incompatible block (complex response inhibition). Contrary to our hypothesis, the two groups did not differ in terms of accuracy. Eggers and Jansson-Verkasalo (2017) using the auditory set-shifting task, with participants of similar age range (6.33–9.83 years old), also found a block × group interaction for complex response inhibition for error percentages but not for speed. The difference between the results may partly be due to the modality used for the stimulus presentation—auditory for the auditory set-shifting task by Eggers and Jansson-Verkasalo (2017) versus visual for the ROO task. Namely, many studies even though not all (Kaganovich et al., 2010), have reported differences in auditory processing between individuals who stutter and who do not stutter (Foundas et al., 2004; Hampton and Weber-Fox, 2008; Liotti et al., 2010; Jansson-Verkasalo et al., 2014; Ismail et al., 2017). The auditory and visual systems differ in terms of how acoustic and visual information are being perceived and processed (e.g., Gomes et al., 2007; Lin et al., 2017). The neural networks engaged in performing auditory and visual tasks are reported to be similar (Bell et al., 2020), but the exact brain regions involved (within the networks) and their contribution may vary according to the stimulus modality (Lee et al., 2014). Furthermore, task-specific differences, such as stimulus presentation time and response type may have also contributed to the different findings between the two tasks. In the current ROO task, the on-screen stimulus presentation time varied between 200 and 6,000 ms, with the stimulus disappearing once a response was given. In the auditory set-shifting task, the stimulus presentation time was shorter, only 100 ms. Also, the rule that had to be applied in the incompatible block of the auditory set-shifting task (press twice after a single tone and once after a double tone) was more demanding compared to that of ROO (press the opposite response key). It could be that CWS, due to these increased demands of the auditory task, were not able to slow down sufficiently to maintain similar accuracy levels as CWNS.

Our hypotheses get support from the study by Anderson and Wagovich (2017), who used two alternation-design tasks with auditory stimuli. They reported CWS exhibiting longer response times as was the case in our study. In contrast to our study, they also reported a lower accuracy for the CWS in the task that targeted complex verbal response inhibition with the use of nonverbal stimuli (animal sounds). The difference between our findings may be due to the domain of stimulus-presentation but also to the task design differences. Additionally, the participants in the Anderson and Wagovich study were preschoolers (3.08–6.08 years old), a period of great development in the EFs but also a period during which CWS were reported to present with impaired sensorimotor learning compared to CWNS (Kim et al., 2020). Finally, our results are corroborating parental questionnaire-based findings (Eggers et al., 2010; Ntourou et al., 2018) as well as Go/NoGo-findings (Piispala et al., 2017, 2018) suggesting weaknesses in inhibitory control in CWS.

In sum, our results showed that CWS were less efficient than CWNS in task conditions that require complex response inhibition as documented by higher response times. We acknowledge that limitations in working memory may have contributed to the slowing down of CWS. Although it may seem more logical to investigate EFs via the auditory domain as it relates more directly to speech and language rather than via the visual domain, the possibility of auditory processing difficulties may interfere with the results, which is not the case when EFs are investigated via the visual domain.

Our second hypothesis was that CWS, compared to CWNS, would be slower and less accurate under cognitive flexibility task conditions. We have evaluated cognitive flexibility by conducting two sets of analyses: (a) compatible block vs. mixed block (compatible trials), and (b) incompatible block vs. mixed block (incompatible trials). In the first set of analyses, comparing the compatible trials, a significant block × group interaction was observed for response times (an index of speed), but not for error percentages. In the second set of analyses, comparing the incompatible trials, as hypothesized, differences were more evident than in the first set of analyses (significant group differences), with CWS being slower than CWNS.

Our findings are partially in agreement with Eggers and Jansson-Verkasalo (2017) who reported limitations for CWS in terms of accuracy but not in terms of speed. In both studies, a mixed-block design task was used, and therefore, comparisons can be made more safely. Our results are consistent with the findings of Eichorn et al. (2018), Anderson et al. (2020), and Eichorn and Pirutinsky (2021), who all reported longer response times for CWS. In the first two studies, alternation-design tasks were used with preschool age children and despite the difference in the age of the participants and the task design, similar results were reported as in the current study. In the third study (Eichorn and Pirutinsky, 2021), a modified version of the Dimension Change Card Sort Task (an alternation design task) was used that included a mixed block. Participants were school-aged children. The authors reported slower performance for the CWS in the mixed block. In both studies (the current and the Eichorn and Pirutinsky’s), the increased slowing down of the CWS compared to the CWNS was observed in the mixed block indicative of a higher performance-cost for the experimental group. A possible explanation for this could be that in mixed blocks, the alternation between set-shifting and no set-shifting is random. Under these conditions, participants experience high levels of uncertainty (Los, 1996) because they are in the position of not knowing if set-shifting or no-set-shifting is required. Due to this uncertainty, additional processing time is required for task-set-reconfiguration (Rogers and Monsell, 1995; Sohn and Carlson, 2000). Task-set-reconfiguration is the loading of the new rule into the working memory and the inhibition of processing of the old rule that no longer applies (Schmitter-Edgecombe and Langill, 2006). Working memory enables the maintenance of the two mental sets in an active state and inhibition enables the shifting between the two mental sets. Together, they are the main contributors to cognitive flexibility (Diamond, 2006).

Our third hypothesis was that the performance of both participant groups would improve with age as a result of EF maturation, as suggested by the ANT literature (Huijbregts et al., 2002; Rommelse et al., 2007; Sterken et al., 2016). This hypothesis was partially corroborated by significant negative correlations between age and performance-costs for the CWS but not for the CWNS for both complex response inhibition (compatible block vs. incompatible block) and cognitive flexibility (compatible block vs. mixed block [compatible part] and incompatible block vs. mixed block [incompatible part]). Despite the significant correlations for the CWS and the non-significant ones for the CWNS, in all cases there was a large overlap of the confidence intervals, something that suggests that the two groups did not behave differently; arguably, with a bigger sample size a negative correlation between age and response times would have reached significance for the CWNS as well. In a previous study, that of Rocha et al. (2019), the EFs of CWS and CWNS (7–12 years old) were evaluated using a paper and pencil performance test. The researchers reported that the younger subgroup of CWS (7–9 years old) required longer response times and made more errors compared to their age-and gender-matched CWNS, while this was not the case when comparisons were made between the two older subgroups (10–12 years old). This finding allowed them to claim a possible slower maturation of EFs in CWS. Other researchers have made similar claims specifically in regard to inhibitory control, that is, CWS may improve and eventually “catch up” in performance the CWNS (Anderson and Ofoe, 2019, p. 309), something that might also be the case for cognitive flexibility. This could explain why studies with older participants (7–11 years old) reported either a comparable performance between CWS and CWNS for inhibitory control (Eggers et al., 2018) or a better performance for the CWS (9–14 years old; Harrewijn et al., 2017). A longitudinal study could elucidate whether the developmental trajectory of these two EFs differs between CWS and CWNS or not.

We also reported a non-significant relationship between stuttering severity and performance-cost although one might speculate that the higher the stuttering severity, the higher the performance-cost would be in both measures of the EF. Kraft et al. (2014) documented a similar type of relationship between parental ratings of executive control and the extreme ends of the stuttering severity range, i.e., very mild versus very severe, but this was less clear for moderate stuttering severity levels. The lack of these extreme stuttering severity levels in our sample (no child with a very mild and only one child with a severe and one with a very severe stuttering severity score) as well as a different way of measuring stuttering severity (parental rating versus SSI-4 score in our sample) might explain this lack of correlation.

The findings of the current study suggest less efficient (slower) complex response inhibition and cognitive flexibility in CWS. This finding aligns with an increasing number of findings suggesting attentional regulation difficulties (Karrass et al., 2006; Jones et al., 2014) and attentional weaknesses in CWS (Eggers and Jansson-Verkasalo, 2017; Eichorn et al., 2018; Anderson et al., 2020). Both inhibitory control (e.g., Engelhardt et al., 2010, 2013) and cognitive flexibility (e.g., Deák, 2004; Crosbie et al., 2009) have been linked to speech-language planning and execution. They facilitate error detection prior to articulation and flexibly shifting one’s attention away from errors during articulation, resulting in increased fluency; attentional weaknesses on the other hand, may result in poor error monitoring, leading to disfluencies (Choo et al., 2020).

The fact that CWS slowed down significantly in both the complex response inhibition and in the cognitive flexibility measures, may reflect CWS needing to reduce their speed more to maintain similar levels of accuracy with CWNS. In other words, they might need additional time to regulate their behavior during attention-demanding situations, possibly also to plan and execute fluent speech and language (Anderson and Ofoe, 2019).

Our results also seem to support the claim of greater heterogeneity in the general stuttering population (e.g., Anderson and Ofoe, 2019; Singer et al., 2020). Despite our strict inclusion criteria, there was greater variability in both speed and accuracy in all blocks and breakdowns of the mixed block for the CWS group (Figure 3).

Although we elaborated on the possibility of less efficient EFs contributing to stuttering, our used methodology was not aimed at causality. Alternative explanations could be that both stuttering and impaired EFs share a common underlying neurological mechanism or that our findings are influenced by previously reported limitations in complex motor skills and sensorimotor learning in people who stutter (e.g., Loucks and De Nil, 2006; Smits-Bandstra et al., 2006; Kim et al., 2020). It is also possible, that CWS slowed down more than CWNS, because they needed additional time for adapting their response style to the needs of the task (see also Eggers et al., 2013) or because they had greater concern about errors (see also Eichorn et al., 2018).

Finally, our results on inhibitory control provide additional insights that may fill the gap between existing behavioral and neurocognitive studies. The results from the existing behavioral studies targeting prepotent response inhibition (Eggers et al., 2013; Piispala et al., 2016; Eggers et al., 2018; except for the Eggers et al., 2013 study in which children as young as 4 years old were included), reported comparable performance for the stuttering and the non-stuttering children, something that does not agree with the results from the neurocognitive studies (Piispala et al., 2017, 2018). In addition, the two behavioral studies that evaluated complex response inhibition, even though they reported limitations in EFs for the CWS group, similarly to the neurocognitive studies, they used auditory tasks with the likelihood of auditory processing difficulties impacting the results (Anderson and Wagovich, 2017; Eggers and Jansson-Verkasalo, 2017). Lastly, our results on cognitive flexibility are also of great importance, given the limited number of existing studies examining this EF in the stuttering population.

The lack of standardized language testing for Greek-speaking Cypriot children or a formal assessment of ADHD might be considered a caveat. However, for none of the participants there was any parental concern about language, learning or any other developmental problems. In addition, all the children’s teachers, and speech-language pathologists (for the CWS) were interviewed to rule out any co-occurring disorders, such as ADHD and, comparison of both participant groups on the different measures of the Bus Story Test did not yield any group-differences.

The male-to-female ratio (18, 1) does not adequately reflect the proportion typical in stuttering (3, 1 to 5, 1). Therefore, one could argue that the current findings are more a reflection of performance in school-age boys who stutter rather than children who stutter in general.

Finally, the additional measure of cognitive flexibility (a comparison of the incompatible trials of Block 3 with the trials of Block 2) goes beyond the suggested standard measure of cognitive flexibility and would benefit from further validation in the future.