Participants

The study used a non-probability convenience sampling strategy. This approach was selected because the study was exploratory and aimed to recruit a heterogeneous participant pool with variability in frustration responses, personality traits, and behavioral tendencies, rather than to estimate population prevalence from a representative sample. The target sample size was determined based on a power analysis for the planned continuous analyses examining associations between frustration responses and individual-difference or task-related variables. Because the number of frustration profiles was not known in advance and was identified empirically through exploratory clustering, the sample size calculation was not based on a prespecified number of groups. Assuming α = .05, 80% power, and a medium expected association of r = .25, the analysis indicated that approximately 123 participants would be required. The initial sample consisted of N = 147 exceeding the target and providing adequate sensitivity for detecting medium-sized associations. Two participants were excluded from the final sample: one because of a technical issue and one because they withdrew before completing the experiment. The final sample consisted of 145 participants, including 100 females (68.97%), with overall ages ranging from 14 to 66 years (M = 28.35, SD = 9.15). The broad age range was permitted because the study was designed as an exploratory investigation of frustration responses in a heterogeneous participant pool rather than as a developmental or age-stratified comparison. Age was not used as an experimental grouping factor; however, age differences across frustration clusters were examined statistically to evaluate whether cluster membership was associated with age. Participants were eligible for inclusion if they were able to understand the study instructions, provide written informed consent or assent with guardian consent when applicable, and complete the iPad-based Frustration Induction Task. Exclusion criteria were withdrawal before completion of the study or technical difficulties that prevented valid completion of the experimental task. Two participants were excluded according to these criteria: one due to technical difficulties that interrupted task completion, and one due to voluntary withdrawal. The recruitment period of this study took place between 2024-11-21 and 2024-12-19. The study received ethical approval from the Swedish Ethical Review Authority (Dnr 2024-04648-01) and was conducted in full accordance with the Declaration of Helsinki. Prior to participation, all individuals received detailed information about the study procedures and their right to withdraw. Written informed consent was obtained from all adult participants. For participants under the age of 18, written consent was obtained from a parent or legal guardian in addition to the participant’s assent.

Recruitment

Participants were recruited via the Karolinska Institutet Psychology Testing Recruitment System, online advertisements (www.accindi.se), and community notice boards to establish a diverse participant pool to the greatest extent possible. The study used partial disclosure because informing participants in advance that the task was designed to induce frustration could have altered their expectations and emotional responses, thereby reducing the validity of the experimental manipulation. The use of partial disclosure, the short duration of the frustration-induction procedure, the expected minimal-risk nature of the task, and the planned debriefing procedure were described in the ethics application and reviewed and approved by the Swedish Ethical Review Authority. Immediately after completing the task and all post-task questionnaires, participants were fully debriefed. The debriefing explained the study’s true purpose, the rationale for using partial disclosure, and the frustration-inducing elements of the FIT. Participants were given the opportunity to ask questions. The study procedure, including the use of partial disclosure and debriefing, was approved by the Swedish Ethical Review Authority.

Participants were informed that they would receive a gift card valued at either 50 or 100 SEK, depending on task performance. All participants received compensation, and the difference between compensation levels was kept small to avoid undue influence or coercion. The performance-contingent component was included as a mild experimental feature to increase task involvement, and goal obstruction may be accompanied by perceived gains or losses that affect the reward system. Table 1 displays a descriptive summary of participants’ personality traits and behavioral tendencies.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Participants’ Descriptive Information.

https://doi.org/10.1371/journal.pone.0353785.t001

Personality traits were evaluated on a validated five-point response scale, BFI-2-XS [24], where participants showed the highest average scores for open-mindedness (M = 3.70, SD = 0.57) and agreeableness (M = 3.36, SD = 0.67). In contrast, neuroticism (M = 2.57, SD = 0.94), extraversion (M = 2.74, SD = 0.75), and conscientiousness (M = 2.78, SD = 0.74) showed moderate average levels. Behavioral tendencies such as internal control, perfectionism, emotional intelligence, and emotional regulation (impulse control difficulties and difficulties engaging) were evaluated on a validated five-point scale. Participants exhibited elevated emotional intelligence (M = 4.35, SD = 0.97) and personal standards for perfectionism (M = 3.32, SD = 0.71), with comparatively lower scores on impulse control difficulties (M = 1.95, SD = 0.78). Other behavioral traits, such as discomfort intolerance, self-esteem, and sensitivity to behavioral inhibition (BIS) and activation (BAS), were measured on four-point scales. Emotional challenges such as low self-esteem (M = 1.63, SD = 0.53) and moderate BIS sensitivity (M = 1.71, SD = 0.50) were observed.

Procedures

The study was conducted in a controlled laboratory environment to ensure consistency across testing conditions. Upon arrival, participants were welcomed and provided an overview of the study’s general purpose as described in the consent form. After giving written informed consent, they completed a multi-stage experiment to assess frustration in response to a navigation-based novel Frustration Induction Task (FIT) [22].

The procedure was divided into three phases. First, in the pre-task assessment, participants completed a baseline questionnaire. Next, participants engaged in the FIT on an iPad. In the final post-task assessment, participants completed a questionnaire evaluating their emotional responses to the FIT, their overall frustration, and individual differences across personality traits and behavioral tendencies. Thus, the procedure followed a fixed sequence for all participants: consent and baseline assessment, completion of the FIT, post-task assessment of frustration and affective responses, individual-difference questionnaires, debriefing, and compensation.

Measures

To assess individual readiness and baseline characteristics before frustration induction, participants completed a pre-task assessment that captured demographic information, current mood, physical energy, sleep quality, hunger levels, and familiarity with navigation tools and environments. The core of the experiment was a ten-trial iPad-based navigation task designed to induce frustration [22]. Using an ABC keyboard layout, participants were asked to enter various address names. The first two trials were free of manipulation, but the remaining eight trials incorporated layered components to gradually increase frustration. These frustration components included a 30-second time limit to induce time pressure, keyboard manipulations to produce case-sensitive errors, distracting visual elements such as running banners and warning banners, and a performance-based reward system that created a loss-framing element; participants earned compensation based on successful completion of each trial, making task failure more consequential. Following each trial, participants responded to a brief in-app questionnaire of five items rated on a 7-point Likert scale (1 = strongly disagree; 7 = strongly agree). They reported perceived difficulty (“I found the trial difficult to accomplish”), familiarity with the address, outcome alignment (“The trial result was as I expected”), frustration (“I found this trial frustrating”), and task motivation (“I am motivated to continue with the next trial”).

Frustration was operationalized as participants’ self-reported emotional response to the obstruction of goal-directed task performance during the FIT. Trial-level frustration was assessed after each trial using the item “I found this trial frustrating,” rated on a 7-point Likert scale from 1 = strongly disagree to 7 = strongly agree. Overall frustration was assessed after the task using a 7-point rating scale ranging from 1 (not frustrated) to 7 (very frustrated). These trial-level and overall ratings served as the primary indicators of participants’ frustration responses.

After the task, participants completed a comprehensive post-task questionnaire assessing emotional responses, behavioral tendencies, personality traits, frustration components, and overall frustration. An open-ended question at the beginning of the post-task questionnaire asked participants what they believed the study’s purpose was. This item was included as an awareness check to assess whether participants had guessed the task’s frustration-induction purpose before debriefing. These responses were not subjected to thematic or qualitative analysis and were not included in the primary statistical analyses. Then, to quantify their overall frustration levels, the participants first rated their experience on a 7-point Likert scale (1 = not frustrated, 7 = very frustrated). Participants also rated their perceived frustration on a 7-point scale from 1 (extremely negative) to 7 (extremely positive). To assess participants’ emotional responses associated with frustration during the task, post-task affective responses were assessed using the Positive and Negative Affect Schedule (PANAS) [25], rating the intensity of 20 affective states on a 5-point scale 1(very slightly or not at all) to 5 (very much). The scale included positive affect items, such as enthusiastic, determined, alert, attentive, and interested, and negative affect items, such as irritable, distressed, hostile, nervous, ashamed, guilty, scared, and afraid. The PANAS was included to characterize the affective profile accompanying frustration after the FIT, rather than to operationally define frustration itself.

To explore behavioral tendencies that may modulate frustration responses, participants completed several validated psychological scales. The internal control scale, Locus of Control [26], assessed beliefs about personal agency and outcome control, using six levels of agreement for items like “When I make plans, I am almost certain to make them work,” “My life is determined by my own actions,” etc. The Frost Multidimensional Perfectionism Scale [27] measured two perfectionism subdomains: concern over mistakes and personal standards, on a 5-point Likert scale from “strongly disagree” to “strongly agree.” Sample items included “If I fail at school, I am a failure as a person,” “I set higher goals than most people,” etc. To measure discomfort intolerance, the Frustration Discomfort Scale [16] was administered, where participants rated the strength of belief on a scale from 0 (absent) to 4 (very strong) for items such as “I can’t stand doing tasks that seem too difficult,” “I need the easiest way around problems.”

Emotion regulation was assessed using two subscales from the Difficulties in Emotion Regulation Scale [28], which evaluated impulse control difficulties and difficulties engaging in goal-directed behavior. Items such as “When I’m upset, I have difficulty concentrating,” “When I’m upset, I lose control over my behaviors,” etc., were rated on a 5-point scale, ranging from 1 (almost never) to 5 (almost always). Emotional intelligence was measured using the Emotional Skills and Competence Questionnaire [29], which assessed the ability to perceive, express, and regulate emotions. Responses ranged from 1 (strongly disagree) to 5 (strongly agree) for items like “I am able to express my emotions well,” “I can maintain a good mood even if something bad happens,” etc. Self-esteem was measured using the Rosenberg Self-Esteem Scale [30], a 10-item measure rated on a 4-point scale, including positively and negatively phrased statements (e.g., “I take a positive attitude toward myself”, “I wish I could have more respect for myself.”)

Motivational sensitivity was assessed using the Behavioral Inhibition System and Behavioral Activation System scales (BIS/BAS) [31]. The BIS scale assesses sensitivity to punishment, threat, and behavioral inhibition, whereas the BAS scales assess reward-related motivation, including reward responsiveness, drive, and fun seeking. In the present study, BIS/BAS scores were used as dimensional indicators of individual differences in motivational sensitivity rather than as diagnostic or psychiatric classifications. The BIS subscale focused on behavioral inhibition (e.g., “Criticism or scolding hurts me quite a bit”, “I have very few fears compared to my friends”). In contrast, the BAS subscales covered reward responsiveness, drive, and fun seeking (e.g., “I crave excitement and new sensations,” “I go out of my way to get things I want,” “I often act on the spur of the moment”), all rated on a 4-point Likert scale ranging from 1 (strong agreement) to 4 (strong disagreement).

Additionally, personality traits were assessed using the Big Five Inventory–2 Extra-Short Form (BFI-2-XS) [24], a brief self-report measure of five broad personality dimensions: extraversion, agreeableness, conscientiousness, neuroticism, and open-mindedness. Each trait was assessed dimensionally, with higher scores reflecting greater endorsement of the corresponding trait. The BFI-2-XS was selected because it provides a brief assessment of broad personality traits while minimizing participant burden in a multi-measure experimental protocol. BFI-2-XS includes a 15-item scale measuring five core personality traits (extraversion, agreeableness, conscientiousness, neuroticism, and open-mindedness). Each trait was assessed via three statements (e.g., “Is full of energy,” “Is reliable,” “Worries a lot”), rated from 1 (disagree strongly) to 5 (agree strongly).

They then evaluated how frustrating each task component was, such as time pressure, keyboard manipulations, visual interference (running banner and warning banner), or environmental distractions like noise or lighting, using the same 1–7 point Likert scale. To assess awareness of task manipulations, they rated their recognition of keyboard errors (“Did you notice the keyboard was not functioning correctly?”) and their confidence that the task had been intentionally manipulated (“How confident are you that the keyboard was manipulated during the task?”), both on a 7-point scale. After completing all measures, participants were fully debriefed and compensated for their task performance.

Data analysis

To analyze frustration clusters in the dataset, we used a combination of software tools and statistical methods. All primary analyses were conducted in Python (version 3.13), using several key libraries for data processing, statistical modeling, and visualization. Pandas was used for data organization and manipulation, while NumPy handled numerical computations. Visualizations, including heatmaps, scatter plots, radar plots, and bar graphs, were created using the matplotlib and seaborn libraries. For clustering analyses, we used SciPy’s linkage and fcluster functions to perform hierarchical clustering with complete linkage and Euclidean distance as the similarity metric. To explore the relationship between questionnaire responses and frustration levels, we applied Partial Least Squares Regression using the PLS regression function from the Scikit-learn (Sklearn) library. Sklearn also supported various data preprocessing tasks throughout the analysis. Because participants were recruited using a non-probability convenience sampling strategy, inferential analyses were interpreted as exploratory rather than as population-representative estimates. In addition, because the number and size of the frustration profiles were empirically derived rather than prespecified, all cluster-based group comparisons were interpreted as exploratory follow-up analyses. Welch’s ANOVA and Games-Howell post hoc tests were used where appropriate because these methods are robust to unequal group sizes and heterogeneity of variance. False discovery rate correction was applied within test families to reduce the likelihood of false-positive findings across multiple comparisons.

The data processing began with cleaning and transforming the raw dataset. Since the original data was collected in spreadsheet format, initial adjustments were made in Microsoft Excel. This included reverse-scoring negatively worded items (for example, personality questions where higher scores indicated a greater degree of that trait) to ensure consistency in variable interpretation. Additional processing steps were performed after the data was imported into Python. Descriptive statistics (means and standard deviations) were reported on the original response scales of their respective questionnaires to facilitate interpretation. For multivariate analyses and graphical visualization, continuous variables were rescaled to a common 0–10 metric using min–max normalization to enable comparisons across measures originally recorded on different response scales. The transformation was applied as , where X is the observed value, X_min and X_max are the theoretical minimum and maximum possible values for that measure, and X′ is the rescaled value. For example, Likert-type responses originally recorded on 1–5 or 1–7 scales were proportionally mapped to the 0–10 metric. This rescaling preserved the relative distribution of each variable while enabling direct comparability across measures in subsequent analyses.

Additional normalization steps were applied to the radar plots to ensure consistent scaling across variables. After the initial normalization, which scaled all values to 0–1, the values were multiplied by 10 to align with the radar plot axes, which ranged from 0 to 10. This procedure was applied to personality traits, behavioral tendencies, perceived task triggers, and emotional response variables to ensure comparability across measures originally recorded on different scales. Standard deviations for each variable were scaled similarly and plotted as dotted lines to illustrate the variability within each cluster; graphical conventions, including color coding and line styles, are specified in the corresponding figure’s legends. For hierarchical clustering, normalized frustration scores were used to generate dendrogram structures, from which three distinct clusters emerged. These were labeled as High, Mid, and Low frustration based on the ranked mean frustration scores of each cluster. To explore relationships between frustration and other factors, we conducted Pearson correlation coefficients and dimensionality reduction with PCA.

To evaluate whether the identified frustration clusters differed across demographic variables, personality traits, behavioral tendencies, perceived task triggers, and emotional responses, we conducted inferential statistical analyses in Python. Gender distribution across clusters was examined using a chi-square test of independence, with Cramér’s V reported as an effect size. Age differences across clusters were examined using Welch’s ANOVA (robust to unequal variances and unequal group sizes), reporting partial eta squared (ηp²).

For conceptually related outcome sets (Big Five traits; perceived frustration triggers; emotional responses; behavioral tendencies), we first conducted multivariate analyses of variance (MANOVAs) with cluster membership as the between-subjects factor and Wilks’ Lambda (Λ) as the primary multivariate test statistic. When the MANOVA indicated a significant multivariate cluster effect, follow-up Welch ANOVAs were performed for each variable within that outcome set. To control for multiple comparisons, false discovery rate correction (Benjamini–Hochberg) was applied within each family of tests (traits, triggers, emotions). For follow-up tests, partial eta squared (ηp²) was reported.

When Welch ANOVAs indicated statistically significant effects after false discovery rate correction, Games–Howell post-hoc comparisons were conducted to examine pairwise differences between clusters. For each comparison, mean differences with corresponding 95% confidence intervals (CIs) and exact p values were reported. Confidence intervals were computed based on the standard error of the mean difference. All inferential tests were two-tailed.

Frustration clusters and demographics

To visualize the separation of the frustration clusters identified through hierarchical clustering, we conducted a principal component analysis (PCA) on the normalized frustration variables. PCA was used solely as a dimensionality-reduction technique for visualization and did not contribute to cluster identification.

The hierarchical clustering identified three frustration clusters: High Frustration (n = 96, 66.2%), Mid Frustration (n = 39, 26.9%), and Low Frustration (n = 10, 6.9%). The proportion of female participants was similar across profiles: 66.7% in the High Frustration profile, 71.8% in the Mid Frustration profile, and 80.0% in the Low Frustration profile, as presented in Table 2. Age did not differ substantially across clusters, with means ranging from 25.0 to 28.8 years. To examine whether demographic characteristics differed across frustration clusters, inferential analyses were conducted. A chi-square test of independence indicated no significant association between gender and frustration cluster, χ² (2) = 0.95, p = .62, Cramér’s V = .08. Similarly, a Welch’s ANOVA revealed no significant difference in age across clusters, F (2, 23.38) = 0.58, p = .57, partial η² = .01 (N = 145). Thus, profile membership was not significantly associated with gender, and no significant age differences were detected across profiles in this sample. Educational background showed modest variation: the Low Frustration cluster had the highest proportion of participants with advanced degrees (40%), compared with 30% in the High group and 28% in the Mid group.

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Demographic information by frustration clusters.

https://doi.org/10.1371/journal.pone.0353785.t002

Fig 1 displays participants’ scores on the first two principal components, color-coded by frustration profile. PC1 explained 32.5% of the variance, and PC2 explained 18.7%, together accounting for 51.2% of the total variance. Participants in the High Frustration group (red) clustered toward the positive end of PC1, while the Low Frustration group (blue) appeared more on the negative side. The Mid Frustration group (green) was distributed between the two. The High Frustration group exhibited the broadest variability, whereas the Low Frustration group showed more compact in the PCA visualization; however, this pattern should be interpreted cautiously because the profile contained only 10 participants.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Principal Component Analysis of frustration clusters.

(1a) Scatterplot of the first two principal components of normalized frustration scores, color-coded by cluster: High Frustration (red), Mid Frustration (green), and Low Frustration (blue). (1b) Each point represents a participant, with point size scaled to the overall frustration score (1–7). (1c) Ellipses indicate the covariance spread of each cluster.

https://doi.org/10.1371/journal.pone.0353785.g001

Personality traits across clusters

To examine whether personality traits differed across frustration profiles, we compared distributions of the Big Five traits (extraversion, agreeableness, conscientiousness, neuroticism, and openness) among the High, Mid, and Low Frustration clusters.

A MANOVA revealed a statistically significant multivariate effect of cluster on the combined Big Five personality traits, Wilks’ Λ = .88, F (10, 276) = 1.88, p = .047, partial η² = .06 (N = 145), indicating modest overall differences in trait profiles across frustration groups. However, follow-up Welch ANOVAs with false discovery rate (FDR) correction did not identify statistically significant differences for any individual trait (all p > .05; Table 3). Although conscientiousness and neuroticism showed nominal differences at the uncorrected level, these effects were not statistically significant after correction.

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Big Five Traits.

https://doi.org/10.1371/journal.pone.0353785.t003

Descriptively, neuroticism displayed the largest separation between clusters in Fig 2. The High Frustration group reported higher neuroticism (M = 7.8, SD = 1.2) compared with the Mid (M = 6.5, SD = 1.0) and Low clusters (M = 5.2, SD = 0.9). Pairwise Games–Howell comparisons indicated that the High cluster differed from the Low cluster (mean difference = 0.957, 95% CI [0.300, 1.614], p = .037), whereas differences between High and Mid clusters were not statistically significant (p = .560). Conscientiousness showed a similar directional gradient (High: M = 8.0, SD = 1.1; Mid: M = 7.2, SD = 0.8; Low: M = 6.8, SD = 0.7). However, because no individual trait remained significant after FDR correction, these patterns should be interpreted as descriptive rather than inferential.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Personality traits across frustration clusters.

(2a) Radar plot of normalized Big Five personality trait scores in High (red), Mid (green), and Low (blue) Frustration clusters. (2b) Neuroticism and conscientiousness showed the largest descriptive differences across clusters, whereas extraversion, agreeableness, and openness remained comparatively similar across profiles.

https://doi.org/10.1371/journal.pone.0353785.g002

Extraversion, agreeableness, and openness were relatively stable across clusters, with small mean differences and confidence intervals spanning zero. Overall, personality differences across frustration profiles were subtle and of limited magnitude.

To assess how situational factors contributed to frustration, we examined participants’ ratings of task-related components in the FIT. The goal was to determine whether specific triggers, such as time pressure, keyboard manipulation, or address complexity, differed in the strength with which they elicited frustration across the High, Mid, and Low Frustration clusters.

We calculated mean and standard deviation values for each trigger. A MANOVA demonstrated a statistically significant multivariate effect of cluster on perceived task-related triggers, Wilks’ Λ = .61, F (20, 266) = 3.67, p < .001, partial η² = .22 (N = 145), indicating substantial differences in trigger sensitivity across frustration profiles. Follow-up Welch ANOVAs with FDR correction revealed statistically significant cluster effects for time pressure, final performance summary, warning banner, ABC keyboard layout, keyboard manipulation, failure point, and running banner (all adjusted p < .05; Table 4). No significant differences were observed for familiarity with the address, long address length, or reward system.

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Frustration Triggers.

https://doi.org/10.1371/journal.pone.0353785.t004

Games–Howell comparisons indicated that the High Frustration cluster consistently reported elevated trigger sensitivity relative to the Low cluster in Fig 3. For example, perceived time pressure was markedly higher in the High group than in the Low group (mean difference = 2.982, 95% CI [1.850, 4.113], p < .001) and also higher than in the Mid group (mean difference = 1.367, 95% CI [0.553, 2.181], p = .005). Similarly, frustration associated with the final performance summary was greater in the High cluster compared with the Low cluster (mean difference = 3.735, 95% CI [1.943, 5.527], p = .005) and the Mid cluster (mean difference = 2.100, 95% CI [1.253, 2.947], p < .001).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Frustration triggers across clusters.

(3a) Radar plot of normalized ratings of FIT components, including address familiarity, address length, time pressure, keyboard manipulation, and reward system disruptions, for High (red), Mid (green), and Low (blue) Frustration clusters.

https://doi.org/10.1371/journal.pone.0353785.g003

Keyboard-related disruptions followed the same pattern. The High cluster reported significantly greater frustration than the Low cluster for keyboard manipulation (mean difference = 2.987, 95% CI [1.277, 4.698], p = .018) and warning banners (mean difference = 3.068, 95% CI [1.739, 4.397], p = .003). In contrast, address familiarity ratings were statistically comparable across clusters, and all pairwise confidence intervals including zero (all p > .60). Descriptive means reflected in this stability (High: M = 4.5, SD = 1.0; Mid: M = 4.6, SD = 0.9; Low: M = 4.4, SD = 0.8). Overall, heightened frustration was characterized by amplified sensitivity to time constraints, performance evaluation, and interface disruptions, whereas familiarity-related task elements did not differentiate the groups.

Behavioral tendencies across clusters

To investigate whether behavioral tendencies differentiated frustration responses, we compared scores for internal control, emotional intelligence, perfectionism, discomfort intolerance, self-esteem, impulse control difficulties, difficulties engaging, and BIS/BAS sensitivity across the three frustration clusters.

A MANOVA examining internal control, perfectionism dimensions, discomfort intolerance, emotional intelligence, self-esteem, impulsive control difficulties, engagement difficulties, and BIS/BAS indices did not yield a statistically significant multivariate cluster effect, Wilks’ Λ = .79, F (24, 262) = 1.37, p = .123, partial η² = .11 (N = 145). Consistent with the multivariate result, follow-up Welch ANOVAs with FDR correction indicated no statistically significant between-cluster differences for individual behavioral tendencies (all adjusted p > .05; Table 5).

Download:

• PNG
  larger image
• TIFF
  original image

Table 5. Behavioral Tendencies.

https://doi.org/10.1371/journal.pone.0353785.t005

Descriptively, emotional intelligence appeared highest in the Low Frustration cluster (M = 8.5, SD = 0.8), followed by the Mid cluster (M = 7.3, SD = 1.0), and lowest in the High cluster (M = 6.2, SD = 1.2). Similarly, perfectionism of personal standards and BIS sensitivity showed directional gradients across clusters. However, pairwise comparisons revealed small to moderate mean differences with confidence intervals frequently spanning zero. For example, the normalized emotional-intelligence mean was highest in the Low Frustration profile and lowest in the High Frustration profile (mean difference = 0.862, 95% CI [0.312, 1.411], p = .022). However, as shown in Fig 4, However, neither emotional intelligence nor any other individual behavioral tendency remained statistically significant after FDR correction. Therefore, these trends should be interpreted descriptively.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Behavioral tendencies across frustration clusters.

(4a) Radar plots of internal control, emotional intelligence, perfectionism (personal standards and concern over mistakes), discomfort intolerance, self-esteem, impulse control difficulties, difficulties engaging, and BIS/BAS sensitivity in High (red), Mid (green), and Low (blue) Frustration clusters.

https://doi.org/10.1371/journal.pone.0353785.g004

Emotional responses across frustration clusters

To evaluate how emotional reactions varied across frustration profiles, we examined positive and negative affective ratings in the High, Mid, and Low Frustration clusters. The question was whether certain emotions, such as attentiveness, interest, or distress, were more strongly associated with high frustration compared to lower frustration states.

A MANOVA revealed a statistically significant multivariate effect of cluster on emotional responses following the task, Wilks’ Λ = .52, F (40, 246) = 2.37, p < .001, partial η² = .28 (N = 145), indicating substantial affective differentiation across frustration profiles. Follow-up Welch ANOVAs with FDR correction identified significant between-cluster differences in multiple negative affective states, including guilt, fear, scared, upset, irritability, hostility, shame, distress, jitteriness, and nervousness (all p < .01; Table 6). Positive engagement-related emotions including enthusiasm, determination, interest, and attentiveness did not show statistically reliable differences after correction, although excited and alert demonstrated significant between-cluster variation (Table 6). Notably, levels of excitement and alertness were higher in the High Frustration cluster relative to the Mid cluster (both p < .05, FDR-corrected), suggesting that the profiles differed in arousal-related positive affect, although the specific pairwise differences require interpretation from the corresponding post hoc comparisons.

Download:

• PNG
  larger image
• TIFF
  original image

Table 6. Emotional Responses.

https://doi.org/10.1371/journal.pone.0353785.t006

Games–Howell comparisons indicated that negative emotions were consistently elevated in the High Frustration cluster relative to the Low cluster. For example, guilt was higher in the High cluster compared with the Low cluster (mean difference = 2.927, 95% CI [2.322, 3.533], p < .001) and compared with the Mid cluster (mean difference = 1.242, 95% CI [0.363, 2.121], p = .018). Similarly, irritability was markedly elevated in the High cluster relative to the Low cluster (mean difference = 4.477, 95% CI [2.741, 6.213], p < .001). Distress showed a comparable pattern, with the High cluster exceeding the Mid cluster (mean difference = 1.908, 95% CI [0.944, 2.872], p < .001) and the Low cluster (mean difference = 3.471, 95% CI [1.222, 5.721], p = .031). In contrast, attentiveness remained stable across groups (High: M = 8.0, SD = 0.7; Mid: M = 8.1, SD = 0.6; Low: M = 8.0, SD = 0.5), with no statistically significant pairwise differences. Overall, as shown in Fig 5, higher frustration was associated with increased negative affect and elevated arousal-related responses, while attentional engagement remained stable across clusters.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 5. Emotional responses across frustration clusters.

(5a) Radar plots of normalized ratings (0–10 scale) of positive and negative emotions, including attentiveness, interest, distress, irritability, guilt, nervousness, and enthusiasm, for High (red), Mid (green), and Low (blue) Frustration clusters.

https://doi.org/10.1371/journal.pone.0353785.g005

Limitations

While this study offers new insights into the psychological mechanisms behind frustration, several limitations warrant mention. First, relying on self-reported measures leaves room for biases, such as participants misjudging their own emotions or trying to present themselves in a better light, even though validated scales were used. Second, the sample size was modest and drawn from a narrow regional pool, which could limit the generalizability of these findings to other populations. Cluster sizes were also imbalanced, particularly for the Low Frustration group (n = 10), which may reduce power for detecting smaller between-cluster effects. Additionally, the large number of outcomes required correction for multiple comparisons; while this reduces false positives, it may obscure smaller effects that warrant replication with preregistered hypotheses. Third, although age did not significantly differ across frustration clusters in the present analyses, participants ranged from adolescence to older adulthood, and developmental differences in cognitive control, emotion regulation, stress physiology, and frustration tolerance may not be fully captured by the current design. Participants at the lower and upper ends of the age range may represent extreme values relative to the sample’s central tendency, and future studies should examine frustration profiles using age-stratified sampling, age-matched groups, or separate developmental cohorts. Fourth, although the self-report instruments used in this study are widely used in psychological research, the broad age range means their psychometric properties may not be identical across adolescents, adults, and older adults. These measures were used as dimensional individual-difference indicators rather than diagnostic or psychiatric tools. Future studies should examine measurement invariance across age groups or use age-specific validated instruments when comparing developmental cohorts. Fifth, the FIT allowed standardized exposure to the same frustration-inducing elements across participants; laboratory-based frustration may differ from real-world frustration. Future studies should examine whether the identified frustration profiles generalize to naturalistic settings. Finally, while the Frustration Task successfully provoked emotional reactions in a controlled setting, it can not fully capture the complexity of frustration as it plays out in real life, where challenges are often longer, less predictable, and shaped by social interactions. Future work should aim to recreate these dynamics in more naturalistic environments to better understand how frustration unfolds outside the lab.