A Portuguese community sample was recruited and invited to participate in an online survey on the relationship between personality and driver behaviour. Inclusion criteria included holding a driving licence, driving regularly for at least three years, and being under 75 years of age (older drivers have been associated with increased prevalence of cognitive, sensory and motor decline, which may affect driving performance and be linked to a higher risk of road traffic accidents (Zhang et al., 2024; SafetyNet, 2009)). In Portugal, the minimum legal driving age is 18 years. Data were collected over a four-month period in 2016, aiming to achieve a balanced distribution across age groups and gender.

All participants completed self-report questionnaires that included socio-demographic information, personal accident history, and validated instruments assessing personality traits and driving behaviour. Responses were anonymous. Prior to participation, individuals were informed about the research objectives and eligibility criteria.

The final sample comprised 747 participants: 417 women (55.8%), with a mean age of 42.13 ± 12.35 years, a mean driving licence tenure of 21.30 ± 11.34 years, and a mean of 20.33 ± 11.33 years of regular driving experience (Table 1). Education level was classified on a seven-point scale, from primary education (1) to doctoral degree (7). Although all levels were represented, most participants held graduate (36.8%) or master’s (30.0%) degrees.

This educational distribution is not representative of the Portuguese population. In 2016, only 23.8% of adults held a higher education degree, and this proportion had increased to merely 31.4% by 2024 (INE, 2025). The over-representation of highly educated participants may be attributed to the online survey format, which tends to attract individuals with greater digital literacy, typically associated with higher educational attainment.

The mean scores (and standard deviations) of the dimensions of the NEO-FFI-20, ImpSS, and DBQ are presented in Table 2. These dimensions correspond to the personality trait variables entered the regression models. The mean and standard deviation of the number of at-fault crashes per year of driving licence are also included in Table 2.

The hierarchical multiple linear regression was applied to three distinct models, considering as dependent variables the three driving behaviour dimensions of the DBQ: IAD, NIE, and Lapses. The aim was to identify the variables influencing each of the three driving behaviours, with particular focus on the NEO-FFI and ImpSS variables.

Four empirical driver profiles were identified (Tables 6 and 7; Figures 2 and 3): Prudent — low Impulsivity and Sensation Seeking, average/high Conscientiousness and Agreeableness (n=67); Regular - moderate across traits (n=429); Distracted/Forgetful - low/average Conscientiousness, high/average Neuroticism and Impulsivity (n=216); and Aggressive - low Agreeableness, high Impulsivity and Sensation Seeking (n=35).

In driving behaviour, the Prudent group exhibited the lowest IAD, NIE, and Lapse scores, while the Aggressive group showed the highest IAD, and the Distracted/Forgetful group showed higher NIE and Lapses. The Kruskal-Wallis test revealed significant group differences, confirmed by Mann–Whitney U tests: Prudent drivers differed significantly from Distracted/Forgetful and Aggressive drivers in all behaviours. The Prudent and Regular groups exhibited significant differences in IAD and NIE behaviours. The Regular and the Distracted/Forgetful presented significant differences in all behaviours. While Distracted/Forgetful and Aggressive groups did not differ significantly.

Hierarchical and k-means clustering identified four driver clusters based on personality traits — Neuroticism, Agreeableness, Conscientiousness, Extraversion, Impulsivity, and Sensation Seeking (Table 8; Figures 4 and 5). Cluster A exhibited medium/low scores for Conscientiousness, medium scores for Agreeableness and Extraversion, medium/high scores for Neuroticism and Sensation-seeking, and a high score for Impulsivity; Cluster B displayed medium/low scores for Neuroticism, Impulsivity and Sensation seeking, medium scores for Conscientiousness, Agreeableness and Extraversion; Cluster C showed low score for Extraversion, medium/low scores for Conscientiousness, Agreeableness, Impulsivity and Sensation seeking and medium/high score for Neuroticism.; and Cluster D demonstrated medium/low score for Neuroticism, medium scores for Agreeableness and Impulsivity, medium/high score for Conscientiousness, and slightly higher scores for Sensation seeking and Extraversion.

Given the number of individuals per cluster, ANOVA was used to compare means between groups, with Tamhane post-hoc tests applied as the variances were not homogeneous. Cluster A showed significant differences from Clusters B and C across all driving behaviours. Between Cluster A and Cluster D, significant differences were observed in NIE and Lapses. Comparisons between Clusters B and C also revealed significant differences across all behaviours, while Clusters B and D, and Clusters C and D, differed significantly only in IAD.

Two models were applied to analyse the number of at-fault crashes per year of driving licence, as described in Section 2.3.5. Table 9 presents the results of the regression model using the empirical profiles as independent variables, which, together with age and gender, explain 7.11% of the variance [F(5,741)= 11.336, p<.0001]. In this model, the Regular profile was used as the reference category. Apart from gender and age, only the Distracted/Forgetful profile showed statistical significance, possibly due to the higher number of observations compared with the Prudent and Aggressive profiles. This result highlights the difference between the Distracted/Forgetful and Regular profiles. The positive value estimated by the model indicates a higher crash risk for this profile compared with the Regular one.

Table 10 presents the results of the regression model using the k-means clusters as independent variables, which, together with age and gender, explain 5.26% of the variance [F(5,741)= 8.234, p<.0001]. The results show that none of the profiles were statistically significant, despite the relatively balanced number of observations across clusters.

Regression analyses indicated that personality traits, although accounting for modest variance, significantly predicted driving behaviour beyond socio-demographic factors. Infractions and Aggressive Driving (IAD) showed the highest explained variance (adj. R² = 0.272), followed by Lapses (adj. R² = 0.133) and Non-intentional Errors (NIE) (adj. R² = 0.101).

Higher Impulsivity, Sensation Seeking, Extraversion and Neuroticism, and lower Agreeableness, predicted higher IAD scores These findings led us to conclude that having a higher tendency to search for intense experiences and feelings, and at the same time having low compassion for others, can explain aggressive behaviours while driving, like becoming impatient with a slow driver and overtaking on the right (illegal manoeuvre in Portugal). The results also show that being talkative and assertive (extraversion), associated with the other traits, can also be correlated with pursuing aggressive behaviours on the road, like showing hostility to other drivers. Nevertheless, previous studies did not consistently confirm the role of Extraversion and Neuroticism, suggesting difficulty in defining a stable predictor set. Male gender and younger age were associated with higher IAD, corroborating earlier findings (Cordazzo et al., 2014; Granié et al., 2020). Importantly, even when controlling for Age, Gender, and Education, the personality traits still contributed significantly to explaining this aggressive behaviour.

The results of the non-intentional errors (NIE) regression also led us to claim that high tendency to act with little or no forethought (high impulsivity), together with low self-discipline (low conscientiousness) and high propensity to ruminate and experience feelings as anxiety (high neuroticism) can contribute to committing errors that can have dangerous outcomes. These results are in line with some previous findings (Tao et al., 2017). Controlling for the effect of Education, personality traits maintained their predictive ability. Gender, age and driving experience were not significant predictors, consistent with some studies (Cordazzo, 2014; Parker et al., 1995) but contradicting others (Deffenbacher et al., 2000; Correia, 2014).

Impulsivity was also the strongest predictor of Lapses, alongside low Conscientiousness and high Neuroticism. The variable Openness to experience also revealed positive association in this regression analysis, putatively showing that people with higher levels of intellectual curiosity, creativity and a preference for novelty have higher tendency to perpetuate errors related to distractibility. Female gender and younger age predicted higher Lapses, supporting existing evidence. When controlling for the socio-demographic variables, personality traits continued to significantly contribute to the model.

Overall, the consistent influence of Impulsivity and Neuroticism across all models underscores their central role in characterising driver behaviour.

Analysis of empirical profiles revealed distinct behavioural and personality patterns. The Aggressive group scored high in Impulsivity and Sensation Seeking but low in Agreeableness, while the Prudent group showed the inverse pattern — high Agreeableness and low Impulsivity and Sensation Seeking (Figure 2). This symmetry was also evident in driving behaviour, with the Regular group occupying an intermediate position (Figure 3). Significant group differences emerged in Infractions and Aggressive Driving (IAD) and Non-intentional Errors (NIE) between Prudent and Aggressive drivers, and in IAD when comparing Prudent with Regular or “Aggressive with Regular groups.

The Distracted/Forgetful group exhibited the highest frequency of NIEs and Lapses, associated with elevated Neuroticism and Impulsivity. Moderate Agreeableness and Sensation Seeking likely explain the lower occurrence of intentional violations. This group displayed higher error and violation levels than Regular drivers, who showed balanced traits and moderate error and violation rates (Figure 3). Comparisons between Distracted/Forgetful and Regular drivers revealed significant differences across all behavioural dimensions. The Distracted/Forgetful group also showed lower Conscientiousness and higher Agreeableness than the Aggressive group, which can explain more unintentional errors and lapses but fewer deliberate violations. Regular drivers represented the majority (n=429), followed by Distracted/Forgetful (n=216), Prudent (n=67), and Aggressive (n=35), reflecting the rarity of extreme behaviours.

K-means clustering produced comparable patterns (Figures 4 and 5). Cluster A- characterised by high Impulsivity and low Conscientiousness, showed the highest IAD, Lapses, and NIE scores, encompassing impulsive/aggressive and distracted drivers. Cluster B represented prudent behaviour with the lowest error levels, while Clusters C and D displayed moderate scores, differing primarily in Extraversion. Cluster sizes were relatively balanced: Cluster B (n=248), Cluster C (n=190), Cluster D (n=183), and Cluster A (n=126). Cluster A aligned most closely with the Aggressive group, and Cluster B with the Prudent group, while Distracted/Forgetful and “Regular groups overlapped with Clusters C and D.

Comparison of both methodologies showed that Cluster A included drivers with the highest mean scores across all behaviours, while the Aggressive group primarily committed intentional violations, with a higher IAD mean (0.709 vs. 0.486). The Prudent group, comparable to Cluster B, displayed the lowest scores and greater behavioural homogeneity. The Distracted/Forgetful group showed elevated NIE and Lapse scores but no significant differences from the “Aggressive group. This profile partially overlapped with Clusters C and D, which were less clearly defined.

The linear regression model results associating the number of at-fault crashes per year of driving licence with the profiles reinforce the conclusion that profiles based on the k-means method do not accurately represent risky behaviour, as no profile showed statistical significance, even with a higher number of observations in each cluster. In contrast, the results of the model applied to the empirical method showed that the Distracted/Forgetful group had a positive relationship with the number of at-fault crashes per year compared with the Regular group. The other groups did not show statistical significance, possibly due to the small number of observations per group.

The positive estimate for the Distracted/forgetful group aligns with expectations, considering Table 6, which shows that this group exhibits average to high scores for Neuroticism and Impulsivity, as well as low to average scores for Conscientiousness. Despite the extensive number of studies examining the association between Driver Behaviour Questionnaire (DBQ) scores and crash involvement (e.g., Bobermin & Ferreira, 2022), to the best of the authors’ knowledge, none has considered profiles derived from the combined dimensions of the NEO-FFI-20, ImpSS, and DBQ. Nevertheless, studies focusing on the relationship between DBQ factors and crash involvement have reported small positive correlations between violations and errors and self-reported crashes (de Winter et al., 2015; de Winter & Dodou, 2010). Assuming the Distracted/Forgetful group can be associated with these behaviours, the findings of the present study are consistent with several previous works.

A major strength of this study lies in its comprehensive integration of validated psychological instruments—the NEO-FFI-20, Impulsive Sensation Seeking Scale, and Driver Behaviour Questionnaire—to construct multidimensional driver behaviour profiles. The empirical profiling approach demonstrated greater predictive validity than k-means clustering, with the Distracted/Forgetful group showing a significant positive association with crash involvement, thereby reinforcing the model’s ecological validity. The identification of four theoretically coherent profiles - Prudent, Regular, Distracted/Forgetful, and Aggressive - highlights meaningful distinctions in personality and driving behaviour, providing novel insights into driver typologies. The large community sample (N=747) and robust psychometric reliability further strengthen the findings.

Nevertheless, limitations include reliance on self-reported data, which may introduce consistency and acquiescence biases, and a sample with an overrepresentation of highly educated individuals, which is not representative of the Portuguese population. Moreover, the complexity of human behaviour is not captured in all its dimensions, limiting the conclusions. In this study we only applied K-mean cluster algorithm. Experiment with other clustering algorithms would be important to validate these results. Additionally, given that road crash events are relatively rare over a driver’s lifetime, self-reported at-fault crashes are typically sparse, thereby reducing the statistical power of the analysis and limiting the ability to detect meaningful association.

Another aspect to consider concerns the temporal context of the dataset. Data were collected in 2016, nevertheless the dataset may still be considered relevant, as the psychological constructs examined—particularly personality traits—are generally regarded as relatively stable across adulthood and repeatedly linked to driving behaviour, as outlined in the introduction. Similarly, the DBQ has shown enduring validity across different traffic contexts since its development. Nonetheless, we acknowledge that the Portuguese driving environment has evolved since 2016, including technological advancements in vehicles, changes in road safety policies, and shifts in traffic density, which may influence the expression or frequency of specific behaviours. Therefore, the findings should not be interpreted as fully representative of current conditions. Instead, they provide valuable insight into the underlying psychological mechanisms shaping driver behaviour, while highlighting the need for updated datasets to confirm contemporary applicability.

Despite these constraints, the study offers an original and methodologically rigorous contribution to understanding personality-based differences in driving behaviour.