A study on the correlation between MBTI dimensions and driving behavior characteristics

Abstract

With the rapid global increase in motor vehicle usage, road traffic injuries have emerged as the leading cause of injury-related deaths worldwide. Within the complex traffic system, although factors such as vehicle performance and road conditions significantly influence driving safety, the driver’s personality traits remain a critical determinant of traffic accidents. Consequently, exploring the intricate relationship between driving behavior patterns and personality traits is essential for understanding the underlying causes of traffic injuries and developing effective intervention strategies. Grounded in the theoretical framework of the Myers-Briggs Type Indicator (MBTI), this study systematically examines the interaction between personality traits and driving behavior. Through an empirical analysis of driving behavior data, this research makes several notable contributions. First, it introduces the “Six Driving Behavioral Facets,” a multidimensional framework for analyzing the relationship between personality traits and driving behavior. Second, the study employs the “inverse chi-square test” to uncover latent patterns in otherwise non-significant results. Using the K-modes clustering algorithm, this study identified significant imbalances in the distribution of MBTI personality dimensions across eight clusters, particularly in the Thinking-Feeling (T-F) dimension. For example, in Cluster 1, Thinking (T) individuals accounted for 10.84% of the total population, compared to 15.09% for Feeling (F) individuals, while in Cluster 5, T individuals represented 17.48%, compared to 10.53% for F individuals. Such pronounced differences in personality distributions across clusters highlight the relevance of MBTI traits in shaping driving behavior patterns. These findings provide theoretical support for personalized traffic management strategies and the optimization of autonomous driving systems.

Introduction

With the rapid increase in motor vehicle ownership, road traffic injuries have become a leading cause of injury-related deaths worldwide. In 2019, road traffic injuries accounted for 29% of all injury-related deaths globally, followed by suicide (16%), falls (16%), and interpersonal violence (11%). Additionally, the decline in the mortality rate from road traffic injuries has been relatively slow, decreasing from 19.1 per 100,000 population in 2000 to 16.7 per 100,000 in 2019—a reduction of just 13%. Despite this decline, the total number of deaths has risen due to global population growth, increasing from 1.2 million per year in 2000 to 1.3 million per year in 2019¹.

Traffic accidents not only threaten personal safety and property but also impose a significant burden on public health and economic development. Within the complex traffic system, although factors such as vehicle performance and road conditions play a role in driving safety, human factors—particularly the personality traits of drivers—remain a core contributor to accidents². Psychologists and traffic researchers have observed that drivers’ psychological states and personality traits are closely associated with their driving behaviors, significantly influencing driving safety³. Driving is a complex behavior that involves decision-making, emotion regulation, risk assessment, and other processes—all of which may be shaped by individual personality traits. Behavioral responses in driving situations are often considered reflective of core personality tendencies⁴. Because driving requires rapid responses, it can reveal a person’s immediate intuitive reactions, which frequently reflect their inherent personality⁵. Although individuals may mask their personality traits in daily life, these traits are likely to manifest in their driving behaviors in complex traffic scenarios⁶.

As autonomous driving technology becomes increasingly prevalent, an in-depth exploration of driving behavior has become essential. Research has demonstrated that personality traits are a key factor influencing trust in autonomous driving⁷, which has prompted further investigations into the role of personality characteristics in driving behavior.

Given the critical role of personality traits in driving behavior, personality theories in psychology provide a theoretical foundation for interdisciplinary research bridging social sciences and engineering. Among existing studies, the Five-Factor Model (FFM) stands out due to its widespread application. Meanwhile, many studies have explored the association between personality and driving behavior without being confined to existing personality theory frameworks. Additionally, despite ongoing controversies, the Myers-Briggs Type Indicator (MBTI) has gradually been incorporated into practical applications due to its unique characteristics.

In the application of the FFM, Arthur et al. found that individuals with high conscientiousness exhibited lower accident rates in driving environments, underscoring the predictive role of conscientiousness in driving behavior⁸. Clarke et al.‘s meta-analysis emphasized that low conscientiousness and low agreeableness were effective predictors of accidents⁹. Similarly, Dahlen et al.‘s theoretical model indicated that high levels of driving anger and low agreeableness were associated with aggressive driving, which in turn led to accidents or violations¹⁰. Liu et al.‘s study categorized unsafe driving behaviors into violations, lapses, and cognitive errors, exploring their relationships with the Big Five personality traits¹¹. Wu et al., focusing on young male drivers, found that risky driving styles were linked to low agreeableness, low conscientiousness, and high exposure to risky situations¹². Wang et al. introduced the concept of a “personality baseline” and observed that high-risk scenarios had a greater impact on open and extraverted drivers¹³. Monteiro et al. tested a mediation model and found that traits such as anger, antisocial tendencies, and sensation-seeking indirectly influenced accident risks through risky driving behaviors¹⁴. Jovanović et al.‘s research revealed a significant association between neuroticism and driving anger, as well as direct links between aggression and the other four personality traits, supporting the utility of assessing overall personality traits to predict aggressive driving tendencies¹⁵.

Beyond the FFM, various innovative methods in personality research have been widely applied to the study of driving behavior. For example, Ulleberg et al., combining personality trait theories with the theory of planned behavior, found that traits such as aggression, altruism, and anxiety indirectly influenced young drivers’ risk-taking behaviors through their attitudes¹⁶. Sümer, in a proposed mediation model, demonstrated that personality factors acted as distal causes, jointly influencing accident occurrence with specific driving behaviors through mediating variables. This study also revealed that psychological symptoms and aggressive traits predicted risky driving behaviors¹⁷. Taubman-Ben-Ari’s Multidimensional Driving Style Inventory (MDSI) classified driving styles into eight dimensions, identifying significant associations between these dimensions and personality traits such as gender and age, thereby providing a methodological foundation for assessing driving styles¹⁸. Kontogiannis examined Greek drivers’ coping strategies under driving stress and found that individuals adopting an “active-confrontational” style exhibited higher violation rates, revealing a direct link between violations and accidents, as well as an indirect effect of aggression through a framework model¹⁹. Gulliver et al. showed that men with high levels of aggression and emotional detachment faced higher accident risks, offering a basis for targeted intervention strategies²⁰. Lucidi et al. identified three subtypes of young Italian drivers and analyzed their differences in driving behaviors using the Driver Behavior Questionnaire (DBQ)²¹. Pearson et al. studied the effects of impulsivity-related traits on risky driving and found that “positive urgency” was a strong predictor, highlighting the need to assess multiple dimensions of impulsivity²². Sucha compared high-risk drivers with revoked licenses to professional drivers without violation records, finding significant differences in personality traits between the two groups²³. Zicat et al. investigated the association between young drivers’ cognitive functions and driving performance, demonstrating that cognitive abilities could predict behaviors such as speeding and lane deviations²⁴. Marín et al. explored the personality traits of violating drivers in the Catalonia region²⁵, while Guo et al. applied Taubman-Ben-Ari’s inventory to Chinese drivers, demonstrating its structural validity and reliability²⁶. Liao et al. modeled driving styles, personality traits, and emotional states using fuzzy logic and random forest algorithms²⁷.

Furthermore, the Myers-Briggs Type Indicator (MBTI), with its unique theoretical perspective, provides a novel avenue for exploring the relationship between personality and driving behavior. Several studies have highlighted significant correlations between MBTI personality types and drivers’ violation behaviors, self-assessments of skills, and actual driving performance. For instance, Park and Lee found that extraverted and perceiving male drivers were more likely to engage in violation driving behaviors²⁸. McPeek et al. observed that extraverted, sensing, and confident older drivers tended to overestimate their abilities in skill assessments²⁹. In contrast, Classen et al.‘s study showed that extraverted and judging older drivers generally demonstrated better driving skills³⁰. These findings suggest that the MBTI framework has been preliminarily applied to the field of driving behavior, offering a novel perspective for examining the relationship between personality traits and driving performance.

Overall, existing research has explored the influence of personality traits on driving behavior through multiple theoretical frameworks, uncovering diverse relationships between personality and driving behavior and establishing a theoretical foundation for predicting driving risks and designing interventions. However, notable limitations remain. First, in terms of methodology, a lack of diversity in sample populations and insufficient contextualization of findings constrain the generalizability of some conclusions, making it difficult to account for different cultural backgrounds and complex driving scenarios. Second, research on personality traits has predominantly focused on static analyses, with limited attention to the dynamic changes in personality traits or the interactions between different dimensions. Furthermore, current studies have not sufficiently examined the specific mechanisms through which personality traits influence driving behavior via mediating variables. These limitations reduce the explanatory power of existing theories in complex driving situations and hinder progress in this field.

As the current mainstream framework for driving behavior research, the FFM has established a relatively mature theoretical ecosystem, providing an important foundation for predicting individual driving behavior. However, despite the wealth of research based on the FFM, its application to practical contexts and its contribution to improving research efficiency remain limited. On one hand, the FFM’s emphasis on continuous dimensional analysis introduces complexity in testing and application, which may hinder large-scale practical implementation. On the other hand, due to the limitations of existing research, the guidance provided by the FFM for personalized driving behavior prediction and the design of intervention measures is still inadequate.

In contrast, the MBTI offers practical advantages not only due to its widespread recognition and intuitive classification system but also because recent studies have demonstrated its reliability and validity in specific research contexts. Its straightforward framework simplifies personality trait descriptions, facilitating large-scale data collection and improving research efficiency. Additionally, its broad social acceptance enhances participant engagement and real-world applicability, providing a solid foundation for exploring personalized driving behavior. These strengths, combined with its growing scientific credibility, make the MBTI a suitable and effective framework for this study.

In response, the present study adopts the MBTI framework as its core perspective to systematically investigate the behavioral characteristics of drivers with different personality types in complex driving scenarios. This approach aims to address theoretical gaps in existing research, further elucidate the intrinsic connections between MBTI personality traits and driving behavior, and provide a theoretical and practical foundation for improving the precision and effectiveness of traffic safety management.

Research design on the relevance between MBTI personality dimensions and driver behavior characteristics

Overview of MBTI

The MBTI personality theory, developed by Myers and Briggs based on Carl Jung’s theory of psychological types, is widely used to assess an individual’s psychological preferences³¹. This theory categorizes personality into four dimensions: (1) Extraversion (E) - Introversion (I), focusing on the direction from which individuals derive energy, with extraverts deriving it from the external world and introverts from within themselves; (2) Sensing (S) - Intuition (N), focusing on how individuals take in information, with sensors perceiving through the five senses and intuitives perceiving through consciousness; (3) Thinking (T) - Feeling (F), focusing on how individuals make decisions, with thinkers relying primarily on logic and analysis, and feelers relying on personal values; (4) Judging (J) - Perceiving (P), focusing on how individuals adapt to the external environment, with judgers preferring to plan and organize in advance, and perceivers preferring to go with the flow and keep options open³¹, as shown in Fig. 1.

Fig. 1

Interpretation of the Four MBTI dimensions.

Research methods

This study adopts an integrated multi-method approach, combining personality testing, questionnaire surveys, in-depth interviews, field observations, and data analysis. This approach aims to ensure both breadth and depth in data collection and analysis.

Personality testing

To measure participants’ MBTI personality types, this study utilized the NERIS Type Explorer^® testing tool, available on the 16Personalities website³², as shown in Fig. 2. This tool, rooted in MBTI theory and enhanced by modern research methods, evaluates individuals across four dimensions: E-I, S-N, T-F, and J-P. The test comprises 60 forced-choice items and is highly standardized.

Fig. 2

MBTI personality test website interface.

To ensure the scientific rigor of the study, the reliability and validity of the NERIS Type Explorer^® tool were assessed using large-scale validation. Key findings are summarized below:

• (1) Internal Consistency: An analysis of 10,000 respondents revealed Cronbach’s α coefficients ranging from 0.75 to 0.87. Specifically, the α coefficient was 0.87 for the E-I dimension, 0.78 for S-I, 0.75 for T-F, and 0.82 for J-P³². These values indicate strong internal consistency within the scales.

• (2) Test-Retest Reliability: A longitudinal analysis involving 2,900 participants, with a 5–7 month interval between tests, showed test-retest reliability coefficients ranging from 0.74 to 0.83. Specifically, the E-I dimension scored 0.83, S-I scored 0.74, T-F scored 0.80, and J-P scored 0.79³². This demonstrates good temporal stability.

• (3) Discriminant Validity: An analysis of 10,000 respondents showed relatively low correlation coefficients between scales (absolute values ranging from 0.09 to 0.37). For instance, the correlation coefficient between S-I and J-P was 0.37³². These results confirm that each scale measures distinct personality dimensions with minimal overlap.

The statistical evidence supports the reliability and validity of the NERIS Type Explorer^® tool, making it a robust instrument for personality type assessment in this study.

Questionnaire survey

A questionnaire survey was employed to explore driving behavior. The survey consisted of 36 multiple-choice items, covering various aspects such as demographic information, personality traits, driving habits, and attitudes toward driving. The questionnaire design was informed by the theoretical framework of the classic DBQ and was optimized to meet this study’s research objectives, enhancing both its representativeness and originality.

The design process followed rigorous academic standards. Prior to formal distribution, the questionnaire underwent expert review by professionals in the transportation field. A small-scale pilot survey was conducted to refine the wording of items, logical structure, and measurement dimensions, ensuring the questionnaire’s content validity, scientific rigor, and practicality.

The finalized questionnaire was distributed through a combination of online and offline channels to encompass a diverse range of driver groups, including professional drivers and learner drivers. To ensure data quality, incomplete responses, those with unreasonably short completion times, or inconsistent answers were excluded. Only high-quality data were retained for analysis.

The questionnaire consisted of two main sections, as summarized in Table 1:

• (1) Basic Information Measurement: This section gathered participants’ MBTI personality types, age, gender, driving experience, and commonly driven vehicle types. These variables were used to characterize the sample and serve as control variables in subsequent analyses, minimizing potential confounding effects.

• (2) Driving Behavior Measurement (DBM): Based on established scales such as the classic DBQ, a combination of situational and frequency-based items was developed to capture driving behavior. Dimensions included speed control, maintaining safe following distances, lane-changing compliance, adherence to traffic rules, responses to traffic congestion, vehicle maintenance, and fatigue-related driving behavior. Situational items simulated real-world driving scenarios to enhance measurement accuracy, while frequency-based items captured the prevalence of specific behaviors.

Table 1 Structure of the driving behavior survey questionnaire.

In-depth interviews

The in-depth interviews aimed to uncover the underlying psychological motivations and situational factors influencing driving behavior using qualitative research methods. A semi-structured design was adopted, incorporating open-ended questions and specific situational discussions. The interviews focused on the following three aspects:

• (1) Causes of Dangerous Driving Behaviors: The interviews explored specific situations in which behaviors such as speeding, running red lights, and fatigued driving occurred, as well as the underlying psychological motivations driving these behaviors.

• (2) Emotion and Stress Management: Drivers’ emotional reactions and coping strategies when encountering traffic congestion, vehicle breakdowns, or other emergencies were analyzed to understand the role of emotional regulation during driving.

• (3) Understanding and Attitudes Toward Traffic Rules: Participants’ understanding of traffic regulations, their willingness to comply, and their subjective evaluations of rule observance were examined to explore the influence of cognitive awareness of rules on actual driving behavior.

All interviews were conducted based on a predetermined outline, with audio recordings and transcriptions prepared for subsequent systematic qualitative analysis.

Field observations

Field observations were conducted to gather data from real-world driving scenarios, aiming to verify the consistency between questionnaire and interview findings while further uncovering the dynamic characteristics of driving behavior. The observations were carried out in the following steps:

• (1) Defining Observation Scenarios and Targets: Researchers observed drivers’ behavioral characteristics during actual driving, including speed, following distance, lane-changing behavior, and turn signal usage. These observations were conducted either through in-vehicle monitoring or using vehicle-installed cameras.

• (2) Collecting Core Behavioral Data: The frequency, duration, and intensity of drivers’ behaviors were recorded systematically.

• (3) Minimizing External Interference: Efforts were made to avoid disrupting drivers’ normal behavior to ensure the authenticity and naturalness of the data. Measures were also taken to minimize potential reactivity biases caused by the observation process itself.

The data collected through field observations were compared with the results from questionnaires and interviews to verify consistency and identify potential sources of discrepancies.

Data analysis

This study employed a combination of quantitative and qualitative analysis methods, integrating statistical tools and modeling techniques to systematically analyze the collected data. The specific methods used are as follows:

• (1) Thematic Analysis of Qualitative Data: The qualitative data obtained from in-depth interviews and field observations were coded to extract key themes related to driving behavior. These themes were then integrated and compared with quantitative data to verify consistency across different data sources and to identify potential sources of discrepancies.

• (2) Correlation Analysis and Pattern Mining: Chi-square tests were used to examine significant relationships between MBTI personality dimensions and driving behavior characteristics. For variables without significant correlations, a reverse screening method was applied to uncover complex relationship patterns and identify underlying interactions between variables.

• (3) Behavior Pattern Analysis Using Clustering: The K-modes clustering algorithm was applied to combine MBTI personality types with driving behavior characteristics, classifying the sample population into groups with similar behavior patterns. In-depth analyses were conducted to examine the characteristic differences between these groups. This approach provided insights into group-specific driving behavior traits and laid the foundation for designing targeted intervention measures.

Research process

The study involved 593 drivers with valid licenses and actual driving experience as research participants. The research process was divided into the following five stages, as illustrated in Fig. 3:

• (1) Preparation: This stage included designing the research framework, specifying the research steps, and selecting appropriate methods. A risk assessment was also conducted to anticipate and address potential challenges.

• (2) Data Collection: Data collection primarily involved the use of standardized MBTI scales and a self-designed driving behavior characteristics questionnaire. Additionally, in-depth interviews and field observations were employed to validate and complement the collected data.

• (3) Data Cleaning: This phase involved data integration and quality control, including data verification, checking, and preliminary evaluation, to ensure the reliability of the dataset for analysis.

• (4) Data Analysis: Chi-square tests were conducted on the screened valid data for an overarching analysis. A reverse screening method was then used to refine the target dataset. Subsequently, K-modes clustering analysis was performed to explore correlations between MBTI personality types and driving behavior characteristics in depth.

• (5) Research Summary: The results from each stage of the analysis were synthesized to draw final conclusions. The research findings were integrated and evaluated comprehensively.

Fig. 3

Research process.

MBTI and driver behavior characteristics overall analysis

Description of participant characteristics

This study primarily focused on middle-aged drivers, striving to achieve a balanced and comprehensive sample in terms of gender ratio, age distribution, and driving backgrounds. Data collection employed a combination of online and offline methods. The sample demonstrated diversity and representativeness in characteristics such as gender, age, and driving experience. During data collection, priority was given to target groups with higher willingness to participate and greater accessibility to enhance data collection efficiency while minimizing potential sampling bias.

A total of 593 initial responses were collected. After data cleaning, 571 valid questionnaires were retained, resulting in an effective response rate of 96.3%. The distribution of the cleaned sample across gender, age, driving experience, and other variables remained largely consistent with the original data, indicating that data cleaning had minimal impact on the sample’s representativeness. The specific sample characteristics are outlined below:

(1) Age Characteristics

To ensure the sample’s applicability and representativeness, this study targeted licensed drivers aged 18 to 60, encompassing the behavioral characteristics of most active driving populations. According to the data collected, the age distribution of the sample showed some randomness but was primarily concentrated within the target range. Specifically, 8.23% of participants were aged 18–25, 36.95% were aged 26–35, 39.93% were aged 36–45, and 14.89% were aged 46 and above. The relatively balanced age distribution reflects the behavioral characteristics of drivers within this range.

(2) Gender Characteristics

The sample included both male and female drivers, with a relatively balanced gender ratio. Males accounted for 56.74% of the sample, while females accounted for 43.26%. This ratio aligns with the typical gender distribution within driving populations and provides an adequate basis for exploring the influence of gender on driving behavior in real-world contexts.

(3) Driving Experience Characteristics

Participants’ driving experience spanned multiple levels, offering a comprehensive representation of drivers with varying levels of expertise. The specific distribution was as follows: 9.46% had less than one year of driving experience, 22.59% had 1–3 years, 39.05% had 4–5 years, and 28.90% had more than five years. This distribution effectively captured the behavioral traits of both novice and experienced drivers.

(4) Vehicle Type Characteristics

To ensure comprehensive coverage of vehicle types, the study investigated participants’ primary vehicle types. Sedans accounted for 72.15% of the sample, SUVs for 18.39%, pickups for 3.68%, commercial vehicles for 3.33%, and other vehicle types for 2.45%. This distribution reflects current trends in vehicle type preferences among driving populations.

The six driving behavior facets

Driving behavior is a complex cognitive and decision-making process that involves not only a driver’s perceptions, judgments, and reactions to the external environment but also the regulation of internal emotions and psychological states. In psychological research, the concept of the “cognitive map” has been widely used to explain how individuals internally represent the spatial organization of the external world³³. According to cognitive map theory, driving behavior can be seen as the external manifestation of an individual’s internal spatial representation. Based on this framework, this study proposes six core facets of driving behavior:

(I) Response Readiness

Cognitive map theory highlights an individual’s ability to navigate the environment, which encompasses not only memory of spatial locations but also sensitivity and adaptability to environmental changes. In driving, response readiness refers to the driver’s ability to quickly adapt to emergencies, such as avoiding collisions or selecting alternative routes during traffic congestion. Drivers with well-developed cognitive maps are better equipped to handle such situations.

(II) Mental Composure

A driver’s psychological state significantly influences their driving behavior. Excessive cognitive load can hinder the effective use of cognitive maps, thereby impairing decision-making. Maintaining a relaxed state is critical for the accuracy and efficiency of cognitive processing. Drivers who remain calm are less likely to experience psychological stress, reducing the likelihood of accidents.

(III) Safety Awareness

Safety awareness refers to a driver’s ability to identify, predict, and respond to potential risks in their environment and is a core component of cognitive map theory. Through cognitive maps, drivers develop an understanding of potential hazards in traffic and adopt preventive measures to minimize accident risks. For instance, in complex traffic scenarios, a driver’s safety awareness determines whether they can identify dangers promptly and take appropriate actions to avoid them.

(IV) Courteous Conduct

Safety awareness refers to a driver’s ability to identify, predict, and respond to potential risks in their environment and is a core component of cognitive map theory. Through cognitive maps, drivers develop an understanding of potential hazards in traffic and adopt preventive measures to minimize accident risks. For instance, in complex traffic scenarios, a driver’s safety awareness determines whether they can identify dangers promptly and take appropriate actions to avoid them.

(V) Risk Inclination

Risk inclination refers to how drivers assess and respond to risks. The representation of risk elements within a driver’s cognitive map directly influences their driving style. For example, drivers with conservative tendencies may overemphasize risk areas in their cognitive maps, leading to cautious driving, while risk-prone drivers may underestimate risks, resulting in more aggressive behavior.

(VI) Emotional Resilience

Emotional resilience is the ability to regulate emotions under pressure or during emergencies. Emotional states have a significant influence on the construction and retrieval of cognitive maps. Drivers with high emotional resilience can maintain the stability of their cognitive maps in stressful situations, leading to more rational decisions. For example, in traffic congestion, emotionally resilient drivers are more likely to remain calm and avoid errors caused by emotional instability.

Distribution of DBMs across the six behavioral facets

To comprehensively assess driving behavior, this study designed 31 DBMs. These DBMs correspond to the six behavioral facets, as outlined in Table 2 below.

Table 2 shows the specific distribution of DBMs across the six facets:

• I. Response Readiness: DBMs 15, 17, 18, 21, 23, 25, and 26.

• II. Mental Composure: DBMs 2, 6, 7, 16, 24, and 29.

• III. Safety Awareness: DBMs 4, 14, 27, and 30.

• IV. Courteous Conduct: DBMs 5, 8, 10, 12, 22, and 31.

• V. Risk Inclination: DBMs 1, 3, 13, 19, and 28.

• VI. Emotional Resilience: DBMs 9, 11, and 20.

Table 2 Classification of DBMs within the six behavioral facets.

Reliability and validity analysis

Reliability analysis primarily focuses on the internal consistency of the questionnaire, i.e., whether different items consistently measure the same concept. To evaluate this, the present study employed Cronbach’s α coefficient, which estimates the average correlation among items in a questionnaire or test. The formula for calculating Cronbach’s α is as follows:

$$\:\alpha\:=\frac{N\stackrel{-}{c}}{\stackrel{-}{v}+(N-1)\stackrel{-}{c}}$$

(1)

where N is the number of items, c̅ is the average covariance among all items, and v̅ is the average variance of all items.

Cronbach’s α ranges from 0 to 1, with values greater than 0.7 generally considered to have acceptable internal consistency. The reliability of the overall questionnaire and each facet was calculated using the statistical software SPSS, with the following results shown in Table 3:

Table 3 Cronbach’s α coefficient of the whole questionnaire and each facet.

The reliability analysis indicated that the overall Cronbach’s α coefficient of the questionnaire was 0.82, demonstrating high internal consistency reliability. The Cronbach’s α coefficients for all facets exceeded the threshold of 0.7, satisfying the general standards for reliability evaluation. Specifically, the “Response Readiness” and “Mental Composure” facets exhibited robust internal consistency reliability, with Cronbach’s α values of 0.84 and 0.82, respectively. Although the “Emotional Resilience” facet showed a relatively lower Cronbach’s α of 0.71, it remained within the acceptable range for reliability assessment.

Validity analysis was conducted using Exploratory Factor Analysis (EFA) as the primary method. Prior to the factor analysis, data suitability tests were performed. The Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy was 0.85, exceeding the recommended threshold of 0.7, indicating that the dataset was appropriate for factor analysis. Additionally, Bartlett’s test of sphericity was significant ($\:{{\upchi\:}}^{2}=1249.13，\text{d}\text{f}=630,\:\:p<0.001$), confirming the presence of significant correlations among variables.

Factor extraction was performed using Principal Component Analysis (PCA), with Varimax rotation employed to optimize the factor structure and enhance interpretability. Based on the criterion of eigenvalues greater than 1, six factors were extracted, collectively accounting for 68.4% of the total variance. This result meets the statistical standards commonly used in social science research. The analysis supported the hypothesized conceptual framework, as all items loaded significantly onto their respective factors, with factor loadings exceeding 0.5, further validating the questionnaire’s structural integrity.

Analysis of questionnaire data characteristics

The driving behavior questionnaire utilized in this study consisted of 31 DBMs, with each item offering 2–4 response options to capture drivers’ behavioral tendencies. By examining the percentage distributions of responses for each option, insights into the preference patterns underlying driving behaviors were obtained.

Results revealed that for some DBMs, a single option was selected significantly more frequently than others, indicating high consistency in drivers’ behavioral traits. In contrast, other DBMs exhibited more diverse response patterns, reflecting a broader distribution of drivers’ views and habits. Detailed statistics on the response distributions for each behavioral measure are presented in Fig. 4.

Fig. 4

Statistical distribution of DBMs in the survey questionnaire.

As shown in the figure, the distribution of responses highlights both the diversity and the commonalities in drivers’ behavioral tendencies. For items with a clear concentration of responses, the results suggest shared norms or widely accepted perceptions of driving risks. On the other hand, the dispersed response patterns provide a basis for exploring individual differences in driving behaviors. These findings offer valuable insights into the correlations between MBTI personality dimensions and specific driving characteristics, which will serve as a focal point for further analysis.

Chi-square test and reverse screening

This study employed the Chi-square test, supplemented by a reverse screening approach, to analyze questionnaire data and identify patterns in the dispersion of behavioral measures. This methodology enabled a more targeted investigation into the variability of driving behaviors across different MBTI personality types.

The Chi-square test is a robust statistical tool used to evaluate the independence between two categorical variables. By comparing observed frequencies with expected frequencies under the null hypothesis of independence, the test determines whether a statistically significant association exists. If the deviation between observed and expected frequencies exceeds a critical threshold, the null hypothesis is rejected, suggesting a potential correlation between the variables.

Reverse Chi-square screening builds on the Chi-square test results by adopting a complementary perspective. Unlike traditional methods that prioritize statistically significant variables, reverse screening focuses on excluding variables with low statistical significance—those whose Chi-square statistics fall below the critical threshold. This method streamlines the analytical model, mitigates overfitting, and prioritizes variables with greater potential for meaningful insights. By concentrating on less statistically prominent variables, reverse screening facilitates the discovery of subtle patterns that might otherwise be overlooked.

In practice, we began by conducting Chi-square tests on all DBMs, computing their respective Chi-square statistics. These statistics were subsequently compared to the critical values associated with their degrees of freedom. DBMs with Chi-square statistics surpassing the critical threshold generally reflect universal or shared characteristics among respondents. On the other hand, DBMs with Chi-square statistics falling below the threshold did not demonstrate statistical significance, potentially suggesting underlying differences among various MBTI personality types. These less prominent items merit further exploration, as they may uncover unique behavioral traits that are either dispersed or context-specific.

The implementation of the Chi-square test and reverse screening in this study followed the steps outlined below:

(1) Determining the expected frequency.

The first step involved calculating the expected frequency for each response option under the null hypothesis of equal probability. The formula used is:

$$E_{i} = \frac{N}{k}$$

(2)

where N represents the total number of choices for all options of a given question by all participants, and k is the number of options provided for that question.

(2) Calculating the chi-square statistic

To quantify the deviation between the observed and expected frequencies, the chi-square statistic is calculated using the following formula:

$$\:{x}^{2}=\sum\:\left(\frac{{({Q}_{i}-{E}_{i})}^{2}}{{E}_{i}}\right)$$

(3)

where Q_i is the observed frequency, and E_i is the expected frequency.

(3) Significance testing.

This step evaluated whether the observed frequency distribution significantly deviated from the expected distribution. The p-values for each DBM were calculated based on the Chi-square distribution table. To address the issue of inflated family-wise error rates (FWER) due to multiple testing, the Bonferroni correction was applied. This adjustment revised the significance level from 0.05 to 0.0016 ($\:{\alpha\:}_{adjusted}$= 0.05/31).

At the adjusted significance level (α = 0.0016), the critical values for the Chi-square test were determined as follows:

For df = 1, the critical value is approximately 10.828.

For df = 2, the critical value is approximately 13.816.

For df = 3, the critical value is approximately 16.266.

Using these thresholds, a significance analysis was conducted on the Chi-square statistics for the 31 DBMs, and the results are presented in Table 4.

Table 4 Chi-square statistics data of 31 DBMs in the driving questionnaire survey.

From the analysis, it was observed that the Chi-square statistics for DBMs numbered [1–10, 12, 14, 15, 20, 24, 27, 30, and 31] were significantly greater than their corresponding critical values, indicating a high level of significance for these items. For DBMs numbered [11, 13, 17, 18, 19, 23, 25, and 26], the Chi-square statistics also exceeded their critical values, but with relatively lower magnitudes, suggesting a less pronounced level of significance. However, for DBMs numbered [12, 16, 21, 22, 28, and 29], the Chi-square statistics fell below their critical values, indicating a lack of significant correlation for these six DBMs.

(4) Reverse screening and analysis focus

To address the limitations of conventional approaches, this study adopted an innovative reverse analysis strategy, focusing on DBMs with low Chi-square statistics that failed to meet the significance threshold.

This approach is grounded in the following theoretical rationale: DBMs with high Chi-square statistics and significant results indicate consistent behaviors among individuals with different personality types. While such consistency highlights commonalities, it may obscure the nuanced mechanisms through which personality traits influence driving behavior at a granular level. Conversely, DBMs lacking statistical significance may point to more complex and subtle associations, reflecting intricate relationships between personality traits and driving behavior. These variations, though statistically inconspicuous, may stem from unique personality characteristics and offer critical insights into the dynamic interplay between personality and behavior.

Therefore, this study refocuses its analytical attention on these six DBMs, which exhibit a lack of statistical significance in their response distributions, categorizing them as Homogeneous-Response Driving Behavior Measures (HR-DBMs). By examining these HR-DBMs in greater detail, the study seeks to uncover potential patterns and nuanced relationships between personality traits and driving behavior.

Cluster analysis of HR-DBMs

Delineation of HR-DBMs

Based on the results of reverse screening using the chi-square test, this study identified DBMs [12, 16, 21, 22, 28, and 29] as HR-DBMs. HR-DBMs are defined as DBMs whose response distributions lack significant correlation, indicating high consistency across the sample and minimal influence from external interference or bias. For ease of subsequent analysis, these HR-DBMs were categorized into three of the Six Behavioral Facets of driving behavior: (II) Relaxed State, (IV) Courtesy and Consideration, and (V) Risk-Taking Tendency, as detailed in Table 5.

Table 5 Classification of HR-DBMs within the six behavioral facets.

Among these, HR-DBMs 16 and 29 reflect the driver’s Mental Composure, HR-DBMs 12 and 22 are associated with the driver’s Courteous Conduct, and HR-DBMs 21 and 28 correspond to the driver’s Risk Inclination. These HR-DBMs retain the original numbering of DBMs, as they represent a subset identified through reverse screening and statistical analysis. To further elucidate the specific meanings of these HR-DBMs and their roles in driving behavior, Table 6 provides the corresponding questionnaire items and their detailed descriptions.

Table 6 Questionnaire items and descriptions of the HR-DBMs.

The existence of these six HR-DBMs suggests the presence of distinct intrinsic correlations between MBTI personality dimensions (including E-I, S-N, T-F, and J-P) and three of the Six Behavioral Facets of driving behavior. This study will further investigate the relationships between these HR-DBMs and MBTI personality traits by utilizing insights derived from their homogeneous response patterns.

K-modes clustering process

To perform cluster analysis on the HR-DBMs in the sample data, this study employed the K-Modes algorithm. Derived from the K-Means algorithm, K-Modes is specifically designed for clustering categorical data. Unlike the K-Means algorithm, which primarily focuses on numerical data, K-Modes is more suitable for analyzing discrete data (such as categorical variables) and can effectively handle non-numerical attributes of sample features. The following are the specific steps involved in the K-Modes algorithm:

(1) Initialization

Based on the three binary features (“Courteous Conduct,” “Risk Inclination,” and “Emotional Resilience”) contained in the research sample, the study selected the number of clusters K = 8 to cover all possible combinations of high and low feature values. Subsequently, eight samples were randomly selected as the initial cluster mode centers, each representing the initial state of a cluster.

(2) Calculation of Dissimilarity Score

For each sample in the dataset, calculate its dissimilarity score with the K = 8 cluster centers to evaluate its similarity to each cluster. The dissimilarity score is defined as follows:

$$d\left( {X,~Y} \right)=\mathop \sum \limits_{{j=1}}^{m} \delta \left( {{x_j},{y_j}} \right)$$

(4)

where $\:X=({x}_{1},{x}_{2},…,{x}_{m})$ is the feature vector of the data point, $\:Y=\left({y}_{1},{y}_{2},…,{y}_{m}\right)$ is the feature vector of the cluster center, m is the total number of features, and $\:\delta\:({x}_{j},{y}_{j})$ is the indicator function, $\:\delta\:=1$ if $\:{x}_{j}\ne\:{y}_{j}$ and $\:\delta\:=0$ if $\:{x}_{j}={y}_{j}$.

This formula represents the number of mismatched features between the data point X and the cluster center Y. A lower score indicates higher similarity between the sample and the cluster center.

(3) Assignment of Data Points to the Nearest Cluster

Based on the calculated dissimilarity scores, assign each data point to the cluster whose center has the smallest dissimilarity score with that point. After this step, the dataset is divided into K = 8 clusters.

(4) Update Cluster Centers

For each cluster, recalculate the cluster center. The new cluster center is determined by the mode (most frequent value) of all samples within the cluster for each feature, ensuring consistency in feature value distribution within the cluster. The update formula is as follows:

$${c_j}=\arg {\hbox{max} _{\nu \in {V_j}}}\sum\nolimits_{{x \in C}} {II\left( {{x_j}=\nu } \right)}$$

(5)

where $\:{c}_{j}$ is the updated value of the cluster center for the j feature, $\:{V}_{j}$ is the set of all possible values for the j feature, C is the set of all samples within the current cluster, and$\:II({x}_{j}=v)$ is the indicator function, taking the value 1 if the feature value $\:{x}_{j}=v$ within the cluster, and 0 otherwise.

The new cluster center maximizes the consistency within the cluster.

(5) Iterative Update:

Repeat steps (2) to (4) until the clustering results no longer change or the objective function converges. The objective function is the sum of dissimilarity scores of all samples within the clusters to their respective cluster centers, defined as:

$$\:Cost=\sum\:_{i=1}^{K}\sum\:_{{X}_{j}\in\:{C}_{i}}d({X}_{j},{Q}_{i})$$

(6)

where $\:K$ is the total number of clusters, $\:{C}_{i}$ is the i cluster, $\:{X}_{j}$ is the j sample within the i cluster, $\:{Q}_{i}$ is the mode of the i cluster; $\:d({X}_{j},{Q}_{i})$ is the dissimilarity score of sample $\:{X}_{j}$ from its cluster center $\:{Q}_{i}$.

Through iterative optimization, the K-Modes algorithm ultimately obtains stable clustering results.

The choice of K in the K-Modes algorithm was guided by a combination of theoretical reasoning, data characteristics, and clustering performance evaluation. The three binary features in the dataset (“Courteous Conduct,” “Risk Inclination,” and “Emotional Resilience”) theoretically yield ($\:{2}^{3}\:$= 8) distinct combinations of high and low feature values. This theoretical framework provided an initial rationale for selecting K = 8, aiming to capture all potential driver behavior patterns.

To validate the choice of K = 8, clustering experiments were conducted with different values of K (e.g., K = 4 to K = 10). The clustering results were evaluated using standard metrics such as total dissimilarity (inertia) and silhouette scores. The results demonstrated that K = 8 achieved an optimal balance between within-cluster homogeneity and between-cluster separability, while also producing a clear representation of distinct behavioral patterns. Additionally, the clustering results for K = 8 were consistent with the theoretical combinations of the binary features, further supporting its suitability for this analysis.

By following the defined steps of the K-Modes algorithm and systematically evaluating the clustering outcomes, this study identified K = 8 clusters as the most appropriate choice. This approach ensured that the clustering process was both theoretically grounded and empirically validated, offering a structured and meaningful representation of the sample data’s behavioral patterns.

Analysis of clustering results

Overall clustering characteristics of HR-DBMs

In this study, the K-Modes algorithm was used to divide the samples into 8 independent clusters, each representing a specific combination of driving behavior characteristics. The specific category features are shown in Table 7.

Table 7 Clustering distribution of the HR-DBMs.

The 8 clusters comprehensively group all participant samples based on their predefined MBTI personality traits. These clusters represent the eight permutations across three of the Six Behavioral Facets developed in this study. The clustering process facilitates a deeper exploration of MBTI personality characteristics within each cluster.

Distribution patterns of MBTI personality dimensions across clusters

After completing the K-Modes clustering process, the study further analyzed the distribution patterns of MBTI personality types across the different clusters. Given the imbalanced distribution of MBTI categories in the overall sample (e.g., a significantly higher number of E types compared to I types), directly using absolute numbers for comparison could introduce bias. To address this, a standardized proportion method was adopted. This method calculates the proportion of a specific category within a cluster relative to the total number of that category in the overall sample, thereby mitigating the influence of population size differences.

The standardized proportion is calculated using the following formula:

$$\:{P}_{i,j}=\frac{{n}_{i,j}}{{N}_{j}}\times\:100\%$$

(7)

where $\:{P}_{i,j}$ is the standardized proportion of MBTI category j in cluster $\:{C}_{i}$, $\:{n}_{i,j}$ is the number of samples belonging to MBTI category j in cluster $\:{C}_{i}$, $\:{N}_{j}$ is the total number of samples belonging to MBTI category j in the overall sample, i = 1,2,…,8 (corresponding to the 8 clusters), and j represents the specific MBTI category (e.g., E, I, S, N, T, F, J, P).

Using this formula, the standardized proportions for each cluster were calculated, as shown in Table 8.

Table 8 The standardized proportion of MBTI personality dimensions in different clusters.

(1) Cluster Analysis Results for the E-I Dimension

Extraversion (E): The highest proportion (23.05%) occurs in Cluster C₁, indicating that extraverted individuals’ driving behaviors are predominantly concentrated in this cluster. In contrast, the lowest proportion (5.58%) is observed in Cluster C₄, suggesting that extraverted individuals are less associated with the driving behaviors represented by this cluster (see Fig. 5).

Introversion (I): The highest proportion (18.87%) is found in Cluster C₃, suggesting that the driving behaviors of introverted individuals align more closely with the characteristics of Cluster C₃. Conversely, the lowest proportion (7.28%) is observed in Cluster C₈, indicating a weaker association between introverted individuals and the driving behaviors represented by this cluster (see Fig. 5).

Fig. 5

Clustering results of E/I dimensions.

(2) Cluster Analysis Results for the S-N Dimension

Sensing (S): The highest proportion (28.85%) is found in Cluster C₁, suggesting that sensing individuals are more likely to exhibit the driving behaviors represented by this cluster. The lowest proportion (3.95%) is observed in Cluster C₈, indicating a weaker connection between sensing individuals and the driving behaviors of this cluster (see Fig. 6).

Intuition (N): The highest proportion (15.09%) occurs in Cluster C₈, implying that intuitive individuals are more commonly associated with the driving behaviors of this cluster. The lowest proportion (9.75%) is found in Cluster C₂, suggesting a less prominent alignment of intuitive individuals with the driving behaviors represented by this cluster (see Fig. 6).

Fig. 6

Clustering results of S/N dimensions.

(3) Cluster Analysis Results for the T-F Dimension

Thinking (T): The highest proportion (18.17%) is observed in Cluster C₆, indicating that thinking individuals dominate the driving behaviors associated with this cluster. The lowest proportion (7.69%) is found in Cluster C₃, suggesting a weaker representation of thinking individuals in this cluster’s driving behaviors (see Fig. 7).

Feeling (F): The highest proportion (17.89%) is observed in Cluster C₂, suggesting that feeling individuals are more strongly aligned with the driving behaviors of this cluster. The lowest proportion (6.67%) occurs in Cluster C₈, indicating a weaker connection between feeling individuals and the driving behaviors represented by this cluster (see Fig. 7).

Fig. 7

Clustering results of T/F dimensions.

(4) Cluster Analysis Results for the J-P Dimension

Judging (J): The highest proportion (26.69%) is found in Cluster C₁, indicating that judging individuals’ driving behaviors are closely associated with this cluster. The lowest proportion (4.99%) is observed in Cluster C₈, suggesting a weaker alignment of judging individuals with the driving behaviors represented by this cluster (see Fig. 8).

Perceiving (P): The highest proportion (17.83%) occurs in Cluster C₄, suggesting that perceiving individuals are more aligned with the driving behaviors represented by this cluster. The lowest proportion (5.22%) is found in Cluster C₃, indicating a weaker association between perceiving individuals and the driving behaviors of this cluster (see Fig. 8).

Fig. 8

Clustering results of J/P dimensions.

Across the four MBTI dimensions, distinct driving behavior patterns are observed among individuals with different personality traits. These findings establish a deeper connection between personality traits and driving behaviors, offering new perspectives and potential intervention points for traffic research.

Research conclusions

Research findings

This study, which focuses on the relationship between MBTI personality theory and driving behaviors, has achieved the following key findings:

• (1) Validation of MBTI theory’s applicability: By analyzing empirical driving behavior data, this study confirmed the feasibility of applying MBTI theory in the field of traffic psychology.

• (2) Development of the Six Driving Behavioral Facets framework: This framework captures the complexity of driving behaviors across multiple dimensions, serving as a novel analytical tool to investigate the relationship between personality traits and behavioral patterns.

• (3) Introduction of the “inverse chi-square test” method: Building on the traditional chi-square test, the study innovatively developed the “inverse chi-square test” method. This approach focuses on data items with insignificant differences (HR-DBMs), uncovering potential patterns hidden within non-significant results.

• (4) Identification of clustering patterns between personality traits and driving behaviors: For homogeneous data items identified through the inverse chi-square test, the study employed the K-Modes clustering algorithm for further analysis. This revealed significant imbalanced distribution patterns in the four MBTI dimensions within each cluster. The polarized distribution patterns statistically validate the systematic association between MBTI traits and driving behavior.

These findings highlight the profound impact of personality traits on driving behavior, offering a theoretical foundation for future research into the interaction between personality traits and autonomous driving technologies.

Practical implications

The findings of this study offer significant practical value for traffic management and safety interventions. Specifically:

• (1) Adaptive driver assistance systems (ADAS): The Six Driving Behavioral Facets framework and the clustering patterns between MBTI personality traits and driving behaviors provide actionable insights for developing ADAS. By incorporating personality traits into these systems, feedback and intervention strategies can be personalized to improve driver adaptability and enhance road safety.

• (2) Personalized warnings and interventions: The study’s findings suggest that personality traits influence drivers’ acceptance of and adaptability to warning mechanisms. This insight can guide the design of personalized warning systems in autonomous vehicles, ensuring better human-machine interaction³⁴.

• (3) Driver training and education programs: Policymakers can leverage personality profiling to design targeted educational campaigns and training programs aimed at enhancing driver performance and reducing traffic risks.

By bridging the gap between theoretical research and practical application, this study contributes to the development of safer and more personalized traffic systems, paving the way for innovations in autonomous vehicle technologies.

Research limitations and future prospects

Although this study has achieved several innovative results, certain limitations exist, and future research can further improve upon the following aspects:

• (1) Expansion of sample scope: The sample in this study was primarily drawn from a specific group, potentially lacking representativeness. Future research could expand the sample scope to include more diverse groups in terms of age, gender, and cultural backgrounds to validate the generalizability of the research conclusions.

• (2) Dynamic exploration of driving behavior variables: The current classification framework for driving behavior characteristics is mainly based on static variables. Future research could attempt to introduce dynamic variables to enhance the ecological validity of the analysis results.

• (3) Cross-theory integration and multi-dimensional modeling: This study primarily explored the association between personality and driving behavior based on MBTI theory. Future research could integrate other psychological theories to construct multi-level comprehensive models, providing a more comprehensive understanding of the influence of personality factors on driving behavior.

In conclusion, this study validated the systematic association between MBTI personality traits and driving behavior patterns using innovative theories and methods. It revealed the mechanisms by which personality traits influence driving behavior, expanding theoretical perspectives in traffic psychology and offering actionable insights for personalized traffic management and safety interventions. These findings hold substantial academic value and practical significance.