Measures and data sources

Departing from conventional metrics, we introduce a Road Crash Severity Index (SI) to address the significant gap in road safety research (Table 1). Whereas prior studies predominantly relied on crash frequency data (e.g., total fatalities), our SI variable—the ratio of deaths to total injuries— explicitly measured the severity of crash outcomes, indicating the relative risk and seriousness of road crashes in terms of human casualties. This measure is not only less susceptible to fluctuations in exposure data but also serves as a crucial indicator of the underlying risk of death in a crash. Consequently, the SI provides a powerful lens through which to evaluate the life-saving efficacy of a nation’s post-crash care and emergency response infrastructure, a dimension often overlooked in socioeconomic models of road safety. We extracted data for this index from the UAE Federal Government database (Bayanati [36], and Al Amir [37]). The study also included road safety indicators and socio-economic variables with high availability and acceptable quality from well-established secondary sources. Using the UAE as a case study, we obtained time-series data on the country’s GDP growth per capita, unemployment rate (the number of unemployed people as a percentage of the labour force), and population density (the number of people living in one square kilometre). These data were obtained from the World Bank Database (World Bank [38]) and the World Population Prospects 2024 database (UNDESA [39]). The variable “FINE” reflects the total fines paid by drivers for traffic violations in UAE dirhams (1 USD = 3.67 dirhams), serving to assess the influence of public behavior, attitudes, and law enforcement on road safety. Higher values indicate improved enforcement and enhanced road safety.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. The study road safety and socio-economic indicators, UAE data 1980–2024.

https://doi.org/10.1371/journal.pone.0341441.t001

The introduction and literature review demonstrate a strong theoretical and empirical consensus on the interdependence between selected socioeconomic variables and road traffic crashes or casualties.

Analytical framework: The ARDL approach.

The Autoregressive Distributed Lag (ARDL) approach and error correction models (ECM) are effective cointegration methodologies that can facilitate the analysis of dynamic short- and long-term relationships between socio-economic variables and road safety indicators. The ARDL model is particularly useful because it allows the inclusion of variables of different orders of integration (I(0) or I(1)) without losing information and provides robust estimates, even in small samples [40]. The ARDL approach can effectively capture the instantaneous and delayed impacts of economic development on road safety outcomes by enabling simultaneous estimation of long-term equilibrium relationships and short-term dynamics. The error correction model complements this by indicating how quickly the variables return to equilibrium after a shock [41,42]. Compared to traditional time-series techniques, which often require the variables to be integrated in the same order and can suffer from issues of overparameterization, ARDL and ECM offer greater flexibility and efficiency, making them particularly advantageous for analyzing complex interactions in socio-economic studies.

Our current investigation tracked socio-economic and road safety indicators from 1980 to 2024. The key indicators used to monitor changes in socio-economic conditions in the UAE are annual GDP per capita (GDP), unemployment rate (UNEMP), and population density per square kilometer (DEN). We used the variable (FINE) to control for the impact of law enforcement on safety outcomes. Equation (1) illustrates the ARDL model that uses the road crash severity index (SI) as the dependent variable. We can formulate similar ARDL models using GDP, UNEMP, DEN, and FINE as dependent variables.

(1)

where , , , , and are the log-transformed road safety severity index, average GDP per capita, unemployment rate, population density, and total annual fines paid for traffic violations, respectively, with corresponding lag lengths of p, q, r, s and z. The difference operator is denoted as Δ. Here, φ_is represents the long-term coefficients of the model and , , , and represent the short-term dynamic coefficients. The white noise error term is denoted .

Equation (1) is estimated using the ordinary least squares (OLS) method to specify the joint significance of the coefficient of lagged variables () using the F-test, comparing the F-statistic to Pesaran et al.‘s critical values [40]. The optimal lag order is selected using the Schwarz information criterion (SC). The null hypothesis of no cointegration relationship between the variables is expressed as Rejecting suggests the presence of a long-term relationship among variables.

If there is cointegration (i.e., a long-term relationship between socio-economic variables and road safety), the relationship between the variables is not a transient situation but rather a more stable one that can be recovered whenever there is a disturbance or shock. What is the speed of adjustment to long-term equilibrium after a deviation occurs in the short term? Consequently, we describe the Granger causality and error correction representation of the ARDL model as follows:

(2)

is the lagged OLS residual is obtained from the long-term model ( The coefficient of , is the speed of adjustment to long-term equilibrium to ensure convergence towards long-term equilibrium, , should be negative and significant. The model was considered unstable and explosive if was positive. If is significant, this also means that the regressors Granger cause the dependent variable in the long term. The causal impact was measured by testing the (t-test) coefficients , and . If there are several lags in the regressors, a joint F-test (Wald test) can be performed to infer the joint short-term causality.

Assessing UAE road safety data

Table 2 summarizes the log-transformed variables used in the analysis (1980–2024), providing essential context for model interpretation. The statistics indicate that GDP per capita exists on a substantially higher logarithmic order of magnitude than the other indicators, while Population Density shows the greatest relative dispersion. The proximity of the mean and median for most variables suggests the log transformation has effectively normalized their distributions, meeting a key precondition for reliable regression analysis and meaningful comparison of variable effects.

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Summary statistics of log-transformed variables (1980–2024).

https://doi.org/10.1371/journal.pone.0341441.t002

As illustrated by the Road Crash Severity Index (Fig 1), the United Arab Emirates experienced a non-linear progression in road safety between 1980 and 2024. Historical data indicate an overarching fluctuating pattern, with significant temporal variations. Initial gains were evident prior to 1990; however, these were subsequently offset by a sustained period of deterioration lasting until approximately 2017. During this nearly three-decade span, the rising index values indicate an increasing severity of road traffic crashes, implying a growing proportion of crashes resulting in fatalities and an elevated relative risk to road users. We observed a notable reversal of this trend since 2017, suggesting a potential positive impact of recent policy interventions. As depicted in Fig 1, the population density of the UAE has increased over time. This can cause variations in road safety owing to factors such as increased pedestrian, cyclist, and motorcyclist congestion, and an increase in the volume of traffic. The UAE’s GDP per capita decreased from 1980 to 2024, reaching its lowest level in the aftermath of the 2008 global financial crisis, but has since begun a modest recovery, as shown in Fig 1. Increased GDP may have a positive impact on road safety because of investments in vehicle standards and road infrastructure. However, rapid economic growth and increased motorization can lead to challenges such as inadequate infrastructure, lack of road safety awareness, and a higher incidence of risky driving behaviours, all of which may have an impact on road safety outcomes. Since the mid-1990s, the UAE unemployment rate has increased slightly, dipping in 2016 before rising again through 2019 and dropping through 2024. Fig 1 also shows that the UAE has improved traffic law enforcement owing to tougher fines for traffic law violations.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Log transformed road crash severity index, population density (people per square kilometers), and growth domestic product (GDP) per capita, total road traffic fines, and unemployment rate, UAE data 1980–2024.

https://doi.org/10.1371/journal.pone.0341441.g001

Assessing long- and short-term dynamic interaction among variables

The ARDL analysis and related tests were based on annual UAE time-series data from 1980 to 2024. lnSI, lnGDP, lnUNEMP, lnDEN, and lnFINE are the log-transformed road safety severity index for the UAE, average annual GDP per capita, unemployment rate, population density, and fines for traffic violations, respectively.

Unit root test

Unit root analysis is the primary step in ARDL analysis, which verifies each series’ stationarity and order of integration to ensure that no variable is I(2) or greater. To satisfy the bounds test assumption for the SARDL models, each variable must be stationary at level I(0) or the first difference I(1). Table 3 shows that all series confirm I(1) using the augmented Dickey–Fuller (ADF) test.

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Results of the Augmented Dickey-Fuller (ADF) unit root tests.

https://doi.org/10.1371/journal.pone.0341441.t003

ARDL bounds cointegration test

Based on the unit root test results (Table 3), which showed that no variable was I(2) or greater, the ARDL bound cointegration test was used to investigate the existence of a cointegration relationship between the UAE’s annual GDP per capita (lnGDP), unemployment rate (LUEMP), population density (lnDEN), traffic violations fines (lnFINE), and road safety as measured by the severity index (SI). The Schwarz information criterion (SC) recommends a maximum lag length of two, Equation (1). In contrast, the Hannan-Quinn (HQ) and Akaike Information criterion (AIC) suggest a maximum lag length of five. Based on these results, the ARDL analysis employed a maximum lag length of two suggested by the SC. The SC criterion is known as a consistent model selector, and tends to select the true model with the highest probability as the sample size increases (Table 4).

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. ARDL bounds tests.

https://doi.org/10.1371/journal.pone.0341441.t004

The ARDL bound test revealed a strong long-term link (cointegration) between socio-economic factors, traffic fines, and road safety when the log severity index (lnSI) was used as the dependent variable. The F-statistic = 5.93 exceeded the upper I(1)-bound critical value [40], as shown in Table 4 (p < 0.05). Similar significant long-term cointegration relationships (p < 0.01) were reported when GDP per capita, unemployment rate and population density were used as dependent variables. The bound test results are inconclusive (the test does not provide a clear answer) for the presence of a long-term relationship when traffic fines are used as a dependent variable, and the F-statistic lies between the I(0) and I(1) bounds (10%) (Table 4).

The results suggest a bidirectional but asymmetric relationship between crash severity and socioeconomic factors, in which crash severity responds more strongly to socioeconomic conditions (evidenced by the robust correlation when treated as the dependent variable, p < 0.01) than vice versa (weaker but still significant reverse correlation). This asymmetry aligns with empirical findings that lower GDP per capita and higher unemployment rates disproportionately increase crash severity owing to factors such as poor infrastructure, riskier driving behaviours, and limited access to safer vehicles [43,44]. Conversely, the feedback effect of crash severity on macroeconomic indicators (e.g., GDP, and unemployment) is subtler and often lagged, operating through indirect channels such as healthcare costs, lost productivity, and reduced economic activity in high-crash regions [44,45].

The persistence of significance in both directions nonetheless points to a systemic interdependence, consistent with theories of nonlinear socio-economic resilience [44]. For instance, studies show that while socioeconomic factors (e.g., education and income inequality) dominate short-term crash outcomes, severe crashes can exacerbate economic disparities over time, creating a reinforcing loop [43,45]. This aligns with Haddon’s framework, which emphasizes the interaction between humans, systems, and product performance in road safety [44].

Granger causal impact of socio-economic conditions on road safety

Table 5 displays the estimated long-term coefficients of the ARDL model in terms of the significant cointegration between road safety (lnSI) and the UAE’s average annual GDP per capita. The estimated average GDP per capita (lnGDP) coefficient is statistically significant (p < 0.01) (Table 5). In the long-term, the average GDP per capita has a significantly positive impact on the severity of road crashes in the UAE. This implies that the increased level of economic growth in the country results in increased road crash fatalities and, thus, a deterioration in road safety, possibly due to increased population income and car ownership. A 1% increase in average GDP per capita results in an approximately 0.94% increase in road crash severity risk and a reduction in road safety in the long term. Ali et al. [30] also reported similar results. They found a positive correlation between GDP and road traffic fatalities in upper-middle-income countries, suggesting that economic growth leads to increased vehicle ownership and higher crash rates due to higher mobility and potential lapses in road safety. Table 5 also shows that a higher population density (lnDEN) is linked to a higher risk of fatal road crashes (p < 0.05), indicating less road safety. In the long term, a 1% increase in population density led to a 0.73% increase in the severity risk. Similar results have been reported by Ageli and Zaidan [46] and Li et al. [47].

Download:

• PNG
  larger image
• TIFF
  original image

Table 5. ARDL long and short-term estimated coefficients.

https://doi.org/10.1371/journal.pone.0341441.t005

The short-term dynamics of the economic and safety indicators were estimated using an error-correction model (Equation (2), set with automatic lag selection in EViews 13), which captures the adjustment process toward the long-run equilibrium relationship identified by the ARDL framework. The coefficient (−0.60) of the error correction term is negative and highly significant (p < 0.01) (Table 5). This determines the speed of adjustment and ensures long-term convergence to an equilibrium relationship. A negative and significant error correction term of −0.60 in the ARDL model indicates that road safety adjusts back to its long-term equilibrium at a relatively rapid rate after any disturbances driven by economic and behavioural factors. The analysis revealed Granger-causal relationships between socioeconomic factors and road safety outcomes, particularly crash severity, in the UAE. These findings highlight the importance of socioeconomic conditions in determining road safety performance and the need for targeted policy interventions to achieve sustainable improvements.

The lagged economic variables—GDP growth (∆lnGDP at lags 1, 2, and 3) and unemployment rate (∆lnUNEMP at lags 1 and 2)—exhibit statistically significant positive coefficients (p < 0.05 or 0.10, Table 5), indicating their delayed influence on crash severity in the UAE. This suggests that past economic fluctuations affect road safety as the system adjusts toward long-term equilibrium. When crash severity exceeds its equilibrium level, prior increases in GDP and unemployment further increase severity in subsequent periods. This delayed effect likely stems from gradual changes in infrastructure, travel demand, and driving behavior. The negative coefficient of −1.9298 for suggests that a higher population density in the current period reduces crash severity, likely due to factors such as slower traffic speeds, better emergency response proximity, or stricter urban traffic enforcement [48,49]. Conversely, the positive lagged effect (2.4704 for implies that higher density in the previous period increases severity, possibly reflecting delayed risks such as congestion-induced driver frustration, accumulated infrastructure strain, or temporal spillovers from increased exposure [50,51]. This dynamic aligns with studies showing that while dense areas benefit from immediate safety infrastructure, lagged effects may emerge from prolonged stressors, such as traffic volume or aging road networks [48,50]. Traffic fines, while theoretically exerting an immediate deterrent effect (as reflected in their negative coefficient in the long term, see Table 5), are statistically insignificant, implying limited long and short-term efficacy in mitigating crash severity.

These findings underscore the interplay between macroeconomic conditions and regulatory measures in shaping road safety outcomes. Lagged socio-economic performance exerts a persistent influence, whereas traffic fines—despite their intended role—appear ineffective in the short term (see [52]).

Short-term implications: Increased GDP per capita and lower unemployment rates from previous periods may strain vehicle safety standards, road infrastructure, emergency response systems, and traffic law enforcement, potentially, exacerbating crash severity.

Long-term Implications: Sustained economic growth could amplify crash severity through higher vehicle ownership, increased travel frequency, adoption of high-power vehicles, and riskier driving behaviours spurred by greater economic affluence.

Residual diagnostics and model stability

Diagnostic tests on the residuals of the model confirmed its statistical adequacy. Standard tests for serial correlation (Breusch-Godfrey LM), heteroscedasticity (Breusch-Pagan-Godfrey), and non-normality (Jarque-Bera) fail to reject their respective null hypotheses at conventional significance levels, indicating that the residuals are well-behaved (Table 5). This result validates the model’s specification and the reliability of the inferred statistical significance. Complementing these findings, CUSUM and CUSUM-of-squares tests for parameter stability were conducted (Figs 2 and 3). Both statistics remain entirely within their 5% critical bounds, providing formal evidence of the model’s structural stability and the constancy of all estimated coefficients over the study period.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Cumulative total (cusum) test of recursive residuals.

Dashed lines show critical bounds at the 5% significance level.

https://doi.org/10.1371/journal.pone.0341441.g002

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Cumulative total of squares (cusumsq) test of recursive residuals.

Dashed lines show critical bounds at the 5% significance level.

https://doi.org/10.1371/journal.pone.0341441.g003

Parameter stability and heterogeneity

Furthermore, the stability of the short-run parameters and error correction term was assessed through recursive estimation. The recursive coefficient plots for all explanatory variables—log GDP per capita, unemployment rate, population density, and traffic violation fines—revealed a pronounced pattern indicative of a robust parameter constancy (Fig 4). Throughout the sample period, the evolutionary path of each coefficient closely adhered to its full-sample estimate, with all values fluctuating narrowly within a band of ± two standard errors around zero. The absence of any discernible structural breaks or trending behavior demonstrates that the influence of these socio-economic factors on road safety outcomes remained stable and homogeneous over the analysis timeframe, thereby affirming the robustness of the estimated ARDL-ECM parameters against temporal heterogeneity.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Recursive estimates of model coefficients.

https://doi.org/10.1371/journal.pone.0341441.g004

Interpretation of results

The empirical findings detail a complex, bidirectional interdependence between socioeconomic conditions and road crash severity in the UAE, characterized by a pronounced asymmetry in influence. The analysis robustly confirms that macroeconomic forces are the primary determinant of road safety outcomes. The significant long-term coefficient for GDP per capita (p < 0.01) suggests that economic growth paradoxically led to deterioration in road safety. This finding aligns directly with the work of Ali et al. [30] in upper-middle-income countries and reinforces the seminal hypothesis that rising income fuel increased vehicle ownership and mobility, potentially outpacing the development of safety regulations and infrastructure. This direct causal pathway is further reinforced by the Granger-causal impact of lagged GDP and unemployment, demonstrating that economic conditions from previous periods exert a persistent, delayed pressure on current crash severity, which is consistent with empirical observations that poorer infrastructure and riskier driving behaviours emerge from adverse socioeconomic conditions [43,44].

Conversely, the feedback effect of crash severity on the macroeconomic system, while statistically significant, operates through more indirect and lagged channels, a nuance that echoes the findings in extant literature [44,45]. The negative and highly significant error correction term (ECT = −0.60, p < 0.01) is pivotal here, indicating a rapid adjustment process that restores the system to its long-term equilibrium after a short-term shock. This dynamic suggests that the socioeconomic environment acts as the dominant attractor state, with crash severity behaving as the responsive variable. This supports the theoretical perspective that the economic impact of crashes is subtler, manifesting through accumulated healthcare costs and lost productivity rather than immediate macroeconomic shocks [43,45], a contrast to the immediate influence of the economy on crash risk.

The results regarding population density reveal a particularly intriguing temporal duality that both confirms and complicates prior research. The immediate, negative impact of current density () on severity aligns with studies by Ageli and Zaidan [46] and others [48,49], which attribute this to slower traffic speeds and better emergency response proximity in dense urban areas. However, the subsequent positive effect of lagged density () introduces a critical temporal dimension, suggesting that the long-term strains of congestion, infrastructure wear, and cumulative exposure eventually outweigh initial safety benefits. This delayed risk effect offers a nuanced explanation for the positive long-term coefficient observed in our model and that of Li et al. [28], suggesting that high density ultimately predicts higher severity, although through a delayed mechanism often overlooked in simpler cross-sectional analyses.

Ultimately, the model’s confirmation of systemic interdependence, consistent with theories of nonlinear socio-economic resilience [44], provides a framework for understanding the limited efficacy of isolated policy measures. The statistical insignificance of traffic fines underscores a finding consistent with the literature [52]: Regulatory deterrents are insufficient to counteract powerful socioeconomic undercurrents. This evidence strongly advocates for a shift in policy strategy, moving beyond reactive measures and towards proactive, system-level interventions that directly address the socioeconomic determinants of risk, such as strategic infrastructure investment and mandatory vehicle safety standards, to achieve sustainable improvements in road safety and break the reinforcing loop between economic disparity and crash outcomes [45].

Theoretical implications

The most striking finding of this study is the positive long-term relationship between GDP per capita and crash severity, contradicting the conventional assumption that economic development inherently improves road safety [21]. Instead, the results support the hypothesis that rapid economic growth, particularly in developing high-income economies like the UAE, can lead to increased vehicle ownership, higher traffic volumes, and riskier driving behaviours before regulatory and infrastructural adaptations catch up [8]. This aligns with the “Smeed’s Law” paradox, which posits that fatalities initially rise with motorization before declining as safety measures mature [50]. The UAE’s mirrors the patterns observed in other upper-middle-income countries, where economic prosperity outpaces the institutional capacity for road safety [51].

Furthermore, Granger-causal relationships highlight the delayed effects of economic variables on crash severity, reinforcing theories of behavioural adaptation [53]. For instance, past GDP growth and unemployment rates continue to influence crash risks in subsequent periods, suggesting that economic booms may encourage risk-taking (e.g., speeding and distracted driving), while economic downturns could strain maintenance budgets and enforcement capabilities.

Our results illuminate the theoretically complex, dualistic nature of the impact of population density on road safety, revealing both immediate benefits and delayed risks [41,44]. This is evidenced by a significant long-term relationship where a 1% increase in density raises crash severity risk by 0.73% (p < 0.05), which is consistent with the existing literature [22,27]. The short-term dynamics further capture this temporal complexity: a contemporary increase in density reduces severity (coefficient: −1.93), likely due to slower traffic speeds and enhanced enforcement proximity [41,43], whereas a positive lagged effect (coefficient: 2.47) suggests a subsequent rise in risk from congestion-induced frustration, accumulated infrastructure strain, or exposure spillover [44,45]. This suggests that the initial protective effects of dense urban design are offset by latent, long-term, and lagged effects, necessitating proactive policies to mitigate these risks.

Rapid economic development in the UAE, while generating wealth, appears to negatively influence road user behaviour, fostering riskier practices such as distracted driving and speeding [21,54]. This effect is compounded by the uniquely diverse demographic landscape. The influx of a multinational workforce introduces a confluence of disparate driving cultures, training standards, and attitudes toward traffic laws, creating a fragmented driving environment that challenges uniform road safety adherence and contributes to increased crash and fatality rates [51,54].

Practical implications

The results of this study have significant implications for policymakers, particularly in rapidly developing countries, where economic growth and road safety goals may appear at odds. First, the finding that traffic fines lack statistical significance in mitigating crash severity suggests that punitive measures alone are insufficient without complementary strategies such as infrastructure upgrades, public awareness campaigns, and stricter vehicle safety standards [55]. Given the UAE’s high GDP per capita, targeted investments in intelligent transportation systems (e.g., speed cameras and automated enforcement) and safer road designs (e.g., pedestrian zones and median barriers) could help mitigate the risks associated with increased mobility [56]. Dong et al. [57] highlighted that smart transportation innovations enhance efficiency and affect inequality through regional impacts and indirect factors, such as energy consumption and emissions efficiency. This suggests a multidimensional approach to road safety research in affluent developing countries, where technology and infrastructure interact with socio-economic conditions to influence outcomes.

Second, the delayed impact of the economic variables implies that road safety policies must adopt a proactive, forward-looking approach. For example, economic stimulus packages could include earmarked funding for road maintenance, while labour market reforms might address driver training and occupational safety in high-risk industries. UAE’s Vision 2030, which emphasizes sustainable development, presents an opportunity to integrate road safety into broader economic planning, such as linking urban density policies with transit-oriented development to reduce reliance on private vehicles [58].

Ultimately, the bidirectional relationship between socioeconomic factors and crash severity necessitates a systemic, multi-sectoral approach. Public health agencies, transport ministries, and economic planners must collaborate to address the root causes of unsafe driving (e.g., income inequality and inadequate public transit) while mitigating the long-term financial costs of crashes. Initiatives like the “Safe System” approach, which prioritize forgiving infrastructure and equitable access to safer mobility, could help break the reinforcing loop between economic disparities and road trauma (OECD [59]).

Limitations and future research directions

The availability and quality of data on road crash casualties may impact the reliability and generalizability of the study. One limitation of this study was the lack of complete data. Only three variables were used to describe the socio-economic situation: GDP per capita, unemployment, and population density. It is important to note that higher levels of education may result in a better understanding of and adherence to traffic laws and regulations, as well as increased economic prospects. Other potential factors and confounding variables that may impact road safety and socio-economic conditions include social and cultural norms, attitudes toward road safety, and cultural behaviours. The use of the relative severity index as a dependent variable in this study has some limitations. This index may not account for all aspects of road crash casualties, such as those that incur no injuries or psychological consequences.

This limitations of the study present clear directives for subsequent scholarly inquiry. To construct a more comprehensive explanatory model, future research must integrate a broader set of variables, encompassing educational attainment, public health indicators, fleet composition, and infrastructural capital. The analytical resolution could be significantly enhanced by employing higher-frequency data, which would strengthen the Granger causality inferences concerning short-term dynamics. We recommend conducting a comparative study across GCC member states to evaluate the external validity of the results. Moreover, a mixed-methods approach is vital; qualitative investigations, including behavioral surveys and policymaker engagement, are necessary to provide a critical context for quantitative relationships and to interpret the complex role of socio-cultural factors and driver psychology.