INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XI November 2025

Logistic Regression Modelling of Road Traffic Accident Severity: A

Study on Driver Characteristics in Zimbabwe

Ever Moyo^1*, Kaitano Dzinavatonga², Zvikomborero Lesley Hakunavanhu³

¹Mathematics and Statistics Lecturer, Department of Mathematical Sciences

²Physics Lecturer, Department of Physics

³Quality Assurance Coordinator, Department of Quality Assurance

^1,3Zimbabwe Open University, Masvingo Region

²Zimbabwe Open University, Harare Region, Zimbabwe

*Corresponding Author

DOI: https://dx.doi.org/10.47772/IJRISS.2025.91100254

Received: 21 November 2025; Accepted: 28 November 2025; Published: 06 December 2025

ABSTRACT

Road traffic accidents (RTAs) remain a major public health concern in Zimbabwe, yet little empirical work has

examined the combined influence of driver behavior, demographic characteristics and environmental factors on

accident severity. This study applied binary logistic regression analysis to a dataset of 500 accident-involved

drivers to identify the key predictors of severe accidents. Frequency distributions summarised the characteristics

of drivers, environmental conditions and vehicle status, while logistic regression quantified their influence on

accident severity. The results showed that six key predictors significantly increased the likelihood of severe

accidents: low driving experience, alcohol use, fatigue, mobile phone use, over speeding and wet road conditions.

Over speeding emerged as the strongest predictor, with drivers who overspeed being four times more likely to

be involved in a severe accident. Although age, gender, time of accident and vehicle condition were not

statistically significant, they exhibited expected directional effects. The full model (M₁) significantly improved

prediction compared to the null model (M0) with Δχ² = 77.3 demonstrating that the included predictors

collectively enhance the model’s explanatory power with respect to predicting accident severity. It demonstrated

an acceptable predictive accuracy with an AUC = 0.720, indicating its effectiveness in distinguishing between

severe and non-severe accidents. The findings emphasize that human behavior remains the most critical

determinant of accident severity in Zimbabwe, with implications for targeted interventions. The study findings

highlight the need for evidence-based interventions focused on speed control, anti-drunk driving enforcement,

fatigue management, mobile phone usage laws, targeted road safety campaigns and improved road infrastructure

especially during wet conditions. Training and awareness programs targeting inexperienced drivers could reduce

severity outcomes. The study contributes valuable insights toward improving road safety strategies in Zimbabwe

and similar contexts.

Keywords: Binary Logistic Regression Analysis, Accident Severity, Driver Characteristics, Predictors, Road

Safety.

INTRODUCTION

Background

Road traffic accidents (RTAs) are among the leading causes of mortality and morbidity worldwide, particularly

in developing countries. Road accidents are a pervasive global issue with profound consequences for individuals,

communities, and economies (Deepthi et al.2023). Zimbabwe has comprehensive road network linking the

different parts of the country and providing access to neighbouring countries for imports and exports. The

Page 3189

www.rsisinternational.org

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XI November 2025

country is experiencing an increase in motorisation while roads have deteriorated resulting in increased road

accidents (Muvuringi, 2012). Nowadays, the problem of road accident rates is one of the most important health

and social policy issues concerning the countries in all continents. Each year, nearly 1.3 million people

worldwide lose their life on roads, and 20–50 million sustain severe injuries, the majority of which require long-

term treatment (Goniewicz et al.,2016). Goniewicz et al.,2016, discovered that the causes of road accidents

include lack of control and enforcement concerning implementation of traffic regulation (primarily driving at

excessive speed, driving under the influence of alcohol and not respecting the rights of other road users (mainly

pedestrians and cyclists), lack of appropriate infrastructure and unroadworthy vehicles. They suggested the

strategies and programmes for improving road traffic such as reducing the risk of exposure to an accident,

prevention of accidents, reduction in bodily injuries sustained in accidents, and reduction of the effects of injuries

by improvement of post-accident medical care.

Foya, (2019) established that the major causes of accidents in Zimbabwe were human error which included

unlicensed driving, texting, alcohol consumption and driving while driving., failure to observe road and that

most of the vehicles were not mechanically safe to be in the roads. Foya, (2019) concluded that all these factors

have caused numerous deaths in the roads and it has become a major cost to insurance companies. State of the

various roads is a major challenge to motorists, and some roads are full of potholes which makes it dangers for

motorists to drive through.

Chibaro et al., (2024) assessed the effects of human factors on road traffic safety in minimizing road traffic

accidents in Harare Metropolitan, Zimbabwe. They tested human factors such as improper age of drivers who

tend to overtake and overspeed recklessly, over speeding, alcohol drinking, corruption and failure to maintain

vehicles as a moderating variable. They identified that excessive speeding and drunk driving while driving were

the worst human behaviour causing accidents in Zimbabwe.They concluded that it is necessary to enforce the

speed limit and deploy speed cameras at significant intersections and problem areas. Munuhwa et al. (2020)

indicated that in Botswana, the RTAs are caused by road users through speeding, unlicensed driving, using cell

phones whilst driving, alcohol and drug abuse, bad state of mind and healthy, non-use of safety belts and

deliberate failure to observe road regulations amongst others. Their findings also indicated that mechanically

faulty vehicles, unmaintained vehicles, old vehicles, and tyre blowouts are vehicle related factors causing RTAs.

Road system conditions involve potholes, stray livestock and road design attributes amongst others. They

recommended educating the public on safe driving habits, punitive policies on road users breaking road traffic

laws and regulations, stringent measures against livestock owners who leave stock straying in highways and

public roads and regular road maintenance and vehicle maintenance to reduce RTAs. Across the world, traffic

accidents cause major health problems and are of concern to health institutions; nearly 1.35 million people are

killed or disabled in traffic accidents every year. This issue is growing; by 2030, road traffic injuries will be the

seventh leading cause of death globally (Ahmed et al.2023).

Road traffic accidents hamper economic growth as they gobble huge financial resources which government can

channel to more urgent developmental programmes. Accident severity can be categorized into three levels,

slight, serious and fatal, depending on the outcome for the driver or passengers. Understanding the determinants

of these severity levels is essential for policymakers, law enforcement agencies and public health authorities.

Road traffic accidents (RTAs) pose significant public health and economic challenges in Zimbabwe.

Understanding factors that influence the severity of accidents can help develop targeted safety interventions.

This study focuses on driver characteristics as predictors of accident severity using logistic regression modelling.

Research Problem

Despite the availability of accident statistics and the prevalence of these accidents, there is limited understanding

of the specific driver characteristics that influence accident severity in Zimbabwe. There is also limited

quantitative research exploring how driver characteristics influence the accident severity. This gap hinders the

development of targeted interventions and policies aimed at reducing the severity and incidence of road

accidents. This study aims to model the relationship between driver characteristics and the severity of road traffic

accidents in Zimbabwe through logistic regression modelling, to inform evidence-based strategies for improving

road safety in Zimbabwe.

Page 3190

www.rsisinternational.org

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XI November 2025

Research Objectives

1. To identify key driver characteristics influencing road accident severity in Zimbabwe.

2. To develop and validate a logistic regression model predicting accident severity based on driver

attributes.

3. To provide recommendations for reducing the occurrence of severe road accidents.

Research Questions

1. What driver characteristics significantly influence the severity of road traffic accidents?

2. How well can logistic regression models predict accident severity outcomes?

3. What policy interventions can be derived from the model results?

Significance of the Study

This study is significant because it sheds light on the characteristics of drivers that influence the severity of

accidents. This information will help road safety authorities and lawmakers create focused interventions and

safety measures. The results will assist in prioritizing resource allocation for road safety programs targeted at

vulnerable groups more effectively. The study enhances data-driven decision making for accident prevention by

using logistic regression modelling. The results can also improve driver education programs by addressing risky

behaviours that are unique to the local population. Additionally, it contributes to academic literature by applying

statistical techniques in the context of road safety in Zimbabwe. Reducing accident severity has socio-economic

benefits through fewer injuries, fatalities and related costs. Furthermore, the study advances regional

understanding of road safety issues in Zimbabwe and comparable settings.

LITERATURE REVIEW

Logistic regression was applied to accident-related data collected from traffic police records to examine the

contribution of several variables to accident severity (Al-Ghamdi, 2002). Michalaki et al., (2015) applied a

generalized ordered logistic regression model to identify the factors affecting the severity of hard shoulder (HS)

and main carriageway (MC) accidents on motorways. They discovered that driver fatigue is one of the factors

increasing the severity of the accidents.

Nowadays, road safety is an issue particularly relevant because of the increasing occurrence of road accidents,

and it is very important to analyse accident severity and the factors influencing it (Eboli et al., (2020). Human

behaviour is a dominant factor in road accidents, contributing to more than 70% of such incidents (McCarty and

Kim ,2024).Chen et al,(2021) investigated the factors affecting the accident severity of drivers with different

driving experience in Shaanxi, China and discovered that novice drivers younger than 30 or older than 55 are

prone to suffer fatal accident, but for experienced drivers, the risk of fatal accident decreases. Paleti et al., (2010)

and Adavikottu and Velaga, (2021) found also that novice drivers under the influence of alcohol and driving on

roads with high-speed limits triggered aggressive driving behaviour leading to severe injuries. Haerani et al.

(2019) suggested that age was a moderate variable in the relationship between personality, driving behaviour

and driving outcomes in the city of Makassar, the capital of the South Sulawesi province in Indonesia.Muvuringi,

(2012) discovered that Zimbabwe’s key risk factors that contribute to RTAs include reckless driving, violation

of traffic laws, damaged vehicles, and bad roads. Chibaro et al., (2024) suggested that excessive speeding and

drunk driving while driving have been identified as the worst human behaviour causing accidents in Zimbabwe.

Millicent et al., (2016) findings reflected that environmental, personal and mechanical factors were the major

driving factors of RTAs in Zimbabwe.

Page 3191

www.rsisinternational.org

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XI November 2025

METHODOLOGY

Research Design

A quantitative, analytical design was employed using secondary accident data on a sample of 500 drivers from

2020 to 2024, to model the relationship between accident severity (dependent variable) and driver characteristics

(independent variables) using Logistic Regression Modelling in Zimbabwe. Data analysis was conducted using

JASP version 0.95.4.0. which supports advanced logistic regression procedures, diagnostics and validation

techniques.

Study Area

The study focused on road traffic accidents occurring across the regions of Zimbabwe. Zimbabwe is a developing

country with diverse driving environments such as urban, peri-urban and rural roads. It is characterised by

variability in road conditions, driver behaviours and enforcement capacity. This made it a suitable setting for

assessing the determinants of accident severity.

Data Source

Secondary accident data were obtained from the official Zimbabwe Republic Police (ZRP) Accident Reports and

the Traffic Safety Council of Zimbabwe (TSCZ) databases from 2020-2024 for 500 drivers. These records

included detailed information on accident type, driver demographics, vehicle factors, environmental conditions

and severity outcome.

Data Collection and Preparation

The data on road traffic accidents were gathered which included variables such as weather conditions, time of

day, vehicle condition, road condition and driver characteristics. The target variable, accident severity was

categorised as a binary outcome (severe and non-severe).

Data Cleaning and Coding

Adata cleaning procedure was done in JASP.0 /1 coding was used in this study because it provides a clear, binary

representation of the categorical variables used, simplifying the modelling process. It also allows the logistic

regression to model the log-odds of the outcome as a linear function of predictors, making it easy to interpret the

results (Dominguez-Almendros et al., 2011). The missing data was promptly inserted. The continuous variables

were automatically dummy coded in JASP. The following coding scheme was employed:

Coding Schemes

Logistic regression in JASP requires correctly coded variables. The coefficient associated with1-coded variable

indicates the change in log-odds when moving from the reference category (0) to the target category (1). The

following conventions were done to both the dependent variable and the categorical predictor variables.

Dependent Variable (0utcome): Accident Severity

This had a binary outcome with 1 = severe accident and 0 = non-severe accident. This coding is suitable for

logistic regression estimates of accident severity because it models the probability of a binary outcome.

Categorical Predictor Variables

The predictors were binary coded and those with more than two classes were automatically dummy coded in

JASP.The categorical predictors were coded numerically (dummy coded) in JASP as follows:

Page 3192

www.rsisinternational.org

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XI November 2025

Table 1: Coding Scheme for the categorical Predictors

Variable

Category

< 25years

25-45years

>45years

Female

Male

code

0

Rationale

Reference

Risk

Age of Driver

1

2

Safer

0

Reference

Risk

Gender

1

<2years

2-5years

>5years

No

0

Reference

Risk

Driving experience

1

2

Safer

0

Reference

Higher Risk

Reference

Higher Risk

Reference

Higher Risk

Reference

Higher Risk

Reference

Risk

Alcohol Use

Yes

1

No

0

Fatigue

Yes

1

No

0

Mobile Phone Use

Over speeding

Time of Accident

Vehicle Condition

Road Condition

Yes

1

No

0

Yes

1

Day

0

Night

Good

Poor

1

0

Reference

Higher Risk

Reference

Higher Risk

1

Dry

0

Wet

1

Data Analysis Procedures

Exploratory Data Analysis

Frequency tables were generated for all the categorical variables. The missing values were checked and manually

inserted.

Checking Logistic Regression Model Assumptions

Correlation and Multicollinearity Check

Multicollinearity is a situation in regression analysis where two or more predictor variables are highly correlated,

meaning they provide overlapping or redundant information about the outcome variable. Multicollinearity

affects interpretation not prediction. It does not influence the overall model fit but it affects the stability,

interpretation and precision of regression coefficients. Proper diagnostic using Tolerance and VIF should always

Page 3193

www.rsisinternational.org

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XI November 2025

be performed before interpreting the logistic regression results. When predictors overlap, it becomes difficult to

determine the unique effect of each variable. In this study multicollinearity diagnostics were conducted using

Tolerance and the Variance Inflation Factor (VIF) in JASP to assess whether the predictor variables in the

accident severity logistic regression model were excessively correlated. Tolerance Values < 0.10 indicate severe

multicollinearity and Tolerance Values < 0.20 indicate a potential concern. Hence Acceptable ranges for

Tolerance Values is any value greater than 0.20. Variance Inflation Factor (VIF) > 5 suggests moderate

multicollinearity and VIF > 10 indicates serious multicollinearity. Hence Acceptance ranges for VIF is any value

less than 5.

Model Specification

A binary logistic regression was employed to model the probability of severe accidents based on vehicle

condition, environmental conditions and driver characteristics.

The Logistic Regression Model

Logistic regression model is a statistical model in which an evaluation is made of the relationship between: a

dependent qualitative, dichotomic variable (binary or binomial logistic regression) or variable with more than

two values (multinomial logistic regression) and one or more independent explanatory variables or covariates,

whether qualitative or quantitative (Dominguez-Almendros et al., 2011).In this study, Logistic regression is a

statistical modelling method used to predict the probability of a binary outcome variable, the accident severity.

Binary Logistic Regression model was used to estimate the probability of an accident occurring and to

understand how the predictor variables (driver age, vehicle condition, gender, driving experience, alcohol

consumption, fatigue, mobile phone use when driving, over speeding, time of accident, road condition) influence

the severity of the road traffic accidents. It enabled us to classify the results into significant factors and non-

significant factors and to compute the Odds Ratio (OR) associated to the covariate/predictor together with the

associated 95% confidence interval (CI) for interpretation (Dominguez-Almendros et al., 2011). The probability

and odds aid in decision making and hypothesis testing.

The Logistic Function

The probability of the event is:

1

(

| )

= (1

=

+

−

1+

(

+

+...+

2

)

0

1

2

The Logit Transformation

The probabilities are bounded between 0 and 1, hence the binary logistic regression model predicts the log-odds

of the outcome:

( )

= ln (₁₋) =

+

₂+. . . +

+

0

1

2

where

= probability of a severe accident,

= driver characteristics, = 1,2,3 … .

= regression coefficients, = 1,2,3 … .

=the residuals

Model Fitting

The Maximum Likelihood Estimation in JASP was used to estimate the model parameters to obtain the

coefficients, standard errors, Odds-Ratios, Confidence Interval and p-values that explains the relationships

Page 3194

www.rsisinternational.org

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XI November 2025

between the variables in the data. Model fitting was assessed using the Pseudo R² Indices (McFadden R²,

Nagelkerke R², Tjur R², Cox & Snell R²) in which higher values indicate a better fit, AIC (Akaike Information

Criterion) and BIC (Bayesian Information Criterion) in which lower values indicate a better fit and The Chi-

Square Distribution, in which a low p-value indicates that the model is a good fit.

Model Evaluation and Validation

The predictive accuracy of the model was evaluated using the AUC (Area Under Curve). It helped to measure

the model’s ability to distinguish between positive and negative classes. If the: AUC > 0.7 -it indicates an

acceptable model.

AUC > 0.8 – it indicates a good model.

AUC > 0.9 -it indicates an excellent model.

Model Interpretation

The interpretation of predictors was done using the Odds Ratio (OR),95% Confidence Interval (CI) and p-values.

The confidence intervals (CI) and the p-values assess the statistical significance of predictors (Dominguez-

Almendros et al., 2011). The interpretation focused on assessing the increase or decrease in the likelihood of

severe accidents given specific driver characteristics. It measures how a predictor affects the odds of the event.

=

where is the probability of a severe accident.

1−

The coefficients ( ) are transformed into ratios using the formula:

=

ℎ

is the regression coefficient, = 1,2,3 … .

If the:

OR > 1: it indicates a positive association with the outcome which means the predictor increases odds of the

event

OR<1: it indicates a negative association with the outcome which means the predictor decreases odds.

OR=1: it means the predictor has no effect.

Model Refinement

All the non -significance predictors were removed from the model to simplify it.The final generated model was

expressed in terms of the significant predictors only.

Model Application

The generated model from the used variables consisted of the significant predictors only. It will be used to predict

accident severity on new data. The high-risk scenarios will be identified for targeted interventions.

RESULTS AND DISCUSSIONS

Frequency Tables

The frequency distributions below describe the characteristics of a sample of 500 drivers involved in road traffic

accidents in Zimbabwe on accident severity predictors stated in each table.

Page 3195

www.rsisinternational.org

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XI November 2025

Table 2: Frequencies for Accident Severity

Accident Severity

Frequency Percent

0

319

181

0

63.8

36.2

0

1

Missing

Total

500

100

Sources: Authors` computation using JASP 0.95.4.0.

Table 2 above shows that 63.8% of the accidents were classified as severity level 1 and 36.2% as non-severe

level 0. Most recorded accidents were non-severe. However, the proportion of severe accidents (36.2%) is still

high, indicating a significant public health and safety concern.

Table 3: Frequencies for Driving Experience

Driving Experience

Frequency Percent

0

86

17.2

26.8

56.0

0

1

134

280

0

2

Missing

Total

500

100

Sources: Authors` computation using JASP 0.95.4.0.

Table 3 above shows that 56% of drivers have the highest level of experience (level 2=>5years) followed by

26.8 (level 1 = 2-5 years) and 17.2% with level 0 = <2years.This indicates that drivers with more experience are

predominant in the data. This suggests that accident involvement is not limited to inexperienced drivers,

experienced drivers also contribute significantly to accident statistics.

Table 4: Frequencies for Age Group

Age Group

Frequency

131

Percent

26.2

52.8

21.0

0

1

264

2

105

Missing

Total

0

500

100

Sources: Authors` computation using JASP 0.95.4.0.

Page 3196

www.rsisinternational.org

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XI November 2025

Table 4 above shows that most drivers are in age group 1 (25-45years) with 52.8 %, followed by age group 0

(<25years) with 26,2 % and age group 2 (>45 years) with 21.0 %. Middle- aged drivers constitute the majority

of accident-involved drivers. Younger and older drivers form smaller but notable proportions which may

influence severity differently.

Table 5: Frequencies for Gender

Gender Frequency Percent

Female 103

Male 397

Missing 0

Total 500

20.6

79.4

0

100

Sources: Authors` computation using JASP 0.95.4.0.

Table 5 above shows more male drivers (79.4) than female drivers (20.6) were involved in accidents. Male

drivers’ dominant accident involvement. This may reflect higher exposure, risk-taking behaviour or driving

frequency among men in Zimbabwe. Gender could be a significant predictor of accident severity.

Table 6: Frequencies for Alcohol Use

Alcohol Frequency

Percent

87.4

12.6

0

No

437

63

Yes

Missing

Total

0

500

Sources: Authors` computation using JASP 0.95.4.0.

Table 6 above shows that 87.4% did not consume alcohol while driving but 12.6% did. A smaller proportion of

drivers were to have used alcohol before the accident. Although a minority variable, alcohol use may strongly

increase the odds of severe accidents.

Table 7: Frequencies for Fatigue

Fatigue

No

Frequency

Percent

82.0

18.0

0

410

90

Yes

Missing

Total

0

500

Sources: Authors` computation using JASP 0.95.4.0.

Page 3197

www.rsisinternational.org

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XI November 2025

Table 7 above shows that 82% reported no fatigue while 18% reported fatigue. Most drivers were not in fatigue,

but fatigue still represents a meaningful portion (18%) and may be associated with severe outcomes due to

slowed reaction times.

Table 8: Frequencies for Mobile Phone Use

Mobile Phone Use Frequency

Percent

79.0

21.0

0

No

395

105

0

Yes

Missing

Total

500

Sources: Authors` computation using JASP 0.95.4.0.

Table 8 above shows that 79.0% did not use a mobile phone while driving and 21% did. One fifth of drivers used

mobile phones while driving. This distracted driving behaviour typically increases accident severity and is a

relevant logistic regression predictor.

Table 9: Frequencies for Over speeding

Over speeding Frequency Percent

No

352

148

0

70.4

29.6

0

Yes

Missing

Total

500

Sources: Authors` computation using JASP 0.95.4.0.

Table 9 above shows that 70.4% did not overspeed and 29.6% did. Nearly one-third of drivers were over

speeding. Since over speeding is directly linked to impact force, it is likely to be a significant predictor of severe

accidents reinforcing the necessity for speed enforcement and public education.

Table 10: Frequencies for Time of Accident

Time of Accident Frequency Percent

Day

335

165

0

67.0

33.0

0

Night

Missing

Total

500

100

Sources: Authors` computation using JASP 0.95.4.0.

Page 3198

www.rsisinternational.org

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XI November 2025

Table 10 above shows that 67.0% occurred during the day and 33.0% at night. Accidents occur more frequently

during the day. However, night-time accidents, though fewer, might be associated with higher severity due to

poor visibility and fatigue.

Table 11: Frequencies for Vehicle Condition

Vehicle Condition Frequency Percent

Good

Poor

401

99

80.2

19.8

0

Missing

Total

0

500

100

Sources: Authors` computation using JASP 0.95.4.0.

Table 11 above shows that 80.2% of vehicles were in good condition while 19.8% were in poor condition. Most

vehicles were in good condition, but poor vehicle condition still accounts for a substantial portion and may

sharply increase the likelihood of severe accidents necessitating vehicle safety checks.

Table 12: Frequencies for Road Condition

Road Condition Frequency Percent

Dry

392

108

0

78.4

21,6

0

Wet

Missing

Total

500

100

Sources: Authors` computation using JASP 0.95.4.0.

Table 12 shows that 78.4% occurred on dry roads and 21,6% on wet roads. Most accidents occurred on dry roads,

but a significant proportion occur on wet surfaces. This highlights the importance of adjusting driving behaviour

based on weather conditions.

Correlation and Multicollinearity Check

Table 13: Multicollinearity Diagnostic Results

Tolerance VIF

0.522

0.516

0.973

0.977

0.965

1.916

1.939

1.028

1.024

1.036

Driving Experience

Age Group

Gender

Alcohol Use

Fatigue

Page 3199

www.rsisinternational.org

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XI November 2025

0.963

0.946

0.984

0.968

0.948

1.038

1.057

1.016

1.034

1.055

Mobile Phone Use

Over speeding

Time of Accident

Vehicle Condition

Road Condition

Table 13 above shows that all predictors: driving experience, age group, gender, alcohol use, fatigue, mobile

phone use, over speeding, time of accident, vehicle condition and road conditions, Tolerance values range from

0.516 to 0.977 and are well above the conventional cutoff of 0.10.VIF values range from 1.024 to 1. 939 which

are also far below the commonly accepted limit of 10. These values are all well within the Acceptable thresholds.

The lowest tolerance (0.516 for Age Group) is far above 0.20, meaning no risk of multicollinearity. The highest

VIF (1.939 for Age Group) is far below 5, indicating a weak collinearity. Several variables (Gender, Alcohol

Use, Fatigue, Mobile Phone Use) show near-perfect tolerance (≈0.97) and very low VIF (≈1.02–1.04),

suggesting that they are statistically independent. The diagnostic results demonstrate that none of the predictor

variables exhibit a problematic multicollinearity. The predictors do not strongly correlate with one another,

meaning that each variable provides unique explanatory information about accident severity. Hence, since the

multicollinearity is low, the logistic regression coefficients are stable, and the standard errors are not inflated.

The effect sizes and significance values can be interpreted with confidence in subsequence regression analyses.

All the predictors were retained in the model without violating the assumptions.

Logistic Regression Results for Accident Severity

Table 14: Coefficients of all the predictors

Model

M₀

Estimate Odds Ratio (OR) 95% Confidence Interval (CI)

p

(Intercept)

-0.567

-1.608

0.274

-0.472

0.506

0.702

0.815

0.629

1.385

0.086

0.123

0.574

0.567

0.200

1.315

0.624

1.659

2.017

2.259

1.875

3.994

1.090

1.131

1.775

(0.473,0.681)

(0.104,0.384)

(0’912,1.897)

(0.449,0.867)

(0.994,2.769)

(1.130,3.603)

(1.367,3.734)

(1.160,3.031)

(2.600,6.136)

(0.713,1.665)

(0.685,1.867)

(1.105,2.851)

<0.01

0.142

0.005

0.053

0.018

0.001

0.010

<0.01

0.691

0.630

0.018

(Intercept)

M₁

Age Group

Driving Experience

Gender (Male)

Alcohol Use (Yes)

Fatigue (Yes)

Mobile Phone Use (Yes)

Over speeding (Yes)

Time of Accident (Night)

Vehicle Condition (Poor)

Road Condition (Wet)

Sources: Authors` computation using JASP 0.95.4.0.

Page 3200

www.rsisinternational.org

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XI November 2025

Logistic Regression Equation with significant predictors

The dependant variable is the Accident Severity = 1 (Severe). The model was fitted using the predictor

coefficients from Table 14 above. The Logistic Regression Model fitted is :

( )

(

)

(

)

(

)

logit

= −1.608 − 0.472 DrivingExperience + 0.702 AlcoholUse-Yes + 0.815 Fatigue-Yes

(

)

(

)

+ 0.629 MobilePhoneUse-Yes + 1.385 Overspeeding-Yes

(

)

+ 0.574 RoadCondition-Wet

1

The corresponding predicted probability Equation of a severe accident is:

1

=

Equation 2

−(−1.608−0.472 +0.702 +0.815 +0.629 +1.385 +0.574

)

6

1

2

3

4

5

1+

Where each

= 1,2,3,4,5,6 ,represents the coded predictor value.

The coefficients of the predictors were used to calculate the Odds Ratio (OR) for interpretation.

Interpretation of Predictors

Using results from Table 14 and Equation 1 and 2 above, the predictors were grouped into significant and non-

significant. All significant predictors have p values less than 0.05 and the Confidence Intervals (CI) exclude 1and

non-significant predictors p-value is greater than 0.05 and the Confidence Intervals (CI) include 1. The

significant predictors were driving experience, alcohol use, fatigue, mobile phone use, over speeding and rod

condition. The non-significant predictors were age group, gender(male), time of accident(night) and vehicle

condition (poor).

The model intercept for the full model, M_1,is -1.608 with OR=0.20, p < 0. 001.This is the baseline odds of

experiencing severe accidents when all predictors are zero. A negative intercept suggests that without any risk

factors, the baseline risk of a severe accident is low.

Significant Predictors (p< 0.05)

Table 13 results indicate that six factors significantly predicted accident severity: Driving Experience, Alcohol

Use, Fatigue, Mobile Phone Use, Over speeding, and Road Condition.

1.Drving Experience (protective: OR = 0.624; 95% CI 0.449–0.867; p = 0.005). - Drivers with low

experience are more likely to be involved in severe accidents. OR=0.624 means higher driving experience

reduces the odds of severe accidents by 37.6% (1-0.624). This means that greater driving experience is a

protective factor. It reduces the likelihood of a severe accident. The protective effect of greater driving experience

aligns with Chen et al. (2021) findings that accident severity was more pronounced among novice and older

drivers.

2.Alcohol Use-Yes (OR = 2.017; 95% CI 1.130–3.603; p = 0.018)– Drivers who consume alcohol before

driving is more than twice as likely to be involved in a severe accident compared to sober drivers. These findings

agree with Paleti et al. (2010) and McCarty & Kim (2024) who emphasize that human-related factors particularly

aggressive or impaired driving, remain the primary contributors to road accidents worldwide. Alcohol

impairment slows reaction time and impairs judgment which increase the severity of crashes. It also supports the

work of Chibaro et al. (2024) which identified drunk driving as a critical factor in causing accidents in

Zimbabwe.

3.Fatigue-Yes (OR = 2.259; 95% CI 1.367–3.734; p = 0.001). Fatigued drivers in our sample have 2.3 times

higher odds of causing a severe accident underscoring fatigue’s impact. Fatigue reduces driver alertness which

increase the severity of crashes. This supports the argument by Eboli et al. (2020) that behavioural characteristics

particularly fatigue account for the largest proportion of accident risk.

Page 3201

www.rsisinternational.org

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XI November 2025

4.Mobile Phone Use-Yes (OR = 1.875; 95% CI 1.160–3.031; p = 0.010)– Using a mobile phone while driving

increases the odds of severe accidents by 88%. This is consistent with Adavikottu & Velaga (2021) and Haerani

et al. (2019) findings that place distraction-related behaviours among the top behavioural contributors to severe

accidents. Therefore, our mobile usage estimate fits well within the trends noted by others regionally to influence

road safety issues.

5.Over Speeding-Yes (OR = 3.994; 95% CI 2.600–6.136; p < 0.01)– Over speeding is the strongest predictor

in our model. Drivers who overspeed are approximately 4 times the odds of a severe crash compared to those

who did not. This large effective size is consistent with regional evidence that speed is among the single strongest

determinants of crash severity This aligns with findings by Michalaki et al. (2015) and Adavikottu and Velaga

(2021) who reported that accident dynamics such as speed significantly increase the likelihood of high-severity

crashes. Similarly, Muvuringi (2012) and Chibaro et al. (2024) identified excessive speeding as one of the most

critical behavioural contributors to accidents in Zimbabwe. The consistent associations demonstrate that speed

management remains a key intervention point for reducing RTA fatalities.

6.Road Condition-Wet (OR = 1.775; 95% CI 1.105–2.851; p = 0.018)– Wet Road conditions increase the odds

of a severe accident by 77.5%. Regional literature commonly cites poor road surface and weather as important

contributors to severe crashes. The increase we observed is consistence with the findings by Chen et al. (2021)

that showed that bad weather and terrain increase the severe accident outcomes. It suggests a meaningful

environmental contribution in addition to behavioural risks.

Non-significant Predictors (p > 0.05)

The non -significant predictors are not statistically significant but they show the directional effects. Factors such

as Age Group, Gender, Time of Accident and Vehicle Condition did not reach statistical significance.

1.Gender-Male (p=0.053, OR=1.659) – Male drivers show 66% higher odds of severe accidents, aligning with

evidence from Paleti et al. (2010) and Haerani et al. (2019) who associated young male drivers with risky

behavioural patterns such as speeding or aggressive driving.

2.Age Group (p=0.142, OR=1.315, Time of Accident-Night (p=0.691, OR=1.090) and Vehicle condition-

Poor (p=0.630, OR=1.131). These variables are not statistically significant in predicting severity in this model,

though not significant, showed expected directional effects. This is consistent with Millicent et al. (2016) who

found that both environmental and mechanical factors contribute to accident prevalence in Zimbabwe at varying

degrees. While not statistically significant, the role of age as a moderating factor in behavioural outcomes, as

highlighted by Haerani et al. (2019), who mentioned age as an influence of the accident severity.

Model Performance

Table 14: Performance metrics

Value

AUC

0.720

Sources: Authors` computation using JASP 0.95.4.0.

An AUC of 0.720 indicates the acceptance predictive accuracy. The value suggests that the model has an

acceptable capacity to distinguish between severe and non- severe accidents based on the given variable.

Model Validation

Page 3202

www.rsisinternational.org

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XI November 2025

Table 15: Model Summary-Accident Severity

2

Model

Deviance

AIC

BIC

df

Δχ²

Nagelkerke

Tjur

Cox &

Snell

McFadde

2

M₀

M₁

654.6

577.2

656.6

599.2

660.8

645.6

494

489

0.00

77.3

0.118

0.196

0.154 0.143

Sources: Authors` computation using JASP 0.95.4.0.

Note. M₁ includes Age Group, Driving Experience, Gender, Alcohol Use, Fatigue, Mobile Phone Use, Over

speeding, Time of Accident, Vehicle Condition, Road Condition.

Table 15 above shows that the model validation results included the deviance and BIC, Δχ² (Chi-square

Difference) and Pseudo R² Measures.

Deviance

It measures the lack of fit of the model in which lower deviance indicates a better fit.The results show that M₀

(null model) has a deviance of 654,6 and M₁(full model with predictors) has a deviance of 577.2. The reduction

in deviance shows that adding the predictors significantly improves the model fit. The model with the driver

characteristics (M₁) fits well the data significantly better than the null model.

AIC (Akaike Information Criterion) & BIC (Bayesian Information Criterion)

Lower AIC (599.2versus 656.6) and BIC (645.6 versus 660.8) in M₁indicate a better model fit. This confirms

that the full model balances goodness of fit and model complexity effectively.

Δχ² (Chi-square Difference)

Table 15 above shows that Δχ² = 77.3 with df = 5 (difference in degrees of freedom between M₀and M₁). This

is the likelihood ratio test, testing whether the full model is significantly better than the null model. It indicates

that the predictors collectively improve the model’s predictive ability.

Pseudo R² Measures

It consisted of McFadden R², Nagelkerke R², Tjur R² and Cox & Snell R². McFadden R² is 0.118 which indicates

a low-moderate fit and it means the model explains 11.8% of the variation in accident severity. Nagelkerke R²

is 0.196 which means the model explains 19.6% of the variation in accident severity. Tjur R² is 0.154 which

shows a moderate discrimination between severe and non-severe accidents. Cox & Snell R² is 0.143 which shows

that the model explains 14.3% of the variation in accident severity. These Pseudo R² values suggest that the

model explains 12-20% of the variation in accident severity which is a better fit, given the complexity of the

accident outcome. The results are consistent with modelling outcomes reported by Al-Ghamdi and 2002;

Michalaki et al., 2015 that used logistic or generalized ordered logistic regression in modelling accident severity.

CONCLUSION

This study has effectively identified the predictors of road traffic accident severity in Zimbabwe using logistic

regression modelling. Key factors such as Over speeding, alcohol use, fatigue, driving experience, mobile phone

use and wet road conditions were found to considerably influence the likelihood of severe accidents. These

behaviours align with global research that identifies human error as the dominant contributor to road crashes.

Although demographic variables such as age, gender, time of accident and vehicle condition did not reach

statistical significance, their directional effects remain consistent with existing literature and may become

significant with larger or more balanced datasets. The logistic regression model provided an acceptable level of

Page 3203

www.rsisinternational.org

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XI November 2025

predictive accuracy and demonstrated that incorporating behavioral and environmental variables significantly

improves the explanation of accident severity. The findings mirror evidence from Zimbabwe and other countries,

reinforcing the conclusion that behavioral interventions, strict enforcement of traffic regulations and targeted

training for inexperienced drivers are essential for reducing severe accident outcomes.

Limitations of Study

Despite producing important findings, this study is subject to some limitations that should be considered when

interpreting the results. The data were obtained from reported road traffic accidents within specific districts,

which may not fully capture all crash types occurring nationally. Minor crashes, unreported incidents, or

accidents occurring in remote rural areas may be under-represented, creating possible selection bias toward more

serious or police-attended cases. Several key predictors are based on self-reported or police-reported behaviours,

including alcohol use, over-speeding, fatigue, and mobile phone use. These behaviours are highly sensitive and

prone to under-reporting due to social desirability bias, fear of legal consequences or recall bias. Under-reporting

biases risk estimates leading to uncertainty in estimated associations. These limitations highlight the need for

future studies incorporating probability sampling, objective measures of driver behaviour and broader

geographical coverage to strengthen the generalisability and causal interpretation.

RECOMMENDATIONS

Based on the findings of this study, the following recommendations are proposed:

1. Strengthen Speed Enforcement

Over speeding was the strongest predictor of accident severity. Deploying more speed cameras and radar systems

at high-risk areas, especially during peak hours, increasing fines and penalties for repeat offenders can improve

the situation.

2. Implement Anti-Drunk Driving Campaigns

Alcohol use doubled the odds of severe crashes. Conducting regular breathalyser checkpoints and implementing

awareness campaigns targeting high-risk groups such as young male drivers can reduce alcohol use while

driving.

3. Address Driver Fatigue

Fatigue significantly increased accident severity. Long-distance drivers are encouraged to do regular rest-breaks

and promote policies that regulate hours of driving, especially for commercial drivers to combat driver fatigue.

Introducing fatigue awareness campaigns during festive and peak travel seasons.

4. Enforce Laws Against Mobile Phone Use While Driving

Mobile phone distractions strongly predict severe crashes. Intensify police monitoring of mobile phone use on

major roads. Promote hands-free alternatives and public educational campaigns.

5. Target Inexperienced Drivers in Training Programs

Low driving experience was associated with higher accident severity. The road traffic authorities may introduce

structured defensive driving courses for new drivers and strengthen the road test to ensure better preparedness.

6. Improve Road Maintenance and Stormwater Management

Wet road conditions increased severity by 77%. The authorities may repair potholes and improve road drainage

systems and install warning signs at slippery or water-logged sections.

Page 3204

www.rsisinternational.org

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XI November 2025

7. Enhance Public Road Safety Education

The traffic safety authorities may implement nationwide campaigns focusing on risky behaviours highlighted in

the study and promoting safe driving practices, particularly during high-risk periods such as night driving and

use social media, radio and community meetings to reach diverse audiences.

8. Strengthen Data Collection and Road Safety Policy Enforcement

Improve accident data reporting systems for accurate policymaking. Align national road safety regulations with

global standards recommended by WHO.

Ethical Considerations

This study adhered strictly to established ethical principles governing the use of secondary data, statistical

modelling and research involving road traffic accidents records in Zimbabwe. The study relied solely on

secondary accident data recorded routinely by traffic police officers during accident investigations. The informed

consent was not required since the data did not involve direct interaction with human participants. However,

ethical best practice dictates that such datasets must be handled responsibly with respect for privacy and

confidentiality. The data were provided in anonymised form and used strictly for academic research purposes.

All identifying information such as names, national identification numbers, vehicle registration numbers, phone

numbers, addresses, and licence numbers was removed prior to the analysis. The researchers did not have access

to any personal identifiers.

The results were presented in aggregate form (percentages, odds ratios and model coefficients), to ensure that

no individual driver or accident could be traced or identified. No attempt was made to use the data for legal,

punitive or discriminatory action against drivers, police officers or transport companies. The study avoided any

interpretations or conclusions that could unjustly stigmatise specific groups such as gender, age categories or

professional drivers. Findings were framed to improve road safety awareness and inform prevention strategies

rather than assign blame. The researchers followed transparent methodological procedures, including accurate

reporting of statistical models, coding schemes, assumptions and limitations. The results were not manipulated

to favour any outcomes. All data transformations (coding, cleaning, recategorisation) were documented clearly.

This transparency ensures replicability and strengthens the trustworthiness of the findings. The logistic

regression models were used to identify risk factors, not to profile or target individual drivers. The models were

validated to avoid misleading interpretations that could influence policy unfairly. Predictive results were

interpreted cautiously and contextualised within broader road safety challenges in Zimbabwe. The findings will

be disseminated responsibly through academic publications without revealing sensitive details. The emphasis

will be on improving public road safety, driver education and accident prevention strategies.

ACKNOWLEDGEMENTS

All sources were acknowledged.

Conflicts of Interests

The authors declare that there was no conflict of interest.

Data Availability

The data used in this study is not publicly available due to the ethical considerations of the participants.

REFERENCES

1. Adavikottu, A., & Velaga, N. R. (2021). Analysis of factors influencing aggressive driver behaviour and

crash involvement. Traffic Injury Prevention, 22(sup1), S21–S26.

Page 3205

www.rsisinternational.org

INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN SOCIAL SCIENCE (IJRISS)

ISSN No. 2454-6186 | DOI: 10.47772/IJRISS | Volume IX Issue XI November 2025

2. Ahmed, S. K., Mohammed, M. G., Abdulqadir, S. O., El-Kader, R. G. A., El-Shall, N. A., Chandran, D.,

Rehman, M. E. U., & Dhama, K. (2023). Road traffic accidental injuries and deaths: A neglected global

health issue. Health Science Reports, 6(5), e1240.

3. Al-Ghamdi, A. S. (2002). Using logistic regression to estimate the influence of accident factors on

accident severity. Accident Analysis & Prevention, 34(6), 729–741.

4. Chen, S., Shao, H., & Ji, X. (2021). Insights into factors affecting traffic accident severity of novice and

experienced drivers: A machine learning approach. International Journal of Environmental Research and

Public Health, 18(23), 12725.

5. Chibaro, M., Munuhwa, S., Mupfiga, C., & Farai, F. C. (2024). Human factors and road traffic safety:

An inquiry into the causes of upsurge in road traffic accidents in Harare Metropolitan Province,

Zimbabwe. Asian Journal of Management, Entrepreneurship and Social Science, 4(03), 92–108.

6. Deepthi, A., Chaithrika, B., Teena, C., Manvitha, G., & Keerthana, G. S. (2023, December). Enhancing

road safety: Predicting severity of accidents. In International Conference on Data Science, Machine

Learning and Applications (pp. 775–783). Springer Nature Singapore.

7. Dominguez-Almendros, S., Benitez-Parejo, N., & Gonzalez-Ramirez, A. R. (2011). Logistic regression

models. Allergologia et Immunopathologia, 39, 295–305.

8. Eboli, L., Forciniti, C., & Mazzulla, G. (2020). Factors influencing accident severity: An analysis by road

accident type. Transportation Research Procedia, 47, 449–456.

9. Foya, D. (2019). Road traffic accidents, causes, trends and preventive measures: The case of City of

Harare (2012–2018). International Journal of Research and Innovation in Social Science, 3(10), 776–

792.

10. Goniewicz, K., Goniewicz, M., Pawłowski, W., & Fiedor, P. (2016). Road accident rates: Strategies and

programmes for improving road traffic safety. European Journal of Trauma and Emergency Surgery,

42(4), 433–438.

11. Haerani, S., Parmitasari, R. D. A., Aponno, E. H., & Aunalal, Z. I. (2019). Moderating effects of age on

personality, driving behavior, and driving outcomes. International Journal of Human Rights in

Healthcare, 12(2), 91–104.

12. https://www.trafficsafety.co.zw.

13. McCarty, D., & Kim, H. W. (2024). Risky behaviors and road safety: An exploration of age and gender

influences on road accident rates. PLoS ONE, 19(1), e0296663.

14. Michalaki, P., Quddus, M. A., Pitfield, D., & Huetson, A. (2015). Exploring the factors affecting

motorway accident severity in England using the generalised ordered logistic regression model. Journal

of Safety Research, 55, 89–97.

15. Munuhwa, S., Govere, E., Samuel, S., & Chiwira, O. (2020). Managing road traffic accidents using a

systems approach: Case of Botswana—Empirical review. Journal of Economics and Sustainable

Development, 10(11), 176–188.

16. Muvuringi, M. P. (2012). Road traffic accidents in Zimbabwe: Influencing factors, impact and strategies.

Vrije University Amsterdam, KIT (Royal Tropical Institute).

Page 3206

www.rsisinternational.org