Accident Analysis and Prevention 66 (2014) 36–42
Contents lists available at ScienceDirect
Accident Analysis and Prevention journal homepage: www.elsevier.com/locate/aap
Dominant role of drivers’ attitude in prevention of road trafﬁc crashes: A study on knowledge, attitude, and practice of drivers in Iran Ramazan Mirzaei a , Nima Hafezi-Nejad b , Mohammad Sadegh Sabagh b , Alireza Ansari Moghaddam a , Vahid Eslami b,c , Fatemeh Rakhshani a,e,∗∗ , Vafa Rahimi-Movaghar b,d,∗ a
School of Medicine/Public Health, Zahedan University of Medical Sciences (ZUMS), Zahedan, Iran Sina Trauma and Surgery Research Center, Tehran University of Medical Sciences (TUMS), Tehran, Iran c Students’ Scientiﬁc Research Center (SSRC), Tehran University of Medical Sciences (TUMS), Tehran, Iran d Research Center for Neural Repair, University of Tehran, Tehran, Iran e Safety Promotion and Injury Prevention Research Center, School of Public Health, Shahid Beheshti University of Medical Sciences, Tehran, Iran b
a r t i c l e
i n f o
Article history: Received 10 July 2013 Received in revised form 28 December 2013 Accepted 12 January 2014 Keywords: Drivers Road trafﬁc crashes Knowledge Attitude Practice Prevention
a b s t r a c t Objective: Evaluating the relation between Iranian drivers’ knowledge, attitude, and practice (KAP) regarding trafﬁc regulations, and their deterministic effect on road trafﬁc crashes (RTCs). Setting: Two cities of Tehran and Zahedan, Iran. Methods: A cross-sectional study was designed. Using a simpliﬁed cluster sampling design, 2200 motor vehicle drivers including 1200 in Tehran and 1000 in Zahedan were selected. Sixty locations in Tehran and 50 in Zahedan were chosen. In each pre-identiﬁed location, 20 adult drivers were approached consecutively. A questionnaire developed by researchers was ﬁlled by each participant. The questionnaire had four sections including items assessing the demographics, knowledge, attitude and practice of drivers toward trafﬁc regulations. Logistic regression analysis was used to evaluate the relationship between the RTCs and KAP variables. Results: The study sample consisted of 619 (28.1%) occupational and 1580 (71.8%) private drivers. Among them, 86.4% were male. The median age was 33.6 ± 10.83. Drivers in Tehran and Zahedan had no signiﬁcant differences between their mean scores of KAP items of the questionnaire. Higher knowledge, safer attitude, and safer practice were associated with a decreased number of RTC. After adjusting for possible confounders, increase of one standard deviation in attitude and practice scores (but not knowledge) resulted in 26.4% and 18.5% decrease in RTC, respectively. Finally, considering knowledge, attitude and practice of drivers in one model to assess their mutual effect, it was shown that only attitude is signiﬁcantly associated with a decrease of RTC (OR = 0.76, P = 0.007). Conclusion: Increase in attitude and practice accompanied with decreased number of RTCs in Iranian drivers. Speciﬁcally, drivers’ attitude had the crucial effect. It is not knowledge and standard trafﬁc education; rather it is how such education is registered as an attitude that translates what is being learned into actions. Without safer attitude, even safer self-reported practice will not result in lower RTCs. © 2014 Elsevier Ltd. All rights reserved.
1. Introduction 1.1. Background
∗ Corresponding author at: Sina Trauma and Surgery Research Center, Sina Hospital, Hassan-Abad Square, Imam Khomeini Avenue, Tehran University of Medical Sciences, Tehran 11365-3876, Iran. Tel.: +98 915 342 2682/216 671 8311; fax: +98 216 670 5140. ∗∗ Corresponding author. E-mail addresses: [email protected]
(R. Mirzaei), [email protected]
(N. Hafezi-Nejad), [email protected]
(M. Sadegh Sabagh), [email protected]
(A. Ansari Moghaddam), [email protected]
(V. Eslami), [email protected]
(F. Rakhshani), v [email protected]
, v [email protected]
, v [email protected]
(V. Rahimi-Movaghar). 0001-4575/$ – see front matter © 2014 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.aap.2014.01.013
An estimated number of 1.2 million people are killed and 50 millions are annually injured in road trafﬁc crashes (RTCs) in the world. This will increase by about 65% over the next 20 years (Peden and Sminkey, 2004). RTC is of the main causes of disability adjusted life years (DALY) both in developed and developing countries (Rasouli et al., 2008; Rahimi-Movaghar et al., 2009). The great burden of RTC describes the need for new studies to unveil its epidemiology, associated factors and preventive strategies (Rahimi-Movaghar, 2010; Moradi et al., 2012).
R. Mirzaei et al. / Accident Analysis and Prevention 66 (2014) 36–42
1.2. Role of knowledge, attitude and practice in RTC Previous studies have emphasized on high risk driving as a key factor in the occurrence of RTCs (Evans, 1996, 2003; Redelmeier et al., 2003). Since inappropriate attitude correlates with risky behavior, drivers’ attitudes toward trafﬁc safety would be an effective determinant of RTC (Nabi et al., 2007). Prior studies have elucidated the role of drivers’ behavior (also called as drivers’ practice) on RTCs. Low practice scores, even in the self-reported questionnaires, were associated with up to 1.5 times more risk of RTC (Ivers et al., 2009). In all, it is proposed that knowledge, attitude, and practice (KAP) of drivers toward trafﬁc regulations, can be the most important determinants of RTCs (Wang et al., 2012; Iversen and Rundmo, 2004; Ali et al., 2011; Moradi et al., 2012). Many behavioral features including the acceptance of higher levels of risk, sensation seeking, prestige seeking, and characteristics such as substance abuse, drunk driving, and not using safety belts contribute to an increased risk of RTC (Meuser et al., 2010; Onyema and Oladepo, 2011; Mcevoy et al., 2006; Nabi et al., 2007; Iribhogbe and Osime, 2008; Shams and Rahimi-Movaghar, 2009). Suitable feedback can change the driving behavior (Evans, 1996, 2003; Redelmeier et al., 2003; Wang et al., 2009). Recent researches highlight the need to modify drivers’ behavior as a major target for trafﬁc safety interventions (Purc-Stephenson et al., 2010). Moreover, it is well recognized that safer behavior is the result of correct attitude and improved knowledge (Wang et al., 2012; MartinovCvejin et al., 1993; Teoh et al., 2004).
consecutively and a previously validated questionnaire was ﬁlled for each participant. Subjects were approached in locations where they were expected to have sufﬁcient waiting time to complete the questionnaire. The locations were identiﬁed by the study group to include public places that were present in all districts. The locations were evenly distributed in the main 22 districts of Tehran and the main 5 districts of Zahedan. If drivers were taxi, bus, or truck drivers, they were approached on their rest breaks or in gas stations. Private drivers were approached inside bank queues. First of all, consultation with two experienced epidemiologist and biostatistician were performed to obtain the optimum stratiﬁcation and minimum skew. In addition, to assure that the responses would not be skewed, based on the place of interviews (interviewing in public places vs. intersections and queues), stratiﬁcation was performed and places were chosen according to the previous samplings in developing countries (Nordfjærn et al., 2011; S¸ims¸eko˘glua et al., 2012), and particularly Iranian drivers (Ozkan et al., 2006). The RAs were trained before approaching the drivers. For illiterate drivers, RAs facilitated the ﬁlling of the questionnaires. All of the participants were adults (≥18 years old) and were drivers. Verbal consent was obtained from the participants prior to inclusion in the study. All subjects who completed the questionnaire were included and none were excluded due to reported mileage or any other characteristics.
2.3. The questionnaire 1.3. Situation in Iran In Iran, RTCs are of the leading causes of disability and mortality (Rahimi-Movaghar et al., 2009). Iran is among the countries with highest number of RTC fatalities in the world (Rasouli et al., 2008). Nevertheless, no previous study has evaluated the association between knowledge, attitude and practice of drivers in the region. For the ﬁrst time, we evaluated the KAP of drivers in two cities of Tehran and Zahedan, Iran. We described the relationships between drivers’ characteristics and RTCs. First, we assessed whether if driver’s KAP are associated with the incidence of RTCs. Then, we aimed to extract the key determinant of RTCs, from knowledge to the ﬁnal act of practice. 2. Materials and methods 2.1. Settings This cross sectional study was performed from March 2010 to March 2011. The study population included a representative sample of drivers in Tehran and Zahedan. Tehran is the capital of Iran and represents a high-income city. Zahedan is categorized as a low income, border-city in Iran. Further details on the demographics, population and gross domestic product (GDP) share of the mentioned cities are presented elsewhere (http://www.amar. org.ir/portals/2/Files1385/kolliostan/tehran/2303.pdf and http:// www.amar.org.ir/portals/2/Files1385/kolliostan/sistan/1103.pdf).
We drafted the questionnaire after reviewing the literature and initially piloted the items with 40 drivers to determine the time needed to answer. In addition, each pilot tester was interviewed to assess the validity of written responses. Cronbach’s alpha was calculated to ensure the reliability of the questions in each section. Questions with Cronbach’s alpha below 15% were discarded from the ﬁnal version. The ﬁnal questionnaire included 16 demographic items, 24 items regarding the knowledge of driving and trafﬁc rules, 26 items determining attitudes toward trafﬁc regulations and 19 items on driving behaviors. The face and content validity of knowledge, attitude, and behavior items were discussed and conﬁrmed by an external committee including police ofﬁcers. The questionnaire is translated into English and could be seen in the supplement. For the analysis of the questionnaires’ items, the maximum scores gained from knowledge, attitude and practice items were set at 91, 26 and 76, respectively. Higher scores indicate a safer KAP. The subjects were also asked for basic demographic questions, including age, gender, educational level, as well as wearing medical glasses, smoking, having driving license and liability insurance. The project was approved by the review board of both Sina Trauma and Surgery Research Center of Tehran University of Medical Sciences, and Zahedan University of Medical Sciences.
3. Statistical analysis 3.1. Univariate analysis
2.2. Study design We used a simpliﬁed method of cluster sampling to represent target population (Bennett et al., 1991). In all, 1200 drivers in Tehran, and 1000 drivers in Zahedan were selected through randomized cluster sampling. Sixty locations in Tehran and ﬁfty locations in Zahedan were chosen. During the visits of our research assistants (RA) to each location, 20 drivers were approached
We used Student’s t test to compare means of the associated factors and KAP (knowledge, attitude, and practice) variables between Tehran and Zahedan. Pooling the data of both cities to evaluate the relationship between KAP variables and RTCs in the last 3 years; employing univariate analysis using Student’s t test (KAP grades were compared in subjects with and without RTC).
R. Mirzaei et al. / Accident Analysis and Prevention 66 (2014) 36–42
3.2. Application of logistic regression modeling Assessing the consistency of the KAP variables and their relation with a dependent outcome has been the subject of several studies since the introduction of the theory of planned behavior (TPB) (Ajzen, 1985). Also, as an applicable statistical measure, regression analyses were frequently used for indicating the KAP variables as determinants of a studied event (including RTC) (Ajzen, 1985; Lawson et al., 2012). Speciﬁcally, in safety sciences, logistic regression has shown to deliver a realistic modeling and is of the widely utilized methods. Numerous studies have adopted logistic regression models in extracting the risk factors of RTC which are rooted in human behavior, including the use of cell phones (Wilson et al., 2003), smoking (Ryb et al., 2007), drunk driving (Labrie et al., 2011) and KAP variables (Koppel et al., 2012). Therefore, we used logistic regression modeling to apply KAP framework in prediction of RTC and make our results comparable and in line with the current literature. Moreover, logistic regression allows us to control the effect of confounding variables. Previous studies have proposed age, gender (Turner and Mcclure, 2003), drivers’ education and licensing (Sivak et al., 1989c; Blows et al., 2005), smoking (Ryb et al., 2007) and wearing medical glasses (Sagberg, 2006) as potential confounders of KAP variables’ effect on RTCs. Including the mentioned characteristics in our demographic variables, four multivariate logistic regression models were conducted. All four models were adjusted for the demographic variables, accordingly (Nordfjærn et al., 2012). Adjustments for demographic variables were performed in a stepwise manner following Jewell (2003). In addition, models 1, 2, and 3 included knowledge, attitude and practice (behavior), respectively. Model 4 included all three KAP variables, concurrently. The ﬁnal model was used to unveil the key factor which has a signiﬁcant association even when the mutual effect of other KAP variables is stratiﬁed. 3.3. Control for bias and drawbacks of logistic regression modeling Despite mentioned advantages of using logistic regression modeling, certain drawbacks of the model should be identiﬁed and controlled as well. In this regard, we followed established guidelines for proper assessment of logistic regression modeling (Kalil et al., 2010). Selection bias is one of the main drawbacks which can be involved in logistic regression modeling. This can be more dramatic when working in a setting of rare events (Steyerberg et al., 2003; Nemes et al., 2009). However, we overcame this issue by acquiring a large sample size of more than 2000 drivers. Increasing the sample size reduces the analytically induced bias and protects against extreme value estimates. In addition, it increased the number of events per variables (EPV) in our study. We considered the number of predictors, the effect size of the coefﬁcients (as described, from the above-mentioned literature) and the correlations among the predictors in calculating our sample size (Courvoisier et al., 2011). By default, the logistic regression module of the SPSS software will not include cases with missing values, in the model. We took a conservative stance in performing complete case analysis and acquiring a large sample size helped us to attain our statistical power in this regard. In all, 996 subjects reported a history of RTC involvement which accounts for 45.3% of all participated drivers. When the number of favorable events falls into lower counts, using exact tests would lead us to a more precise estimate (Steyerberg et al., 2003). However, our study was not vulnerable to this type of bias. From a statistical perspective, when comparing the results of the events vs. non-events
(those with and without RTC as the binary choice of our logistic regression models), the nearer we get to the 1:1 ratio, the greater power of study we will gain; within the same general sample size. The next important point for deriving a logistic regression model is the overall calibration of our model. A model will not be reliable (and thus, accounting on its predictors would be defective, even if a predictor reveals a signiﬁcant effect size) unless being generally calibrated. The calibration of each logistic regression model is in means of concordance between the model’s predictions and the observed probabilities. Technically, it can be estimated by the Hosmer–Lemeshow X2 test for the goodness of ﬁt. All our models were evaluated to have acceptable calibration. Without calibration, big amounts of bias can be involved in a logistic regression modeling. In addition, we conducted an internal validation using 100 bootstrap simulations for the models. The procedure and choice of 100 samples were performed following Steyerberg et al. (2001, 2003). Multicollinearity of the model variables was ruled out by observing the tolerance and the variance inﬂation factor (VIF). In the ﬁnal model all variables had VIF ≤ 1.5. Finally, we suggest external validation of our ﬁnal logistic regression model in separate data set; to conﬁrm its applicability for generalized interpretations.
3.4. Standardization and comparability of the methods/results Certain strategies were undertaken to ensure standardization and comparability of our methods/results with the current literature. Previous studies have highlighted the role of cross-cultural differences on the association of the RTCs and KAP variables (Ozkan et al., 2006; Lund and Rundmo, 2009; Nordfjærn et al., 2011). Thus, they emphasize on the country/region speciﬁc platforms for KAP studies. In line with the suggested recommendations, a domestically designed, reliability tested and validated questionnaire on the KAP of the drivers regarding the road trafﬁc regulations was piloted and implemented in this study. The framework of the questionnaire was similar to earlier KAP studies on the trafﬁc safety (Iversen and Rundmo, 2004; Nordfjærn et al., 2012). Similar questions have been previously experienced among developing countries (Nordfjærn et al., 2011; S¸ims¸eko˘glua et al., 2012) and speciﬁcally a small sample of Iranians (Ozkan et al., 2006). However, in order to achieve meaningful changes and to ensure that our results are comparable with other questionnaires and surveys, individuals’ grades for each KAP variable were converted into changes in standard deviations (SDs) and then were included in the logistic regression models. Normality of the variables was approved by Kolmogorov– Smirnov test. We used ordinal SD categorized KAP variables for our analysis. Changes in SDs, were calculated in a manner similar to the standard scores. Brieﬂy, for each of the KAP variables, mean and SDs of the sample were obtained. T-score was used as the standard score and was calculated as t = (x − mean)/SD. The participants were scored based on their ranking and placement in the sample population. An individual’s standardized score describes the number of SDs, where he/she places, above or below the mean of the populations. Hosmer–Lemeshow X2 test was checked to conﬁrm the calibration of the models (P value > 0.01, X2 < 20) (Lemeshow and Hosmer, 1982). Choosing logistic regression models and adjustments for the demographic variables were also to facilitate the comparability of our results with other studies and questionnaires. A type I error was set at 0.05 for all tests. The collected data was analyzed using SPSS software (v.11.0 and 18.0, SPSS Inc., IL, USA).
R. Mirzaei et al. / Accident Analysis and Prevention 66 (2014) 36–42 Table 1 Principle characteristics of the study population. Variable
Number of participants
18–24 25–34 35–44 45–54 55–64 65 and higher
493 827 464 306 99 11
22.4 37.6 22.1 13.9 4.5 0.5
Illiterate or elementary Less than college College or higher
140 905 1155
6.3 41.1 52.5
Having driving license
Having liability insurance
Medical glasses while driving
Yes No Not applicable
536 108 1556
24.4 4.9 70.7
Table 3 Univariate analysis of knowledge, attitude and practice of drivers with and without RTC. Variable
Without RTC (No. 1204)
With RTC (No. 996)
T score (P value)
Knowledge Attitude Practice
55.61 ± 10.24 14.55 ± 7.12 63.80 ± 7.76
54.17 ± 11.05 12.92 ± 7.91 62.08 ± 8.02
−3.15 (0.002) −5.09 (