Ethical consideration
Data used in this study are publicly available data from the WHO Global Status Report on Road Safety and do not need approval from a Human Research Ethics Committee.
Definition of 2-3 wheelers
Motorized 2-3 wheelers refers to powered two- and three- wheelers [3]. According to the WHO definition “motorized 2-3 wheelers or powered two- and three-wheelers (PTWs) are motor-operated two- and three-wheeled vehicles, powered by either a combustion engine or rechargeable batteries”. These included motorcycles, scooters, e-bikes, tricycles motor- rickshaws or e-rickshaws [3].
Data collection
Data were retrieved from the WHO Global Status Reports on Road Safety for years 2007, 2010, 2013, and 2016, which were published in 2009, 2013, 2015, and 2018, respectively [1, 9, 10, 13]. These reports had data on 178, 182, 180, and 175 countries, respectively, with complete data on motorized 2-3 wheelers death on 115 (64.6%), 123 (67.6%), 117 (65.0%), and 122 (69.7%) countries, respectively. The area of countries was retrieved from the infoplease.com website [15].
Studied variables
Studied variables included the percentage of estimated helmet-wearing rate, the effectiveness of helmet law enforcement, the effectiveness of speed law enforcement, estimated road traffic deaths rate per 100 000 population, percentage of motorized 2-3 wheelers death, country population, Gross National Income (GNI) per capita in US dollars, number of registered vehicles, and percentage of motorized 2-3 wheelers in each country.
The percentage of motorized 2-3 wheelers death included all riders (drivers or passengers). The percentage of estimated helmet-wearing rate in our study included all riders. However, if data was not available on all riders, we used instead the reported percentage of estimated helmet-wearing rate of drivers. Information on the overall effectiveness levels of both helmet law enforcement and speed limit enforcement were scored on a scale of 0 to 10, where 0 is “not effective” and 10 is “highly effective” based on professional consensus in each country.
Data entry
Data collected during all studied years were coded and entered into the MS Excel program in two formats: vertical data format (the same variables in all studied years were entered into a single column in order of years with an added year variable), and horizontal data format (each variable in each studied year entered into a separate column). Data were rechecked for accuracy and consistency and exported into SPSS for analysis.
Calculations
The population density was calculated by dividing the total population by country area (number of people/mile square). The number of motorized 2-3 wheelers was calculated by multiplying the percentage of motorized 2-3 wheelers by the total number of registered vehicles. Motorized 2-3 wheelers death rate was calculated by multiplying the percentage of motorized 2-3 wheelers death by the estimated traffic road death rates per 100 000 population. Vehicle per person ratio was calculated by dividing the total number of registered vehicles by the total population. Motorized 2-3 wheelers per person ratio was calculated by dividing the total number of motorized 2-3 wheelers by total population.
Statistical analysis
Mortality rates were highly skewed to the left. Accordingly, median (interquartile range, IQR) were used in reporting the data. We used a mixed linear model (MLM) to assess the factors affecting motorized 2-3 wheelers death rates over time. Death rates were transformed to a normal distribution to fulfill the requirements of the MLM. Log transformation had the best normal distribution over time and within each year and was used as the outcome variable.
The MLM analyses data of repeated measures (years) of each country (subject) separately, taking into account both the slope and intercept of each linear line of a country (within-subjects correlation). Studying the slope has an important advantage of addressing missing data and the nonlinear relationship between different factors. MLM assumes a normal distribution of the outcome (dependent) variable. The logarithmic transformation of death rate was the dependent variable of the MLM model, while independent covariates do not need to have a normal distribution which can be ordinal, continuous, or categorical data.
The used MLM model was a strict unstructured, main-effects model with repeated measures. It included a fixed effect, type III sum of squares error (due to the unbalanced data), and random effects for the independent variables (factors and covariates). These strict requirements assume that the variance of each studied year and the covariance (correlation) between the studied independent factors are different. The change of the outcome dependent variable (death rate) was studied over time by entering the studied years as a categorical factor (factor = year) while independent variables as covariates. These included continuous variables (population density, GNI per capita, vehicle per person ratio, and motorized 2-3 wheelers per person ratio) and ordinal variables (speed law enforcement (0-10) and helmet law enforcement (0-10)). We tested different interactions in the model, excluded non-significant and included significant interactions in the final MLM model. Accordingly, the interaction between vehicle per person ratio and motorized 2-3 wheelers per ratio was added to the final main effects model.
After achieving the results of the final MLM model, several univariate post-hoc analyses were performed to explain our findings. Spearman rank correlation test was used to study the correlation between different continuous or ordinal variables. Wilcoxon signed-rank test was used to compare the continuous or ordinal data of two dependent groups. Friedman test was used to compare the continuous or ordinal data of more than two dependent groups. Mann-Whitney U test was used to compare the continuous or ordinal data of two independent groups, while Kruskal Wallis test was used to compare the continuous or ordinal data of more than two independent groups. Data were analyzed with the IBM SPSS Statistics version 26 (SPSS Inc, Chicago, IL, USA). A p-value of less than 0.05 was accepted as statistically significant.