category: literaturenote citekey: haghaniroadsafetyresearchcontext2022 title: "Road safety research in the context of low- and middle-income countries: Macro-scale literature analyses, trends, knowledge gaps and challenges" authors: "Haghani, Milad; Behnood, Ali; Dixit, Vinayak; Oviedo-Trespalacios, Oscar" year: 2022 date: 2022-02-01 2022-02-01 doi: 10.1016/j.ssci.2021.105513 publication: Safety Science url: "https://www.sciencedirect.com/science/article/pii/S0925753521003568" zotero_key: QIIFSYVB zotero_storage: FR29DA7N collections: doktoritöö / HLO folder: 001_artiklid/Road Safety Indicators firstAuthor: "Haghani, Milad"

status: converted

Contents lists available at ScienceDirect

Safety Science

journal homepage: www.elsevier.com/locate/safety

Road safety research in the context of low- and middle-income countries: Macro-scale literature analyses, trends, knowledge gaps and challenges

Milad Haghani a,* , Ali Behnood ^b , Vinayak Dixit ^a , Oscar Oviedo-Trespalacios ^c

^a School of Civil and Environmental Engineering, The University of New South Wales, UNSW Sydney, Australia
^b Lyles School of Civil Engineering, Purdue University, USA
^c Centre for Accident Research & Road Safety-Queensland (CARRS-Q), Queensland University of Technology (QUT), Australia

ARTICLE INFO

Keywords: Road trauma Road crashes Traffic safety Road safety

Low- and middle-income countries

ABSTRACT

Road users in low- and middle-income countries (LMICs) are overrepresented in road trauma statistics. Despite the relative success of many high-income countries (HICs) in reducing deaths on their roads, not much tangible progress has been made in LMICs. Also, on the research front, the vast majority of road safety knowledge has been emerging from institutes of HICs. Considering significant differences in driving culture, legislation, and traffic law enforcement between LMICs and HICs, it seems essential that research on road safety within LMICs intensifies beyond the existing rate to produce the much-needed local knowledge and to develop initiatives that meet their safety needs and upgrade their practices. To facilitate this, here, the landscape and temporal trends of road safety research in LMICs are analysed while contrasting them with those of the general scholarly literature on road safety. It is estimated that slightly less than 10% of the road safety research has been undertaken in the contexts of LMICs, which is extremely disproportionate considering the fact that most road traffic deaths and injuries occur in LMICs. Questionnaire-based research on socio-psychological aspects of driving, cycling, and walking as well as statistical modelling of road crash data seem to have made up the dominant focus of LMIC researchers within the recent years. Areas of road safety research that are underrepresented in LMIC studies are also identified in this work. Patterns of authorship and co-authorship in LMIC studies are also analysed at the level of countries, organisations, and authors. It is hoped that this effort can contribute to further invigoration of road safety research in LMICs and to highlighting the current knowledge gaps, while also giving better recognition to active road safety researchers of LMICs, and thereby, prompting more international collaborations in this domain.

1. Introduction

*Road safety research in the context of low- and middle-income countries as an underrepresented domain*—Findings from scholarly research on road accidents have always played a crucial and fundamental role in developing safety interventions and determining its effectiveness, and as such, have been vital in informing road safety strategy and policies around the world. However, it has been long established that the effectiveness of such interventions could be, to significant degrees, context dependant and modulated by cultural and social norms as well as population characteristics and their economic conditions. The fact that a road safety intervention has worked in a certain setting and with a certain population does not necessarily guarantee that the prevention be effective to similar degrees in populations with different socioeconomic

characteristics. Noting this issue alongside the fact that the majority of the body of scholarly research findings related to road safety have originated from high-income countries (HICs) highlights a broad research gap: Road safety research in low- and middle-income countries (LMICs)—described as "a neglected research area" (Perel et al., 2007) should be given more attention and priority (Heydari et al., 2019). This would be a key step in producing road safety knowledge specific to various cultural conditions and ultimately devising policies with empirically established local relevance and tailored to specific populations of interest.

*Aims and motivation of the study*—Motivated by the aforesaid issue as well as the disproportionate burden of global road trauma that populations of LMICs are currently shouldering, here in this work, we look at the landscape of road safety research in the context of LMICs at a broad

E-mail address: milad.haghani@unsw.edu.au (M. Haghani).

^* Corresponding author.

scope while contrasting it with the general body of road safety literature. This work is, in fact, the first scoping review of the LMICs road safety research at such broad scale. In undertaking this large-scale investigation, our main indicators are bibliometric aspects of the research articles published in road safety literature. This includes indicators such as titles, abstracts, keywords lists, years of publication, authors lists, author affiliations and countries, and lists of references. Given the size of the literature that we are covering, our analyses will mainly be undertaken at a macro scale, although these macro-level analyses are also supplemented by a more detailed scoping review of the field. We mainly focus on the following questions in this study:

(i) What portion of the road safety literature is dedicated to LMICs and how big is this subset?
(ii) How does the landscape of road safety research in LMICs contrast with that of the general road safety literature and what are the respective knowledge gaps?
(iii) What are the temporal trends in road safety research in the context of LMICs?
(iv) Which countries is this research mostly originating from?
(v) Which authors and organisations are most active in producing road safety research in the context of LMICs?
(vi) What are the patterns of collaborations between researchers affiliated with LMICs and those of HICs?

*Potential impact*—By investigating these questions, we aim this work to achieve, or at least contribute to, the goal of invigorating road safety research in LMICs, identify potential overlooked areas of research in those countries, and facilitate scientific collaborations between authors of LMICs and HICs in order to further advance this research front.

*Structure of the paper and employed methods*—The structure of the rest of the paper is as follows. Section 2 provides summary statistics of road safety trauma in LMICs as compared with those of HICs, further delineating the motivation of this work. In Section 3, building on a previous method for sourcing the literature of road safety (Haghani et al., 2021a), we introduce a systematic search method that enables us to approximate and obtain both the general literature of road safety as well as that of the LMICs. These will constitute the datasets of research articles that are analysed throughout the paper. Section 4 presents the results of the study. It provides general statistics based on these datasets and analyses patterns of authorship and co-authorship in LMIC studies at various levels of aggregation (Section 4.1) Also, using the methodology of Visualisation of Similarities (VOS) between objects developed by Van Eck and Waltman (2007), in Section 4.2, we analyse major divisions of road safety research in LMIC and the most influential studies within each division. These divisions are also contrasted with those of the general structure of the road safety literature. In Section 4.3, using the methodology of Document Co-citation proposed by Chen (2004), we analyse temporal patterns of road safety research in LMICs. Building on the outcomes of the document co-citation analysis, we also provide a scoping review of the field. Section 5 provides some discussions on the findings and Section 6 concludes the findings and offers areas for future research.

2. Background

LMICs bear road trauma disproportionately —According to the latest figures provided by the World Health Organization (2018) (WHO), an excess of 1.35 million people are fatally injured every year due to road crashes. Along with the rising world population, this number has been consistently on the rise each year, while the rate of road accident deaths relative to the size of world population has been stubbornly constant, despite the efforts undertaken around the world to increase human safety on roads (Bhatti and Ahmed, 2014; Forjuoh, 2003; Stewart et al., 2012; Vecino-Ortiz et al., 2014). Road traffic injuries now constitute the eighth leading cause of death across all age groups and the leading cause of death for children and young adults. Across all the regions, the risk of being fatally injured is the highest in Africa and South East Asia (World Health Organization, 2018). There is no doubt that low- and middleincome countries (LMICs) have been bearing a disproportionate burden of road fatalities (Bener et al., 2003; Bhalla and Shotten, 2019; Nantulya and Reich, 2003; Wegman, 2017). The numbers provided by the WHO have established an undeniable correlation between countries' level of income and their risk of road traffic deaths (Herman et al., 2012a). According to WHO, 90% of road accident deaths occur in LMICs, while these countries only make up 82% of the world's population and 54% of the world's level of motorisation. An individual in a low-income country is more than three times likelier to die from a road accident compared to an individual in a high-income country (HIC) (World Health Organization, 2018). As Fig. 1 shows, while LMICs make up 85% of the world population and only 60% of registered motor vehicles around the world, they bear nearly 93% of all road traffic deaths. Perhaps more importantly, according to Fig. 2, while more than 50% of HICs have experienced reductions in the number of road traffic deaths between 2013 and 2016, only 23.5% of middle-income countries and zero low-income country have managed to reduce deaths on their roads. In other words, countries of different income levels are experiencing different trends in terms of mitigation of road traffic injuries and that further highlights the divide that exist in road safety practices and safety cultures across countries of different income levels.

*Distribution of road accident deaths by user types in LMICs and HICs*—Fig. 3 visualises the income level of countries across the world according to the World Bank definition while also presenting the distribution of road accident deaths by road user types across different regions of the world. An interesting pattern that is observable is the fact that in Africa where there is a higher concentration of low-income countries, pedestrian and cyclist fatalities make up 44% of all deaths, well above the corresponding figure for other regions of the world (McIlroy et al., 2019). However, differences of road safety concerns across HICs and LMICs are not limited to the risk of death on the road per se or the type of road users who are at more risk. They exhibit noticeable differences in other aspects such as post-crash care. According to WHO, the proportion of people who get injured in a road accident and die before reaching a medical facility is over two times larger in an LMIC than an HIC (World Health Organization, 2018).

3. Methods and data

*The source of data*—The two dataset of references that form the basis for the analyses of this work were obtained from the Web of Science (WoS) Core Collection. One dataset represents the general literature of road safety and the other represents the subset of LMIC studies. The following method was used to develop two search queries that can output these datasets. This method, in abstract terms, is based on exporting the entire content of the five specialty road safety journals while excluding the minority of their articles that are not road safety journals, and also sourcing individual road safety papers from anywhere

Fig. 1. Proportion of vehicle registration, road traffic deaths and population across levels of country income according to World Health Organization (2018).

Fig. 2. Percentage of countries within each income level that recorded significant (above 2%) change in number of road accident deaths between 2013 and 2016 according to World Health Organization (2018).

else in the WoS (i.e., published by journals other than the five specialty journals) through an extensive term-based search query. This method results in the general dataset of road safety papers. Then from this dataset, we extract articles in whose title or abstract or keyword list include the name of an LMIC. This subset constitutes the dataset of LMIC articles. This methodology has been visualised in Fig. 4.

*Search query design to obtain general road safety literature*—In designing the search query that was briefly described above, first the contents of the five specialty road safety journals (Accident Analysis & Prevention, Transportation Research Part F, Journal of Safety Research, Traffic Injury Prevention, Analytic Methods in Accident Research) were individually isolated in Scopus and their list of top keywords were individually extracted. These lists of raw terms were subsequently used as the basis for developing term combinations and search units (we refer to a search unit as a term or a combination of terms that can independently produce outcomes in our search query). These units were examined one by one to determine whether they return false positives (i. e., their respective outcomes were sampled and examined in the WoS), and if so, they were either removed or modified to be made more restrictive and specific in order to eliminate the likelihood of false positives. Search units that were approved using this screening test were then combined with one another using the Boolean operator OR. This makes up the term-based component of the search strategy. From the list of raw keywords obtained from each journal, a set of terms and term combinations were also formed that represent non-road safety content of the specialty journals. This list was used in order to exclude irrelevant articles when exporting the full content of specialty journals (in the journal-based part of the search strategy). We refer to this as a termbased filter. The filter part of the query is combined with the rest of the query using the operator NOT. The purpose of this filter is to disallow irrelevant articles of the specialty journals (which constitute a very small set for each journal) into the dataset. The search query described above returned 27, 956 items in the WoS in early 2021 which could be the best estimate of the size of road safety literature at the time.

*Search query modification to obtain the subset of LMIC road safety literature*—Then, in order to extract the subset of LMIC studies from this dataset, the list of the name of LMICs were obtained from the information provided by The Organisation for Economic Co-operation and Development. These names were combined with one another using the operator OR, and then being treated as a unit itself, it was combined with the general body of the search query using the operator AND. The purpose of this unit is to allow only articles of road safety literature that have mentioned the name of a LMIC in their title, abstract or keyword list. The full content of this search query has been provided as an Online Supplementary Material of this article. The content can be simply copied and pasted into the Advanced Search section of the WoS to reproduce the data. The search query described above (with no time span restriction) returned 2646 items in the WoS in February 2021 which could be the best estimate of the size of LMIC road safety literature at the time.

Fig. 3. Income level of countries across the world according to the World Bank definition and distribution of road accident deaths by user types according to World Health Organization (2018).

Fig. 4. An abstract visualisation of the search query method used to obtain the dataset of references of road safety articles in the context of LMICs.

See Online Supplementary Material of this article for a full list of these references.

*The nature of the data*—For both datasets, the full bibliometric information of all items was exported in the text files to be used for further analyses. This includes title of each article, details of their authors, their affiliations, year of publication, journal title, conference information, citation count, document type, abstract of the each article, list of keywords, funding details, and list of references of each article.

4. Analyses and results

4.1. General statistics, authorships, and co-authorships in LMIC road safety literature

*General statistics of the LMIC road safety literature*—The literature of road safety research in the context of LMICs (N = 2646) is estimated to have constituted nearly 9.5% of the general size of the road safety research (N = 27,956). Three main types of scholarly documents in the LMIC sector are Article (82.5%, as opposed to 73.3% of this document type in the general literature), Proceedings Paper (15.65%, as opposed to 21.91% in the general literature) and Review (1.55%, as opposed to 2.3% in the general literature). Nearly 15.57% of LMIC papers have been published by Accident Analysis and Prevention, followed by Traffic Injury Prevention (8.9%), Transportation Research Part F (5.8%), International Journal of Injury Control and Safety Promotion (3.1%), Journal of Safety Research (2.3%), Transportation Research Record (2.1%), Injury Prevention (1.32%), International Journal of Environmental Research and Public Health (1.3%), PLOS One (1.3%), Safety Science (1.1%) and Journal of Transportation Safety & Security (1.0%).

*Distribution of contributions across countries*—Fig. 5 visualises the extent of contribution of authors affiliated from organisations of different countries to LMIC as well as general road safety papers. This is based on the involvement of authors listed on the papers, regardless of whether authors have been listed as a first author or a co-author. In the general literature, authors of the following countries have had the largest number of scholarly contributions (in terms of the number of published studies): United States (n = 8554), China (n = 2944), Australia (n = 2327), Canada (n = 1744), Germany (n = 1631), England (n = 1604), France (n = 1015) and Sweden (n = 952). Within the subdomain of LMIC studies, however, authors of the following countries have been listed most frequently: China (n = 650), United States (n = 473), India (n = 215), Iran (n = 206), Australia (n = 163), Brazil (n = 156), Malaysia (n = 145) and South Korea (n = 136). It goes without saying that authors based in LMICs often collaborate with authors affiliated with organisations of HICs on these studies, hence the presence of authors from HICs such as United States and Australia in the latter list provided above. Additionally, many road safety researchers in LMICs have completed their training in HICs such as Australian and the United States.

*How fast the LMIC literature is expanding*—The rate of accumulation of research papers in both general road safety literature and the LMIC subdomain since 1970 can be visualised in Fig. 5. While earliest appearances of LMIC articles can be traced to 1970 s (with first studies such as Emenalo et al. (1977), analysing road accident data in Zambia), notable presence of LMIC papers in the road safety literature is only observable since 2008. In 2008, the portion of LMIC papers in the general literature

Fig. 5. Number of documents from authors of different countries to the general road safety literature (top) and LMIC subset (middle). The bottom figure shows the number of such papers over time since 1970.

for the first time rose to 7.7%. Since 2008, this percentage has been fluctuating over the years, but the general trend has been on the rise. In 2020, an estimated 15.7% of the papers published on road safety can be linked to research undertaken in relation to LMICs, and this is the highest proportion recorded since 1970. Among the earliest LMIC papers, contributions of authors from African countries such as Nigeria (Asogwa, 1980, 1992; Jegede, 1988; Oluwasanmi, 1993), South Africa (Fernie, 1982; Flisher et al., 1993), Libya (Mekky, 1984) and Kenya (Odero, 1995) seem notably represented. Appendix A lists the oldest LMIC papers according to the WoS record.

*Collaboration of countries in LMIC research*—Fig. 6 provides a different perspective into the country affiliation of authors within the LMIC subset by visualising patterns of co-authorships at the level of countries. Outcomes have been obtained from VOSviewer, the scientometric software

Fig. 6. Network of country collaborations on LMIC road safety papers. The colours represent the average year of publications from authors of each country.

(Van Eck and Waltman, 2010). In this node-and-link visualisation, the spatial proximity of the countries on the map as well as the thickness of links between them are indicative of the extent of collaborations between their authors on LMIC publications. The colour of node visualisation is representative of the average year of publication with the involvement of authors of each country. The most substantial link of collaboration exists between authors affiliated with institutes of China and the United States (n = 136, meaning that on 136 LMIC papers, authors of China and United States organisations have been listed jointly). Overall, authors of the United States institutes have been common collaborators with almost all LMIC countries on their publications, potentially explained by the fact that U.S. is one of the main hubs of transport research. After the United states, most frequent collaborators of Chinese authors have been authors from institutes of Australia and Canada. There are a group of countries (visualised on the right-hand side of the map whose authors have exclusively collaborated with researchers based in the United Stated. This includes researchers of institutes in Philippines, Haiti, Iraq, Egypt, Namibia, Zambia and Guyana. Overall, most of the strong links of collaboration are between an LMIC and an HIC, and collaborations across LMICs are not very notable on this network, although they do exist. Among the LMICs whose authors have a strong presence on this network (i.e., more than 50 documents), contributions with involvement of authors from Colombia (n = 61) are relatively the youngest, with their year of publication averaging 2018.03. In the same category, articles with involvement of authors from Turkey, Nigeria, South Africa and Mexico, all averaging around 2013, are relatively the oldest.

*Collaboration of organisations in LMIC research*—Patterns of coauthorship in the LMIC sub-domain of road safety literature can also be analysed at a less aggregate scale and at the level of organisations of authors. Fig. 7 presents clusters of collaborations between organisations in this dataset of research articles. The cluster have been determined by the VOS method implemented in VOSviewer (Van Eck and Waltman, 2007, 2010). Overall, ten different clusters of collaborations were identified. Each cluster on the map is presented with a different colour. Nodes represents organisations and the size of each node is proportional to the number of studies on which the authors of the organisation have been listed. In a subset of these clusters, the lead organisation is in fact not the one based in an LMIC and is rather an organisation of an HIC. This, for example, is the case with cluster #2 whose leading organisation

Fig. 7. Network of collaborations between organisations in LMIC road safety papers.

is Johns Hopkins Bloomberg School of Public Health (United States) (n = 52), or cluster #6 whose leading organisation is Queensland University of Technology (Australia) (n = 34), or cluster #9 whose leading organisation is Monash University (Australia) (n = 39). Amongst clusters of collaborations that are largely populated by Chinese organisations, the followings have had the largest number of studies in this dataset: Southeast University (n = 72), Tongji University (n = 60), Tsinghua University (n = 57), Chang'an University (n = 43) and Wuhan University of Technology (n = 35). Other institutes that are based in LMICs and whose researchers have contributed substantially to this literature are University of Sao ˜ Paulo (Brazil, n = 33), Indian Institute of Technology (India, n = 27), Shahid Beheshti University of Medical Sciences (Iran, n = 28), Beijing University of Technology (China, n = 23), Iran University of Science and Technology (Iran, n = 20) and Seoul National University (South Korea, n = 20).

*Collaboration of authors in LMIC research*—Collaboration patterns can also be investigated in further disaggregate levels by treating authors as the entities of the analyses using the VOS method (Van Eck and Walt man, 2007, 2010). Fig. 8 provides a visualisation of the network of author collaborations in the LMIC literature. The map includes major authors of this reference dataset, those that have at least four items to their name in the literature of LMIC road safety. Nodes in this network represent authors and their size is proportional to the number of documents from their respective author. Each link represents the existence of co-authorships between two authors in the LMIC literature. The number of links originating from the node of an author represents the number of co-authors that they have had on this network and on LMIC topics. Each link has a strength that is representative of the number of joint articles between the two authors. The accumulation of these link strengths across all the links that originate from a node is referred to as total link strength.

A total of 18 distinct clusters of co-authorships were identified using

this analysis. The spatial proximity of node makes it impossible to visualise a label for all nodes (authors) of the network, and therefore, the VOS algorithm inevitably skips some labels. To unpack the full content of this network, however, Appendix B tabulates the list of authors of each cluster, along with their respective institutes and countries, the number of links, total link strength and number of documents in the LMIC literature. Not all authors of this network are affiliated with an LMIC. The table also specified authors based in LMICs using an * sign. The size of each cluster (i.e., the total number of its authors) N, has also been contrasted with the size of LMIC authors of the cluster, n. Out of the total of 189 authors of this network, 115 authors (nearly 60% of all authors) are those affiliated with institutes of LMICs and the rest are affiliated with organisations based in HICs. The proportion of LMIC authors varies substantially across different clusters. Some clusters such as cluster #1 are predominantly made up of HIC authors (more than 60%), whereas, some clusters predominantly or even exclusively include LMIC authors, such as cluster #11 (whose all authors are from China). Within the clusters with predominantly LMIC authors, almost invariably, a dominant country or region is identifiable. Clusters #2, #4, #10, #11, #16 and #18 are mostly made up of authors affiliated with Chinese institutes. Cluster #3 has been created because of the creation of a close network of collaboration between a number of authors from Turkey, Greece, Kosovo and Estonia. Cluster #5 represents collaborations between a cohort of authors from Iran and Serbia. In cluster #6, the presence of authors from Malaysia is notable. Cluster #12 is associated with activities originated mostly from countries in the continent of America, particularly Mexico and Brazil. All LMIC authors of cluster #15 are also from Brazil. Cluster #13 presents a mixture of authors from Iran, India and South Korea. The author with largest number of publications in this dataset is Hyder, Adnan A. (of The George Washington University).

Fig. 8. Network of collaborations between authors on road safety papers in the contexts of LMICs.

4.2. Divisions of road safety research in the context of LMICs

*Divisions of the LMIC literature based on bibliographic coupling of articles*—This section aims to identify various major divisions of scholarly road safety research in the LMIC contexts and contrast the landscape of that research with that of the road safety literature in general. In doing so, we resort to two main indicators: the similarity of references at the level of individual articles (also known as bibliographic coupling (Kessler, 1963; Martyn, 1964; Weinberg, 1974)) as well as patterns of term co-occurrence in the titles and abstracts of individual articles (Bornmann et al., 2018; Haghani, 2021; Haghani and Bliemer, 2021; Haghani et al., 2021b; Haghani et al., 2021c; Sedighi, 2016; Van Eck and Waltman, 2007). By applying the method of Visualisation of Similarities (VOS) (Van Eck and Waltman, 2007) to the reference lists of articles in the general literature of this field, six distinct clusters (including major clusters and one small cluster) of bibliographically coupled articles became identifiable (Fig. 9, top part). In this visualisation, the size of node visualisation is proportional to the number of citations received by individual articles, and spatial proximity of articles (nodes) is indicative of their degree of similarities and colours define clusters. Groups of articles whose reference lists are similar to one another can form a cluster. When the method is applied to the sub-dataset of LMIC papers, however,

Fig. 9. Clusters of road safety articles based on the similarity of their reference lists (i.e., bibliographic coupling) in the general literature (top map) and the subset of LMIC studies (bottom map).

only five clusters (including four major clusters and one small cluster) were identified (Fig. 9, bottom part). Among these, clusters #1 (red) and cluster #4 (pink) that respectively represent observational studies of road traffic accident in LMICs and cultural and psychological aspects of drivers in LMICs are on average the oldest clusters. Cluster #2 (green), representing applications of statistical models in crash severity analysis in LMICs is mostly made up of younger articles. This analysis allows us to identify influential LMIC clusters within each cluster. These studies (determined based on the total number of citations to the articles) have

been listed in Appendix C.

*Divisions of the LMIC and general road safety literature based on cooccurrence of terms in titles and abstracts of articles*—A more tangible way for contrasting the structure/divisions of the road safety literature in general as well as that of the LMICs is through the analysis of simultaneous occurrences of key terms in their titles and abstracts. This way, key terms that often occur in the same articles can form a cluster and that cluster could represent a division of the field. Inspection of the semantic content of the cluster can give tangible insights into the nature

Fig. 10. Clusters of co-occurred terms in the general literature (top) and the subset of LMIC studies (bottom).

of the research division that it represents. Fig. 10 provides outcomes of this analysis for the general literature (top part) as well as the LMIC subdomain (bottom part). In this visualisation, each node represents a single key term or term combination. The size of visualisation is proportional to the overall occurrences of the term in the titles and abstracts of articles of its respective dataset.

*The general road safety literature*—When the method is applied to the dataset that represents the general road safety literature, six distinct clusters become identifiable. Inspection of the content of these clusters shows that these are representative of research in the areas of (#1) road safety legislation and policy and observational statistics of road trauma and road traffic death/injury prevalence (yellow), (#2) technological methods of advancing vehicle safety (pink), (#3) applications of statistical modelling methods in crash frequency and severity analysis (orange), (#4) experimental studies of driver behaviour using simulators (green), (#5) driver psychology, perception and attitude (grey) and (#6) biomechanics aspects of crashes, crash reconstruction and vehicle crash worthiness (blue). Among these, cluster #1 is generally the oldest cluster (in that the articles where the terms that represent this cluster have appeared in are on average older), while clusters of driving safety technology (#2) and statistical modelling (#3) are on average younger. See Appendix D where hybrid maps with overly of average age of the terms have been provided.

*The subset of LMIC road safety literature*—The same method, when applied to the subset of LMIC articles, only produces three major clusters. For each of these clusters, an approximate counterpart can be identified in the general literature. Cluster #1 (orange) is nearly similar to cluster #3 of the general literature. Cluster #2 demonstrates content that is closely comparable to that of cluster #1 of the general literature and cluster #3 is similar, in nature, to cluster #5 of the general literature. This cluster, representing questionnaire-based studies of driver behaviour, represents a major of body of road safety studies undertaken in the context of LMICs. In fact, the term "questionnaire" per se is the

single most frequent term used in the title and abstract of LMIC articles (n = 292). In contrast to the general literature, divisions of driving safety technology, vehicle crashworthiness and simulator-based studies do not show a strong presence in the LMIC sub-domain. Similar to the general literature, however, the division of road trauma prevalence/statistics is relatively the oldest division in the LMIC subdomain too (see Appendix D).

4.3. Scoping review and temporal analysis of LMIC literature

*The document co-citation methodology*—This section provides a more disaggregate view (compared to the previous analysis based on term cooccurrence) into the content of road safety research in LMIC contexts by providing a scoping review of the field assisted by a document cocitation analysis (Chen, 2004, 2006). The analysis also embodies a temporal component through which we document the progression of LMIC studies over the last thirty years. The document co-citation analysis is based on identifying patterns of referencing. References that are jointly cited by third articles are said to be co-cited. And those that are frequently co-cited could represent a stream of research studies that are thematically related to one another; hence they can form a cluster. By analysing the content of the co-cited references of each cluster as well as the citing references that have created those clusters, one can identify major streams of research activities in this field, here, LMIC road safety sector. The document co-citation methodology of Chen (2004) provides a network with different time slices (i.e., with duration of a year) which can allow us to analyse temporal progression of road safety knowledge in LMIC contexts.

*Document co-citation maps in LMIC literature*—The outcome of this analysis as applied to the dataset of LMIC studies has been provided in Figs. 11–13. The nodes in each of these figures represent cited (or more precisely co-cited) references. These are studies that have frequently been references by LMIC papers. They may or may not themselves be

Fig. 11. The network document co-citation analysis in the literature of road safety studies in the context of LMICs.

Fig. 12. State of road safety research in the context of LMICs over the last six years. Readers can refer to the Online Supplementary Material of this paper to access a dynamic year-by-year visualisation of this map.

LMIC articles. Groups of these references that are frequency co-cited form a cluster of co-cited references. These cluster receive a label. This label is determined automatically through the algorithm of Chen (2006) using a log-likelihood ratio method. The method inspects the titles of citing articles (which are all LMIC studies) and extracts common noun phrases from those titles. Each citing article of a cluster has a different coverage of the references of the cluster, which is equal to the number of the references of that cluster that have appeared in the reference list of that article. Noun phrases extracted from articles with larger coverage receive higher log-likelihood score. And at the end, the noun or noun phrase with the largest score is selected as the label of the cluster. However, it should be noted that this label, while objectively determined, is a very abstract representation of the content and nature of the cluster. As a result, it is essential that a range of terms with high score be considered for each cluster instead of the label only. Here, a total of eight major clusters of co-cited documents were identified. These clusters are detailed in Table 1 through their ID, size (S), Mean Year (MY) of their cited references, and their Silhouette Score (SS) (a measure of the homogeneity of the cluster), a list of top terms extracted from the title of the citing references; and a list of citing references with the highest coverage of the cluster. The amount of coverage of the cluster, the main focus of the study as well as the country of focus of each of these citing articles have also been specified in brackets opposite to each reference.

*Temporal dynamics in LMIC literature*—The network presented in Fig. 11 has been colour-coded based on the average year of the references of clusters with brighter colours representing younger clusters. In the Online Supplementary Material of this paper, the year-by-year development of this network has also been visualised in the form of a video. Each time slice of that visualisation shows parts of the network that have been most active during that year by visualising the corresponding links in a visually salient way. Fig. 12 also provides a static representation of such year-by-year analysis by showing the state of this network over the last six years. A visual inspection of this set of outputs makes it clear that the LMIC literature can be decomposed to three different temporal segments: a classic segment (that seems to be no longer active), an intermediate segment and a contemporary segment (that specifies trending topics of LMIC studies). Fig. 13 forgoes the feature of spatial proximity of clusters in favour of visualising the cited references of each cluster against a timeline. Larger nodes in this visualisation represent references that have received more local citations from LMIC papers.

Cluster #0, road traffic injury, is the largest cluster of LMIC literature and can be attributed to the intermediate segment of this field, meaning that the cluster is not highly active at the current time, but its activities have not gone extinct either. The cluster predominantly represents observational and descriptive studies of determinants of road

Fig. 13. A timeline view of the document co-citation network for LMIC road safety studies.

trauma in LMICs (Abegaz and Gebremedhin, 2019; Adeloye et al., 2016; Ahmed et al., 2016; Batool et al., 2012; Fleiter and Watson, 2016; Grimm and Treibich, 2013; Sabzevari et al., 2016; Tumwesigye et al., 2016; Vissoci et al., 2017; Wang et al., 2008) (mostly excluding studies that use statistical/econometric modelling methods), crash, injury and fatality prevalence in LMICs (Asefa et al., 2015; Caixeta et al., 2009; Garg and Hyder, 2006; Gomez-Salazar ´ et al., 2017; Nguyen et al., 2018), estimations of road traffic cost in LMICs (Anti´c et al., 2011), prevalence of traffic rule violations in LMICs (Ferdosian et al., 2015; Gomez-Salazar ´ et al., 2017; Tavafian et al., 2011b; Vecino-Ortiz et al., 2014; Wainiqolo et al., 2016; Zamani-Alavijeh et al., 2011), risk quantification (Kogani et al., 2020) and identification of risk factors (Ackaah et al., 2020; Fararouei et al., 2017; Kogani et al., 2020; Saadat and Karbakhsh, 2010), socio-demographic aspects of road traffic trauma and cohort studies (Behzadnia and Shahmohammadi, 2016; Chen et al., 2013; Jaung et al., 2009; Koekemoer et al., 2017; Kovess-Masfety et al., 2017; Kulkarni et al., 2013; Lin et al., 2013; Mazaheri et al., 2016; Nabipour et al., 2015), epidemiology studies of road traffic crashes (Adejugbagbe et al., 2015; Bakhtiyari et al., 2014), regional correlates (Rahman et al., 2016; Rockett et al., 2017) and road traffic interventions and traffic legislation in LMICs (Bishai et al., 2008; Damsere-Derry et al., 2019; De Andrade et al., 2008; Jordaan et al., 2005; Klair and Arfan, 2014; Wesson et al., 2016). This cluster could be attributed to the division of LMIC literature that was characterised by the yellow cluster in Fig. 10 based on patterns of term co-occurrence.

Cluster #0 is thematically intertwined with cluster #3, motorcycle crashes, as indicated by their spatial proximity on the map. They both belong to the major division of LMIC studies that was identified through key term analysis and releveled as the yellow cluster on the respective map on Fig. 10. They also both belong to the intermediate temporal segment of the field, although cluster #3 seems to be rather younger. This cluster embodies key issues in the contexts of LMIC populations

such as risky motorcycle riding behaviours (Nguyen-Phuoc et al., 2020c; Nguyen-Phuoc et al., 2020e; Rusli et al., 2020), mobile phone use among motorcycle drivers (Nguyen-Phuoc et al., 2020d; Truong and Nguyen, 2019; Truong et al., 2018; Widyanti et al., 2020) and four-wheel vehicle drivers (Bastos et al., 2020; Wang et al., 2020b) and among different cohorts of drivers such as students or taxi drivers (Truong and Nguyen, 2019; Truong et al., 2019), motorcycle helmet use for adults (Akaateba et al., 2014; Bachani et al., 2013; Bachani et al., 2017a; Kanitpong et al., 2008; Kauky et al., 2015; Kumphong et al., 2018b; Li et al., 2020c; Lunnen et al., 2015; Nimako Aidoo et al., 2018; Satiennam et al., 2020; Setty et al., 2020; Siebert and Lin, 2020; Tarigan and Sukor, 2018; Tosi et al., 2016) and child passengers (Ederer et al., 2016; Kulanthayan et al., 2020; Merali and Bachani, 2018). This also includes studies beyond the prevalence of helmet use and its effectiveness in death/ injury prevention, those investigating effectiveness of helmet law enforcement in LMICs (Kumphong et al., 2018a; Supramaniam et al., 1984) or helmet promotion campaigns in LMICs (German et al., 2019). Another topic embodied by cluster #3 is the issue of drunk driving in LMIC populations (Bhalla et al., 2013; Fouch´e et al., 2018; Zhang et al., 2014a). This includes time series analyses and evaluation of effectiveness of legislation in this area in LMICs, (Li et al., 2017; Volpe et al., 2017) particularly in Brazil (Jomar et al., 2019). In fact, the network of document co-citation presents a cluster of co-cited references that is collective outcome of a large cohort of road safety studies predominantly in the context of Brazilian driver populations (Magalh˜ aes et al., 2011; Salvarani et al., 2009). This constitutes cluster #5, Sao Paulo, that is also very closely related to both cluster #0 and cluster #3. All three of them have been visualised in close proximity and all three belong to the intermediate temporal segment of this literature. This cluster has investigated topics such as the use of recreational and illicit drugs (Breitenbach et al., 2012; Takitane et al., 2013), and alcohol (Campos et al., 2013) including the effect of reducing legal limits

Table 1 Clusters of co-cited references created by road safety articles that reported research in the context of LMIC. The table only lists citing articles of each cluster with the

	highest amount of coverage of the references.
ID S MY SS	Top themes	Citing articles with largest coverage (coverage, topic, county of study)
ID = 0 S = 149	road traffic injury; road traffic crashes; health belief model; seat belt use; Yaounde Douala road	Dandona et al. (2006b) (18, traffic injuries, India); Sobngwi-Tambekou et al. (2010) (16, traffic crashes, Cameroon); Dandona et al. (2006a) (16, risky behaviour, India); Dandona

MY = 2002 SS = 0.776 section; public health concern; urban India

ID = 1 S = 122 MY = 2003 SS = 0.944

aberrant driving behaviour; driving anger; cross-cultural difference; accident involvement; aggressive driving; driving skill; personality trait; driving style

(2006) (15, policy making, India); Nguyen-Phuoc et al. (2020b) (13, turn signal use, Vietnam); Li et al. (2008b) (12, helmet use, China); Bhatti et al. (2010) (11, traffic crashes, Cameroon); Zhang et al. (2014b) (11, fault and severity analysis, China); Kayani et al. (2014) (11, road crashes, Pakistan); Das et al. (2012) (10, road crashes and alcohol, India); Razzak et al. (2011) (9, costs of traffic injuries, Pakistan); Herman et al. (2012b) (9, traffic injuries, Fiji); Ali et al. (2011) (9, seatbelt, Iran); Juillard et al. (2011) (9, traffic injuries, Cameroon); Mohammadi (2009b) (9, road fatalities, Iran); Zhang et al. (2011) (9, traffic injuries, China); Mir et al. (2013) (9, commercial vehicles, Pakistan); Bachani et al. (2012a) (9, traffic injuries, Kenya); Zamani-Alavijeh et al. (2010) (9, risk taking, Iran); Huang et al. (2011) (8, seatbelt, China); Bakhtiyari et al. (2015) (8, traffic crashes, Iran); Amoh-Gyimah et al. (2017) (8, crash severity, Ghana); McGreevy et al. (2014) (8, traffic injuries, Cameroon); Ram and Chand (2016) (8, risk perception, India); Andreuccetti et al. (2011) (8, BAC limit, Brazil); Gururaj (2008) (8, traffic deaths and injuries, India); Barffour et al. (2012) (8, public data, India); Tavafian et al. (2011a) (8, seatbelt, Iran); Dastoorpoor et al. (2016) (7, traffic mortality, Iran); Xiong et al. (2016) (7, injury severity, China); Routley et al. (2011) (7, seatbelt, China (compared to Australia)); Lund and Rundmo (2009) (7, risk perception, Ghana (compared to Norway)); Sadeghi-Bazargani et al. (2016) (7, traffic crashes, Iran); Perez-Nunez et al. (2014) (7, traffic injuries, Mexico); Mohammadi (2009a) (7, traffic crashes, Iran); Dandona et al. (2008) (7, data underreporting, India); Majdzadeh et al. (2008) (6, traffic injuries, Iran); Lin et al. (2013) (6, traffic injuries, China); Dandona et al. (2011) (6, traffic injuries, India); Bhatti and Ahmed (2014) (6, policy making, Pakistan); Eze et al. (2013) (6, traffic mortality, Nigeria); Oluwadiya et al. (2009) (6, traffic crashes, Nigeria); Wiebe et al. (2016) (6, traffic fatalities, Zambia); Nordfjaern and Rundmo (2009) (6, traffic risk perception, Ghana (compared to Norway)); Teye-Kwadjo (2017) (6, pedestrian injuries, Ghana), Bazargan-Hejazi et al. (2016) (6, traffic injuries, Iran); Mashreky et al. (2010) (6, traffic injuries, Bangladesh); Herman et al. (2012a) (6, traffic injuries, Papua New Guinea); Mabunda et al. (2008) (6, pedestrian fatalities, South Africa); Lateef (2011) (6, traffic injuries, Pakistan); Azetsop (2010) (6, traffic injuries, Kenya); Tetali et al. (2013) (6, helmet use, Pakistan); Patel et al. (2016) (6, traffic injuries, Rwanda); Aghamolaei et al. (2011) (6, helmet use, Iran); Wang and Chan (2016) (6, traffic fatalities, China) Shen et al. (2018a) (22, aggressive driving behaviour, China); Shen et al. (2018b) (21, positive driving behaviour, China); Mehdizadeh et al. (2018) (21, aberrant driving behaviour, Iran); Nordfjaern et al. (2014a) (17, road traffic culture, Russia, India, Ghana, Tanzania, Uganda, Turkey, Iran); Stanojevic et al. (2018) (17, driving behaviour questionnaire, Bulgaria, Romania, Serbia); Han and Zhao (2020) (17, bus driver behaviour, China); Nordfjaern and Simsekoglu (2014) (16, aberrant driving behaviour, Turkey); Useche et al. (2019) (15, driving anger, Colombia); Peng et al. (2019) (15, aberrant driving behaviour, China); Mohamadi Hezaveh et al. (2018) (15, driving behaviour questionnaire, Iran); Qu et al. (2014) (14, Dula dangerous driving index; China); Hezaveh et al. (2018) (14, bicycle rider behaviour questionnaire, Iran); Shao et al. (2020) (13, aberrant driving behaviour, China); Zhang et al. (2018b) (13, traffic climate scale, China); Zhang et al. (2019d) (13, driving anger & aberrant behaviour; China); Nordfjaern et al. (2014b) (12, road traffic culture, Turkey); Xu et al. (2018b) (12, driving skill & driving behaviour, China); Ersan et al. (2020) (12, aggressive and positive driving behaviour, Estonia, Greece, Kosovo, Russia, Turkey); Zheng et al. (2019b) (11, cyclist personality, China); Maslac et al. (2018) (11, driving behaviour questionnaire, Serbia); Hussain et al. (2020) (11, aberrant driving behaviour, Pakistan); Ozkan et al. (2006a) (11, driving behaviour questionnaire, Iran, Turkey); Olandoski et al. (2019) (11, driving anger expression inventory, Brazil); Ozkan et al. (2006b) (11, driving skill inventory, Iran, Turkey); Wang et al. (2019a) (11, cycling behaviour questionnaire, China); Yang et al. (2013) (11, risky driving behaviour, China); Wang et al. (2018e) (11, driving style inventory, China); Poo et al. (2013) (10, driving style inventory, Argentina); Shi et al. (2020) (10, ride-hailer aberrant behaviour, China); Hussain et al. (2019) (10, truck drivers risky driving, Pakistan); Mehdizadeh et al. (2019) (10, taxi and truck driver behaviour, Iran); Shirmohammadi et al. (2019) (10, driving behaviour and skill, Iran); Xu et al. (2014) (9, traffic violations, China); Ge et al. (2015) (9, driving anger expression inventory, China); Shams et al. (2020) (9, sleep quality and risky driving, Iran); Long and Ruosong (2019) (9, driving style inventory, China); Arafa et al. (2020) (9, driving behaviour, Egypt); Sarbescu et al. (2014) (9, aggressive driving, Serbia, Romania); Dinh et al. (2020) (9, pedestrian behaviour, Vietnam); McIlroy et al. (2020a) (9, pedestrian behaviour, Bangladesh, China, Kenya, Thailand, Vietnam); McIlroy et al.

(2020b) (9, pedestrian behaviour, Bangladesh, China, Kenya, Thailand, Vietnam); Chu et al. (2019) (9, traffic climate, China); Sahebi et al. (2019) (9, aberrant driving behaviour, Iran); Parishad et al. (2020) (9, driver behaviour questionnaire, Iran); Ng

(continued on next page)

et al. (2013) (9, seatbelt, Malaysia)

| Table 1 (continued ) | | |----------------------|--| |----------------------|--|

ID	Top themes	Citing articles with largest coverage (coverage, topic, county of study)
S MY SS
ID = 2 S = 109 MY = 2009 SS = 0.881	crash severity; driver injury severity; intersection-related crashes; examining; pedestrian crashes; injury severity analysis	Park and Ko (2020b) (23, pedestrian-vehicle crashes, South Korea); Mukherjee and Mitra (2020c) (22, pedestrian risk factors, India); Belloumi and Ouni (2019) (20, crash severity, Tunisia); Zeng et al. (2019) (19, spatial correlation, China); Mukherjee and Mitra (2020d) (18, pedestrian crashes, India); Adanu et al. (2020) (17, unobserved heterogeneity, Namibia); Park and Bae (2020b) (16, pedestrian-vehicle crashes, South Korea); Wahab and Jiang (2019a) (14, machine learning applications, Ghana); Behnood et al. (2016) (13, effect of unsafe behaviour on crash severity, Iran); Mukherjee and Mitra (2020e) (13, intersection safety analysis, India); Zeng et al. (2020a) (12, effect of weather on crash severity, China); Park and Bae (2020a) (12, pedestrian-vehicle crash frequency, South Korea); Rahimi et al. (2020) (12, truck crashes, Iran); Kamruzzaman et al. (2014) (11, ordered probit or injury severity, Bangladesh); Waseem et al. (2019a) (11, heterogeneity in means and variances for motorcycle crashes, Pakistan); Hu et al. (2020) (11, pedestrian crashes, China); Cantillo et al. (2020) (11, traffic crash severity, Colombia); Salum et al. (2019) (11, motorcycle crashes, Namibia); Chang et al. (2019b) (11, unobserved heterogeneity in motorcycle crash analysis, China); Park and Ko (2020a) (11, pedestrian-vehicle crashes, South Korea); Chen et al. (2020b) (10, latent class model application, China); Musa et al. (2020) (10, road condition and accident severity, Malaysia); Hosseinpour et al. (2014) (10, road condition and accident severity, Malaysia); Se et al. (2020) (9, single-vehicle crash severity, Thailand); Wahab and Jiang (2019b) (9, motorcycle crash severity, Ghana); Sheykhfard et al. (2020) (9, structural equation modelling for pedestrian accidents, Iran); Azadeh et al. (2016) (8, driver decision-making style and injury severity, Iran); Chang et al. (2019a) (8, human factors and crash severity, China); Besharati and Tavakoli Kashani (2018) (8, pedestrian crashes, Iran); Wang et al. (2019c) (8, truck crash severity, China); Xie et al. (2020) (8, mixed ordered probit application, China); Zhang and Hassan (2019b) (8, mixed ordered
ID = 3 S = 99 MY = 2008 SS = 0.899	motorcycle crashes; mobile phone use; motorcycle helmet use; planned behaviour; car driver; traffic violation; young novice drivers; drunk driving	probit application, Egypt) Rusli et al. (2020) (16; risky motorcyclist behaviour; Malaysia); Nguyen-Phuoc et al. (2020e) (12, risky motorcyclist behaviour, Vietnam); Akaateba et al. (2015) (11, helmet use, Ghana); Manan et al. (2017) (10, motorcyclist speeding, Malaysia); Nguyen-Phuoc et al. (2020b) (10, turn signal use, Vietnam); Li et al. (2008b) (9, helmet use, China); Bhalla and Mohan (2015) (9, children cyclists, India); Nguyen-Phuoc et al. (2020a) (9, risky behaviour of motorcycle taxis, Vietnam); Manan et al. (2020) (9, motorcyclist red light running, Malaysia); Waseela and Laosee (2015) (8, motorcyclist crashes, Maldives); Hassan et al. (2017a) (8, risky behaviour of powered two-wheeler riders, India); Hassan et al. (2017b) (7, risky behaviour of powered two-wheeler riders, India); Manan and Varhelyi (2015) (7, motorcyclist behaviour at access points, Malaysia); Wadhwaniya et al. (2017) (7, helmet use, India); Hung et al. (2008a) (7, helmet use, Vietnam); Li et al. (2020b) (7, risky behaviour of non-motorised vehicle users, China); Oviedo-Trespalacios and Scott-Parker (2017) (6, Young Novice Driver Scale, Colombia); Nguyen et al. (2020) (6, mobile phone use, Vietnam); Ngueutsa and Kouabenan (2017) (6, risk perception, Cameroon); Joewono and Susilo (2017) (6, young motorcyclists, Indonesia); Kulanthayan et al. (2020) (6, helmet use, Malaysia); Li et al. (2008a) (6, helmet use, China); Hung et al. (2008b) (6, helmet use, Vietnam); Sukor et al. (2017) (6, motorcycle facilities, Malaysia); Hernandez ´ et al. (2016) (6, helmet use, Colombia); Tulu et al. (2015) (5, pedestrian crashes, Ethiopia); Bachani et al. (2017b) (5, drunk driving, Cambodia); Zhang et al. (2014a) (5, drunk driving, China); Salum et al. (2019) (5, motorcycle crashes, Tanzania); Nguyen-Phuoc et al. (2020d) (5, mobile phone use, Vietnam); Bui et al. (2020) (5, motorcycle rider questionnaire, Vietnam); Tosi et al.
ID = 4 S = 83 MY = 2005 SS = 0.858	crash frequency; negative binomial; random effect; safety performance; non-urban section; rural intersection; spatial autocorrelation; spill-over effect	(2020) (5, young novice drivers scale, Argentina, Colombia, Mexico); Truong et al. (2016b) (5, motorcyclist mobile phone use, Vietnam) Mukherjee and Mitra (2020c) (17, pedestrian crashes using negative binomial model, India); Wen et al. (2019a) (15, spatial autocorrelation in crash frequency data, China); Hou et al. (2018b) (14, random effect negative binomial model, China); Wen et al. (2019b) (14, effect of weather condition on crash frequency, China); Weng et al. (2019) (13, spatial–temporal interaction effect, China); Zeng et al. (2020b) (13, underreporting and spatial correlation, China); Zeng et al. (2019) (10, spatial generalized ordered logit model; China); Tang et al. (2020) (10, correlated random parameters negative binomial, China); Oh et al. (2009) (9, safety performance of signalised rural intersections, South Korea); Ye et al. (2009) (9, safety of rural intersections, Georgia); Ar´evalo-Tamara ´ et al. (2020) (8, crash frequency, Colombia); Rusli et al. (2018a) (8, random parameters negative binomial model, Malaysia); Dereli and Erdogan (2017) (8, Poisson and negative binomial model, Turkey); Singh et al. (2016) (7, tree and random effect negative binomial models, India); Soltani and Askari (2017) (7, autocorrelation of crashes, Iran); Mitra and Bhowmick (2020) (7, intersection safety, India); Wu et al. (2020) (7, truck drivers violations, China); Wang and Feng (2019) (7, hotspot identification, China); Ye et al. (2018) (7, generalized event count model, South Korea); Ma et al. (2017) (6, random effect negative binomial model, China); Zeng et al. (2020a) (6, weather condition and crash severity, China); Lord et al. (2010) (6, crash data under-dispersion, South Korea); Mahmud et al. (2019) (6, crash probability modelling, Bangladesh);
ID = 5 S = 78 MY =	Sao Paulo; southern Brazil; drug; traffic accident; Brazilian motorcycle courier; illicit drug; detecting alcohol	Truong et al. (2016a) (6, negative binomial panel data models, Vietnam) da Silva et al. (2012) (9, road accidents, Brazil); Bacchieri and Barros (2011) (8, history of traffic accidents, Brazil); Das et al. (2012) (8, alcohol, drugs and traffic crashes, India); Marin-Leon et al. (2012) (7, trends in traffic accidents, Brazil); Yonamine et al. (2013)

(continued on next page)

Table 1 (continued )

ID S	Top themes	Citing articles with largest coverage (coverage, topic, county of study)
MY SS
2006 SS = 0.887		(7, alcohol and drug use in truck drivers, Brazil); De Boni et al. (2012) (6, drunk driving, Brazil); Bachani et al. (2012a) (6, traffic injuries, Kenya); De Boni et al. (2011) (6, alcohol and drug use, Brazil); Sant'Anna et al. (2013) (6, Motorcycle accidents, Brazil); Bachani et al. (2012b) (5, helmet use, Cambodia); Barffour et al. (2012) (5, road safety data availability, India); Mir et al. (2013) (5, commercial vehicles, Pakistan); Neto et al. (2012) (5, traffic accident mortality trends, Brazil); Tran et al. (2012) (5, drunk driving, Vietnam)
ID = 7 S = 49 MY = 2008 SS = 0.924	signalised intersection; crossing behaviour; red-light running behaviour; planned behaviour; traffic signal violation; e-bike users; electric bike users	Yang et al. (2015a) (10, pedestrian waiting times, China); Zhang et al. (2016a) (10, traffic signal violations, China); Zhou et al. (2016) (9, pedestrian crossing violations, China); Yang et al. (2015b) (8, cyclists and electric bike riders crossing behaviour, China); Zhang et al. (2016b) (8, pedestrian red-light running behaviour, China); Guo et al. (2018b) (8, cyclists red-light running behaviour, China); Yang et al. (2018) (7, e bike users, red-light running, China); Wang et al. (2018a) (7, risky behaviour of e-bike riders, China); Yang et al. (2016) (7, traffic wardens, China); Guo et al. (2017) (6, e-biker behaviour, China); Guo et al. (2018a) (6, e-biker license registration, China); Yan et al. (2016) (6, red-light running, China); Wang et al. (2018c) (5, child pedestrian behaviour, China); Onelcin and Alver (2015) (5, illegal pedestrian crossing, Turkey); Yuan et al. (2017) (5, injury severity of e-bike users, China); Wan et al. (2016) (5, metro passenger risky behaviour, China); Demiroz et al. (2015) (5, illegal road crossing, Turkey); Shen et al. (2020) (5, delivery riders red-light running, China); Wang et al. (2017) (5, e-bike related fatal crashes, China); Zhang et al. (2018d) (5, e-bike injuries, China)
ID = 8 S = 26 MY = 2012 SS = 0.879	fatal pedestrian crashes; pedestrian crossing behaviour	Mukherjee and Mitra (2020a) (13, risk factors for fatal pedestrian crashes, India); Mukherjee and Mitra (2020d) (13, risk factors for fata pedestrian crashes, India); Mukherjee and Mitra (2020c) (12, pedestrian risk factors, India); Mukherjee and Mitra (2020b) (11, pedestrian signal violating behaviour, India); Mukherjee and Mitra (2020e) (9, pedestrian safety at intersections, India); Zhang et al. (2019a) (4, pedestrian safety at crosswalks, China); Mukherjee and Mitra (2019) (4, pedestrian safety at signalised intersections, India); Sheykhfard and Haghighi (2020) (4, pedestrian crossing safety, Iran)

(Andreuccetti et al., 2011) and drinking law enforcement (Nunes and Nascimento, 2012) in Brazilian populations. This cluster is also associable to the major research division determined by the yellow cluster in Fig. 10.

Cluster #7, signalised intersection, is also rather closely related to clusters #0, 3 and 5 discussed above, although this cluster can be considered a crossover between the intermediate and contemporary segment of the LMIC road safety research. Same proposition can also be extended to cluster #8, fatal pedestrian crashes, that is even a younger cluster and can be considered a cluster derived from cluster #7 and a rather contemporary segment of this field. These two are also both attributable largely to the yellow division of Fig. 10. The topics of intersection safety and crossing behaviour and overall the safety of vulnerable road users constitute central and overarching themes of these two clusters (Kumar et al., 2019; Li et al., 2020b; Mitra and Bhowmick, 2020; Quistberg et al., 2014; Wang et al., 2018b; Yuan and Chen, 2017). These two clusters collectively embody topics of LMIC road safety research such as safety of electric bikes at intersections (Bai et al., 2015; Wang et al., 2018d), pedestrian crossing behaviour and safety at crosswalks and intersections (Al Bargi et al., 2017; Amoako et al., 2014; Anti´c et al., 2016; Hashemiparast et al., 2017; Murat et al., 2017; Pawar and Patil, 2015; Poo ´ et al., 2018; Quistberg et al., 2015; Zafri et al.; Zhang et al., 2017) particularly the issue of gap acceptance (Zafri et al., 2020b; Zhuang and Wu, 2014) and phone use (Syazwan et al., 2017; Zhang et al., 2019c), cyclists safety at crosswalks and intersections (Chen et al., 2018; Petzoldt et al., 2017), and cyclists red-light running behaviour (Bai and Sze, 2020; Fraboni et al., 2018; Kadali et al., 2015). Applications of theory of planned behaviour are prevalent in the studies of these two clusters (Zhou et al., 2009b).

Cluster #1, aberrant driving behaviour, is the only cluster of cocited documents that can be attributed predominantly to the classic social psychology division of the field, the division that is characterised by the grey in Fig. 10. The cluster is relatively young, has been visualised in the contemporary segment of the map, and according to the supplementary video as well as Fig. 11, it can be deemed as one of the

trending/hot topics of road safety research in LMIC contexts. The cluster has been well and truly active over the last six years of development of road safety knowledge in LMICs. Major topics embodied by citing articles that have created this cluster include aberrant driving behaviour (Luo and Shi, 2020; Mohamed and Lotfi, 2016; Scott-Parker and Oviedo-Trespalacios, 2017; Shi et al., 2010; Warner and Åberg, 2014; Zhang et al., 2015), driving anger (Ersan et al., 2019; Escan´es and Poo, ´ 2018; Fei et al., 2019; Hern´ andez-Hernandez ´ et al., 2019; Sullman et al., 2015; Zhang et al., 2018c), cross-cultural differences in driving skills (Ozkan et al., 2006b; Uzumcuoglu et al., 2020), driving style (Wang et al., 2018e), risky driving (Disassa and Kebu, 2019; Wang et al., 2014b), risky cycling (Useche et al., 2018; Wang et al., 2020a; Zheng et al., 2019a) and the effect of personality traits on driving (Nordfjærn et al., 2015) of LMIC driver populations. Various questionnaires have also been developed or existing ones have been validated in the contexts of LMIC populations by studies of this cluster. This includes the Driving Anger Scale (Escan´es and Poo, ´ 2018; Li et al., 2014; Stephens et al., 2016), The Driving Anger Expression Inventory (Ge et al., 2015; Olan doski et al., 2019), the Manchester Driver Behaviour Questionnaire (Ang et al., 2019; Mohamadi Hezaveh et al., 2018; Oluwadiya et al., 2020; Parishad et al., 2020; Stanojevic et al., 2018; Zhang et al., 2013b), the Behaviour of Young Novice Drivers Scale (BYNDS) (Oviedo-Trespalacios and Scott-Parker, 2017; Tosi et al., 2020), the Attitudes toward Traffic Safety Scale (Trogolo ´ et al., 2019), the (multi-dimensional) Driving Style Inventory (Holman and Havarneanu, 2015; Long and Ruosong, 2019; Poo et al., 2013; Wang et al., 2018e), Driving Skill Inventory (Xu et al., 2018b), the Motorcycle Rider Behaviour Questionnaire (Sakashita et al., 2014), Pedestrian Behaviour questionnaire (McIlroy et al., 2019), Bicycle Rider Behaviour questionnaire (Hezaveh et al., 2018), and Chinese Cycling Behaviour questionnaire (Wang et al., 2019a).

Two further clusters also constitute the contemporary segment of the LMIC literature. This includes cluster #4, crash frequency and an even younger and trendier cluster, cluster #2, crash severity. These two clusters are closely related to one another as apparent from their close proximity on the map of Fig. 11. In fact, they both are associable to the

division of statistical modelling of road traffic crashes, a division that was identified in previous sections through a term co-occurrence analysis and reflected in the orange cluster of the respective map in Fig. 10. Both of these streams of research can be considered as hot topics of road safety research in LMIC contexts as they have exhibited a boost of activity over the last six years (see Fig. 12). While the overarching theme of this stream lies within the methodology of its underlying studies; i.e., applications of statistical/econometric methods to road crash data, they collectively embody a variety of topics within that central theme. This includes topics such as injury severity in vehicle–pedestrian crashes (Agustiyani et al., 2020; Amoh-Gyimah et al., 2017; Chung, 2018; Chung et al., 2017; Kashani and Besharati, 2017; Mujalli et al., 2019; Ouni and Belloumi, 2018; Song et al., 2017; Verzosa and Miles, 2016; Zafri et al., 2020a; Zhang et al., 2014b), motorcycle crashes (Cunto and Ferreira, 2017), vehicle-object accidents (Li et al., 2019; Peng et al., 2018), rear-end crashes (Chen et al., 2015a; Zhang and Hassan, 2019c), head-on crashes (Hosseinpour et al., 2014), crashes involving buses (Nasri and Aghabayk, 2020; Park et al., 2019), crashes involving trucks (Chen et al., 2015b; Chen et al., 2020a; Rahimi et al., 2020; Wang and Prato, 2019; Xu et al., 2019), crashes in rural (Chen et al., 2020b; ChikkaKrishna et al., 2017; Glavi´c et al., 2016; Ma et al., 2015; Tulu et al., 2015; Wang et al., 2014a) versus urban roads (Miqdady and de Ona, 2020; Zhang et al., 2020), rental versus non-rental crashes (Tay and Choi, 2017), crashes in mountainous freeways/highways (Huang et al., 2018; Rusli et al., 2018b), crashes in tunnels (Hou et al., 2018a), alcohol-influenced crashes (Chen et al., 2016; Wu et al., 2016b), and daytime versus night-time crashes (Zhang and Hassan, 2019a).

Methods that have been used by these studies also cover a broad range of econometric models including multinomial logit (Celik and Oktay, 2014; Wu et al., 2016a; Zhang et al., 2019b), mixed logit (Adanu et al., 2020; Wu et al., 2014) (and mixed logit models with heterogeneity in means (Waseem et al., 2019a)), latent-class logit (Li et al., 2019), ordered logit (Asare and Mensah, 2020), ordered probit (Kardar and Davoodi, 2020; Zhang et al., 2018a), negative binomial regression (Kim et al., 2007; Mukherjee and Mitra, 2020c; Ture Kibar et al., 2019), Poison models (Lord et al., 2010), Bayesian models (Besharati et al., 2020; Guadamuz and Aguero-Valverde, 2019; Lee et al., 2010; Park et al., 2010; Zeng et al., 2017), spatial regression analysis (Rhee et al., 2016; Satria et al., 2020; Shariat-Mohaymany and Shahri, 2017; Wang et al., 2019b) as well as machine learning techniques (Effati et al., 2015), genetic algorithms (Hashmienejad and Hasheminejad, 2017), randomforest classifiers (Jung et al., 2016; Li et al., 2020a), decisions trees (Rusli et al., 2018b; Scott-Parker and Oviedo-Trespalacios, 2017; Taamneh, 2018) and survival theory (Xu et al., 2018a).

Articles of road safety in LMIC contexts that have contributed the most to the creation of these clusters (i.e., those that have had largest coverage of each cluster) have been listed in Table 1 in the descending order based on the amount of their coverage.

5. Discussions

In this section, we discuss the findings of the analyses presented earlier as well as their implications and their connection to the six research questions (questions (i)-(vi)) that were previously outlined in the introduction of this work. The current study reports the first macroscale analyses of road safety research in LMICs. The landscape of road safety research shows that research activity in the African continent has been limited (question (iv)). This is a sparkling contrast when considering that Africa is the region with the poorest road safety performance in the world in terms of rates of road traffic death per 100,000 population (World Health Organization, 2018). Research activity can be considered an indicator of capacity building, analysis and evaluation of approaches to improve safety, so the fact that this is lacking is a concern. Governments and the international community need to invest in road safety research to support the development of initiatives tailored to meeting their safety needs and upgrade their practices. Copying models developed in western HICs may not provide the best understanding of road safety determinants (Oviedo-Trespalacios et al., 2021).

Only in the last 15 years, we have seen an increase in road safety research in LMICs (questions (i) and (iii)). However, it is important to notice that the growth in terms of the number of publications is not as accelerated as in the rest of the world. This is surprising because in many LMICs countries there are incentives to publish research in ranked international journals. A potential explanation for this finding is that there is lag in resources to conduct high-quality research that could contribute to the development of evidence-based road safety policy and initiatives.

The number of authors working in LMICs countries' research is considerably smaller than the number of researchers working in high income countries (question (v)). This means that jurisdictions with better road safety performance (high income), have generally a higher rate of road safety researcher developing and publishing research than jurisdiction with poorer road safety performance (LMICs). This can be explained in part for the limited capability to conduct research that can be observed in LMICs, as they generally have a lower number of professionals with a Ph.D. than high-income countries. Nevertheless, there are groups of leading researchers working in LMICs, some of them not necessarily based in LMICs. This is a consequence of the globalisation of research and science and we still do not understand how the COVID-19 shift towards more digital interactions will influence this even further. However, it is important to note that there are not many connections among researchers, which suggests that research in LMICs context occurs largely in isolation (question (vi)). This lack of connectivity may be a consequence of the lack of large-scale funding to support high quality research initiatives in LMICs.

Regarding country collaboration, there are some marked patterns of collaboration across countries, with Australia being potentially the most connected country in road safety research in LMICs. Some of these collaborations have a historical/colonialist basis such as Spain working with LMICs such as Peru and Argentina or England working with LMICs such as Kenya and Bangladesh. There is a need to globalise road safety research more to guarantee cross-pollination of road safety approaches. Additionally, it is important to explore collaborations across LMICs with similarity in terms of region and culture given that knowledge across these jurisdictions is more likely to be relevant (question (vi)).

The title analysis reveals that research about China is a key theme in the published research of LMICs. The large amount of activity in China shows the sophistication of the technology and investment of the Chinese government in road safety issues. A country that is emerging in the title analysis is India which has also experienced terrific development in the research and transport space. Although it is true that these two countries have an important portion of the worldwide population, which potentially justify this large number of publications, it is important to consider that there are nearly eighty LMICs which also need evidence to develop better road safety approaches (question (iv)).

The title and abstract analysis shed light on the main topics discussed in LMICs road safety research. There are three distinctive clusters in the dataset which appear to be driven mostly by the research approach: questionnaire-based research (grey), road traffic injury analysis (orange), and descriptive crash data analysis (yellow). This group highlights most of the research conducted in LMICs that there is a plethora of methodologies used in LMICs road safety research. However, it also shows that more sophisticated approaches are not widely consolidated (question (ii)). For example, there are not naturalistic studies in LMICs with counted small-scale exceptions in countries such as Brazil (Bastos et al., 2021). The grey cluster formed around self-reported research also shows that the most common themes for this research involve the application of behavioural scales such as driver behaviour questionnaire (DBQ) and the study of psychometric properties of questionnaires.

However, this appears to be very driver-centric rather than comprehensive to all road users. The orange cluster formed around traffic injury analysis highlights the on-going focus of research on under-theinfluence driving across countries such as China and Colombia. There also seems to be a significant amount of work on issues such as helmet and seatbelt wearing. Helmet wearing motorcycle riders is an on-going focus of concern in many LMICs. Finally, the yellow cluster which shows research crash data analysis appear to be oriented to understand crash severity and determinants. Speed management and crossing behaviour of pedestrians appear to be the focus of attention. From looking at this cluster, it is evident that LMICs road safety research has been drivercentric with only growing interest in the recent five years on active travel such as pedestrians and electric bike riders. Overall, there is a need to advocate for more research resources in active travel and sustainable forms of transport safety in LMICs. The current extent of work is insufficient and largely limited compared to driver behaviour research, which raises questions about the availability of evidence to advocate for more safety measures to protect vulnerable road users.

Two main sources of data for safety analysis in LMICs are questionnaire-based data and reported crash data (i.e., data collected after occurrence of a crash). Although these two sources of data can provide useful information towards improving the safety of road users, they have their own shortcomings. Regarding the questionnaire-based data, although it provides a quick and inexpensive method for collecting the information, it could have issues such as selectivity-biased problem, dishonest answers, unconscientious responses, accessibility issues, questionnaire fatigue, hidden agenda of the respondents, and differences in interpretation and understanding. In terms of reported crash data, trained personnel and systematic program are required to collect such data. Reported crash data provides more reliable source of data for safety analysis compared to questionnaire-based data. However, the availability and quality of reported crash data is one the concerns of safety researchers in LMICs (Wegman, 2017). It is also worth noting that the emerging sources of data, which have been recently used to obtain more realistic crash data compared to traditional sources, are not readily available in these countries (question (ii)). These emerging sources of data include, but not limited to, naturalistic driving data, Crash Outcome Data Evaluation System (CODES), and Event Data Records (EDRs).

Overall, the clusters of "motorcycle crashes", "fatal pedestrian crashes", and "signalized intersection" show the importance of the safety of vulnerable roadway users in LMICs (question (ii)). Pedestrians and motorcyclists are the vulnerable roadway users since they have less protection compared to other users (Lin and Kraus, 2009; Rod et al., 2021) which makes their safety of utmost importance for transportation agencies and safety researchers. The safety of these users become more important at locations where they share facilities and travel routes with motorized users (e.g., signalized and non-signalized intersections).

6. Conclusions, applications and future research directions

6.1. Summary of findings

It was estimated that slightly less than 10% of the entire body of road accident publications are linked to LMICs contexts. An important conclusion when comparing the clusters of general and LMICs road safety research is that the sophistication of LMICs road safety research is limited. There are a limited number of themes and approaches that have been taken. For example, issues such as distraction and drowsiness are virtually non-existent in LMICs and advanced statistical techniques such as Bayesian approaches or heterogeneity models have only been implemented in a very limited number of applications. Finally, if possessed with the question, what is missing in LMICs? It appears that, unlike HICs, active safety technologies have not been the focus of attention in LMICs, which highlights the adoption of these technologies is going to be largely delayed or even inadequate since research provides the groundwork to develop adequate road safety policy. Whilst issues such as advanced driver support systems, intelligent transport systems and autonomous vehicles are common in global road safety research, these are virtually non-existent in the LMICs. We also noted a lack of qualitative studies in LMICs. A potential explanation for this is the difficulty that authors in LMICs have publishing this type of research in international journals that are generally edited by editors based in HICs. We expect that this trend changes as qualitative research is important and of the interests of the international readership of the journals. Indeed, we argue that with the growth of capacity in LMICs work made in these jurisdictions are going to be cited with more frequency.

6.2. Potential applications

In addressing the abovementioned gaps in local road safety knowledge in the context of LMICs, we believe that further international collaborations between LMICs as well as between authors of LMICs and those of major universities in HICs could be a key. An important step in developing such collaborations could potentially be better recognising 'who is doing what' in road safety research in LMICs. This could be particularly important in relation to the authors affiliated with institutes of LMICs who are active in this domain but may generally receive lesser recognition compared to researchers in HICs. It is hoped that the identification of influential studies, influential authors and patterns of collaborations, as reported in this work (Figs. 8 and 9 as well as Appendices B and C), can make a contribution towards this end.

6.3. Future research directions

Note that while our scoping review covers a list of nearly 500 LMICrelated articles in its reference list, the underlying data used for the analysis of LMIC research patterns contains nearly 2700 items This means that these studies clearly could not all be cited in a single article. However, guided by the current analysis, subsequent studies could focus on certain subsets of LMIC road safety research (e.g., the subdomain of statistical analysis of accident severity in LMICs, or the subdomain of questionnaire-based research in LMICs) and analyse those areas in more details. This can be done in the form of conventional narrative or scoping reviews since the size of reference dataset would be smaller. In obtaining such sub-samples of references and isolating them from the rest of the body of LMIC research, the dataset as well as the search strategy of this work (all provided as Online Supplementary Material), could be used as a foundation. The methodology of this work could also be potentially adopted for investigating other domains of transportation research in the context of LMICs.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgements

The authors wish to thank two anonymous reviewers of this work for their constructive remarks during the peer review process. While undertaking this study, Milad Haghani's was a recipient of an Australian Research Council Discovery Early Career Researcher Award [DE210100440] funding from the Australian Government. Dr. Oscar Oviedo-Trespalacios was also the recipient of an Australian Research Council Discovery Early Career Researcher Award [DE200101079] funded by the Australian Government.

Appendix A. Oldest LMIC papers

Title	Authors	Journal	LMIC
Analysis of road traffic accidents data in Zambia	Emenalo et al. (1977)	Accident Analysis &	Zambia
		Prevention
Statistical estimation of individual motor-vehicle driver accident liability based on an analysis of	Williford and Murdock	Accident Analysis &	Georgia
mean time intervals between accidents of Georgia drivers	(1978)	Prevention
The crash helmet legislation in Nigeria - a before-and-after study	Asogwa (1980)	Accident Analysis &	Nigeria
		Prevention
Multilevel road accident exposure sampling system for South-Africa	Fernie (1982)	Accident Analysis &	South Africa
		Prevention
An estimate of the fatal accident frequency rate from mountaineering fatalities in Peru	Ridden (1983)	Accident Analysis &	Peru
		Prevention
Fatal motorcycle accidents and helmet laws in peninsular Malaysia	Supramaniam et al.	Accident Analysis &	Malaysia
	(1984)	Prevention
Road traffic accidents in rich developing-countries - the case of Libya	Mekky (1984)	Accident Analysis &	Libya
		Prevention
Accidental death and disability in India - a stocktaking	Mohan (1984)	Accident Analysis &	India
		Prevention
2-wheeler injuries in Delhi, India - a study of crash victims hospitalized in a neurosurgery ward	Mishra et al. (1984)	Accident Analysis &	India
		Prevention
An analysis of road traffic fatalities in Delhi, India	Mohan and Bawa	Accident Analysis &	India
	(1985)	Prevention
Patio-temporal analysis of road traffic accidents in Oyo state, Nigeria	Jegede (1988)	Accident Analysis &	Nigeria
		Prevention
The effect of open-back vehicles on casualty rates - the case of Papua-New-Guinea	Nelson and Strueber	Accident Analysis &	Papua-New
	(1991)	Prevention	Guinea
The occurrence and driver characteristics associated with motor-vehicle injuries in Addis Ababa,	Dessie and Larson	Journal of Tropical Medicine	Ethiopia
Ethiopia	(1991)	and Hygiene
A perspective on road fatalities in Jeddah, Saudi Arabia	Bener and Jadaan	Accident Analysis &	Saudi Arabia
	(1992)	Prevention
Road traffic accidents in Nigeria - a review and a reappraisal	Asogwa (1992)	Accident Analysis &	Nigeria
		Prevention
Risk-taking behavior of cape peninsula high-school-students 0.6. Road-related behavior	Flisher et al. (1993)	South African Medical	South Africa
		Journal
Road accident trends in Nigeria	Oluwasanmi (1993)	Accident Analysis &	Nigeria

Application of Smeed formula to assess development of traffic safety in Jordan	Gharaybeh (1994)	Prevention Accident Analysis &	Jordan

		Prevention
Road traffic accidents in Kenya - an epidemiologic appraisal	Odero (1995)	East African Medical Journal	Kenya
China takes to the roads	Roberts (1995)	British Medical Journal	China
Some implications of driver training for road accidents in Gaborone	Oladiran and Pheko	Accident Analysis &	Gaborone
	(1995)	Prevention

Appendix B. Clusters of co-authorships. Number of links represent the number of co-authors on the network while link strengths represent the number of co-authored publications between pairs of authors of the network. Total link strength is a cumulative representation of the strengths of all links originating from the node of each author. Authors of each cluster are listed alphabetically. The * sign specifies authors affiliated with institutes based in an LMIC. N is the number of authors in the cluster and n is the size of the subset of authors of the cluster affiliated with organisations based in LMICs.

Author	Authors (number of links, total link strength, number of documents)
cluster

#1	Bachani, Abdulgafoor M. Johns Hopkins Bloomberg School of Public Health, United States (15, 31, 16); Ballesteros, Michael F., National Center for Injury
N = 19	Prevention and Control, United States (5, 19, 7); Bhalla, Kavi, University of Chicago, United States (9, 13, 6); Bishai, David, Johns Hopkins Bloomberg School of
n = 7	Public Health, United States (9, 14, 5); Brijs, Tom, Hasselt University, Belgium (2, 5, 4); Duan, Leilei*, Chinese Centre for Disease Control and Prevention, China (3, 3,
	4); Gupta, Shivam, Johns Hopkins Bloomberg School of Public Health, United States (7, 12, 7); Hyder, Adnan A., The George Washington University, United States
	(19, 65, 30); Kobusingye, Olive*, Makerere University, Uganda (1, 1, 5); Li, Qingfeng, Johns Hopkins Bloomberg School of Public Health, United States (7, 11, 6);
	Mohan, Dinesh*, Indian Institute of Technology, India (2, 5, 5); Parker, Erin M., Centers for Disease Control and Prevention, United States (5, 12, 6); Roehler,
	Douglas R., Centers for Disease Control and Prevention, United States (5, 15, 6); Sann, Socheata*, Handicap International, Cambodia (7, 12, 4); Stevens, Kent A.,
	Johns Hopkins Bloomberg School of Public Health, United States (2, 5, 4); Tiwari, Geetam, Indian Institute of Technology, India (2, 3, 8); Wang, Yuan, Chinese
	Center for Disease Control and Prevention, China (6, 7, 4); Wets, Geert, Hasselt University, Belgium (2, 5, 4); Ye, Zhirui*, Southeast University, China (2, 3, 5)
#2	Du, Wei, Australian National University, Australia (5, 8, 6); Fei, Gaoqiang*, Southeast University, China (3, 3, 6); Fleiter, Judy J., International F´ed´eration of Red
N = 16	Cross & Red Crescent Soci´et´es, Switzerland (3, 3, 6); King, Mark J., Queensland University of Technology, Australia (4, 5, 6); Ozanne-Smith, Joan, Monash
n = 9	University, Australia (6, 15, 6); Qin, Yu*, Renmin University of China, China (3, 11, 4); Routley, Virginia, Monash University, Australia (3, 11, 4); Senserrick,
	Teresa, Queensland University of Technology, Australia (2, 2, 4); Stallones, Lorann, Colorado State University, United Stated (6, 23, 8); Wu, Changxu*, Tsinghua
	University, China (2, 2, 7); Wu, Ming*, Jiangsu Provincial Center for Disease Control and Prevention, China (10, 20, 6); Xiang, Henry, The Ohio State University
	College of Medicine, United States (7, 19, 6); Yang, Jie, Chang'an University, China (8, 12, 4); Zhang, Guangnan, Sun Yat-Sen University, China (2, 2, 8); Zhang,
	Xujun, Southeast University, China (6, 23, 8); Zheng, Xiaoying*, Peking University, China (4, 7, 4)
#3	Azik, Derya*, Middle East Technical University, Turkey (17, 67, 4); Danelli-Mylona, Vassiliki, R.S.I. Road Safety Institute Panos Mylonas, Greece (17, 67, 4); Ersan,
N = 16	Ozlem*, Middle East Technical University, Turkey (17, 67, 4); Georgogianni, Dimitra, R.S.I. Road Safety Institute Panos Mylonas, Greece (17, 67, 4); Kacan,
n = 11	Bilgesu, Middle East Technical University, Turkey (17, 67, 4); Krasniqi, Ema Berisha, Kosovo Association of Motorization, Kosova (17, 67, 4); Krasniqi,
	Muhamed, Kosovo Association of Motorization, Kosova (17, 67, 4); Makris, Evangelos, R.S.I. Road Safety Institute Panos Mylonas, Greece (17, 67, 4); Oz, Bahar,
	Middle East Technical University, Turkey (17, 67, 5); Pashkevich, Anton, Tallin University of Technology, Estonia (17, 67, 4); Pashkevich, Maria, Tallin

(continued on next page)

(continued )

(17, 67, 4) #4 Abdel-Aty, Mohamed, University of Central Florida, United States (8, 14, 9); Hu, Lin, Changsha University of Science and Technology, China (1, 1, 4); Huang, N = 16 Helai, Central South University, China (12, 18, 13); Lee, Jaeyoung, Central South University, China (9, 11, 6); Li, Yibing, Tsinghua University, China (1, 3, 4); n = 12 Ma, Yongfeng, Southeast University, China (1, 1, 5); Peng, Yichuan, Tongji University, China (2, 6, 4); Sayed, Tarek, University of British Colombia, Canada (1, 1, 5); Wang, Jie, Central South University, China (4, 6, 4); Wang, Xuesong, Tongji University, China (6, 13, 23); Wen, Huiying, South China University of Technology, China (7, 16, 7); Yang, Xiaobao, Beijing Jiaotong University, China (2, 6, 5); Yu, Rongjie, Tongji University, China (2, 5, 4); Yuan, Quan, Tongji University, China (5, 10, 7); Zeng, Qiang, South China University of Technology, China (8, 18, 7); Zhang, Xuan, Sam's Club Technology, United States (5, 12, 4) #5 Antic, Boris, University of Belgrade, Serbia (2, 11, 8); Jovanovic, Dragan, Technical College of Applied Sciences Urosevac, Serbia (5, 9, 6); Khanjani, Narges, N = 15 Kerman University of Medical Sciences, Iran (3, 9, 9); Khorasani-Zavareh, Davoud, Shahid Beheshti University of Medical Sciences, Iran (1, 1, 4); Lipovac, Krsto, n = 11 Belgrade University, Russia (3, 10, 8); Moradi, Ali, Hamadan University of Medical Sciences, Iran (2, 3, 4); Nabipour, Amir Reza, Kerman University of Medical Sciences, Iran (2, 8, 5); Nazari, Seyed Saeed Hashemi, Shahid Beheshti University of Medical Sciences, Iran (1, 2, 5); Pesic, Dalibor, University of Belgrade, Serbia (3, 15, 12); Saadat, Soheil, University of California, United States (1, 1, 4); Sadeghi-Bazargani, Homayoun, Tabriz University of Medical Sciences, Iran (2, 2, 12); Soori, Hamid, Shahid Beheshti University of Medical Sciences, Iran (4, 6, 11); Stanojevic, Predrag, Technical College of Applied Sciences Urosevac, Serbia (3, 7, 4); Stephens, Amanda N., Monash University, Australia (1, 2, 4); Sullman, Mark J. M., Massey University, New Zealand (5, 10, 6) #6 Ambak, Kamarudin, Tun Hussein Onn University of Malaysia, Malaysia (4, 6, 5); Chen, Cong, University of South Florida, United States (5, 15, 8); Chen, Feng, N = 13 University of Texas at Dallas, United States (6, 9, 7); Ci, Yusheng, Harbin Institute of Technology, China (3, 8, 4); Daniel, Basil David, (4, 8, 7); Ishak, Siti n = 7 Zaharah, Tun Hussein Onn University of Malaysia, Malaysia (3, 3, 4); Prasetijo, Joewono, Tun Hussein Onn University of Malaysia, Malaysia (5, 14, 12); Sukor, Nur Sabahiah Abdul, Universiti Sains Malaysia, Malaysia (1, 1, 4); Sun, Jian, Tongji University, China (1, 1, 6); Wu, Qiong, University of Hawaii, United States (4, 14, 7); Zainal, Zaffan Farhana, Tun Hussein Onn University of Malaysia, Malaysia (2, 8, 6); Zhang, Cunbao, Wuhan University of Technology, China (1, 4, 5); Zhang, Guohui, University of Hawaii, United States (9, 27, 15) #7 Chen, Yikai, Hefei University of Technology, China (4, 10, 4); Damsere-Derry, James, Queensland University of Technology, Australia (4, 16, 9); Ebel, Beth E., N = 11 University of Washington, United States (3, 7, 5); Fleiter, Judy, International F´ed´eration of Red Cross & Red Crescent Soci´et´es, Switzerland (2, 5, 4); He, Jie, n = 4 Southeast University, China (6, 10, 7); King, Mark, Queensland University of Technology, Australia (9, 30, 19); Mock, Charles N., University of Washington, United States (2, 5, 4); Palk, Gavan, Queensland University of Technology, Australia (2, 12, 6); Quistberg, D. Alex, Drexel University, United States (1, 2, 4); Rusli, Rusdi, Universiti Teknologi, Malaysia (2, 4, 4); Shi, Qin, Hefei University of Technology, China (4, 10, 4) #8 Castro, Candida, University of Granada, Spain (2, 2, 4); Chan, Alan H. S., City University of Hong Kong, China (3, 6, 6); Crundall, David, Nottingham Trent N = 11 University, United Kingdom (3, 5, 5); Hu, Guoqing, Central South University, Changsha, China (8, 18, 9); Li, Li, Tsinghua University, China (2, 3, 4); Ning, Peishan, n = 4 Central South University, Changsha, China (4, 10, 4); Schwebel, David C., University of Alabama at Birmingham, United States (6, 15, 11); Sheppard, Elizabeth, University of Nottingham, United Kingdom (1, 3, 4); Yang, Jikuang, (1, 1, 4); Zhang, Tingru, Chalmers University of Technology, Sweden (2, 5, 4); Zhang, Wei, Tsinghua University, China (7, 8, 10) #9 Choi, Jaisung, University of Seoul, South Korea (1, 4, 5); Daniel Ledesma, Ruben, Universidad Nacional de Mar del Plata, Argentina (2, 2, 5); De Gruyter, Chris, N = 10 RMIT University, Australia (4, 16, 9); Diep Ngoc Su, The University of Danang, Vietnam (2, 7, 5); Nguyen, Hang T. T., University of Transport and n = 4 Communications, Vietnam (3, 14, 8); Oviedo-Trespalacios, Oscar, Queensland University of Technology, Australia (7, 15, 14); Scott-Parker, Bridie, University of the Sunshine Coast, Australia (1, 5, 5); Tay, Richard, RMIT University, Australia (3, 6, 10); Truong, Long T., Latrobe University, Australia (3, 14, 9); Useche, Sergio A., University of Valencia, Spain (2, 2, 4) #10 Bai, Lu, Southeast University, China (5, 7, 4); Guo, Yanyong, Southeast University, China (5, 10, 6); Liu, Pan, Southeast University, China (5, 16, 9); Lu, Jian, N = 10 Tongji University, China (3, 6, 8); Wang, Chen, Southeast University, China (5, 12, 12); Wang, Wei, Georgia Institute of Technology, United States (5, 11, 7); Wang, n = 9 Yonggang, Chang'an University, China (1, 1, 9); Wu, Yao, Southeast University, China (2, 6, 4); Xing, Yingying, Tongji University, China (2, 5, 4); Xu, Chengcheng, Southeast University, China (9, 22, 14) #11 Feng, Zhongxiang, Hefei University of Technology, China (5, 17, 10); Lyu, Nengchao, Wuhan University of Technology, China (2, 5, 5); Ma, Changxi, Lanzhou N = 9 Jiaotong University, China (5, 6, 4); Sze, N. N., The Hong Kong Polytechnic University, Hong Kong (7, 8, 5); Wan, Ping, East China Jiaotong University, China (1, 4, n = 9 4); Wang, Kun, Hefei University of Technology, China (4, 14, 7); Wu, Chaozhong, Wuhan University of Technology, China (6, 17, 16); Zhang, Hui, Wuhan University of Technology, China (3, 8, 4); Zhang, Weihua, Hefei University of Technology, China (4, 14, 6) #12 Chandran, Aruna, Manipal University, India (11, 40, 12); De Boni, Raquel, Federal University of Rio Grande do Sul, Brazil (2, 5, 4); Hidalgo-Solorzano, Elisa, N = 8 National Institute of Public Health, Mexico (5, 20, 7); Hijar, Martha, ST CONAPRA, Mexico (6, 33, 11); Lunnen, Jeffrey C., Johns Hopkins Bloomberg School of n = 5 Public Health, United States (8, 28, 7); Pechansky, Flavio, Federal University of Rio Grande do Sul, Brazil (5, 14, 12); Perez-Nunez, Ricardo, Instituto Nacional de Salud Pública, Mexico (6, 32, 11); Sousa, Tanara, The University of Melbourne, Australia (4, 8, 4) #13 Besharati, Mohammad Mehdi, Iran University of Science and Technology, Iran (2, 8, 7); Kashani, Ali Tavakoli, Iran University of Science and Technology, Iran N = 7 (2, 8, 8); Lee, Dongmin, University of Seoul, South Korea (2, 6, 6); Mitra, Sudeshna, Indian Institute of Technology, India (2, 9, 12); Mukherjee, Dipanjan, n = 6 Indian Institute of Technology, India (1, 8, 8); Oh, Jutaek, Korea National University of Transportation, South Korea (2, 6, 4); Washington, Simon, The University of Queensland, Australia (7, 13, 12) #14 Haghighi, Farshidreza, Babol Noshirvani University of Technology, Iran (2, 6, 5); Hezaveh, Amin Mohamadi, University Of Tennessee, United States (3, 7, 4); N = 7 Nordfjaern, Trond, Norwegian University of Science and Technology, Norway (6, 22, 15); Rundmo, Torbjorn, Norwegian University of Science and Technology, n = 2 Norway (3, 11, 6); Shariat-Mohaymany, Afshin, Iran University of Science and Technology, Iran (1, 2, 4); Sheykhfard, Abbas, Delft University of Technology, The Netherlands (2, 6, 5); Simsekoglu, Ozlem, Nord University Business School, Norway (4, 16, 11) #15 Andreuccetti, Gabriel, University of Sao Paulo Medical School, Brazil (6, 25, 10); De Carvalho, Heraclito Barbosa, University of Sao Paulo, Brazil (5, 18, 6); N = 6 Gjerde, Hallvard, Oslo University Hospital, Norway (5, 13, 7); Leyton, Vilma, University of Sao Paulo Medical School, Brazil (6, 26, 11); Munoz, Daniel Romero, n = 5 University of Sao Paulo Medical School, Brazil (5, 14, 4); Yonamine, Mauricio, University of Sao Paulo Medical School, Brazil (5, 14, 4) #16 Ge, Yan, University of Chinese Academy of Sciences, China (5, 43, 14); Qu, Weina, University of Chinese Academy of Sciences, China (5, 46, 16); Sun, N = 6 Xianghong, University of Chinese Academy of Sciences, China (4, 29, 10); Zhang, Kan, University of Chinese Academy of Sciences, China (6, 45, 15); Zhang, n = 6 Qian, Chinese Academy of Sciences, China (5, 16, 8); Zhao, Wenguo, Chinese Academy of Sciences, China (4, 14, 4) #17 Lajunen, Timo, Traffic Research Centre of Finland Ltd., Finland (24, 92, 18); Ozkan, Turker*, Middle East Technical University, Turkey (23, 88, 13); Parker, N = 5 Dianne, University of Manchester, United Kingdom (3, 11, 4); Summala, Heikki, University of Helsinki, Finland; Warner, Henriette Wallen, The Swedish National	Author cluster	Authors (number of links, total link strength, number of documents)
		University of Technology, Estonia (17, 67, 4); Shubenkova, Ksenia, Kazan Federal University, Russia (17, 67, 4); Solmazer, Gaye, Middle East Technical University, Turkey (17, 67, 4); Uzumcuoglu, Yesim, Middle East Technical University, Turkey (19, 56, 5); Xheladini, Gentiane*, Kosovo Association of Motorization, Kosova































































n = 1 Road and Transport Research Institute, Sweden (2, 5, 4)
#18		Ma, Jianming, Texas Department of Transportation, United States (5, 12, 7); Rong, Jian, Beijing University of Technology, China (3, 11, 8); Wu, Yiping, Beijing
N = 4 University of Technology, China (3, 8, 4); Zhao, Xiaohua*, Beijing University of Technology, China (4, 12, 8)
n = 3

Appendix C. Most cited LMIC papers of each cluster (clusters are based on similarity of references, and are in relation to Fig. 9, bottom part)

Title	Authors (year)	Journal	LMIC	Citation count
Cluster #1
Risk factors associated with traffic violations and accident severity in China Road traffic deaths, injuries and disabilities in India: current scenario	Zhang et al. (2013a) Gururaj (2008)	Accident Analysis & Prevention National Medical Journal of India	China India	137 92
Risky behavior of drivers of motorized two wheeled vehicles in India Understanding on-road practices of electric bike riders: An observational	Dandona et al. (2006a) Du et al. (2013)	Journal of Safety Research Accident Analysis & Prevention	India China	67 65
study in a developed city of China Logistic regression analysis of pedestrian casualty risk in passenger vehicle collisions in China	Kong and Yang (2010)	Accident Analysis & Prevention	China	62
Reducing the legal blood alcohol concentration limit for driving in developing countries: a time for change?	Andreuccetti et al. (2011)	Addiction	Brazil	61
Sleep Habits and Accident Risk Among Truck Drivers: A Cross-Sectional Study in Argentina	Perez-Chada et al. (2005)	Sleep	Argentina	60
The influence of anger, impulsivity, sensation seeking and driver attitudes on risky driving behaviour among post-graduate university students in Durban, South Africa	Bachoo et al. (2013)	Accident Analysis & Prevention	South Africa	57
Improper motorcycle helmet use in provincial areas of a developing country	Li et al. (2008b)	Accident Analysis & Prevention	China	56
Pedestrians' injury patterns in Ghana	Damsere-Derry et al. (2010)	Accident Analysis & Prevention	Ghana	54
Magnitude and categories of pedestrian fatalities in South Africa Road Traffic Injury in China: A Review of National Data Sources Exploring the relationship between development and road traffic injuries: a	Mabunda et al. (2008) Ma et al. (2012) Garg and Hyder (2006)	Accident Analysis & Prevention Traffic Injury Prevention European Journal of Public	South Africa China India	49 48 47
case study from India Perceptions of traffic risk in an industrialised and a developing country	Nordfjaern and Rundmo (2009)	Health Transportation Research Part F	Ghana	47
Cluster #2 A simultaneous equations model of crash frequency by collision type for	Ye et al. (2009)	Safety Science	Georgia	100
rural intersections A Multinomial Logit Model of Pedestrian–Vehicle Crash Severity	Tay et al. (2011)	International Journal of Sustainable Transportation	South Korea	92
Corridor-level signalized intersection safety analysis in Shanghai, China using Bayesian hierarchical models	Xie et al. (2013)	Accident Analysis & Prevention	China	72
Exploring the effects of roadway characteristics on the frequency and severity of head-on crashes: Case studies from Malaysian Federal Roads	Hosseinpour et al. (2014)	Accident Analysis & Prevention	Malaysia	70
Multivariate spatial models of excess crash frequency at area level: Case of Costa Rica	Aguero-Valverde (2013)	Accident Analysis & Prevention	Costa Rica	59
Accident analysis with aggregated data: The random parameters negative binomial panel count data model	Coruh et al. (2015)	Analytic Methods in Accident Research	Turkey	56
Temporal distribution of motorcyclist injuries and risk of fatalities in relation to age, helmet use, and riding while intoxicated in Khon Kaen, Thailand	Nakahara et al. (2005)	Accident Analysis & Prevention	Thailand	54
Analyzing freeway crash severity using a Bayesian spatial generalized ordered logit model with conditional autoregressive priors	Zeng et al. (2019)	Accident Analysis & Prevention	China	52
Extension of the Application of Conway-Maxwell-Poisson Models: Analyzing Traffic Crash Data Exhibiting Underdispersion	Lord et al. (2010)	Risk Analysis	South Korea	49
A multinomial logit analysis of risk factors influencing road traffic injury severities in the Erzurum and Kars Provinces of Turkey	Celik and Oktay (2014)	Accident Analysis & Prevention	Turkey	49
Factors affecting motorcyclists' injury severities: An empirical assessment using random parameters logit model with heterogeneity in means and variances	Waseem et al. (2019b)	Accident Analysis & Prevention	Pakistan	48
Bayes classifiers for imbalanced traffic accidents datasets Cluster #3	Mujalli et al. (2016)	Accident Analysis & Prevention	Jordan	46
Young driver risky behaviour and predictors of crash risk in Australia, New Zealand and Colombia: Same but different?	Scott-Parker and Oviedo Trespalacios (2017)	Accident Analysis & Prevention	Colombia	59
Why the government should be blamed for road safety	Jing et al. (2020)	International Journal of Occupational Safety and Ergonomics	China	26
Mobile phone use among motorcyclists and electric bike riders: A case study of Hanoi, Vietnam Cluster #4	Truong et al. (2016b)	Accident Analysis & Prevention	Vietnam	22
Cross-cultural differences in driving behaviours: A comparison of six countries	Ozkan et al. (2006a)	Transportation Research Part F	Iran, Turkey	177
The effect of conformity tendency on pedestrians' road-crossing intentions in China: An application of the theory of planned behavior	Zhou et al. (2009a)	Accident Analysis & Prevention	China	124
Social psychology of seat belt use: A comparison of theory of planned behavior and health belief model	Simsekoglu and Lajunen (2008)	Transportation Research Part F	Turkey	102
Cross-cultural comparisons of traffic safety, risk perception, attitudes and behaviour	Lund and Rundmo (2009)	Safety Science	Ghana	97
The "genetics" of driving behavior: parents' driving style predicts their children's driving style	Bianchi and Summala (2004)	Accident Analysis & Prevention	Brazil	94
	Nordfjærn et al. (2011)	Journal of Risk Research		70 (continued on next page)

(continued )

Title	Authors (year)	Journal	LMIC	Citation count
A cross-cultural comparison of road traffic risk perceptions, attitudes towards traffic safety and driver behaviour			India, Ghana, Tanzania, Uganda
Cross-cultural comparison of drivers' tendency to commit different aberrant driving behaviours	Warner et al. (2011)	Transportation Research Part F	Turkey	62
Effects of Personality on Risky Driving Behavior and Accident Involvement for Chinese Drivers	Yang et al. (2013)	Traffic Injury Prevention	China	61
Traffic Safety for Electric Bike Riders in China: Attitudes, Risk Perception, and Aberrant Riding Behaviors	Yao and Wu (2012)	Transportation research Record	China	54
Aberrant driving behaviors: A study of drivers in Beijing	Shi et al. (2010)	Accident Analysis & Prevention	China	54
Cross-cultural differences in driving skills: A comparison of six countries	Ozkan et al. (2006b)	Accident Analysis & Prevention	Iran, Turkey	53
Influence of traffic enforcement on the attitudes and behavior of drivers	Stanojevic et al. (2013)	Accident Analysis & Prevention	Serbia, Northern Kosovo	53
Cluster #5
A cross-sectional observational study of helmet use among motorcyclists in Wa, Ghana	Akaateba et al. (2014)	Accident Analysis & Prevention	Ghana	37
Cultural foundations of safety culture: A comparison of traffic safety culture in China, Japan and the United States	Atchley et al. (2014)	Transportation Research Part F	China	34
Cross-sectional study of road accidents and related law enforcement efficiency for 10 countries: A gap coherence analysis	Urie et al. (2016)	Traffic Injury Prevention	India, Morocco, Argentina, South	31
Interaction between socio-demographic characteristics: Traffic rule violations and traffic crash history for young drivers	Alver et al. (2014)	Accident Analysis & Prevention	Korea, Turkey	30

Appendix D. Hybrids maps of term co-occurrence with an overlay of the average year of publication for each detected term. The top map is corresponding with the general road safety literature and the bottom map is corresponding with the LMIC subset of the literature