Factors Influencing Rural Freight Transport in Bangladesh: Insights from Generalized Linear and Additive Models

Abstract

Freight transportation is an important research topic in the transportation domain of high-income countries, but middle-income and low-income countries lack quality research on this, especially for heterogeneous roadway freight movement on narrower rural roads. This study addresses this issue in northern Bangladesh, analyzing the volume of five dedicated freight vehicle types across 1,162 roadway segments. A negative binomial generalized linear model (NB-GLM), along with a supporting negative binomial generalized additive model (NB-GAM), was found suitable to better explain the interrelation between freight volume and different predictors like area, population, household size, solvency rate, crest width, embankment height, and international roughness index. Freight movement was notably higher on market days. The NB-GLM highlighted significant linear effects of socioeconomic attributes, whereas the NB-GAM revealed strong nonlinear influences of roadway and spatial characteristics. Distinct variations were observed for predictor significance in statistical terms across different vehicle classes. These findings serve as a decision-support tool for policy makers wanting to implement targeted interventions, including temporal zoning, last-mile surface funds, and village freight consolidation hubs. The results provide a robust framework for predicting freight movements in regions with similar economic conditions, aiding in sustainable road network development and maintenance planning.

Keywords

rural freight transport sustainable transport logistics and mobility generalized additive models (GAMs)generalized linear models (GLMs)

Introduction

Freight transportation is a critical component of both national and global economies, playing a significant role in economic development, trade, and sustainability. It accounts for approximately 9.5% of the U.S. gross domestic product (GDP) and supports the import and export of about $3.6 trillion in international trade ( 1 ). The sector is also a major contributor to greenhouse gas emissions, responsible for about 8% of emissions in the United States and 41% of total transport emissions globally, necessitating efficient and sustainable management to meet environmental objectives ( 1 , 2 ). Freight transportation networks, characterized by their complexity and redundancy, are essential for ensuring resilience against disruptions, as demonstrated in studies of network redundancy and the impact of events like the COVID-19 pandemic ( 3 , 4 ). The design and operation of these networks require strategic decisions about service frequency, mode, and routing to maintain competitiveness and reliability in a highly competitive market ( 5 ). Furthermore, the development of logistics hubs and the consumption of industrial space are closely linked to freight activity, underscoring the importance of integrating transportation services with industrial planning ( 6 ). The future of freight transport is poised for transformation, driven by the need to reduce emissions and improve efficiency, with innovative concepts and technologies being explored to reconcile growth with environmental goals ( 7 ). Additionally, freight logistics can contribute to sustainability by internalizing externality costs, which may encourage shifts to more efficient modes like rail, potentially lowering overall freight costs ( 8 ). Overall, freight transportation is indispensable for economic growth and sustainability—not only in the Global North but also in low- and middle-income countries—requiring careful planning and innovation to address its environmental impact and enhance its efficiency and resilience ( 9 , 10 ). Although regional freight generation models can be found in academia in the context of emerging economic powerhouse like India ( 11 ), in the case of its neighbors, like Bangladesh, where the population is heavily dependent on the rural transportation system given its vast agricultural production zones, a research gap exists.

The study of rural roadway freight transportation is essential for understanding and enhancing the socioeconomic development of rural areas, particularly in low- and middle-income countries, where there are budget constraints. Rural roads are pivotal in improving access to local markets, which can significantly boost household incomes by facilitating the sale of agricultural products and the purchase of modern inputs such as modern agricultural tools and fertilizers, as demonstrated in studies from Ghana and Cambodia ( 12 , 13 ). These roads also play a vital role in achieving Sustainable Development Goals by linking rural communities to essential services such as education and healthcare, thereby contributing to broader socioeconomic benefits ( 14 ). The construction and maintenance of rural roads can lead to increased economic activity, as evidenced by the iRoad program in Sri Lanka, which showed enhanced local economic activity through improved road connectivity ( 15 ). However, the benefits of rural road investments are not uniformly distributed, often favoring those closer to the roads, which necessitates careful planning and optimization to ensure equitable access and benefits for remote populations ( 16 , 17 ). Furthermore, rural road infrastructure is integral to disaster resilience, as it reduces vulnerability to storm damage and facilitates efficient disaster relief efforts ( 18 ). In China, rural infrastructure, including roads, has been shown to drive economic development through improved living environment governance, highlighting the multifaceted impact of such investments ( 19 ). Despite these benefits, challenges remain, such as the need for sustainable maintenance practices and the integration of alternative fuels to mitigate the impact of fluctuating fuel prices on transportation costs ( 12 , 20 ). As freight vehicles contribute to heavy loadings of these roads, studies on freight vehicle demand and the number of vehicles counted on the roads can be particularly beneficial, especially for middle- or low-income countries, where budget constraints are a significant challenge.

By analyzing freight movements, particularly in rural areas, stakeholders can identify inefficiencies, enhance service delivery, and support policy making aimed at fostering sustainable rural development ( 21 ). Modeling freight counts plays a particularly crucial role in supporting targeted infrastructural developments in rural areas by providing insights into transportation patterns and optimizing resource allocation. For instance, the integration of weight-categorized truck flow data can identify high-demand zones for freight movement, thereby informing infrastructure investments that enhance connectivity and reduce inefficiencies, such as empty or partially loaded truck trips ( 22 ). Additionally, rural road optimization models, which consider both vehicular and walking access, ensure that even the most remote populations benefit from improved road networks, thus fostering economic growth and resilience ( 23 ). By integrating passenger and freight transportation, as proposed in the demand-driven service model, rural areas can achieve better vehicle capacity utilization and cost efficiency, ultimately supporting sustainable development ( 24 ). Collectively, these modeling approaches facilitate informed decision making that targets infrastructural improvements effectively in rural contexts ( 15 ). Again, improved road infrastructure can significantly enhance access to markets, thereby increasing the value of farmland and benefiting poorer households, although it may not sufficiently address income inequality ( 16 ). Additionally, understanding the speed behavior of freight vehicles under various road–weather conditions can inform the development of weather-responsive traffic management strategies, improving safety on rural highways ( 25 ). The analysis of traffic collisions on rural roads also highlights the need for targeted safety measures, as truck driving is associated with high accident rates ( 26 ). Moreover, optimizing rural road planning through multimodal travel models ensures that even remote populations benefit from new infrastructure, enhancing overall accessibility and economic opportunities ( 17 ). Finally, temporal analysis of crash factors in rural settings provides insights into evolving safety risks, enabling the development of more effective measures to reduce large-truck crashes ( 27 ).

This research initiative was thus undertaken to explore the determinants of freight vehicle movement in the rural areas of a lower-/middle-income country. Bangladesh, located in South Asia, was selected as the case study location. As budget constraints in middle- and low-income countries call for careful consideration when investing in freight transport-related infrastructural development (e.g., geometry and structural design of both paved and earthen roads, access roads to ferry and cargo boat jetties, approach roads to bridges, stop/resting locations for trucks, refueling station locations, weighing stations, road signage and markings), a comprehensive freight vehicle count model would be extremely useful for a country like Bangladesh. The main objective was to understand the interrelation between different types of freight vehicle count and several socioeconomic, spatial, and roadway-related variables using a robust modeling technique. A little research was found on freight transportation in Bangladesh (shipping and port-related studies were not considered here). One study developed an origin–destination synthesis model with mode choice for Bangladesh as part of developing a freight demand model for the country ( 28 ). Another research initiative studied the efficiency of rail freight transport in the country ( 29 ). Finally, two studies provide a detailed description of a novel method to build freight models in data-scarce regions like Bangladesh by combining optimal sampling planning and smart data collection (surveys of businesses) with advanced modeling to estimate freight flows, mode choices, and even empty trips for better policy planning ( 30 , 31 ). No other studies focusing on freight transportation in Bangladesh could be found.

Literature Review

Literature researching the interrelation between vehicle count and different socioeconomic, spatial, temporal, and infrastructure-related variables is widely available. However, very few studies address particular count models for freight vehicles only. Predicting rural freight vehicle count involves a variety of variables, as highlighted across multiple studies. Spatial variables are notably significant, with road characteristics and spatial autocorrelation playing crucial roles in traffic volume prediction models. For instance, spatial variables were identified as the most significant group in a study using a random forest algorithm for truck traffic volume prediction on Florida’s interstate highways, demonstrating the importance of spatial data in forecasting models ( 32 ). Similarly, spatial autocorrelation was found to be significant in zero-inflated negative binomial models for freight trip generation, emphasizing the need to consider spatial interactions in predictive models ( 33 ). The use of segment-based regression kriging has shown improved accuracy in predicting heavy vehicle traffic volumes by accounting for spatial characteristics and the homogeneity of road segments, which are crucial for understanding the impact of heavy vehicles on road maintenance ( 34 ). Economic development levels and the presence of large freight traffic generators (LTGs) also influence freight vehicle counts, as these factors are associated with higher freight demand and significant truck trips, respectively ( 35 ). Furthermore, the integration of spatiotemporal data, as seen in the TripCFDM (Trip-based Collective Flow Diffusion Model), enhances the estimation of freight trip distribution by utilizing aggregate data along with trip data, thereby improving the accuracy of intrazonal freight trip predictions ( 36 ). These studies collectively underscore the importance of incorporating spatial, economic, and infrastructural variables in models to accurately predict freight vehicle counts.

Traditional continuous dependent variable assumed regression models can be misleading in finding out the underlying interrelations between an integer count data and different factors. Whereas generalized linear models (GLMs) and generalized additive models (GAMs) are particularly useful for vehicle count data owing to their ability to handle complex data characteristics such as spatial clustering, nonlinear relationships, and overdispersion, which are common in traffic data. The negative binomial GAM (NB-GAM) model, for instance, effectively addresses these issues by incorporating a smooth spatial trend surface to control for spatial unobserved heterogeneity, which is crucial for accurate vehicle count predictions in regions with varying traffic patterns ( 37 ). This model’s flexibility in capturing nonlinear effects is beneficial in traffic scenarios where the relationship between traffic density and vehicle counts is nonlinear, such as the diminishing returns of agglomeration economies in highly dense areas ( 37 ). Furthermore, GAMs can be integrated with other methodologies to enhance vehicle count accuracy. For example, the use of deep-learning models for real-time vehicle detection and counting, as seen in various studies, can complement GAMs by providing high-precision data inputs. These models, such as those using YOLO (You Only Look Once) and deep convolutional neural networks, offer robust detection capabilities even under challenging conditions like low illumination or occlusion ( 38 – 40 ). Furthermore, the integration of machine learning techniques with GAMs can improve the reliability of vehicle count estimates by leveraging inexpensive and scalable data sources, such as Bluetooth sensors and probe vehicle data, which are crucial for real-time traffic management ( 41 ). Overall, the adaptability of GLMs and GAMs in incorporating spatial and temporal variations, along with their ability to integrate advanced detection technologies, makes them powerful tools for analyzing and predicting vehicle count data in complex traffic environments.

This research is thus novel in several ways. Firstly, it explores freight vehicle movement in the rural regions of Bangladesh; secondly, it considers several socioeconomic and roadway characteristics in understanding the freight vehicle count of a low-/middle-income country. Thirdly, it incorporates a combined analysis of regression-based NB-GLMs and NB-GAMs to accurately capture the underlying relation between the dependent variable and predictors. The findings could be beneficial in predicting freight vehicle movements in other regions of Bangladesh and even other countries with similar economic conditions, which in turn could help formulate policies like rural and semiurban road network development, maintenance, and environmental conservation, as freight vehicle movement is directly linked with those issues. The following sections will cover the methods employed and the data collected, the formal data analysis, and will discuss the results, concluding with an overview on future research directions.

Methods

Study Location and Data

Bangladesh, a South Asian country covering 147,570 km², is bordered by India and shares a minor border with Myanmar. It lies adjacent to the Bay of Bengal and is characterized as a low-lying delta formed by major rivers, making it highly fertile yet flood-prone. The administrative structure consists of 8 divisions, 64 districts, 495 subdistricts (upazilas), and 4,554 unions, alongside 12 city corporations and 330 municipalities ( 42 , 43 ). Being one of the most densely populated countries globally, it has a population of approximately 164 million people, resulting in a density of 1,142 people per square kilometer ( 44 ). The country’s economy is predominantly agricultural, with agriculture employing a large portion of the labor force and contributing significantly to the gross national product ( 45 ). Despite this, Bangladesh faces challenges in food security, exacerbated by climate change and natural disasters, which frequently disrupt agricultural productivity ( 44 , 45 ). The GDP per capita has seen growth, increasing steadily from USD 203 in 1975 to USD 641 in 2010 and presently is around USD 2,700, yet it remains low compared with global averages. The spatial distribution of the economy of the country can be understood from the poverty distribution map of the country in Figure 1. Key agricultural activities include rice and fish production, with Bangladesh ranking fourth and sixth globally in these sectors, respectively. Industrial activities are centered around textiles, food manufacturing, and emerging sectors like basic metals and machinery, which are crucial for economic growth ( 46 ). The agricultural sector, however, faces constraints such as inadequate irrigation, input shortages, and climate impacts, which hinder productivity, especially in lagging regions ( 47 ). Addressing these challenges through improved agricultural practices and policy support is essential for sustainable development and economic growth in Bangladesh ( 46 ).

Figure 1.

Poverty map of Bangladesh ( 48 ).

Bangladesh’s transportation system is a complex network that plays a crucial role in the country’s economic and social development. However, the system faces significant challenges, including congestion, inadequate infrastructure, and environmental concerns. Investments in the transportation sector have enhanced productivity and economic development, contributing significantly to GDP growth. The transport sector also plays a crucial role in enhancing exports, marketing agricultural goods, and boosting industrial production ( 49 ). Bangladesh faces severe transportation challenges as a result of rapid urbanization, high vehicular growth, and inadequate traffic management. These issues have led to significant congestion and environmental problems ( 50 ). The rural road network is predominantly administered by the Local Government Engineering Department (LGED). LGED oversees approximately 4,718 upazila roads (UZRs), which generally pass through and connect comparatively developed, dense rural areas, which have a wider crest width, that is the total top surface width of a roadway embankment, including driving lanes and shoulders (dimension not consistent for all UZRs, but the variations are minimal); 8,074 union roads (UNRs), which generally pass through and connect less dense rural areas that tend to be less economically developed and have a narrower crest width than UZRs (dimension not consistent for all UNRs but the variations are minimal); and 134,236 village roads (VRs). The vehicle count data utilized in this study were obtained from the Rural Connectivity Improvement Project of LGED, which is financially supported by the Asian Development Bank. An extensive analysis involving 30,000 road segments (UZRs, UNRs, and VRs), with a total span of 355,000 km, was performed as part of this project. The data collection process took place between 2023 and 2025. Classified vehicle count surveys at these segments were conducted manually, whereby enumerators recorded the presence of various vehicles from 7:00 a.m. to 7:00 p.m., on both market and nonmarket days. Market here refers to days of the week when, typically, vendors from the local and neighboring areas gather to sell their produce at wholesale and retail outlets. On these days, the market zones usually experience high vehicle traffic. The categories of vehicles surveyed, along with a brief description, are summarized in Table 1. Dedicated freight vehicle–related information can be found in Table 2 and a schematic illustration in Figure 2. The designed freight-vehicle payloads, dimensions, and purchasing costs in local markets of Bangladesh (including taxes) were obtained from the manufacturer’s manual and website. Modified payloads of all types of vehicles and the hiring cost of each class on a per-kilometer basis were determined by a field survey carried out at Tejgaon truck terminal. This survey included a short questionnaire (face-to-face) administered to three major freight service providers.

Table 1.

Types of Vehicles Considered for Volume Count

Vehicle type	Description
Auto rickshaw	Three-wheeled battery/CNG-driven vehicles
Bicycle	All two-wheeled pedal cycles
Car	All types of sedans
Light commercial vehicle^a	Mostly locally converted sports utility vehicles (SUVs) for goods movement
Heavy truck^a	Three or more axles, multiaxle tandem trucks, container carriers, and articulated vehicles
Large bus	>40 seats and >36-ft chassis
Medium truck^a	Two-axle rigid trucks >3-ton payload
Micro bus	Up to 16 seats
Minibus	16 to 39 seats and <36-ft chassis
Motorcycle	All two-wheeled motorized vehicles
Rickshaw	Three-wheeled passenger nonmotorized vehicle
Rickshaw van^a	Three-wheeled cargo nonmotorized vehicle
Small truck^a	Two-axle rigid <3-ton payload
Tempo	Large passenger and cargo-carrying auto tempo/van and four-wheeler human haulers
Utility car	Four-wheel drive vehicles, SUVs

Note: CNG = Compressed Natural Gas.

Dedicated freight vehicle

Table 2.

Dedicated Freight Vehicle Characteristics

Vehicle type	Designed payload^a (tons)	Modified payload^b (tons)	Typical dimension^b (length [ft] x width [ft])	Price^a (USD)	Hiring cost^b (USD/km)
Light commercial vehicle	1.3	1.56	16 x 6	8,000	0.2–0.28
Heavy truck	19.5	27.3	28 x 8	42,160	0.64–1.0
Medium truck	10	16.65	20 x 7.5	14,800	0.4–0.56
Small truck	3	4.5	15 x 5.5	11,200	0.28–0.36
Rickshaw van	na	0.95	8 x 3.5	320	0.16–0.32

Note: na = not applicable.

Based on vehicle manufacturer and structural/dead weight of local vehicles.

Based on the field survey.

Figure 2.

Dedicated freight vehicles.

The data acquisition process transpired between the years 2023 and 2025. Additional information incorporated variables such as the width of the roadway carriageway, the height of the embankment, and international roughness index (IRI) measurements. Given the extensive volume and heterogeneity of the data, northern Bangladesh was selected as the focal area for this investigation, specifically comprising one administrative division (Rangpur) out of the total eight divisions in the country (Figure 3). This geographical region is recognized for its significant productivity in agricultural and livestock outputs, thus enabling the research to derive insightful findings that can ultimately be contrasted with other regions across the nation. As per the LGED’s classification, this division is further partitioned into the Rangpur and Dinajpur regions. Within these two regions there are 8 districts and 57 Upazilas. For the purposes of this research, a total of 1,162 road segments with an aggregate length of 11,248 km were selected, based on their usability, of which 544 segments (with a total length of 4,186 km) are designated as UZRs, whereas 618 segments (with a total length of 3,331 km) are categorized as UNRs.

Figure 3.

Study location (colored red within the map of Bangladesh) ( 51 ).

Model Structure

To understand the dynamics behind freight vehicle movement, several available variables were selected for testing. The predictor variables are listed in Table 3. Apart from LGED, several demographic, geographic, and economic data were collected from the Bangladesh Bureau of Statistics (a government agency responsible for surveying and storing national data of Bangladesh). The predicted variable or the vehicle count data per day is the number of vehicles passed through a selected roadway section on any given day in the study period. Data was collected on both market day and nonmarket day for the same roadway segments. The crest width (measured in meters and is the usable top width of roadway for vehicle movement and emergency stops) of the pavement can directly restrict access of wide-body vehicles and plays a crucial role in freight transportation; thus, it is included in the model. Embankment height (measured in meters) of roads in rural Bangladesh is associated with frequent upward and downward slopes, which prevent the usage of nonmotorized and heavily loaded freight vehicles in high roadway embankment-dominated areas. The variation of roadway embankment is high in Bangladesh in closely spaced areas because of the presence of many waterbodies compared to any other part of the world. The IRI value (the higher the value, the rougher the surface is) of the road indicates the roughness of the pavement surface, which in turn can indicate the extensive usage of heavy vehicles and unsuitable roads for lighter vehicles, especially nonmotorized ones. Market and nonmarket days show significant variations in roadway traffic in rural Bangladesh, which is why they are kept as a dummy variable in the model. As rural Bangladesh is dominated by two types of roadway standards according to LGED, the dummy UZR (Upazilla Road) variable is kept to check the variations in rural roadway freight traffic volume. In the context of predicting rural freight volume in Bangladesh at the Upazila level, several other geographic, demographic, and economic variables are essential because they collectively capture the region’s aggregate supply and demand dynamics, which are the core drivers of freight generation. Population (in numbers) and area (in square kilometers) are deliberately maintained as separate independent variables rather than combined into a density ratio (population per square kilometers) because they proxy for two fundamentally distinct economic drivers: population determines the scale of consumption demand (inbound freight for staples), whereas area, in a rural context, acts as a direct proxy for arable land availability and agricultural production capacity (outbound freight). Merging them into a density metric would conflate these effects and mathematically obscure the positive marginal impact of land size on freight generation. Complementing these, household size (number of people per family) is critical because large joint families in Bangladesh often function as consolidated economic production units, providing the labor force necessary to generate marketable agricultural surplus. Finally, Solvency and literacy rate serve as proxies for purchasing power and economic maturity; higher solvency drives can demand, for example, heavier non-food items like construction materials, while literacy correlates with the adoption of modern farming technologies, thereby increasing the density and commercial complexity of the freight network.

Table 3.

Variable Description

Variable	Unit	Source
Vehicle count (daily traffic)	Number	RCIP Project, LGED
Crest width	Meters	RCIP Project, LGED
Roadway embankment height	Meters	RCIP Project, LGED
IRI	Value	RCIP Project, LGED
Market/nonmarket day	Binary (1/0)	RCIP Project, LGED
UZR/UNR	Binary (1/0)	RCIP Project, LGED
Area (of the upazila)	Square kilometers	GIS section, ICT unit, LGED
Population (of the upazila)	Number	BBS ( 52 )
Household size (of the upazila)	Average number of persons per house	BBS ( 52 )
Literacy rate (of the upazila)	Percent of total population	BBS ( 52 )
Solvency (of the upazila)	Percent of total population	BBS ( 52 )

Note: RCIP = Rural Connectivity Improvement Project; LGED = Local Government Engineering Department; IRI = international roughness index; UZR = upazila roads; UNR = union roads; GIS = geographic information system; ICT = information and communications technology; BBS = Bangladesh Bureau of Statistics.

Vehicle counts were modeled separately using negative binomial GLMs (NB-GLMs) and NB-GAMs. Although multiple vehicle categories were observed on the same roadway segments, the counts were not compositional nor constrained to sum to a fixed total. Increases in one vehicle type did not automatically imply decreases in others, and multiple vehicle classes increased simultaneously in response to common demand factors. Shared influences were accounted for through common covariates, and the models assumed conditionally independent arrival processes. Therefore, separate count models were assumed to be methodologically appropriate and consistent. A multicollinearity analysis check was performed to strengthen this assumption and is shown in Figure S1 of the Appendix.

The modeling starts with testing the ordinary least squares (OLSs) regression. The general equation can be written as shown in Equation 1,

y_{k} = β_{0} + β_{1} x_{1} + β_{2} x_{2} + \dots \dots . β_{n} x_{n} + ϵ

(1)

where

$y_{k}$ is observed traffic count for the vehicle type, $k$ ;

$x_{n}$ is explanatory variable (crest width, population, etc.);

$β_{n}$ is model parameter; and

$ϵ$ is error term.

Market day/nonmarket day and UZR/UNR variables are coded as binary (market day = 1, UZR = 1, all else 0). Considering the count data type outcome variable, two different types of GLMs are considered. Firstly, the Poisson regression-based (with log link) and secondly, the negative binomial regression-based approach. The Poisson GLM for traffic counts of vehicle type, $k$ , can be written as

\begin{matrix} y_{k} ~ Poisson (μ_{k}) \\ \ln (μ_{k}) = β_{0} + β_{1} x_{1} + β_{2} x_{2} + \dots \dots . β_{n} x_{n} + ϵ \end{matrix}

(2)

where $μ_{k} = E [y_{k}]$ is expected mean count.

Here, the log link function ensures that the linear predictor is mapped to a strictly positive expected count. The NB-GLM approach can be expressed by Equation 3 (basically, the main equation is the same as Equation 2),

\begin{matrix} y_{k} ~ NB (μ_{k}, θ) \\ \ln (μ_{k}) = β_{0} + β_{1} x_{1} + β_{2} x_{2} + \dots \dots . β_{n} x_{n} + ϵ \\ Var (y_{k}) = μ_{k} + \frac{μ_{k}^{2}}{θ} \end{matrix}

(3)

where $θ$ is the dispersion parameter that allows variance to exceed the mean. As with the Poisson GLM, the log link function ensures positive fitted values. Although the NB-GLM effectively accounts for overdispersion in count data, it assumes strictly linear relationships between predictors and response. To capture potential nonlinear associations in infrastructure, demographic, and socioeconomic variables, the NB-GAM can allow smooth functions that flexibly model complex patterns in the data. It can be structured as follows (Equation 4):

\begin{matrix} y_{k} ~ NB (μ_{k}, θ) \\ \ln (μ_{k}) = β_{0} + \sum_{j = 1}^{p} s_{j} (x_{j}) + \sum_{m = 1}^{q} γ_{m} z_{m} \end{matrix}

(4)

where $s_{j} (.)$ are smooth functions (e.g., splines) of continuous covariates, $x_{j}$ ; $z_{m}$ are linear (parametric) covariates (e.g., indicator variables), with coefficients, $γ_{m}$ ; $j$ is index for smooth functions; $p$ is number of predictors (covariates) in the model; $m$ is number of smooth terms in the GAM; and $q$ is number of parameters for a given smooth function (basis coefficients). The log link enforces $μ_{k} > 0$ via $μ_{k} = \exp (β_{0} + \sum_{j} s_{j} (x_{j}) + \sum_{m} γ_{m} z_{m})$ . To compare among different models prepared, the Akaike information criterion (AIC), R², pseudo R², Cragg–Uhler (Nagelkerke) R², dispersion, deviance explained (%), and so forth, were used in relevant cases. All the modeling works were conducted using RStudio 2024.04.2. Data processing was performed on a system running Windows 11 with an Intel Core i7 processor (2.30GHz) and 32GB of installed memory.

Results

The summary statistics of the data can be found in Table 4. Considerable variation existed across districts, with Dinajpur reporting the longest network (1,787 km) and the highest traffic volumes, whereas Kurigram had the shortest network (644 km) and correspondingly lower volumes.

Table 4.

Summary Statistics of Data

		Road segments (n)		Road length (km)		Vehicle count (market day)		Vehicle count (nonmarket day)
District	Upazila (n)	UZR	UNR	UZR	UNR	Freight	Passenger	Freight	Passenger
Dinajpur	13	128	136	1,000	787	125,793	336,454	99,920	286,440
Gaibandha	6	56	52	437	245	59,879	111,329	48,327	93,284
Kurigram	8	43	74	286	358	16,143	168,977	13,360	152,197
Lalmonirhat	5	50	59	344	269	15,125	116,085	12,113	100,600
Nilphamari	6	72	67	515	323	62,439	168,907	50,319	145,797
Panchagarh	5	45	66	419	368	51,845	171,949	41,036	144,515
Rangpur	8	90	99	713	519	91,032	217,178	74,383	181,528
Thakurgaon	5	60	65	472	463	50,607	245,088	40,896	218,578
			Freight vehicle density (n/km)			Freight vehicle share (%)		Freight vehicle volume increase on market day (%)
District	Road type		Nonmarket day		Market day	Nonmarket day	Market day	Freight vehicle volume increase on market day (%)
Dinajpur	UZR		33		42	26.01	27.34	24.11
Dinajpur	UNR		84		107	25.79	27.15	26.79
Gaibandha	UZR		37		46	32.54	33.72	23.75
Gaibandha	UNR		131		162	35.00	35.65	23.98
Kurigram	UZR		20		25	6.54	7.05	27.69
Kurigram	UNR		21		25	9.77	10.85	15.71
Lalmonirhat	UZR		13		16	10.03	10.49	26.60
Lalmonirhat	UNR		29		35	11.20	12.24	23.87
Nilphamari	UZR		28		34	21.58	22.84	23.98
Nilphamari	UNR		112		139	27.71	29.06	24.13
Panchagarh	UZR		48		59	21.64	22.29	22.62
Panchagarh	UNR		57		74	22.60	24.03	29.91
Rangpur	UZR		33		41	24.45	25.29	21.93
Rangpur	UNR		98		120	31.90	32.05	22.60
Thakurgaon	UZR		31		38	13.95	14.68	24.27
Thakurgaon	UNR		57		70	16.99	18.86	23.45

Note: UZR = upazila road; UNR = union road.

Traffic volumes, expressed as vehicle count per day, demonstrated a marked distinction between freight and passenger movement, and between market and nonmarket days. Passenger traffic dominated across all districts, with Dinajpur, Rangpur, and Thakurgaon showing the heaviest flows, often exceeding 200,000 passenger vehicles per day on market days. Freight traffic, although comparatively smaller, showed substantial sensitivity to market days. For instance, Dinajpur recorded nearly 126,000 freight vehicles on market days, compared with 99,920 on nonmarket days, reflecting a notable increase of around 24%. This pattern was consistent across districts, indicating that freight activity was significantly elevated during market days. The density and share of freight vehicles further highlighted spatial and temporal differences. On UZRs, freight vehicle density generally ranged between 13 and 48 vehicles per kilometer on nonmarket days and rose to between 16 and 59 vehicles per kilometer on market days. On UNRs, density was much higher, with several districts such as Gaibandha and Nilphamari exceeding 100 vehicles per kilometer. Correspondingly, freight share in total traffic was slightly higher on UNRs compared with UZRs across most districts.

Firstly, a correlation analysis was run to check for multicollinearity: area and population were found to be highly correlated (Figure 4). As a result, the population variable was removed (household size was kept as a proxy for this demographic characteristic). The counterintuitive results from the correlation matrix, such as in the pairing of literacy rate and solvency, reflected the decoupling of formal education from traditional rural wealth in Bangladesh. Although solvency is often derived from ancestral land ownership and agricultural output, higher literacy is frequently concentrated among younger generations transitioning into the service sector, who may not yet possess significant land-based assets. Again, the negative correlation between population and house size accurately captured the spatial demographic transition. High-population hubs and growth centers in rural Bangladesh typically favored nuclear family structures, whereas larger, extended families remained the norm in lower-population, remote agricultural zones to meet manual labor requirements for farming.

Figure 4.

Correlation matrix.

Model comparison across OLSs, Poisson GLM, NB-GLM, and NB-GAM frameworks revealed a consistent advantage of NB-GAMs for all vehicle categories (Table 5). The Poisson GLMs, although producing relatively high pseudo-R² and Nagelkerke values, were characterized by extreme overdispersion, with dispersion ratios exceeding acceptable limits (e.g., 49.97 for light commercial vehicles, 182.87 for rickshaw vans, and 50.96 for small trucks). This systematic inflation confirmed the unsuitability of the Poisson assumption given the overdispersed nature of the count data. OLS regressions achieved modest R² values (0.12 to 0.20), but these models were not theoretically appropriate for discrete, skewed count outcomes and failed to provide dispersion metrics or explain deviance. NB-GLMs improved dispersion control (close to 1.0 across all vehicles), but they exhibited low explanatory power, with pseudo-R² values near zero (0.02 to 0.04) and very limited deviance explained (≤0.22). By contrast, the NB-GAM consistently balanced goodness-of-fit and parsimony. AIC values were consistently lower than Poisson GLMs and comparable to or slightly better than NB-GLMs, while simultaneously yielding higher values for deviance explained (ranging from 0.28 for heavy trucks to 0.38 for rickshaw vans and 0.37 for light commercial vehicles). In addition, the use of smooth terms enabled NB-GAMs to capture nonlinear effects in infrastructural and socioeconomic predictors. Dispersion values across NB-GLMs and NB-GAMs remained close to unity, demonstrating a well-calibrated variance structure. Although pseudo-R² values were occasionally negative (rickshaw van, small truck), this could be attributed to penalization effects and did not outweigh the consistently superior deviance explained and dispersion control. Taken together, the performance indicators justified the choice of NB-GAM as the most suitable framework for modeling vehicle counts. It successfully addressed the overdispersion problems inherent in Poisson GLMs, outperformed OLS in explanatory adequacy, and improved on NB-GLMs by accommodating nonlinearities while maintaining appropriate dispersion. Thus, NB-GAM offered the most reliable and interpretable balance of model fit, variance control, and explanatory capability across all vehicle categories. Observed versus predicted counts for different NB-GAMs can be found in Figure 5.

Table 5.

Model Performance Analysis

Vehicle type	Model form	AIC	R ²	Pseudo R²	Cragg–Uhler (Nagelkerke) R²	Dispersion	Deviance explained
Rickshaw van	OLS	6,729	0.20	na	na	1.07	na
Rickshaw van	Poisson GLM	439,426	na	0.25	1.00	182.87	0.25
Rickshaw van	NB-GLM	30,427	na	0.02	0.24	1.13	0.22
Rickshaw van	NB-GAM	29,909	na	−0.64	−155.25	1.11	0.38
Light commercial vehicle	OLS	8,005	0.12	na	na	1.82	na
Light commercial vehicle	Poisson GLM	123,834	na	0.13	1.00	49.97	0.13
Light commercial vehicle	NB-GLM	17,707	na	0.02	0.15	1.18	0.13
Light commercial vehicle	NB-GAM	16,956	na	0.07	0.41	1.16	0.37
Heavy truck	OLS	6,295	0.15	na	na	0.87	na
Heavy truck	Poisson GLM	29,046	na	0.17	0.92	10.79	0.19
Heavy truck	NB-GLM	9,893	na	0.04	0.17	0.97	0.17
Heavy truck	NB-GAM	9,664	na	0.47	0.98	0.97	0.28
Medium truck	OLS	7,556	0.14	na	na	1.50	na
Medium truck	Poisson GLM	72,224	na	0.11	0.98	28.18	0.12
Medium truck	NB-GLM	15,437	na	0.02	0.12	1.15	0.10
Medium truck	NB-GAM	14,822	na	0.19	0.76	1.13	0.32
Small truck	OLS	7,725	0.18	na	na	1.62	na
Small truck	Poisson GLM	127,630	na	0.17	1.00	50.96	0.19
Small truck	NB-GLM	19,830	na	0.02	0.19	1.19	0.16
Small truck	NB-GAM	19,269	na	−0.06	−0.60	1.17	0.35

Note: AIC = Akaike information criterion; OLS = ordinary least squares; NB-GLM = negative binomial generalized linear model; NB-GAM = negative binomial generalized additive model; na=not applicable.

Figure 5.

Observed versus predicted counts for different negative binomial generalized additive models: (a) light commercial vehicle, (b) heavy truck, (c) medium truck, (d) rickshaw van, and (e) small truck.

The model estimates for NB-GLM and NB-GAM are given in Tables 6 and 7, respectively. The NB-GLM can help understand the positive–negative trend of the variables used, whereas the NB-GAM can help understand the nonlinearity of continuous variables. Across all models, market day and UZR consistently demonstrated positive and statistically significant coefficients in the NB-GLM, indicating stable effects across vehicle classes. For light commercial vehicles, the NB-GLM identified significant negative effects of area and household size, with solvency level contributing positively. In contrast, crest width, embankment height, and IRI were nonsignificant in the linear form.

Table 6.

NB-GLM Estimates

Variable	Rickshaw van	Light commercial vehicle	Heavy truck	Medium truck	Small truck
Intercept	6.92***	6.99***	−4.66***	−1.66	3.62***
Crest width	0.07***	0.06	0.07	−0.09	0.07*
Embankment Height	−0.03	−0.02	0.14	0.44	−0.07
IRI	−0.06***	−0.03	0.1***	−0.01	−0.01
Area (log/linear)	0.00***	−0.51***	0.00***	0.00	0.00***
Household size	−1.07***	−0.60**	0.56*	0.1	0.02
Literacy rate	0.04***	−0.01	0.03*	0.03**	−0.03***
Solvency level	0.01***	0.03***	0.00	0.02***	0.03***
Market day	0.21***	0.10	0.61***	0.34***	0.26***
UZR	0.60***	0.85***	1.18***	1.01***	0.70***

Note: NB-GLM = negative binomial generalized linear model; IRI = international roughness index; UZR = upazila roads.

p < .05; **p < .01; ***p < .001.

Table 7.

NB-GAM Estimates

Variable	Rickshaw van	Light commercial vehicle	Heavy truck	Medium truck	Small truck
Intercept	5.12***	2.17***	0.11	1.62***	2.73***
Market day	0.21***	0.15**	0.67***	0.41***	0.28***
UZR	0.58***	0.79***	1.04***	1.09***	0.50***
Crest width	edf = 3.03 χ² = 27.28 p < 0.001***	edf = 3.53 χ² = 11.86 p = 0.021*	edf = 3.72 χ² = 29.92 p < 0.001***	−0.041	edf = 5.83 χ² = 38.08 p < 0.001***
Embankment Height	edf = 3.72 χ² = 9.09 p = 0.073	edf = 5.60 χ² = 46.50 p < 0.001***	edf = 2.92 χ² = 6.50 p = 0.114	edf = 3.86 χ² = 29.65 p < 0.001***	edf = 4.85 χ² = 23.60 p = 0.001**
IRI	edf = 2.70 χ² = 13.25 p = 0.007**	edf = 5.60 χ² = 21.11 p = 0.003**	edf = 1.01 χ² = 6.06 p = 0.015*	edf = 4.60 χ² = 13.55 p = 0.024*	edf = 7.71 χ² = 65.64 p < 0.001***
Area	edf = 8.58 χ² = 189.37 p < 0.001***	edf = 8.30 χ²129.72 p < 0.001***	edf = 7.63 χ² = 77.37 p < 0.001***	edf = 8.23 χ² = 207.63 p < 0.001***	edf = 6.99 χ² = 41.58 p < 0.001***
Household size	edf = 8.43 χ² = 219.60 p < 0.001***	edf = 8.47 χ² = 137.50 p < 0.001***	edf = 7.36 χ² = 68.33 p < 0.001***	edf = 8.56 χ² = 253.59 p < 0.001***	edf = 8.77 χ² = 222.10 p < 0.001***
Literacy rate	edf = 8.60 χ² = 117.99 p < 0.001***	edf = 8.31 χ² = 269.98 p < 0.001***	edf = 3.45 χ² = 41.62 p < 0.001***	edf = 8.70 χ² = 155.24 p < 0.001***	edf = 8.71 χ² = 220.56 p < 0.001***
Solvency level	edf = 8.83 χ² = 273.30 p < 0.001***	edf = 8.58 χ² = 224.70 p < 0.001***	edf = 8.45 χ² = 102.77 p < 0.001***	edf = 7.54 χ² = 146.05 p < 0.001***	edf = 8.26 χ² = 202.55 p < 0.001***

Note: NB-GAM = negative binomial generalized additive model; edf = effective degrees of freedom; IRI = international roughness index; UZR = upazila road.

p < .01; ***p < .001.

However, the NB-GAM highlighted strong nonlinear responses for crest width and embankment height, and highly flexible smooths for socioeconomic factors (area, household size, literacy rate, solvency level) with large effective degrees of freedom (edf) values (in general, an edf close to 1 suggests linearity), suggesting complex relationships. The heavy truck’s NB-GLM departed from that of light commercial vehicles by producing a negative intercept and significant positive coefficients for IRI, household size, and literacy rate. No other roadway-related predictors were influential. The NB-GAM added further nuance: crest width and embankment height showed nonlinear smooths, whereas IRI remained largely linear. Socioeconomic variables again displayed pronounced nonlinearity with edf values above 3. For medium trucks, the NB-GLM reported a strong positive effect of embankment height and a marginal negative role of crest width, whereas literacy rate and solvency level contributed positively. Other variables remained negligible in linear form. In the NB-GAM, both embankment height and IRI exhibited distinct nonlinearities, and the socioeconomic predictors revealed smooth structures with high edf values, underscoring nonlinear dependence. In the small-truck model, the NB-GLM highlighted significant contributions from crest width (positive), area (negative), and literacy rate (negative), with solvency level being positive. No other predictors were linearly important. The NB-GAM captured nonlinear influences of crest width, embankment height, and IRI, alongside highly curved patterns for all socioeconomic indicators. For rickshaw vans, the NB-GLM showed negative effects from IRI, area, and household size, contrasting with positive roles for literacy rate and solvency level. In the NB-GAM, crest width and IRI were strongly nonlinear, whereas embankment height was only marginally significant. Socioeconomic predictors again dominated with very high edf values, indicating complex nonlinear behavior.

To give an example, the estimated smooth functions for the heavy truck model (NB-GAM) are shown in Figure 6. Crest width displayed a relatively flat trend at lower values with broadening uncertainty, followed by a gradual increase at higher widths. Embankment height exhibited a concave relationship, remaining stable up to mid-range values and then declining at greater heights. The smooth for IRI transitioned to a steady positive trend beyond moderate roughness levels. Area demonstrated one of the most complex nonlinear effects, with multiple oscillations including local peaks around intermediate sizes and dips at larger areas, indicating that its influence was not monotonic. Household size remained largely flat across most of its range, but exhibited sharp, localized increases toward the upper end. Literacy rate was strongly nonlinear, beginning with a negative slope at lower values, flattening at mid-range, and then rising progressively at higher literacy levels. Collectively, the smooth functions confirmed that the effects of roadway and socioeconomic variables on freight vehicle counts could not be captured adequately by simple linear terms. The estimated edf reported in the model supported these patterns, with high values for area, household size, and literacy rate (edf > 7), indicating complex nonlinearity, whereas crest width and embankment height showed moderate nonlinearity (edf ∼3 to 5).

Figure 6.

Smooth partial effects of continuous predictors on heavy truck counts from the NB-GAM, fitted using thin plate regression splines: (a) crest width, (b) embankment height, (c) international roughness index, (d) area, (e) household size, and (f) literacy rate.

Discussion

Model Interpretation

The modeling results revealed distinct associations between roadway, spatial, and socioeconomic attributes with rural freight vehicle movement in Bangladesh. Across all vehicle categories, market day and UZRs consistently exerted significant positive effects, underscoring the temporal and spatial concentration of freight flows. This reflects the market day or periodic market culture that is central to rural life in Bangladesh, where freight demand is not a steady daily stream but a rhythmic pulse. The notably higher sensitivity of heavy and medium trucks to market days compared with rickshaw vans and light commercial vehicles suggests that these cycles are the primary triggers for regional-scale logistics (longer trips) rather than just local neighborhood shopping (shorter trips). Although rickshaw vans handled daily local needs and smaller volumes, the larger motorized fleet, specifically heavy and medium trucks, was only fully utilized when the commercial volume generated on market days reached a peak. The reliance on UZRs further confirmed that these roads act as the essential corridors for all vehicles, though the impact was most pronounced for heavy trucks, confirming their role as the backbone of national/regional goods distribution. The nonlinear effects revealed in the models suggest that road attributes do not affect freight simply or gradually; instead, they operate on important threshold effects that vary by vehicle type. For instance, the significant nonlinearity for crest width across light commercial vehicles, heavy trucks, and small trucks suggests a physical unlocking point. A road may need to reach a specific width, likely broad enough for a truck to safely pass a nonmotorized vehicle, before a major shift from manual rickshaw vans to motorized trucks can occur. The embankment-height relationship reflected different needs: it was highly significant for light commercial vehicles and medium trucks, which often carry processed or perishable goods. In a country like Bangladesh, which is prone to flooding, once a road is built high enough to remain above the anticipated flood level, it becomes a reliable year-round asset for these commercial vehicles. With regard to road roughness (IRI), rickshaw vans were negatively affected by even minor increases in roughness, whereas trucks showed a more complex nonlinear relationship. This implies that although heavy trucks are forced to use rough roads owing to a lack of alternatives, there is a clear tipping point where the surface quality becomes too poor even for them, forcing a total stop or diversion to avoid vehicle damage and extra maintenance costs.

The way household size related to different vehicles provided a window into how the rural economy is changing at the family level. The finding that larger households were less likely to use rickshaw vans or light commercial vehicles but were significantly more likely to use heavy trucks suggests that these families often function as collective farming units or small enterprises. As a household grows, it likely moves away from supplying goods for small-scale, daily domestic shopping, which usually relies on rickshaws, and shifts toward industrial-scale agricultural production for export to larger cities. This bulk-buy and bulk-sell behavior is highly nonlinear for almost every vehicle type, meaning that transport choice changes drastically once a household reaches a certain size or has enough labor to produce at a commercial scale. Smaller households, by contrast, remained the primary users of rickshaw vans and light commercial vehicles for individual consumption and smaller-scale trade. Socioeconomic factors such as literacy and solvency also acted as vital catalysts, but they affected vehicle types differently. Higher solvency levels led to increased movement of medium and small trucks, likely because of the consumption of processed goods and modern building materials. Literacy rate had a strong positive association with rickshaw vans and heavy trucks, possibly suggesting that more educated rural residents are better integrated into both local retail markets and larger-scale commercial supply chains. These families were more likely to participate in modern, high-value industries like poultry or aquaculture, which require frequent and specialized motorized transport. The models showed that for all vehicle types, this shift was nonlinear, indicating a take-off point where a community’s combined wealth and education triggered a permanent move toward faster and more sophisticated transport modes.

Finally, the findings for the area variable suggested that freight thrives in compact clusters rather than spread-out regions. Light commercial vehicles were particularly sensitive to this, showing a negative relationship with larger areas, which implies they operate most efficiently in dense, localized environments. In the dense landscape of Bangladesh, logistics were most efficient for all motorized vehicles when households and markets were close together. Larger areas may hinder the logistic efficiency of trucks because of the long distances and lower volume of goods, which often forces reliance back onto localized rickshaw vans for basic transport. This suggests that the spatial concentration of activity is just as important as the quality of the roads themselves in determining which vehicles will be used.

Policy Applications and Implications

The insights derived from this study offer a strategic roadmap for LGED and other organizations to refine rural transport policy. By pivoting from uniform paving to targeted, threshold-based interventions, these recommendations address specific bottlenecks in crest width, embankment height, and maintenance prioritization. The following applications integrate infrastructural upgrades with regulatory and land-use strategies to modernize rural logistics and maximize economic returns on Bangladesh’s ongoing investments in rural education and infrastructure.

Critical Link Upgrade

The identification of specific threshold effects for road crest width and embankment height suggests that LGED should move away from uniform paving projects in favor of critical link upgrades, as crest width acts as an unlocking point for motorized trucks. Engineers/policy makers should prioritize widening specific bottleneck segments, such as narrow bridges or stretches passing through village markets, to a minimum standard that allows a medium truck to pass a rickshaw van without one vehicle leaving the pavement. This targeted widening would facilitate the bulk-buy and bulk-sell logistics of larger, more productive rural households.

Climate-Synchronized Design Standard

The nonlinear relationship between embankment height and light commercial vehicles requires a climate-synchronized design standard. Rather than a standard height for all roads, the modeling suggests that raising roads to a certain height above the highest local flood level would yield a disproportionate increase in year-round commercial vehicle reliability, specifically benefiting the delivery of perishable goods and e-commerce.

Last-Mile Surface Fund

Maintenance strategies should also be differentiated by vehicle sensitivity. Since rickshaw vans are disproportionately hindered by road roughness (IRI), a last-mile surface fund should be established to provide frequent bituminous carpeting or patching for minor UNRs that connect houses to markets. This would support the small-scale commerce of lower-income families.

Structural Axle-Load Capacity Focus

For the UZRs, where heavy trucks are concentrated, the focus must shift from surface smoothness to structural axle-load capacity. This would ensure that the backbone of the regional freight network does not suffer from structural failure during the peak harvest periods or during market-day surges.

Temporal Zoning of the Rural Road Network

On the regulatory side, the rhythmic pulse of market-day freight calls for temporal zoning of the rural road network. Local authorities could implement a freight priority window on market days, where heavy and medium trucks are encouraged to move during specific early-morning or late-evening hours to minimize conflict with the rickshaws and light commercial vehicles used for domestic shopping.

Village Freight Microhubs

To further support the rising literacy and solvency levels of rural residents who drive the demand for motorized trucks, the government should incentivize the creation of village freight consolidation points. These microhubs, located near dense household clusters, would allow for the grouping of small agricultural surpluses into single heavy-truck loads, reducing the number of half-full or empty trips and lowering overall transport costs for rural entrepreneurs.

Nucleated Market Hubs and Land-Use Policies

The consistently negative relationship between area and motorized freight movement, particularly for light commercial vehicles, suggests that expansive spatial extent acts as a barrier to logistic efficiency. In the context of rural Bangladesh, this finding implies that land-use policies should actively discourage ribbon developments—where residential and commercial structures are spread thinly along road stretches—because such patterns increase travel times and reduce the profitability of motorized transport. Instead, policy makers should promote nucleated market hubs that concentrate larger population clusters and commercial activities into smaller, manageable areas. By reducing the physical area over which goods must be distributed, the logistics network can better support the bulk-buy and bulk-sell behaviors of larger households and the high-value supply chain needs of more solvent and literate residents.

This strategy of spatial concentration, combined with the unlocking of infrastructure thresholds like crest width and embankment height, allows for a more cost-effective transition from slow, manual transport to a modern, motorized freight system that rewards a low-/medium-income country’s rural education and economic investment, which are basically the backbone of these types of economies.

Conclusion

The analysis of rural freight vehicle movements across northern Bangladesh demonstrates the complex interplay between roadway characteristics, socioeconomic characteristics, and temporal variations in travel demand. By employing both NB-GLMs and NB-GAMs, the study revealed that although certain correlates, such as market day effects and the importance of UZRs roads, are consistently significant across vehicle types, the strength and form of these relationships vary substantially. The GLM results highlighted a set of key linear associations, indicating that household size, area, literacy, and solvency systematically influence freight activity. However, the additive modeling provided deeper insights by capturing nonlinear dependencies, particularly for road geometry, pavement condition, and socioeconomic indicators, which could not be adequately represented by linear models. Taken together, these findings suggest that rural freight demand is not only a function of road network attributes but also shaped by underlying demographic and economic contexts. The presence of strong nonlinear effects emphasizes the necessity of flexible modeling frameworks to better reflect real-world transport dynamics. The outcomes contribute to a more nuanced understanding of rural freight flows in Bangladesh and point to the need for integrated planning approaches that address both physical infrastructure and socioeconomic development to sustain and enhance rural mobility.

Supplemental Material

sj-docx-1-trr-10.1177_03611981261454272 – Supplemental material for Factors Influencing Rural Freight Transport in Bangladesh: Insights from Generalized Linear and Additive Models

Supplemental material, sj-docx-1-trr-10.1177_03611981261454272 for Factors Influencing Rural Freight Transport in Bangladesh: Insights from Generalized Linear and Additive Models by Md. Amin Al Noor and Sadia Mehanaz in Transportation Research Record

Footnotes

Acknowledgements

The authors gratefully acknowledge the assistance of the Rural Connectivity Improvement Project of the Local Government Engineering Department of Bangladesh in providing the necessary data required to conduct this research.

Author Contributions

The authors confirm contribution to the paper as follows: study conception and design: Md. A. Al Noor; data collection: S. Mehanaz, Md. A. Al Noor; analysis and interpretation of results: Md. A. Al Noor, S. Mehanaz; draft manuscript preparation: Md. A. Al Noor, S. Mehanaz. All authors reviewed the results and approved the final version of the manuscript.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Md. Amin Al Noor

Supplemental Material

Supplemental material for this article is available online.

References

Wang

Nozick

Gearhart

Modeling Ocean, Rail, and Truck Transportation Flows to Support Policy Analysis. Maritime Economics & Logistics, Vol. 20, No. 3, 2018, pp. 327–357. https://doi.org/10.1057/s41278-018-0108-x.

Fry

Kanemoto

Fraser

Nansai

Global Freight Transport Emissions Responsibility. Environmental Science & Technology, Vol. 58, No. 43, 2024, pp. 19231–19242. https://doi.org/10.1021/acs.est.4c05658.

Jansuwan

Chen

Analysis of Freight Transportation Network Redundancy: An Application to Utah’s Bi-Modal Network for Transporting Coal. Transportation Research Part A: Policy and Practice, Vol. 151, 2021, pp. 154–171. https://doi.org/10.1016/j.tra.2021.06.019.

Gonzalez

J. N.

Camarero-Orive

González-Cancelas

Guzman

A. F.

Impact of the COVID-19 Pandemic on Road Freight Transportation – A Colombian Case Study. Research in Transportation Business & Management, Vol. 43, 2022, p. 100802. https://doi.org/10.1016/j.rtbm.2022.100802.

Wieberneit

Service Network Design for Freight Transportation: A Review. OR Spectrum, Vol. 30, No. 1, 2007, pp. 77–112. https://doi.org/10.1007/s00291-007-0079-2.

Lindsey

Mahmassani

H. S.

Mullarkey

Nash

Rothberg

Regional Logistics Hubs, Freight Activity and Industrial Space Demand: Econometric Analysis. Research in Transportation Business & Management, Vol. 11, 2014, pp. 98–104. https://doi.org/10.1016/j.rtbm.2014.06.002.

Macharis

Nocera

The Future of Freight Transport. European Transport Research Review, Vol. 11, No. 1, 2019, p. 21. https://doi.org/10.1186/s12544-019-0359-3.

Havenga

J. H.

Simpson

Z. P.

Freight Logistics’ Contribution to Sustainability: Systemic Measurement Facilitates Behavioural Change. Transportation Research Part D: Transport and Environment, Vol. 58, 2018, pp. 320–331. https://doi.org/10.1016/j.trd.2016.08.035.

Simpson

Z. P.

Havenga

J. H.

Witthöft

I. E.

Aritua

A Methodology for Disaggregated Freight Demand Modeling in Emerging Economies. In Freight Transport Modeling in Emerging Countries ( Kourounioti

Tavasszy

Friedrich

, eds.), Amsterdam, Elsevier, pp. 55–84.

10.

Kourounioti

Tavasszy

L. A.

Friedrich

Freight Transport Modeling in Emerging Countries. Amsterdam, Elsevier, 2021.

11.

Dhulipala

Patil

G. R.

Regional Freight Generation and Spatial Interactions in Developing Regions Using Secondary Data. Transportation, Vol. 50, No. 3, 2023, pp. 773–810. https://doi.org/10.1007/s11116-021-10261-w.

12.

Asomani-Boateng

Fricano

R. J.

Adarkwa

Assessing the Socio-Economic Impacts of Rural Road Improvements in Ghana: A Case Study of Transport Sector Program Support (II). Case Studies on Transport Policy, Vol. 3, No. 4, 2015, pp. 355–366. https://doi.org/10.1016/j.cstp.2015.04.006.

13.

Takada

Morikawa

Idei

Kato

Impacts of Improvements in Rural Roads on Household Income through the Enhancement of Market Accessibility in Rural Areas of Cambodia. Transportation, Vol. 48, No. 5, 2021, pp. 2857–2881. https://doi.org/10.1007/s11116-020-10150-8.

14.

Kaiser

Barstow

C. K.

Rural Transportation Infrastructure in Low- and Middle-Income Countries: A Review of Impacts, Implications, and Interventions. Sustainability, Vol. 14, No. 4, 2022, p. 2149. https://doi.org/10.3390/su14042149.

15.

Chen

Nanayakkara

Rural Road Connectivity and Local Economic Activity: Evidence from Sri Lanka’s IRoad Program. Transport Policy, Vol. 144, 2023, pp. 49–64. https://doi.org/10.1016/j.tranpol.2023.09.022.

16.

Jacoby

H. G.

Access to Markets and the Benefits of Rural Roads. The Economic Journal, Vol. 110, No. 465, 2000, pp. 713–737. https://doi.org/10.1111/1468-0297.00562.

17.

Heyns

A. M.

Banick

Optimisation of Rural Roads Planning Based on Multi-Modal Travel: A Multi-Service Accessibility Study in Nepal’s Remote Karnali Province. Transportation, Vol. 51, No. 2, 2024, pp. 567–613. https://doi.org/10.1007/s11116-022-10343-3.

18.

Keller

G. R.

Rural Roads Vulnerability Reduction Assessment, Mitigation Measures, and Training. Natural Hazards Review, Vol. 3, No. 4, 2002, pp. 139–147. https://doi.org/10.1061/(ASCE)1527-6988(2002)3:4(139).

19.

Jiao

How the Rural Infrastructure Construction Drives Rural Economic Development Through Rural Living Environment Governance—Case Study of 285 Cities in China. Frontiers in Environmental Science, Vol. 11, 2023, p. 1280744. https://doi.org/10.3389/fenvs.2023.1280744.

20.

Gao

Tong

Exploration of the Impact of Rural Road Maintenance and Management on Rural Economy Based on VAR Model. Applied Mathematics and Nonlinear Sciences, Vol. 9, No. 1, 2024, pp. 1–12. https://doi.org/10.2478/amns.2023.1.00211.

21.

Tawfik

Shah

Analysis of Freight Movements in the San Joaquin Valley. Mineta Transportation Institute, San Jose, CA, 2023.

22.

Zhao

Balusu

S. K.

Sheela

P. V.

Pinjari

A. R.

Eluru

Weight-Categorized Truck Flow Estimation: A Data-Fusion Approach and a Florida Case Study. Transportation Research Part E: Logistics and Transportation Review, Vol. 136, 2020, p. 101890. https://doi.org/10.1016/j.tre.2020.101890.

23.

Heyns

A. M.

Banick

24.

Yang

Chu

Wang

Feasibility on the Integration of Passenger and Freight Transportation in Rural Areas: A Service Mode and an Optimization Model. Socio-Economic Planning Sciences, Vol. 88, 2023, p. 101665. https://doi.org/10.1016/j.seps.2023.101665.

25.

Yasanthi

R. G. N.

Mehran

Alhajyaseen

W. K. M.

Modelling Speed Behaviour in Rural Highways: Safety Analysis of Driving Under Adverse Road-Weather Conditions. PLoS One, Vol. 16, No. 8, 2021, p. e0256322. https://doi.org/10.1371/journal.pone.0256322.

26.

Lee

Jeong

B. Y.

Comparisons of Traffic Collisions between Expressways and Rural Roads in Truck Drivers. Safety and Health at Work, Vol. 7, No. 1, 2016, pp. 38–42. https://doi.org/10.1016/j.shaw.2015.11.005.

27.

Biglari

Kofi Adanu

Jones

A Sequel to “Comprehensive Analysis of Single- and Multi-Vehicle Large Truck at-Fault Crashes on Rural and Urban Roadways in Alabama”: Accounting for Temporal Instability in Crash Factors. Accident Analysis & Prevention, Vol. 206, 2024, p. 107723. https://doi.org/10.1016/j.aap.2024.107723.

28.

Kalahasthi

Holguín-Veras

Yushimito

W. F.

A Freight Origin-Destination Synthesis Model with Mode Choice. Transportation Research Part E: Logistics and Transportation Review, Vol. 157, 2022, p. 102595. https://doi.org/10.1016/j.tre.2021.102595.

29.

Nasirullah

Mortuza

Md. Hossain

Cost-Benefit Analysis for Freight Transportation of Bangladesh. In 4th Annual Paper Meet and 1st Civil Engineering Congress, Dhaka, Bangladesh, 2011, pp. 145–151.

30.

Holguín-Veras

Kalahasthi

L. K.

Ismael

Yushimito

W. F.

Herrera-Dappe

Hoque

Md. . Integrated Data Collection and Modeling with Freight Origin–Destination Synthesis: Application to Bangladesh. Case Studies on Transport Policy, Vol. 20, 2025, p. 101456. https://doi.org/10.1016/j.cstp.2025.101456.

31.

Kalahasthi

L. K.

Sutar

Yushimito

W. F.

Holguín-Veras

Optimal Sampling Plan for Freight Demand Synthesis with Mode Choice: A Case Study of Bangladesh. Transportation Research Record: Journal of the Transportation Research Board, 2023. 2677(12): 718–730. https://doi.org/10.1177/03611981231169527.

32.

Mahdavian

Shojaei

Salem

Laman

Eluru

Oloufa

A. A.

A Universal Automated Data-Driven Modeling Framework for Truck Traffic Volume Prediction. IEEE Access, Vol. 9, 2021, pp. 105341–105356. https://doi.org/10.1109/ACCESS.2021.3099029.

33.

Middela

M. S.

Ramadurai

Incorporating Spatial Interactions in Zero-Inflated Negative Binomial Models for Freight Trip Generation. Transportation, Vol. 48, No. 5, 2021, pp. 2335–2356. https://doi.org/10.1007/s11116-020-10132-w.

34.

Song

Wang

Wright

Thatcher

Felix

Traffic Volume Prediction with Segment-Based Regression Kriging and Its Implementation in Assessing the Impact of Heavy Vehicles. IEEE Transactions on Intelligent Transportation Systems, Vol. 20, No. 1, 2019, pp. 232–243. https://doi.org/10.1109/TITS.2018.2805817.

35.

Xiao

Yang

Bao

Xia

Exploring the Spatiotemporal Characteristics of Freight Transportation Using Truck GPS Data: A Case Study of Lianyungang City. IEEE Access, Vol. 12, 2024, pp. 181899–181906. https://doi.org/10.1109/ACCESS.2024.3504480.

36.

Kawasaki

Namba

Oka

Dulebenets

M. A.

Freight Trip Distribution Using Spatiotemporal Aggregate Data: A Modified Collective Flow Diffusion Model-Based Approach. Transportation Research Interdisciplinary Perspectives, Vol. 21, 2023, p. 100904. https://doi.org/10.1016/j.trip.2023.100904.

37.

Basile

Benfratello

Castellani

Geoadditive Models for Regional Count Data: An Application to Industrial Location. Geographical Analysis, Vol. 45, No. 1, 2013, pp. 28–48. https://doi.org/10.1111/gean.12001.

38.

Narayanan

Varier

Bhupathi

Simhadri Kavali

Mohana Ramakanth Kumar

Sreelakshmi

Vehicle Turn Pattern Counting and Short Term Forecasting Using Deep Learning for Urban Traffic Management System. IEEE Access, Vol. 13, 2025, pp. 8585–8593. https://doi.org/10.1109/ACCESS.2025.3526880.

39.

Mandal

Mussah

A. R.

Jin

Adu-Gyamfi

Artificial Intelligence-Enabled Traffic Monitoring System. Sustainability, Vol. 12, No. 21, 2020, p. 9177. https://doi.org/10.3390/su12219177.

40.

Kutlimuratov

Khamzaev

Kuchkorov

Anwar

M. S.

Choi

Applying Enhanced Real-Time Monitoring and Counting Method for Effective Traffic Management in Tashkent. Sensors, Vol. 23, No. 11, 2023, p. 5007. https://doi.org/10.3390/s23115007.

41.

Aljamal

M. A.

Abdelghaffar

H. M.

Rakha

H. A.

Real-Time Estimation of Vehicle Counts on Signalized Intersection Approaches Using Probe Vehicle Data. IEEE Transactions on Intelligent Transportation Systems, Vol. 22, No. 5, 2021, pp. 2719–2729. https://doi.org/10.1109/TITS.2020.2973954.

42.

Bangladesh National Portal. Union-List. https://www.bangladesh.gov.bd/site/view/union-list/. Accessed June 24, 2025.

43.

LGED. Administrative Area of LGED. https://proinfo.lged.gov.bd/lged/. Accessed June 27, 2025.

44.

Ahmed

Muir

J. F.

Garnett

S. T.

Bangladesh Needs a “Blue–Green Revolution” to Achieve a Green Economy. AMBIO, Vol. 41, No. 2, 2012, pp. 211–215. https://doi.org/10.1007/s13280-011-0160-6.

45.

Chen

L. C.

Chaudhury

R. H.

Demographic Change and Food Production in Bangladesh, 1960-74. Population and Development Review, Vol. 1, No. 2, 1975, p. 201. https://doi.org/10.2307/1972221.

46.

Huq

M. T.

Ichihashi

Prospective Accelerating Sectors to Attain Sustainable Development in Bangladesh Economy: Findings from a Sectoral Approach Using Input-Output Analysis. Sustainability, Vol. 15, No. 3, 2023, p. 2651. https://doi.org/10.3390/su15032651.

47.

Quddus

Kropp

J. D.

Constraints to Agricultural Production and Marketing in the Lagging Regions of Bangladesh. Sustainability, Vol. 12, No. 10, 2020, p. 3956. https://doi.org/10.3390/su12103956.

48.

Bangladesh Bureau of Statistics, World Bank, and World Food Programme. Poverty Maps of Bangladesh 2010: Key Findings. World Bank Group, Washington, D.C., 2014. https://doi.org/10.1596/20425.

49.

Yasmin

Akter

The Long Run Association Between Transport and Economic Growth in Bangladesh: An Econometric Analysis. International Journal for Multidisciplinary Research, Vol. 6, No. 1, 2024. https://doi.org/10.36948/ijfmr.2024.v06i01.11713.

50.

Bari

M. M.

Sustainable Transport Developments for Dhaka City. In Urban Transition – Perspectives on Urban Systems and Environments ( Wallhagen

Cehlin

, eds.), IntechOpen, London, 2023. https://doi.org/10.5772/intechopen.111857

51.

Wikipedia. Rangpur Division. https://en.wikipedia.org/wiki/Rangpur_Division. Accessed March 9, 2026.

52.

Bangladesh Bureau of Statistics. Latest Publications. https://bbs.gov.bd/. Accessed June 20, 2025.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.31 MB

0.00 MB