Modelling of Active and Latent Attributes Based on Traveler Perspectives: Case of Port City of Douala

Anastasia Ojong Maayuk-Okpok, Yin Ming

College of Transport and Communication, Shanghai Maritime University, Shanghai, China.

DOI: 10.4236/wjet.2023.111012                

Abstract

A growing stream of study stresses the relevance of subjective elements in understanding the hierarchy of preferences that underpin individual travel behavior. The purpose of this study is to evaluate the impact of various factors on mode choice. To achieve this, a multinomial logit model (MNL) was used to analyze the relationships between mode choice and three classes of attributes; Combined Active and Latent, Active only and Latent only attributes. The data used are derived from surveys in the port city of Douala, Cameroon as a case study. Results stipulated that, the combined attributes model performed better than both active only attributes and latent only attributes models. Likewise, latent only attributes model performed better than active only attributes model. The advantage of modelling all three groups is for better selection of the most relevant attributes, and this is very relevant in understanding travel behavior of individuals and mode choice decisions.

Keywords

Multinomial logit Model, Latent Attributes, Mode Choice, Individual Behavior, Active Attributes

1. Introduction

Mode choice is very significant in transport planning, and it closely deals with choice behavior which impacts policy making directly [1] . Customarily, transportation parameters and socioeconomic characteristics of service users have been used to model and analyze the choice of travel mode [2] . The costs, destinations, capacities, frequency, and other attributes of the modes, as well as the nature of the traveler (in the case of passenger transport) and their destinations, may influence mode selection [3] . Transportation modeling is used to assess the effects of behavioral changes as well as the consequences of infrastructure investments [4] [5] . The tools available are becoming more complicated, and there are an increasing number of criteria, aspects, and stakeholders to consider [6] . Traditional modeling techniques have been supplemented with activity based models, which are used to analyze traffic impacts and travel behavior, with the selection of the most relevant modeling parameters being a major challenge [7] . A growing stream of study stressed the relevance of subjective elements in understanding the hierarchy of preferences that underpin individual travel behavior throughout the last decade. It backs up the idea that attitudes and behaviors might play a role in mode selection. These subjective or latent elements, on the other hand, cannot be directly observed; they can be extracted from other observable variables, such as replies to survey questions regarding attitudes, perceptions, or decisions. Most of the existing study on mode choice dwells on active variables and their effects on choice behavior but these active variables are more tilted in the econometric analysis of utility and mode choice and due to their availability and predictive nature they are very easy to model especially in transport systems where the service level is high and well developed. Some recent studies have looked at the latent variables and their effects on mode choice behavior and expressed the challenges involved in acquiring data with this information. Notwithstanding, they also clearly showed the extent to which it is important to include these variables directly and not just assume them to be represented or included in the error term as unobserved attributes. The downside is that most of these existing works used the latent variables for single trip purposes and due to the fact that the determinants of mode choice are found to differ across trip purposes, it is not very valid to generalize the results from studies considering only single purpose. Therefore, this study presents a clear case, incorporating the important attributes as depicted by the respondents of the study area in a very unpredictive transport system which is still being developed keeping in mind the importance of including individual perceptions in transport planning and policy making decisions for better service provision. In their study [8] , they investigated the idea of user perception of safety, comfort, and accessibility and how these may have effects on transport mode choice. The study [9] , supports the idea that individual latent preferences do play a significant role in mode choice behavior in urban transport. They discovered that, in addition to standard cost and benefit factors, individual beliefs and qualities about some aspects of transportation, such as flexibility, comfort, safety, and symbolic-affective nature, influence urban travel behavior [9] . Their study was limited to work trips. Furthermore, age, gender, employment status, and the number of young children have all been found to be significant explanatory factors in the psychological profiles of respondents for both studies.

In this study, we plan to find out the factors that influence the travel behavior for the respondents in the city of Douala, Cameroon. Especially, at this period where the population is increasing rapidly with influx from neighboring cities dealing with civil unrest. According to recent knowledge, the traffic congestion is very great even with the opening and building of new transport infrastructures like that of the bridge of Bonaberi which has been known to be one of the biggest bottlenecks. It is believed that with the comprehensive implementation of various policies, laws, and regulations, as well as econometric modeling, traffic demand management aims to achieve a stable and healthy balance of the traffic system by providing guidance and limits of travel behavior, adjusting residents' trip distribution, and alleviating the contradiction between supply and demand of traffic in order to achieve the set target of traffic system operation efficiency improvement, congestion ease, and pollution reduction [10] . This study takes particular interest in understanding the influence and extent of latent and active attributes for the different choice of transport modes. That is, why some individuals prefer some modes over others. This study looks at attributes in three different categories; first, the active attributes which are characteristic of the individual’s socioeconomic life and household; second, the latent attributes of the modes as perceived by the individual and third, a combination of both active and latent attributes. We belief that transport planners need this information to better incorporate traveler’s choice behavior for a more accurate policy and public transport service provision. It is not always about building new infrastructures, sometimes, all it takes is to include user’s perception on qualitative dimensions to improve performance of public transport services and the entire transport system [11] .

The main objective of the study is to investigate the role of active and latent attributes of respondents based on the data collected using stated and revealed preference (SP/RP) survey. The data is analyzed using multinomial modelling and attribute fitting analysis in R. The reason for the attribute fitting analysis is to show the inferred impact of latent, active and a combination of both on traveler’s behavior in mode choice for the respondents in this study area. We discovered that, sometimes even the poor will disregard cost in comparison to other attributes for better satisfaction. Due to the time in which this study was done, the data collected was very limited as the Covid-19 pandemic is still in action. Also, getting personal information from respondents was not very easy as there is a lot of crises in Cameroon. So, the data had a lot of missing information to be considered representative. In future, a more representative and complete analysis can be done with a larger sample size.

The next part of this paper describes the study area and collected data of the respondents. Section 3 gives the mode choice model specification, while section 4 gives the model estimations and the results. In the last section, we discuss and draw conclusions and make recommendations for future research developments.

2. Study Area and Data Analysis

2.1. Study Area

The area considered for this paper is Douala, the economic capital and chief port of Cameroon. It is situated on the southeastern shore of the Wouri River estuary, on the Atlantic Ocean coast about 130 miles (210 km) west of Yaoundé [12] . Douala, being a port and a natural gateway for the entry of imports and the exit of exports, is home to most Cameroon’s industrial and service operations, accounting for more than half of the country’s economic activity and industrial production. The geographical situation, on the other hand, is adverse. The city deals with a natural environment that is severely limited. In swampy places, as well as on the slopes of streams and natural drainage basins, several unplanned neighborhoods have sprung [13] . For purpose of this study, the area will be divided into 10 main zones (See Figure 1).

2.1.1. Institutional Framework of Urban Transport in Douala—Multiple Players and Minimal Coordination

Five urban and one rural arrondissement make up the city of Douala. A municipality that consists of an elected mayor and a municipal council oversees each arrondissement [13] . The Douala Urban Community is governed by an appointed government representative and is made up of multiple representatives chosen by the arrondissements. The municipal council of the Community is made up of several officials elected by the arrondissements. At the municipal level, the Urban Community has control over issues such as parking, main roads, signal maintenance, urban planning, and urban development. Even though the limits defining who is responsible for what are not often obvious, the Urban Community shares its authority with the central agencies and their local delegations. The Public Works Ministry oversees administering the renovation of the bridge over the Wouri River, and the City Ministry oversees the creation of the future Urban Development Master Plan, while the Transport Ministry gives licenses for transportation. The Urban Community has no authority to intervene in the arrondissements’ issuance of several transport licenses.

2.1.2. Road System and Public Transport Supply

The road infrastructure struggles to keep up with urbanization in Douala, as it does in most Sub-Saharan cities. Paved roadways are primarily found in the city’s center. The Wouri river, the Bessengué rail station, the Bassa industrial

Figure 1. Map of survey locations for study area.

park, and the former airport serve as obstacles around the city’s core. Access to the city’s core for vehicles going from Bonaberi or the east side mostly follows four routes: the Wouri Bridge from north to south, the Ndokoti, North Akwa Road, and the main highway (Axe Lourd). Notwithstanding, the supply of public transport is very diversified ranging from taxis, bendskins (motorbike taxis), SOCATUR buses, cargos (minibuses and light trucks), and unregistered cabs. In this paper we refer to taxis (commercial cars), bendskins (commercial bikes), SOCATUR buses (Public bus) and every other fall in the “others” category. The commercial cars form part of the declining transport modes in this area as the cars are most often very old and not very convenient as they are often overloaded and carry several people going to different locations in same area, but their prices are relatively stable and determined by their union. The commercial bikes on the other hand are the booming transport mode in this area as they are easily accessible even though they are not very safe especially on main routes, their prices are also very expensive. The public buses are limited in supply, no definite schedule and not very accessible but it is the cheapest mode of transport in this area. Some individuals prefer walking, but that too is not very safe as there are no designated walkways. Residents of remote neighborhoods and the city’s outermost regions have a harder time getting access to the transportation system. The access conditions for the poor in these locations are marginally worse than for the non-poor [13] . Figure 2 and Figure 3 present the Mode share of the study area for the respondents of this study based on their origin and destination respectively. The two figures for the origin and destination mode shares show that commercial bikes top the chart of usage followed by commercial cars then public bus and others. It is clearly seen that the commercial bikes access more locations than commercial cars and public bus. The major reason for this is the fact that accessible roads into streets are limited and some are in terrible conditions leaving the people with more of commercial bikes than commercial cars and public bus.

Figure 2. Mode share of study area based on origin of respondents.

Figure 3. Mode share of study area based on destination of respondents.

2.2. Data Analysis

This study uses both quantitative and qualitative analysis methods as well as primary and secondary data obtained by use of Stated Preference/Revealed Preference survey techniques and interviews where necessary. The collected data has information on the individuals, their household characteristics, daily trip characteristics, their perceived perspective on the various mode of transport at their disposal, and so on.

2.2.1. Socio-Economic Characteristics

The socio-economic characteristics of the respondents in this study are presented in Table 1. Table 1 shows the characteristics of the respondents in the study area. From the table, it is possible to get the percentage of female respondents to male as 55% and 45% respectively.

2.2.2. Trip Distribution by Mode

Figure 4 presents the mode choice distribution for the respondents. It shows the % of each mode option based on the survey response. The commercial bike % stands at 44.96%, which is the highest and is followed by commercial cars, with 29.46%; the public bus % of 10.08% is the lowest of all the modes.

2.3. Household Characteristics

The key characteristics of households include household size, household income, the number of cars owned by each household, the number of workers, etc. these characteristics are those observed to impact mode choice behavior most.

2.3.1. Household Size

As the size of the household increases, so does the frequency of trips [5] . The breakdown of the household structure for the chosen dataset is provided in Figure 5. In the data acquired one household declared size as 6 and that was left out as an outlier. From the figure below, household size of 3 constitutes the largest share with a 36% as it is the densest including respondents from all study zones

Table 1. Socio-economic characteristics of respondents.

Figure 4. Mode choice distribution for respondents.

in the area. Similarly, household size of 4 with 29% followed by household size of 2 with 26%. The single individual households and those of 5 or more constitute less than 10%.

Figure 5. Household size distribution in study area.

2.3.2. Household Income

Due to the insecurity in the country, it was a bit difficult getting information about the household income of the respondents. So, it was easier getting the ranges for income and from the ranges average values were obtained to ease the process. According to [14] , the legal minimum wage a person should earn in Cameroon is 36,270 CFA francs per month, but a lot of people still earn far less. Douala being the economic capital and the major port city in Cameroon makes it possible for most to be self-employed which enables them to earn more. From Table 2, it is seen that 25% of the respondents earn below 53,000 Fcfa average income while 75% earn from 88,000 Fcfa and above. This is clearly shown in Figure 6.

Table 3 gives the distribution of car ownership of respondents with regards to average household income. In the city center, houses are more expensive than in the outskirts [15] . Most people prefer to stay in cheaper areas to avoid the high pricing of houses and this weighs more on transportation mode and cost. The reason why in Douala there is no free ride, most car owners offer rides to individuals going their same direction at a cost to use their earnings for fuel and other car charges.

2.3.3. Individual Characteristics and Mode Choice

The individual characteristics such as age, gender and so on contribute also to mode choice behavior. The age group of respondents and mode choice distribution can be seen in Figure 7. The 18 - 25 years age group dominates the commercial bikes and commercial car, while the 36 - 45 years age group dominates other modes (own cars, walking, own bikes, etc.). The public bus mode is dominated by the 26 - 35 years group and avoided by the 36 - 45 years group.

In this study, the percentage of female and male respondents is 55% and 45% respectively. Mode choice with regards to gender of respondents is shown in Figure 8. It is seen that female respondents use commercial bikes, commercial cars, and other modes more than male respondents and the male respondents slightly use more of public bus than the female respondents.

Table 2. Average household income (Fcfa) for the respondents.

Table 3. Car ownership with average household income (Fcfa) of respondents.

Figure 6. Average household income (Fcfa) for respondents in study area.

2.4. Latent Attributes of the Different Modes of Transport as Perceived by Respondents

For this study, we chose the Level of service (LOS), Availability (AV), Safety (S), Affordability (AF), Accessibility (AC), Flexibility (F) as the latent attributes for the model and the respondents gave their perceptions on the three main modes (commercial bike (CB), commercial car (CC), public bus (PB)) for this study. Figure 9 presents the visual perception of the respondents on a scale of 1 - 5

Figure 7. Age group of respondents and mode choice distribution.

Figure 8. Gender of respondents and mode choice distribution.

Figure 9. Respondents’ perception on the selected latent attributes of the different modes.

with 5 being the highest and 1 the lowest of the latent attributes. The affordability of public bus (AFPB) and the safety of the public bus (SPB) tops the chart on the level 5, meaning the public bus is perceived by the respondents to be most affordable and safe but it is the least used mode as per Figure 4. The availability of commercial bike (AVCB) and accessibility of commercial bike (ACCB) tops the chart on the level 5, meaning the respondents perceive the commercial bikes to be the most available and accessible mode of transport and this is backed in Figure 4. Most of the respondents rate the level of the attributes with regards to commercial car as within level 3 and 4. Accessibility of public bus (ACPB) tops the chart on level 1. Seemingly, Safety of commercial bikes (SCB) is on the high side in level 1 and most of the respondents rated affordability of commercial bikes from level 3 down to level 1, but why is it that commercial bikes are still the most used mode is the reason we want to model these attributes to be able to answer that.

3. Multinomial Mode Choice Modelling

Mode choice modelling has been seen to be very significant in any transport system for many years now with changes either on the data being analyzed, the contributing attributes or the method used for analysis. This study aims to investigate the effects of both latent and active attributes in the mode choice behavior of the respondents in the study area considered. It models active attributes, latent attributes, and a combination of both active and latent attributes to see the degree of impact of these on mode choice decision making using the backward selection method.

3.1. Model Formulation

The basic assumption here is that not only traditional and objective attributes like travel time, travel cost, distance, income, or household size have effects on mode choice, but some subjective or latent attributes like safety, flexibility, comfort, and so on, also have effects on choice of mode. Since information about these latent attributes is difficult to get, they are most often left out or compensated for by person specific factors in traditional choice models. But, by including preference variables directly into choice models, these compensations can be improved [16] . The utility of the individual is typically expressed as a linear function of the trip’s features weighted by coefficients that aim to capture the relative importance of those attributes as experienced by the individual. A possible representation of a utility function of a mode m, mathematically, can be shown in Equation (1).

$U_{mi}={\beta }_1{X}_{mi1}+{\beta }_2{X}_{mi2}+\cdots +{\beta }_k{X}_{mik}$ (1)

where, $U_{mi}$ is the net utility function of mode m for individual i; $X_{mi1},\cdots ,X_{mik}$ are k number of attributes of mode m for individual i; ${\beta }_1,\cdots ,{\beta }_k$ , are k number of coefficients or weights for each attribute. Since the behavior of the decision maker cannot be determined with certainty, an error component is included in the model to represent the unrepresented or unobserved components of utility. Therefore,

$U_{mi}=V_{mi}+{\epsilon }_{mi}$ (2)

where, $V_{mi}$ is the observed component of utility of mode m for individual i; ${\epsilon }_{mi}$ is the unobserved (error) component of utility of mode m for individual i. It is possible for the observed utility components to be viewed as a function of the characteristics for the available modes only if the decision maker’s attributes are disregarded. Consequently, it might be possible to employ a single utility function for everyone. For the same reason, it is also possible to consider the error component of the utility to be independent of socioeconomic traits. If the error component has zero mean and an extreme value distribution, the net utility function can be given as:

$U_m=V_m+{\epsilon }_m$ (3)

Therefore, if there are “n” alternative modes available, the probability of an individual selecting mode m, such that $m\in n$ , is based on its associated utility function $U_m$ , such that.

$U_m\ge U_n$ (4)

where, $U_m$ represents utility of mode alternative m; and $U_n$ represents utility of any mode alternative in the set of available modes. Briefly put, a person chooses the option that has the highest utility as shown in Equation (4). However, it is difficult to fully comprehend how different factors influence a person’s decision-making. As seen in Equation (3), this is resolved by include the unobserved components in the error term and combining with the observed components. The mathematical structure of a discrete choice model is determined by the assumptions made for the error components of the utility function for each alternative. The Multinomial Logit Model (MNL) is based on the following specific assumptions:

• The error components are extreme value (or Gumbel) distributed,

• The error components are identically and independently distributed across alternatives,

• The error components are identically and independently distributed across individuals [17] .

The MNL has a simple closed-form mathematical structure because of the three assumptions [17] described above. However, these assumptions leave the MNL model with the IIA property at the individual level, which is the model’s worst flaw [18] . The MNL gives the choice probabilities of each alternative as a function of the systematic portion of the utility of all the alternatives. The general Equation (5) for the probability of choosing an alternative “m” ( $m=1,2,\cdots ,n$ ) from a set of n alternatives is:

$Pr\left(mi\right)=\frac{\exp \left(V_{mi}\right)}{{\displaystyle {\sum }_{m=1}^n\exp \left(V_{mi}\right)}}$ (5)

where, $Pr\left(mi\right)$ is probability of utility for a mode choice (m) by individual (i); $V_{mi}$ is the utility of individual (i) choosing mode (m), $V_n$ is the systematic component of the utility of the set alternative (n). The MNL describes the relationships between the independent and dependent variables and expresses these relationships in terms of utility.

3.2. Model Specification

Household information from the main data gathered from respondents in the study region is coded and used as variables in the model. Table 4 gives the description of the attributes/variables used in the model. Two broad classes of variables; active/objective and latent/subjective variables sub-divided into; trip characteristics, household characteristics and mode characteristics (gotten from individual perceptions using survey data). The model’s chosen variables are derived from prior theoretical and empirical research on mode choice model analysis carried out by other scholars. As a result, the final variable definition based on statistical testing is reached here.

4. Model Estimation

The attributes described in Table 4 were considered for the formulation of the utility function of commercial bikes, commercial cars, and public bus. The total variables were used in the first case scenario and the backward selection approach was used to investigate the active only scenario and the latent only scenario for the study area being considered. Running the MNL model in R, the utilities are given as a logarithmic function of the reference level. In this case, 1 (commercial bikes) is used as the reference level. Table 5 and Table 6 shows the coefficient and Standard errors of both active and latent variables in the model respectively.

Using the coefficients in Table 5, we can write the 2 utility equations following from Equation (1).

$\begin{array}{c}\ln \left(\frac{P\left(2\right)}{P\left(1\right)}\right)=9.417239+\left(-0.02287\ast \text{ATT}\right)+0.001521\ast \text{ATC}\\ +\cdots +0.365315\text{AVPB}\end{array}$ (6)

$\begin{array}{c}\ln \left(\frac{P\left(3\right)}{P\left(1\right)}\right)=11.504+\left(-0.0708\ast \text{ATT}\right)+0.00o409\ast \text{ATC}\\ +\cdots +0.304697\text{AVPB}\end{array}$ (7)

Equation (6) is logarithm of the probability that mode 2 (commercial car) is chosen versus the probability that mode 1 (commercial bike) is chosen. That is the log-odds of mode 2 chosen versus mode 1 chosen. While Equation (7) gives the logarithm of the probability that mode 3 (public bus) is chosen versus the probability that mode 1 (commercial bike) is chosen. Literally, the log-odds of mode 3 chosen versus mode 1chosen. Looking at the coefficients of the variable, those with positive values have positive impact on th log-odds and those with

Table 4. Description of attributes/variables used in model.

negative values have negative impacts on the log-odds. For example, ATT, AHI, SCC, AFCB, FCC, FPB, ACCB, ACPB, AVCB and AVCC have negative impacts on both log-odds; ACCC has negative impact on the log-odds of equation 6 and

Table 5. Coefficients of the variables in the combined model.

Table 6. Standard errors of the variables in the combined model.

a positive impact on the log-odds in Equation (7). The remaining variables, ATC, NOC, HS, SCB, SPB, AFCC, AFPB, FCB and AVPB have positive impacts on both log-odds.

We can derive the probabilities from Equation (6) and Equation (7) as follows. Let Y₁ and Y₂ be the utility values obtained from Equation (6) and Equation (7) respectively.

$Y_1=\ln \left(\frac{P\left(2\right)}{P\left(1\right)}\right)\omega \to \frac{P\left(2\right)}{P\left(1\right)}={\text{e}}^{Y_1}$ (8)

$Y_2=\ln \left(\frac{P\left(3\right)}{P\left(1\right)}\right)\to \frac{P\left(3\right)}{P\left(1\right)}={\text{e}}^{Y_2}$ (9)

$\frac{P\left(2\right)+P\left(3\right)}{P\left(1\right)}={\text{e}}^{Y_1}+{\text{e}}^{Y_2}$ (10)

$\frac{P\left(1\right)}{P\left(1\right)}={\text{e}}^{Y_1}+{\text{e}}^{Y_2}$ (11)

Table 7. Probabilities of the observations in the combined model.

$\frac{1}{P\left(1\right)}=1+{\text{e}}^{Y_1}+{\text{e}}^{Y_2}$ (12)

Since $P\left(1\right)+P\left(2\right)+P\left(3\right)=1$ , it implies that

$P\left(1\right)=\frac{1}{1+{\text{e}}^{Y_1}+{\text{e}}^{Y_2}}$ (13)

$P\left(2\right)=\frac{{\text{e}}^{Y_1}}{1+{\text{e}}^{Y_1}+{\text{e}}^{Y_2}}$ (14)

$P\left(3\right)=\frac{{\text{e}}^{Y_2}}{1+{\text{e}}^{Y_1}+{\text{e}}^{Y_2}}$ (15)

4.1. Misclassification of the Combined Model

The probabilities of the combined model are given in Table 7. These probabilities are used to compare the model prediction to the data information, and the result is the misclassification matrix as seen in Table 8. The diagonals (14, 30, 49) gives the number of times the model classified correctly, and the rest of the value shows the misclassification.

The percentage in which there is misclassification between model and data is given by this formular.

$C=\left(1-\frac{{\displaystyle {\sum }_{i=1}^{3}d{m}_i}}{{\displaystyle {\sum }_{j=1}^9{m}_j}}\right)\times 100$ (16)

where, C is the percentage of misclassification between the model and data; $d{m}_i$ is the diagonal values for the misclassification matrix; $m_j$ is the misclassification matrix. For this model the value obtained was 21.85%. This means that the model was more right than wrong in predicting the data.

4.2. Two Tailed Z-Test to Check Model Fitting

A two tailed Z-test was performed to check for model fitting.

Looking at Table 9, with the P level set at 0.5, all the selected attributes fit for this data with only the second value for ATC (0.74021) exceeding the 0.5 level.

Table 8. Misclassification/confusion matrix table for the combined model.

Table 9. P-Values of both active and latent attributes in the combined model.

Table 10. Residual deviance and % misclassification for all three models.

The results of the Active Attributes Model and the Latent Attributes Model are shown in the appendix. Comparing the Residual Deviance and the percentage of misclassification of all three models shows that; the combined model has a lower misclassification percentage of 21.85% followed by the Latent attribute model with a percentage of 35.3% and lastly, the active attribute model of 39.5%. The combined model also has the lowest residual deviance of 141.9737 while the active attribute model has residual deviance of 196.4175, the latent attribute model has residual deviance of 174.0741; these are all seen in the Table 10.

5. Conclusion

The purpose of this study is to evaluate the impact of various factors on mode selection. To achieve this, we used a multinomial logit model (MNL) to analyze the relationships between mode choices of the individuals and three classes of attributes; Combined Active and Latent, Active only and Latent only attributes. The information about the characteristics of the individual, household, the trip and perceptions on level of service of the modes by the individuals, derived from surveys conducted in the study area. Tests were conducted using modeling approaches to demonstrate that the models statistically fit the data. Additionally, review and interpretation of the variable estimations were done. From the results obtained, it was realized that some variables perform better when combined with others as the results of the combined model show a better fit than the Active only and latent only models. Though it is quite difficult to obtain information on latent variables, it is very important as they relate more information on individual perceptions and based on this case study data, the latent only model proved to be fitter than the active only and it predicted the data better with a misclassification % of 35.3% while the active only had a 39.5% of misclassification.

6. Recommendation

The result of this research goes a long way to explain why individuals in this study area prefer commercial bikes to public bus, which is cheaper and safer but least accessible, least available. It also gives insight to why many individuals prefer to own a car and even sometimes use the car for offering paid rides to others in this area as the cheap modes are not very accessible and the accessible modes are not very cheap and less safe. This information can aid traffic demand management in better improving public bus systems to reduce traffic congestion and achieve a stable and healthy balance in the traffic system. Building new transport infrastructures is not the answer to all traffic congestion problems.

The relevance of this study adds to the stress on the relevance of subjective/latent attributes in understanding hierarchy of preferences that underpin individual travel behavior and further research can be done in identifying and selection of most relevant modelling parameters. The main limitation of this study was getting a representative dataset for the study since the process was disturbed by the civil unrest and Covid-19 pandemic. Future research will be helpful in further identifying numerous variables considering the effect on mode choice through expanded analysis of a big dataset. This will also allow for more recent techniques of modelling to be used. Finally, this research results in the creation of a mode demand model prototype that is based on microsimulation.

Acknowledgements

The authors thank the inhabitants of the city of Douala for their participation in the data collection process, Shanghai Maritime University for the software resources and all who supported morally.

Ethical Approval

The manuscript includes a survey study of inhabitants in the port city of Douala from 18 years and above of age. It, therefore, follows all ethical norms and no minors involved.

Appendices

Appendix A. Probabilities of the Observations for the Combined Model

Appendix B. Coefficients of Active Attribute Model

Appendix C. Standard Errors of Active Attribute Model

Appendix D. Misclassification/Confusion Matrix Table for the Active Attribute Model

Appendix E. P-Values of Active Attributes Model

Appendix F. Probabilities of the Observations for the Active Attributes Model

Appendix G. Coefficients of Latent Attribute Model

Appendix H. Standard Errors of Latent Attribute Model

Appendix I. Probabilities of the Observations for the Latent Attributes Model

Appendix J. Misclassification/Confusion Matrix Table for the Latent Attribute Model

Appendix K. P-Values of Latent Attributes Model

Conflicts of Interest

The authors declare no conflicts of interest regarding the publication of this paper.

References

[1]	Seyedabrishami, S. and Shafahi, Y. (2013) A Joint Model of Destination and Mode Choice for Urban Trips: A Disaggregate Approach. Transportation Planning and Technology, 36, 703-721. https://doi.org/10.1080/03081060.2013.851507
[2]	Amoroso, F., et al. (2010) A Demand-Based Methodology for Planning the Bus Network of a Small or Medium Town. European Transport, 44, 41-56.
[3]	Teravaninthorn, S. and Raballand, G. (2008) Transport Prices and Costs in Africa: A Review of the Main International Corridors. World Bank, Washington DC.
[4]	Ben-Akiva, M. and Boccara, B. (1995) Discrete Choice Models with Latent Choice Sets. International Journal of Research in Marketing, 12, 9-24. https://doi.org/10.1016/0167-8116(95)00002-J
[5]	Khan, O. (2007) Modelling Passenger Mode Choice Behavior Using Computer Aided Stated Preference Data. PhD Thesis, Queensland University of Technology, Brisbane.
[6]	Torok, B., et al. (2020) Representing Autonomated Vehicles in a Macroscopic Transportation Model. Periodica Polytechnica Transportation Engineering, 48, 269-275. https://doi.org/10.3311/PPtr.13989
[7]	Bergantino, S., et al. (2013) Taste Heterogeneity and Latent Preferences in the Choice Behavior of Freight Transport Operators. Transport Policy, 30, 77-91. https://doi.org/10.1016/j.tranpol.2013.08.002
[8]	Yanez, D., et al. (2010) Inclusion of Latent Variables in Mixed Logit Models: Modelling and Forecasting. Transportation Research Part A, 44, 744-753. https://doi.org/10.1016/j.tra.2010.07.007
[9]	Bergantino, A.S. and Catalano, M. (2016) Individual’s Psychological Traits and Urban Travel Behaviour. International Journal of Transport Economics, 43, 341-359.
[10]	Wang, F. and Ross, C. (2018) Machine Learning Travel Mode Choices: Comparing the Performance of an Extreme Gradient Boosting Model with a Multinomial Logit Model. Journal of the Transportation Research Board, 2672, 35-45. https://doi.org/10.1177/0361198118773556
[11]	Claudia, N.B., Uwe, D., Yishen, L. and Harris, S. (2017) Transport Policies and Development. The Journal of Development Studies, 53, 465-480. https://doi.org/10.1080/00220388.2016.1199857
[12]	T. E. o. E. Britannica (1999, September 1) Britanica. https://www.britannica.com/place/Douala
[13]	SITRASS (2004) Poverty and Urban Mobilty. African Region World Bank, Douala.
[14]	Minimum-Wage.org (2022). https://www.minimum-wage.org/international/cameroon
[15]	Cao, J. and Cao, X.S. (2017) Comparing Importance-Performance Analysis and Three-Factor Theory in Assessing Rider Satisfaction with Transit. Journal of Transport and Land Use, 10, 837-854. https://doi.org/10.5198/jtlu.2017.907
[16]	Maria, V.J., Tobias, H. and Per, J. (2004) Latent Variables in a Travel Mode Choice Model; Attitudinal and Behavioural Indicator Variables. Swedish National Road and Transport Research Institute, Linkoping.
[17]	Frank, S.K. and Chandra, B. (2006) A Self Instructing Course in Mode Choice Modeling: Multinomial and Nested Logit Models. Georgia Institute of Technology, Georgia.
[18]	Bhat, C. (1995) A Heteroscedastic Extreme Value Model of Intercity Mode Choice. Transportation Research Part B: Methodological, 29, 471-483. https://doi.org/10.1016/0191-2615(95)00015-6