Daily Travel Mode Choice Considering Carbon Credit Incentive (CCI)—An Application of the Integrated Choice and Latent Variable (ICLV) Model

Abstract

There have been many implementations of carbon credit incentives (CCIs) for promoting green travel, but research on quantifying the effectiveness remains limited. To fill this gap, this study focuses on residents’ daily transportation mode choices under the incentive of carbon credits by employing an integrated choice and latent variable (ICLV) model to adequately address the role of attitudinal variables related to carbon credits, such as perceived usefulness, perceived ease of use, and behavioral intentions. Data from a questionnaire survey show that the ICLV model provides a richer and more nuanced understanding of the green mode choice than a traditional multinomial logit (MNL) model, where the AIC value of the ICLV model (3901.17) is smaller than that of the MNL model (3910.09). Carbon credits exhibit diverse impacts across various modes of eco-friendly transportation and specific demographic groups. Commuting trips reveal noteworthy positive associations between carbon credits and the use of bicycles as well as metro/bus services. Moreover, carbon credits exert a more pronounced influence on individuals with higher education levels, older age groups, and owners of new energy vehicles, whereas their impact on high-income individuals is relatively constrained. Furthermore, perceptions of carbon credits are pivotal, with perceived utility emerging as the most influential factor. The results provide a scientific basis for formulating more effective policies regarding carbon credit and incentive measures in the future.

1. Introduction

In recent years, incentivizing green travel modes has become a hot topic in the field of transportation [1,2,3,4]. Both designing appropriate incentivizing measures to guide travelers towards green travel mode choices and understanding the influencing factors leading to these choices are of importance for several reasons. Firstly, the global transportation industry has already become the fastest growing source of carbon emissions [5]. Secondly, incentives to switch to green travel can reduce private car usage [6], thereby alleviating traffic congestion and improving transportation efficiency [7]. Besides these practical reasons, studying the process of how individuals choose green travel modes can enrich the travel behavior theory, which can further guide the implementation of relevant low-carbon policies [8].

Environmental awareness campaigns and educational initiatives were often used to promote green travel. However, these approaches may not provide sufficient motivation for individuals to change their travel behavior due to insufficient incentives [9]. Carbon credit systems, as a new mechanism of carbon emission consumption reduction regarding people’s daily lives, are an innovative approach and an extension from the existing carbon emission trading system among countries or companies to individual persons. It uses carbon credits as an incentive for rewarding and quantifying the carbon reduction contribution of individual residents. The rewards further encourage whole societies to participate in low-carbon lifestyles. Therefore, this paper focuses on studying the incentive effect of carbon credits on green travel and their various influencing factors.

Travelers’ acceptance of carbon credits, as well as green travel, is related to an individual’s preference and attitude. Although incorporating attitudinal variables into mode choice models are considered in the previous research [2,10,11], measurement errors associated with latent variables are still not well handled in discrete choice models. To fill this gap, we employ an integrated choice and latent variable (ICLV) model focusing on the role of attitudinal variables, which are treated as the latent variables, to well handle this problem. The ICLV model provides a useful modeling framework to adequately analyze structural relationships between sociodemographic characteristics, attitudes, and mode choice behavior [1,12]. Through highlighting the roles of attitudes and acceptance, the present work provides deeper insights into the intricate mechanism of daily travel mode choice considering carbon credit incentives (CCIs), which has received limited attention in previous research. Moreover, it elucidates the direct and indirect impact of travelers’ demographic and socioeconomic attributes on their mode choice, primarily through the influence of their attitudes.

The rest of this manuscript is organized as follows. Section 2 presents a literature review of carbon credits and illustrates the reason for applying the ICLV model in this study by comparing previous studies based on utilizing carbon credits to incentivize green mode choice behavior. The Section 3 gives a detailed description of data used in this research, followed by an explanation and application of the ICLV model in Section 4. Section 5 presents the results and discussion, which show how latent variables and observable variables influence the process of travelers’ choice of different green travel modes with incentives. Finally, the conclusions are drawn in Section 6.

2. Literature Review

The concept of carbon credit originated from the California Zero-emission-vehicle Act in 1990, promoting low-emission vehicles through mandatory market mechanisms. In 1996, individual carbon credits trading was proposed to encourage emission reduction [13]. Tradable Energy Quotas (TEQs) were allowed for trading between low- and high-emission contributors. Personal Carbon Allowances, Household Carbon Trading, and Tradable Fuel Permits were proposed between 2005 and 2008 [9,14,15]. Research in this field then shifted towards individual carbon trading and its impact on energy consumption choices. In the context of transportation, by providing individuals with incentives and rewards for choosing eco-friendly modes of travel, such as public transportation or cycling, the carbon credit system aims to promote sustainable behaviors and reduce carbon emissions.

Consequently, the carbon credit system’s influence on travel behavior has gained attention lately. In 2011, personal tradable carbon permits were introduced for road transportation [16]. The emission reduction effects between the carbon credit system and carbon tax policies were also compared, highlighting the incentivizing power of carbon credit on low-carbon travel [17]. Later on, more empirical studies and implementation cases of incentivizing green travel through carbon credits emerged, which can be found in Xiamen, Nanjing, and Wuxi in China. These research and applications of the development in the carbon credit system have demonstrated its impact on travel behavior choices.

However, previous studies have ignored the significance of individual attitudes and preferences in modeling the effect of CCIs on mode choice decisions. Although some early research on travel behavior either directly incorporated attitudinal scores from Likert scale surveys into discrete choice models [2,11] or utilized factor scores derived from observed indicators [10], both of the above approaches have inherent methodological limitations in adequately capturing the complex relationship between attitudes and mode choice. The former treats Likert scores as error-free predictors of underlying attitudes. However, the underlying attitudes themselves, rather than the survey responses, may have a causal relationship with choice behavior [12]. For the latter approach, simply extracting data from the scale results does not effectively handle the measurement errors in latent variables [10]. Moreover, these approaches fail to adequately consider the proved potential association between an individual’s sociodemographic characteristics and attitudes discovered in related research [18,19]. As indicated in previous behavior-related studies, the significance of individual attitudes and preferences cannot be neglected as research on travel behavior [20,21,22] increasingly uncovers the influence of attitudes and preferences, such as attitudes towards transportation modes, environmental concerns, and perceived safety, in shaping travel decisions.

To fill up this gap, the ICLV model is proposed to address the ignored associated attitudinal variables in travel behavior modeling [1,12,22,23,24,25]. Specifically, the ICLV model combines a latent variable model with a discrete choice model to explore the structural relationship between observed and unobserved variables, as well as the measurement relationship between latent variables and outcomes within a unified framework. Its advantage in understanding mode choice behavior by incorporating attitudes and preferences has already been demonstrated [26,27], resulting in a comprehensive understanding on the influence of attitudinal factors on travel decisions, which could also be impacted by unconventional green travel incentive policies, such as individualized marketing strategies and campaigns [28]. Additionally, unobserved variables such as psychological factors and attitudes [29] can be well incorporated in the ICLV model to investigate the motivating effects of carbon credits on green travel choice. This will fill the gap both theoretically by enriching the travel behavior theory in circumstances of green travel incentivization and practically by providing fundamental essentials for making transport policies.

3. Data Description

3.1. Data Collection

The data utilized in this study were collected via an online questionnaire in 2023. Finally, 345 travelers’ data from, originally, 600 respondents were selected as valid ones through a series of rules. These rules encompassed, but were not limited to, considerations such as whether the time taken to complete the questionnaire falls within an appropriate range (neither too short nor too long), whether responses to internal questionnaire items exhibit consistency (such as alignment between possessing a driver’s license and engaging in car travel), and whether the reliability of responses in the questionnaire’s scale section is dependable (such as whether answers predominantly cluster around mid-range values). Due to the random sampling method employed in this study, along with a sufficient sample size (345 valid respondents), the sample is considered to possess a degree of representativeness. We specifically chose individuals who had prior exposure to CCIs or had knowledge of CCIs. This careful selection process was designed to ensure that the respondents had a reasonable level of assurance in obtaining rewards through CCI measures. Moreover, it is essential to note that rewards are an integral part of the CCI. Participants in our survey have actively engaged with carbon credits by using them, making it highly likely that they have benefited from these incentives.

The participants who were surveyed had the complete freedom to discontinue their participation in the survey if they felt uncomfortable sharing any information that could compromise their privacy or if they encountered any offensive content. The participation was voluntary and the surveyed persons were accordingly paid remuneration for their participation.

Records relating to respondents’ sociodemographic profile, travel-related information, and perceptions of carbon credit in daily travel were collected. Sociodemographic data included their gender, age, education level, occupation, income, marital status, and the number of children being raised. Travel-related information contained driver’s license possession status, driving experience, the frequency of driving, and the energy type of the car. Based on Extended Planned Behavior Theory [30], perceptions of carbon credit in this survey cover six latent variables, which are subjective norms, degree of cognition, perceived usefulness, perceived ease of use, environmental awareness, and travel behavioral intentions, to analyze respondents’ perceptions of carbon credit. Subject norms refer to the common behaviors, attitudes, and values expected from respondents regarding a carbon credit system. Degree of cognition refers to the degree of respondents’ understanding of carbon credits. And, a five-point Likert-type scale was applied on the investigation of preference and attitude regarding carbon credits, where one to five represents strongly disagree, disagree, moderately agree, agree, and strongly agree, respectively. This study primarily focused on car travel and did not survey all modes of transportation for the respondents within a single day, aiming to model the transition from driving to green transportation modes for individuals with varying driving frequencies.

3.2. Sociodemographic and Travel-Related Features of Respondents

Table 1. Statistics of the sociodemographic and travel-related features (345 valid respondents).

Variable (Abbr. in Models)	Description	Statistics
Sociodemographic
Gender	% Females (otherwise: males)	53.72%
Age	% Mean age of respondents (years old)	34
	% ≤20 years old	0.65%
	% 21–30 years old	40.13%
	% 31–45 years old	53.07%
	% 46–60 years old	5.83%
	% More than 60 years old	0.32%
Education level (edu)	% Primary school and below	0%
	% Junior high school	0.65%
	% High school/vocational school	2.59%
	% Bachelor’s degree/college diploma	90.94%
	% Master’s degree and above	5.83%
Average monthly income	% Mean (US dollars)	1360
	% Min (US dollars)	342.5
	% Max (US dollars)	2741.1
Marital status	% Married (otherwise: unmarried)	82.85%
Children	% People with one or more children	73.56%
Travel related
License possession (d_license)	% People have a driver’s license	93.85%
Driving experience (d_year)	% Mean (year)	6.11
	% ≤1 year	2.76%
	% 1–5 years	44.14%
	% 6–10 years	40.00%
	% More than 10 years	13.10%
Frequency of driving (d_frequence)	% ≤twice a week	14.14%
	% 3–6 times a week	46.90%
	% Once–twice a day	32.76%
	% More than twice a day	6.21%
Energy type of vehicles (d_type)	% Conventional fuel (otherwise: new energy)	61.38%

illustrates the summary statistics of the valid surveyed participants, whose data are used in the later research. More than four-fifths of the respondents are employees of enterprises and institutions, who have relatively regular income and travel patterns [31], and are one of the main participants in transportation systems [32]. More than 90% of participants have a driver’s license, and more than 80% of the respondents make a trip by car at least once every two days. Especially, more than half of the respondents drive conventional fuel vehicles. Therefore, these respondents are apposite samples to study the possibility of choosing green travel modes induced by carbon credits.

3.3. Perceptions of Carbon Credits

For investigating respondents’ travel behavior intentions, six captured perceptions of carbon credits are incorporated as latent variables in a daily travel mode choice model.

Table 2. Responses to perception statements. (α = 0.89 > 0.7).

Indicator (Abbr. in the Models)	5-Point Likert-Type Scale
	1 (%)	2 (%)	3 (%)	4 (%)	5 (%)
Subjective norms (SN)
I: I will accept the recommendation of relatives and friends to use the carbon credit platform.	1.29	1.94	12.3	54.69	29.77
II: I will accept the recommendation of experts and scholars to use the carbon credit platform.	0.32	5.83	18.77	46.93	28.16
III: I will accept the promotion of government agencies to use the carbon credit platform.	0.65	1.62	9.71	45.63	42.39
Degree of cognition (DC)
I: I understand how carbon credits encourage green travel.	0.65	2.91	18.77	60.52	17.15
II: I know the advantages of carbon credits incentivizing green travel.	0.65	3.24	17.15	51.46	27.51
III: I know the policy support for the carbon credit system.	1.62	4.53	31.07	41.75	21.04
Perceived usefulness (PU)
I: Carbon credit platforms can reduce carbon emissions.	0.65	2.59	9.39	51.13	36.25
II: Carbon credit platforms can reduce my daily travel expenses.	0.32	2.91	21.68	46.28	28.8
III: Carbon credit platforms can reduce the time cost of my daily travel.	0.65	10.36	25.89	44.01	19.09
Perceived ease of use (PEOU)
I: Carbon credit platforms are easy to use.	2.41	6.53	20.62	51.2	19.24
II: Carbon credit platforms can accurately record my green travel behavior.	3.09	8.25	12.03	43.99	32.65
III: Carbon credit platforms can provide me with satisfactory rewards.	2.06	10.31	31.27	34.71	21.65
IV: Carbon credit platforms can motivate me to choose green travel behavior.	3.09	8.59	18.9	44.33	25.09
Environmental awareness (EA)
I: I know what carbon reduction is.	0.32	4.53	26.21	58.9	10.03
II: I understand why carbon reduction is necessary.	0.32	6.15	18.12	45.95	29.45
III: I know ways to reduce carbon emissions.	1.62	8.74	32.69	44.66	12.3
Behavioral intentions (BI)
I: I support the carbon credit system in daily transportation.	0.97	2.27	11.97	42.07	42.72
II: I wish to have more carbon credit platforms to use.	0.32	2.59	8.74	43.04	45.31

Notes: under 5-point Likert-type scale, 1, 2, 3, 4, and 5, respectively, represent strongly disagree, disagree, moderately agree, agree, and strongly agree.

demonstrates the proportion of respondents’ response to their perceptions related to the six variables of carbon credit.

The majority of surveyed respondents were likely to be influenced by recommendations from external sources and, therefore, chose to use carbon credit platforms. Although over 70% of the surveyed respondents had some extent of knowledge or experience with carbon credit platforms, there was still nearly 40% of individuals who were unaware of the current situation of promotion and policy support for the carbon credit system, according to the response data to the third question concerning the degree of cognition of CCI in the Likert scale. Almost all of the surveyed participants (over 90%) believed that the carbon credit system was beneficial in reducing carbon emissions in transportation. However, when it came to whether the carbon credit system could reduce travel time and economic costs, at least half of the respondents remained skeptical. While over 60% of the surveyed participants believed that the carbon credit system could have a certain incentivizing effect on them, but close to half of the respondents doubted whether carbon credit platforms could provide satisfying rewards. Over half of the surveyed participants had a strong environmental awareness and supported the implementation of carbon credit systems in daily transportation.

3.4. Trip Data and Mode Share Regarding Travel Behavioral Intentions

The travel mode choices under different trip times and distances are used as dependent variables for investigating respondents’ travel behavioral intentions under the attitude and preference of carbon credits. The travel behavioral intentions involve three categories of green travel modes, i.e., walking, cycling, and bus/metro, which are described by three corresponding travel mode choice models, respectively. In each discrete choice model, the dependent variable is a binary one, that is whether the respondent can be induced by CCIs to choose the corresponding green travel mode. The comparison of the induction effects on each green travel mode can be also made with driving as the reference.

Table 3. Summary statistics of the trip data and mode share regarding travel behavioral intentions.

Variable	Summary
Trip data	Induced to choose walking by carbon credit	Induced to choose cycling by carbon credit	Induced to choose bus/metro by carbon credit	Not induced (remain driving)
Trip time (mean)	18.18 min	30.43 min	43.93 min	56.32 min
Trip distance (mean)	2.79 km	4.83 km	25.11 km	32.05 km
Mode choice for each trip purpose(abbr. in the models)	Induced to choose walking by carbon credit (otherwise not)	Induced to choose cycling by carbon credit (otherwise not)	Induced to choose bus/metro by carbon credit (otherwise not)	----
Commuting (Pur1)	43.60%	54.71%	74.53%	----
Recreation (Pur2)	38.39%	39.62%	44.34%	----
Shopping (Pur3)	80.09%	72.17%	63.21%	----
Hospital (Pur4)	17.06%	20.75%	33.49%	----
Personal affairs (Pur5)	15.63%	11.32%	27.34%	----

demonstrates statistics of the trip data and mode share regarding travel behavior intentions. It indicates that the larger the average travel time or trip distance of individuals, the more difficult a driving trip changes to a green travel mode, even under CCIs. Regarding each trip purpose, commuting is the easiest one which can be changed to bus/metro due to the incentivizing effect of carbon credits. The proportions of choosing walking, cycling, and metro/bus for commute trips are 43.6%, 54.7%, and 74.53%, respectively, demonstrating a stronger effect of CCIs for using the bus/metro. Carbon credits perform a stronger incentive effect on changing driving to public transportation modes rather than the other two green travel modes for almost all travel purposes, except for shopping trips. The integration of public transportation with commuters’ mobile devices through app platforms allows for easy implementation of carbon credits, which strengthens public transportation’s role in handling a substantial portion of commuting trips [33].

Table 3 indicates that carbon credits vary in terms of the strength of their incentives, resulting in different trip times and distances for the three green travel modes. The distribution of trip time and distance are shown in Figure 1 and Figure 2. As trip time and distance increase, the proportion of carbon-credit-induced choices initially rises to a maximum before decreasing. To enhance the effect of CCIs and promote sustainable transportation, it is crucial to choose the appropriate green travel mode based on each commuter’s specific trip range. It is essential to consider the specific trip range of each commuter. Commuters tend to respond positively to CCIs for shorter trips where the advantages of green travel modes are more evident. However, for longer trips, factors such as comfort and convenience often lead to a preference for personal vehicles, diminishing the influence of CCIs. To maximize the positive impact of carbon credits, the strategy should involve customized incentives to align with individual travel patterns. This customization ensures that incentives cater to the unique needs and preferences of commuters based on their trip lengths. Ultimately, the success of CCI in influencing travel behavior is contingent on this tailored approach, where shorter trips may require different incentives compared to longer ones.

4. Methodology

Different from sociodemographic and travel behavior, the six perceptions are not observable. As a result, they must be considered as latent variables. In order to include them in the mode choice model, one common approach is to add factor scores derived from attitude statements directly into the utility of the choices. However, this approach cannot avoid potential risks, including measurement errors and endogeneity biases, which have been highlighted in previous studies [12,23,24]. To solve these risks, the ICLV model is proposed to combine a discrete choice model with one or more latent variables.

In order to specifically study the inductive effect of carbon credits on different green travel modes, we selected three categories of green travel modes: walking, cycling, and bus/metro, and established three corresponding travel choice models, respectively. In each discrete choice model, the dependent variable was binary: whether the person was induced by carbon credits to choose a green travel mode (walking, cycling, and bus/metro) or not. In order to facilitate the horizontal comparison of the induction effect of carbon credits on different green travel modes, we made an assumption that the default travel mode of all surveyed travelers was driving.

The entire framework of the ICLV model is shown in Figure 3. Within the ICLV framework, exogenous variables exert an influence on latent variables (LVs). The structural equation model (SEM) captures the relationship between these LVs and incorporates another model, known as the measurement model. The measurement model utilizes the LVs as predictors to estimate the survey responses of individuals who have concerns in terms of attitude and perception statements [34]. In our case, the measurement model connects LVs to statements regarding the perception of carbon credit in the daily travel of respondents. Three models corresponding to the three different incentivized green travel modes (i.e., walking, cycling, metro/bus) are proposed with a similar framework. The discrete choice model describes the utility of each alternative travel mode based on trip-specific, sociodemographic, and attitudinal characteristics.

For each individual $n\left(\mathrm{i}\mathrm{n} \mathrm{o}\mathrm{u}\mathrm{r} \mathrm{c}\mathrm{a}\mathrm{s}\mathrm{e}, n=1,\dots ,345\right)$, there is a set of mutually exclusive alternatives $i\left(i=0:\mathrm{n}\mathrm{o}\mathrm{t} \mathrm{i}\mathrm{n}\mathrm{d}\mathrm{u}\mathrm{c}\mathrm{e}\mathrm{d},1:\mathrm{i}\mathrm{n}\mathrm{d}\mathrm{u}\mathrm{c}\mathrm{e}\mathrm{d}\right)$. Based on our previous assumptions, the ICLV model for the three different incentivized green travel modes can be defined as $i=0$, representing the choice of driving, and $i=1$ representing one of the choices of walking, cycling, or metro/subway in each corresponding model. The utility maximization function is defined as in Equation (1).

$$Y_{n,i}=\left\{\begin{array}{c}1, if U_{n,i}=ma{x}_j\left\{U_{n,j}\right\}\\ 0, otherwise\end{array}\right.$$

(1)

where $Y_{n,i}$ represents the indicator variable for individual $n$ choosing mode option $i$. If a traveler chooses travel mode $i$, the value of $Y_{n,i}$ is $1$, resulting in utility $U_{n,i}$ as shown in Equation (2); otherwise, $0$.

$$U_{n,i}=V_{n,i}+{\epsilon }_{n,i}\cdots$$

(2)

where $V_{n,i}$ is a systematic component (observable) of the utility function and ${\epsilon }_{n,i}$ is a random disturbance (error term). In the ICLV model, which incorporates the effects of latent variables on mode utility, $V_{n,i}$ is represented in Equation (3).

$$V_{n,i}=a_i\mathit{S}+b_i{\mathit{Z}}_i+c_i{\mathit{X}}^{\mathrm{*}}$$

(3)

where $\mathit{S}$ is the observed variable vector representing the socioeconomic attributes of individual travelers. ${\mathit{Z}}_{\mathit{i}}$ is the observed variable vector representing the attributes of each travel mode. ${\mathit{X}}^{\mathrm{*}}$ is the $1\times l$ dimensional latent variable vector. In our case, $l=6$; specifically, it includes subjective norms, degree of cognition, perceived usefulness, perceived ease of use, environmental awareness, and behavioral intentions. ${\mathit{a}}_{\mathit{i}},{\mathit{b}}_{\mathit{i}},{\mathit{c}}_{\mathit{i}}$ are the unknown parameter vectors to be estimated.

In the structural equation of the latent variable model, it is assumed that the relationship between the observed variable vector and the latent variable vector is a linear function [35]. In this study, the structural equation represents the relationship among the latent variables as shown in Equation (4).

$${\mathit{X}}^{\mathrm{*}}=\mathit{X}\mathit{\Gamma }+\mathit{\chi }$$

(4)

where $\mathit{X}$ represents the observed variable vector of $1\times k$ dimensional socioeconomic attributes that influence the latent variables. In our case, $k=11$; specifically, it includes the following socio-economic attributes: gender, age, education level, occupation, income, marital status, the number of children raised, driver’s license possession, driving experience, the frequency of driving, and the energy type of the car driven. $\mathit{\Gamma }$ is the $k\times l$ dimensional unknown parameter matrix to be estimated. $\mathit{\chi }$ is the error term.

For the measurement equation in the latent variable model, we also assume a linear function as the estimation function [36]. Therefore, the mathematical expression of the measurement equation is as follows (Equation (5)):

$$\mathit{y}={\mathit{X}}^{\mathrm{*}}\mathit{\Lambda }+\mathit{\nu }$$

(5)

where $\mathit{y}$ is the $1\times q$ dimensional observed variable vector of the latent variables. $q$ represents the number of indicators for each latent variable. $\mathit{\Lambda }$ is the $l\times q$ dimensional unknown parameter matrix to be estimated. $\mathit{\nu }$ is the error term. For the error terms $\mathit{\epsilon },\mathit{\chi },\mathit{\nu }$, in order to facilitate subsequent model parameter estimation, it is assumed that they satisfy the following relationship, as shown in Equations (6) and (7).

$$E\left(\mathit{\epsilon }{\mathit{\chi }}^{\prime }\right)=E\left(\mathit{\epsilon }{\mathit{\nu }}^{\prime }\right)=E\left(\mathit{\nu }{\mathit{\chi }}^{\prime }\right)=0$$

(6)

$$E\left(\mathit{\epsilon }{\mathit{\epsilon }}^{\prime }\right)=\mathit{\Xi },E\left(\mathit{\chi }{\mathit{\chi }}^{\prime }\right)=\mathit{\Psi },E\left(\mathit{\nu }{\mathit{\nu }}^{\prime }\right)=\mathit{\Theta }$$

(7)

The structural equation models and corresponding discrete choice models for the three green travel modes form the basis of the ICLV model for daily travel mode choice considering CCIs. Specifically, Equations (4) and (5) form the structural equation model that describes the relationship among the latent variables. Equations (1) and (2) form the discrete choice model. In our case, the structural equation can be represented specifically as follows (Equations (8) and (9)):

$$\begin{array}{c}{X}_{n,l}^{\mathrm{*}}={\gamma }_{l1}gender+{\gamma }_{l2}age+{\gamma }_{l3}edu+{\gamma }_{l4}work+{\gamma }_{l5}income+{\gamma }_{l6}marriage+{\gamma }_{l7}child+{\gamma }_{l8}{d}_{license}\\ +{\gamma }_{l9}{d}_{year}+{\gamma }_{l10}{d}_{frequency}+{\gamma }_{l11}{d}_{type}+\chi \end{array}$$

(8)

$$l=SN,DC,PU,PEOU,EA,BI$$

(9)

To estimate unknown parameters of the ICLV model, the maximum likelihood method is used. We let $\mathit{U}={\left(U_1,U_2,\cdots U_J\right)}^{\prime },\mathit{\epsilon }={\left({\epsilon }_1,{\epsilon }_2,\cdots {\epsilon }_J\right)}^{\prime }$, Equations (2) and (3) can be rewritten as Equation (10):

$$\mathit{U}=\mathit{A}\mathit{s}+\mathit{Z}\mathit{b}+\mathit{C}{\mathit{X}}^{\mathbf{*}}+\mathit{\epsilon }$$

(10)

where $\mathit{A}=\left[\begin{array}{c}{a}_1^{\prime }\\ a_2^{\prime }\\ \cdots \\ a_J^{\prime }\end{array}\right],\mathit{Z}=\left[\begin{array}{c}{Z}_1^{\prime }\\ Z_2^{\prime }\\ \cdots \\ Z_J^{\prime }\end{array}\right],\mathit{C}=\left[\begin{array}{c}{C}_1^{\prime }\\ C_2^{\prime }\\ \cdots \\ C_J^{\prime }\end{array}\right];$ for estimation, set $a_J=c_J=0$.

Here, it is assumed that vector $\mathit{q}=\left(y^{\prime },X^{{\mathrm{*}}^{\prime }},U^{\prime }\right)$ follows a multivariate normal distribution with covariance matrix ${\Omega }_1$ and mean vector $m_1$. Let $\mathit{\rho }$ be the vector of unknown parameters under this condition, $d$ represents the vector of indicator variables where the element in the row $j$ is denoted as ${\mathrm{d}}_j=1$, and $\Phi$ represents the probability density function of a $J$ dimensional normal distribution. When the sample size is $n$, the likelihood function for the unknown parameter $\rho$ can be calculated as shown in Equations (11) and (12).

$$l\left(\mathit{\rho }\right)=\sum _{i=1}^n{d}_i\ln PR\left(d_i=j,\left.y_i\right|X_i,s_i,z_{ij}\right)$$

(11)

$$\begin{array}{c}PR\left({\mathrm{d}}_i=1,\left.y_i\right|X_i,s_i,z_{ij}\right)\\ ={\int }_{SN}{\int }_{DC}\cdots {\int }_{BI}\left[{\int }_{-\mathrm{\infty }}^{\mathrm{\infty }}{\int }_{-\mathrm{\infty }}^{U_{i1}}\cdots {\int }_{-\mathrm{\infty }}^{U_{ij}}\Phi \left(U_{i1},\cdots ,U_{ij}\right)d{U}_{i1}\cdots U_{ij}\right]d{X}_i^{{\mathrm{*}}^{\prime }}\end{array}$$

(12)

To intuitively analyze the impact of travelers’ acceptance of carbon credits on their choice of green travel modes as in the aforementioned ICLV model, a new latent variable named $ACCEPT$ is introduced. This categorical variable is used to directly describe the acceptance level of carbon credit platforms for each surveyed individual. The specific definition of the variable is as shown in Equation (13).

$$ACCEP{T}_n=\left\{\begin{array}{c}1,con \\ 2,netural\\ 3,pro\end{array}\right.$$

(13)

For individual n in the structural model, regression weights for predicting the unobserved variables from the observed variables can be calculated by Equation (14).

$$\mathit{W}=\mathit{B}{\mathit{S}}^{-1}$$

(14)

where $\mathit{W}$ is the matrix of regression weights, $\mathit{S}$ is the matrix of covariances among the observed variables, and $\mathit{B}$ is the matrix of covariances between the unobserved and observed variables. The fitted results of all measurement variables from the six latent variables in the ICLV model are utilized. Using the estimated weight matrix from the structural equation, the weighted score for each individual’s acceptance level of carbon credits is calculated. As the acceptance level for each respondent is represented with a five-point Likert scale, the acceptance levels are discretized based on interval distribution to obtain the acceptance categories for the surveyed individuals.

The Biogeme Python programming package [37] is used to estimate the ICLV model. Traditional multinomial logit (MNL) models are also run to compare the results with the ICLV model in order to check its superiority.

5. Results and Discussion

Table 4. Estimated results of the ICLV and MNL models (Green travel mode: walking).

	Model 1		Model 2		Model 3
	Estimate	z-Statistic	Estimate	z-Statistic	Estimate	z-Statistic
Constant	4.940 ***	10.51	4.945 ***	10.7	4.519 ***	10.79
Trip-specific variables
Travel time	−0.949 ***	−15.5	−0.948 ***	−15.5	−0.947 ***	−15.49
Travel distance	−0.874 ***	−14.79	−0.874 ***	−14.78	−0.872 ***	−14.77
Pur3	1.909 ***	12.61	1.908 ***	12.61	1.904 ***	12.6
Pur4	−1.104 ***	−7.29	−1.104 ***	−7.29	−1.101 ***	−7.28
Pur5	−1.203 ***	−7.86	−1.203 ***	−7.86	−1.201 ***	−7.85
Sociodemographic variables
Gender	0.188 **	2.07	0.208 **	2.31	0.202 **	2.24
Age	0.133 *	1.7	0.126 *	1.61	0.119 *	1.52
Income	−0.082 **	−2.29	−0.082 **	−2.28	−0.070 **	−1.98
D_type	0.120 *	1.37	0.118 *	1.34	0.154 **	1.78
Latent variables
DC	−4.389 *	−1.14	-	-	-	-
PU	4.571 **	1.16	-	-	-	-
PEOU	0.182 *	1.63	-	-	-	-
ACCEPT	-	-	0.202 ***	2.25	-	-
LL(Start)	−2075.97		−2075.97		−2075.97
LL(Final)	−1647.01		−1647.53		−1650.05
AIC	3320.03		3317.06		3320.10
BIC	3398.75		3383.67		3380.66

Notes: * p < 0.1; ** p < 0.05; *** p < 0.01; Reference groups for trip purpose: Pur2(recreation) for model 1, 2, 3.

,

Table 5. Estimated results of the ICLV and MNL models (Green travel mode: cycling).

	Model 1		Model 2		Model 3
	Estimate	z-Statistic	Estimate	z-Statistic	Estimate	z-Statistic
Constant	0.389	0.57	0.243	0.36	−0.165	−0.25
Trip-specific variables
Travel time	−0.221 ***	−4.83	−0.221 ***	−4.82	−0.220 ***	−4.82
Travel distance	−0.428 ***	−8.9	−0.428 ***	−8.9	−0.427 ***	−8.89
Pur1	0.752 ***	5.55	0.751 ***	5.55	0.749 ***	5.55
Pur3	1.537 ***	10.71	1.535 ***	10.71	1.532 ***	10.69
Pur4	−0.780 ***	−5.24	−0.779 ***	−5.24	−0.777 ***	−5.23
Pur5	−1.474 ***	−8.81	−1.473 ***	−8.81	−1.469 ***	−8.8
Sociodemographic variables
Age			0.077 *	1.08	0.096 *	1.28
Edu	0.312 **	2.26	0.327 **	2.37	0.321 ***	2.34
Income	−0.103 ***	−3.18	−0.094 ***	−2.94	−0.072 **	−2.22
Marriage	0.260 **	2.01	0.248 **	1.93	0.272 **	2.11
D_year					−0.058 *	−1.06
Latent variables
PU	5.380 **	1.97	-	-	-	-
SN	−4.106 **	−1.86	-	-	-	-
ACCEPT	-	-	0.230 ***	2.87	-	-
LL(Start)	−2119.97		−2119.97		−2119.97
LL(Final)	−1937.58		−1939.47		−1943.04
AIC	3901.17		3902.94		3910.09
BIC	3979.95		3975.66		3982.81

Notes: * p < 0.1; ** p < 0.05; *** p < 0.01; Reference groups for trip purpose: Pur2(recreation) for model 1, 2, 3.

and

Table 6. Estimated results of the ICLV and MNL models (Green travel mode: bus/metro).

	Model 1		Model 2		Model 3
	Estimate	z-Statistic	Estimate	z-Statistic	Estimate	z-Statistic
Constant	−0.254	−0.4	−0.435	−0.71	−0.971 *	−1.63
Trip-specific variables
Travel time	−0.178 ***	−3.95	−0.178 ***	−3.94	−0.177 ***	−3.93
Travel distance	0.164 ***	3.64	0.164 ***	3.64	0.163 ***	3.63
Pur1	1.061 ***	7.54	1.057 ***	7.53	0.990 ***	7.34
Pur2	−0.264 **	−2.04	−0.263 **	−2.03
Pur3	0.517 ***	3.92	0.516 ***	3.92	0.468 ***	3.64
Pur4	−0.728 ***	−5.46	−0.726 ***	−5.45	−0.773 ***	−5.87
Sociodemographic variables
Age	0.170 ***	2.51	0.142 **	2.12	0.136 **	2.02
Edu	0.206 *	1.56	0.212 *	1.61	0.200 **	1.53
Income	−0.064 **	−2.09	−0.051 **	−1.7	−0.033 *	−1.12
Latent variables
EA	0.130 *	1.08	-	-	-	-
PU	6.077 **	1.89	-	-	-	-
PEOU	0.167 **	1.96	-	-	-	-
SN	−4.606 *	−1.81	-	-	-	-
ACCEPT	-	-	0.286 ***	3.72	-	-
LL(Start)	−2165.19		−2165.19		−2165.19
LL(Final)	−2055.84		−2060.79		−2069.78
AIC	4141.69		4145.59		4159.56
BIC	4232.59		4218.31		4220.16

Notes: * p < 0.1; ** p < 0.05; *** p < 0.01; Reference groups for trip purpose: Pur5 (affairs) for model 1, 2, 3.

illustrate the estimated results of all ICLV models and MNL models that are used in the situation of daily travel mode choice behavior considering CCIs (our case uses a data set of a sample size n = 345). Each model includes only variables that are significantly associated with the utility. Model 1 represents the ICLV model, where all six latent variables related to awareness and attitudes regarding carbon credits are included. Model 2 represents the ICLV model with the acceptance level (i.e., ACCEPT) as a latent variable, rather than the latent variables that are used in Model 1. Model 3 is the MNL model as the baseline, which does not include any latent variables. The results are further explained from the perspective of trip purpose, demographic features, and perception related to attitude and preference as follows. Based on the model fitting results in Table 4, Table 5 and Table 6 (AIC and BIC values), despite the ICLV model incorporating more variables, the AIC values for the ICLV model are consistently smaller than those of the corresponding MNL models. This indicates that the ICLV model not only explains more variables (i.e., the latent psychological variables in this study) but also outperforms the basic MNL model in terms of goodness of fit.

5.1. Carbon Credits’ Effect on Different Trip Purpose

For commuting trips, the utility term for the walking mode is not significant, while for cycling and the metro/bus, it shows a significant positive correlation (0.752 and 1.061). For other trip purposes like recreation or medical trips, there is a significant negative correlation with the induction effect from carbon credits. It means that when having such a trip purpose, travelers prioritize personal needs, convenience, comfort, and time costs over CCIs.

However, an exception exists for shopping trips as travelers for shopping are more inclined to accept CCIs. This is due to their higher concern for economic factors, such as the case that implementing carbon credits on app platforms of large shopping malls could specifically target these travelers. Additionally, shopping trips typically happen in cases of shorter distances, making walking or cycling a more feasible alternative of green travel mode.

5.2. Carbon Credits’ Effect on Various Demographics

Higher education levels show a positive correlation with the effect of CCIs across all green travel modes, with the most significant impact observed for cycling. This aligns with existing research on cycling preferences among educated commuters [38].

Conversely, the effect of CCIs is limited among high-income individuals. While income positively correlates with the acceptance level of carbon credits in the latent variable model, the final choice of green travel modes reveals a negative correlation between income and the effect of CCIs. High-income individuals prioritize travel convenience and comfort, making them less sensitive to material incentives provided by carbon credits. These individuals would like to financially gain less to make their travel more convenient and comfortable, which also means that these people may be those would like to pay for their ideal or accustomed trips. To attract high-income people to choose green transportation options, expanding the range of incentive measures beyond material rewards, such as personal recognition and other non-material incentives, should be considered.

It is also found that carbon credits have a stronger incentive effect on older individuals. Regardless of the specific green travel mode, the results from the ICLV models consistently show that age is positively correlated with the effect of CCI. Considering the current implementation of carbon credit platforms, the material incentives provided to travelers are relatively limited and often only older travelers satisfy an accumulative travel frequency or mileage. Younger individuals, compared to older ones, may prefer more direct or substantial material incentives. Therefore, carbon credits can be improved with more customized and attractive styles targeting young individuals.

New energy vehicle owners are potential targets for carbon credit-induced green travel as they show a higher acceptance of carbon credits, according to the results from the latent variable model. The ICLV model also indicates a stronger effect of CCIs on new energy vehicle owners compared to drivers of traditional fuel-powered cars, though this positive correlation is not significant for all green travel modes. For walking, owners of new energy vehicles are more inclined to choose this green travel option under the inducement of carbon credits. However, the incentive effect is not as significant for the other two green travel options. Public transportation (bus/subway) is only one-tenth as effective as walking to new energy vehicle owners. As for cycling, the model results indicate that there is no significant relationship between the possession of new energy vehicles and the effect of CCIs.

Gender differences affect the effect of CCIs. For cycling and walking, carbon credits have a stronger effect on male individuals than on female individuals. This may be due to females prioritizing safety and comfort in their travels. However, for public transportation modes like buses and subways, the gender difference in the incentive effect is relatively small, possibly because public transportation is perceived as safe and convenient for both genders.

Travelers’ driving habits pose challenges to the effectiveness of CCIs in promoting green travel. Indeed, despite the original intention of carbon credits to encourage travelers to reduce driving cars, the results indicate that many travelers already have relatively fixed driving habits. This poses significant resistance to the effectiveness of carbon credits. For all green travel options, the model results show that the frequency of driving and the driving age of travelers are generally negatively correlated with the effect of CCIs, regardless of statistical significance. Although many variables related to driving habits were considered, they were ultimately excluded from the final ICLV models due to a lack of statistical significance. However, it can be inferred that commuters’ habitual driving behavior may resist the induction effect of carbon credits despite the lack of statistical significance in the result.

5.3. Carbon Credits’ Effect on Perceptions

Perceived usefulness has the most significant impact on the effect of CCIs for green travel. It positively correlates with this effect across all travel modes, indicating that individuals who perceive the benefits of using carbon credits are more motivated to adopt green travel modes. Their belief in the value and effectiveness of carbon credits drives their decision making and acceptance of this incentive.

Environmental awareness and perceived ease of use weakly positively correlate with the effect of CCIs. Higher environmental awareness makes individuals more receptive to using carbon credits as an incentive, aligning with their environmental values. Additionally, a user-friendly process for earning and redeeming carbon credits encourages engagement in green travel behavior.

Subjective norms, however, negatively correlate with the effect of CCIs. Individuals’ perception of social norms and others’ expectations influence their acceptance of carbon credits. Lower social support or expectations regarding the use of carbon credits for green travel may decrease individuals’ inclination to adopt the incentive. This negative correlation suggests the need for improved promotion and publicity of the carbon credit system to enhance its acceptance in society.

Unexpectedly, when the target green travel mode is walking, the degree of carbon credit cognition negatively correlates with the incentive effect, contrary to expectations. The possible reasons are that the current carbon credit system lacks systematic means of recording low-carbon travel behaviors, especially for walking, which involves short-duration and random travel. Improving the monitoring system for green travel behaviors when dynamically accumulating carbon credits for sustainable transportation choices is important.

The exclusion of the behavioral intention variable from all ICLV models due to a lack of statistical significance is noteworthy. It highlights the gap between public environmental intentions and actual behaviors, which is a limitation in many environmental policies. In practical terms, the findings emphasize the need for policymakers, organizations, and program designers to focus on the design and implementation of effective incentive systems to encourage green behavior. This could involve reevaluating and enhancing existing incentive structures or developing innovative approaches that align with individuals’ motivations and preferences.

Beyond our specific study, this observation has broader implications for the field of behavioral economics and environmental psychology. It underscores the complexity of human decision making in the context of sustainability and highlights the importance of considering external motivators and incentive mechanisms in designing effective interventions.

Although the acceptance level of carbon credits was quantified through latent variables, it does not directly equal to an incentive effect. The effective transformation of low-carbon intentions to behaviors requires reasonable incentive mechanisms, attractive reward schemes, educational campaigns, and other measures. Enhancing the actual effectiveness of carbon credits will encourage more people to choose green travel modes and achieve genuine environmental goals.

6. Conclusions and Future Research

This study explored daily travel mode choices with carbon credits as incentives using an ICLV model. Perceptions of carbon credits are included in the model as latent variables to examine its influence on travelers’ travel mode choices. The performance of the ICLV model is then compared with a traditional MNL model. Data from the questionnaire were used in the estimation and demonstrated the superiority of the ICLV model. The model results show that the ICLV model provides deeper insights into daily travel mode choices with carbon credits as incentives. Major conclusions obtained in the present work are summarized as follows. First, carbon credits were effective in promoting public transportation among commuters. Meanwhile, the impact of carbon credits varied across green travel modes, emphasizing the need to customize incentives according to individual trip features. Second, demographics also show a significant association with the incentive effect. Specifically, travelers with higher education levels show a positive correlation with the incentive effect, while higher incomes show a negative one. Older individuals were more responsive to the carbon credit system, and gender differences were also observed. Moreover, perceptions of carbon credits played a crucial role in travelers’ transport mode choices, where perceived usefulness has the most significant impact, followed by environmental awareness and ease of use. Subjective norms had a negative correlation, suggesting the need for improved societal attitudes through promotion. Behavioral intentions were not significant, highlighting the importance of effective incentive mechanisms. As can be concluded, customizing incentives and improving monitoring systems can enhance the role of carbon credits in promoting sustainable transportation choices.

Although online surveys may be less reliable than offline ones, our data meet validity requirements. A limitation of this study is the sample size, which could be improved by an additional survey. The samples may not fully cover all potential sociodemographics and travel attributes; yet, the model applied in this paper still provided valuable insights by analyzing carbon credits’ inducement effect on specific green travel modes with separate discrete choice models. Although the sociodemographics and travel attributes of the data may deviate slightly from the real world, this study proposed an approach that can be effectively used for a comparative analysis of carbon credits’ impact on diverse travel behaviors. For future research, exploring different incentive types of CCIs and employing longitudinal studies to track behavior changes over time may help us to better understand the incentive effect of CCIs on green travel. Establishing a comprehensive choice model that includes the overall mode choice distribution influenced by CCIs is also necessary in subsequent research.