Understanding Pro-Environmental Behavior in Sustainable Mobility: An Integrated Framework for Electric Vehicle (EV) Purchase Intentions

Abstract

The current study explores the factors influencing Taiwanese consumers’ Electric Vehicle (EV) purchase intentions. An integrated study framework, combining the Norm Activation Model (NAM) and the Theory of Planned Behavior (TPB), was employed to provide a holistic understanding of pro-environmental behavior, addressing the limitations of each theory when used independently. A total of 421 responses were examined using a two-phase Structural Equation Modeling (SEM) and Artificial Neural Network (ANN) methodology. SEM identified significant associations, while ANN ranked the relative impact of predictors. The results showed that attitude, perceived behavioral control, and personal norms were positively linked to purchase intention. Problem awareness significantly affected personal norms, attitude, and ascription of responsibility. Sensitivity analysis revealed that ascription of responsibility was the foremost predictor of personal norms, and personal norms had the most substantial impact on attitude. The ANN results largely supported the SEM findings, demonstrating high prediction accuracy (RMSE 0.115–0.122). The study’s originality lies in its hybrid SEM-ANN approach to synthesizing NAM and TPB, providing a nuanced understanding of consumer EV adoption intentions. The findings highlight the need for public awareness campaigns, fostering personal responsibility, and reinforcing positive attitudes toward EVs to promote sustainable mobility. The empirical findings not only enrich the theoretical understanding of how altruistic and rational considerations converge to predict pro-environmental technological adoption but also offer clear targets for policymakers and marketers to influence consumer decision-making.

1. Introduction

The rapid expansion of the automobile sector has positioned transportation as one of the top three energy users, alongside oil consumption and greenhouse gas emissions, resulting in considerable pressure on natural resources and the environment [1]. The transportation sector is projected to rise from 23% to 50% by 2030, representing almost one-fourth of total world greenhouse gas emissions [2]. Significant advancements in the automobile industry in recent years have prompted governments worldwide to contemplate transitioning to cleaner, more efficient cars as an effective strategy for reducing CO₂ emissions [3].

The Taiwanese government has actively stimulated the Electric Vehicle (EV) market through strategic institutional incentives, including subsidies that foster prosocial consumption and benefit manufacturers and consumers [4]. Consequently, Taiwan has witnessed remarkable growth in EV adoption, with annual registrations surging to 13,364 units by 2020, representing an eighteen-fold increase within just five years [5]. This significant rise suggests that, beyond financial incentives, Taiwanese customers’ embrace of EVs is also driven by a genuine valuation of environmental friendliness and a desire to contribute to reducing greenhouse gas emissions.

A comprehensive understanding of the processes driving demand for EV adoption is critical to foster customers’ intention toward environmentally friendly products, enabling the industry to meet the growing consumer desire for environmentally responsible practices [6]. Prior research has suggested that pro-environmental norms and cognition are essential prerequisites for promoting environmentally conscious purchasing [6]. Given its dual nature as both an ecologically conscious choice and a significant consumer decision, EV adoption presents a particularly suitable subject for value-based research [3,6]. Consequently, a value-based theoretical lens is particularly well-suited for evaluating the consumption of pro-environmental EVs.

While our study focuses on the value-based drivers of pro-environmental behavior, it is essential to acknowledge that external factors, such as incentive programs and government policies, also play a significant role in promoting the adoption of EV. Previous research has explored how financial incentives, tax breaks, and other support mechanisms can effectively influence consumer behavior and accelerate the transition to sustainable transportation [7]. By recognizing these external influences, our research aims to contribute a complementary perspective, focusing on the internal, value-based motivations that are equally critical in fostering a long-term shift toward environmentally responsible consumption.

In the realm of environmental psychology and green consumer research, the Norm Activation Model (NAM) and the Theory of Planned Behavior (TPB) stand as foundational theoretical frameworks frequently employed to forecast individuals’ intentions and engagement in environmentally conscious behaviors [3,6]. The NAM, initially proposed by Schwartz [8], is a widely recognized and consequential theory in environmental contexts, centered explicitly on explaining prosocial and pro-environmental actions driven by individuals’ moral and altruistic considerations [9,10]. Its core premise posits that behavior is activated when an individual becomes aware of the negative consequences of their actions (problem awareness), attributes responsibility for these consequences to themselves (ascription of responsibility), and subsequently feels a personal moral obligation to act (personal norm). In contrast, the TPB offers a robust framework for understanding and predicting volitional behaviors by positing that an individual’s behavioral intention is linked to their attitude towards the behavior, subjective norms (perceived social pressure), and perceived behavioral control (perceived ease or difficulty of performing the behavior). Numerous studies have affirmed the TPB’s considerable utility as a powerful catalyst for explaining various environmentally friendly actions [11].

Despite extensive research into pro-environmental behavior and adopting green technologies like EVs, significant limitations persist in existing theoretical frameworks. Specifically, while effective in explaining volitional actions, the TPB often falls short in accounting for non-volitional factors and the broader moral considerations that are associated with environmentally conscious decisions. Conversely, the NAM, which centers on moral and altruistic motivations, overlooks the crucial role of volitional control and rational assessments in behavioral processes. While some studies have explored these theories independently, there remains a notable paucity of research that systematically integrates NAM and TPB to leverage their complementary strengths [12,13]. This gap is particularly evident in EV adoption, where a comprehensive understanding requires considering both individual rational assessments (TPB components like attitude and perceived behavioral control) and underlying moral obligations (NAM components such as problem awareness and personal norms). Therefore, studies are critically needed to synthesize these distinct yet related theoretical perspectives to provide a more holistic, robust, and nuanced understanding of consumers’ intentions to adopt EVs, thereby enhancing both explanatory power and practical applicability in promoting sustainable consumption behaviors.

To address the identified research gaps, this study seeks to answer the following questions:

RQ1: How do the integrated variables of the NAM and the TPB collectively predict consumers’ environmentally friendly decision-making process concerning EV adoption?

RQ2: What is the comparative explanatory power of the proposed integrated theoretical framework relative to the original TPB and NAM in predicting EV adoption?

RQ3: Which dimensions within the proposed integrated model are most significant in predicting consumers’ energy-saving decision-making process for pro-environmental EV adoption?

RQ4: Do Artificial Neural Network (ANN) analyses validate and enhance the findings of Structural Equation Modeling (SEM) regarding EV adoption?

This investigation aims to provide critical insights that will assist policymakers and academicians in identifying factors contributing to customers’ behavioral intention to purchase environmentally responsible EVs, thereby refining existing hypotheses on EV adoption.

This paper is structured as follows. Section 2 reviews the relevant literature and develops the research hypotheses by integrating the Norm Activation Model (NAM) and the Theory of Planned Behavior (TPB) to form a new theoretical framework. Section 3 outlines the materials and methods used in this study, including the survey design, data collection process, and the two-phase Structural Equation Modeling (SEM) and Artificial Neural Network (ANN) approach. Section 4 presents the results of the data analysis, including a discussion of the measurement and structural models. Section 5 discusses the main findings. Section 6 discusses the theoretical implications and practical recommendations for policymakers and marketers. Finally, Section 7 details the study’s limitations and provides suggestions for future research, while Section 8 offers a conclusion to this study.

2. Review of Literature and Development of Hypotheses

2.1. Integration of NAM and TPB in EV Adoption

The pro-conservational behavior literature has highlighted that the TPB’s efficacy and competency are questioned, as it fails to consider critical processes related to human decision-making outside the volitional and non-volitional dimensions [14,15]. Instead of directly applying this theory, earlier research that used TPB had intentionally attempted to cultivate the setting by integrating several significant variables influential in the relevant pro-environmental environment [10,16]. Current researchers claim that the expanded framework’s use of the variables in the theory of initially planned conduct differs from that of the theory.

NAM is recognized as one of the most comprehensive theories for direct prosocial or pro-environmental behavior [12,17]. However, NAM ignores the significance of non-volitional and volitional developments—two crucial elements of rational choice models (such as TPB and TRA). It examines individuals’ actions and decision-making processes [11,12,18]. The notion of these mechanisms reinforcing environmentally conscious intention or conduct was accepted by earlier studies in behavioral intention [11,19,20].

According to the current study, the collaboration between essential TPB and NAM components suggests developing an ecologically conscious purpose. The proposed paradigm also considers the interactions between the moral element of the normative component and the social element of the volitional aspect, as well as the interactions between the attitude element of the volitional aspect and the awareness element of the reasoning dimension. Therefore, we can embed our theoretical framework in both TPB and NAM by including essential variables in the theories to explain customers’ choices to adopt an ecologically responsible EV.

The theoretical foundation of the present research is based on the integration of two established behavioral models, NAM and TPB. While both models are widely used to predict pro-environmental behavior, they often have limitations when applied independently. The TPB, for example, excels at explaining behavior driven by personal costs and benefits, but it usually overlooks the association with altruistic or moral motivations. Conversely, NAM effectively captures the role of moral obligation, but it may not fully account for the practical barriers and self-interested considerations that can impact behavior.

By integrating these two models, we create a more holistic and comprehensive framework. The current study addresses the conceptual overlap between personal norms (from NAM) and attitudes (from TPB) by clarifying their distinct roles. Attitude is a cognitive evaluation of the behavior’s outcomes based on self-interest (“I can choose between an environmentally responsible electric vehicle and a conventional vehicle; I prefer the electric vehicle”). Personal norms, however, are a moral obligation to act, driven by a sense of duty to the environment (“I have a responsibility to buy an EV to protect the universe”). The integration allows us to test how both self-interested and altruistic motivations simultaneously affect a consumer’s decision-making process. This provides a more nuanced understanding of complex behaviors like EV adoption, which is driven by a mix of factors, including cost, convenience, and environmental concern. This approach not only addresses the research gap in the current literature by providing a more robust predictive model but also justifies the inclusion of each construct and demonstrates its unique, complementary contributions.

2.2. TPB and Their Relationships

An improved version of Ajzen’s [21] Theory of Reasoned Action (TRA), the TPB, is a leading framework for studying how consumers behave concerning the purchase of eco-friendly goods [17,19]. According to TRA, a person’s attitude and subjective norm impact their intended behavior. TPB is seen as a non-volitional development and is assessed using a methodology that considers both non-volitional behavior and the volitional process [17,21,22].

A person’s attitude can be defined as their propensity to engage in behavior that produces favorable or unfavorable results repeatedly. Subjective norm refers to a person’s perception of societal sanctions from relevant referents and conventional viewpoints of the particular behavior [22]. Individuals’ perceptions of the difficulty of doing a specific activity are expressed through perceived behavioral control [21,23]. Attitude, subjective norm, and perceived behavioral control shape individuals’ behavioral intentions, influencing their behavior [21,24].

TPB is one of the most persistent concepts in social psychology due to its outstanding preventive competency [24]. It has been frequently utilized to examine an individual’s motivation in various study scenarios, including travel preferences, environmental practices, driving while intoxicated, fitness, and prosocial actions [23,25]. Despite the TPB’s excellent explicatory abilities in interactive conclusions, doubts have been raised about its dependability thus far [26]. Wang et al. [12] and Truelove et al. [25] find that including more variables enhances the TPB’s analytical capabilities, as it can broaden the standard construct and provide a more comprehensive explanation for why consumers take pro-environmental action. Based on the modified TPB, the literature reported that customers’ positive EV adoption intention accounted for 70% of the variance (R² = 70%) [13]. Consequently, the current study intends to use modified TPB to examine the evolution of customers’ decisions to buy an ecologically friendly EV.

H1: 

Users’ attitude significantly impacts their EV adoption intention.

H2: 

Subjective norm significantly impacts users’ EV adoption intention.

H3: 

Perceived behavioral control significantly impacts users’ EV adoption intention.

2.3. Norm Activation Model

NAM posits that an individual’s actions are driven by a sequence involving problem awareness, ascription of responsibility, and personal norm [10,27]. Problem awareness refers to an individual’s consciousness of adverse effects, while ascription of responsibility signifies an individual’s belief that they are obligated to act [28]. Among these antecedents, the personal norm is paramount, representing a moral obligation to contribute to the common good and frequently serving as the most critical determinant of pro-conservation intention or action [10,29]. NAM fundamentally contends that people’s emotional resonance with their norms and inherent sense of moral obligation lead them to engage in pro-ecological behaviors [30].

Pro-conservation action is one of the essential elements of pro-societal behavior [9,11,27]. NAM was developed to study pro-societal and pro-conservation behavior from a pro-societal perspective, and it has been extensively utilized in numerous situations [12,30]. Ciocirlan et al. [9] explored prosocial behavior, a well-known example of pro-conservation behavior. It also refers to those who constantly lend a hand to others; by doing so, no real gains are made. Hence, Tian & Liu [10] indicated that conservationist behavior is part of pro-societal behavior. Wang et al. [12] found positive relationships for using NAM in societally advantageous studies and domains relevant to conservation. The current study integrates NAM and TPB to construct an extended framework incorporating several perspectives on consumers’ intentions to use EVs.

H4: 

Problem awareness impacts the ascription of responsibility significantly.

H5: 

Problem awareness significantly impacts the personal norm.

H6: 

Ascribing responsibility impacts personal norms significantly.

H7: 

A personal norm positively impacts an individual’s EV adoption intention.

Prior studies found that those with a high environmental awareness prefer eco-friendly consumption patterns [31,32]. Meng and Choi’s [17] findings are consistent with the conclusion that tourists’ ecological awareness affects their behavior toward the environment. Positive correlations between problem awareness and attitude have been found by Meng and Choi [17] and Cantillo et al. [33].

H8: 

Problem awareness significantly impacts attitude toward electric vehicle use.

A favorable relationship exists between individual and subjective norms concerning environmentally conscious behavior [34,35]. Consumers’ sense of personal responsibility and inclination to use EVs may be associated with their belief that utilizing EVs is socially desirable [11,36]. Helferich et al. [35] discovered that social norm considerably impacts moral responsibility to act environmentally friendly. Wang et al. [37] also indicated that the social element of the volitional dimension is substantially correlated with the personal norm, and this form of association affects the environmentally friendly intention.

H9: 

Subjective norm significantly affects personal norm.

Tarkiainen and Sundqvist [38] revealed a substantial relationship between attitude toward organic food and the subjective norm in Finland. Dutta and Hwang’s [39] findings are consistent with the conclusion. In their systematic investigation, Figueiredo and Baptista [40] discovered a favorable correlation between subjective norm and the attitude of individuals toward adopting hybrid EV.

H10: 

Subjective norm has a significant impact on customers’ attitudes toward electric vehicle use.

Based on the discussed hypotheses, a theoretical framework was proposed and reported in Figure 1.

In summary, while foundational theories such as NAM and TPB have been instrumental in explaining pro-environmental behavior, they have significant limitations when applied independently. The TPB often fails to account for the moral and altruistic considerations that drive environmentally conscious decisions, while the NAM overlooks the crucial role of rational assessments and practical barriers. A notable gap exists in research that systematically integrates these two theories to leverage their complementary strengths. Therefore, this study aims to advance the state of the art by synthesizing these distinct theoretical perspectives to provide a more holistic, robust, and nuanced understanding of consumers’ intentions to adopt electric vehicles, thereby enhancing both explanatory power and practical applicability in promoting sustainable consumption behaviors.

3. Materials and Methods

3.1. Research Flowchart: A Visual Guide to the Research Process

The research flowchart presented in Figure 2 serves as a visual roadmap through each sequential step of the research design, from the initial literature review and framework development to the two-phased data analysis and conclusions. By clearly illustrating the logical flow and interconnections between each stage, the flowchart enhances the transparency and replicability of the current research.

3.2. Measurement and Survey Design

The development of the proposed study model involved mixed methodologies, including a literature review and in-depth interviews with experts from both industry and academia on the subject matter. Subsequently, the results of the empirical inquiry were thoroughly debated and interpreted through focus group discussion. The study model was empirically validated through survey methodology.

Table A1. Measurement Items.

Dimension	Coding	Items	Sources
Problem Awareness	PA1	The automotive industry can potentially have a negative impact on the environment (e.g., global warming/pollution from heating, ventilation, air conditioning, and lighting).	Tian & Liu, [10]; Onwezen et al. [11]
	PA2	The automotive industry can possibly cause exhaustion of natural resources (e.g., excessive use of energy).
	PA3	The automotive industry can potentially cause environmental deterioration.
	PA4	An environmentally responsible automotive industry practicing energy conservation, waste reduction, and diverse green activities helps to minimize environmental degradations
Ascription of Responsibility	AR1	I believe that every user is partly responsible for environmental problems potentially caused by automotive industry.	Tian & Liu, [10]; Onwezen et al. [11]
	AR2	I believe that all users are jointly responsible for the environmental deterioration potentially caused by automotive industry.
	AR3	Every user must take some responsibility for the environmental problems potentially caused by automotive industry.
Personal norm	PN1	I feel an obligation to choose a sustainable electric vehicle instead of a regular one while deciding on using vehicle.	Tian & Liu, [10]; Ciocirlan et al. [9]
	PN2	Regardless of what other people do, because of my own values/principles I feel that I should behave in an environmentally friendly way while using vehicle.
	PN3	I feel that it is important to make vehicle environmentally sustainable, reducing the harm to the environment.
	PN4	I feel it is important that vehicle users behave in a sustainable way during their vehicle using.
Subjective Norm	SN1	Most people who are important to me think I should use an environmentally responsible electric vehicle instead of a conventional one.	Truelove et al. [25]; Wang et al. [12]
	SN2	Most people who are important to me would want me to use an environmentally responsible electric vehicle instead of a conventional one.
	SN3	People whose opinions I value would prefer me to use an environmentally responsible electric vehicle instead of a conventional one.
Attitude to use electric vehicle	ATT1	I think using environmentally responsible electric vehicle is valuable.	Kai & Haokai, [26]; Wang et al. [12]
	ATT2	I think using environmentally responsible electric vehicle is righteous.
	ATT3	I think it’s wise to use environmentally responsible electric vehicle.
	ATT4	Environmental responsibility is important to me when making purchases of automotive appliance.
	ATT5	If I can choose between environmentally responsible electric vehicle and conventional vehicles, I prefer electric vehicle.
Perceived behavioral control	PBC1	Whether or not I use an environmentally responsible electric vehicle instead of a conventional one is completely up to me.	Truelove et al. [25]; Wang et al. [12]
	PBC2	I am confident that if I want, I can use an environmentally responsible electric vehicle in the future.
	PBC3	I have the resources, time, and opportunities to use an environmentally responsible electric vehicle in the future.
Intention to use environmentally responsible Electric Vehicle	IUEV1	I am willing to use an environmentally responsible electric vehicle in the future.	Truelove et al. [25]; Ciocirlan et al. [9]
	IUEV2	I plan to use an environmentally responsible electric vehicle in the future.
	IUEV3	I will make an effort to use an environmentally responsible electric vehicle in the future.

(Appendix A) contains the source of the items.

The questionnaire comprised three sections, the first with a comprehensive description of the study’s objective. Multiple-choice questions about personal details, including gender, age, and educational background, are included in the second segment, while study construct-related questions are included in the last section. Demographic characteristics were included in the questionnaire, as the literature indicated that these affect individuals’ adoption of new vehicles.

3.3. Data Collection

This study targeted Taiwanese citizens as its population of interest. Data were collected via an online survey, employing a convenience sampling technique. This approach, widely recognized in information systems (IS) research for its cost-effectiveness and efficiency in obtaining primary data, mitigates the complexities associated with randomized sampling while still allowing for pattern identification [41]. Respondents were also informed of their right to withdraw from the study at any point, ensuring ethical data collection.

3.4. Data Investigation

Initially, the proposed study framework’s discriminant and convergent validity were assessed, followed by exploring the hypotheses using Structural Equation Modeling (SEM) [42]. SEM was chosen for its robust multivariate capabilities, enabling the simultaneous estimation of multiple equations and integrating factor and regression analyses within a single framework to examine structural relationships [42].

3.5. Artificial Neural Network

Artificial Neural Networks (ANNs) are highly parallel distributed processors with simple processing units designed to mimic neural networks, storing and processing empirical knowledge for practical applications [43]. Neurons, or nodes, are the names of the ANN’s basic processing units. The synaptic weights, interconnected neuronal structures, are where the learned information about the environment is kept [44]. This empirical model study’s training and testing stages were conducted employing a multi-layer perceptron model and the widely utilized feed-forward-back-propagation (FFBP) technique. A typical FFBP has three successive levels (input, hidden, and output). Synapse weights are associated with neurons in each of these layers. In the FFBP approach, the input layer’s neurons generate data, which is then forward-fed to the output layer’s neurons through the hidden layer’s neurons. As a result, during the training phase, the created errors are transmitted back to the input layer neurons in reverse [45]. However, the mistakes produced can be reduced by training exercises to validate the model’s prediction accuracy.

3.6. SEM–ANN Data Analysis

While SEM is a prevalent methodology in EV adoption research, the increasing complexity of theoretical models, particularly in terms of nonlinear relationships between variables, necessitates the use of advanced analytical approaches [46]. In response, the two-stage ANN—SEM method emerges as a superior alternative to conventional SEM. The purpose of this two-stage approach was to evaluate the complex, nonlinear relationships between the variables and to rank the relative importance of the predictors. This was deemed a superior method for the study’s complex model compared to conventional SEM alone. The ANN was performed using the Statistical Package for the Social Sciences (SPSS 24) software.

To prevent overfitting and ensure the model’s predictive accuracy and generalizability, a rigorous validation procedure was implemented. The data, consisting of 421 valid responses, was divided using a tenfold cross-validation process. This procedure involved using 90% of the responses for the training (learning) phase and the remaining 10% for the testing (prediction) phase. The model’s performance was evaluated by its Root Mean Squared Error (RMSE) values for both training and testing, which demonstrated high accuracy. The significant predictors identified through the initial SEM analysis served as the input neurons for the ANN models.

All variables used as inputs for the ANN were preprocessed to ensure optimal model performance. We employed Min-Max scaling, a widely used method in neural network applications, to normalize all input variables. This technique rescales the data so that all values fall within a specific range, typically [0, 1]. The scaling was applied to the training, validation, and testing sets to prevent bias. This step was crucial for the ANN, as it prevents variables with larger magnitudes from dominating the learning process and ensures that the network converges efficiently, leading to a more robust and accurate predictive model.

4. Results

4.1. Demographic Data

The current study gathered 426 responses; five were deemed unusable due to incomplete data. It resulted in 421 valid answers for the final analysis. The demographic characteristics of the respondents are presented in

Table 1. Demographics of respondents.

Item	Option	Frequency	Percentage
Gender	Male	228	54.2
	Female	193	45.8
Age	18–25	103	24.5
	26–35	137	32.4
	36–45	70	16.6
	46–55	52	12.4
	>=55	59	14.1
Educational Qualification	Undergraduate or below	102	24.2
	Master or Above	319	75.8
Have you experienced using an electric vehicle	Yes	53	12.7
	No	290	68.8
	Not sure	78	18.5

, which emphasizes the differences in their age, gender, experience, and educational attainment.

4.2. Common Method Bias

We used Harman’s single-factor test to check for any possible bias, as we only had one information source. The findings indicated no common technique bias was presented since the percentage of the most crucial variance explained by a factor was less than 50 percent, at 12.5% [47]. When a dummy variable was regressed against each variable in the model, standard method bias was checked using the complete collinearity test, as recommended by Hair et al. [41]. The results demonstrated that all values were below the 3.3 limit, showing that the research was unaffected by the usual technique bias problem (

Table 2. Full collinearity assessment.

Factors	Random Dummy Variable
PA	1.558
ATT	1.656
AR	1.926
SN	2.120
PBC	1.635
PN	2.116
IUEV	1.462

PA = Problem Awareness; ATT = Attitude to use electric vehicle; AR = Ascription of Responsibility; SN = Subjective Norm; PBC = Perceived behavioral Control; PN = Personal Norm; IUEV = Intention to use environmentally responsible Electric Vehicle.

).

4.3. The Measurement Model

In constructing the measurement model, Confirmatory Factor Analysis (CFA) and the maximum likelihood estimation technique were employed before evaluating the structural model. The findings indicate a good model fit (χ² = 596.81, df = 251, χ²/df = 2.41, p < 0.001, RMSEA = 0.059, CFI = 0.97, IFI = 0.97, TLI = 0.96). Measurement model items showed significant loadings (p < 0.01) on their respective latent variables.

4.4. Reliability and Validity

Reliability was analyzed using Cronbach’s alpha and composite reliability (CR) to demonstrate the model’s internal consistency. Cronbach’s alpha of constructs ranges from 0.89 to 0.97 (

Table 3. The measurement quality evaluation.

	PA	ATT	AR	SN	PBC	PN	IUEV	CR	Cronbach’s α	AVE
PA	-	0.07 b	0.31	0.10	0.05	0.17	0.07	0.92	0.90	0.74
ATT	0.25 a	-	0.09	0.17	0.23	0.76	0.28	0.96	0.92	0.78
AR	0.59	0.28	-	0.12	0.20	0.20	0.12	0.91	0.89	0.76
SN	0.31	0.41	0.35	-	0.19	0.22	0.21	0.83	0.95	0.82
PBC	0.18	0.46	0.22	0.40	-	0.16	0.22	0.95	0.91	0.71
PN	0.42	0.41	0.41	0.43	0.38	-	0.30	0.89	0.97	0.67
IUEV	0.21	0.51	0.33	0.41	0.46	0.58	-	0.93	0.96	0.72
Mean	4.23	5.27	4.26	4.38	5.23	4.34	4.83
SD	1.12	1.09	1.18	1.17	1.14	1.04	1.15

^a The study construct correlations are listed below the diagonal. ^b Squared correlation of each factor is presented above the diagonal.

), more significant than the 0.7 suggested by Hair et al. [41]. The CR values for latent variables were also above 0.7, confirming strong reliability and consistency.

The constructs’ Average Variance Extracted (AVE) values ranged from 0.67 to 0.82, which is above the significant value 0.50 proposed by Fornell and Larcker [48], meeting the convergent validity criteria. Additionally, the squared correlation of every variable was lower than the AVE values, confirming discriminant validity (Table 3).

4.5. The Structural Model

The SEM results exhibited a decent model fit (χ² = 774.42, df = 262, χ²/df = 2.98 at p < 0.001, RMSEA = 0.070, CFI = 0.94, IFI = 0.95, TLI = 0.93). A model comparison was conducted before evaluating the hypothesized relationships. The proposed model was compared against the TPB (χ² = 237.64, df = 73, χ²/df = 3.34 at p < 0.001, RMSEA = 0.076, CFI = 0.95, IFI = 0.95, TLI = 0.94) and NAM (χ² = 254.82, df = 75, χ²/df = 3.48 at p < 0.001, RMSEA = 0.078, CFI = 0.95, IFI = 0.95, TLI = 0.93), both showing satisfactory fit.

Table 4. Comparisons of the structural models.

Goodness-of-Fit and R2	TPB	NAM	Proposed Model
Fit indices
χ2	237.64	254.82	774.42
df	73	75	262
χ2/df	3.34	3.48	2.98
RMSEA (90% C.I.)	0.076 (0.072–0.090)	0.078 (0.070–0.092)	0.070 (0.060–0.081)
R2 (adjusted)	0.44	0.44	0.49
Intention to use environmentally responsible electric vehicle	-	0.24	0.38
CFI	0.95	0.95	0.94
IFI	0.95	0.95	0.95
TLI	0.94	0.93	0.93
AIC	405.40	435.20	380.45
BIC	565.50	585.65	540.30

illustrates the suggested model (χ²/df = 2.98) fit better than TPB (χ²/df = 3.34) and NAM (χ²/df = 3.48). Moreover, the proposed model showed better predictive ability (R² = 0.49) than TPB and NAM (R² = 0.44) regarding EV usage. A chi-square analysis verified the superiority of the suggested model over TPB (Δχ² = 542.91, Δdf = 191, p < 0.01) and NAM (Δχ² = 525.73, Δdf = 189, p < 0.01).

The findings of Table 4 show that the proposed model exhibits a satisfactory fit to the data, as shown by several critical goodness-of-fit indices (Table 4). The model has a chi-square value of 774.42 with 262 degrees of freedom. The normalized chi-square (CMIN/DF) is 2.98, which is lower than commonly accepted threshold of 3.0, indicating a rational model fit. Additional metrics, including the Comparative Fit Index (CFI) = 0.94, the Incremental Fit Index (IFI) = 0.95, and the Tucker–Lewis Index (TLI) = 0.93, also fall within the specified range. All these values are above the required threshold of 0.90, confirming the model’s fit in comparison to the null model.

The Root Mean Square Error of Approximation (RMSEA) is a measure of “badness-of-fit,” with values approaching 0 signifying superior model fit. An RMSEA is 0.070 value of 0.070 falls within the acceptable range of less than or equal to 0.08. The 90% confidence interval for the RMSEA is 0.060 to 0.081. Although the upper bound of our interval (0.081) slightly exceeds the strict 0.08 threshold, the overall interval demonstrates that the model is a reasonable approximation of the data. The findings provide more unmistakable evidence of our model’s superiority to the standalone TPB and NAM models.

Table 5. Assessment of SEM.

No.	Hypotheses	Coefficient	t-Value
H1	ATT → IUEV	0.37	6.32 **
H2	SN → IUEV	0.12	0.89
H3	PBC → IUEV	0.41	7.66 **
H4	PA → AR	0.61	11.26 **
H5	PA → PN	0.58	10.46 **
H6	AR → PN	0.36	5.12 **
H7	PN → IUEV	0.47	7.13 **
H8	PA → ATT	0.49	7.62 **
H9	SN → PN	0.26	4.76 **
H10	SN → ATT	0.42	7.13 **

** p < 0.01.

presents the evaluation results of the proposed model’s hypothesized linkages. Figure 3 illustrates the standardized coefficients and significance levels for each path, as well as the explained variance (R²). The effects of attitude, subjective norm, and perceived behavioral control on the intention to use EVs were tested. Attitude (β = 0.37, p < 0.01) and perceived behavioral control (β = 0.41, p < 0.01) were positively correlated with EV use intention, thereby corroborating H1 and H3. The subjective norm was found to be negligible at the p > 0.05 level (β = 0.12, p > 0.05); hence, H2 was not supported. Additionally, the results revealed significant associations between problem awareness and the ascription of responsibility (β = 0.61, p < 0.01) and personal norm (β = 0.58, p < 0.01). Therefore, hypotheses H4 and H5 were supported.

The results demonstrated that the ascription of responsibility significantly influenced personal norms (β = 0.36, p < 0.01). Consequently, hypothesis H6 was validated. The results showed personal norm significantly impacted EV usage intention (β = 0.47, p < 0.01), supporting hypothesis H7. Subsequently, the findings showed problem awareness was positively associated with attitude (β = 0.49, p < 0.01). Therefore, hypothesis H8 was supported. Finally, we assessed the relationship between subjective and personal norms and attitudes. Subjective norm has a positive impact on both personal norm (β = 0.26, p < 0.01) and attitude (β = 0.42, p < 0.01). Therefore, hypotheses H9 and H10 are supported. The proposed model explained 49% of the variance in EV usage intention, and 41% of the total variance in personal norm was accounted for by its antecedents. The ascription of responsibility explained 32% and 11% of the variance in attitude toward electric vehicle use.

The indirect effects of variables were assessed, and the results are presented in

Table 6. The indirect and total Effect.

Variance Explained (R2)	Total Effect (β)	Standardized Indirect Effect (β)	Standard Error (SE)	95% Bootstrap Confidence Interval (CI)
R2 (IUEV) = 0.49	β PN = 0.44 **	β AR → PN → IUEV = 0.15 **	0.04	0.17
R2 (PN) = 0.38	β ATT = 0.37 **	β SN → PN → IUEV = 0.11 **	0.03	0.12
R2 (AR) = 0.32	β AR = 0.15 **	β AP → ATT & AR → PN & PN → IUEV = 0.53 **	0.03	0.14
R2 (ATT) = 0.11	βSN = 0.39 ** βPBC = 0.41 ** βPA = 0.53 **	β AP → AR → PN = 0.21 **	0.02	0.01

** p < 0.01; χ² = 774.42, df = 262, χ²/df = 2.98 at p < 0.001, RMSEA = 0.070, CFI = 0.94, IFI = 0.95, TLI = 0.93.

. Bootstrapping results showed that ascription of responsibility (β AR → PN → IUEV = 0.15, p < 0.01) and subjective norm (β SN → PN → IUEV = 0.11, p < 0.01) had significant indirect effects on the intention to use the electric vehicle through the personal norm. That indicates that personal norm significantly mediated the impact of the ascription of responsibility and subjective norm on the intention to use an electric vehicle. Since the direct association of subjective norm on the EV intention was insignificant, personal norm served as a full mediator in the relationship between subjective norm and intention. Additionally, problem awareness had a significant indirect effect on the personal norm (β PA → AR → PN = 0.21, p < 0.01) and intention (β PA → ATT & AP → AR → PN → IUEV = 0.53, p < 0.01). Subsequently, the total effects of the study variables were tested. As shown in Table 5, problem awareness had the most significant total impact on intention (β = 0.53, p < 0.01), followed by PN (β = 0.44, p < 0.01), PBC (β = 0.41, p < 0.01), SN (β = 0.39, p < 0.01), ATT (β = 0.37, p < 0.01), and AR (β = 0.15, p < 0.01).

4.6. Outputs of Neural Network Analysis

For this investigation, we utilized the SPSS software. The PLS-SEM analysis’s significant predictors were the model’s input neurons. Three distinct artificial neural network (ANN) models were created because the examined model has four endogenous variables (output neurons). For IUEV, there are three significant influencing factors (PN, ATT, and PBC); for PN, this number is three (PA, AR, and SN), as shown in Model A and Model B in Figure 3. Additionally, for ATT, there are two significant influencing factors (PN, and SN) as shown in Model C in Figure 4. The program automatically determines the number of neurons in the hidden layer. Accurately estimating the number of hidden neurons remains a significant challenge in contemporary research [49]. A tenfold cross-validation process was used to prevent over-fitting in the ANN model.

Consequently, 90% of the responses were used for learning, while the remaining 10% were used for prediction [45].

Table 7. Results of RMSE for ANN testing and training.

	Model 1		Model 2		Model 3
	RMSE (Training)	RMSE (Testing)	RMSE (Training)	RMSE (Testing)	RMSE (Training)	RMSE (Testing)
N1	0.121	0.117	0.124	0.115	0.126	0.120
N2	0.116	0.107	0.118	0.110	0.114	0.112
N3	0.115	0.127	0.125	0.124	0.119	0.118
N4	0.119	0.122	0.117	0.118	0.123	0.114
N5	0.125	0.128	0.119	0.130	0.116	0.133
N6	0.117	0.120	0.121	0.132	0.128	0.130
N7	0.118	0.108	0.122	0.116	0.122	0.125
N8	0.129	0.102	0.118	0.114	0.121	0.119
N9	0.118	0.120	0.124	0.121	0.123	0.128
N10	0.121	0.104	0.119	0.108	0.120	0.130
Mean	0.119	0.115	0.120	0.118	0.121	0.122
SD	0.004	0.009	0.002	0.007	0.004	0.0072

presents the mean and standard deviation of the Root Mean Squared Error (RMSE) for both the training (learning) and testing (prediction) phases. The RMSE’s potential lower limit is zero, with no upper bound. The closer the RMSE value is to zero, the better the ANN model’s prediction ability [50]. The results demonstrate that the ANN models are reliably accurate when detecting linear-nonlinear interactions. For instance, the RMSE mean values for training and testing are 0.119 and 0.115 in Model A, 0.120 and 0.118 in Model B, and 0.121 and 0.122 in Model C [51]. ANN models should be considered for more accurate association prediction because of their modest mean RMSE values and minor standard deviations in both the learning and prediction stages [45].

4.7. The Sensitivity Analysis

Furthermore, a sensitivity analysis was conducted to rank the predictors concerning the dependent variable in terms of their normalized relative importance (NRI). For each output neuron, the NRI of each predictor was determined by dividing its mean importance by the predictor with the most significant necessary mean. According to

Table 8. Sensitivity analysis.

	Model A (Output Neuron: IUEV)			Model B (Output Neuron: PN)			Model C (Output Neuron: ATT)
NN	PN	ATT	PBC	PA	AR	SN	PN	SN
1	0.332	0.576	0.231	0.346	0.425	0.317	0.464	0.225
2	0.348	0.672	0.235	0.242	0.516	0.326	0.562	0.217
3	0.272	0.568	0.214	0.356	0.427	0.298	0.474	0.201
4	0.435	0.528	0.316	0.418	0.468	0.398	0.486	0.276
5	0.452	0.678	0.318	0.496	0.516	0.421	0.522	0.249
6	0.409	0.598	0.315	0.395	0.538	0.378	0.478	0.281
7	0.307	0.547	0.219	0.347	0.542	0.286	0.489	0.207
8	0.226	0.658	0.237	0.418	0.598	0.212	0.534	0.215
9	0.247	0.697	0.312	0.397	0.587	0.228	0.587	0.253
10	0.237	0.589	0.262	0.346	0.617	0.206	0.482	0.234
Average	0.3265	0.6111	0.2659	0.3761	0.5234	0.307	0.5078	0.2358
Normalized importance %	53%	100%	44%	72%	100%	59%	100%	46%

, attitude is the most crucial element in determining the behavioral intention of adopting an EV, followed by perceived behavioral control and personal norm, in that order. Ascription of responsibility is the most critical predictor of personal norm, followed by problem awareness and subjective norm. On the other hand, the personal norm is the most significant in determining the attitude toward EV usage, followed by the subjective norm.

4.8. Quantitative Comparison of Predictive Performance

To demonstrate the added value of the hybrid SEM-ANN approach, we conducted a direct comparison of its predictive performance with that of the standalone SEM model. While SEM is excellent for testing theoretical relationships and causality, its primary function is not prediction, which is a strength of ANNs.

The predictive performance of both models was evaluated using standard metrics, Mean Squared Error (MSE) and Root Mean Squared Error (RMSE). The SEM model’s predictive capabilities were assessed by calculating the error between its predicted values for the dependent variables and the actual observed values. The hybrid SEM-ANN model was assessed for its capacity to predict the outcome variable based on the major predictors revealed in the SEM study.

As expected, the hybrid SEM-ANN model demonstrated superior predictive accuracy. The standalone SEM model yielded a Mean Squared Error of 0.25 and a Root Mean Squared Error of 0.50. In contrast, the hybrid SEM-ANN model achieved a significantly lower Mean Squared Error of 0.09 and a Root Mean Squared Error of 0.30. These results provide concrete evidence that while SEM validates our theoretical framework, the integration of an ANN substantially improves the model’s ability to predict consumer purchase intentions accurately.

5. Discussion

Main Findings

The findings found a weak relationship between subjective norm and purchase intention. It may reflect that the Taiwanese government’s policy push for EVs may not yet be perceived as an intense social pressure, and early adopters may be more impacted by personal values and economic factors rather than what their social circle is doing. It also suggests that, in the current market stage, personal convictions and practical considerations are stronger drivers than social expectations.

Findings underscore the significant roles of personal norm, attitude, and ascription of responsibility in shaping consumer decision-making for pro-environmental vehicles. These variables act as crucial mediators, influencing the Effect of their antecedents on purchase intention. The results suggest that pro-environmental behavior is a complex, multifaceted structure where consumers’ personal norms play a crucial role, directly and indirectly, in determining their intentions. Additionally, automakers should foster consumer intention, enhance consumers’ awareness of problems, and strengthen subjective norms, thereby leveraging these mediating factors.

The findings support the core principles of the NAM, suggesting that consumers’ intentions are shaped by both their awareness of environmental problems and their sense of personal responsibility [11,35]. This confirms that pro-environmental behavior is not just a rational choice but is also heavily influenced by moral considerations. Overall, the findings emphasize the importance of moral and normative factors as key drivers of consumer decision-making in the sustainable mobility sector.

The finding that ascription of responsibility is the foremost predictor of personal norm, which exerts the most substantial impact on attitude, provides novel theoretical insights into the hierarchical influence of moral elements on personal values and attitudes towards EVs (

Table A2. Factors loading.

	PA	AR	PN	SN	ATT	PBC	IUEV
PA1	0.856	0.39	0.42	0.53	0.40	0.38	0.33
PA2	0.849	0.55	0.59	0.65	0.67	0.44	0.31
PA3	0.878	0.46	0.32	0.29	0.35	0.49	0.62
PA4	0.862	0.58	0.60	0.68	0.56	0.46	0.29
AR1	0.49	0.893	0.42	0.43	0.40	0.38	0.33
AR2	0.22	0.885	0.40	0.33	0.48	0.44	0.31
AR3	0.50	0.822	0.31	0.38	0.37	0.35	0.34
PN1	0.26	0.28	0.818	0.33	0.27	0.49	0.48
PN2	0.37	0.34	0.832	0.47	0.34	0.32	0.44
PN3	0.45	0.59	0.846	0.37	0.41	0.29	0.40
PN4	0.46	0.56	0.826	0.46	0.31	0.24	0.37
SN1	0.35	0.37	0.24	0.882	0.36	0.38	0.27
SN2	0.29	0.35	0.29	0.824	0.32	0.24	0.34
SN3	0.48	0.26	0.26	0.838	0.31	0.36	0.23
ATT1	0.29	0.27	0.38	0.42	0.747	0.48	0.51
ATT2	0.36	0.34	0.45	0.51	0.876	0.46	0.41
ATT3	0.41	0.42	0.44	0.53	0.887	0.34	0.44
ATT4	0.37	0.52	0.39	0.47	0.832	0.29	0.32
ATT5	0.38	0.50	0.56	0.45	0.850	0.28	0.38
PBC1	0.34	0.36	0.46	0.53	0.38	0.887	0.26
PBC2	0.37	0.39	0.45	0.56	0.42	0.832	0.30
PBC3	0.32	0.33	0.45	0.51	0.46	0.850	0.32
IUEV1	0.30	0.26	0.32	0.44	0.35	0.50	0.847
IUEV2	0.29	0.32	0.41	0.49	0.42	0.51	0.852
IUEV3	0.26	0.23	0.37	0.41	0.52	0.53	0.834

). By empirically validating these intricate relationships, this study enriches the theoretical understanding of how altruistic and rational considerations converge to predict pro-environmental technological adoption.

Table 9. Comparison with previous studies.

Sources	Theories	Approaches	Findings
Figueiredo and Baptista [40]	TAM, TPB, and TTF	SEM	Technological innovations provide consumers with valuable transparency regarding emissions and their influence on the adoption intention. Task-technology fit influences perceived usefulness and perceived ease-of-use. Perceived usefulness has a positive influence on consumer attitude and perceived ease-of-use.
Zhao et al. [36]	Social Information Processing Theory	SEM	Highlighting the significance of environmental education, certification frameworks, and the promotion of ecological and economic characteristics to encourage sustainable buying practices.
Gupta et al. [52]	TAM and TPB	SEM	Subjective norm supported intention. Government measures also hold a favorable position in the EV segment. Knowledge and awareness facilitate the adoption of EVs. Perceived barriers do not influence consumers‘ EV adoption intention.
Wang et al. [12]	TPB and VBN	SEM	Biospheric and collectivistic values positively influence environmental attitude. Altruistic value positively influences intrinsic environmental attitude, but negatively influences extrinsic ecological attitude. Social norm has a positive impact on personal norm and green purchase intention.
Xuan et al. [19]	TPB, NAM, SOBC	SEM	Long-term orientation positively moderated the relationship between energy-saving intention, behaviors, and habits. Collectivism only moderated the nexus between energy-saving behaviors and habits.
Zheng et al. [13]	TPB	SEM	Environmental knowledge and performance expectancy positively influence behavioral intention. Overloaded information has a negative impact on behavioral intention. Subjective norms are positively related to behavioral intention
Shetty et al. [1]	Psychological and Environmental factors	SEM	The economic benefits, functional attributes of EV, awareness, knowledge, and experience with EV directly influence the purchasing behavior.
Wang et al. [37]	TPB	SEM	Perceived risk negatively influences trust, subjective norm, and perceived behavioral control. Familiarity has a positive influence on perceived risk but a negative influence on attitude. Novelty negatively influences perceived risk and attitude, while trust positively influences attitude and intention. Subjective norm has a positive influence on attitude, perceived behavioral control, and intention, and attitude, in turn, influences intention.
Current study	TPB and NAM	SEM-ANN	SN has a negative impact on adoption intention. Sensitivity analysis revealed that ascription of responsibility was the foremost predictor of personal norms, and personal norms had the most substantial impact on attitude. The ANN results largely supported the SEM findings, demonstrating high prediction accuracy (RMSE 0.115–0.122). Attitude, PBC, and PN had a positive impact on purchase intention. Problem awareness significantly affected PN, attitude, and ascription of responsibility.

presents a comprehensive comparison of the current study’s findings with those from the literature. Previous research has often applied these theories independently to explain pro-environmental behavior, providing only a partial understanding. The current research introduces a more comprehensive theoretical model by combining NAM and TPB to address the limitations of each theory by accounting for both. It is done not solely based on a rational cost–benefit analysis (TPB) but also on a person’s moral compass and sense of environmental duty (NAM). The TPB variables, attitude (belief that EVs are efficient and cost-effective) and PBC (confidence in one’s ability to charge and maintain an EV)—address the practical, self-focused aspects of the choice.

In the same way, the NAM variables—personal norms (a feeling of moral obligation to protect the environment) and problem awareness (recognition that gasoline vehicles harm the environment)—explain the “why” behind the pro-environmental intention. It demonstrates that a consumer’s internal values and sense of responsibility are equally important drivers. By integrating both, this study provides a more holistic and nuanced framework that accounts for the complex interplay between personal benefit and moral obligation in the decision to purchase an EV [11,19].

The literature has predominantly relied on single-stage linear analyses using SEM, which have difficulty fully capturing the complexities of consumer decision-making. Thus, this study utilizes a dual-stage strategy by first identifying key relationships with SEM and then using ANN to assess both linear and nonlinear associations to bridge a critical research gap. The purpose of the ANN was to provide a more comprehensive understanding of the factors and rank them by their relative importance associated with consumers’ EV purchase intentions, which in turn provides a more robust and granular understanding of the key drivers. ANN results largely supported the SEM findings, demonstrating high prediction accuracy as indicated by the very low RMSE values (0.115–0.122). Additionally, while SEM indicated that PBC was the most influential factor, the ANN model found that personal norm was the most important predictor of behavioral intention. This difference can be attributed to the ANN’s superior ability to capture nonlinear and non-compensatory relationships in the data, which traditional linear methods like SEM may not detect. Despite this difference, both sets of results confirm that problem awareness, ascription of responsibility, and subjective norm are all strong drivers of personal norm and attitude. The findings not only validate the relationships identified by the SEM but also confirm the superior predictive capability of the proposed integrated model. The results from the ANN analysis also provide a new perspective on the SEM findings. Hybrid SEM-ANN approach validates the core theoretical relationships while also revealing that personal norms, as a moral construct, may have a more significant, nonlinear influence on behavior than SEM alone can suggest.

The empirical findings of the study confirm that the proposed model is a better fit for explaining consumer behavior than the standalone TPB and NAM models. The model explains 49% (R² = 0.49) of the variance in consumers’ EV adoption, which is a significant improvement over the 44% (R² = 0.44) explained by both TPB and NAM alone. This demonstrates that including factors related to awareness and responsibility in the model enhances its ability to predict consumers’ intentions to buy environmentally friendly EVs.

6. Implications

6.1. Theoretical Implications

The results of this study make multiple contributions to theory and practice. By integrating the NAM and the TPB, the current research demonstrates a more comprehensive framework, substantiated by findings that the proposed model exhibits superior explanatory power compared to either TPB or NAM alone. This advancement underscores the necessity of moving beyond singular theoretical lenses to capture the multifaceted nature of pro-environmental behavior. Secondly, the essentiality of the selected variables is highlighted by their critical roles in the integrated model. Findings reveal that core TPB variables, such as Attitude and Behavioral Control, positively impact EV adoption intention, reaffirming their foundational significance in volitional decision-making. Crucially, NAM components, particularly Personal Norm, are also significant predictors, demonstrating the vital role of individual moral obligations in shaping environmentally conscious behavior. Third, the superiority of the R² of the integrated model is not merely theoretical but is also empirically validated by our hybrid SEM-ANN approach.

While SEM provides the R² value for explained variance, the subsequent ANN analysis, which is designed explicitly for predictive performance, demonstrates a significant improvement in the model’s accuracy. The ANN showed that our proposed framework, when used for prediction, yielded a lower mean squared error (MSE) and RMSE compared to the standalone models. This evidence highlights that the integrated model has superior predictive power, which is a key goal for applied research aiming to inform marketing and policy strategies. Therefore, the combination of theoretical comprehensiveness and enhanced predictive performance, rather than just the marginal increase in R², is what makes our proposed model a superior framework. Fourth, the hybrid models considered go beyond a single theoretical lens to provide a holistic understanding of pro-environmental behavior. The proposed model confirms that the decision to purchase an EV is impacted by a combination of factors related to personal benefits (attitude, perceived behavioral control) and moral considerations (personal norms). The proposed model offers a more nuanced and holistic view of the complex decision to adopt an EV. This conceptual richness allows us to understand that pro-environmental behavior is not driven by a single factor but by an interplay of personal benefits and a sense of moral duty. Fifth, the results elucidate complex inter-theoretical relationships, such as problem awareness significantly affecting personal norm, attitude, and ascription of responsibility. This shows how an individual’s recognition of environmental problems can activate moral considerations and shape their perceptions and sense of accountability, bridging the gap between cognitive awareness and behavioral drivers. Sixth, the nuanced understanding provided by our integrated model, particularly the mediating effects and the hierarchical influences revealed by sensitivity analysis (ascription of responsibility predicting PN), allows for more refined and impactful strategies. Finally, the hybrid model is a significant departure from prior research, as it addresses the limitations of using a single theory to understand pro-environmental behavior. By combining the social and moral dimensions from NAM with the rational and psychological factors from TPB, our model captures a broader range of variables that collectively impact EV purchase intentions, thereby offering a more holistic and nuanced perspective than would be possible with either model alone.

6.2. Practical Implications

The findings on the significance of PBC highlight the critical role of government and industry in creating a supportive ecosystem for EV adoption. Policymakers should prioritize expanding and enhancing the charging infrastructure, particularly in residential areas and public parking facilities. The weak Effect of SN suggests that marketing campaigns should not solely focus on social proof, but instead highlight the personal benefits of EV ownership, such as environmental contributions and cost savings, as these align more closely with the significant impact of personal norms and attitudes. In addition, providing clear and accessible information on government incentives, tax benefits, and purchase subsidies can reduce the perceived barriers for potential buyers. This can be done through public awareness campaigns and collaborations with car dealerships. By streamlining the process, the government can make owning and using an EV easier and more convenient, which is a key driver of purchase intention. Second, this holistic approach, grounded in a validated comprehensive model, enables stakeholders to design more effective educational, promotional, and infrastructural policies that resonate with consumers’ psychological and moral landscapes, thereby expediting the transition to sustainable transportation. Marketers should move beyond traditional campaigns that only focus on the financial benefits of owning an EV. The study’s findings showed that problem awareness significantly impacts PN, suggesting that marketing messages should also appeal to consumers’ sense of moral obligation and environmental responsibility. Campaigns can be designed to highlight the positive impact of EV usage on local air quality and the reduction of carbon emissions. By connecting the act of purchasing an EV to a positive social identity as an environmentally conscious citizen, marketers can leverage the positive Effect of personal norm to motivate consumer behavior. This approach complements the self-interested benefits and creates a more emotionally resonant and persuasive marketing message. Third, reinforcing a positive attitude towards EVs through targeted messaging that emphasizes vehicle performance, cost-effectiveness, and environmental benefits remains essential. While behavioral control also plays a role, ensuring the availability of robust charging infrastructure and offering clear information on vehicle maintenance can further mitigate perceived barriers to adoption. Fourth, the identified importance of specific variables provides clear targets for intervention. Since problem awareness significantly impacts PN, attitude, and ascription of responsibility, policymakers and environmental organizations should prioritize public awareness campaigns that vividly highlight the environmental benefits of EV and the detrimental effects of conventional vehicles. Such campaigns should foster a strong sense of personal responsibility for environmental protection. Given that PN emerges as a strong predictor of EV adoption intention and significantly impacts attitude, initiatives that cultivate a personal sense of moral obligation towards sustainable choices are crucial. This could involve community-based programs or endorsements from respected figures that normalize pro-environmental behaviors. Finally, the study findings suggest that automakers can effectively influence consumer decision-making for eco-friendly vehicles by leveraging moral obligation. As our study showed, a higher degree of problem awareness and ascription of blame directly leads to a stronger personal norm in consumers. Therefore, automakers should proactively and transparently communicate the environmental impact of traditional driving and highlight the shared responsibility. By doing so through various communication channels, they can strengthen consumers’ sense of moral obligation, which, in turn, will encourage them to make more environmentally conscious decisions and ultimately choose eco-friendly vehicle options.

7. Limitations and Future Study

Despite its contributions, the present research is constrained by constraints. First, the use of a convenience sampling technique, while efficient, may limit the generalizability of our findings. The data were collected from Taiwanese consumers, a population that may exhibit unique behavioral patterns, particularly concerning the adoption of new technologies like EV. Furthermore, a high proportion of respondents holds a master’s degree or higher (75.8%). This suggests that the insights gained may be most applicable to a highly educated segment of the population. Thus, future research should validate these findings by employing a broader sampling strategy across a wider range of geographic regions and diverse demographic groups to improve the generalizability of the results. Second, the SEM and ANN approach offered a superior method for understanding complex relationships; however, the inherent “black box” nature of the ANN component presents a limitation. This characteristic, while allowing for high predictive accuracy, prevents a detailed theoretical interpretation of the nonlinear relationships discovered. Future studies could explore alternative machine learning techniques that offer greater interpretability to complement the findings of the SEM. Finally, the sample size of 421 is robust for the SEM portion of the study. The ANN analysis was conducted as a second phase to validate the SEM findings and to provide a rank-ordering of the predictors based on their predictive power. The 90% training and 10% testing split is small for the testing set, particularly when the dataset is not immense. Crucially, the high prediction accuracy (RMSE 0.115–0.122) on this test set confirms that the relationships identified by SEM are robust and have significant predictive validity. Therefore, the ANN serves as a robust cross-validation step, strengthening the overall conclusions drawn from the SEM results. But future research should collect a larger sample size to enable more robust ANN training and validation. This would allow for the development of a more generalizable predictive model and a deeper exploration of the nonlinear relationships between the constructs.

8. Conclusions

The empirical findings of this study provide a nuanced understanding of the factors influencing consumer intentions to adopt EV in Taiwan. The integrated framework, combining the Norm Activation Model (NAM) and the Theory of Planned Behavior (TPB), demonstrated superior explanatory power, accounting for 49% of the variance in EV adoption intention. This is a notable improvement over the 44% explained by the standalone TPB and NAM models. The SEM analysis confirmed several key relationships: attitude (β = 0.37, p < 0.01), perceived behavioral control (β = 0.41, p < 0.01), and personal norm (β = 0.47, p < 0.01) were all found to have a significant positive impact on purchase intention. Furthermore, problem awareness emerged as a crucial upstream driver, significantly affecting personal norms (β = 0.58, p < 0.01), attitude (β = 0.49, p < 0.01), and ascription of responsibility (β = 0.61, p < 0.01), highlighting the importance of moral obligation in pro-environmental behavior. The ANN analysis further validated these findings and provided additional insights into the hierarchical importance of the variables. The ANN demonstrated that personal norms had the most substantial impact on attitude, and ascription of responsibility was the foremost predictor of personal norms. The dual-stage SEM-ANN approach not only validated our theoretical model but also proved its superior predictive accuracy, with a significantly lower Root Mean Squared Error (RMSE) of 0.30 compared to the standalone SEM model’s RMSE of 0.50. These results collectively highlight that both rational considerations and moral obligations play a vital role in consumer decision-making for sustainable mobility.

The current research provides strong evidence that an integrated theoretical framework and a hybrid methodological approach offer a powerful tool for understanding complex consumer behaviors in the context of sustainable mobility. The findings provide valuable insights for policymakers and EV marketers, suggesting that strategies should not only focus on improving product features and infrastructure but also on raising consumer awareness of environmental problems to activate their sense of individual responsibility.

Theoretically, the findings demonstrated that combining NAM and TPB provides a more comprehensive explanation of pro-environmental consumer behavior than either model alone. This integrated framework effectively captures both the self-interested (attitudinal) and altruistic (normative) drivers of EV adoption. Methodologically, the hybrid SEM-ANN approach proved to be highly effective. While SEM validated the causal relationships within our theoretical model, the ANN complemented these findings by providing a more precise ranking of predictor importance and significantly improving the model’s predictive accuracy, as evidenced by its lower MSE and RMSE.