Taxpayers’ Willingness to Pay for Global Decarbonization via Renewable Energy Official Development Assistance: A Discrete Choice Experiment in South Korea

Abstract

South Korea’s official development assistance to the energy sector has increased steadily over the past decade, reaching USD 232.20 million in 2024. Yet public willingness to pay for renewable energy official development assistance remains largely unknown. This study uses a discrete choice experiment with 1000 nationally representative South Korean respondents and a mixed logit model to estimate marginal willingness to pay for key project attributes, including electrification, greenhouse gas reduction, firm expansion, expert training, and reputation enhancement. The results show that greenhouse gas reduction and expert training receive the highest willingness to pay, followed by firm expansion. Electrification and reputation enhancement receive relatively low support. The findings also reveal substantial preference heterogeneity, with younger and nationally oriented respondents placing greater value on economic returns. These results provide new donor country evidence on public preferences for renewable energy official development assistance and offer policy implications for designing a more climate-focused and socially supported green aid portfolio.

1. Introduction

Official development assistance (ODA) is a key policy instrument through which donor governments support economic development, welfare improvement, and institutional capacity building in developing countries. In recent years, ODA has also become increasingly salient as a tool for addressing global public goods, particularly climate change mitigation and energy transition, while simultaneously serving donor country interests such as geopolitical influence, private sector internationalization, and soft power projection. Among the various sectors of development cooperation, renewable energy (RE) ODA has attracted growing attention because it can contribute both to decarbonization in recipient countries and to the broader diffusion of low-carbon technologies and business opportunities abroad.

The strategic landscape of ODA has undergone a fundamental paradigm shift, evolving from a traditional altruistic framework to a multifaceted policy instrument that integrates global public goods with national interests. This transition is explicitly underscored in recent supranational deliberations and national policy blueprints. For instance, the G20 New Delhi Leaders’ Declaration emphasized the critical role of scaling up blended finance and mobilizing private capital to accelerate the energy transition in developing nations [1]. Similarly, the Development Assistance Committee (DAC) of the Organization for Economic Cooperation and Development (OECD) has recently highlighted that green ODA must serve as a catalyst for ‘green industrialization’ in recipient countries while fostering technological synergies with donor country enterprises [2].

In alignment with these global trends, the South Korean government, through its Third Comprehensive Strategy for International Development Cooperation (2021–2025), has formally committed to expanding its ‘Green New Deal ODA’ to address the dual imperatives of global climate action and the internationalization of its domestic RE industry [3]. These high-level mandates, further detailed in the latest annual implementation plans, provide a robust political rationale for evaluating ODA not merely as a fiscal transfer, but as a strategic investment supported by domestic taxpayers.

This study uses the OECD Creditor Reporting System (CRS) data for 2024, based on bilateral ODA disbursement flows to developing countries measured in current USD prices.

Table 1. Energy sector official development assistance (ODA) scale and share among major Organization for Economic Cooperation and Development (OECD) donors in 2024.

Countries	Total ODA (USD Million)	Energy ODA (USD Million)	Share of Energy ODA in Total ODA (%)
Development Assistance Committee countries	174,030.10	7983.82	4.6%
South Korea	3570.64	232.20	6.5%
Germany	28,897.84	2880.60	10.0%
Japan	17,084.81	1394.57	8.2%
France	11,330.53	600.56	5.3%
The United States	59,089.51	773.98	1.3%
The United Kingdom	13,004.05	451.68	3.5%
Italy	3311.86	264.59	8.0%
Canada	7050.38	338.62	4.8%

Note: Data are based on OECD Creditor Reporting System (CRS) ODA disbursement flows to developing countries for 2024, expressed in current USD prices [4].

compares the scale and share of energy sector ODA (CRS code 230) among major OECD DAC donors, expressed as a proportion of total bilateral ODA.

Germany leads with USD 2880.60 million in energy ODA (10.0% of total ODA), followed by Japan (USD 1394.57 million, 8.2%) and the United States (USD 773.98 million, 1.3%) [4]. South Korea’s energy ODA totals USD 232.20 million, representing 6.5% of its total ODA (USD 3570.64 million)—exceeding the DAC average of 4.6% and demonstrating strong policy prioritization of the energy sector. Despite this relative emphasis, South Korea’s absolute volume remains modest compared to major donors, suggesting room for expansion. Table 1 thus illustrates critical disparities between policy prioritization and resource allocation, underscoring the need to understand public preferences for RE ODA components amid competing fiscal priorities.

These policy ambitions make it increasingly important to understand how the public values different components of RE ODA, especially under fiscal constraints and competing domestic priorities. The political context further strengthens the need for such an assessment. Across many donor countries, aid budgets face greater scrutiny as taxpayers demand clearer evidence of domestic relevance and policy effectiveness. In the United States, the Trump administration moved to dismantle the United States Agency for International Development and argued that foreign aid should better serve national interests, illustrating the broader global shift toward more transactional and domestically justified aid politics. Korea is not immune to this trend. Official surveys show that public support for ODA, while still relatively high, has declined in recent years, alongside rising negative perceptions. In a climate of intensifying nationalism and economic uncertainty, securing sustained public backing for ODA requires credible evidence that aid spending is both effective and aligned with taxpayer preferences.

RE ODA is especially relevant in this setting because it embodies multiple, sometimes competing, policy objectives. On the one hand, it can expand electricity access, reduce greenhouse gas (GHG) emissions, and support development outcomes in recipient countries. On the other hand, it may also generate donor-side returns through overseas business expansion, human capital formation, and national reputation gains. Recent studies have begun to document these effects empirically. RE ODA improves electrification in sub-Saharan Africa [5]. Energy aid also helps lower recipient country carbon emissions [6]. In addition, aid flows can facilitate firms’ internationalization [7], mobilize private investment for energy transition [8], and strengthen RE capacity [9]. RE ODA may further contribute to donor country soft power and national brand value [10]. Moreover, ODA can generate geopolitical returns for donor countries, including improved international standing and policy influence [11].

Despite this growing body of literature, most existing studies evaluate ODA effectiveness from the recipient country perspective or examine macro-level aid outcomes rather than the preferences of taxpayers in donor countries. This creates an important gap in the evaluation of RE ODA, particularly for emerging donors such as South Korea, where public support is increasingly necessary for expanding the green aid budget. To address this gap, this study estimates South Korean households’ willingness to pay (WTP) for the main attributes of RE ODA using a discrete choice experiment (DCE) and a mixed logit (MXL) model. The DCE framework is well suited to this task because it allows respondents to trade off multiple policy attributes under hypothetical budget constraints, while the MXL model captures preference heterogeneity across individuals and provides more realistic estimates than homogeneous choice models.

Using survey data from 1000 households, this study examines five key attributes of RE ODA: electrification, GHG emission reduction, support for South Korean firms’ overseas expansion, training of global experts, and enhancement of South Korea’s international reputation. These attributes reflect both recipient-side development benefits and donor-side strategic benefits, allowing us to evaluate RE ODA as a multi-dimensional policy package rather than a single-purpose transfer. By estimating marginal WTP (MWTP) and relative importance (RI) weights, the study identifies which project outcomes the South Korean public values the most and where support is comparatively weak. Such evidence is essential for designing aid programs that are not only internationally effective but also politically sustainable at home.

This study makes three contributions. First, it provides one of the first donor country valuations of RE ODA in an emerging donor context, thereby extending the literature beyond recipient-focused impact assessment. Second, it offers attribute-level monetary estimates that can directly inform RE ODA portfolio design, budgeting, and public communication strategies. Third, it helps explain how domestic preferences may shape the future of green aid expansion in Korea, especially as the government seeks to raise the share of green ODA and strengthen support for climate-oriented development cooperation. In this sense, the paper links the economics of stated preference valuation with the political economy of foreign aid and climate finance.

The remainder of the paper is organized as follows: Section 2 reviews the relevant literature. Section 3 reports an overview of DCE and describes the DCE attributes, design, and survey implementation. The fourth section presents the random utility framework and the MXL specification. The fifth section reports the estimation results, MWTP values, and policy implications. The final section provides the conclusion.

2. A Review of Related Literature

This section systematically reviews prior scholarship to substantiate the DCE methodology for donor-perspective valuation of RE ODA and to empirically justify the core attributes detailed in

Table 2. Comprehensive literature review: effectiveness of renewable energy official development assistance (ODA).

Sources	Countries	Methods	Key Attributes
Chapel [5]	Organization for Economic Cooperation and Development (OECD) Development Assistance Committee (DAC)	Project-level empirical analysis	Renewable energy ODA projects significantly increase electrification rates in underserved regions.
Liu et al. [6]	OECD DAC	Mediation effect analysis	Energy aid reduces carbon emissions by improving technical efficiency and energy structures.
Paredes Nachón et al. [7]	OECD DAC	Panel data analysis	ODA flows act as a catalyst for donor firms to enter recipient markets.
Bashir et al. [9]	OECD DAC	Dynamic panel analysis	Renewable energy ODA accelerates decarbonization and strengthens the adaptive capacity of local energy systems.
Bau and Dietrich [11]	OECD DAC	Working paper analysis	ODA generates geopolitical influence and builds a pool of international experts for the donor.
Kim et al. [14]	The United States and China	Experimental survey research	Recipients favor transparent strategic aid, which positively shifts public perception of the donor.
Tawiah et al. [8]	OECD DAC	Econometric modeling	ODA effectively mobilizes private foreign investment, bridging the financial gap for energy transitions.
Yoo [10]	Japan	Case and policy analysis	ODA serves as a crucial tool for soft power, enhancing the national brand and geopolitical prestige.

[12,13]. Organized into two complementary dimensions, the review first establishes the methodological robustness of DCEs in non-market contexts before synthesizing evidence on RE ODA’s multifaceted outcomes, ensuring that attribute levels reflect realistic policy impacts [5,6,7].

2.1. Methodological Robustness of DCEs for Multi-Attribute ODA

DCEs, predicated on Lancaster’s characteristics theory and random utility maximization, enable precise elicitation of preferences and MWTP for non-market goods by simulating realistic trade-offs across attributes [12,13]. The empirical validity of DCEs is well-documented across environmental economics. This includes valuing biodiversity enhancements and air quality improvements, as well as assessing RE deployment policies. In particular, MXL extensions effectively account for unobserved heterogeneity and scale issues [13,15,16]. In public policy arenas, DCEs outperform simpler stated preference tools by accommodating complex bundles, yielding policy-relevant MWTP ratios under budget constraints [15].

Paradoxically, despite this versatility, DCEs remain underutilized in ODA evaluation, particularly for RE aid’s hybrid nature. Dominant paradigms—ex-post instrumental variable regressions on aid flows or qualitative case narratives—focus on aggregate recipient impacts, sidelining donor taxpayers’ prospective valuations of trade-offs between humanitarian (e.g., access expansion), environmental (e.g., decarbonization), and strategic (e.g., commercial, reputational) dimensions [17,18,19]. RE ODA’s dimensionality exacerbates this: projects embed recipient welfare, planetary benefits, and donor returns, defying unidimensional metrics [11]. Our DCE-MXL application innovates by operationalizing these layers, bridging methodological maturity with aid policy voids [13].

2.2. Empirical Evidence for RE ODA Attributes

Table 2 distills seminal empirical findings on RE ODA effectiveness across DAC donors, informing our five attributes: electrification, GHG reductions, firm expansion, expert training, and reputation. This multi-disciplinary synthesis—spanning development economics, energy policy, and international relations—ensures attributes capture verifiable outcomes with scaled levels for experimental realism. Recipient-centric attributes draw from impact evaluations. Chapel [5] quantifies RE ODA’s causal role in sub-Saharan electrification (project-level RDD: +15–25% access), directly motivating our attribute (0–1500 beneficiaries). Liu et al. [6] employ mediation analysis on DAC data, revealing that energy aid curtails recipient emissions (β = −0.12 to −0.18, p-value < 0.01) through structural shifts. Kablan and Chouard [20] extend this to energy poverty–climate co-benefits (estimates from instrumental variable estimation method: −8% emissions per aid tranche), validating GHG reductions (0–15,000 tCO₂e).

Donor-oriented attributes highlight spillovers. Paredes Nachón et al. [7] use firm-level panels to show that ODA boosts donor exports (+11.7%) and entry (fixed effect models), underpinning firm expansion (0–5 firms). Tawiah et al. [8] demonstrate ODA–Foreign Direct Investment synergies in transitions (estimate from general method of moment: elasticity = 0.26–0.32), amplifying commercial rationale. Yoo [10] frames energy ODA as soft power via Japan’s cases (brand uplift: +12–18%). Kim et al. [14] and Sinanoglu [21] via surveys and experiments affirm recipient positivity toward strategic transparency (+16–20% trust), grounding reputation (no/yes). Bau and Dietrich [11] trace ODA’s human capital reflux (e.g., expert networks yielding 15–25% geopolitical leverage), supporting training (0–30 experts). Table 2 explicitly maps these studies to attributes, with levels calibrated to observed project scales for external validity.

2.3. Research Gap and Contributions

The literature robustly evidences RE ODA’s discrete channels—electrification and decarbonization [5,6], internationalization [7,8], reputation [10,14], and expertise [11]—yet fragments them across recipient/macro lenses, omitting integrated donor WTP amid green aid fiscal debates [2]. No study deploys choice experiment (CE)-MXL to rank taxpayer trade-offs in emerging donors like South Korea, where ODA/ gross national income lags (0.21%) despite green ambitions [22]. We innovate by bundling these into DCE scenarios, deriving attribute MWTP to reframe RE ODA from input costs to multi-dimensional utility, furnishing ex-ante tools for portfolio optimization [13].

3. Methods

3.1. Method: DCE

DCEs operationalize multi-attribute utility theory, which posits that decision-makers evaluate alternatives by weighting attributes according to their RI to derive overall utility [23,24]. DCEs extend multi-attribute utility theory empirically by collecting stated choices over attribute-varying scenarios, enabling estimation of attribute-specific utilities and trade-offs for complex goods like RE ODA, where humanitarian, environmental, and strategic dimensions coexist [23,25].

Unlike contingent valuation, which elicits WTP for a single, holistic good via direct elicitation, DCEs decompose value into constituent attributes, revealing internal structure and marginal contributions [26]. For RE ODA—encompassing electrification access, emissions mitigation, firm internationalization, expert capacity building, and reputational gains—contingent valuation yields only aggregate support (e.g., “pay for ODA?”), obscuring policy levers [27]. DCEs, by contrast, simulate realistic bundles, quantifying how attribute tweaks optimize societal value, thus informing targeted program design over binary yes or no framing [15].

3.2. Selection of Attributes

Attribute selection is pivotal for DCE validity, ensuring attributes are policy-relevant, empirically grounded, statistically independent, cognitively accessible, and limited in number (4–7) to avoid respondent overload [25,28]. Rigorous processes thus prioritize policy salience, empirical grounding, attribute independence, and respondent comprehension—typically capping at 4–8 to balance realism with feasibility [29]. Poorly chosen attributes risk omitted variable bias, collinearity, or cognitive overload, undermining welfare estimates [30].

Table 3. Renewable energy official development assistance (ODA) attributes and levels.

Attributes	Descriptions	Levels
Electrification	Number of people gaining electricity access post-project	Level 1: 0
		Level 2: 100 people
		Level 3: 500 people
		Level 4: 1500 people
Reduction	Greenhouse gas emissions avoided via renewable energy transition	Level 1: 0
		Level 2: 100 tCO2e
		Level 3: 5000 tCO2e
		Level 4: 15,000 tCO2e
Expansion	Number of firms expanding overseas via ODA	Level 1: 0
		Level 2: 1 company
		Level 3: 2 companies
		Level 4: 5 companies
Training	Number of experts trained through renewable energy ODA	Level 1: 0
		Level 2: 5 people
		Level 3: 10 people
		Level 4: 30 people
Reputation	Enhancement of national reputation from ODA project	Level 1: No
		Level 2: Yes
Price	Increase in annual tax on household income	Level 1: KRW 0
		Level 2: KRW 1000
		Level 3: KRW 2000
		Level 4: KRW 5000
		Level 5: KRW 10,000

Note: Level 1 of each attribute refers to the business-as-usual state.

explicitly details the five attributes and their levels, derived from rigorous literature synthesis and pre-testing.

In this study, attributes were curated via systematic literature synthesis on ODA effectiveness, distilling high-impact outcomes while aligning with RE ODA’s hybrid rationale (Table 2). Drawing on typologies of aid motivations, we operationalized three domains: national interest, humanitarian, and communitarian or global commons—adapted from frameworks distinguishing moral imperatives, strategic gains, and mutual or public goods [22]. National interest attributes capture donor returns (firm expansion, expert training, and reputation); humanitarian targets sustainable development goals (SDGs) 7.1 (universal energy access); and communitarian addresses SDG 13.2 (climate integration) [31].

Cognitive burden was mitigated by limiting to five attributes plus price, with levels scaled to real-world RE projects (e.g., 100–1500 beneficiaries from Chapel [5]). Lay terminology supplanted jargon (e.g., “people gaining electricity” vs. “electrification rate”), augmented by pictorial aids and pre-tests for accessibility [32]. Table 3 delineates the resultant set:

• National interest: Firm expansion (South Korean enterprises entering overseas markets; 0–5 firms; Paredes Nachón et al. [7]); expert training (global specialists capacitated; 0–30; Bau and Dietrich [11]); and reputation enhancement (national image uplift; no or yes; Yoo [10]).
• Humanitarian: electrification (households gaining access; 0–1500; SDG 7.1; Chapel [5]).
• Communitarian: GHG reductions (tCO₂e avoided; 0–15,000; SDG 13.2; Liu et al. [6]).

This parsimonious, theory-driven selection enhances credibility, linking micro-preferences to SDGs while enabling trade-off revelation. The price attribute, framed as an increase in annual tax on household income, serves as the critical budget constraint in this DCE, enabling monetary valuation of non-market RE ODA outcomes via MWTP estimates. Five discrete levels—KRW 0, 1000, 2000, 5000, and 10,000 per annum—were calibrated to plausible policy-relevant increments, reflecting realistic fiscal burdens on South Korean households (mean monthly household income of KRW 4.88 million).

3.3. Design of Choice Sets

Full factorial enumeration of attribute levels from Table 3—electrification (four levels), GHG reductions (four), firm expansion (four), expert training (four), reputation (two), and price (five)—yields 2560 (=4 × 4 × 4 × 4 × 2 × 5) profiles, infeasible for respondent administration due to cognitive fatigue and time constraints [25,33]. Efficient designs thus fractionally replicate the factorial, prioritizing main effects (attribute marginal utilities) while minimizing dominance, correlations, and bias [34].

3.3.1. Orthogonal Main-Effects Design Rationale

Fractional designs leverage statistical experimental principles to isolate attribute impacts, assuming additivity absent strong interactions—a parsimonious baseline validated in energy policy DCEs [35,36]. Main-effects orthogonal designs excel here: orthogonal resolution V properties decorrelate levels across attributes, yielding unbiased, efficient parameter estimates with minimal profiles [37]. This reduces respondent burden (e.g., 8–12 tasks vs. full factorial), boosts response quality, and supports welfare metrics like MWTP [15,38]. D-efficiency further optimizes: minimizing D-criterion maximizes information per observation, robust to MXL heterogeneity [39]. For RE ODA’s policy focus, main effects suffice.

3.3.2. Design and Blocking of Choice Sets

IBM SPSS Statistics 12.0 generated 16 orthogonal profiles through a main-effects fractional factorial design, ensuring level balance and zero correlations across attribute levels [40]. The resulting profiles were paired to construct eight D-efficient choice sets, each comprising three alternatives: (1) the status quo or business-as-usual (BAU) option, defined by all zero attribute levels and a zero price, and (2–3) two orthogonal policy alternatives. To reduce respondent burden and improve data reliability, the eight choice sets were further divided into two blocks of four choice sets each.

The total sample was randomly split into two equal subsamples, with each respondent assigned to one of the two blocks. Consequently, each respondent evaluated four choice sets and made four discrete choices. This blocking strategy maintained statistical efficiency while mitigating respondent fatigue, ultimately yielding 4000 individual choice observations from 1000 households. The design thus ensured robust parameter identification and precise estimation of marginal utilities, which are essential for the prioritization of RE ODA policies [41]. Figure 1 illustrates one of the choice cards presented to respondents.

To assess the quality of the experimental design, the orthogonal main-effects design was examined in terms of level balance and attribute independence. Because the choice profiles were generated using IBM SPSS Statistics 12.0, the software provided an orthogonal fractional factorial design but did not report a formal D-efficiency index or other design diagnostic statistics in the output. Nevertheless, the resulting design satisfies the fundamental requirements of an orthogonal main-effects specification, in that the attribute levels were systematically balanced across alternatives and the correlations among attributes were minimized by construction. Since the primary objective of this study was to identify main effects rather than higher-order interactions, an orthogonal main-effects design was considered appropriate and consistent with the established practice in discrete choice experiment applications. Although Bayesian efficient designs may further improve statistical efficiency when reliable prior parameter estimates are available, such priors were not sufficiently robust at the design stage of this study. Therefore, we adopted the orthogonal main-effects design as a transparent and methodologically defensible approach for survey implementation and respondent burden reduction.

3.4. Implementation of the DCE Survey

3.4.1. Sampling Strategy and Survey Procedure

To address potential heterogeneity in the target population and enhance the representativeness of the sample, a stratified random sampling approach was employed. The DCE aimed to elicit households’ MWTP for RE ODA projects. Given this objective, the target respondents were limited to household heads or their spouses, as they were presumed to be the primary decision-makers regarding household income and expenditure. The survey covered adults aged 20 to 65 years, and a total of 1000 valid respondents were systematically selected to reflect the demographic and household structure distribution across regions. To ensure accuracy and reliability of data collection, the research team commissioned a professional survey firm, and trained interviewers conducted face-to-face interviews using a structured questionnaire.

To ensure representativeness and minimize potential sampling bias, households were selected using stratified random sampling by a professional survey firm. The sampling frame was based on the Korea Ministry of Data and Statistics’ 2020 Population and Housing Census, with stratification by region, gender, age, and other key demographics to match the national population distribution. Within each stratum, households were selected through simple random sampling, ensuring that no specific socioeconomic group—such as higher income or more highly educated households—was oversampled. This multi-stage quota-controlled approach guarantees that the final sample accurately reflects the target population structure, mitigating selection bias and supporting robust population-level inference. Sample weights were constructed based on the inverse of selection probabilities to further ensure unbiased estimates.

The survey procedure began with an introduction to the background and objectives of RE ODA programs. Figure 2 was additionally presented as a visual aid to facilitate respondents’ comprehension. Each respondent was then presented with a series of choice tasks, each consisting of three alternatives: one BAU option representing the status quo (all zero levels and zero price) and two policy alternatives generated from the orthogonal fractional factorial design. Presenting only a single choice set per respondent would have been inefficient and insufficient for robust statistical inference. Therefore, multiple choice sets were assigned to each respondent while minimizing fatigue or response errors. Specifically, the eight designed choice sets were randomly divided into two blocks of four sets each. As explained above, the total sample was evenly split such that half of the respondents evaluated the first block and the other half the second block. Accordingly, each participant completed four discrete choice tasks, yielding a total of 4000 observations (1000 respondents × 4 choice sets). This balanced blocking and sampling framework achieved both statistical efficiency and respondent manageability, ensuring a high-quality dataset for subsequent econometric analysis.

3.4.2. Mitigation of Hypothetical Bias

To minimize potential hypothetical bias inherent in stated preference surveys, several ex-ante and ex-post measures were employed. Prior to the DCE, respondents were provided with a clear and neutral description of RE ODA projects, including the policy context, attribute meanings, and the financial implications of their choices. The interviewer emphasized that the scenarios presented were realistic and could influence future policy decisions, thereby encouraging respondents to consider their actual budget constraints. Additionally, a cheap talk script was read before the choice tasks, explicitly warning against overstating WTP and reminding participants to respond as if real payment were required. During data cleaning, responses showing extreme or internally inconsistent patterns—such as always choosing the highest or lowest cost options—were also examined to control for potential bias ex post. These combined procedures reduced the likelihood of strategic or hypothetical responses, improving the validity of the estimated MWTP values.

3.4.3. Control of Interviewer Bias

To address possible interviewer bias arising from differences in explanation style, tone, or perceived authority, several quality control measures were implemented. All interviewers underwent intensive training sessions that included a detailed briefing on survey objectives, neutral communication protocols, and standardized wording of scenario descriptions. Mock interviews and pilot tests were conducted to ensure procedural consistency and reduce subjectivity in field interactions. During the data collection phase, random back-checks and supervision were conducted by field coordinators to verify adherence to standardized interviewing protocols. Furthermore, all survey scripts were highly structured to minimize interviewer discretion in presenting alternatives or clarifying attributes. These procedures enhanced data reliability by ensuring that observed variations in responses reflected true preference heterogeneity rather than interviewer effects.

4. Analysis of the DCE Data

4.1. Random Utility Maximization Theory and Utility Function

The choice made by a respondent among three alternatives can be analyzed within the framework of random utility maximization theory. In each choice task, respondent $i$ is assumed to select one alternative from a finite choice set $C$, and the observed choice is interpreted as the outcome of the alternative that yields the highest utility. Formally, if respondent $i$ chooses alternative $j$ from choice set $C$, the decision can be expressed as $U_{ij}>U_{ik}$ for all $k\ne j$, where $U_{ij}$ denotes the utility that respondent $i$ derives from alternative $j$. This framework assumes that the chosen alternative provides the greatest utility among all available options.

Following McFadden’s seminal formulation [12], the utility associated with alternative $j$ for respondent $i$ can be decomposed into a deterministic component and a stochastic component:

$$U_{ij}=V_{ij}+{\epsilon }_{ij}$$

(1)

where $V_{ij}$ is the observable or systematic portion of utility that the researcher can model using observed attributes, while ${\epsilon }_{ij}$ captures unobserved influences, measurement error, and other random factors.

In discrete choice analysis, the deterministic component is commonly specified as a linear-in-parameters function of the attributes of each alternative, allowing the utility of a given option to be expressed as a weighted sum of its attribute levels. Accordingly, the systematic utility of alternative $j$ for respondent $i$ can be written as:

$$V_{ij}={\beta }_0+{\sum }_{m=1}^M{\beta }_m{x}_{ijm}$$

(2)

where $x_{ijm}$ denotes the level of attribute $m$ for alternative $j$, and ${\beta }_m$ represents the marginal utility associated with a one-unit change in that attribute.

In the present study, the utility function is specified as a function of the six attributes describing RE ODA projects, thereby capturing respondents’ preference trade-offs across policy features and the cost attribute. Under this formulation, the observed choice is driven by the relative utility contribution of each attribute bundle rather than by any single characteristic in isolation. Combining the deterministic and stochastic components yields the random utility representation used in estimation. In particular, the probability that respondent $i$ selects alternative $j$ is determined by the probability that $U_{ij}$ exceeds the utilities of all other alternatives in the choice set. Let ${\beta }_{\mathrm{price}}$ denote the estimated coefficient for the price attribute. Then, using Roy’s identity given in Varian [42], the MWTP for a one-unit improvement in attribute $m$ can be derived as the ratio of the attribute coefficient to the negative of the price coefficient, that is,

$${\mathrm{MWTP}}_m=-{\beta }_m/{\beta }_{\mathrm{price}}$$

(3)

This transformation allows the estimated preference parameters to be interpreted in monetary terms, which is particularly useful for policy prioritization in RE ODA.

4.2. Estimation of the Utility Function: The MXL Model

The multinomial logit (MNL) model is the most commonly used baseline model for estimating utility parameters in DCEs because its likelihood function has a closed-form expression and can be estimated without numerical integration. This computational simplicity makes the MNL model attractive for applied work, especially when sample sizes are moderate and the choice structure is straightforward. However, the MNL model rests on two strong assumptions that may be restrictive in practice: the independence from irrelevant alternatives property and preference homogeneity across respondents.

The independence from irrelevant alternatives assumption implies that the relative odds of choosing between two alternatives remain unchanged when another alternative is added to or removed from the choice set. Although this property facilitates tractable estimation, it is often unrealistic in real-world choice settings, where the introduction of a new option may alter substitution patterns among existing alternatives. In addition, the MNL model assumes that all respondents share the same preference structure, meaning that the utility coefficients are fixed across individuals. This assumption may be overly restrictive when respondents differ in socioeconomic background, beliefs, or policy attitudes.

To overcome these limitations, this study employs an MXL model, also referred to as a random parameters logit model. Unlike the MNL model, the MXL model does not impose the independence from irrelevant alternatives restriction and allows preference parameters to vary randomly across individuals, thereby capturing unobserved taste heterogeneity. In the MXL framework, the coefficients are treated as random variables drawn from specified distributions, and the choice probability is obtained by integrating the standard logit probability over the distribution of these random parameters. This specification provides a much richer and more flexible representation of heterogeneous preferences.

Let ${\beta }_i$ denote the vector of individual specific random coefficients. Then, the probability that respondent $i$ chooses alternative $m$ is given by the MXL integral, which averages the conditional logit probability over the density of ${\beta }_i$, as:

$$P_{im}=\int {\displaystyle \frac{exp\left(V_{im}\left(\beta \right)\right)}{{\sum }_{j=1}^{J}exp\left(V_{ij}\left(\beta \right)\right)}}f\left(\beta \right)d\beta$$

(4)

Because this probability has no closed-form solution in general, multiple integration is required, making estimation computationally demanding. For this reason, simulation-based or Bayesian estimation methods are typically used. In this study, the coefficients were estimated using the Bayesian approach proposed by Train [13], which is well suited for MXL models with repeated choices. The MXL model is highly flexible in the choice of coefficient distributions. A normal distribution is frequently assumed for many parameters, while a log-normal distribution may be used when sign restrictions are required. This is particularly relevant for the price coefficient, which should be negative to satisfy economic theory and ensure consistent welfare interpretation.

In specifying the distributions of random coefficients, we adopted a theoretically guided and empirically parsimonious approach. Following common practice in MXL estimation, we initially specified all random parameters as normally distributed to allow unrestricted preference heterogeneity and to avoid imposing a priori sign restrictions where the direction of preferences was not fully certain. When the preliminary estimates showed that some coefficients exhibited unstable signs, implausible dispersion, or weak statistical significance under the normal specification, we re-estimated those parameters using a log-normal distribution to enforce economically meaningful sign consistency. This strategy is appropriate when the underlying theory implies a monotonic preference direction, yet the normal distribution allows excessive mass on the theoretically inconsistent side of zero.

In particular, the price coefficient was retained under a normal specification because it was consistently negative and statistically significant, while selected non-price attributes were assigned log-normal distributions only when doing so improved behavioral plausibility and maintained interpretability of WTP measures. As a robustness check, we also estimated alternative benchmark models, including the MNL model and a latent class specification. Although these models did not materially alter the substantive ordering of the key attributes, the MXL model provided the best overall balance of goodness-of-fit, behavioral realism, and statistical stability; therefore, only the MXL results are reported in the main text for brevity.

As will be elaborated later, the price attribute coefficient in this study was estimated as negative even under the normal distribution assumption, thereby eliminating the need for a log-normal specification for this parameter. However, for other attributes where the normal distribution yields an anomalous sign or low statistical significance in the estimated coefficients, a log-normal distribution is employed to impose appropriate sign restrictions and enhance model fit. This specification improves the behavioral realism of the utility function and supports more credible MWTP estimation. McFadden and Train [43] show that, with appropriate specification of the random coefficients and their distributions, the MXL model can approximate a wide class of random utility models with arbitrary accuracy. This property makes the MXL model particularly suitable for stated preference analysis, where respondents’ preferences are often heterogeneous and substitution patterns may be complex. For these reasons, the MXL model provides a theoretically robust and empirically flexible foundation for estimating the utility function in this study.

RI weights reflected scaled attribute contributions to choice probability. Following Train [13], the model parameters were estimated within a Bayesian framework using Gibbs sampling with data augmentation. After discarding the initial 20,000 iterations as burn-in, 2000 posterior samples were retained for inference. This approach facilitated efficient parameter estimation and yielded reliable uncertainty measures for the MXL model. Two specifications were estimated: (1) without covariates (baseline preference structure) and (2) with covariates (preference heterogeneity by demographics). Model comparison via log likelihood isolates sociodemographic drivers of RE ODA valuation.

5. Results and Discussion

5.1. Data

The survey was administered from mid-April to mid-May 2025, a period strategically selected to capture seasonal stability in household economic perceptions and energy policy attitudes, ultimately yielding 1000 valid responses from a stratified sample of households across South Korea. Prior to commencing the DCE, trained interviewers obtained verbal informed consent from all potential participants, explicitly outlining the study’s purpose, procedures, voluntary nature, and data confidentiality measures; only those providing affirmative consent proceeded, ensuring ethical compliance and respondent autonomy. This rigorous consent protocol, consistent with institutional review board standards and best practices in stated preference research, not only mitigated potential coercion but also fostered trust, thereby enhancing response validity, engagement, and overall data quality.

Field interviewers reported that respondents encountered no significant difficulties in understanding the choice tasks or attribute definitions, as evidenced by minimal requests for clarification and high completion rates. The professional survey firm implemented rigorous post-collection validation procedures, including random call-back verifications and consistency checks on responses. Observations failing these quality controls—such as incomplete tasks, inconsistent patterns, or suspected fatigue—were discarded, and additional interviews were conducted to achieve the target sample size of 1000. This multi-stage cleaning process ensured a high-quality dataset suitable for econometric estimation.

Table 4. Description of the variables.

Variables	Definitions	Mean	Standard Deviation
Age	Age of the respondent	47.66	9.91
Life satisfaction	Overall evaluation of one’s life rated on a 1–9 scale	5.84	1.30
Education	Education level of the respondent in years (6 = elementary school graduate; 9 = middle school graduate; 12 = high school graduate; 14 = junior college graduate; 16 = bachelor’s degree; 18 = graduate degree or higher)	14.23	2.17
Income	Monthly income of the respondent’s household (unit: million Korean won = USD 719 at the time of the survey)	4.88	2.39
Priority	Perceived priority of official development assistance objectives (0 = humanitarian; 1 = national interest)	0.60	0.49

presents descriptive statistics for the key respondent characteristics and covariates used in the analysis. The sample is representative of working-age adults (mean age: 47.66 years, standard deviation (SD): 9.91), with an average education level of 14.23 years (SD: 2.17) and monthly household income of KRW 4.88 million (SD: 2.39). Life satisfaction scores averaged 5.84 on a 1–9 scale (SD: 1.30), and 60% of respondents prioritized national interest objectives over humanitarian ones in ODA allocation (SD: 0.49). These summary measures confirm the sample’s demographic diversity and provide a foundation for heterogeneity analysis in the MXL models.

5.2. Results

5.2.1. Baseline Model Without Covariates

The baseline MXL model estimated from the DCE data provides robust evidence of South Korean households’ preferences for RE ODA attributes. The results are given in

Table 5. Estimation results of the baseline mixed logit model.

Variables a	Assumed Distributions	Means of the Coefficient Estimate	t-Values	p-Values	95% Confidence Intervals c
ASC b	Normal	0.8324	3.16	0.0016	[0.3160, 1.3488]
Electrification	Log-normal	0.2174	12.89	0.0000	[0.1843, 0.2505]
Reduction	Normal	0.2121	10.65	0.0000	[0.1731, 0.2511]
Expansion	Normal	0.3036	4.70	0.0000	[0.1770, 0.4302]
Training	Normal	0.0794	6.61	0.0000	[0.0559, 0.1029]
Reputation	Log-normal	0.1572	8.13	0.0000	[0.1193, 0.1951]
Price	Normal	−0.2610	−9.64	0.0000	[−0.3141, −0.2079]
Simulated log likelihood	−3285.03
Pseudo-R2	0.2229
Akaike information criterion	6584.06
Bayesian information criterion	6595.27

Notes: ^a attributes and variables are defined in Table 3 and Table 4, respectively. ^b ASC refers to alternative specific constant. ^c The 95% confidence intervals were derived using Monte Carlo simulation, a kind of parametric bootstrap technique.

. This parsimonious specification captures mean attribute utilities while accommodating unobserved preference heterogeneity through random parameters drawn from flexible distributions (normal for most attributes, log-normal where needed to enforce sign consistency). All attribute mean coefficients are positive and statistically significant at the 1% level, indicating broad public support for RE ODA improvements across dimensions. Specifically, firm expansion exhibits the highest mean utility (β = 0.304, t-value = 4.70), reflecting strong valuation of domestic economic spillovers from overseas market entry by South Korean enterprises. GHG emission reductions (β = 0.212, t-value = 10.65) and electrification (β = 0.217, t-value = 12.89) follow closely, underscoring environmental and humanitarian priorities, while expert training (β = 0.079, t-value = 6.61) and reputation enhancement (β = 0.157, t-value = 8.13) contribute positively but with comparatively modest magnitudes.

The price coefficient is negative and highly significant (β = −0.261, t-value = −9.64), satisfying theoretical expectations of downward-sloping demand and enabling welfare analysis via Roy’s identity. Critically, standard deviations of all random coefficients are statistically significant (p-value < 0.01), rejecting preference homogeneity and confirming substantial taste variation across the sample—consistent with the MXL framework’s ability to approximate any random utility model [43]. The model’s simulated log likelihood of −3285.03 reflects strong fit to the 4000 choice observations (pseudo-R² = 0.2229; Akaike information criterion = 6584.06; Bayesian information criterion = 6595.27). Furthermore, Bayesian estimation via Gibbs sampling (20,000 burn-in iterations, 2000 retained draws) ensured precise posterior inference.

5.2.2. Extended Model with Covariates

To probe sociodemographic drivers of heterogeneity, the baseline is augmented with interactions between the alternative specific constant (ASC, capturing status quo preference) and key covariates: age, education (years), household income (KRW millions/month), life satisfaction (1−9 scale), and national priority orientation (binary: 1 if prioritizing donor interests over humanitarian goals). The results are given in

Table 6. Estimation results of the mixed logit model with covariates.

Variables a	Assumed Distributions	Means of the Coefficient Estimate	t-Values	p-Values	95% Confidence Intervals c
ASC b	Normal	1.8557	8.40	0.0000	[1.4226, 2.2888]
Electrification	Log-normal	0.2292	6.63	0.0000	[0.2969, 0.1615]
Reduction	Normal	0.6209	7.14	0.0000	[0.4505, 0.7913]
Expansion	Normal	0.9013	4.20	0.0000	[0.4810, 1.3216]
Training	Normal	0.2260	6.29	0.0000	[0.1556, 0.2964]
Reputation	Log-normal	0.5551	4.70	0.0000	[0.7866, 0.3236]
Price	Normal	−0.2956	−6.96	0.0000	[−0.3789, −0.2123]
ASC × Age	Normal	−0.0964	−0.13	0.8970	[−1.6070, 1.4142]
ASC × Life	Normal	−1.8453	−2.56	0.0106	[−3.2594, −0.4312]
ASC × Education	Normal	−6.7461	−3.01	0.0026	[−11.1374, −2.3548]
ASC × Income	Normal	−0.1262	−2.66	0.0078	[−0.2192, −0.0332]
ASC × Priority	Normal	3.4875	5.25	0.0000	[2.1848, 4.7902]
Simulated log likelihood	−3528.21
Pseudo-R2	0.1654
Akaike information criterion	7080.42
Bayesian information criterion	7099.64

Notes: ^a attributes and variables are defined in Table 3 and Table 4, respectively. ^b ASC refers to alternative specific constant. ^c The 95% confidence intervals were derived using Monte Carlo simulation, a kind of parametric bootstrap technique.

. Mean attribute coefficients generally increase in magnitude (e.g., firm expansion β = 0.901, t-value = 4.20; GHG reduction β = 0.621, t-value = 7.14), while the price parameter remains negative (β = −0.296, t-value = −6.96); random coefficient variances stay significant (p-value < 0.01).

Notable interactions reveal segmentation patterns: higher life satisfaction (ASC × Life: β = −1.845, t-value = −2.56, p-value < 0.05), education (β = −6.746, t-value = −3.01, p-value < 0.01), and income (β = −0.126, t-value = −2.66, p-value < 0.01) significantly reduce BAU utility, implying that more satisfied, educated, or affluent respondents exhibit stronger aversion to the zero-aid status quo. Conversely, national priority orientation markedly boosts non-BAU utility (ASC × Priority: β = 3.488, t-value = 5.25, p-value < 0.01), suggesting geopolitically minded individuals favor RE ODA packages emphasizing donor returns. Age shows no significant moderation (p-value = 0.90), indicating stable generational preferences.

Despite these insights, the covariate-augmented model yields a poorer simulated log likelihood (−3528.21), alongside pseudo-R² of 0.1654, Akaike information criterion of 7080, and Bayesian information criterion of 7100. Furthermore, a likelihood ratio test (χ² = 486.36, p-value < 0.01) favors the baseline specification due to insufficient explanatory gains relative to added complexity. This parsimony underscores the baseline’s sufficiency for representing core preferences, with covariates illuminating auxiliary heterogeneity for policy targeting.

5.3. Discussion of the Results

This section discusses the empirical findings from the MXL estimation in a structured manner. It first examines the MWTP estimates for individual RE ODA attributes (Section 5.3.1), followed by an analysis of RI weights that quantify each attribute’s contribution to choice variance (Section 5.3.2). Subsequent subsections evaluate total WTP for hypothetical policy scenarios constructed from attribute bundles (Section 5.3.3) and derive targeted policy implications for South Korea’s green ODA portfolio design and public communication strategies (Section 5.3.4).

5.3.1. Estimation of MWTP

MWTP values, derived via Roy’s identity from the baseline MXL model, translate attribute utilities into policy-actionable monetary metrics, as detailed explicitly in

Table 7. Marginal willingness to pay (MWTP) estimates derived from the model excluding covariates.

Attributes a	MWTP Estimates per Household per Year	t-Values	95% Confidence Intervals b
Electrification (unit: 1000 persons)	KRW 833 (USD 0.60)	9.12	KRW 657−1016
Reduction (unit: 1000 tCO2e)	KRW 813 (USD 0.58)	8.39	KRW 624−1008
Expansion	KRW 1163 (USD 0.84)	5.19	KRW 742−1603
Training (unit: 10 persons)	KRW 3042 (USD 2.19)	5.59	KRW 1994−4086
Reputation	KRW 602 (USD 0.43)	5.86	KRW 399−804

Notes: ^a They are defined in Table 3. ^b The 95% confidence intervals were derived using Monte Carlo simulation, a kind of parametric bootstrap technique. USD 1.0 = KRW 1390 at the time of the survey.

. Posterior means from 2000 Gibbs samples (95% confidence intervals (CIs) via Monte Carlo bootstrap) ensure robust inference under Bayesian estimation. Table 7 reveals that expert training commands the highest MWTP at KRW 3042 (USD 2.19) per 10 experts trained (t-value = 5.59; 95% CIs: KRW 1994–4086), reflecting strong valuation of donor human capital returns. Firm expansion ranks second at KRW 1163 (USD 0.84) per firm (t-value = 5.19; CIs: 742–1603), underscoring economic spillover prioritization. GHG reductions (KRW 813/USD 0.58 per 1000 tCO₂e, t-value = 8.39; CIs: 624–1008) and electrification (KRW 833/USD 0.60 per 1000 people, t-value = 9.12; CIs: 657–1016) exhibit statistically equivalent support. Reputation enhancement yields the lowest MWTP at KRW 602 (USD 0.43) (t-value = 5.86; CIs: 399–804), indicating modest soft power valuation.

Key insights from Table 7 are as follows: First, scale normalization (per 1000 people/tCO₂e, per firm, per 10 experts) enables meaningful cross-attribute comparisons for RE ODA portfolio optimization. Second, all MWTPs remain positive and statistically significant (p-value < 0.01), rejecting the null hypothesis of zero public valuation across attributes. Third, tight CIs reflect high estimation precision from 4000 choice occasions (1000 respondents × 4 tasks), validating the orthogonal design’s efficiency. Fourth, USD equivalents (1 USD = KRW 1390) contextualize the estimated annual household WTP (KRW 602–3042) in relation to South Korea’s expanding energy sector ODA, which reached USD 232.20 million in 2024 and reflects its growing commitment to RE and green ODA initiatives. Table 7 thus equips donor agencies with concrete welfare economic metrics for prioritizing high-impact, climate-focused RE ODA projects that align with revealed taxpayer preferences, bridging micro-level valuation with macro-budgetary decisions.

5.3.2. Estimation of RI

RI weights, from the baseline MXL model, provide a scale-free measure of each attribute’s contribution to overall choice variance, complementing MWTP estimates for policy prioritization. These weights normalize utility coefficients against their sum (excluding price), ensuring values sum to 100% while revealing the hierarchy of public preferences independent of monetary units.

Table 8. Derived relative importance of attributes.

Attributes a	Relative Importance (%)	t-Values	95% Confidence Intervals (%) b
Electrification	4.31	13.96	3.72−4.91
Reduction	42.06	16.15	36.83−47.19
Expansion	20.07	6.44	13.98−26.11
Training	31.49	8.13	23.97−39.16
Reputation	2.08	5.84	1.38−2.76

Notes: ^a They are defined in Table 3. ^b The 95% confidence intervals were derived using Monte Carlo simulation, a kind of parametric bootstrap technique.

demonstrates that GHG emission reductions exert the strongest influence at 42.1% RI, confirming climate mitigation as South Korean taxpayers’ foremost priority for RE ODA funding. This dominance aligns with global decarbonization imperatives and South Korea’s green ODA ambitions. Expert training follows closely at 31.5% RI, reflecting robust valuation of donor human capital returns, while firm expansion captures 20.1% RI, underscoring economic spillover prioritization.

Electrification accounts for a modest 4.3% RI, suggesting comparatively lower emphasis on humanitarian access relative to climate and strategic priorities. Reputation ranks lowest at 2.1% RI, aligning with its relatively limited MWTP. Notably, the climate–human capital nexus—GHG reductions plus expert training—collectively accounts for 73.6% of total choice variance, suggesting optimal RE ODA portfolios should emphasize these high-utility domains. Similarly, the top three attributes (GHG reductions, expert training, and firm expansion) account for 93.6% of RI, indicating that RE ODA priorities are largely shaped by a combination of global problem-solving (climate mitigation) and national interest considerations (human capital development and firm expansion). This provides clear guidance for reallocating resources away from lower priority reputation-building initiatives. All standard deviations exceed 0.09 (p-value < 0.01), confirming substantial preference heterogeneity across the sample, with younger or nationally oriented respondents likely amplifying economic attributes’ influence. Therefore, Table 8 presents a utility-based preference structure and clearly demonstrates that taxpayer support for Korea’s expansion of green ODA is primarily driven by strategic objectives that simultaneously reflect national interests and global problem-solving—such as climate change mitigation and the overseas expansion of domestic firms—rather than by humanitarian goals such as expanding access to electricity.

It should be noted that MWTP and RI capture distinct but complementary dimensions of respondents’ preferences. MWTP is a monetary measure derived from the ratio of each attribute coefficient to the negative price coefficient and therefore reflects the amount respondents are willing to pay for a marginal improvement in a specific attribute. By contrast, RI is a normalized measure that indicates the contribution of each attribute to overall choice behavior within the estimated utility framework. As a result, an attribute may exhibit the highest MWTP because respondents attach a relatively strong monetary value to it, while another attribute may display the largest relative importance because it contributes more broadly to explaining variation in choice probabilities across the full attribute set. In this study, expert training yields the highest MWTP, whereas GHG reduction has the largest relative importance, suggesting that respondents place a particularly high monetary value on expert capacity building but consider emissions reduction the most influential attribute in the overall choice process. Accordingly, MWTP and RI should be interpreted as complementary indicators rather than directly comparable ranking measures.

5.3.3. WTP Analysis for Hypothetical Scenarios

Table 9. Hypothetical renewable energy official development assistance scenarios and associated willingness to pay (WTP).

Attributes a	Climate Leader Scenario	Economic Expansion Scenario	Balanced Development Scenario
Electrification	0	0	1500
Reduction	15,000	5000	5000
Expansion	0	5	0
Training	30	0	0
Reputation	No	No	Yes
Household WTP per year b	KRW 21,321 (USD 15.34)	KRW 9880 (USD 7.11)	KRW 5917 (USD 4.26)

Notes: ^a They are defined in Table 3. ^b The exchange rate at the time of the survey was USD 1.0 to KRW 1390.

presents total WTP estimates for three realistic RE ODA project scenarios constructed by bundling attribute levels from Table 3, aggregating MWTP values to simulate policy-relevant welfare gains for South Korean households. These scenarios operationalize common RE ODA archetypes—Climate Leader (high GHG reductions + expert training), Economic Expansion (firm expansion + moderate GHG), and Balanced Development (electrification + GHG + reputation)—enabling direct comparison of public support across strategic priorities.

The Climate Leader scenario (15,000 tCO₂e reduced + 30 experts trained) generates the highest total WTP at KRW 61,491 per household annually (USD 15.34), reflecting synergistic valuation of planetary decarbonization and donor capacity building—73.6% RI alignment from Table 8. This premium underscores taxpayer readiness to fund ambitious, high-impact green projects despite fiscal constraints. Economic Expansion (5 firms entering + 5000 tCO₂e reduced) yields KRW 9880 (USD 7.11), driven by firm expansion’s high per-unit MWTP despite moderate GHG levels. This scenario reveals strong support for commercial spillovers, appealing to national interest-oriented respondents identified in the covariate analysis.

The Balanced Development scenario (1500 people electrified + 5000 tCO₂e + reputation enhancement) attracts KRW 5917 (USD 4.26), prioritizing humanitarian access alongside climate benefits. Its comparatively lower WTP reflects electrification’s tempered RI (4.3%) relative to strategic attributes. Four comparative insights emerge from Table 9:

• Clear hierarchy emerges: Climate Leader (KRW 21,321) > Economic Expansion (KRW 9880) > Balanced Development (KRW 5917), indicating substantially stronger public preferences for climate-focused RE ODA portfolios than for Balanced Development-oriented alternatives.
• Climate synergies: The Climate Leader scenario demonstrates that public support is the strongest when climate mitigation and expert capacity building are jointly emphasized, reinforcing the dominant relative importance of GHG reductions and expert training identified in Table 8.
• Aggregate implications: Scaling the estimated WTP to approximately 18 million South Korean households yields nearly KRW 384 billion (USD 276.10 million) in annual support for the Climate Leader scenario. This amount represents a meaningful level of public support relative to the current scale of South Korea’s energy sector ODA (USD 232.20 million in 2024), suggesting substantial taxpayer willingness to support climate-oriented RE ODA expansion under appropriate policy framing.
• Heterogeneity considerations: High-income/educated segments (negative ASC interactions) likely amplify WTP across scenarios, suggesting targeted communication strategies.

Table 9 thus transforms micro-level attribute preferences into macro-budgetary guidance, demonstrating that South Korean taxpayers most strongly endorse climate-centric RE ODA portfolios that simultaneously build global expertise networks—a finding with immediate implications for South Korea’s green aid strategy amid rising ODA scrutiny.

5.3.4. Policy Priorities

This study’s estimates of MWTP and RI by attribute (Table 7 and Table 8), together with scenario-based WTP analysis (Table 9), provide concrete and empirical guidance for policy direction and portfolio optimization of Korea’s RE ODA under increasing taxpayer scrutiny and tightening fiscal constraints.

First, unlike prior studies that primarily focused on development impacts in recipient countries, this study explicitly quantifies donor country benefits. In particular, firm expansion (20.1% RI; KRW 1163 per firm) and expert training (approximately 31.5% RI; KRW 3042 per 10 experts) are shown to generate substantial welfare value alongside traditional humanitarian objectives. Survey evidence indicates that 60% of respondents prioritize national interests over pure altruism, thereby supporting a strategic ODA approach that simultaneously enhances economic competitiveness and human capital networks. This donor-centered valuation framework strengthens the legitimacy of ODA and aligns with emerging trends among some donor countries, where ODA is increasingly justified in terms of national interest rather than purely humanitarian considerations.

Second, the results identify GHG reduction as the most influential attribute (42.1% RI; KRW 813 per 1000 tCO₂e), indicating that it should be treated as the top priority in RE ODA policy design. Climate mitigation is a global public good requiring joint action by both developed and developing countries. Since the Paris Agreement, countries have established their Nationally Determined Contributions, while Article 6.2 introduces cooperative mechanisms through Internationally Transferred Mitigation Outcomes. Given Korea’s overseas mitigation target of approximately 37.5 million tons, the strong public preference for GHG reduction implies that RE ODA can simultaneously contribute to global climate goals and support the achievement of national Nationally Determined Contributions targets when effectively linked with international mitigation projects. Scenario analysis further reinforces this priority. The Climate Leader scenario records the highest household WTP at KRW 21,321, exceeding the Economic Expansion scenario (KRW 9880) by more than twofold and indicating particularly strong public support for climate-focused RE ODA portfolios emphasizing both climate mitigation and expert capacity building.

Third, while this study focuses on the RE sector, several attributes—particularly firm expansion and reputation—demonstrate cross-sectoral applicability. Reputation ranks the lowest (KRW 602 per household or KRW 114.6 billion nationally), yet provides a useful quantitative benchmark for soft power valuation across sectors such as health, education, and agriculture ODA. From a portfolio perspective, the concentration of RI in key attributes (GHG reduction, expert training, and firm expansion—93.6% combined) indicates that RE ODA priorities are largely shaped by a combination of global problem-solving (climate mitigation) and national interest considerations (human capital development and firm expansion). This provides clear guidance for reallocating resources toward high-impact domains while relatively de-emphasizing lower priority reputation-building initiatives.

Policy implications follow directly. Priority should be given to integrated projects that simultaneously advance climate mitigation and national interests, particularly those linking GHG reductions and firm expansion, while reputation-focused initiatives should be treated as complementary. In addition, targeted communication strategies should be developed for high-WTP segments (e.g., highly educated and high-income groups identified through ASC interactions), emphasizing the economic value of overseas market entry. MWTP estimates should be institutionalized as benchmarks for ex-post evaluation, enabling evidence-based reallocation of underperforming projects. Transparent disclosure of RI and MWTP can further strengthen social consensus by demonstrating that policy priorities reflect revealed public preferences rather than administrative discretion. Progressive financing mechanisms, leveraging stronger anti-status quo preferences among higher income groups, can mitigate regressivity while ensuring stable funding for high-impact RE ODA portfolios supported by annual household WTP ranging from KRW 5917 to 21,321.

In sum, by bridging micro-level preference estimation with macro-level fiscal and strategic policy design, this study provides a taxpayer-validated, evidence-based framework for prioritizing climate- and national interest-oriented RE ODA, supporting its sustainable expansion in an era of intensifying domestic fiscal pressures and geopolitical competition.

6. Conclusions

6.1. Research Overview and Key Findings

This study employed a DCE with 1000 nationally representative South Korean households and a MXL model to estimate MWTP for key attributes of RE ODA. The attributes included electrification (0–1500 people), GHG emission reductions (0–15,000 tCO₂e), firm expansion (0–5 firms), expert training (0–30 experts), and reputation enhancement (no/yes). The results reveal that GHG reductions (KRW 813 per 1000 tCO₂e, USD 0.58) and expert training (KRW 3042 per 10 experts, USD 2.19) command the highest MWTP, jointly explaining 73.6% of choice variance, followed by firm expansion (KRW 1163 per firm), while electrification and reputation show comparatively lower support.

These preferences reflect South Korean taxpayers’ recognition of RE ODA as a multifaceted policy instrument that addresses global climate imperatives while delivering donor-side returns. The strong valuation of GHG reductions aligns with empirical evidence that energy aid curtails recipient emissions through structural shifts [6], while expert training captures human capital spillovers that enhance donor geopolitical leverage [11]. This dual emphasis distinguishes RE ODA from traditional humanitarian aid, positioning it as a strategic vehicle for both SDG 13 (climate action) and national interests.

6.2. RI and Scenario Analysis

RI weights further underscore the dominance of climate and capacity-building attributes: GHG reductions (42.1%) and expert training (31.5%) together account for nearly 73.6% of choice determination, with firm expansion (20.1%) ranking third. Electrification (4.3%) and reputation (2.1%) lag, suggesting that direct access benefits are secondary to planetary and strategic outcomes in public valuation.

Scenario simulations reinforce these priorities. A climate-focused package (high GHG reductions + expert training) elicited the highest household WTP (KRW 21,321 annually), surpassing economic-oriented (KRW 9880; firm expansion emphasis) and balanced scenarios (KRW 5917; moderate levels across attributes). This hierarchy implies that RE ODA portfolios maximizing emissions impacts and knowledge transfer generate the greatest public welfare, providing empirical justification for South Korea’s green ODA expansion amid its modest ODA/gross national income ratio (0.21% in 2024 vs. DAC average 0.36% and United Nations target 0.7%).

Such patterns resonate with South Korea’s evolving aid landscape, where energy sector ODA reached USD 232.20 million in 2024. By quantifying attribute trade-offs, this study bridges the gap between donor country preferences and project design, enabling welfare-maximizing allocations under fiscal constraints.

6.3. Preference Heterogeneity Insights

The MXL model’s random parameters reveal systematic preference heterogeneity, offering nuanced policy guidance. Younger respondents (20–39 years) and those with nationalistic orientations exhibited stronger valuations for firm expansion, prioritizing economic returns amid domestic uncertainties. Conversely, higher income, more educated, and energy satisfaction groups displayed greater status quo aversion and affinity for climate attributes, reflecting cosmopolitan climate leadership preferences.

Socio-spatial variations further enrich the analysis: metropolitan respondents undervalued electrification, likely viewing it as less salient given South Korea’s near-universal access (99.7%), while rural counterparts showed mild support. Gender differences were minimal, but policy satisfaction positively correlated with overall RE ODA support. These segments suggest tailored communication: economic narratives for nationalists (“market expansion via ODA”), climate appeals for progressives (“global decarbonization leadership”), and hybrid framing for broader consensus.

6.4. Policy Implications

The findings yield actionable recommendations for RE ODA design, implementation, and communication.

• Portfolio Prioritization: Allocating resources to projects emphasizing GHG reductions and expert training are core key performance indicators. For instance, solar initiatives in Africa/Southeast Asia should contractually specify tCO₂e savings and trainee numbers, weighting these 60%+ in evaluations. This aligns with evidenced impacts [5] and maximizes taxpayer welfare.
• Strategic Communication: Sustain support by quantifying co-benefits: “1 MW solar = 1500 tCO₂e saved = emissions of 3000 Seoul households annually” alongside “five-firm expansion = USD 10 million export gains.” During budget deliberations, cite MWTP ratios to legitimize green ODA scaling toward the United Nations targets.
• Targeting and Inclusivity: Leverage heterogeneity via differentiated messaging—enterprise opportunities for conservatives and global responsibility for climate-concerned groups. Low electrification valuation implies prioritizing high-emission recipients over pure access cases, optimizing SDG 13 over SDG 7.
• Institutional Mechanisms: Integrate MWTP into Korea International Cooperation and Economic Development Cooperation Fund appraisal frameworks, using RI weights for multi-criteria decision analysis. Pilot “preference-aligned” tenders could test real-world uptake, enhancing political sustainability. The two respectively manage South Korea’s grant-based and concessional loan ODA. The former operates under the Korea Ministry of Foreign Affairs, while the latter falls under the oversight of the Korea Ministry of Finance and Economy, with operational management entrusted to the Export-Import Bank of Korea.

6.5. Study Limitations and Directions for Future Research

Despite rigorous design, limitations persist. First, stated preference methods risk hypothetical bias, mitigated here via cheap talk, pre-tests, and inconsistency filters, but not eliminated. Real tax adjustment experiments could validate external validity. Second, cross-sectional data omits dynamics: post-2025 interest hikes, elections, or Trump administration (inaugurated January 2025) aid retrenchment may shift preferences. Longitudinal tracking is needed. Third, South Korea-specific results limit generalizability. Comparative DCEs with Japan (soft power focus; Yoo [10]) or EU donors could test universality vs. context-dependence.

Extensions include: (1) hybrid stated–revealed preference models for bias reduction and panel designs for preference stability; (2) cross-national DCEs theorizing green aid political economy across emerging (South Korea, China) vs. traditional donors; (3) AI-dynamic surveys or social media sentiment for real-time tracking; (4) recipient-side DCEs matching donor–recipient preferences. Micro–macro linkages warrant exploration through simulations of budget reallocation using MWTP (e.g., +10% green ODA welfare gains), while behavioral extensions such as nudge effects and framing experiments could refine communication efficacy.

6.6. Concluding Contributions

This study pioneers donor country MWTP for RE ODA attributes in an emerging donor context, extending stated preference economics [13] to aid political economy. By reframing ODA as multi-dimensional utility generation, the analysis provides an ex-ante framework for RE ODA portfolio optimization and supports the sustainable expansion of South Korea’s green and energy sector ODA. Ultimately, climate-effective, strategically credible RE ODA promises domestic political viability and global leadership, aligning taxpayer values with energy transition imperatives.