Back to All Research
Reported study summary · DCE + mixed logit + latent class

Consumer Preferences for Electric Vehicles Aesthetic Design and Technology: A Discrete Choice Experiment in Vietnam

This page summarizes a discrete choice experiment on young Vietnamese consumers, focusing on how they trade off EV functional performance and aesthetic design.

Authors: Hy Chung Dai, Anh Minh Hoang Department of Business, Swinburne Vietnam - FPT University Hanoi Corresponding email: hydcsws00482@fpt.edu.vn Study setting: Hanoi and Ho Chi Minh City, Vietnam Final analytic sample: n = 212 Segment shares: 65% pragmatic, 35% aesthetic
Background DCE Design Results MWTP Simulation
Section 1

Background and Research Gap

Vietnam's transport transition occurs in a motorcycle-dominant system where private mobility remains fossil-intensive and car ownership is still comparatively low relative to two-wheelers (Huu and Ngoc, 2021; Khuat, 2015; VAMM, 2025). Prior EV studies in Vietnam focus mainly on intention drivers, cost burden, and infrastructure readiness.

The gap addressed here is comparative valuation: how aesthetic cues perform relative to functional levers in explicit product trade-offs. The contribution is incremental but important for emerging markets, integrating aesthetic attributes into a standard DCE framework rather than treating EV adoption as purely utilitarian (Jones et al., 2013; Pham et al., 2022).

Section 2

Research Questions

RQ1

How do aesthetic design attributes compare with functional technological attributes in shaping EV preferences among young Vietnamese consumers?

RQ2

Does this cohort contain distinct preference segments with different valuations of design versus functional features?

Section 3

Literature Synthesis

Utility and decision theory

Interpretation: EV buyers do not optimize one feature; they trade across competing bundles of price, performance, and style.

Theory: Random Utility Theory models choice as systematic utility plus random error, so coefficients indicate how strongly each attribute shifts selection probability (Manski, 1977; McFadden and Train, 2000).

DCE implication: The experiment must force explicit bundle choices, not isolated attribute ratings.

Aesthetic processing in product evaluation

Interpretation: Visual form influences first-pass judgment before detailed technical comparison.

Theory: Product appearance triggers fast affective cues that can raise or lower perceived quality and desirability (Bloch, 1995; Norman, 2004; Creusen and Schoormans, 2005).

DCE implication: Design attributes cannot be treated as decorative controls; they must be directly parameterized.

Functional constraints in EV adoption

Interpretation: Cost, range, and charging reliability remain structural barriers even for younger and innovation-oriented cohorts.

Theory: Strong functional performance lowers perceived risk and increases practical utility in adoption decisions (Hackbarth and Madlener, 2013; Hoang et al., 2022; Chitturi et al., 2008).

DCE implication: Functional levels must be realistic enough to capture plausible sacrifice/benefit trade-offs.

Symbolic and emotional meaning

Interpretation: EVs can signal identity, aspiration, and status beyond transport utility.

Theory: Symbolic consumption research shows status and self-image can shift how buyers value design and brand cues (Levy, 1959; Rezvani et al., 2015; Xia et al., 2022).

DCE implication: Aesthetic levels should be estimated jointly with functional levels to avoid omitted-value bias in design effects.

Moderators relevant to Vietnam

Interpretation: Young urban Vietnamese consumers are likely to combine technology openness with image sensitivity in a motorcycle-dominant transition context.

Theory: Prior ownership, status orientation, and environmental concern can change both attribute weights and segment membership (Buhmann and Criado, 2023; Gauer et al., 2025; Ha et al., 2023).

DCE implication: Segmentation models are necessary; average coefficients alone are insufficient for strategy design.

Section 4

DCE Design

How to read this section: each design choice is linked to an inferential trade-off between realism, identification quality, and respondent burden.

The experiment uses a Taguchi L27 orthogonal array to identify main effects for eight three-level attributes while keeping task load manageable for field respondents. This is a pragmatic design decision: it sacrifices some efficiency relative to D-efficient optimization but preserves orthogonality and operational simplicity in a long instrument (Hensher et al., 2015).

Profiles were grouped into 9 tasks with 3 alternatives each because triadic sets typically balance realism and cognitive tractability better than larger boards in stated-choice contexts (Hess et al., 2012). Attribute levels were selected to represent credible Vietnamese EV market ranges while preserving variance large enough for utility separation.

Effects coding was used for categorical aesthetics so every level receives interpretable coefficients and monetary conversion remains coherent in MWTP space (Bech and Gyrd-Hansen, 2005). Because DCE responses are hypothetical, quality controls and bias caveats are central to interpretation: minimum completion-time filtering, dominance checks, and explicit realism framing reduce but do not eliminate hypothetical bias (Hensher, 2010; Lusk and Schroeder, 2004; Xhakollari et al., 2025).

Table 1. DCE attributes and levels
Category Attribute Levels
Functional Purchase price 300m, 650m, 900m VND
Functional Driving range 200, 350, 500 km
Functional Fast-charge time (to 70%) 60, 30, 15 minutes
Functional Running cost (per 100 km) 50k, 100k, 150k VND
Functional Charging station availability 5,000; 50,000; 150,000 stations
Aesthetic Body shape Streamlined, balanced, angular/muscular
Aesthetic Front fascia / lighting Closed signature, semi-closed, conventional grille
Aesthetic Color Neutral, metallic, vivid
Example EV choice task layout used in the study survey
Example survey screen presenting three EV alternatives per choice task; shown to document instrument structure.
Contextual EV charging environment image
Illustrative charging environment used to contextualize range and charging-availability attributes.
Section 5

Survey and Sample

The target population was private-vehicle users in Hanoi and Ho Chi Minh City. Recruitment combined online panel and street intercept modes with soft quotas on gender and broad age bands. Quality control included minimum completion time checks and an attention/dominance task.

From 620 invitations, 240 completions were obtained. Exclusion rules removed 28 cases (18 below time threshold, 7 failed dominance trap, 3 partials), yielding the final analytic sample of n = 212.

Invitations

620

Completions

240

Excluded

28

Final sample

212

Mean age

22.9 ± 4.3

Table 2. Descriptive sample profile
Variable Key values
Gender 63.2% male, 36.8% female
EV experience 70.8% none, 18.9% test-driven, 10.4% owner
Income band 34.9% low, 44.8% middle, 20.3% high
Car ownership 60.4% no car, 37.3% one car, 2.4% two or more
Psychographic indicators Status seeking mean 0.29; environmental concern mean 0.39 (binary above-median coding)
Section 6

Econometric Model

How to read this section: the model stack moves from average behavior to heterogeneity, then to actionable segmentation.

The model stack follows Random Utility Theory and uses staged specifications so each model answers a distinct question instead of overloading one specification (Manski, 1977; McFadden and Train, 2000).

Utility decomposition

Uni = Vni + εni

Systematic utility

Vni = β′ Xni

MWTP conversion

MWTPk = - βk / βprice

Stage 1: MNL

Used as the baseline because it provides transparent directional coefficients and a reference likelihood before heterogeneity terms are introduced.

Stage 2: MXL

Introduced to capture continuous heterogeneity in valuation. Price is log-normal (sign-constrained) for theoretical consistency, while range and aesthetics are random-normal to allow bidirectional dispersion (Revelt and Train, 1998; McFadden and Train, 2000).

Stage 3: LCM

Complements MXL by yielding discrete managerial segments. Class count is selected using fit criteria and interpretability, enabling segment-specific pricing and communication translation (Greene and Hensher, 2003).

Categorical aesthetics are effects-coded for level interpretation. MWTP uncertainty is reported with Krinsky-Robb simulation because ratio-based money metrics inherit coefficient uncertainty non-linearly and need simulation-based intervals (Krinsky and Robb, 1986).

Section 7

Results

How to read this section: separate statistically robust signals from directional but imprecise effects before drawing managerial conclusions.

Model fit improves from MNL to MXL to the two-class LCM, indicating non-trivial heterogeneity beyond average utility. Across models, price and charging/range attributes remain the strongest and most stable determinants. Aesthetic terms are positive in several specifications but not uniformly precise across classes.

MNL log-likelihood

-1529.96

MXL log-likelihood

-1528.23

LCM (2-class) log-likelihood

-1523.26

Class 1 share

65% pragmatic adopters

Class 2 share

35% aesthetic enthusiasts

Table 3. Selected latent class coefficients (C = 2)
Attribute Pragmatic adopters (65%) Aesthetic enthusiasts (35%)
Price (100m VND) -0.436*** -0.372***
Driving range (100 km) 0.584*** 0.670***
Fast-charge time (15 min) -0.419** -0.392***
Fast charging stations (50,000) 0.191 (ns) 0.225***
Running cost (50k VND) -0.120 (ns) -0.224**
Body: streamlined 0.189 (ns) 0.253 (ns)

Statistically strong signals

  • Higher purchase price consistently reduces choice probability.
  • Longer range and faster charging materially increase utility.
  • Class heterogeneity is real, with distinct cost-sensitivity profiles.

Directional but imprecise signals

  • Streamlined body design is valued positively in both classes but not precisely estimated in the LCM.
  • Some fascia/color effects vary across classes and should be interpreted as exploratory.
  • Design relevance is meaningful but not empirically dominant over core functional constraints in this dataset.

Interpretation discipline: class contrast is strongest on functional sensitivity (price, range, charging). Design channels appear economically relevant for positioning, but coefficient precision does not support universal "design-first" claims.

Section 8

MWTP and Scenario Simulation

Calculator output is coefficient-linked to reported segment MWTP values (million VND). The interval display is an additive approximation from attribute-level confidence bounds and should be interpreted as a communication aid, not a full joint-simulation confidence surface.

How to read this section: outputs are valuation translations of estimated coefficients, not demand forecasts or observed market-price elasticities.

Computation contract

range_units = delta_range_km / 100
charge_units = delta_charge_minutes / 15
station_units = delta_stations / 50000
body_unit = streamlined ? 1 : 0

WTP_pragmatic = 133.9 × range_units + 96.0 × charge_units + 43.7 × station_units + 43.3 × body_unit
WTP_aesthetic = 180.0 × range_units + 105.3 × charge_units + 60.5 × station_units + 67.9 × body_unit

150 km

Unit conversion: every 100 km contributes class-specific MWTP.

30 min

Unit conversion: every 15-minute reduction contributes MWTP.

100,000

Unit conversion: every 50,000 stations contributes MWTP.

Pragmatic adopters (point estimate)

480.3 million VND

Approx. 95% additive CI: [31.8, 928.8]

Aesthetic enthusiasts (point estimate)

601.6 million VND

Approx. 95% additive CI: [232.7, 970.5]

Segment premium gap (Aesthetic - Pragmatic)

121.4 million VND

Preloaded manuscript scenario: +150 km range, -30 min charge time, +100,000 stations, no streamlined premium.

Approximation note: interval aggregation assumes additive independence of attribute-level bounds. It is intentionally conservative for communication and not a replacement for full parametric simulation.

Section 9

Latent Class Profiles

Class 1 · 65%

Pragmatic adopters

Higher price sensitivity and stronger emphasis on functional certainty, especially range and practical operating constraints.

  • Price coefficient: -0.436***
  • Range coefficient: +0.584***
  • Charging station term positive but not significant
  • Streamlined body preference positive but imprecise

Class 2 · 35%

Aesthetic enthusiasts

Lower price sensitivity with continued functional priorities and some design receptivity, though not all design terms are statistically robust.

  • Price coefficient: -0.372***
  • Range coefficient: +0.670***
  • Fast-charging stations and running cost significant
  • Streamlined design premium positive but imprecise
Table 4. Class profile highlights
Characteristic Pragmatic adopters Aesthetic enthusiasts
Mean age 25.1 21.2
Male share 55.1% 75.7%
Middle/high income 31.9% 54.1%
Any EV experience 18.8% 47.3%
High environmental concern 60.9% 44.6%
Section 10

Discussion and Implications

Theoretical implication

The evidence supports a dual-evaluation pathway: utilitarian utility dominates baseline adoption calculus, while symbolic/aesthetic value adds differentiated willingness-to-pay in specific segments (Norman, 2004; Rezvani et al., 2015).

Product strategy implication

For the larger pragmatic class, affordability and range are high-confidence strategy levers. For the aesthetic class, design-led variants are justified as premium options, but pricing should reflect the wider uncertainty around some aesthetic coefficients.

Policy implication

Policy remains strongest on functional levers (cost support and charging access), while communication can integrate identity and aspiration framing as secondary accelerators rather than substitutes for infrastructure readiness (Hoang et al., 2022; Truong et al., 2024).

Section 11

Limitations and Future Research

  • Sample scope is young and urban; generalizability to older/rural populations remains untested.
  • Stated-preference design is vulnerable to hypothetical bias despite quality controls.
  • Motorcycle-centric mobility culture was not explicitly modeled as a structural moderator.
  • Some aesthetic coefficients are imprecise in class-specific estimates, limiting strong monetary claims.
  • Future work should combine revealed-preference behavior, longitudinal tracking, and emotion-linked design mechanisms.
Section 12

Conclusion

The study indicates that EV preference among young Vietnamese consumers is driven primarily by functional certainty, with a meaningful minority that also values design expression. Segment-aware pricing, product architecture, and communication can therefore outperform single-track EV positioning.

For market acceleration, strategy should align technology affordability and charging support with selective design differentiation rather than treating aesthetics as either irrelevant or universally dominant.

Section 13

Selected References

  • Bech, M. and Gyrd-Hansen, D. (2005). Effects coding in discrete choice experiments.
  • Bloch, P.H. (1995). Seeking the ideal form: product design and consumer response.
  • Chitturi, R., Raghunathan, R. and Mahajan, V. (2008). Delight by Design: Hedonic versus utilitarian benefits.
  • Buhmann, K.M. and Criado, J.R. (2023). Consumers' preferences for electric vehicles: status and reputation.
  • Creusen, M.E.H. and Schoormans, J.P.L. (2005). Roles of product appearance in consumer choice.
  • Gauer, V.H. et al. (2025). Automobility engagement and EV preferences among car buyers.
  • Greene, W.H. and Hensher, D.A. (2003). A latent class model for discrete choice analysis.
  • Hackbarth, A. and Madlener, R. (2013). Consumer preferences for alternative fuel vehicles.
  • Hensher, D.A. (2010). Hypothetical bias, choice experiments, and willingness to pay.
  • Hensher, D.A., Rose, J.M., and Greene, W.H. (2015). Applied Choice Analysis (2nd ed.).
  • Hess, S., Hensher, D.A., and Daly, A. (2012). Revisiting respondent fatigue in stated-choice experiments.
  • Hoang, T.T., Pham, T.H., and Vu, T.M.H. (2022). EV purchase decision in Vietnam.
  • Huu, D.N. and Ngoc, V.N. (2021). Urban transport status and transition to electric mobility in Vietnam.
  • Jones, L.R. et al. (2013). Incentives, technology, and electric vehicle adoption in Vietnam.
  • Khuat, V.H. (2015). Traffic Safety Strategies for Vietnam.
  • Krinsky, I. and Robb, A.L. (1986). On approximating the statistical properties of elasticities.
  • Lusk, J.L. and Schroeder, T.C. (2004). Incentive compatibility in choice experiments.
  • Manski, C.F. (1977). The structure of random utility models.
  • McFadden, D. and Train, K. (2000). Mixed MNL models for discrete response.
  • Norman, D. (2004). Emotional Design.
  • Pham, V.T. et al. (2022). Factors influencing EV purchase intention in Vietnam.
  • Revelt, D. and Train, K. (1998). Mixed logit with repeated choices.
  • Rezvani, Z., Jansson, J., and Bodin, J. (2015). EV adoption research review and agenda.
  • Truong, N. et al. (2024). Barriers to electric car and motorcycle adoption in Vietnam.
  • VAMM (2025). Vietnam motorcycle sales report (Q4 and full-year 2024).
  • Xhakollari, V., Asioli, D., and Nayga, R.M. (2025). Mitigating hypothetical bias in choice experiments.
  • Xia, Z., Wu, D., and Zhang, L. (2022). Economic, functional, and social factors in EV adoption.

Note: This page provides selected references for decision relevance. The full bibliography is available in manuscript materials.

Back to All Research