Seaplane Adoption in Greece: From Ordinal Intent to Firth-Penalized Logistic Inference

Abstract

This study examines how sentiments and perceptions in Greece relate to seaplane adoption, building on prior work on Greek users’ emotions and attitudes toward seaplane services. Using survey data from N = 443 respondents (N = 373 used in the logistic model), we estimate a binary logistic regression for the intention to choose a seaplane. Perceived comfort and safety (F3) is the dominant predictor, substantially increasing the odds of adoption (e.g., OR = 6.67, 95% CI [4.09, 11.35]; robust under Firth penalization). In the full MLE model, emotion dummies (Freedom, No feelings) are not statistically significant relative to Joy; Fear exhibits quasi-complete separation, so its MLE coefficient is not interpretable (penalized results are provided as sensitivity). Model performance indicates acceptable discrimination (AUC = 0.782, 95% CI [0.734, 0.829]). Better perceived comfort and safety are critical for broader seaplane use in island and coastal regions.

1. Introduction

A seaplane (hydroplane) is an aircraft that can take off from and land directly on water. Across island and coastal networks, seaplanes offer a critical connectivity link and time savings, notably where traditional airports are nonexistent or remote [1]. Yet the mode is hampered by high costs, sensitivity to weather, limited facilities, and heavy regulation [2].

In nations like Canada, the Maldives, and Norway, seaplanes are embedded in transport provision, chiefly for island and remote areas. They enhance regional connectivity and stimulate tourism [3,4]. Safety, however, is central due to combined aviation and maritime risks. Understanding passenger preferences is crucial to seaplane adoption. Mode choice depends not only on functional or economic factors—time, cost, convenience—but also on psychological and emotional factors, including perceived risk and trust [5,6,7]. Prior research shows that comfort, safety, and integration with the wider transport system strongly influence traveler behavior [8,9]. In island environments, emotions such as trust, fear, or joy can significantly affect decisions to use a given mode [10,11]. This suggests that traditional models focusing solely on functional attributes may be insufficient; an analytical framework that considers both cognitive (functional/economic) and emotional determinants is needed—particularly in settings with limited prior user exposure to seaplanes—so that the quantitative influence of psychological elements on travel behavior can be identified.

In Greece, recent work points to gaps in infrastructure and regulation, and to the value of technology and operations research for improving island connectivity [1,2]. Siskos, Maravas, and Mau [12] report that political, economic, technological, and social factors act both as catalysts and as obstacles to widespread application. Strengthening this mode could improve national connectivity, attract visitors, and benefit local communities [1,13]. Nevertheless, obstacles persist, such as bureaucracy. There are infrastructure gaps, as well as a need for effective promotion. The absence of clear market positioning continues to delay progress. Lengthy procedures for licensing and legalizing water-aerodrome routes are a typical example. In 2023, Hellenic Seaplanes presented the first seaplane in the Sporades as a pilot demonstration. The expansion to other areas (e.g., the Ionian and the Northern Aegean) has been gradual, leaving connectivity issues [13]. At present, services remain at a pilot or early-deployment stage, with potential destinations under evaluation.

More recently, Sitzimis et al. [11] examined local populations’ attitudes and perceptions toward seaplanes in Greece using quantitative methods. Participants considered seaplanes more environmentally friendly than airplanes and ships, more economical, and important for improving island connectivity; they also expected water-aerodrome routes to enhance tourism. The authors emphasized the need to investigate factors affecting the intention to choose seaplanes, including emotional determinants.

Building on this background, the present study examines which independent variables are associated with travelers’ likelihood of choosing a seaplane. Unlike prior work focused mainly on functional or economic factors [6,7], we also include emotional factors such as fear, joy, and neutrality (absence of a specific emotion), thereby proposing a more comprehensive model of travel behavior. In the empirical analysis, we quantify these relationships using logistic regression and discuss the method’s assumptions and limitations. This approach enriches the literature on mode choice and provides a more detailed understanding of seaplane adoption in island and coastal environments.

The results matter for both firms and policymakers. For firms, they can shape targeted campaigns that stress flexibility, access, and the travel experience, adapted to local needs to build acceptance and steady demand [10]. For policymakers, integrating public views could simplify procedures, accelerate infrastructure development, and justify targeted subsidies on low-demand routes. Given their flexibility and relatively low cost, seaplanes could play a strategic role in serving thin markets and remote islands [3,4].

The article is structured as follows. Section 2 reviews the literature. Section 3 details the data and methods; Section 4 reports the results; Section 5 interprets them and outlines policy implications; Section 6 offers the conclusion.

2. Literature Review

2.1. Mode of Transport Choice

When selecting a mode of transport, numerous factors shape passenger decisions, alongside elements tied to individual preferences. In sum, determinants of traveler behavior vary with market conditions and modal characteristics. Psychological and emotional factors increasingly influence choices.

Gärling and Axhausen [5] argued that travel choices reflect not only perceived attributes but also the feelings tied to them. A broad line of work places comfort and safety at the core. In bus settings, Eboli and Mazzulla [8] showed that service quality—especially cleanliness—drives satisfaction. Khalid et al. [14], in student scenarios, showed that comfort and safety affect mode selection. Braathen [3] emphasized air transport’s essential role for remote and island areas, with safety and reliability as foundations, and Fageda, Suarez-Aleman, Selebrisky, and Fioravanti [4] outlined policy and regulatory levers for sustainable air connectivity—context within which seaplanes fit.

Göransson and Andersson [9] argue that reliability, comfort, safety, and system integration drive public transport attractiveness, though their effect on ridership can be limited [15]. Before the pandemic, travel time and distance predominated; during lockdown, personal safety concerns and personality traits gained importance [16]. Zhang et al. [17] combined discrete choice models with machine learning. Time, cost, and convenience shaped commuter-rail choice. Machine learning improved predictive accuracy, whereas discrete choice models better captured behavior. Using a latent class logit, Li et al. [6] uncovered convenience- and price-oriented segments; environmental awareness and car ownership were additional drivers [6,17].

For the Greek islands, Rigas [10] reported that emotions such as comfort and trust matter in choices between ships and aircraft, supporting their inclusion in adoption models. Moreover, treating behavior as mere trip frequency is insufficient; frequency must be combined with psychological and perceptual variables to explain choices [17,18].

2.2. Determinants of Seaplane Choice

Recent work offers insights into infrastructure, service configuration, and perceptions toward seaplanes. Emotions matter more when information is limited, according to Loewenstein and Lerner [7], which fits early adoption of seaplanes. Across multimodal work, seaplanes are framed as vital in island settings such as Greece. Using a SWOT-based exploratory survey, Andrade, Kalakou, and Da Costa [1] identified the need for better infrastructure and a cohesive regulatory scheme. Ballis, Moschovou, Pagonakis, and Zachariadis [2] examined growth opportunities, emphasizing how operations research and technology could enhance island connectivity. Siskos, Maravas, and Mau [12] provided a PESTLE analysis showing political, economic, technological, and social factors as both opportunities and barriers to large-scale implementation.

Sitzimis, Dimou, Kourgiantakis, and Kanellis [11] examined user perceptions in Greece, revealing emotional differences (safety, joy, fear) that motivate including emotional factors in logistic models of seaplane adoption. Overall, the literature underscores infrastructure, service quality, psychological elements, and behavior. A gap remains regarding the quantitative influence of emotional responses on travel mode choices in Greece.

While general determinants of mode choice are well documented [6,7], seaplane-specific evidence is limited. The present study extends prior work by quantitatively focusing on predictors of seaplane use intention, indicating that determinants for seaplanes differ from general models and merit separate analysis.

3. Methods and Data

3.1. Data Validity and Reliability Check

To investigate the research question, we included 13 independent factors: sex (Q1), age (Q2), marital status (Q3), education (Q4), occupation (Q5), residential area (Q6), trip frequency (Q7), purpose of travel (Q8), preferred mode of transport (Q9), feelings (Q12) (see Table 1), as well as F1, F2, and F3. F1 (development of the local economy) has been linked to regional cohesion [3,12]. F2 (ecological and economic impacts) reflects sustainable mobility trends [6,12]. F3 (comfort and safety) is consistently highlighted as critical in transport mode choice [9,12]. Previous work also suggested that feelings, trip frequency, and F3 may be important predictors [11]. To ensure completeness, however, all questionnaire variables were analyzed.

Table 1. Independent variables used in the logistic regression model and justification of their inclusion.

Index Category	Variable	Description	Bibliographic Documentation
Demographic	Q1—Sex	Binary variable (male/female)	Gender differences have been documented as important in the way people choose transportation modes [10].
Demographic	Q2—Age	Categorical variable (18–24, 25–39, 40–54, 55+)	Age influences the willingness to adopt new modes of transportation [5].
Demographic	Q3—Marital status	Univariate categorical variable	Family obligations can shape different transportation needs.
Demographic	Q4—Education	Education level	Education has been linked to increased environmental awareness and adoption of new practices [6].
Demographic	Q5—Occupation	Employment status category	Occupational roles are associated with travel frequency and preferences [15].
Demographic	Q6—Residential area	Urban/island/rural area	Geographical location influences access to and perception of seaplanes [3].
Behavioral	Q7—Trip frequency	Categories (rarely, 1 time/year, 2–4 times/year, >4 times/year)	Frequency is related to familiarity with new forms of transportation [18].
Behavioral	Q8—Purpose of travel	Leisure/work/personal reasons	Purpose influences the trade-off between comfort and cost [8].
Behavioral	Q9—Preferred mode of transport	Categorical variable	Previous preference is related to travel habits [9].
Psychological	Q12—Feelings	Categories (joy, fear, freedom, no feeling, mixed feelings, safety)	The role of emotions is decisive in decisions about new forms of transportation [7,12].
Continuous factors	F1—Local development	Perception of seaplanes’ contribution to the local economy	Its contribution to regional cohesion has been highlighted [3].
Continuous factors	F2—Ecological/economic impacts	Perception about ships/airplanes	It is related to the environmental sensitivity of users [6].
Continuous factors	F3—Comfort and safety	Perception of seaplanes’ comfort/safety compared to other means	It has been recorded as a determining factor in mode choice [9].

Source: [11].

Reliability analysis (Cronbach’s α) for the multi-item constructs (F1–F3) yielded α = 0.824 for F1 (good), α = 0.685 for F2 (borderline acceptable), and α = 0.600 for F3 (moderate), based on the modeling sample (n = 373). Content validity followed Sitzimis et al. [11,19]. Item wordings, response anchors, and scoring for F1–F3 are provided in Appendix A.2.

The survey was distributed via electronic form to a sample of island and mainland residents. The data collection period took place from June to September 2024. Participants were informed about the purpose of the study (voluntary participation) and filled out the questionnaire in about 10 minutes. The completion rate was high. This enhanced the reliability of the responses.

Socio-demographic variables (gender, age, marital status, education, occupation, area of residence) were included in the logistic regression. Although examined, most showed limited predictive power. Demographics are thus important for describing the sample but not for explaining seaplane choice. Psychological and perceptual factors (e.g., emotions, comfort/safety) proved more influential. Travel decisions depend more on perceptions and feelings than on socio-demographic characteristics.

3.2. Logistic Regression

The dependent variable is Q11 (F4): “How likely would you be to choose a seaplane for your travel if this option existed?” This ordinal variable was dichotomized by assigning 0 to “not at all,” “slightly,” and “moderately,” and 1 to “very” and “very much.” Values of 0 represent a Negative or Neutral Aspect (NNA), and values of 1 a Positive Aspect (PA). This simplifies interpretation but discards some information. Robustness checks with alternative cut-offs and with ordinal specifications are reported in Appendix A.3.

Accordingly, a binomial logistic regression model was applied [19,20]. Logistic regression is widely used for modeling transport mode choice probabilities [18]. Analyses were run in R 4.5.1, with both a full Maximum Likelihood Estimation (MLE) and a Firth-penalized variant for robustness (see Table A5 for MLE coefficients, Table A6 for Firth-penalized results and Supplementary Materials).

Model assumptions were verified. For continuous predictors (F1–F3), the linearity-in-the-logit assumption was tested using standard approaches and showed no material deviations. Multicollinearity was absent (Tolerance > 0.1, VIF < 10) [21,22,23]. Some sparse categories (e.g., low-frequency feelings responses) produced quasi-complete separation in MLE; penalized (Firth) logistic regression was therefore used as a sensitivity check. Regarding sample size, rules of thumb suggest at least 20 cases per predictor with a minimum of 60 cases overall [24]. With ~15 parameters including dummies, our dataset of N = 443 respondents (N = 373 modeled) was sufficient.

3.3. Model Fit Checks and Procedure

We report the full specification and sensitivity analyses in Appendix A.3. To assess the effect of each predictor, Wald tests were employed [19,20,25,26]. Key performance and calibration/discrimination metrics are presented in the Results section, with additional calibration diagnostics in Table A1 and Table A2 (see Figure A1 for ROC and Figure A2 for calibration). Odds ratios (Exp(B)) were used for interpretation: values < 1 indicate reduced odds of adoption as the predictor increases. Values > 1 indicate higher odds, holding other variables constant [26] (Appendix A.4, Table A5 and Table A6 and Supplementary Materials).

4. Results

The basic statistical characteristics of the indicators clarify the socio-demographic profile of the sample and highlight the distribution of emotions and behavioral factors (see Table 2). Comparison with general population data allows an assessment of representativeness [27]. Rather than characterizing the sample as “largely representative,” we note clear imbalances: individuals aged 40–54 and those with higher education are overrepresented, and regional concentration is high (Crete, Attica). According to Statistics Greece’s 2021 census [27], 19% of the Greek population is 65 years or older, whereas our age binning uses 55+; thus, this comparison is not directly equivalent. In our sample, 14.4% are 55+. At the same time, 84.4% of respondents hold a higher education degree, consistent with other transport surveys where more educated citizens participate more often [10]. We therefore refrain from claiming full representativeness and interpret population inferences with caution. The effective modeling sample was N = 373 (listwise availability in the logistic model) and is considered sufficient for analysis.

Table 2. Descriptive statistics of independent variables.

Index Category	Variable	Distribution (%)
Demographic	Q1—Sex	47% Male–53% Female
Demographic	Q2—Age	10.8% ≤24, 30.9% 25–39, 43.8% 40–54, 14.4% 55+
Demographic	Q3—Marital status	50.6% Married with children, 7% Married without children, 39.5% Single, 2.3% Divorced, 0.7% Divorced with children
Demographic	Q4—Education	14.9% High school, 0.7% Post-secondary, 42.2% University, 42.2% Postgraduate/PhD
Demographic	Q5—Occupation	34.5% Private sector, 32.5% Public sector, 15.6% Self-employed, 10.8% Student, 4.5% Retired, 2% Unemployed
Demographic	Q6—Residential area	34.1% Crete, 29.3% Attica, 9.3% Central Macedonia, 4.7% South Aegean, 3–4% each other region
Behavioral	Q7—Trip frequency	0.5% Rarely, 24.6% Once/year, 48.1% 2–4 times/year, 26.9% >4 times/year
Behavioral	Q8—Purpose of travel	43.8% Leisure, 13.3% Business, 16% Personal, 8.8% Leisure and business, 8.8% Leisure and personal, 7% All reasons, 2.3% Personal and business
Behavioral	Q9—Preferred mode of transport	47.6% Airplane, 38.6% Car, 12.9% Ship, 0.9% Combination
Psychological	Q12—Feelings	38.8% Freedom, 26.6% Joy, 17.4% None, 13.8% Fear, 2.9% Mixed, 0.5% Safety

Source: [11].

As a descriptive control, younger people (25–39) reported slightly higher joy than the 55+ group, while fear was marginally more common in the 55+ group (minor differences, no inductive control). The pattern is consistent with the role of perceived safety in less familiar modes. For full percentages see Table 2. In the text we highlight only key sample imbalances.

4.1. Model Fit and Classification Performance

All model-fit metrics are reported in Table 3. (see Figure A1 for ROC and Figure A2 for calibration).

Table 3. Model fit and classification (binary logistic regression for F4 = Q11).

Metric	Value
Sample (used rows)	N = 373
Classification cut-off	0.50
Overall accuracy	0.729
PA sensitivity (accuracy_pa)	0.828
NNA specificity (accuracy_nna)	0.575
AUC (95% CI)	0.782 [0.734, 0.829]
Brier score	0.181
Hosmer–Lemeshow χ2 (df = 8), p-value	8.140, p = 0.420

Notes: Alternative cut-offs and separation diagnostics are provided in Appendix A.3. All metrics computed on used rows (N = 373).

4.2. Model Specification and Reference Coding

The estimated model uses Joy as the reference emotion and “Rarely” as the reference for trip frequency. All dummy indicators are coded 0/1. Details on separation and Firth estimation appear in Section 3.2 and Appendix A.3 and Appendix A.4.

4.3. Regression Coefficients and Effect Sizes (See Table 4)

From the full MLE model:

• F3 (comfort and safety) is a strong positive predictor: OR = 6.67, 95% CI [4.09, 11.35], p < 0.001 (Appendix A.4, Table A5 and Supplementary Materials).
• Emotion dummies (Freedom, None) are not statistically significant vs. Joy (p = 0.164 and p = 0.058, respectively). Fear was dropped by MLE due to separation (very sparse and highly predictive responses).

From the Firth (penalized) logistic sensitivity (same rows/predictors after dropping constants):

• F3 remains significant (OR = 6.31, 95% CI [3.90, 10.64], p < 0.001) (Appendix A.4, Table A6).
• Trip frequency “up to 1 time/year” shows a negative association (OR = 0.51, p = 0.043) relative to “Rarely”; other trip dummies and emotion dummies are not significant.

These results underscore the central role of perceived comfort/safety (F3) for seaplane adoption, while emotions relative to Joy do not retain significance once other factors are controlled—and separation cautions against over-interpreting sparse categories such as Fear.

Table 4. Key predictors of seaplane adoption (Binary logistic regression for F4 = Q11) (R results on used rows, N = 373. Reference categories: Feelings = Joy; Trip frequency = Rarely. Odds ratios (OR) shown with 95% CI).

Predictor	Model	OR [95% CI]	p-Value	Notes
F3 (Comfort and Safety)	MLE	6.67 [4.09, 11.35]	<0.001	Strong positive effect
F3 (Comfort and Safety)	Firth	6.31 [3.90, 10.64]	<0.001	Robust to separation
Trip frequency: up to 1 time/year (vs. Rarely)	Firth	0.51 [0.27, 0.98]	0.043	Modest negative association
Freedom (vs. Joy)	MLE	—	0.164	Not significant
None (vs. Joy)	MLE	—	0.058	Borderline, NS
Fear (vs. Joy)	MLE	—	—	Dropped (quasi-complete separation in MLE)

Notes: Some dummy categories exhibited quasi-complete separation in the MLE (using Firth’s penalized logistic regression alleviated this). We used a 0.50 classification threshold. Further checks appear in Appendix A.3 and Appendix A.4.

Table 4. Key predictors of seaplane adoption (Binary logistic regression for F4 = Q11) (R results on used rows, N = 373. Reference categories: Feelings = Joy; Trip frequency = Rarely. Odds ratios (OR) shown with 95% CI).

Predictor	Model	OR [95% CI]	p-Value	Notes
F3 (Comfort and Safety)	MLE	6.67 [4.09, 11.35]	<0.001	Strong positive effect
F3 (Comfort and Safety)	Firth	6.31 [3.90, 10.64]	<0.001	Robust to separation
Trip frequency: up to 1 time/year (vs. Rarely)	Firth	0.51 [0.27, 0.98]	0.043	Modest negative association
Freedom (vs. Joy)	MLE	—	0.164	Not significant
None (vs. Joy)	MLE	—	0.058	Borderline, NS
Fear (vs. Joy)	MLE	—	—	Dropped (quasi-complete separation in MLE)

Notes: Some dummy categories exhibited quasi-complete separation in the MLE (using Firth’s penalized logistic regression alleviated this). We used a 0.50 classification threshold. Further checks appear in Appendix A.3 and Appendix A.4.

5. Discussion

Persistent licensing and coordination delays have hindered the transition from pilot training to regular seaplane operations. Policy should focus on how to boost perceived comfort and safety through tech upgrades, clear maintenance standards, and targeted crew training. Also, cutting fear by using transparent information, demos, and test flights. These facts support prior evidence that comfort and safety materially shape mode choice [3,4,6,8,9,15,18].

Shifting just 10% of people from “Fear” to “No feelings” would lift expected uptake by roughly 1–2 percentage points (see Table A8 for predicted probability scenarios and Supplementary Materials). This means that better risk communication, social proof, and positive direct exposure to the service can nudge hesitant users. This complements—not replaces—the bigger lever of strengthening comfort and perceived safety (F3).

The pattern that negative or ambivalent affect suppresses adoption is consistent with behavioral decision theory. When a mode is unfamiliar, affective cues weigh heavily in choice [5,7]. Although trip frequency showed an overall effect in some specifications, its individual categories were not reliably different from the reference once emotions and perceptions were included (and sparse cells raised separation concerns). This implies that what travelers feel and believe about the mode matters more than how often they travel. Among respondents with no clear emotion, practical exposure (mini flights, open-day walkthroughs, high-fidelity visuals) should raise awareness and comfort.

On islands such as Skopelos and Alonissos, residents are engaged to travel by ship to reach hospitals or public services. The duration is often prohibitive. A seaplane route could reduce the journey to less than half an hour [28]. This would improve residents’ sense of safety and trust. Comfort and safety (F3) are operational necessities in daily travel, not theoretical appraisals.

A practical deployment plan can move forward in four stages: (i) brief demonstration flights with public safety briefings; (ii) display service design cues that indicate reliability (standardized checklists, wayfinding, boarding choreography); (iii) coordination with health services and municipalities on scaling use; and (iv) timed feedback cycles (post-flight micro surveys) for comfort and perceived safety (F3).

Reforms to permitting policies alone are insufficient. Focus on investing resources, both real and perceived, in terms of safety/comfort (in relation to aircraft, components, workforce training, and communication). For low-demand island links, targeted subsidies may be warranted [4]. Make the rollout align with the segmented market, addressing both cognitive and affective drivers in light of the implementation efforts. Involve a collaborative strategy with local stakeholders to build buy-in within the same organization. Utilize pilot services in conjunction with structured, repeated feedback to establish trust. The emotional variable was elicited after a video demonstration; collecting in situ data during the demonstration flights would enhance the external validity.

Overall, the results suggest that seaplanes could add value to Greece’s multimodal system—especially to islands—if safety and comfort are considered in the operating processes and awareness of user emotion is acknowledged by operators and authorities.

6. Conclusions

This study examined whether seaplanes can be a feasible mode of transportation for the Greek population. Building on Sitzimis et al. [11], we used binary logistic regression to identify demographic, emotional, and perceptual factors associated with individuals’ propensity to choose a seaplane. There are important implications for transport policy, the design of the delivered service, and tourism development. In summary, strengthening comfort and safety is essential.

Policy should move on two fronts. First, it needs to strengthen real safety and comfort (fleet technology, maintenance, operating standards, training). Second, it should improve perceived safety and comfort (transparent communication, wayfinding, service design that signals reliability). Streamlining water aerodrome licensing and considering targeted incentives on thin island routes are also important. Finally, seaplanes should be made part of the overall transport–tourism mix and should be co-designed with local actors to build credibility.

Limitations apply. The sample covers local residents, not international visitors. Price and income were not modeled, though they shape substitution with ferries and short-haul air services. The design is cross-sectional. Longitudinal data would track change gradually. The emotion measure came from a video stimulus; in situ experience sampling after demonstration flights would strengthen external validity.

In addition to regular services, seaplanes can be integrated into emergency/medical transport to small islands, where time is crucial. Interconnecting tickets with ferries and regional buses reduces perceived effort and improves trust, complementing safety communication [29].

Seaplanes can bolster Greece’s transport mix on island and coastal links. Success requires attention to safety, comfort, and reliability (technical aspect) but also to trust, emotions, and experience (human aspect). Communication should address feelings as well as facts.

Appendix A.1. Questionnaire (Full Instrument)

The complete questionnaire used in the survey is included to ensure transparency and reproducibility.

• Questionnaire: Perceptions of Seaplanes in Greece
• Section A: Demographic Information
• Q1. Gender
• ☐ Male ☐ Female
• Q2. Age
• ☐ 18–24 ☐ 25–39 ☐ 40–54 ☐ 55 and above
• Q3. Marital status
• ☐ Single ☐ Married without children ☐ Married with children
• Q4. Educational level
• ☐ Primary/Secondary education ☐ High school ☐ University/College degree ☐ Master’s/Doctorate
• Q5. Employment status
• ☐ Public sector employee ☐ Private sector employee ☐ Self-employed ☐ Retired ☐ Unemployed ☐ Student
• Q6. Region of residence
• (Open-ended)
• Section B: Travel Habits
• Q7. Frequency of travel per year
• ☐ Rarely ☐ Up to 1 time ☐ 2–4 times ☐ More than 4 times
• Q8. Main purpose of travel (tick all that apply)
• ☐ Leisure ☐ Business ☐ Visiting relatives/friends ☐ Other
• Q9. Usual means of transportation
• ☐ Airplane ☐ Ship ☐ Road transport (car, bus, train)
• Section C: Perceptions of Seaplanes
• Response scales:
• —Yes/No/Maybe items: 1 = No, 2 = Maybe, 3 = Yes
• —Likert items: 1 = Not at all … 5 = Very much
• Q10. How safe do you consider seaplanes compared to other means of transport?
• Q11. How likely would you be to choose a seaplane for your travel if this option existed?
• Q12. Please watch the seaplane flight video (link). Which feeling does it evoke most strongly in you?
• ☐ Joy ☐ Freedom ☐ Fear ☐ None/No particular feeling ☐ Mixed feelings ☐ Safety
• Q13. Which regions of Greece are most suitable for the development of water aerodromes?
• (Open-ended—e.g., Crete, Cyclades, Ionian Islands, Sporades, etc.)
• Q14. Seaplanes provide a more environmentally sustainable option than airplanes. (Yes/Maybe/No)
• Q15. Seaplanes provide a more environmentally sustainable option than ships. (Yes/Maybe/No)
• Q16. Seaplanes will be more cost-effective than airplanes. (Yes/Maybe/No)
• Q17. Seaplanes will be more cost-effective than ships. (Yes/Maybe/No)
• Q18. Seaplanes will improve connections between islands and mainland Greece. (1–5)
• Q19. Seaplanes will improve connectivity among islands. (1–5)
• Q20. Seaplanes will provide a more comfortable journey than airplanes. (1–5)
• Q21. Seaplanes will provide greater comfort than ships. (1–5)
• Q22. Establishing water aerodromes will enhance tourism and regional development. (1–5)
• Q23. Establishing water aerodromes will create new jobs in the region. (1–5)
• Q24. Establishing water aerodromes will attract foreign investment. (1–5)

Appendix A.2. Items, Anchors and Scoring

• Response scales

• Yes/Maybe/No (Q14–Q17): coded as 1 = No, 2 = Maybe, 3 = Yes and then mapped to 1/3/5 for comparability.
• Likert (Q10, Q20–Q24): 1 = Not at all … 5 = Very much. Higher values indicate stronger agreement/perception.

• Constructs

• F1—Local development = mean (Q22, Q23, Q24).
• F2—Ecological and economic = mean (Q14, Q15, Q16, Q17) after 1/3/5 mapping.
• F3—Comfort and safety = mean (Q10, Q20, Q21).

• Outcome

• F4 (Q11) dichotomized: 0 for 1–3 (Not at all/Slightly/Moderately), 1 for 4–5 (Very/Very much).

• Feelings (Q12) categories and reference

• Categories: Joy (reference), Freedom, Fear, None/No particular feeling, Mixed feelings, Safety.
• Dummies in models: freedom, fear, none, mixed, safety (Joy = reference).

• Question wording (EN/GR synopsis)

• Q10: perceived safety of seaplanes vs. other modes (1–5).
• Q11: likelihood of choosing a seaplane if available (1–5).
• Q12: dominant feeling after video (Joy/Freedom/Fear/None/Mixed/Safety).
• Q14–Q17: environmental and cost comparisons vs. airplanes/ships (Yes/Maybe/No).
• Q20–Q21: comfort vs. airplanes/ships (1–5).
• Q22–Q24: tourism, regional development, jobs/investment from water aerodromes (1–5).

• Reliability (Cronbach’s α; used rows, n = 373)

• F1 (Local development): α = 0.824 (3 items)
• F2 (Ecological and economic): α = 0.685 (4 items)
• F3 (Comfort and safety): α = 0.600 (3 items)

Appendix A.3. Model Performance and Robustness

Table A1. Overall classification metrics (MLE, n = 373).

Metric	Value
Overall accuracy	0.729
Accuracy (NNA = 0)	0.575
Accuracy (PA = 1)	0.828
AUC (95% CI)	0.782 [0.734, 0.829]
Brier score	0.181
Hosmer–Lemeshow χ2 (df = 8), p-value	8.140 (p = 0.420)

Table A1. Overall classification metrics (MLE, n = 373).

Metric	Value
Overall accuracy	0.729
Accuracy (NNA = 0)	0.575
Accuracy (PA = 1)	0.828
AUC (95% CI)	0.782 [0.734, 0.829]
Brier score	0.181
Hosmer–Lemeshow χ2 (df = 8), p-value	8.140 (p = 0.420)

Figure A1. ROC curve showing discrimination of the logistic model (AUC = 0.782).

Figure A2. Calibration plot across 10 deciles of predicted probabilities.

Table A2. Confusion matrix at classification cut-off = 0.50 (rows = observed, columns = predicted).

Observed\Predicted	0	1
0	84	62
1	39	188

Table A2. Confusion matrix at classification cut-off = 0.50 (rows = observed, columns = predicted).

Observed\Predicted	0	1
0	84	62
1	39	188

Table A3. Cut-off sensitivity analysis (overall accuracy, class-specific accuracies).

Cut-Off	Overall acc	Accuracy NNA	Accuracy PA
0.40	0.737	0.925	0.445
0.50	0.729	0.575	0.828
0.60	0.686	0.705	0.658

Table A3. Cut-off sensitivity analysis (overall accuracy, class-specific accuracies).

Cut-Off	Overall acc	Accuracy NNA	Accuracy PA
0.40	0.737	0.925	0.445
0.50	0.729	0.575	0.828
0.60	0.686	0.705	0.658

Table A4. Calibration by deciles (observed PA vs. mean predicted probability).

Decile Mean ŷ	Observed PA
0.17	0.16
0.32	0.27
0.44	0.43
0.52	0.68
0.60	0.70
0.66	0.79
0.70	0.84
0.77	0.88
0.83	0.86
0.91	0.89

Table A4. Calibration by deciles (observed PA vs. mean predicted probability).

Decile Mean ŷ	Observed PA
0.17	0.16
0.32	0.27
0.44	0.43
0.52	0.68
0.60	0.70
0.66	0.79
0.70	0.84
0.77	0.88
0.83	0.86
0.91	0.89

Appendix A.4. Effects and Presentation Tables

Odds ratios and confidence intervals are summarized in Table 4; this appendix retains only coefficients (B), SE, CI, and p-values.

Table A5. Maximum Likelihood Estimation (MLE)—coefficients (B), SE, z, p.

Term	B	SE	z	p
Intercept	−4.915	0.964	−4.64	<0.001
F1	0.105	0.115	0.84	0.401
F2	−0.166	0.160	−1.08	0.279
F3	1.897	0.260	7.45	<0.001
Feel_Freedom	0.022	0.251	0.09	0.929
Feel_None	−0.446	0.309	−1.44	0.150
Trip_1	−0.630	0.311	−2.03	0.042
Trip_2	0.152	0.274	0.56	0.578
Trip_3	Dropped (separation)

Table A5. Maximum Likelihood Estimation (MLE)—coefficients (B), SE, z, p.

Term	B	SE	z	p
Intercept	−4.915	0.964	−4.64	<0.001
F1	0.105	0.115	0.84	0.401
F2	−0.166	0.160	−1.08	0.279
F3	1.897	0.260	7.45	<0.001
Feel_Freedom	0.022	0.251	0.09	0.929
Feel_None	−0.446	0.309	−1.44	0.150
Trip_1	−0.630	0.311	−2.03	0.042
Trip_2	0.152	0.274	0.56	0.578
Trip_3	Dropped (separation)

Table A6. Firth logistic regression (penalized likelihood)—coefficients (B), SE, 95% CI, p.

Term	B	SE	95% CI (Low)	95% CI (High)	p
Intercept	−4.777	1.044	−6.901	−2.769	<0.001
F1	0.103	0.124	−0.143	0.347	0.411
F2	−0.160	0.171	−0.503	0.175	0.352
F3	1.841	0.253	1.361	2.365	<0.001
Feel_Freedom	0.056	0.269	−0.475	0.586	0.836
Feel_None	−0.463	0.329	−1.116	0.185	0.161
Trip_1	−0.668	0.330	−1.324	−0.021	0.043
Trip_2	0.180	0.288	−0.387	0.748	0.533

Table A6. Firth logistic regression (penalized likelihood)—coefficients (B), SE, 95% CI, p.

Term	B	SE	95% CI (Low)	95% CI (High)	p
Intercept	−4.777	1.044	−6.901	−2.769	<0.001
F1	0.103	0.124	−0.143	0.347	0.411
F2	−0.160	0.171	−0.503	0.175	0.352
F3	1.841	0.253	1.361	2.365	<0.001
Feel_Freedom	0.056	0.269	−0.475	0.586	0.836
Feel_None	−0.463	0.329	−1.116	0.185	0.161
Trip_1	−0.668	0.330	−1.324	−0.021	0.043
Trip_2	0.180	0.288	−0.387	0.748	0.533

Table A7. Average Marginal Effects (AME).

Continuous Predictors	AME
F1	+0.022
F2	-0.037
F3	+0.408
Binary predictors
Freedom	+0.005
None	-0.095
Trip_1	-0.124
Trip_2	+0.029
Trip_3	+0.077

(AMEs = average discrete change for 0→1 in binaries; average slope for continuous.)

Table A7. Average Marginal Effects (AME).

Continuous Predictors	AME
F1	+0.022
F2	-0.037
F3	+0.408
Binary predictors
Freedom	+0.005
None	-0.095
Trip_1	-0.124
Trip_2	+0.029
Trip_3	+0.077

(AMEs = average discrete change for 0→1 in binaries; average slope for continuous.)

Table A8. Representative predicted probabilities (Baseline: Joy and Rarely; F1–F3 at sample means).

Scenario	Predicted Probability
Baseline (Joy, Rarely)	0.589
Freedom vs. Joy	0.594
None vs. Joy	0.494
Trips: up to 1	0.465
Trips: 2–4	0.617
Trips: >4	0.664

Table A8. Representative predicted probabilities (Baseline: Joy and Rarely; F1–F3 at sample means).

Scenario	Predicted Probability
Baseline (Joy, Rarely)	0.589
Freedom vs. Joy	0.594
None vs. Joy	0.494
Trips: up to 1	0.465
Trips: 2–4	0.617
Trips: >4	0.664