با همکاری انجمن اقتصاد کشاورزی ایران
نوع مقاله : مقالات پژوهشی به زبان انگلیسی
نویسندگان
1 گروه آموزشی اقتصاد کشاورزی.دانشکده کشاورزی.دانشگاه تبریز
2 گروه آموزشی اقتصاد کشاورزی، دانشکده کشاورزی، دانشگاه تبریز، ایران.
3 گروه آموزشی اقتصاد کشاورزی، دانشکده کشاورزی، دانشگاه تبریز.ایران
کلیدواژه‌ها
عنوان مقاله English
نویسندگان English
Extended Abstract
1. Introduction
The global organic food market has expanded considerably over the past decade, yet a puzzling contradiction persists across many emerging economies: consumers routinely report favourable attitudes toward organic products while actual purchase behaviour remains surprisingly low. This attitude–behaviour gap is especially pronounced in Iran, where certified organic market share stands below two percent despite growing urban demand and widespread health and environmental awareness. Understanding why this gap endures is not merely an academic question; it carries direct implications for sustainable development goal twelve and for designing evidence‑based interventions in sanction‑affected, collectivist societies.
Traditional behavioural theories, including the Theory of Planned Behavior and Social Learning Theory, have struggled to explain this gap in structurally constrained environments. These frameworks typically assume linear, additive relationships and cannot easily accommodate non‑linear thresholds, conditional interactions, or the subtle role of social desirability bias (SDB) in shaping what people say versus what they actually do. Moreover, in economies under prolonged sanctions—characterised by hyperinflation, currency devaluation, and eroded household purchasing power—structural constraints intersect with cultural pressures toward conformity, creating a uniquely complex decision landscape.
This study addresses three core questions. First, can interpretable machine learning models predict consumer trust levels and uncover the hidden architecture of the attitude–behaviour gap more effectively than conventional linear models? Second, do interactions between willingness to pay (WTP) and SDB produce identifiable economic and psychological thresholds that must be jointly satisfied before trust can translate into actual purchasing behaviour? Third, how can explainable AI methods validate the conditional moderation mechanism hypothesised in the proposed Trust Amplification Model (TAM), which integrates economic feasibility, digital peer influence, and collectivist norms?
2. Materials and Methods
The study was conducted in Tabriz, Iran—a city that offers a valuable analytical context due to the coexistence of growing organic product availability and persistent affordability pressures linked to international sanctions. A multi‑phase, mixed‑mode survey was administered during 2024. Data collection combined in‑person structured questionnaires with online distribution, followed by convenience and snowball sampling to capture a broad cross‑section of urban grocery decision‑makers. Inclusion criteria were age eighteen years or older, residency in Tabriz, and having purchased food within the previous six months. Ethical approval was obtained from the University of Tabriz Ethics Committee, and all participants gave informed consent. The final analytical sample comprised 450 respondents after preprocessing.
The target variable—consumer trust—was operationalised as a composite measure combining a single five‑point Likert item with a validated four‑item trust scale, treated as an ordinal variable with five ordered classes. Predictors were organised into two categories. The first included demographic and behavioural features: age, income, education, household size, gender, marital status, healthy purchasing behaviour, prior organic purchase experience, digital review‑seeking, awareness of healthy products, perceptions of sanctions, presence of vulnerable household members, and willingness to pay. The second category comprised latent psychometric constructs: Social Desirability Bias, Digital Peer Influence, Environmental Attitude, and Willingness to Pay as a premium percentage.
Confirmatory Factor Analysis demonstrated excellent model fit (χ²/df = 2.41, CFI = 0.962, TLI = 0.954, RMSEA = 0.052, SRMR = 0.044). Cronbach's alpha ranged from 0.811 to 0.902, composite reliability exceeded 0.80, and average variance extracted surpassed 0.50, confirming internal consistency and convergent validity.
A two‑model strategy was adopted. First, a Random Forest classifier served as a benchmark. Second, an ordinal XGBoost classifier was fine‑tuned as the primary analytical model. A 5×5 nested cross‑validation framework was used to prevent data leakage, with SMOTE applied exclusively within training folds to address class imbalance. For interpretability, SHAP provided global feature importance rankings, and partial dependence plots visualised non‑linear marginal effects and interactions, particularly between SDB and WTP.
3. Results, Discussion and Implications
Predictive performance. XGBoost consistently outperformed both Random Forest and ordinal logistic regression across all metrics. The final model achieved a weighted F1 score of 0.682, Cohen's weighted kappa of 0.641, and a strong ROC AUC of 0.955—a seventeen percent improvement over the linear baseline. This gap empirically confirms that trust formation and the attitude–behaviour gap in this context are governed by non‑linear, higher‑order dynamics that traditional econometric models cannot adequately capture.
The dual‑threshold mechanism. The partial dependence contour plot reveals that high trust does not arise from additive effects of economic and psychological factors. Instead, it requires two thresholds to be crossed simultaneously. First, an economic feasibility threshold: at a price premium below approximately twenty‑five percent, the predicted probability of high trust remains near zero—even for individuals with strong pro‑environmental or conformity tendencies. Willingness to pay functions as a structural gatekeeper. Second, a psychological activation threshold: among respondents who surpass the twenty‑five percent premium, high trust only materialises when Social Desirability Bias exceeds approximately 4.0. Below that point, even economically capable consumers do not reliably exhibit high trust.
This dual‑threshold structure reframes SDB not as measurement noise or a simple direct predictor, but as a stealth moderator—a latent psychological force that becomes behaviourally relevant only once financial constraints are sufficiently relaxed. The finding resolves a long‑standing paradox: surveys in many emerging markets routinely report strong pro‑environmental attitudes alongside persistently low purchase rates. The answer is not that consumers are insincere, but that normative motivations remain dormant until economic barriers are lowered.
Social and digital dynamics. Post‑SMOTE analysis shows that marital status, social media opinions, and healthy purchasing behaviour achieve F1 scores above 0.91, surpassing WTP. Married consumers, particularly those with children, display higher trust. Digital platforms, including Instagram and Telegram, serve as influential channels for observational learning, partially substituting for weak institutional certification credibility. However, even strong digital peer influence cannot overcome the twenty‑five percent economic threshold; above that threshold, it amplifies trust, but below it, it merely generates cognitive dissonance.
Policy implications. Because the twenty‑five percent price premium operates as the primary structural barrier, the most effective policy lever is to reduce effective price premiums through targeted subsidies, tax reductions, or income‑based vouchers. Second, blockchain‑based certification systems can provide immutable, verifiable proof of organic origin—a technological substitute for weak institutional trust. Third, marketing strategies should shift from symbolic status messaging toward pragmatic, health‑oriented communication delivered through verified digital channels, leveraging peer influence as credible social proof.
Conclusion. This study demonstrates that combining behavioural theory with interpretable machine learning offers a scalable, mechanism‑based framework for understanding sustainable consumption under economic constraints. For emerging economies grappling with the attitude–behaviour gap, the path forward lies not in more awareness campaigns alone, but in systematically removing the economic gate that keeps normative motivations from turning into action. The Trust Amplification Model, with its dual‑threshold logic and reconceptualisation of social desirability bias as a conditional moderator, provides both a theoretical contribution and a practical tool for strengthening trust and promoting organic food adoption in structurally challenging environments.
کلیدواژه‌ها English
