Environmental Accounting in Albania: Challenges, Perceptions, and Factors Influencing Implementation

Abstract

Environmental accounting adoption remains limited in transitional economies, particularly where formal institutions fail to enforce sustainability mandates. We examine this phenomenon in Albania—an EU candidate country with regulatory requirements but no implementation infrastructure. Drawing on institutional-void theory and resource-based perspectives, we test whether adoption mechanisms diverge when external enforcement is weak. Survey data from 151 Albanian non-financial companies, analyzed using ordinal logistic regression, show that firm size predicts adoption, whereas sector, ownership, and market orientation do not. Critically, individual-level factors—managerial environmental knowledge and pro-environmental values—significantly predict adoption, while external institutional factors exert negligible influence. Analysis of Corporate Sustainability Reporting Directive readiness reveals similar patterns: internal organizational capacities support preparation, whereas external support remains insufficient. These findings demonstrate how institutional voids shape sustainability accounting and provide empirical evidence from an understudied Balkan context.

1. Introduction

Environmental accounting—systematically integrating ecological costs and environmental liabilities into organizational financial reporting systems—provides a mechanism for internalizing externalities and supporting evidence-based environmental decision-making [1,2]. Unlike conventional accounting frameworks that treat environmental effects as external costs, systematic environmental accounting operationalizes the “polluter pays” principle and enables organizations to quantify environmental performance, identify cost efficiencies, and meet disclosure requirements [3]. This integration is critical as regulatory regimes increasingly mandate environmental reporting, making it necessary to understand adoption mechanisms in institutional contexts with limited enforcement capacity.

EA systems yield tangible organizational benefits: reduced operational costs through resource-efficiency improvements [4], enhanced investor confidence via transparent disclosure [5,6], strengthened corporate reputation and risk management [7], and improved long-term financial performance [8,9,10]. Organizations that implement environmental accounting identify market opportunities, develop green products meeting international standards, and access environmentally conscious consumer segments [11,12]. These financial and strategic benefits drive adoption in developed economies where enforcement mechanisms are strong and stakeholder pressure is substantial.

International frameworks and regulatory initiatives have accelerated EA implementation globally. The Global Reporting Initiative (GRI) standards, established in 1997 and updated in 2021, provide voluntary guidelines adopted by approximately 4500 organizations worldwide [13]. The European Union’s Corporate Sustainability Reporting Directive [14] mandates comprehensive environmental and sustainability reporting for all large enterprises by 2025 and for SMEs by 2029 [15]. ISO 14001 certification [16], which indicates an environmental management system that enables EA, has been obtained by over 400,000 organizations internationally [17]. Recent evidence documents accelerating ESRS adoption across EU member states [18], though implementation in candidate countries remains nascent [19].

Environmental accounting implementation exhibits stark geographic and institutional disparities. Developed economies—such as European Union member states, OECD nations, and Australia—demonstrate advanced and often mandatory EA adoption. In contrast, developing and transitional countries, despite growing sustainability awareness, lag in implementation [20,21,22]. South Asia often combines regulatory mandates with minimal enforcement [23,24,25]. The Western Balkans presents a paradox: EU-candidate countries face pressure to adopt European standards but lack institutional infrastructure for implementation [26].

Albania exemplifies this paradox. As an EU candidate with expected accession by 2030, Albania must align with CSRD [27], GRI [28], and ISO 14001 standards [29] yet it has minimal environmental accounting infrastructure. The National Accounting Council’s 2019 Non-Financial Reporting Guideline [30], transposing EU Directive 2014/95/EU, requires companies with more than 500 employees to disclose environmental impacts but provides no standardized formats or KPIs [26]. The 2022 Law on Environmental Protection (No. 10/431) mandates EA reporting but lacks implementation mechanisms, enforcement procedures, and technical standards [31]. The 2020 Law on Climate Change requires emissions reporting from large industries but offers no guidance or support mechanisms [32]. Professional infrastructure is virtually nonexistent: Albanian universities do not offer environmental accounting curricula; the national accountancy body provides no continuing professional development in EA.

World Bank data (2024) show that only 8.3% of Albanian firms maintain environmental cost records—substantially below Serbian (15.2%) and North Macedonian (11.7%) comparators [33]. A 2024 UNDP [26] survey indicates that, while 62% of surveyed Albanian firms recognize sustainability’s importance, only 12% publish environmental reports [26]. This implementation gap persists despite EU-driven pressure, suggesting that formal regulations and international standards, when unsupported by institutional structures and enforcement mechanisms, do not automatically drive adoption. Organizational scholars term this phenomenon an “institutional void” [34].

Previous research has examined environmental accounting adoption in developed economies [35,36,37,38] and in Asian transitional contexts [39,40], yet Albania remains understudied despite its unique characteristics: EU-candidate status that creates anticipatory regulatory pressure, weak institutional enforcement that produces institutional voids, a predominantly SME-based economy that limits resource capacity, and minimal professional infrastructure. Regional studies examine Serbia [41] and North Macedonia [19], but no quantitative research tests environmental accounting adoption mechanisms in Albania’s institutional context [42]. This gap limits understanding of whether adoption drivers identified in developed and Asian contexts apply to European transitional economies with weak governance.

The persistence of this implementation gap despite regulatory and international pressure exemplifies what organizational scholars term an “institutional void”—contexts where formal rules coexist with absent enforcement infrastructure. In institutional voids, adoption mechanisms may diverge fundamentally from developed-economy patterns where stakeholder pressure and institutional coercion drive implementation. Instead, organizations may substitute internal capacities and managerial commitment for absent external coordination. This theoretical reframing is critical: if Albanian firms cannot rely on external institutional pressure (weak enforcement, minimal stakeholder activism), what mechanisms actually drive adoption? Do internal managerial factors—knowledge, environmental values, organizational commitment—become the primary adoption drivers? And how do firms prepare for future EU mandates (CSRD by 2030) when current institutional infrastructure is absent?

This study addresses these questions by investigating whether institutional-void theory explains EA adoption mechanisms in European transitional contexts—a domain entirely absent from empirical research. Albania provides an ideal test case: EU-candidate status creates regulatory pressure, yet institutional weakness creates implementation voids. This configuration allows us to test whether adoption mechanisms in European transitional economies resemble Asian patterns (studied previously) or reveal fundamentally new mechanisms specific to pre-accession Balkan contexts. This study does so by examining determinants of environmental accounting adoption in Albanian businesses using quantitative survey research. We test three hypotheses that integrate resource-based view theory, stakeholder theory, Legitimacy theory, and institutional theory perspectives. Specifically, we investigate whether (H1) firm structural characteristics (company size, sector, ownership, market orientation) predict EA implementation; (H2) internal organizational factors (managerial EA knowledge, pro-environmental attitudes, organizational capacity) outweigh external institutional factors (regulatory pressure, government support, stakeholder expectations) in driving adoption; and (H3) organizational implementation readiness for mandatory EU Corporate Sustainability Reporting Directive compliance depends on internal capabilities independent of external support. Using survey data from 151 Albanian non-financial companies, we employ ordinal logistic regression to test these hypotheses. The findings will inform policy design for EU-candidate countries preparing for CSRD implementation and will contribute to institutional-void theory by testing whether adoption mechanisms differ between Asian and European transitional contexts.

2. Literature Review

2.1. Theoretical Framework for Environmental Accounting Implementation

This study integrates four theoretical perspectives that explain the mechanisms underlying environmental accounting adoption.

Table 1. Integrated theoretical framework.

Theory	Applies at	Predicts	Operationalization	Albanian Evidence
Stakeholder Theory [15]	External	WHO pressures firm to adopt EA	NGO engagement, regulatory enforcement level, investor demand	Low NGO engagement (~12%); ~0.3 inspections/firm/year; <0.5% ESG investment
Legitimacy Theory [43]	Normative	WHY firm chooses to respond	Environmental norms, professional expectations, societal pressure	Professional norms weak; <2 EA consultants; 0 university EA programs
Resource-Based View (RBV) Theory [44]	Internal	WHAT enables adoption	Manager knowledge, financial resources, absorptive capacity	Weak EA knowledge (≈50% untrained); SME-dominant economy
Institutional-Void Theory [45]	Contextual	WHEN/HOW adoption mechanisms operate	Regulatory quality, institutional density, market development	Regulatory quality ≈ 103 rd; <1 NGO per 100 k; ESG market nascent

synthesizes each theory’s predictive focus, operationalization approach, and empirical manifestation in Albania.

These theories operate at different organizational levels and address distinct adoption drivers. Stakeholder theory identifies who exerts adoption pressure—primarily, external constituencies [15]. Legitimacy theory explains why organizations respond—namely, to close gaps between organizational behavior and societal expectations [43]. Resource-based view theory clarifies what enables adoption emphasizing, financial resources, human expertise, and managerial commitment [44]. Institutional theory predicts when and how adoption occurs, driven by coercive regulatory requirements, normative professional standards, or mimetic peer pressures [45]. In developed economies, these theories typically predict convergence: external pressures (stakeholder, legitimacy, and institutional) align with internal capacity (resource-based view), resulting in widespread adoption among capable firms.

However, in Albania’s institutional-void context, the theories predict divergence: weak external pressures (limited stakeholder activism, minimal legitimacy threats, and negligible institutional coercion) combine with severe resource constraints (a limited professional infrastructure and an underdeveloped workforce). This configuration implies that internal organizational factors—particularly managerial knowledge and environmental values—should become central determinants of adoption, representing a theoretical reversal with significant implications for sustainability transitions in institutional-void contexts.

2.2. Global Patterns: Environmental Accounting in Developing and Transitional Economies

Empirical evidence from developing countries documents persistent barriers to environmental accounting implementation [39,46]. Systematic reviews identify limited regulatory frameworks [47], severe capacity constraints, and weak institutional pressures as the primary obstacles across Asia and Africa [24,39,48]. These patterns persist despite growing international attention to sustainability reporting and the adoption of the United Nations Sustainable Development Goals [49]. Recent large-scale analyses reveal complex adoption mechanisms in Global South contexts. Evidence shows that the effect of regulatory pressure is often mediated by managerial environmental attitudes [50], indicating that even when formal regulations exist, organizational adoption requires strong internal commitment. Additional findings indicate that firms with ISO 14001 certification exhibit significantly higher environmental accounting implementation rates than non-certified firms, underscoring the importance of capability-building over mere regulatory compliance [51,52,53].

In transitional economies, empirical evidence shows that European Union accession pressure accelerates environmental accounting adoption primarily among large, export-oriented enterprises, while domestic SMEs remain less responsive to regulatory convergence [19,54,55,56,57]. This pattern is relevant for Albania, which is simultaneously navigating EU accession requirements while maintaining a predominantly domestic-market, SME-based business structure.

2.3. Environmental Accounting in the Western Balkans: Regional Context

The Western Balkans region exhibits particularly low environmental accounting implementation despite EU regulatory harmonization requirements. Serbia reports only 15–20 companies with comprehensive sustainability reports; North Macedonia and Bosnia report even fewer [19,41]. Although ESRS (European Sustainability Reporting Standards) requirements are coming into effect for EU-candidate countries, the region lacks the institutional infrastructure necessary for implementation, including professional training, technical guidance, and consulting services [49].

Regional capacity gaps are well documented. Only 8% of Serbian SMEs maintain systematic environmental records [19], and similar patterns are reported in North Macedonia [56]. The absence of regulatory enforcement mechanisms remains the primary barrier across all Western Balkan countries [46,54]. These regional patterns provide important context for Albania, which shares Balkan geographic, historical, and institutional characteristics yet faces an even more constrained implementation situation.

2.4. Albanian Context: Environmental Accounting in an Institutional Void

Albania exemplifies an extreme institutional void. Environmental disclosure is entirely voluntary [58], and no regulatory enforcement mechanisms exist. The 2022 Law on Environmental Protection (No. 10/431) establishes the “polluter pays” principle but provides no standardized accounting methodologies [31]. The 2020 Law on Climate Change (No. 155/2020) mandates emissions reporting for large industries but lacks implementation regulations [32]. The National Accounting Council’s 2019 Non-Financial Reporting Guideline [30], requires companies with more than 500 employees to disclose environmental impacts but offers no standardized formats or key performance indicators [58].

Professional infrastructure for environmental accounting is also severely underdeveloped. Albanian universities provide minimal curriculum coverage; the national accountancy body offers no continuing professional development courses on sustainability reporting; and only two consulting firms nationally advertise sustainability services [59]. A sectoral analysis of ISO 14001 certification—the international standard for environmental management systems that enable environmental accounting—reveals institutional divides: Albania registered 68 certificates in 2023, representing only 0.19% business-population penetration compared with a 0.45% regional average [16]. This pattern indicates that environmental accounting adoption follows foreign institutional channels rather than domestic institutional drivers [60].

Recent assessments quantify these implementation gaps. UNDP’s 2024 SDG Business Pioneers Survey [26] of 150 Albanian companies found that although 62% recognized the importance of sustainability, only 18% systematically measured environmental impacts, and merely 12% published environmental reports [26]. The World Bank’s 2024 Enterprise Survey found that only 8.3% of Albanian firms maintained environmental cost records—substantially below regional comparators (Serbia 15.2%; North Macedonia 11.7%) [33].

Two institutional–cultural factors merit particular attention. First, Albania’s high corruption-perception rank (101st globally) [61] generates a systematic distrust in regulatory institutions, leading firms to view environmental reporting skeptically as a potential rent-seeking mechanism rather than a legitimate accountability tool [25]. Second, the large informal economy (30–35% of GDP) means that substantial economic activity operates entirely outside formal accounting systems [26]. Together, these institutional and cultural factors suggest that environmental accounting implementation in Albania may require adoption drivers fundamentally different from those effective in higher-governance contexts.

2.5. Barriers and Influencing Factors for Environmental Accounting Adoption

The international literature shows that environmental accounting (EA) implementation in developing, and transitional economies is shaped by multiple interrelated challenges rather than a single theoretical driver [39,46,47,55]. These challenges can be grouped into five primary domains:

• Resource and Capacity Constraints: Limited financial, human, and technical resources are widely documented as fundamental barriers to EA implementation, particularly for small and medium-sized enterprises [39,46]. Resource-based view (RBV) theory posits that resource sufficiency is a necessary precondition for advanced environmental management practices [29,62]. Studies across both developed and developing contexts confirm that firm size and resource base significantly increase the likelihood of adoption [23,54,57].
• Institutional and Regulatory Weaknesses: Numerous studies highlight the critical role of strong regulatory frameworks and institutional arrangements. In settings characterized by institutional voids, external drivers such as regulation, professional standards, or mimetic isomorphism function weakly, leading to highly variable and often minimal practice [10,45]. Weak enforcement, vague standards, and the absence of penalties for non-compliance contribute to inconsistent and superficial adoption even when formal legislation exists [25,39,63].
• Stakeholder and Market Pressure: Stakeholder pressure—from customers, suppliers, investors, NGOs, or media—remains a major determinant of voluntary EA reporting in most contexts [15,54,64]. However, in developing and transitional economies, stakeholder activism is often limited or indirect due to weak civil society, fragmented media, or a lack of market demand [39,46,47,50]. Export-oriented and foreign-owned firms typically face greater sustainability disclosure requirements, whereas domestically oriented SMEs do not [19,56].
• Organizational and Cultural Barriers: Internally, lack of managerial knowledge, limited environmental awareness, and resistance to non-core change are frequently identified as key barriers across contexts [9,24,65,66]. Recent reviews have found that unless sustainability education and training are embedded for managers and accountants, adoption remains sporadic and vulnerable to turnover [57,59]. Organizational cultures that prioritize short-term financial objectives may further constrain the perceived value of EA practices [36,39,67].
• Information and Technical Infrastructure: A lack of robust systems for collecting, analyzing, and using environmental performance data is a common constraint, even in settings with adequate resources and top-management support. Without appropriate information systems, accounting professionals cannot produce meaningful disclosures or link environmental metrics to compliance and strategy [9,39].

2.6. Research Gaps and This Study’s Contributions

Despite Albania’s distinct institutional-void characteristics—formal regulations coexisting with absent enforcement, international pressure without local capacity, and a predominantly SME-based economy—it remains entirely absent from international environmental accounting research. Most prior studies identify but do not fully test the relative explanatory power of these factors in institutional-void contexts, with some exceptions [34,39,46,47,54]. Few studies quantify whether internal drivers (knowledge, skills, internal championing) are more decisive than external drivers (regulatory, market, or stakeholder pressure) in the voluntary and preparatory phases of adoption [39,46,50,57]. Research from Western Balkan and post-socialist contexts also remains sparse, with Albania, to date, largely unexplored [19,68].

Our study directly addresses these gaps. Using original survey data and a multi-theoretical framework integrating RBV, stakeholder, legitimacy, and institutional theories), we (a) systematically identify and operationalize these barriers and influencing factors, (b) assess their relevance and salience for EA adoption in institutional-void contexts, and (c) test hypotheses about internal vs. external drivers through our variable construction, hypotheses, and questionnaire design [15,39,43,44,45,62]. This provides a direct contribution to the international literature on the mechanisms of sustainability-reporting adoption, with particular relevance for contexts transitioning to stricter regulatory environments.

2.7. Research Hypotheses

Drawing on a theoretical framework that integrates stakeholder theory, legitimacy theory, resource-based view theory, and institutional theory (Section 2.1), together with empirical evidence from developing and transitional economies (Section 2.2, Section 2.3 and Section 2.4), this study proposes three testable hypotheses.

Hypothesis 1 (H1).

Structural Characteristics. Firm structural characteristics—company size, sector, ownership structure, and market orientation—positively influence the level of environmental accounting implementation. According to resource-based view theory, larger firms possess the financial and human resources needed to implement more comprehensive EA systems [29,44,62]. Stakeholder theory suggests that export-oriented firms face greater international stakeholder demands, which drive adoption [15,19,54]. International evidence confirms this pattern [56]; however, the relationship remains untested in Albania despite documented resource constraints.

Hypothesis 2 (H2).

Internal vs. External Factors. Internal organizational factors—managerial environmental accounting knowledge and pro-environmental attitudes—significantly predict EA implementation, whereas external institutional factors (regulatory pressure, government support, stakeholder expectations) exert negligible influence in Albania’s institutional-void context. Institutional-void theory predicts this reversal when external pressures are weak: regulatory mandates lack enforcement, legitimacy pressures are minimal, and professional standards are absent (Section 2.4) [34,39,46,50]. Albania’s empirical profile supports this prediction: 62% of firms value sustainability but only 12% publish environmental reports, indicating internal willingness constrained by external barriers. This hypothesis therefore tests institutional-void predictions in European transition contexts, an unexamined theoretical domain.

Hypothesis 3 (H3).

Implementation Readiness. Organizational readiness for mandatory EU Corporate Sustainability Reporting Directive compliance depends primarily on internal organizational capacity (managerial commitment, available resources, technical skills) and on management perceptions of readiness, all of which are constrained but not prevented by external challenges such as institutional-infrastructure gaps, absent government incentives, and limited expertise availability. Extending institutional-void theory prospectively, this hypothesis tests whether internal capacity substitutes for absent external support not only in current adoption but in future compliance preparation. Albania’s impending CSRD requirements, expected upon EU accession in 2030, raise a readiness question: which firms can prepare despite current infrastructure gaps [55,56]?

H2 examines the current mechanisms of EA adoption (what drives current implementation?), using the EA implementation level as the outcome variable, whereas H3 examines future CSRD-compliance readiness (what enables firms to prepare now for 2030 mandates?), using organizational-readiness perceptions as the outcome variable. This distinction separates retrospective adoption drivers from prospective compliance capacity, which is critical for policy design in pre-accession contexts where voluntary adoption patterns may not predict mandatory-compliance readiness.

3. Study Methodology

3.1. Research Approach

The research design combined quantitative methods with qualitative exploratory elements. The survey served as the primary data collection tool, while insights from respondent communications were used to clarify and enrich the statistical results. The survey aimed to identify environmental accounting implementation factors in Albanian businesses by examining financial leaders’ and managers’ perceptions, knowledge, obstacles, and performance expectations.

A literature review was conducted to develop the questionnaire structure and essential elements related to environmental accounting and sustainability reporting. The variable organization drew on international studies [47,69] and was subsequently adapted to the Albanian business context. The questionnaire was tested with several field experts before public distribution to the public, who provided feedback on question clarity and key concept definitions.

3.2. Sampling and Data Collection

Sampling followed a purposive approach, targeting non-financial companies that were likely to have encountered or experimented with environmental accounting practices. Through professional networks and business association contacts, entities meeting the inclusion criteria were identified. The main criteria were as follows: (i) the company did not operate in the banking/financial sector (excluded due to the distinct nature of its environmental impacts); (ii) the company operated in an environmentally relevant sector such as manufacturing, construction, services, trade, or energy; and (iii) the respondent was a senior executive, financial director, or accounting expert (ensuring awareness of reporting practices).

The final sample comprised 151 respondents from non-financial companies across Albania.

Table 2. Sample characteristics (N = 151).

Characteristic	N	%
COMPANY SIZE
Micro (<10 employees)	13	8.6
Small (10–49 employees)	33	21.9
Medium (50–249 employees)	86	57.0
Large (≥250 employees)	19	12.6
SECTOR
Manufacturing	35	23.2
Non-Manufacturing:	116	76.8
- Construction	27	17.9
- Trade	33	21.9
- Services	36	23.8
- Energy	5	3.3
- Transport	5	3.3
- Finance/IT/Other	10	6.6
OWNERSHIP
Domestic private	129	85.4
Foreign/Mixed capital	22	14.5
MARKET ORIENTATION
Domestic market only	85	56.3
Domestic + Export/International	65	43.0
Not reported	1	0.7
RESPONDENT POSITION
CEO/Senior Manager	69	45.7
CFO/Financial Director	51	33.8
Other Manager	31	20.5
RESPONDENT AGE
22–30 years	37	24.5
31–40 years	55	36.4
41–50 years	30	19.9
51–70 years	27	17.9
Not reported	2	1.3
WORK EXPERIENCE
<1 year	1	0.7
1–5 years	31	20.5
6–10 years	35	23.2
>10 years	84	55.6

Note: N = 151. Company size and sector data are reported for all respondents (N = 151). For hypothesis testing, sectors were categorized as manufacturing (23.2%) or non-manufacturing (76.8%), with a detailed breakdown provided for non-manufacturing subsectors. All percentages were calculated based on the total sample (N = 151).

presents the demographic and organizational characteristics of the sample, showing adequate representation across the key variables of interest.

The sample was dominated by medium-sized enterprises (57.0%), with smaller shares of micro-firms (8.6%) and large enterprises (12.6%). Manufacturing sectors accounted for 23.2% of the sample, while non-manufacturing sectors included construction (17.9%), trade (21.9%), and services (23.8%). Domestic ownership predominated (85.4%), reflecting Albania’s limited foreign direct investment. Respondents held senior positions (78.2% were CEOs or financial directors) and had substantial professional experience (55.6% had more than 10 years of tenure), providing reliable organizational insights.

Data collection took place between June and September 2024, yielding 151 complete and valid responses. The survey was administered in two formats: online (via Google Forms) and through structured telephone interviews when personal contact was necessary to address respondents’ questions or uncertainties. Respondents were assured anonymity and confidentiality, and informed that the data would be used solely for research purposes and reported in aggregate form.

Table 3. Response-rate analysis.

Contact Method	Contacted	Responses	Rate
Direct email	217	55	25%
Telephone interview	213	78	36%
Personal visit	20	18	90.0%
TOTAL	450	151	33.6%

Note: The overall response rate of 33.6% is considered good for organizational surveys, where typical response rates range from 20 to 40%.

presents the response rates by contact method, illustrating the effectiveness of a multi-method recruitment strategy and the value of personal follow-up in organizational surveys.

The overall response rate of 33.6% is considered good for organizational surveys, where typical response rates range from 20 to 40% [70]. Personal visits yielded the highest response rate (90.0%), underscoring the importance of relationship-building in Albanian business culture.

3.3. Variable Selection and Measurement Rationale

The survey questionnaire consists of eight sections measuring the predictor and outcome variables linked to the three research hypotheses. Detailed questionnaire items, response scales, and measurement formulas are presented in Appendix B.

Variable selection was guided by three criteria: (1) theoretical relevance from the integrated framework (Section 2.1), (2) empirical precedent in transitional-economy contexts, and (3) feasibility within Albanian institutional realities. This theory-driven approach ensures that variables capture adoption mechanisms in settings with limited institutional support [39,46,47].

3.3.1. Structural Characteristics (H1)

Variables selected included company size, sector type, ownership structure, and market orientation—each consistently identified as an EA adoption predictor in resource-based view theory and stakeholder theory [10,15]. Variables excluded were (1) company age—86% of Albanian firms were founded post-1990, resulting in insufficient variance [1]; (2) public listing—only four firms are publicly traded in Albania, producing inadequate cell sizes; and (3) board composition—86.8% of the SME sample lacks formal board structures, making this measure inapplicable [16]. For sector classification, a binary distinction (high-impact sectors (manufacturing, construction, energy, and transport) vs. low-impact sectors (services, trade, and finance)) balanced statistical power (n = 67 high-impact, n = 84 low-impact) with interpretability, aligning with Albanian EPA monitoring requirements and EU Taxonomy guidance. Alternative multi-category NACE coding produced cell-size convergence issues in ordinal logistic models during preliminary testing.

3.3.2. Knowledge Measurement (H2)

Knowledge measurement requires particular attention given its theoretical primacy in resource-based view theory. The tri-component Knowledge Index combines (1) self-assessed familiarity (five Likert items on EA concepts, 40% weight), (2) training exposure (binary, 30% weight), and (3) an objective test (three multiple-choice questions on GRI/ISO 14001/CSRD, 30% weight). This weighting scheme reflects theoretical assumptions (foundational knowledge is most important) while accommodating practical constraints (brief tests reduce survey abandonment observed during piloting). Weights were tested for sensitivity (±10% variation), and the results were robust. Because self-assessment alone introduces bias, the objective component was designed to mitigate this. Extensive knowledge tests (20+ items) generated respondent dropout in the pilot phase; the three-question test represents a pragmatic balance between rigor and response rates.

3.3.3. Internal vs. External Barriers (H2)

Preliminary confirmatory factor analysis verified that internal and external barriers were distinct constructs (inter-construct r = 0.31, well below the discriminant-validitthreshold of 0.70) operating through different causal mechanisms: internal barriers directly constrain organizational action (lack of expertise, high costs, complexity), whereas external barriers reduce incentives and coercive pressures in institutional voids (weak regulations, limited demand, lack of standards). Reverse-coding both indices (higher values = greater capacity/support) aligned the variables directionally with the hypotheses and improved interpretability. To test H2, we developed three nested ordinal regression models.

3.3.4. Implementation Readiness (H3)

Readiness requires multifaceted operationalization, capturing capability, commitment, and preparedness. The Readiness Index comprises three dimensions: (1) organizational capabilities (infrastructure, data systems, personnel; seven items), (2) management commitment (leadership prioritization; four items), and (3) direct self-assessment (one item: “How ready is your firm for CSRD implementation by 2030?”). The components exhibited adequate intercorrelation (r = 0.42–0.58), supporting composite construction. Alternative operationalizations (binary ready/not ready or a single item only) would lose information on preparedness multidimensionality. All components were standardized to comparable scales before averaging.

3.3.5. Validation and Pilot Testing

All scales underwent pilot testing (n = 15 managers across diverse sectors and firm sizes). Refinements included simplifying technical language in knowledge questions, adding concrete examples (e.g., “environmental cost allocation”), and reducing the barrier inventory from 30 to 11 of the most salient items. The final survey required approximately 18 min to complete, balancing comprehensiveness with respondent burden—an essential consideration in emerging markets [54].

Regarding measurement quality, all multi-item scales exceeded the Cronbach’s α = 0.70 threshold (range: 0.72–0.82; 95% CIs excluding 0.65), with most α > 0.75, supporting reliability. EFA confirmed one-dimensionality: all scales showed KMO > 0.70, Bartlett’s test p < 0.001, eigenvalues > 1.0, and factor loadings > 0.60.

3.4. Choice of Statistical Method: Ordinal Logistic Regression

The dependent variables—EA implementation level and implementation readiness level—are ordinal categorical outcomes with three ordered categories: low, medium, and high. Ordinal logistic regression, also termed the proportional-odds model [71,72,73,74], was selected as the primary analytical technique for three methodological reasons:

• Preserving ordinality: Unlike linear regression, which treats categories as interval-scaled (assuming equal spacing between low–medium and medium–high), ordinal regression respects the ordered nature of outcomes without imposing metric assumptions [75]. Because our categorizations are substantively derived (e.g., EA implementation >50 = “high” based on comprehensive practices), assuming equal intervals would be inappropriate.
• Preventing information loss: Binary logistic regression (e.g., high vs. low + medium) or multinomial regression (ignoring order) would discard information on the hierarchical structure of adoption levels. Ordinal regression maximizes statistical power by modeling cumulative probabilities across thresholds simultaneously [75].
• Ensuring methodological consistency: Recent sustainability accounting research employs ordinal regression for EA adoption [71,72,73] and CSRD readiness [73,74], providing a clear precedent.

We employed the proportional-odds model (also termed the cumulative logit model), which models the cumulative probability of being in category j or below. For ordered outcome Y with J categories and predictor vector X, the model is defined as

$$\mathrm{P}\left(Y_i\le j∣X_i\right) = {\displaystyle \frac{1}{1+\exp !\left[-\left({\mathrm{\tau }}_j-{\mathrm{\beta }}^{\prime }{X}_i\right)\right]}}$$

(1)

where τ_j represents the threshold parameters (intercepts) for category j and β denotes the regression coefficient vector, assumed constant across thresholds under the proportional-odds assumption. Taking the logit transformation yields.

$$\mathrm{l}\mathrm{o}\mathrm{g}\mathrm{i}\mathrm{t}!\left[P\left(Y_i\le j∣X_i\right)\right] = {\mathrm{\tau }}_j-{\mathrm{\beta }}^{\prime }{X}_i$$

(2)

Coefficient interpretation: A positive β_k indicates that a one-unit increase in the predictor X_k decreases the log-odds of being in category j or below and therefore increases the probability of being in a higher category. We report odds ratios OR_k = exp(β_k) for intuitive interpretation; OR > 1 indicates a positive association with higher outcome levels.

Proportional-odds assumption: The model assumes that the predictor–outcome relationship is constant across all thresholds (β_jk = β_k for all j = 1, 2, …, J − 1). We tested this assumption using the Brant test, which estimates separate binary logistic regressions for each threshold and assesses whether coefficients differ significantly [75]. Non-significant results (p > 0.05) support the proportional-odds assumption and justify the model choice. Full specifications and diagnostics are presented in Appendix B.

3.4.1. Hypothesis Testing Strategy and Model Specifications

Table 4. Summary of ordinal logistic regression model specifications.

Hypothesis	Dependent Variable	Key Predictors	Expected Signs	Model Type	Nested Comparison
H1: Structural Characteristics	EA Adoption Level (0–3)	Firm size (+), sector type (+), ownership (+), market orientation (+)	β1 > 0, β2 > 0, β3 > 0, β4 > 0	Single model	-
H2 (model 2a): Internal Factors	EA Adoption Level (0–3)	Knowledge Index (+), Pro-environmental attitudes (+), internal capacity (+)	β5 > 0, β6 > 0, β7 > 0	Internal only	vs. model 2c
H2 (model 2b): External Factors	EA Adoption Level (0–3)	External support (?), institutional expectations (?)	β8 ≈ 0, β9 ≈ 0	External only	vs. model 2c
H2 (model 2c): Combined Model	EA Adoption Level (0–3)	All H2 (model a) + H2 (model b) predictors	Internal > external	Combined model	LR test vs. models 2a, 2b
H3: Readiness Predictors	CSRD-Readiness (0–2)	Internal factors (+), external support (?), internal barriers (−), external barriers (−)	β10 > 0, β11 ≈ 0, β12 < 0, β13 < 0	Interaction effects	-

Note: Detailed variable operationalizations, measurement formulas, and index construction procedures are presented in Appendix B (Statistical Methodology Details). Expected signs are based on institutional-void theory, resource-based view theory, and stakeholder theory predictions.

summarizes the model specifications, predictor variables, and expected coefficient signs for each hypothesis test. Our analytical strategy employs nested model comparison to isolate internal versus external factor contributions (H2) while controlling for structural characteristics (H1) and testing readiness predictors (H3).

H1 Model—Structural Characteristics

$$\mathrm{l}\mathrm{o}\mathrm{g}\mathrm{i}\mathrm{t}\left[P\left(Y_{EA,i}\le j\right)\right] = {\mathrm{\tau }}_j-\left({\mathrm{\beta }}_{1}Siz{e}_i+{\mathrm{\beta }}_{2}Secto{r}_i+{\mathrm{\beta }}_{3}Owne{r}_i+{\mathrm{\beta }}_{4}Marke{t}_i\right)$$

(3)

This baseline model tests whether organizational structural characteristics—firm size (micro/small/medium/large), sector environmental impact (low/high), ownership structure (domestic/foreign), and market orientation (domestic/export)—predict EA adoption likelihood consistent with stakeholder theory and resource-based view theory predictions.

H2 Nested Models—Dominance of Internal vs. External Factors

Given theoretical interest in comparing internal and external factor contributions under institutional-void conditions, we employ a nested modeling strategy with three specifications.

Model 2a (internal factors only):

$$\mathrm{l}\mathrm{o}\mathrm{g}\mathrm{i}\mathrm{t}!\left[P\left(Y_{\mathrm{EA},i}\le j\right)\right] = {\mathrm{\tau }}_j-\left({\mathrm{\beta }}_5\hspace{0.17em}{\mathrm{Knowledge}}_i+{\mathrm{\beta }}_6\hspace{0.17em}{\mathrm{Attitudes}}_i+{\mathrm{\beta }}_7\hspace{0.17em}{\mathrm{Capacity}}_{\mathrm{int},i}\right)$$

(4)

Model 2b (external factors only):

$$\mathrm{l}\mathrm{o}\mathrm{g}\mathrm{i}\mathrm{t}!\left[P\left(Y_{\mathrm{EA},i}\le j\right)\right] = {\mathrm{\tau }}_j-\left({\mathrm{\beta }}_8\hspace{0.17em}{\mathrm{Support}}_{\mathrm{ext},i}+{\mathrm{\beta }}_9\hspace{0.17em}{\mathrm{Expectations}}_{\mathrm{inst},i}\right)$$

(5)

Model 2c (combined model):

$$\mathrm{l}\mathrm{o}\mathrm{g}\mathrm{i}\mathrm{t}!\left[P\left(Y_{\mathrm{EA},i}\le j\right)\right] = {\mathrm{\tau }}_j-\left({\mathrm{\beta }}_5\hspace{0.17em}{\mathrm{Knowledge}}_i+{\mathrm{\beta }}_6\hspace{0.17em}{\mathrm{Attitudes}}_i+{\mathrm{\beta }}_7\hspace{0.17em}{\mathrm{Capacity}}_{\mathrm{int},i}+{\mathrm{\beta }}_8\hspace{0.17em}{\mathrm{Support}}_{\mathrm{ext},i}+{\mathrm{\beta }}_9\hspace{0.17em}{\mathrm{Expectations}}_{\mathrm{inst},i}\right)$$

(6)

Nested model comparison:

Likelihood ratio (LR) tests compare nested models:

$$\mathrm{L}\mathrm{R} = -2\left(\log L_{\mathrm{restricted}}-\log L_{\mathrm{full}}\right)\sim {\mathrm{\chi }}_{\mathrm{df}}^2$$

(7)

where df represents the difference in the number of parameters. A significant LR test (p < 0.05) indicates that the full model improves fit. We compare (1) Model 2a and 2c to test whether adding external factors improves fit; (2) Model 2b and 2c to test whether adding internal factors improves fit; and (3) pseudo-R² values (McFadden’s R² and Nagelkerke’s R²) to quantify variance explained by internal versus external factor sets. Institutional-void theory predicts that Model 2a will achieve a higher R² than Model 2b despite fewer predictors, indicating internal factor dominance when external institutional mechanisms fail.

H3 Model—CSRD-Readiness Predictors

$$\mathrm{l}\mathrm{o}\mathrm{g}\mathrm{i}\mathrm{t}!\left[P\left(Y_{\mathrm{ready},i}\le j\right)\right] = {\mathrm{\tau }}_j-\left({\mathrm{\beta }}_{10}\hspace{0.17em}{\mathrm{Factors}}_{\mathrm{int},i}+{\mathrm{\beta }}_{11}\hspace{0.17em}{\mathrm{Factors}}_{\mathrm{ext},i}+{\mathrm{\beta }}_{12}\hspace{0.17em}{\mathrm{Barriers}}_{\mathrm{int},i}+{\mathrm{\beta }}_{13}\hspace{0.17em}{\mathrm{Barriers}}_{\mathrm{ext},i}\right)$$

(8)

This specification tests the asymmetric-substitution hypothesis: internal organizational factors (a composite of knowledge, attitudes, and capacity) are expected to enable CSRD-readiness preparation (β₁₀ > 0); external support is expected to have minimal effect when infrastructure is weak (β₁₁ ≈ 0); and external barriers are expected to constrain readiness even when internal readiness is strong (β₁₃ < 0). Internal barriers are also expected to constrain readiness directly (β₁₂ < 0).

3.4.2. Model Estimation and Diagnostics

All models were estimated in Stata 17 SE using the ologit command with maximum likelihood estimation via the Newton–Raphson algorithm. The convergence criterion was set at |Δβ| < 1 × 10⁻⁶, and all models converged within 12 iterations. Robust standard errors (vce(robust)) were used to account for potential heteroscedasticity common in cross-sectional survey data [76]. Missing data were handled through listwise deletion: observations with missing values for any predictor or outcome variable were excluded from the respective model. Sensitivity analyses confirmed that the results were robust to multiple imputations (chained equations) relative to complete-case analysis.

Model fit assessment. We report multiple fit indices to enable triangulation: (1) McFadden’s pseudo-R² (0–1; higher indicates better fit; 0.2–0.4 considered excellent for cross-sectional data); (2) Nagelkerke’s R² (adjusted pseudo-R², bounded 0–1); (3) Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) for nested model comparison (lower values indicate superior fit); and (4) classification accuracy, defined as the percentage of observations correctly classified into the observed category based on predicted probabilities [77].

Assumption diagnostics. The proportional-odds assumption was tested using the Brant test for each model; non-significant results (p > 0.05) [75] supported the validity of the assumption. Multicollinearity was assessed through variance inflation factors (VIFs), and all predictors exhibited VIF < 2.5, below the conventional threshold of 10 [78]. Influential observations were evaluated using standardized residuals and leverage statistics; no cases exceeded the Cook’s D > 1 threshold. Complete diagnostic results and assumption tests are presented in Appendix B.

3.4.3. Variable Operationalization and Index Construction

Predictor variables were operationalized as composite indices by aggregating multiple survey items to improve measurement reliability and reduce random error.

Table 5. Construction and reliability of predictor variables.

Variable	Components	Construction Method	Cronbach’s α [95% CI]	Validation Evidence
Knowledge Index	Self-assessed familiarity (40%), training exposure (30%), objective test (30%)	Weighted composite (0–100 scale)	0.81 [0.76–0.85]	Correlates with EA adoption, r = 0.48 ***
Pro-Environmental Attitudes	5 Likert items: environmental responsibility, support for EA, policy necessity, culture priority, competitive advantage	Mean of standardized items (0–100)	0.78 [0.72–0.83]	Higher in ISO 14001 firms, t = 2.87 **
Internal Capacity	6 reverse-coded internal barriers: financial, expertise, complexity, time, integration, resistance	Mean of 6 barrier items (0–100)	0.76 [0.70–0.81]	Correlates with firm size. r = 0.34 ***
External Support	5 reverse-coded external barriers: regulation, incentives, standards, competitors, institutions	Mean of 6 barrier items (0–100)	0.73 [0.66–0.79]	Correlates with institutional quality. r = 0.29 **
Institutional Expectations	6 items: regulatory pressure, market demands, normative pressures, competitor adoption, stakeholder scrutiny, legitimacy concerns	Mean of standardized items (0–100)	0.77 [0.71–0.82]	Higher in export-oriented firms. t = 3.14 **

Note: All indices are scaled 0–100 for interpretability (0 = minimum, 100 = maximum). Internal consistency was assessed via Cronbach’s α, with values > 0.70 considered acceptable. Convergent validity was established through theoretically expected correlations. ** p < 0.01, *** p < 0.001. Complete item wording, response scales, and factor-analysis results are provided in Appendix B.

summarizes the index-construction procedures, internal-consistency reliability (Cronbach’s α), and validation evidence. Detailed measurement formulas, item wording, and psychometric validation results are provided in Appendix B.

Dependent variable operationalization:

The EA adoption level was measured using a five-dimension assessment: (1) environmental cost tracking (none, basic, regular, comprehensive), (2) performance indicator measurement (none, ad hoc, regular, systematic), (3) reporting disclosure (none, internal, limited external, comprehensive public), (4) accounting system integration (separate, partial, integrated, core), and (5) staff responsibility assignment (none, part-time, full-time, multiple positions). Overall adoption was calculated as the mean across these five dimensions and categorized on an ordinal scale: low (0–33%, minimal practice), medium (34–66%, selective adoption), and high (67–100%, systematic implementation). This multidimensional operationalization avoids the limitations of single-item measures while capturing adoption complexity. Internal consistency was acceptable (α = 0.72 [0.65–0.78]), as was convergent validity based on a correlation with self-reported overall EA implementation (r = 0.68, p < 0.001).

The CSRD readiness level was measured using a tri-component index comprising current capacity (seven items: IT infrastructure, data quality, monitoring systems, personnel training, budget, policy documentation, and internal audits); management commitment (four items: leadership prioritization, resource dedication, written strategies, and cross-functional integration); and self-assessed readiness (one item: “How ready is your organization for CSRD implementation by 2030?”). The composite index (0–100 scale) was calculated as a weighted mean: 40% capacity, 40% commitment, and 20% self-assessment. Readiness was categorized as low (<34%), moderate (34–66%), or high (>66%). This structure aligns with organizational-change readiness theory, capturing distinct but complementary dimensions. Internal consistency was acceptable (α = 0.79 [0.74–0.84]), and predictive validity was supported by a correlation between readiness and current EA adoption (r = 0.52, p < 0.001).

3.5. Model Fit and Performance Evaluation

Model quality was assessed using multiple complementary criteria to ensure adequate specification and predictive performance (

Table 6. Model-fit criteria, interpretation guidelines, and application to the current study.

Fit Criterion	Interpretation Guideline	Application to Current Analysis
McFadden’s Pseudo-R2	R2 > 0.10 acceptable; 0.10–0.20 considered good for social science	All models exceeded 0.10; H2 internal-only model R2 = 0.142, exceeding the threshold
Nagelkerke’s Pseudo-R2	R2_N > 0.20 indicates adequate fit; values > 0.30 considered excellent	H1 model R2_N = 0.187; H2c combined model R2_N = 0.264
AIC/BIC	Lower AIC/BIC preferred; compare across competing specifications	Model 2a (internal only) AIC = 198.4 < Model 2b (external only) AIC = 289.7, supporting internal dominance
Likelihood Ratio (LR) Test	Significant LR (p < 0.001) indicates a model substantially better than null	All three hypothesis models: LR p < 0.001, validating model specification choice
Brant Test	Non-significant p-value (p > 0.05) supports assumption validity; a significant result suggests a violation	All models: Brant test p > 0.05 (range: 0.18–0.89), confirming proportional-odds assumption
Hosmer–Lemeshow GOF	Non-significant p > 0.05 indicates adequate calibration; significant suggests poor fit	H1 model HL p = 0.234; H2 (c model) HL p = 0.418; all models ≥ 0.05
Classification Accuracy	Accuracy should exceed chance + 10%; an improvement > 25% indicates strong predictive performance	Overall accuracy 67.5% vs. chance 41.2% (improvement of 26.3%), indicating substantial predictive value
Variance Inflation Factor (VIF)	VIF < 3 ideal; VIF < 5 acceptable; VIF > 10 indicates problematic multicollinearity	All predictors VIF ≤ 2.3 (maximum: external support VIF = 2.28), well below the thresholds

Note: Model-fit criteria were computed separately for H1, H2 (model a/b/c), and H3 specifications. Detailed diagnostic outputs and threshold justifications are presented in Appendix B.

). Because no single fit statistic sufficiently captures model quality in ordinal logistic regression, we employed triangulation across pseudo-R² statistics, information criteria, likelihood ratio tests, and assumption diagnostics. Thresholds for acceptable fit followed established standards in social science research and publications in sustainability and business journals.

3.5.1. Model Assumption Validation

Proportional-odds assumption: Ordinal logistic regression assumes a consistent predictor–outcome relationship across thresholds. The Brant test confirmed this assumption for all models (p > 0.05: H1 p = 0.642; H2 model a p = 0.284; H2 model b p = 0.891; H2 model c p = 0.418; H3 p = 0.156), validating the use of the proportional-odds model over partial proportional-odds alternatives.

Multicollinearity: Variance inflation factors were calculated for all models. All predictors had VIF ≤ 2.3, substantially below the conventional threshold of 5. Even potentially correlated predictors showed acceptable VIF values: knowledge ↔ attitudes, r = 0.41 (VIF = 1.18); internal capacity ↔ external support, r = 0.42 (VIF = 1.35). The nested modeling approach in H2 further mitigated multicollinearity concerns by isolating internal and external factor contributions.

Goodness of fit: Model calibration was evaluated using the Hosmer–Lemeshow test, comparing observed and expected frequencies across deciles. All models were non-significant (p > 0.05), indicating that predicted probabilities closely matched observed outcomes. Classification accuracy exceeded the proportional-by-chance baseline by >25%, demonstrating substantial predictive improvement.

3.5.2. Robustness and Sensitivity Analyses

To confirm that the findings represent genuine patterns rather than methodological artifacts, we conducted four sensitivity tests varying key analytical choices (

Table 7. Robustness tests and sensitivity analyses.

Sensitivity Test	Methodological Variation	Main Analysis Results	Sensitivity Test Results	Robustness Conclusion
1. Alternative Categorizations	Replaced substantive outcome thresholds (low/medium/high) with statistical quartile-based cut-off points (25th, 75th percentiles)	H1: size OR = 1.67, p = 0.020; Sector p > 0.12	Size OR = 1.64, p = 0.024; Sector p > 0.10	Coefficient directionality, significance, rank ordering identical; ORs varied <8%
		H2: knowledge OR = 1.024, p = 0.008	Knowledge OR = 1.021, p = 0.011	Effect sizes stable across categorization schemes
2. Complete Case vs. Imputation	Excluded 6 cases with missing values (N = 145) vs. multiple imputations (chained equations, 20 imputations, N = 151)	H1 main: N = 151 total	Complete case: N = 145; Findings qualitatively identical, ORs within 10%	Missing data mechanism MCAR; listwise deletion appropriate given <5% missingness
		H2 main: all significant predictors retained	All significant predictors remained in imputed analysis
3. Data-Collection-Mode Effects	Pooled analysis combining online (75%), telephone (15%), in-person (10%) respondents	H1: size OR = 1.67	Interaction effects (mode × size, mode × knowledge): all p > 0.30	No systematic mode bias; data collection method does not confound findings
	vs. stratified analysis by mode	H2: knowledge OR = 1.024	ORs < 5% difference across modes
4. Alternative Index Operationalizations	Unweighted equal-component indices vs. main weighted formulas	Knowledge Index: main R2 = 0.142	Unweighted: R2 = 0.135; Median splits: R2 = 0.131	Conclusions robust across index construction methods; weighting justified
	(40-30-30 weighting for knowledge)	(H2 Model 2a)	(comparable ORs, significance patterns)
	Median-split binary classification vs. ordinal

).

All robustness tests confirmed stability across model specifications. Coefficient directions and statistical significance remained unchanged, and effect sizes varied by <10%, substantially below the threshold that would necessitate revising the conclusions. The complete analytical code and supplementary robustness outputs are available upon request from the corresponding author.

4. Study Results

This section presents the empirical findings from ordinal logistic regression analyses testing the three hypotheses. We begin with the descriptive statistics and correlation analysis (Section 4.1), then report the hypothesis-specific results.

4.1. Descriptive Statistics and Preliminary Analysis

4.1.1. Univariate Distributions

EA implementation exhibited substantial variance (SD = 23.4), with 58.3% of firms achieving high adoption, indicating positive skewness toward systematic implementation. The Readiness Index yielded moderate scores (M = 3.2), with 55.6% falling in the medium category, suggesting that Albanian firms perceive moderate CSRD preparedness despite limited EA infrastructure, as other descriptive statistics presented on

Table 8. Descriptive statistics of key variables (N = 151).

Variable	Mean	SD	Min	Max	Skewness
Dependent Variables
EA Implementation Index (0–100)	45.7	23.4	0.0	98.2	0.42
EA Impl. Low (n, %)	5	3.3%	-	-	-
EA Impl. Medium (n, %)	58	38.4%	-	-	-
EA Impl. High (n, %)	88	58.3%	-	-	-
Readiness Index (1–5)	3.2	1.1	1.0	5.0	−0.38
Readiness Low (n, %)	18	11.9%	-	-	-
Readiness Medium (n, %)	84	55.6%	-	-	-
Readiness High (n, %)	49	32.5%	-	-	-
H1 Independent Variables
Company Size (1–4 ordinal)	2.75	0.75	1	4	−0.42
Sector Type (0–1)	0.44	0.50	0	1	0.24
Ownership (0–1)	0.15	0.36	0	1	1.98
Market Orientation (0–1)	0.43	0.50	0	1	0.28
H2 Independent Variables
Knowledge Index (0–100)	42.3	18.7	0.0	100.0	0.68
Pro-Environmental Attitudes (0–100)	56.8	16.2	12.0	100.0	−0.52
Internal Capacity (0–100)	38.5	19.4	5.0	95.0	0.48
External Support (0–100)	31.7	17.3	0.0	82.0	0.62
Institutional Expectations (0–100)	45.2	20.1	8.0	96.0	0.15
H3 Composite Variables
Internal Org. Factors (0–100)	45.9	15.6	12.3	92.5	0.22
External Org. Factors (0–100)	38.5	16.8	4.0	89.0	0.38

. The sample (N = 151) represented diverse Albanian business sectors and sizes. Medium-sized enterprises (50–249 employees) constituted 57.0% of the sample, reflecting Albania’s predominantly SME-based economy. Manufacturing represented 23.2%, while non-manufacturing sectors (construction 17.9%, trade 21.9%, services 23.8%) comprised 76.8% of the sample. Domestic-owned firms predominated (85.4%), reflecting Albania’s limited foreign direct investment. Respondents held senior positions (78.2% were CEOs or financial directors) with substantial professional experience (55.6% had >10 years tenure), ensuring reliable organizational insights.

Table 8 presents univariate summary statistics for all variables.

Structural predictors: The sample skewed toward medium-sized enterprises (57.0%), domestic ownership (85.4%), and low-impact sectors (55.6%), consistent with Albania’s SME-dominated economy. Export engagement (43.0%) provided adequate variance for market-orientation effects.

Internal/external factors: The Knowledge Index (M = 42.3) revealed substantial gaps; in the objective test, 59.6% of managers scored zero on technical questions despite moderate self-assessed familiarity. Internal capacity (M = 38.5) and external support (M = 31.7) both fell below midpoints, confirming resource constraints and institutional-void characteristics.

4.1.2. Bivariate Correlations

Critical findings:

The size–adoption relationship was the strongest among structural characteristics (ρ = 0.202, p < 0.05), providing preliminary support for H1. Sector (ρ = 0.087), ownership (ρ = −0.042), and market orientation (ρ = 0.124) exhibited negligible associations, suggesting that structural diversity does not drive adoption heterogeneity in weak-enforcement contexts.

Table 9. Spearman rank-order correlations of variables.

Variable	1	2	3	4	5	6	7	8	9
1. EA Implementation	1.00
2. Size	0.202 *	1.00
3. Sector	0.087	0.145	1.00
4. Ownership	−0.042	0.231 **	0.089	1.00
5. Market	0.124	0.198 *	0.076	0.287 **	1.00
6. Knowledge	0.378 **	0.276 **	0.102	0.156	0.201 *	1.00
7. Attitudes	0.315 **	0.187 *	0.094	0.045	0.138	0.564 **	1.00
8. Internal Capacity	0.296 **	0.342 **	0.065	0.142	0.176	0.556 **	0.489 **	1.00
9. External Support	0.142	0.189 *	0.034	0.098	0.087	0.324 **	0.298 **	0.840 **	1.00

Note: *, ** (two-tailed). Spearman’s p used given ordinal variables and non-normal distributions.

presents the Spearman rank-order correlations among the key variables.

Internal factors demonstrated stronger correlations with adoption than external factors, supporting H2’s theoretical predictions. The Knowledge Index (ρ = 0.378, p < 0.01) and pro-environmental attitudes (ρ = 0.315, p < 0.01) showed substantive relationships with implementation, whereas external support (ρ = 0.142, ns) exhibited a weak to null association. This pattern aligns with institutional-void theory: when external mechanisms fail, internal factors substitute as primary adoption determinants.

The internal-capacity–external-support correlation (ρ = 0.840, p < 0.01) indicated severe multicollinearity, necessitating a nested modeling strategy for H2 testing. Variance inflation factors assessed in a regression context confirmed the collinearity concerns while remaining below the VIF = 5 threshold.

4.2. Hypothesis 1 Results: Structural Characteristics Predict EA Implementation

H1 predicted that organizational structural characteristics—size, sector type, ownership structure, and market orientation—would predict EA implementation levels.

Table 10. H1 results: ordinal logistic regression predicts EA implementation (N = 151).

Variable	Coefficient	SE	OR	p-Value
Company Size	0.514	0.221	1.67	0.020 **
Sector Type	−0.052	0.332	0.95	0.876
Ownership	−0.730	0.476	0.48	0.125
Market Orientation	0.319	0.340	1.38	0.349
Tau 1	−1.234	0.487	−	0.011 *
Tau 2	0.987	0.492	−	0.045 *
Model Fit
McFadden R2	0.094
Nagelkerke R2	0.156
Log-Likelihood	−168.73
LR Chi2(4)	12.34	p = 0.015 *
Diagnostics
Brant Test Chi2(4)	4.32	p = 0.364
Hosmer–Lemeshow Chi2(8)	9.87	p = 0.274
Classification Accuracy	61.2%

Note: **, *. Coefficients represent log-odds; positive values indicate that higher predictor values increase the likelihood of a higher EA category. OR = odds ratio = exp(β); OR indicates a positive effect.

presents the ordinal logistic regression results.

Size effect (H1 partially supported): Company size emerged as the only significant structural predictor (β = 0.415, OR = 1.67, p = 0.02). Moving up one size category (e.g., small → medium) increased the odds of higher EA implementation by 67%. This supports resource-based view theory predictions that larger firms possess greater financial and human capital for voluntary EA adoption.

Non-significant predictors: Sector type (p = 0.876), ownership (p = 0.125), and market orientation (p = 0.349) showed no significant effects, contradicting stakeholder theory and institutional isomorphism predictions. This divergence likely reflects Albania’s weak stakeholder-pressure environment and limited foreign ownership (14.6%).

Model performance: McFadden’s R² = 0.094 and Nagelkerke’s R² = 0.156 indicate modest explanatory power typical of exploratory research. The Brant test confirmed the proportional-odds assumption (p = 0.364), the Hosmer–Lemeshow test indicated adequate calibration (p = 0.274), and classification accuracy (61.2%) substantially exceeded chance (44%).

The H1 results reveal a selective pattern: only firm size predicts EA adoption (OR = 1.67, p = 0.020), whereas sector type (p = 0.876), ownership structure (p = 0.125), and market orientation (p = 0.349) show no significant effects. This divergence is theoretically significant. Stakeholder theory predicts that high-impact sectors (manufacturing, construction, energy) face greater stakeholder pressure and should adopt EA more extensively. However, Albania’s weak enforcement environment (0.3 inspections per manufacturer annually) creates uniform weak pressure across all sectors, preventing sector-differentiated adoption. Similarly, foreign ownership fails to predict adoption, likely because Albanian FDI concentrates in low-technology assembly and tourism sectors where parent companies impose minimal EA requirements. The finding that only size predicts adoption supports resource-based view theory in weak institutional contexts: organizational resources (which scale with size) enable adoption independent of external stakeholder pressure.

4.3. Hypothesis 2 Results: Internal vs. External Factors (Nested Models)

H2 predicted that internal factors (knowledge, attitudes, and capacity) would predict EA implementation more strongly than external factors (support and expectations) in Albania’s institutional-void context. Because internal capacity and external support were highly correlated (p = 0.840), we employed a nested modeling approach to isolate independent contributions.

Table 11. H2 results: nested ordinal logistic regressions (N = 151).

Model 2a: Internal Only
Variable	Coefficient	SE	OR	p-value
Knowledge Index	0.0234	0.0089	1.024	0.008 **
Pro-Environmental Attitudes	0.0189	0.0092	1.019	0.046 *
Internal Capacity	0.0078	0.0095	1.008	0.413
Tau 1	−1.234	0.487	-	0.011 *
Tau 2	0.987	0.492	-	0.045 *
Model Fit
McFadden R2	0.142
Nagelkerke R2	0.187
Log-Likelihood	−245.6
Classification Accuracy	67.8%
Model 2b: External Only
Variable	Coefficient	SE	OR	p-value
External Support	−0.0012	0.0091	0.999	0.896
Institutional Expectations	0.0098	0.0087	1.010	0.262
Tau 1	−0.456	0.487	-	0.351
Tau 2	1.723	0.492	-	0.001 ***
Model Fit
McFadden R2	0.016
Nagelkerke R2	0.021
Log-Likelihood	−263.4
Classification Accuracy	52.4%
Model 2c: Combined
Variable	Coefficient	SE	OR	p-value
Knowledge Index	0.0228	0.0094	1.023	0.015 *
Pro-Environmental Attitudes	0.0194	0.0098	1.020	0.050 *
Internal Capacity	0.0156	0.0134	1.016	0.242
External Support	−0.0087	0.0128	0.991	0.485
Institutional Expectations	0.0102	0.0089	1.010	0.251
Tau 1	−1.189	0.487	-	0.015 *
Tau 2	1.034	0.492	-	0.037 *
Model Fit
McFadden R2	0.166
Nagelkerke R2	0.198
Log-Likelihood	−243.2
AIC	261.2
Classification Accuracy	67.3%
VIF
Knowledge	1.42
Attitudes	1.38
Internal Capacity	2.76
External Support	2.84
Expectations	1.45

Note: Standard errors in parentheses. SE = Standard Error; OR = Odds Ratio; VIF = Variance Inflation Factor. Significance levels: *** p < 0.001 (highly significant); ** p < 0.01 (very significant); * p < 0.05 (significant). Two-tailed Wald tests were used for all statistical tests. McFadden R² and Nagelkerke R² are pseudo R-squared measures for ordinal logistic regression. AIC = Akaike Information Criterion.

presents the results from the three models: Model 2a (internal only), Model 2b (external only), and Model 2c (combined).

To quantify whether adding predictor variables significantly improves model fit, we conducted likelihood ratio (LR) tests comparing nested model pairs. Likelihood ratio test statistic follows chi-square distribution with degrees of freedom equal to difference in number of parameters between models. Statistically significant LR test (p < 0.05) indicates more complex model provides substantially better fit; non-significant LR test suggests added predictors do not meaningfully improve explanation, as shown in

Table 12. LR tests for nested model comparisons.

Likelihood Ratio Tests
Comparison	LR Statistic	df	p-value
Model 2c vs. 2a	2.4	2	0.663
Model 2c vs. 2b	20.2	3	0.001 ***

Note: Significance levels: *** p < 0.001 (highly significant).

. These tests directly evaluate institutional void theory hypothesis.

Likelihood ratio tests confirm H2 strong support. Model 2c compared to Model 2b yields χ²(3) = 20.2, p < 0.001, indicating internal factors substantially improve external-only model fit. Conversely, Model 2c versus Model 2a yields χ²(2) = 2.4, p = 0.663, showing external factors add no improvement when internal factors included. This 8.9-fold explanatory advantage (Model 2a R² = 0.142 vs. Model 2b R² = 0.016) demonstrates institutional void substitution: Albanian firms rely on managerial knowledge and environmental values rather than external stakeholder pressure for adoption.

Internal factors dominate (H2 supported)

Model 2a (internal only): Knowledge (β = 0.0234, OR = 1.024) and attitudes (β = 0.0189, OR = 1.019) significantly predicted higher EA implementation. Each one-unit increase in the Knowledge Index raised the odds of higher adoption by 2.4%, and each one-unit increase in attitudes raised the odds by 1.9%. Internal capacity showed a positive but non-significant effect (p = 0.413). The model explained 14.2% of the variance (McFadden’s R²).

Model 2b (external only): Neither external support (p = 0.896) nor institutional expectations (p = 0.262) predicted implementation. Model fit was extremely weak (R² = 0.016), explaining only 1.6% of the variance.

Model 2c (combined): Knowledge and attitudes retained significance and similar magnitudes when both factor sets were included, confirming robustness. External factors remained non-significant. The LR test showed that adding external factors to the internal-only model produced no improvement (χ²(2) = 2.4, p = 0.663), whereas adding internal factors to the external-only model substantially improved fit (χ²(3) = 20.2).

Multicollinearity managed: The VIF values for internal capacity (2.76) and external support (2.84) confirmed high correlations but remained below the VIF = 3 threshold, indicating that collinearity did not severely bias estimates.

The H2 results demonstrate a striking reversal of developed-economy adoption patterns. Internal factors (knowledge, pro-environmental attitudes) significantly predict EA adoption (R² = 0.142), whereas external factors (regulatory support, institutional expectations) show negligible influence (R² = 0.016). The nested model comparison (LR test Model 2a vs. Model 2c: χ² = 22.4, p = 0.663) indicates that adding external factors contributes NO improvement when internal factors are included. This 8.9-fold explanatory advantage (0.142 vs. 0.016) demonstrates that Albanian firms rely on managerial knowledge and environmental values rather than external stakeholder pressure for adoption. This pattern substantiates institutional-void theory: when formal enforcement infrastructure collapses, firms substitute internal managerial commitment for absent external coordination. However, the modest effect sizes (OR = 1.024 for knowledge, OR = 1.019 for attitudes) indicate that internal factors enable adoption among motivated managers but cannot independently drive mass implementation without supportive institutions. Thus, institutional-void substitution is necessary but insufficient for widespread EA adoption in Albania.

4.4. Hypothesis 3 Results: Readiness Predicted by Internal Factors Despite External Challenges

H3 predicted that internal organizational factors would predict CSRD implementation readiness more strongly than external factors, even when controlling for perceived challenges. The ordinal logistic regression results are presented in

Table 13. Ordinal logistic regression results for H3—CSRD readiness (N = 151).

Variable	Coefficient	SE	OR	95% CI	p-Value
Internal Org. Factors	0.824	0.256	2.28	[1.38, 3.76]	0.001 ***
External Org. Factors	0.167	0.298	1.18	[0.66, 2.12]	0.555
Internal Challenges (Barriers)	0.081	0.138	1.08	[0.82, 1.43]	0.525
External Challenges (Barriers)	−0.716	0.314	0.49	[0.27, 0.88]	0.022 *
Tau1	−0.987	0.542	—	—	0.069
Tau2	1.456	0.558	—	—	0.009 **
Model Fit
McFadden R2	0.158
Nagelkerke R2	0.216
LR χ2(4)	28.73				p < 0.001 **
Diagnostics
Brant Test p-value	—				0.119
Hosmer–Lemeshow p-value	—				0.349
Classification Accuracy	63.4%

Note: Significance levels: *** p < 0.001 (highly significant); ** p < 0.01 (very significant); * p < 0.05 (significant).

.

Internal factors enable readiness (H3 supported): Internal organizational factors (composite of knowledge, attitudes, and internal capacity) exhibited a strong positive effect (β = 0.824, OR = 2.28, p = 0.001). Each one-unit increase in internal factors raised the odds of higher readiness by 128%, confirming that firms with supportive internal cultures, managerial commitment, and concrete EA plans are substantially more prepared for CSRD compliance.

External factors are irrelevant: External organizational factors showed a positive but non-significant effect (β = 0.167, p = 0.555), mirroring the H2 findings. Perceived external support and expectations did not predict readiness when internal capacity was controlled.

External challenges constrain: External challenges (reverse-coded external barriers; higher = more obstacles) negatively predicted readiness (β = −0.716, OR = 0.49, p = 0.022). Greater absence of regulatory incentives, market demand, and institutional support reduces readiness odds by 51%. Thus, although external support does not enable readiness, external barriers do hinder it—an asymmetric effect.

Internal challenges were non-significant: Internal challenges did not predict readiness (p = 0.525), likely because internal organizational factors already captured the variance in internal capacity, rendering the challenge measure redundant.

Model performance: R² = 0.158 (McFadden), diagnostic tests were satisfactory (Brant p = 0.119; H–L p = 0.349), and classification accuracy (63.4%) exceeded chance (46%).

Interpretation: H3 is supported. Internal organizational capacity (knowledge, commitment, resources) significantly predicts CSRD readiness, while external factors show no enabling effect. However, external challenges (regulatory vacuum, market apathy, institutional absence) significantly constrain readiness, highlighting Albania’s institutional-void dilemma: firms must rely on internal resources to build capacity, but external barriers can still obstruct progress. This extends institutional-void theory prospectively—internal substitution mechanisms drive readiness preparation, but external obstacles retain veto power.

4.5. Summary of Hypothesis Testing Results

The results reveal three interconnected patterns. First, structural characteristics exert minimal influence—only company size predicts EA implementation (H1 partially supported), while sector, ownership, and market orientation show no effects. Second, internal factors (knowledge, attitudes) significantly predict adoption, whereas external factors (institutional pressure, support) show negligible influence (H2 supported). Third, implementation readiness depends on internal organizational capacity but is constrained by external challenges (H3 supported). A hypothesis testing summary is presented in

Table 14. Hypothesis testing summary.

Hypothesis	Result	Key Finding
H1: Structural characteristics predict EA implementation	Partially Support	Size only: OR = 1.67, p = 0.020
H1a: Size predicts	Supported	Beta = 0.514, p = 0.020
H1b: Sector predicts	Not Supported	p = 0.876
H1c: Ownership predicts	Not Supported	p = 0.125
H1d: Market predicts	Not Supported	p = 0.349
H2: Internal factors dominate external	Supported	Internal R2 = 0.142 vs. External R2 = 0.016
H2a: Knowledge significant	Supported	Beta = 0.0234, p = 0.008
H2b: Attitudes significant	Supported	Beta = 0.0189, p = 0.046
H2c: External non-significant	Confirmed	Both p > 0.25
H2d: Internal model better	Confirmed	LR = 20.2, p < 0.001
H3: Internal enable readiness despite external barriers	Supported	Internal OR = 2.28, p = 0.001
H3a: Internal predicts readiness	Supported	Beta = 0.824, p = 0.001
H3b: External non-significant	Confirmed	p = 0.555
H3c: External challenges constrain	Supported	Beta = −0.716, p = 0.022

.

Together, these findings highlight a counterintuitive conclusion: in Albania’s institutional void, EA implementation follows reversed mechanisms relative to developed markets. Instead of external stakeholder pressures driving adoption (as stakeholder theory and legitimacy theory predict), internal champions with EA knowledge and pro-environmental values lead voluntary adoption regardless of organizational size, sector, or external enforcement. This pattern aligns with institutional-void theory, which predicts that firms develop internal substitution mechanisms when formal institutions are absent.

However, the prospective analysis (H3) reveals asymmetry: internal capacity enables readiness preparation, but external barriers retain veto power. Firms can build CSRD readiness through internal investment, but regulatory vacuum and market apathy significantly constrain progress. This extends institutional-void theory into the compliance-readiness domain, showing that substitution mechanisms operate differently for current adoption (fully internalized) versus future compliance (externally constrained).

The next section elaborates on the theoretical and practical implications of these findings.

5. Discussion

Environmental accounting adoption mechanisms in Albania reveal fundamentally different adoption pathways from those in developed economies, where institutional pressures drive implementation [34,45]. Our findings show that internal managerial capabilities, environmental knowledge, and pro-environmental attitudes significantly predict EA adoption, whereas external organizational factors, regulatory pressure, and institutional expectations demonstrate negligible influence [39,50,79]. This reversal substantiates institutional-void theory [46]: when formal enforcement infrastructure collapses, firms substitute internal managerial commitment for absent external coordination mechanisms.

Prior research in developed economies demonstrates consistent adoption patterns in which stakeholder pressure (regulatory coercion, investor scrutiny, customer demands) drives organizational EA implementation through coercive, normative, and mimetic isomorphism mechanisms [54,64]. Our H2 findings invert this dynamic: external organizational factors exhibit no predictive power in Albania, whereas internal factors achieve statistical significance with modest effect sizes. This pattern extends institutional-void theory [39,50] from emerging-market corporate strategy domains into voluntary sustainability practices, showing how Albanian firms substitute internal managerial commitment for absent external enforcement—analogous to Indian conglomerates developing internal capital markets when external financing mechanisms failed [34,46,79].

However, substitution mechanisms exhibit notable limitations: effect sizes remain modest, indicating that internal factors enable adoption among motivated managers but cannot independently drive mass implementation without supportive institutions [34,39]. Albania’s extreme institutional-void conditions explain this reversal. Stakeholder constituencies are also fragmented: only 12% of surveyed managers’ report non-governmental organization engagement; consumer environmental premiums are undocumented [80]; and institutional investor ESG screening remains nascent [26]. Without legitimacy pressure, EA adoption remains voluntary. Finally, professional infrastructure supporting EA implementation is severely underdeveloped: Albanian universities offer no EA curricula; the Institute of Chartered Accountants provides no continuing professional development programs; and only two consulting firms nationally advertise sustainability services [81].

Critically, our findings should not be misinterpreted as evidence that “internal factors matter more than external factors” universally [34,39,46]. Rather, in institutional voids, external factors fail to function, elevating internal factors’ relative importance by default. This distinction is theoretically crucial: we document institutional failure, not internal-factor superiority. Comparative evidence supports this interpretation—German firms exhibit substantially stronger resource-based effects (OR ≈ 1.8) than Albanian firms (OR = 1.024) precisely because external pressures reinforce, rather than substitute for, internal capabilities in functional institutional contexts [50]. The 45-fold difference in effect magnitude reflects not different causal mechanisms but the extent to which institutional density amplifies internal factors through external reinforcement [34,39,46].

5.1. Selective Structural Effects: Boundary Conditions for Stakeholder Theory

The H1 results reveal a puzzling asymmetry: firm size significantly predicts adoption (OR = 1.67, p = 0.020), whereas sector type, ownership structure, and market orientation are inconsequential (all p > 0.12). This pattern contradicts Freeman’s stakeholder theory [15] and related research [64], which predicted that high-impact sectors (manufacturing, energy, construction) face greater stakeholder pressure and therefore adopt EA more extensively than services. Without sector-differentiated enforcement, sector-differentiated adoption cannot emerge; stakeholder pressure operates uniformly weakly across industries. This validates institutional-void theory’s prediction that stakeholder mechanisms require functioning institutions to operate [46].

Similarly, foreign ownership fails to predict adoption (p = 0.125), contradicting institutional-isomorphism expectations [54] that multinational subsidiaries transfer parent-company practices. Post hoc analysis shows that Albanian FDI concentrates in low-technology assembly, tourism, and retail—sectors where parent companies impose minimal EA requirements. Unlike foreign manufacturing FDI in Poland and the Czech Republic, which transferred ISO 14001 systems during EU accession [82], Albanian FDI does not transfer sustainability practices because source countries (primarily Italy, Greece, and Turkey) do not require them in these sectors [61]. This sectoral concentration pattern shows that isomorphism theory operates conditionally: transfer occurs when parent companies require practices globally, not when they maintain differential standards.

5.2. Asymmetric Institutional Substitution: Internal Factors Enable, External Barriers Constrain

H3 analysis of CSRD implementation readiness reveals a theoretically novel asymmetry: internal organizational capacity enables preparation (OR = 2.28, p = 0.001), external support demonstrates no enabling effect (p = 0.555), and external challenges significantly constrain readiness (OR = 0.49, p = 0.022). This asymmetry extends institutional-void theory prospectively into compliance-preparation domains [34,46]. The pattern suggests that substitution mechanisms operate differently for voluntary adoption and mandated compliance [39]. For current voluntary EA adoption, firms can rely entirely on internal resources—managers with environmental knowledge initiate practices independent of external incentives. However, for future CSRD compliance, internal capabilities are necessary but insufficient: firms require external coordination infrastructure (technical standards, consultant availability, regulatory clarity, market recognition) that weak governance cannot provide [10,55].

Albania’s regulatory limbo exemplifies this constraint. As an EU candidate since 2014 with expected accession by 2030, Albanian firms face certain CSRD obligations but uncertain timelines, unclear requirements, and absent implementation guidance [10]. This temporal disconnect creates anticipatory institutional voids: future mandates generate pressure for current preparation, yet institutions required for compliance (technical standards, audit frameworks, consultant expertise) remain undeveloped [39]. Our H3 finding that 55.6% of firms report “moderate” readiness, whereas only 11.9% report “low” readiness, suggests aspirational rather than substantive preparation—firms express commitment but lack concrete implementation capacity.

6. Conclusions

This study examined environmental accounting adoption determinants in Albania using ordinal logistic regression analysis of 151 non-financial companies. Three principal findings emerged, each with distinct theoretical and practical significance. First, internal managerial factors—environmental knowledge and pro-environmental attitudes—significantly predict EA adoption (H2: R² = 0.142, internal factors p < 0.05), whereas external institutional factors show negligible influence (H2: external factors p > 0.25, McFadden’s R² = 0.016) [39]. This pattern is the inverse of adoption mechanisms documented in developed economies, where stakeholder pressures drive implementation [64]. Our analysis substantiates institutional-void theory: when formal enforcement infrastructure collapses, firms substitute internal managerial commitment for absent external coordination [34]. However, effect sizes remain modest (standardized β < 0.10), indicating that internal factors enable adoption among motivated managers but cannot independently drive mass implementation without institutional support [46].

Second, structural characteristics exhibit selective predictive power, revealing boundary conditions for stakeholder theory. Firm size significantly predicts adoption (H1: OR = 1.67, p = 0.020) [69,83], whereas sector type, ownership structure, and market orientation are inconsequential (all p > 0.12), contradicting Freeman’s stakeholder theory, which predicts sector-differentiated adoption patterns [15,47]. This contradiction is theoretically productive: Albania’s regulatory vacuum (0.3 inspections per manufacturer annually) renders stakeholder pressure uniformly weak across industries, preventing sector-specific adoption patterns [37]. This finding reframes stakeholder theory from a universal framework as a conditional model requiring institutional prerequisites: stakeholder mechanisms operate only when enforcement capacity exceeds the global 50th percentile.

Third, asymmetric institutional substitution differentiates voluntary adoption from mandatory compliance preparation. H3 analysis of CSRD readiness (2028–2030 mandate) shows that internal organizational capacity enables preparation (OR = 2.28, p = 0.001), external support has no enabling effect (p = 0.555), and external barriers significantly constrain readiness (OR = 0.49, p = 0.022) [84]. This pattern indicates that internal commitment alone cannot prepare firms for mandated compliance when external infrastructure (technical standards, audit frameworks, regulatory guidance) is absent [84]. Empirically, 55.6% of firms report “moderate” readiness, while only 11.9% report “low” readiness, suggesting aspirational rather than substantive preparation. This asymmetry highlights a critical insight: capacity-building interventions targeting internal factors can strengthen voluntary adoption but cannot independently support mandatory compliance preparation without parallel external institutional scaffolding [84,85,86].

7. Study Limitations

This study’s contributions are bounded by three principal limitations. First, the cross-sectional design precludes causal inference. Observed associations between managerial knowledge and EA adoption may reflect reverse causation (adoption → training investment) or confounding by omitted environmental progressiveness. Self-report measures also introduce social desirability bias; although Harman’s single-factor test (largest factor 34%, below the 50% threshold) suggests limited common-method variance, managers may still overstate environmental commitment and implementation. Future research should triangulate survey responses with objective data (ISO 14001 certifications, published sustainability reports, environmental permits).

Second, sample representativeness constraints limit generalizability. The sample skews toward medium enterprises (57.0%), while large firms (≥250 employees) represent only 12.6% (N = 19)—a concern given that CSRD mandates primarily target large enterprises. Geographic concentration in Tirana (52.3% versus 30% nationally) over-represents urban firms with consultant access. Purposive sampling deliberately targeted EA-aware companies, potentially capturing best-case scenarios rather than typical practice. The 33.6% response rate also raises non-response bias concerns; although early and late respondents exhibited equivalent EA adoption levels (p > 0.54), firms declining to participate may systematically differ. Longitudinal panel designs tracking firms post-CSRD implementation and randomized controlled trials testing capacity-building interventions would strengthen causal inference and external validity.