Behavioural Determinants of Credit Appraisal: An Integrated Analysis of Risk Attitude, Experience, and Loan Officers’ Attributes in Indian Banks

2.1.1 Implication for this Study

Macro and bank-level remedies are necessary but insufficient. Micro-level behaviour of credit officers at loan origination – how they gather, weight and interpret borrower information – matters for preventing the build-up of future NPAs (Das 2023; RBI 2024).

2.2.1 Implication for this Study

Behavioural models must be extended from borrower behaviour to lender cognition; controlled experiments and standardized fictitious-case designs are effective for isolating the influence of experience and risk attitude on information acquisition and decision accuracy (Payne 1976; Rosman and O’Neill 1993).

2.3.1 Implication for this Study

Even as banks deploy better predictive models, human loan officers continue to play a gatekeeping role. Understanding which human attributes (diagnostic skills, domain knowledge, stress-tolerance, etc.) complement algorithmic outputs is essential for effective, fair, and explainable lending systems.

3.1.1 Measurement of Risk Attitude

Risk attitude was measured using a 22-item scale adapted from Hedelin and Sjoberg (1995), derived from the original 28-item instrument to better reflect Indian credit practices (Annex 4). Reliability testing yielded Cronbach’s α = 0.72, indicating acceptable internal consistency for exploratory behavioural research (Nunnally and Bernstein 1994). Items captured both risk-averse and risk-seeking orientations in lending contexts, scored on a five-point Likert scale (1 = Strongly Disagree to 5 = Strongly Agree). Reverse coding of aversion items produced a composite index where higher scores denoted stronger risk-taking propensity.

3.1.2 Measurement of Information Acquisition

Information-acquisition behaviour was captured through 15 borrower-related categories identified from prior credit-appraisal literature (Annex 5). Respondents rated the probability of consulting each information type on a seven-point scale (0 = Not Probable to 6 = Very Probable). The composite Information variable – the mean across all categories – represented the overall breadth of information used (Cronbach’s α = 0.86).

3.1.3 Analytical Approach

Kendall’s W was first applied to assess inter-rater agreement in the ranking of information categories, followed by the Kruskal-Wallis test to compare information-acquisition patterns across risk-attitude groups (risk-averse, neutral, risk-taking), experience levels, and gender. The objective was to determine whether risk attitude meaningfully alters information search intensity or selectivity.

3.2.1 Experimental Design

Each respondent evaluated two standardised fictitious loan proposals (Annex 6) representing opposite financial profiles:

1. Case A: Creditworthy borrower (loan ₹ 29.4 lakh)

2. Case B: Bankrupt borrower (loan ₹ 23.5 lakh)

Both cases contained 74 information indicators (Annex 7) grouped into four categories – balance-sheet indicators (21), financial ratios indicators (25), income-statement indicators (11), and qualitative/non-financial indicators (17). Participants freely selected any indicators before issuing an approval or rejection decision; selections were auto logged (1 = viewed, 0 = not viewed).

3.2.2 Data Derivation

Two dependent variables were constructed:

1. Information-acquisition breadth: mean proportion of indicators accessed within and across categories.

2. Order-effect index: Spearman correlation between the sequence of viewed indicators and their presentation order; coefficients ≥ 0.85 signified passive (order-dependent) acquisition.

Decision outcomes (approve/reject) for each case were coded dichotomously to compute Type I and Type II error rates and inter-group consensus (Cramer’s V).

3.2.3 Analytical Techniques

The analysis proceeded in three phases corresponding to hypotheses H3–H6:

1. Regression Analysis – Multiple regression with Information as dependent variable and Experience, Gender, and Risk Attitude as predictors examined the determinants of self-reported information breadth.

2. ANOVA/Kruskal–Wallis – Between-group comparisons tested whether experience level significantly affected the number and type of indicators accessed.

3. Order-Effect and Decision-Outcome Analysis – Spearman rank correlations captured sequence dependence; Chi-square and Cramer’s V measured agreement and bias in final lending decisions.

3.2.4 Measurement Logic

The fictitious-case framework allowed objective observation of decision behaviour under controlled, equivalent information conditions – removing contextual noise inherent in real-case data.

3.3.1 Phase 1: Identification of Attributes

Unstructured interviews (Annex 8) were conducted with 26 senior loan officers (≥15 years’ experience) across public and private banks. Guided by the grounded-theory approach (Glaser and Strauss 1967), transcripts were coded to extract recurring themes such as diagnostic ability, sincerity, analytical reasoning, decisiveness, adaptability, and stress-tolerance. This yielded 27 distinct attributes (C1–C27) clustered into four preliminary domains: cognitive capability, basic abilities, interpersonal traits, and behavioural discipline. The complete list of 27 attributes is:

1. Cognitive Capability

   (Attributes reflecting mental sharpness, analytical thinking, and decision-making skills)

   1. C1. Diagnostic abilities

   2. C2. Assumption of responsibility

   3. C3. Ability to master decision aids

   4. C5. Decisiveness

   5. C11. Information acquiring skill

   6. C12. Intuition

   7. C16. Perceptiveness

   8. C17. Judgement of human nature

   9. C18. Accuracy

   10. C19. Curiosity

   11. C26. Ability to know what is relevant

2. Basic Abilities

   (Attributes denoting foundational skills and functional competence)

   1. C4. Current knowledge

   2. C6. Energy and interest

   3. C7. Adaptability

   4. C9. Experience

   5. C10. Education

   6. C13. Initiative and creativity

   7. C23. Ability to focus on sales

   8. C25. Being a good teacher

3. Interpersonal Traits

   (Attributes relating to social competence, empathy, and communication)

   1. C8. Negotiation skills

   2. C14. Stubbornness

   3. C21. Warmth and friendliness

   4. C22. Common sense

   5. C24. Sincerity

   6. C27. Communication skills

4. Behavioural Discipline

   (Attributes reflecting reliability, emotional control, and professional attitude)

   1. C15. Stress tolerance

   2. C20. Reliability

3.3.2 Phase 2: Survey Administration and FAHP Application

A structured questionnaire (Annex 9) was then distributed to 337 respondents – 155 senior loan officers and 182 outsourced loan officers – who rated each attribute’s importance on a seven-point Likert scale. The two-group design facilitated both between-group and within-group analyses of perceived attribute significance. Further, the 27 attributes identified in the interview phase were subsequently refined and prioritised using the Fuzzy Analytic Hierarchy Process (FAHP) to assess their relative importance in credit decision-making.

3.3.3 Phase 3: Quantitative Analyses

1. Between-Group Comparison (Kruskal–Wallis): tested inter-group differences in importance rankings.

2. Within-Group Consensus (Split-Sample Replication): assessed intra-group homogeneity by correlating half-sample response profiles (r̄ = 0.16 for seniors; r̄ = 0.32 for outsourced loan officers).

3. FAHP Prioritisation: To derive relative weights, the Fuzzy Analytic Hierarchy Process combined pairwise-comparison matrices with triangular fuzzy numbers (l, m, u). All consistency ratios (CR < 0.10) confirmed logical coherence.

3.3.4 Phase 4: Model Development

Synthesising the attribute hierarchy with behavioural findings from Sections 4.1–4.2 yielded the Loan officer Decision Competence Model (LODCM). The LODCM conceptualises decision competence as an interaction among:

1. Cognitive inputs (risk attitude, experience),

2. Information-processing behaviours (acquisition breadth, order, and judgement), and

3. Underlying personal attributes (capability, abilities, volitional control).

This integrative model connects who the loan officer is, with how the loan officer thinks and what decisions are ultimately made, providing a comprehensive behavioural framework for credit-risk appraisal.

Collectively, these three methodological components – risk-attitude analysis, experience-based experimentation, and attribute prioritisation – establish the empirical foundation for the behavioural findings presented in Section 4. The next section discusses the results of these analyses and synthesises them into the Loan officer Decision Competence Model.

4.1.1 Interpreting Kendall’s W (Inter-Rater Agreement)

The Information variable, representing the average breadth of information acquisition across the 15 categories, yielded a mean of 2.45, indicating that on average, respondents used approximately 49 % of the available information during credit assessment (Table 2). The Kendall’s W = 0.37 indicates a moderate level of inter-rater agreement among loan officers. In this context, inter-rater agreement refers to the degree of similarity in how different loan officers rank and prioritise information categories. The W value closer to 1 would imply strong consensus; hence, the result shows that while loan officers share some common preferences (e.g. financial statements are widely emphasised), considerable individual variation remains in how information is weighted during appraisal.

4.1.2 Formation of Information Acquisition Groups

To further benchmark information acquisition behaviour, respondents were categorised into three groups:

1. Low information acquirers

2. Moderate information acquirers

3. High information acquirers

These groups were generated using a tertile split of the composite information variable. That is, the distribution of scores across respondents was divided into three equal segments. This standard classification approach enables comparative analysis without imposing subjective thresholds and is widely applied in behavioural research. Among high information acquirers, the three most emphasised information types remained the same: financial statements, banking track record, and technical evaluation reports.

4.1.3 Experience Categories Clarification

The experience groups referenced in subsequent analysis were based on years of professional credit appraisal experience, as documented in the respondents’ profiles:

1. Low Experience – 2–5 years of professional experience – Early-stage procedural learning phase.

2. Moderate experience – 5–8 years of professional experience – Stable operational and evaluative competence development.

3. High experience – Greater than 8 years of professional experience – Advanced lending judgement and institutional expertise.

These categories reflect established progression stages in credit appraisal roles, consistent with HR grading and training pathways in Indian banking institutions.

4.1.4 Kruskal–Wallis Results

The Kruskal–Wallis test was applied to determine whether risk attitude groups (risk-averse, risk-neutral, risk-taking), experience levels, and gender differed in their information acquisition behaviour (Table 3). The analysis revealed no statistically significant differences across risk attitude groups in terms of the type of information acquired; thus, Hypothesis 1 was rejected.

For measuring risk attitudes, the study employed a modified 22-item scale refer Table 4 (derived from Hedelin and Sjoberg 1995) adapted from the original 28-item instrument. The full list of items, along with retained and excluded items, is provided in Annex 5: Derivation of 22-Item Risk Attitude Scale. Reliability analysis of the adapted 22-item Risk Attitude Scale yielded a Cronbach’s alpha of 0.72, indicating acceptable internal consistency. This suggests that the retained items are sufficiently correlated and measure a common underlying construct – the risk attitude of loan officers. Given that values above 0.70 are generally considered acceptable for behavioural and social science research (Nunnally and Bernstein 1994; Hair et al. 2010), the scale was deemed reliable for further analysis.

The classification of respondents into risk groups was based on questionnaire responses (see Annex 1 & Annex 3), yielding three categories:

1. Risk-averse loan officers (n = 54),

2. Risk-neutral loan officers (n = 49), and

3. Risk-taking loan officers (n = 56).

Kruskal–Wallis test results (Table 3) showed no significant differences across these risk-attitude groups regarding the type of information acquired, leading to the rejection of Hypothesis 1.

4.2.1 Hypothesis 3: Level of Experience Significantly Influences Information Acquisition Behaviour

To further investigate the determinants of information acquisition behaviour, a multiple linear regression model was employed, where the Information variable served as the dependent variable, and Experience, Gender, and Risk Attitude were included as independent variables. The Gender variable was coded as binary (female = 1; male = 0), while Experience and Risk Attitude were treated as continuous predictors. Preliminary diagnostic tests confirmed that the model satisfied all key assumptions, including normality of residuals, absence of multicollinearity, lack of autocorrelation, and homoscedasticity, thereby validating the robustness of the regression results. The detailed regression coefficients are presented in Table 5, while Table 6 provides the corresponding correlation matrix for the variables incorporated in the model.

4.2.2 Construction of the Dependent Variable

The Information variable represents the overall extent of information acquisition behaviour exhibited by loan officers during the credit evaluation process. It was derived by averaging the self-reported frequency ratings across 15 borrower-related information categories (see Annex 5: Derivation of 15 Information Sets). Each category was rated on a seven-point probability scale, ranging from Not Probable (0) to Very Probable (6), indicating the likelihood that an loan officer would obtain that specific type of information during a credit appraisal. The composite score, calculated as the mean of these 15 items, thus reflects the breadth and depth of information acquisition by each respondent. A higher score corresponds to more extensive and detailed information-gathering behaviour. The reliability of this composite index was confirmed to be strong, with a Cronbach’s alpha coefficient of 0.86, signifying high internal consistency among the 15 items.

4.2.3 Regression Results

Table 5 presents the results of the multiple regression analysis using Information as the dependent variable and Experience, Gender, and Risk Attitude as predictors.

4.2.4 Correlation Analysis

To assess potential interrelationships among the study variables, a correlation matrix was generated and is presented in Table 6.

The regression model explains approximately 92 % of the variance (Adjusted R² = 0.92) in information acquisition behaviour, indicating a strong explanatory power. Among the predictors, Experience exhibited a significant negative effect on information acquisition (β = –0.035; p < 0.01), suggesting that as professional experience increases, the breadth of information acquisition tends to decline. This pattern supports earlier findings that less experienced loan officers compensate for limited practical exposure by gathering more extensive data to minimise perceived uncertainty.

Neither Gender (β = 0.081; p > 0.05) nor Risk Attitude (β = 0.294; p > 0.05) emerged as significant predictors of information acquisition, indicating that while individual differences in gender and risk preference exist, they do not significantly alter the overall information-acquisition pattern. Collectively, these results demonstrate that experience exerts a significant inverse influence on the scope of information acquisition, while gender and risk attitude show limited explanatory relevance.

The next section extends this analysis by examining the role of loan officers’ experience in credit-granting decisions through controlled fictitious case scenarios (Annex 6).

4.2.5 Examining the Role of Loan officers’ Experience in Credit-Granting Decisions Through Controlled Fictitious Case Scenarios

This segment is organised in two parts: (A) Decision Framing and (B) Decision Making. The first part (Decision Framing) examines how loan officers frame and process information during loan evaluation, while the second part (Decision Making) reports on the final lending decisions and the insights drawn from the post-assessment questionnaire.

4.2.6 Clarification of Computation and Interpretation

The fictitious-case analysis (Annex 6 & 7) was conducted with a subsample of 61 loan officers (selected as per availability) drawn from the broader dataset. These respondents were divided into three pre-defined experience categories based on their professional designations and tenure in credit appraisal:

1. Out-sourced Loan officers (n = 19) – typically 2–5 years of experience.

2. Junior Loan officers (n = 18) – 5–8 years of experience.

3. Senior Loan officers (n = 23) – over 8 years of experience.

Each respondent was asked to assess two standardised fictitious loan applications, one from a financially sound firm (Case A) and another from a financially distressed firm (Case B).

Each case contained 74 discrete information indicators, systematically grouped into four categories:

1. Balance Sheet Indicators (21 items)

2. Financial Ratios (25 items)

3. Income Statement Indicators (11 items)

4. Qualitative Information (17 items)

During each case evaluation, loan officers could freely select which indicators they wished to consult before making a sanction or rejection decision. The system automatically recorded their selections as binary entries (1 = indicator viewed; 0 = indicator not viewed). The mean proportion of indicators accessed within each category was then computed for every participant, and the results were averaged across each experience group.

For example, when the mean value for Balance Sheet indicators among out-sourced loan officers is 0.63, it implies that on average, 63 % of the 21 available balance-sheet indicators were consulted. This equates to approximately 13 indicators per respondent (0.63 × 21 ≈ 13). Hence, the “mean total number of acquired indicators” represents the average count of all indicators accessed by the respondents within each experience category out of the 74 available indicators.

4.2.7 Interpretation of Results

Contrary to the earlier regression results (Section 4.2.1), where less experienced loan officers appeared to acquire more information in self-reported data, the fictitious-case experiment revealed an opposite trend. Here, senior loan officers accessed significantly more indicators (M = 65.13, SD = 8.02) than both junior (M = 55.11) and out-sourced loan officers (M = 48.22). This discrepancy arises from the difference in measurement context. In the survey data, information acquisition was self-reported, reflecting perceived rather than actual behaviour. In the fictitious-case experiment, information acquisition was objectively logged from participants’ real-time selections, capturing behavioural evidence rather than attitudinal responses. Senior loan officers’ broader information retrieval in the experimental setting likely reflects their greater cognitive organisation, domain knowledge, and confidence in filtering relevant data rather than superficial search behaviour.

Furthermore, statistically significant pairwise differences (noted by superscripts in Table 7) indicate that senior loan officers accessed a higher proportion of balance-sheet and income-statement indicators than the other groups. These two categories are considered critical for credit evaluation accuracy, suggesting that experience refines not only the amount but also the quality and selectivity of information acquired.

Overall, Table 7 demonstrates that professional experience affects both the breadth and strategic focus of information acquisition under controlled decision environments. Although the self-report data (Table 5) suggested declining breadth with experience, the experimental findings reveal that senior loan officers’ information use is broader yet more targeted, consistent with the cognitive-expertise literature (Payne 1976; Rosman and O’Neill 1993).

4.2.8 Section A: Decision Framing

4.2.9 Results of ANOVA Tests

The first ANOVA, with the total volume of information acquired as the dependent variable, revealed a statistically significant between-subjects effect of experience level, F(2, 53) = 6.33, p < 0.01, MS = 2,683.58, while other effects were non-significant. A conservative post hoc analysis (Tamhane’s test) identified the following pairwise differences:

1. Senior loan officers acquired significantly more information than junior loan officers (mean difference = 10.03, p = 0.05) and out-sourced loan officers (mean difference = 16.91, p < 0.01).

2. The difference between junior and out-sourced loan officers was not statistically significant.

These results suggest that although information quantity does not increase linearly with experience, senior loan officers exhibit a distinct behavioural pattern, combining higher total acquisition with greater discrimination in indicator selection.

4.2.10 Category-wise ANOVA Results

Subsequent analyses examined the proportion of acquired indicators within each of the four information categories. The results revealed significance between-subjects effects for several categories are:

1. Balance Sheet Indicators: MS = 0.89, F(2, 53) = 9.60, p < 0.001

2. Income Statement Indicators: MS = 0.57, F(2, 53) = 4.95, p = 0.01

3. Financial Ratio Indicators: MS = 0.76, F(2, 53) = 4.78, p = 0.01

4. Qualitative Information Indicators: MS = 0.42, F(2, 53) = 3.16, p = 0.04

In addition, a significant interaction effect between loan case and experience group was observed for the Balance Sheet category, MS = 0.03, F(2, 53) = 2.94, p = 0.02, implying that the selection pattern of balance-sheet indicators differed depending on both the credit case and the level of professional experience. The remaining interactions across the other three categories were non-significant, suggesting that the observed effects were primarily driven by experience rather than by contextual case differences.

Post hoc Tamhane’s pairwise comparisons revealed the following statistically significant differences among the experience groups:

1. Senior versus Junior loan officers: greater analytical depth in Financial Ratio indicators (mean difference = 0.25, p = 0.03).

2. Senior versus Out-sourced loan officers: higher emphasis on Balance Sheet indicators (mean difference = 0.30, p = 0.003).

3. Junior versus Out-sourced loan officers: stronger focus on Balance Sheet indicators (mean difference = 0.22, p = 0.04).

4. Senior versus Out-sourced loan officers: modest but significant difference in Qualitative Information acquisition (mean difference = 0.17, p = 0.05).

These outcomes collectively indicate that as experience increases, loan officers tend to rely more on quantitatively verifiable financial data (balance sheet, ratios, income statements), while also maintaining limited but meaningful attention to qualitative assessments such as managerial competence or project feasibility.

4.2.11 Interpretation and Theoretical Implications

The ANOVA findings indicate that experience significantly affects the structure rather than the volume of information acquisition. Senior loan officers exhibited both higher overall indicator use and greater focus on financial statement data, while out-sourced loan officers displayed wider variability and less consistent depth of analysis. These outcomes align with the cognitive-expertise theory, which asserts that experts rely on mental schemas and pattern-based recognition to efficiently navigate complex data environments (Chi et al. 1988). Experienced loan officers appear to activate well-organised cognitive frameworks that allow them to identify the most relevant signals (e.g. balance sheet stability, financial ratios), whereas less experienced officers tend to gather data more diffusely, reflecting exploratory rather than diagnostic behaviour.

Therefore, although the hypothesis assumed a reduction in total information volume with experience, the empirical results revealed the opposite: experienced loan officers acquire a broader yet more selective range of information. This nuance underscores that expertise in credit analysis does not necessarily entail less data use, but rather more strategic and cognitively efficient data acquisition.

Consequently, Hypothesis 4 is rejected, as experience did not reduce information acquisition volume; instead, it enhances the precision, hierarchy, and cognitive organisation of information processing.

4.2.12 Section B – Decision Making

4.2.13 Analytical Design and Measures

Each respondent’s two binary decisions (approve/reject) yield a four-box decision matrix, analogous to a confusion matrix:

1. Type I Error (False Positive): Approving a loan for a bankrupt firm (over-optimism).

2. Type II Error (False Negative): Rejecting a loan for a creditworthy firm (over-cautiousness).

These two error types indicate opposite behavioural tendencies: risk-seeking (Type I) versus risk-averse (Type II).

Further, Chi-square tests, Cramer’s V, and error rates were used to test for group-wise differences in decision consistency and bias. Table 9 shows the descriptive statistics and crosstab results (Table 10).

A Chi-square test confirmed significant differences in decision outcomes across experience groups:

Cramer’s V = 0.35 (p = 0.03) indicates a moderate association between experience group and lending decision (see Table 12).

The actual outcomes of the fictitious firms were known. Accuracy and bias were analysed using separate four-box matrices for each experience group (Table 13a–c).

4.2.14 Interpretation of Error Patterns

A comparison across groups reveals distinct error tendencies:

1. Out-sourced loan officers show higher Type I error (false approvals) – an optimism bias, possibly due to limited exposure to default risk.

2. Senior loan officers show higher Type II error (false rejections) – a rejection bias or risk-averse conservatism reflecting accumulated caution from professional experience.

3. Junior loan officers occupy an intermediate position, displaying inconsistency in both error types.

These results confirm that experience significantly shapes the nature and direction of decision bias. While seniors demonstrate more consistent and cautious decisions (as seen in Chi-square tests), this caution sometimes leads to excessive rejection of creditworthy borrowers – a typical trade-off in experienced lending judgment.

4.2.15 Confidence Ratings and Group Comparison

Participants also rated their confidence in each decision on a 7-point scale. No significant differences emerged among the groups (M = 3.98, SD = 1.33), implying that confidence was not necessarily aligned with decision accuracy. Senior loan officers, despite showing higher rejection rates, did not express greater confidence than their less experienced counterparts. The analyses collectively demonstrate that experience level significantly affects both the consistency and bias pattern of credit decisions. Chi-square and effect-size statistics show a significant relationship between experience and decision outcome, and the four-box matrices reveal systematic differences in error types (Table 14):

Thus, Hypothesis 6 is supported: experience significantly influences the level of agreement and accuracy in lending decisions. However, the direction of this influence reveals that experience promotes caution and procedural consistency rather than improved predictive accuracy. This finding aligns with dual-process models of decision-making (Kahneman and Klein 2009), which suggest that accumulated experience strengthens intuitive caution rather than analytical calibration. Senior loan officers appear to have internalised institutional risk norms, prioritising error avoidance over approval accuracy, thereby achieving higher inter-group consensus but lower optimal accuracy.

4.3.1 Desired Attributes of Loan officers

The behavioural and cognitive literature identifies the personal attributes of experts as key determinants of professional performance under uncertainty. In the context of credit evaluation, these attributes determine how loan officers perceive, interpret, and act on borrower information. Foundational works by Shanteau (1992a, 1992b) postulate that expert judgment in dynamic environments depends not merely on accumulated knowledge but on the ability to apply diagnostic reasoning and context-sensitive intuition. Similarly, Einhorn and Hogarth (2003) and Kahneman and Klein (2009) highlighted that decision quality in ill-structured problems emerges from the interplay between intuitive recognition (rapid, experience-based insight) and analytical control (structured reasoning), both of which are influenced by underlying personal dispositions.

In this study, unstructured interviews (Annex 8) with 26 senior loan officers revealed 27 attributes considered vital for professional competence in credit appraisal. Consistent with Shanteau’s (1992a) and Kaur and Singh (2022) findings, diagnostic ability and domain knowledge were viewed as the most fundamental attributes of senior loan officers, enabling them to filter relevant cues, interpret financial patterns, and discriminate between genuine borrower strength and superficial indicators. Attributes such as assumption of responsibility, decisiveness, adaptability, and initiative were also emphasised, reflecting professional accountability and situational judgment – both essential in credit risk contexts where decisions entail substantial economic and reputational consequences. Furthermore, interpersonal and affective attributes such as warmth, perceptiveness, sincerity, and stress-tolerance were consistently cited. These findings align with earlier behavioural research (e.g. Payne et al. 1993; Slovic 1987) suggesting that tolerance for ambiguity, curiosity, and emotional composure underpin stable decision-making in complex lending environments. Hence, the portrait of a competent loan officer extends beyond technical mastery to encompass balanced emotional and relational intelligence.

A structured questionnaire (Annex 9) was subsequently administered to 337 respondents -comprising 155 senior and 182 outsourced loan officers – to quantify the relative importance of the 27 attributes identified in the interview phase. Developed from the qualitative findings and adapted from Glaser and Strauss (1967), the instrument required respondents to rate each attribute on a seven-point Likert scale ranging from “Not Important” to “Extremely Important.” This two-group design enabled both between-group and within-group comparisons, forming the empirical foundation for the Fuzzy Analytic Hierarchy Process (FAHP) prioritisation of loan officers’ behavioural and professional attributes.

The 27 attributes identified from interviews were subsequently refined through the Fuzzy Analytic Hierarchy Process (FAHP), which prioritised them according to perceived importance in actual decision-making. Attributes marked “#” in Table 15 were retained following FAHP prioritisation; those marked “*” were rephrased or adapted during survey design (Annex 9). This classification mirrors the tripartite framework of competence – cognitive, behavioural, and interpersonal – found in prior literature on expert decision systems.

The empirical findings reaffirm Shanteau’s (1992a) model of expertise, where contextual sensitivity combined with cognitive control defines professional competence. Attributes such as diagnostic skill, perceptiveness, and adaptability demonstrate this synthesis: loan officers with stronger contextual awareness and structured analytical judgment can efficiently process borrower information without cognitive overload. Notably, 12 attributes overlapped with those reported in the A&S (1979) framework, underscoring theoretical continuity. This included decisiveness, assumption of responsibility, experience, curiosity, and stress-tolerance, among others. Their recurrence across decades and contexts reinforces the stability of these behavioural determinants of effective financial decision-making.

4.3.2 Linking Attributes to Decision Behaviour (Integration with Hypotheses 4–6)

The identification of these attributes also deepens the interpretation of the empirical results presented earlier under Decision Framing and Decision Making (Hypotheses 4–6). The cognitive and behavioural tendencies observed in those analyses – such as the risk-aversion of senior loan officers, their structured yet selective information acquisition, and their institutional order-following behaviour are closely aligned with the attribute clusters derived here.

For instance:

1. The diagnostic and analytical abilities (#) correspond to the broader yet targeted information acquisition found among senior loan officers (Hypothesis 4).

2. Responsibility, decisiveness, and stress-tolerance (#*) parallel the rejection bias and caution observed in senior groups’ final lending outcomes (Hypothesis 6).

3. Curiosity and adaptability (#*) explain why junior and out-sourced loan officers displayed more exploratory (though inconsistent) information patterns (Hypothesis 5).

4. Perceptiveness and sincerity (#*), finally, connect to the interpersonal dimension of credit evaluation, reflecting the importance of relational judgment in interpreting borrower intent and credibility.

Thus, the “desired attributes” framework not only validates the empirical patterns identified in the decision experiments but also provides a psychological foundation for understanding why those patterns emerge. In sum, these attributes encapsulate the human architecture of expertise in credit judgment – linking cognitive skill, behavioural style, and institutional discipline into a cohesive model of professional decision competence.

4.3.3 Loan officers’ Attributes Measurement – Between-Group Analysis

Building on Section 4.3.1 (attributes elicitation via interviews and FAHP) and the behavioural evidence from Decision Framing/Decision Making (Sections 4.2.2 Part A and B), we compared how Senior Loan officers (SLO) and Outsourced Loan officers (OLO) rate the 27 attributes. Items were rated on a 7-point Likert scale. Internal consistency of the 27-item instrument was α = 0.82, which is acceptable for behavioural research using adapted scales.

4.3.4 Computation of “Order” and “Rank”

1. Order (second column) is the fixed item sequence in the questionnaire. It is common to both groups and reflects the FAHP-informed content layout (see Annex 9). It is not computed from means; it simply records the presentation order used in measurement to facilitate cross-group comparison.

2. Rank (SLO/OLO) is the within-group internal priority of each attribute, computed separately for each group by sorting that group’s attribute means from highest (Rank = 1) to lowest (Rank = 27). Ties, when present, were assigned the average of tied rank positions.

Thus, Order = instrument sequence shared by both groups; Rank = within-group importance based on that group’s own means.

4.3.5 Descriptive Contrasts

Compared to OLO (group mean range 3.02–4.87), SLO reported systematically higher endorsement (group mean range 3.92–5.18) and lower dispersion (average SD across items: SLO 0.84 vs OLO 1.42). This indicates that seniors not only value these attributes more but also agree more among themselves about that valuation.

4.3.6 Classification of Three Significant Classes (and which Items are in Each)

We grouped attributes by their between-group (SLO vs OLO) Kruskal–Wallis p-values:

1. Class A – “Comparably evaluated” (non-significant, p ≥ 0.05; 5 items) Analytical skills; Experience; Accuracy; Ability to know what is relevant; Judgement of human nature.

2. Class B – “Moderately different” (0.01 ≤ p < 0.05; 3 items) Ability to master decision indicators; Knowledge; (none others at p<0.05 only).

3. Class C – “Strongly different” (p < 0.01; 19 items).01 ≤ p < 0.001 (double asterisks): Assumption of responsibility; Education; Common sense. p < 0.001 (***): Decisiveness, Energy and interest, Adaptability, Negotiation skills, Intuition, Initiative & creativity, Stubbornness, Stress tolerance, Perceptiveness, Curiosity, Reliability, Warmth & friendliness, Ability to focus on sales, Sincerity, Being a good teacher, Communication skills, Current knowledge (p < 0.001 in table).

This study used the attribute-wise KW tests, not a pooled test, to classify each attribute’s between-group separability: Class A = no evidence of difference; Class B = modest evidence; Class C = strong evidence.

4.3.7 Usage of Sub-population in this Section

1. Between-group comparison (Table 16): two independent respondent groups from the study sampling frame (see Section 3.2): SLO (bank-employed credit seniors) and OLO (third-party evaluators). All respondents in each group rated all 27 items.

2. Internal rank and dispersion: computed within each group from that group’s item means and SDs.

3. Split-half consensus diagnostic (odd/even subsamples): performed within each group to probe internal agreement beyond average ranks. We randomly split each group into odd-ID versus even-ID halves, computed pairwise correlations of item profiles across halves for each respondent, and averaged them. The average within-group consensus was r̄ = 0.18 (SD = 0.13), indicating weak internal alignment once we move beyond headline means.

4.3.8 Interpretation and Linkage to Earlier Findings

Three patterns align with earlier behavioural results:

1. Higher endorsement + lower dispersion among Seniors mirrors their broader-but-selective information acquisition (H4) and conservative lending stance (H6).

2. Many large SLO–OLO gaps (Class C) concentrate in diagnostic/volitional attributes (decisiveness, adaptability, stress tolerance, curiosity, etc.) – the same levers that differentiated information depth and error patterns in the fictitious cases.

3. No between-group differences on core cognitive architecture (Analytical skill, Experience, Accuracy, knowing what is relevant, Judgement of human nature) match the shared professional baseline noted in Section 4.2: both groups understand the fundamentals; what changes with professional socailisation is the weighting and consistency with which those fundamentals are applied.

Using a FAHP-informed, reliable 27-item instrument (α = 0.82), we find that Seniors consistently rate decision-critical attributes higher and with less dispersion than Outsourced loan officers. Attribute-wise nonparametric tests show 19 strong, 3 moderate, and 5 non-significant between-group differences. Importantly, the non-significant set comprises foundational cognitive skills shared across groups, while strong differences cluster in diagnostic, volitional, and interpersonal domains – precisely those that, in earlier sections, explained the selective breadth of information use (H4) and risk-averse decision patterns (H6) among Seniors.

4.3.9 Loan officers’ Attributes Measurement – Within-Group Analyses

This subsection complements the between-group results (Section 4.3.2) by testing internal consensus within each respondent group – Senior Loan officers (SLO) and Outsourced Loan officers (OLO) – on the 27 attributes identified in Section 4.3.1 and operationalised in Table 15. It aligns with the study’s behavioural arc: from Decision Framing and Decision Making (Sections 4.2.2 Part A and B) to underlying personal attributes (Sections 4.3.1).

4.3.10 Split-Sample Consensus Correlations

To quantify within-group concordance, this study used a split-sample (odd/even ID) replication procedure:

1. Split each group (SLO, OLO) into two non-overlapping halves by respondent ID (odd vs. even).

2. For each respondent i, compute a 27-vector of item scores x _i (C1–C27).

3. Compute that respondent’s correlation with the mean profile of the opposite half in the same group (i.e. odd i vs. even-mean; even i vs. odd-mean). Use Pearson’s r on the 27 items (Likert treated as quasi-interval; robustness checked with Spearman – pattern unchanged).

4. Repeat the split (permute IDs) several times; average each respondent’s r across replications.

5. Summarise the distribution of respondent-level r’s by group (mean and SD); compare groups using a Kruskal–Wallis (KW) test on respondent-level correlations (non-parametric due to non-normality).

Note on “categories disentangled.” Earlier, we resolved the attribute space into three dimensions (not four) via EFA (next subsection). The split-sample consensus here is orthogonal to that step: it does not add a fourth category; it measures agreement on the existing 27 items within each group.

4.3.11 Results

1. Within SLO: mean internal correlation r ˉ = 0.16 (SD = 0.09) → weak consensus.

2. Within OLO: r ˉ = 0.32 (SD = 0.16) → still modest, but significantly higher than SLO.

3. Between-group test on respondent-level correlations: KW mean ranks = SLO 625.80 versus OLO 1270.33; χ ² = 652.72, p < 0.001.

Despite seniors’ higher mean endorsements and lower dispersion across items (Section 4.3.2), their profile-to-profile similarity is lower than OLO’s. This indicates idiosyncratic weighting among seniors (consistent with expertise differentiation), whereas OLO show more homogeneous (template-like) profiles.

4.3.12 Reducing the Attribute Space and Prioritising Weights: EFA → FAHP

4.3.13 FAHP Synthesis: Local and Global Weights

The study aggregated fuzzy priorities to obtain (i) set-level weights (A, B, C) and (ii) local sub-attribute weights within each set; the global weight equals (set weight × local sub-attribute weight). The final ranking ties back to Table 15 via C-codes.

At the attribute set level, Basic abilities (Set B) dominate (48.7 %), followed by Capability (Set A) (27.34 %) and Interpersonal abilities (Set C) (23.96 %). At the sub-attribute level, the highest global priorities are C24 Sincerity (8.88 %), C11 Information acquiring skill (8.78 %), and C5 Decisiveness (8.48 %) – a triad that maps tightly to the behavioural patterns observed in Sections 4.2 (structured search, procedural discipline, conservative decision thresholds).

Within-group consensus is modest in both cohorts and significantly higher in OLO than SLO, indicating template-like homogeneity among outsourced evaluators versus senior idiosyncrasy consistent with differentiated expertise. EFA yields three coherent dimensions; FAHP prioritisation shows Basic abilities as most influential (48.7 %), led by C11 Information acquiring skill, while Capability (27.34 %) is anchored by C24 Sincerity and C4 Current knowledge, and Interpersonal abilities (23.96 %) by C5 Decisiveness. These weights align with the earlier behavioural evidence: seniors’ selective breadth of information (H4), order-constrained processing (H5), and risk-averse decision consensus (H6).

The empirical results indicate a stable pathway from personal attributes (FAHP-weighted C-codes) to information behaviours (selective breadth with institutional sequencing) and, ultimately, to lending outcomes (Type-I/II error trade-offs by experience). This pathway is consistent across survey and experimental evidence (H4–H6) and is anchored by the highest-weighted attributes (C24, C11, C5) as proximal drivers of decision discipline and diagnostic depth. On this basis, we advance an integrative Loan officer Decision Competence Model (IDCM) that (a) organises attributes into Set A: Capability, Set B: Basic Abilities, and Set C: Interpersonal/Volitional, (b) maps these sets onto Decision Framing (information scope and selectivity) and Sequencing (order dependence), and (c) predicts Decision Making outcomes (approval/rejection propensity and error structure). The model yields clear, testable propositions for future estimation and out-of-sample validation.

The collective findings from the between-group, within-group, and Fuzzy AHP analyses provide strong empirical support for the exploratory proposition (P1) that specific behavioural attributes significantly influence lending decisions. As observed in Table 16, several attributes – including sincerity (χ ² = 34.35, p < 0.001), adaptability (χ ² = 42.50, p < 0.001), decisiveness (χ ² = 39.96, p < 0.001), stress tolerance (χ ² = 38.19, p < 0.001), and perceptiveness (χ ² = 41.96, p < 0.001) – exhibited statistically significant differences between senior and outsourced loan officers, indicating that behavioural and interpersonal traits play a decisive role in credit-related judgement. Further, the Fuzzy AHP results (Tables 17–20) quantified these effects by assigning the highest relative importance to Information Acquiring Skill (C11; 8.78 %), Sincerity (C24; 8.88 %), Decisiveness (C5; 8.48 %), and Perceptiveness (C16; 7.88 %) – together explaining more than 33 percent of total decision-weight in the Loan officer Decision Competence Model (IDCM). These convergent results confirm that lending performance is not solely driven by analytical capacity but is deeply conditioned by underlying behavioural and cognitive attributes. Accordingly, Proposition P1 – stating that specific behavioural attributes (diagnostic ability, judgement, knowledge, and interpersonal competence) significantly influence lending decisions – is empirically supported in this study.