Summary

• Adds Study 013: Keh, H. T., Foo, M. D., & Lim, B. C. (2002). "Opportunity Evaluation under Risky Conditions: The Cognitive Processes of Entrepreneurs." Entrepreneurship Theory and Practice, 27(2), 125–148.
• Cross-sectional survey of N = 77 founders of top SMEs in Singapore, measuring four cognitive biases (overconfidence, illusion of control, planning fallacy, belief in small numbers) and their effects on risk perception and opportunity evaluation via a standardized business vignette.
• Implements 5 findings (F1–F5) from the paper's calibration test and Table 4 regressions:
  • F1: Entrepreneurs are overconfident (one-sample t-test, mean items outside 90% CI > 1)
  • F2: Risk perception negatively predicts opportunity evaluation (OLS, β = −0.50)
  • F3: Illusion of control negatively predicts risk perception (multiple OLS, β = −0.76)
  • F4: Illusion of control positively predicts opportunity evaluation (multiple OLS, β = 0.40)
  • F5: Belief in small numbers positively predicts opportunity evaluation (multiple OLS, β = 1.17)

Files added (14 files, 1,811 lines)

studies/study_013/
├── index.json
├── README.md
├── source/
│   ├── Keh-Foo-Lim-2002-Opportunity-Evaluation.pdf
│   ├── metadata.json
│   ├── specification.json
│   ├── ground_truth.json
│   └── materials/
│       ├── section_a_risk_propensity.json      (5 forced-choice gamble items)
│       ├── section_b_cognitive_biases.json      (7 Likert items: 2 filler, 2 planning fallacy, 3 illusion of control)
│       ├── section_c_overconfidence.json        (10 confidence-interval items with verified answers)
│       └── section_d_case_vignette.json         (Mr. Tan vignette + 4 risk perception + 3 opportunity evaluation + 1 open-ended)
└── scripts/
    ├── config.py          (CustomPromptBuilder + StudyStudy013Config)
    ├── evaluator.py       (OLS regressions for F2–F5, one-sample t-test for F1)
    ├── study_utils.py     (shared scoring, D8 coding, response parsing)
    └── stats_lib.py       (p-value parsing utilities)

Implementation notes

• Overconfidence answer key: All 10 confidence-interval items verified against Yearbook of Statistics Singapore 2000, Changi Airport Group records, LTA Vehicle Quota Tender Results, and SingStat residential datasets.
• Persona variation: Each of 77 trials generates a randomized entrepreneur profile (age, sex, race, education, business size, founder status) sampled from the paper's Table 2 demographics to produce individual differences needed for regression analyses.
• Evaluator: F2 uses simple OLS (OE ~ risk_perception). F3–F5 use multiple OLS with controls [overconfidence, small_numbers, planning_fallacy, illusion_of_control, risk_propensity, age], matching the paper's Model 1 and Model 2 specifications.
• D8 open-ended item: Coded as belief in small numbers (+1) vs. statistical reasoning (−1) via keyword matching. Made optional in the prompt with neutral framing to avoid biasing AI responses toward data requests.

Test plan

• Structure verification passes (5/5 checks)
• D8 coding edge cases pass (10/10 tests)
• Full pipeline with synthetic data: all 5 findings computed, replication verdicts generated
• CI verify_study.sh study_013 validation
• CI build_studies_index.py index building

🤖 Generated with Claude Code