# W3-실증: Initial Results **발송일**: Tuesday, November 12, 2024 (저녁 6-8pm) **Subject**: [Empirical] Week 3: Initial Results --- Dear Charlie and Scott, **Progress This Week:** ✅ **Models 1-2 Estimated** *Model 1 (Reversal hypothesis):* ``` logit(Funding_Success_it) = β₀ + β₁·Vagueness_i + β₂·SeriesB_t + β₃·(Vagueness × SeriesB) + Controls ``` **Results:** - β₁ (Vagueness main effect) = -0.028 (p = 0.04) → Vague firms struggle at Series A ✓ - β₂ (SeriesB main effect) = -1.35 (p < 0.001) → Series B harder overall ✓ - β₃ (Vagueness × SeriesB) = +0.042 (p = 0.02) → Reversal confirmed ✓ **Interpretation:** - At Series A: 1 SD increase in vagueness → 14% lower funding probability - At Series B: Same vagueness → 8% *higher* funding probability (among Series A survivors) - **Reversal pattern supported**: Vagueness hurts early, helps later *Model 2 (Integration cost moderator):* ``` Model 1 + β₇·(Vagueness × SeriesB × High_Integration_Cost) ``` **Results:** - β₇ (Three-way interaction) = +0.063 (p = 0.08) → Marginally significant - Direction consistent with hypothesis (hardware firms benefit more from vagueness) **Interpretation:** - Hardware firms with vague promises: +12% Series B success vs. precise hardware firms - Software firms with vague promises: +5% Series B success vs. precise software firms - **Integration cost moderates reversal**: Effect stronger for high-commitment ventures ✓ ✅ **Robustness Checks Initiated** *Alternative vagueness measures:* - Word count variance (high word count = vague) → Results hold - Hedge word frequency ("approximately," "around") → Results hold - Technical jargon density (low jargon = vague) → Results weaker but directionally consistent *Sample sensitivity:* - Exclude top/bottom 5% vagueness → Results hold - Exclude firms with <$1M Series A → Results strengthen (larger effect sizes) *Diagnostics:* - VIF < 2.5 for all predictors (no multicollinearity) - Cook's D < 0.5 for all cases (no influential outliers) - Pseudo R² = 0.28 (reasonable fit for logistic model) ⏳ **In Progress** - Visualization: Reversal curves (vagueness × funding probability by stage) - Visualization: Moderator effects (hardware vs. software trajectories) - Table construction: Tables 1-3 for draft --- **Next Week Target:** Complete visualizations (Figures 1-2), draft results section connecting findings to Clockspeed/S-curve frameworks. Best, Angie --- ## 작성 가이드 (당신이 채울 때) **실제 결과로 교체할 부분:** 1. **모든 계수와 p-values**: β₁ = -0.028 (p = 0.04) → 실제 회귀 결과 2. **해석 수치들**: "14% lower" → 실제 marginal effects 계산값 3. **Pseudo R²**: 0.28 → 실제 모델 fit **만약 결과가 가설과 반대면:** ``` **Results:** - β₁ (Vagueness main effect) = +0.018 (p = 0.32) → Not significant - β₃ (Vagueness × SeriesB) = -0.012 (p = 0.54) → Reversal not supported **Interpretation:** - Hypothesis not supported in this sample - Possible explanations: [list 2-3 reasons] - Alternative specifications under exploration ``` **정직하게 보고하되, constructive하게:** - 결과가 안 나와도 괜찮음 ("The goal is to connect hypothesis to test" - Scott's words) - "Not supported" ≠ "Failed" - 대안 설명 제시 **만약 analysis가 아직 안 끝났으면:** ``` ⏳ **Models 1-2 Running** - Logistic regression syntax validated - Convergence achieved for Model 1 - Model 2 estimation in progress (computational issues with three-way interaction) **Preliminary results (Model 1 only):** - β₁ = [preliminary estimate] (significance pending bootstrap SE) **Next steps:** - Complete Model 2 by end of week - Run robustness checks next week ``` **톤:** - Factual, 숫자 중심 - 결과에 대한 해석은 간단히 (깊은 논의는 대면 미팅에서) - "Supported ✓" or "Not supported" 명확히 **핵심:** 가장 정직하게, 실제 결과를 보고하는 것이 목표.