narrative summary

analysis/new analysis Aug 2025/analayis11_2020_nooutliers.ipynb

@@ -1791,25 +1791,55 @@
    "id": "1f786261",
    "metadata": {},
    "source": [
-    "## narrative\n",
+    "## Narrative Summary of Reporting Delay Analysis\n",
     "\n",
-    "In the analytic sample (N = 11,376; after IQR outlier trimming), predicted reporting delays declined after 2020 across spill types and rurality. Using a Poisson GLM with a full C(spill_type) × C(Period) × C(rurality) interaction and parametric bootstrap (B = 2,000) for uncertainty, the estimated reduction in predicted delay (Before 2020 − 2020 and After) ranged from about 0.22–0.46 days (≈5–11 hours). For example, Historical spills in urban areas show a median reduction of 0.460 days (95% bootstrap CI 0.404–0.517, p < 0.001), and Recent spills in urban areas show a median reduction of 0.299 days (95% CI 0.222–0.375, p < 0.001).\n",
+    "### Analytic Sample and Model\n",
     "\n",
-    "Comparing spill types, Recent spills tend to have longer predicted delays than Historical spills, with the largest and most robust gap in urban settings. After 2020, the Recent − Historical difference in Urban areas is 0.291 days (95% CI 0.244–0.341, p < 0.001), equivalent to roughly 7.0 hours. Differences by spill type are smaller and often not statistically distinguishable in some Suburban and Rural cells before 2020 (e.g., Before 2020 Suburban: 0.020 days, 95% CI −0.222–0.244, p ≈ 0.85).\n",
+    "- **Sample size:** N = 11,376 (after IQR outlier trimming)\n",
+    "- **Model:** Poisson GLM with full interaction: `C(spill_type) × C(Period) × C(rurality)`\n",
+    "- **Uncertainty:** Parametric bootstrap (B = 2,000)\n",
     "\n",
-    "Limitations: the primary inference is model‑dependent (Poisson fit showed underdispersion, Pearson χ2/df ≈ 0.69), so results rely on the bootstrap procedures used to quantify uncertainty; a nonparametric case bootstrap (B = 1,000) produced broadly similar group CIs. Small sample sizes in some SpillType×Period×rurality cells limit precision for those specific comparisons, and reported p‑values are unadjusted for multiple contrasts; therefore emphasize effect sizes and bootstrap CIs in policy interpretation.\n",
+    "### Main Findings\n",
     "\n",
-    "Figure caption Figure X. Predicted reporting delays (days) for each Spill Type × Period × rurality cell with parametric bootstrap (B = 2,000) 95% confidence intervals. Points show bootstrap medians of the model‑predicted group means; vertical bars are 2.5th–97.5th percentiles from the bootstrap distribution."
+    "- **Reporting delays declined after 2020** across spill types and rurality.\n",
+    "- **Estimated reduction (Before 2020 − 2020 and After):**  \n",
+    "    - Range: **0.22–0.46 days** (≈5–11 hours)\n",
+    "    - **Historical spills, Urban:**  \n",
+    "        - Median reduction: **0.460 days**  \n",
+    "        - 95% bootstrap CI: **0.404–0.517**  \n",
+    "        - p < 0.001\n",
+    "    - **Recent spills, Urban:**  \n",
+    "        - Median reduction: **0.299 days**  \n",
+    "        - 95% CI: **0.222–0.375**  \n",
+    "        - p < 0.001\n",
+    "\n",
+    "- **Spill Type Comparison:**  \n",
+    "    - Recent spills generally have longer predicted delays than Historical spills.\n",
+    "    - **Largest and most robust gap:** Urban settings after 2020\n",
+    "        - **Recent − Historical, Urban, 2020 and After:**  \n",
+    "            - Difference: **0.291 days** (≈7.0 hours)  \n",
+    "            - 95% CI: **0.244–0.341**  \n",
+    "            - p < 0.001\n",
+    "    - **Suburban and Rural cells before 2020:**  \n",
+    "        - Differences smaller and often not statistically significant  \n",
+    "        - Example: Suburban, Before 2020: **0.020 days** (95% CI −0.222–0.244, p ≈ 0.85)\n",
+    "\n",
+    "### Limitations\n",
+    "\n",
+    "- **Model dependence:** Poisson fit showed underdispersion (Pearson χ²/df ≈ 0.69)\n",
+    "- **Bootstrap reliance:** Results rely on bootstrap procedures for uncertainty quantification\n",
+    "- **Sensitivity:** Nonparametric case bootstrap (B = 1,000) produced similar group CIs\n",
+    "- **Small sample sizes:** Some SpillType × Period × rurality cells have limited precision\n",
+    "- **Multiple comparisons:** p-values unadjusted; emphasize effect sizes and bootstrap CIs for policy\n",
+    "\n",
+    "---\n",
+    "\n",
+    "### Figure Caption\n",
+    "\n",
+    "**Figure X.** Predicted reporting delays (days) for each Spill Type × Period × rurality cell with parametric bootstrap (B = 2,000) 95% confidence intervals.  \n",
+    "Points show bootstrap medians of the model-predicted group means; vertical bars are 2.5th–97.5th percentiles from the bootstrap distribution."
    ]
   },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "eeed5308",
-   "metadata": {},
-   "outputs": [],
-   "source": []
-  },
   {
    "cell_type": "code",
    "execution_count": 49,