# Prespecified Florida–Niño 3.4 comparison
Recorded 2026-09-30, after source availability and coverage checks, before association calculations.

- Question: Is monthly Niño 3.4 associated with Florida statewide precipitation, and does that association vary by season or a short Niño-leading lag?
- Florida is selected for a physical, a priori reason: ENSO-related subtropical jet and cool-season storm-track variability. No other states will be screened.
- Rainfall proxy: NOAA Climate at a Glance statewide monthly precipitation, including snow water equivalent; inches converted to mm with 25.4 mm/in. Statewide averaging can hide local patterns.
- Exposure: NOAA PSL monthly Niño 3.4, ERSST v6, 5°N–5°S and 170°W–120°W, anomaly baseline 1981–2010, °C. This is a monthly index, not the three-month ONI.
- Outcome dates: January 1950–December 2025. Preceding Niño values are retained for lag alignment. No 2026 outcomes.
- Comparisons: DJF, MAM, JJA, SON defined by outcome calendar month, each at lags 0, 1, 2 months. Positive lag means earlier Niño predicts later precipitation. Primary comparison: DJF lag 0. All 12 comparisons will be reported; no strongest-model selection.
- Estimand: monthly precipitation change in mm per 1°C higher Niño anomaly, after calendar-month mean removal within 1950–2025; equivalent to month fixed effects plus a single Niño predictor within each season. Also report correlation between the month-centered variables. This is not a season-total regression.
- Direct contrasts: DJF minus JJA slopes at each lag; lag 1 minus lag 0 and lag 2 minus lag 0 within each season. DJF–JJA at lag 0 is the main seasonal contrast. Adjacent lags are marginal models, not independent lag effects.
- Uncertainty: paired circular moving-block bootstrap of whole calendar years, five-year blocks, 4,000 replicates, seed 20260930. Use the same year resample across all 12 models, re-estimating calendar-month means. This preserves short-range dependence except at block joins; repeated bootstrap draws are statistical resampling, not climate simulations.
- Report pointwise 95% percentile intervals and approximate simultaneous 95% max-standardized-deviation bootstrap intervals across 12 slopes and 11 contrasts (23 estimates). Do not use naive independent-month significance tests.
- Fixed sensitivity checks: three- and seven-year bootstrap blocks, and removing a common linear time trend in addition to calendar-month means from predictor and outcome. No window, lag, state, or transform search.
- Checks: unique contiguous months, expected counts and units, no sentinel/missing observations in selected period, correct lag dates, 228 months per season, manual slope/alignment checks, reproducible script execution and archive inspection.
- Constraints: public monthly tables; local CPU analysis; no climate simulations, extra plugins, or remote computing. Observational associations do not identify intervention effects.
- Delivery: preserved raw inputs and URLs/checksums, monthly aligned CSV, complete estimates and contrasts, executable Python analysis, checks, and concise report.
