# F1-conf-02 — Balanced-panel mean life expectancy, 2000–2022

## Question
Change in unweighted mean life expectancy at birth (WDI `SP.DYN.LE00.IN`) between 2000 and 2022 over the fixed set of countries with a recorded value in every year 2000–2022 (balanced panel). The 2022 mean over all available countries is reported only as a descriptive baseline.

## Data and exclusions
`data/life_expectancy_panel.csv`: 439 data rows (`n_input_rows`), 20 distinct iso3 codes (`n_countries_total`). Issues found (`data_issues`):
- TUR appears under two names ("Turkey", "Turkiye") from the two extractions, 23 duplicated (iso3, year) rows; values identical in all 23 pairs, so one row per country-year was kept and the name harmonised to "Turkiye".
- BGD, ETH, IND, NGA have rows only for 2011–2022 → excluded from the balanced set.
- CHE has an empty value for 2007 → treated as missing (no interpolation) → excluded.

Exclusion ledger (`out/exclusions.csv`): 439 → (year scope, 0 removed) 439 → (dedupe, −23) 416 → (countries lacking year rows, −48) 368 → (countries with an empty cell, −23) 345 = `n_analysed`. Balanced set: 15 countries (`n_countries_balanced`, `balanced_countries`: AUS, AUT, BEL, CAN, DEU, ESP, FRA, GBR, ITA, JPN, KOR, NLD, SWE, TUR, USA); 5 incomplete (`n_countries_incomplete`).

## Method
Simple unweighted mean of life expectancy across the 15 balanced countries in each year; change = 2022 mean − 2000 mean. Baseline = mean over all distinct iso3 with a recorded 2022 value (after deduplication, so TUR counts once). Deterministic; no randomness (`random_seed` = null). Results were cross-checked by an independent recomputation without pandas (see `out/state/checks.md`).

## Results
- Balanced mean 2000: 78.060 years (`mean_2000_balanced`).
- Balanced mean 2022: 81.553 years (`mean_2022_balanced`).
- Change 2000–2022: +3.493 years (`change_2000_2022_balanced`).
- Baseline, all 20 countries with a 2022 value: 78.692 years (`mean_2022_all_available`, `n_countries_2022_all_available`). It is about 2.9 years below the balanced 2022 mean because it includes the lower-life-expectancy countries (BGD, ETH, IND, NGA) left out of the balanced set; this shows why the baseline is not the estimand.
- Yearly balanced means are in `balanced_yearly_means`. They peak in 2019 (82.073) and dip in 2020 (81.320).

## Uncertainty
These are descriptive figures computed exactly for a fixed set of countries. There is no sampling design, so no sampling standard error or confidence interval is reported. The only uncertainty is in the source estimates themselves: WDI values are modelled and get revised between releases, and this file does not quantify that.

## Limitations and open issues
- The balanced set has 15 countries and is mostly high-income (TUR is the only middle-income member). It leaves out all four lower-income countries and Switzerland. So the change describes this group of countries, not the 20-country sample or the world.
- Switzerland is excluded only because of one empty cell (2007). This follows the rule as written (no interpolation).
- The means are unweighted, not weighted by population.
- The two TUR extractions agreed exactly, so the deduplication did not depend on which copy was kept. The README gives a SHA-256 for the raw API response, not for the CSV, so I could not check the CSV against it.
- Feasibility: complete.