spb/traite-evaluation-qc

1id,chapitre,description,script,test,resultat_attendu,resultat_obtenu,statut,date2CALC-36-01,36,Vérité du choc simulé : ATT global / cohorte A / cohorte B / fenêtre 2×2 (profil δ(k) déclaré),analysis/python/cases/ch36_panel.py,analysis/python/tests/test_ch36.py::test_plan_et_verite,"0.0454 / 0.0484 / 0.0356 / 0.04625",idem,verifie,2026-10-313CALC-36-02,36,DiD 2×2 cohorte A (témoins Montréal) : cellules 10.1379/10.2450/10.3696/10.4234 ; δ = 0.0534 (ET HC1 0.0201) ; variante 35 témoins 0.0468 (0.0177) ; placebo 2019T3 0.0155 (0.0101),analysis/python/cases/ch36_panel.py,analysis/python/tests/test_ch36.py::test_did_2x2,"0.0534 ; 0.0468 ; 0.0155",idem,verifie,2026-10-314CALC-36-03,36,TWFE statique : effets communs 0.0905 (ET 0.0047 / HC1 0.0053 / grappes 0.0198) ; RMR×trimestre 0.0702 (0.0054 / 0.0064 / 0.0264) ; within bidirectionnel = 0.0905 ; rapports 4.2 et 4.9,analysis/python/cases/ch36_panel.py,analysis/python/tests/test_ch36.py::test_twfe_et_erreurs_types,"0.0905 ; 0.0702","idem (linearmodels : 0.090460, ET grappes 0.0196 ; statsmodels cluster : 0.0198 ; HC1 identique 0.005289)",verifie,2026-10-315CALC-36-04,36,Étude d'événement TWFE (RMR×t) k = −8…8 : coefficients et ET en grappes du tableau 36.4 ; pente pré (k −7…−2) 0.0011,analysis/python/cases/ch36_panel.py,analysis/python/tests/test_ch36.py::test_event_study,"k=3 : 0.0410 (0.0138) ; k≥8 : 0.0817 (0.0270)",idem,verifie,2026-10-316CALC-36-05,36,Décomposition de Goodman-Bacon cas 4 : poids 0.6658/0.2721/0.0476/0.0146 ; β 0.0844/0.1224/0.0080/0.0403 ; Σ s·β = TWFE = 0.0905 (identité 1e-10),analysis/python/cases/ch36_panel.py,analysis/python/tests/test_ch36.py::test_bacon_cas4,"Σ poids = 1 ; pondéré = TWFE",idem,verifie,2026-10-317CALC-36-06,36,Callaway–Sant'Anna (témoins même RMR) : A 0.0528 (boot 0.0154) ; B 0.0406 (0.0183) ; global 0.0481 (0.0116) ; variante tous témoins A 0.0509 / B 0.0521 / global 0.0514,analysis/python/cases/ch36_panel.py,analysis/python/tests/test_ch36.py::test_callaway_santanna,"0.0481 (vrai 0.0454)",idem,verifie,2026-10-318CALC-36-07,36,Contrôle synthétique cohorte A : effet post 0.0363 (vrai 0.0484) ; RMSPE pré 0.0031 / post 0.0397 ; ratio 12.9 rang 1/36 ; 11 poids > 0.5 % (P02 0.142 … P05 0.022),analysis/python/cases/ch36_panel.py,analysis/python/tests/test_ch36.py::test_controle_synthetique_cas4,"0.0363 ; p = 1/36",idem,verifie,2026-10-319CALC-36-08,36,Discontinuité spatiale (simulation transversale, saut vrai 0.05) : h 0.5 → 0.034 (0.034 ; n 93) ; h 1 → 0.0316 (0.0247 ; 202) ; h 2 → 0.058 (0.0183 ; 400) ; naïf 0.1136,analysis/python/cases/ch36_panel.py,analysis/python/tests/test_ch36.py::test_discontinuite,"toutes à < 1 ET de 0.05",idem,verifie,2026-10-3110CALC-36-09,36,Laboratoire (effets croissants) : TWFE 0.0233 (0.0061) vs vrai 0.0692 ; comparaison interdite −0.0356 (poids 0.3265) ; CS 0.0601 (0.0073) ; cohortes 0.0893 / 0.0310,analysis/python/cases/ch36_panel.py,analysis/python/tests/test_ch36.py::test_laboratoire,"TWFE à > 7 ET de la vérité",idem,verifie,2026-10-3111CALC-36-10,36,Sensibilité tendances parallèles violées (+0.01/an) : 2×2 0.0734 (= 0.0534 + 0.02) ; TWFE 0.1227 ; pente pré 0.0036 ; TWFE + tendance 0.0609 ; 2×2 détendancé 0.0426 ; témoins : 0.0468/0.0534/0.0445/0.0308,analysis/python/cases/ch36_panel.py,analysis/python/tests/test_ch36.py::test_sensibilite_et_temoins,"biais 2×2 = 0.02 exactement",idem,verifie,2026-10-3112CALC-36-11,36,Solutions : ex1 0.05 → 5.13 % ; ex2 within 0.20 vs MCO 0.425 ; ex3 4.64/6.16/2.12 et ρ implicite 0.56 ; ex5 poids 0.5/0.5 et TWFE 0.005 vs ATT 0.0675 ; ex6 0.045/0.0817/−0.015 ; ex7 variantes SC 0.0452 (Montréal) / 0.0375 (hors) / placebo −0.0055,analysis/python/cases/ch36_panel.py,analysis/python/tests/test_ch36.py::test_solutions_exercices + test_solution_exercice_7,voir solutions/ch36.tex,idem,verifie,2026-10-3113CALC-36-12,36,Fonctions génériques traite.panel : within = indicatrices (1e-8) ; grappes unitaires = HC1 ; TWFE retrouve δ constant ; event study retrouve profil ; Bacon identité + CS corrige TWFE biaisé ; SC retrouve poids exacts (0.5/0.3/0.2) ; RDD retrouve saut 0.1,analysis/python/traite/panel.py,analysis/python/tests/test_ch36.py (7 tests génériques),tolérances déclarées,idem,verifie,2026-10-3114