1 Introduction

Jun Shao, Eric Slud, Yang Cheng, Sheng Wang et Carma Hogue

Précédent | Suivant

L’Annual Survey of Public Employment and Payroll (ASPEP) des États-Unis fournit des estimations courantes de l’emploi et de la rémunération à temps plein et à temps partiel dans les administrations publiques d’État et locales par fonction (par exemple, enseignement primaire et secondaire, enseignement supérieur, services de police, services de protection contre l’incendie, administration financière, services judiciaires et juridiques, etc.). Cette enquête a pour champ d’observation les administrations publiques d’État et locales (89 526 selon le Census of Governments de 2007), qui englobent les comtés, les villes, les cantons, les administrations appelées « districts spéciaux » et les districts scolaires. L’ASPEP, qui est la seule source de données sur l’emploi dans le secteur public par fonction administrative et catégorie d’emploi, fournit des données sur le nombre et la rémunération des employés à temps plein et à temps partiel, ainsi que le nombre d’heures travaillées par les employés à temps partiel. Habituellement, la collecte des données débute en mars et se poursuit pendant environ sept mois, en prenant la période de paye incluant le 12 mars comme période de référence.

Soit U MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadwfaaa a@39D6@  la population finie de N MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaad6eaaa a@39CF@  unités subdivisée en H MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadIeaaa a@39C9@  strates, U 1 ,, U H , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadwfada WgaaWcbaGaaGymaaqabaGccaaISaGaeSOjGSKaaGilaiaadwfadaWg aaWcbaGaamisaaqabaGccaGGSaaaaa@3FE2@  où U h MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadwfada WgaaWcbaGaamiAaaqabaaaaa@3AEF@  contient N h MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaad6eada WgaaWcbaGaamiAaaqabaaaaa@3AE8@  unités et N 1 ++ N H =N. MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaad6eada WgaaWcbaGaaGymaaqabaGccqGHRaWkcqWIVlctcqGHRaWkcaWGobWa aSbaaSqaaiaadIeaaeqaaOGaeyypa0JaamOtaiaac6caaaa@42D3@  Le plan de sondage habituel de l’ASPEP est un plan avec probabilité proportionnelle à la taille (PPT), où les strates sont construites en se basant sur l’État et le type d’administration publique, à savoir le comté, le sous-comté (grande ou petite ville), le district spécial ou le district scolaire. La taille de chaque unité (administration publique) est mesurée par la masse salariale totale, et l’échantillonnage est effectué indépendamment dans les diverses strates. En 2009, on a élaboré un plan d’échantillonnage modifié, qui comprend la division de certaines strates U h MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadwfada WgaaWcbaGaamiAaaqabaaaaa@3AEF@  en deux sous-strates, U h1 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadwfada WgaaWcbaGaamiAaiaaigdaaeqaaaaa@3BAA@  et U h2 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadwfada WgaaWcbaGaamiAaiaaikdaaeqaaaaa@3BAB@  contenant N h1 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaad6eada WgaaWcbaGaamiAaiaaigdaaeqaaaaa@3BA3@  et N h2 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaad6eada WgaaWcbaGaamiAaiaaikdaaeqaaaaa@3BA4@  unités, respectivement, où U h1 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadwfada WgaaWcbaGaamiAaiaaigdaaeqaaaaa@3BAA@  contient les unités de petite taille (Cheng et coll. 2009). L’idée était d’économiser des ressources et de réduire le fardeau de réponse en sélectionnant dans U h1 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadwfada WgaaWcbaGaamiAaiaaigdaaeqaaaaa@3BAA@  un échantillon plus petit sous le plan modifié que sous le plan habituel. Soit S hj MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadofada WgaaWcbaGaamiAaiaadQgaaeqaaaaa@3BDC@  un échantillon PPT de taille n hj MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaad6gada WgaaWcbaGaamiAaiaadQgaaeqaaaaa@3BF7@  provenant de U hj , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadwfada WgaaWcbaGaamiAaiaadQgaaeqaaOGaaiilaaaa@3C98@   j=1,2, MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadQgacq GH9aqpcaaIXaGaaiilaiaaikdacaGGSaaaaa@3DC8@   n h1 + n h2 = n h . MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaad6gada WgaaWcbaGaamiAaiaaigdaaeqaaOGaey4kaSIaamOBamaaBaaaleaa caWGObGaaGOmaaqabaGccqGH9aqpcaWGUbWaaSbaaSqaaiaadIgaae qaaOGaaiOlaaaa@434F@  Notons que n h1 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaad6gada WgaaWcbaGaamiAaiaaigdaaeqaaaaa@3BC3@  peut encore être plus grand que n h2 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaad6gada WgaaWcbaGaamiAaiaaikdaaeqaaaaa@3BC4@  , parce que N h1 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaad6eada WgaaWcbaGaamiAaiaaigdaaeqaaaaa@3BA3@  est habituellement beaucoup plus grand que N h2 . MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaad6eada WgaaWcbaGaamiAaiaaikdaaeqaaOGaaiOlaaaa@3C60@

Pour l’unité iU, MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadMgacq GHiiIZcaWGvbGaaiilaaaa@3CF8@  soit y i MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadMhada WgaaWcbaGaamyAaaqabaaaaa@3B14@  une variable étudiée clé (p. ex., l’emploi à temps plein, la rémunération à temps plein, l’emploi à temps partiel, la rémunération à temps partiel, les heures travaillées à temps partiel), x i MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadIhada WgaaWcbaGaamyAaaqabaaaaa@3B13@  une variable auxiliaire, disons la même variable que y i MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadMhada WgaaWcbaGaamyAaaqabaaaaa@3B14@  provenant du recensement le plus récent, et soit z i MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadQhada WgaaWcbaGaamyAaaqabaaaaa@3B15@  la covariable utilisée comme variable de taille dans l’échantillonnage PPT. Les valeurs des covariables x i MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadIhada WgaaWcbaGaamyAaaqabaaaaa@3B13@  et z i MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadQhada WgaaWcbaGaamyAaaqabaaaaa@3B15@  sont observées pour tout iU, MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadMgacq GHiiIZcaWGvbGaaiilaaaa@3CF8@  tandis que y i MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadMhada WgaaWcbaGaamyAaaqabaaaaa@3B14@  est observée uniquement pour chaque unité i échantillonnée.

L’estimateur de Horvitz-Thompson du total inconnu Y= iU y i MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadMfacq GH9aqpdaaeqaqabSqaaiaadMgacqGHiiIZcaWGvbaabeqdcqGHris5 aOGaaGPaVlaadMhadaWgaaWcbaGaamyAaaqabaaaaa@43BD@  est

Y ^ HT = h j i S hj y i / π i ,      (1.1) MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiqadMfaga qcamaaBaaaleaacaqGibGaaeivaaqabaGccqGH9aqpdaaeqbqabSqa aiaadIgaaeqaniabggHiLdGcdaaeqbqabSqaaiaadQgaaeqaniabgg HiLdGcdaaeqbqabSqaaiaadMgacqGHiiIZcaWGtbWaaSbaaeaacaWG ObGaamOAaaqabaaabeqdcqGHris5aOWaaSGbaeaacaWG5bWaaSbaaS qaaiaadMgaaeqaaaGcbaGaeqiWda3aaSbaaSqaaiaadMgaaeqaaaaa kiaaiYcacaWLjaGaaCzcaiaaxMaacaGGOaGaaGymaiaac6cacaaIXa Gaaiykaaaa@55AB@

π i MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiabec8aWn aaBaaaleaacaWGPbaabeaaaaa@3BD3@  est la probabilité d’inclusion d’ordre un de l’unité i MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadMgaaa a@39EA@  dans S hj , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadofada WgaaWcbaGaamiAaiaadQgaaeqaaOGaaiilaaaa@3C96@  une fonction connue des z i . MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadQhada WgaaWcbaGaamyAaaqabaGccaGGUaaaaa@3BD1@  Pour utiliser la variable auxiliaire x i MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadIhada WgaaWcbaGaamyAaaqabaaaaa@3B13@  et accroître la précision de l’estimation de Y, MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadMfaca GGSaaaaa@3A8A@  l’approche assistée par modèle (Särndal, Swensson et Wretman 1992) a été adoptée. L’application de la régression dans chaque échantillon S hj MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadofada WgaaWcbaGaamiAaiaadQgaaeqaaaaa@3BDC@  conduit à l’estimateur par la régression de Y MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadMfaaa a@39DA@  de la forme

Y ^ reg,2 = h j [ N hj Y ^ hj N ^ hj + β ^ hj ( X hj N hj X ^ hj N ^ hj ) ],      (1.2) MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiqadMfaga qcamaaBaaaleaacaqGYbGaaeyzaiaabEgacaaISaGaaGOmaaqabaGc cqGH9aqpdaaeqbqabSqaaiaadIgaaeqaniabggHiLdGcdaaeqbqabS qaaiaadQgaaeqaniabggHiLdGcdaWadaqaamaalaaabaGaamOtamaa BaaaleaacaWGObGaamOAaaqabaGcceWGzbGbaKaadaWgaaWcbaGaam iAaiaadQgaaeqaaaGcbaGabmOtayaajaWaaSbaaSqaaiaadIgacaWG QbaabeaaaaGccqGHRaWkcuaHYoGygaqcamaaBaaaleaacaWGObGaam OAaaqabaGcdaqadaqaaiaadIfadaWgaaWcbaGaamiAaiaadQgaaeqa aOGaeyOeI0YaaSaaaeaacaWGobWaaSbaaSqaaiaadIgacaWGQbaabe aakiqadIfagaqcamaaBaaaleaacaWGObGaamOAaaqabaaakeaaceWG obGbaKaadaWgaaWcbaGaamiAaiaadQgaaeqaaaaaaOGaayjkaiaawM caaaGaay5waiaaw2faaiaaiYcacaWLjaGaaCzcaiaaxMaacaGGOaGa aGymaiaac6cacaaIYaGaaiykaaaa@6984@

X hj = i U hj x i , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadIfada WgaaWcbaGaamiAaiaadQgaaeqaaOGaeyypa0ZaaabeaeqaleaacaWG PbGaeyicI4SaamyvamaaBaaabaGaamiAaiaadQgaaeqaaaqab0Gaey yeIuoakiaaykW7caWG4bWaaSbaaSqaaiaadMgaaeqaaOGaaiilaaaa @4884@   Y ^ hj = i S hj y i / π i , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiqadMfaga qcamaaBaaaleaacaWGObGaamOAaaqabaGccqGH9aqpdaaeqaqabSqa aiaadMgacqGHiiIZcaWGtbWaaSbaaeaacaWGObGaamOAaaqabaaabe qdcqGHris5aOWaaSGbaeaacaWG5bWaaSbaaSqaaiaadMgaaeqaaaGc baGaeqiWda3aaSbaaSqaaiaadMgaaeqaaaaakiaacYcaaaa@4A00@   X ^ hj = i S hj x i / π i , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiqadIfaga qcamaaBaaaleaacaWGObGaamOAaaqabaGccqGH9aqpdaaeqaqabSqa aiaadMgacqGHiiIZcaWGtbWaaSbaaeaacaWGObGaamOAaaqabaaabe qdcqGHris5aOWaaSGbaeaacaWG4bWaaSbaaSqaaiaadMgaaeqaaaGc baGaeqiWda3aaSbaaSqaaiaadMgaaeqaaaaakiaacYcaaaa@49FE@   N ^ hj = i S hj 1/ π i , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiqad6eaga qcamaaBaaaleaacaWGObGaamOAaaqabaGccqGH9aqpdaaeqaqabSqa aiaadMgacqGHiiIZcaWGtbWaaSbaaeaacaWGObGaamOAaaqabaaabe qdcqGHris5aOWaaSGbaeaacaaIXaaabaGaeqiWda3aaSbaaSqaaiaa dMgaaeqaaaaakiaacYcaaaa@488E@  et

β ^ hj = i S hj ( x i X ^ hj / N ^ hj ) y i / π i i S hj ( x i X ^ hj / N ^ hj ) 2 / π i . MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiqbek7aIz aajaWaaSbaaSqaaiaadIgacaWGQbaabeaakiabg2da9maalaaabaWa aabeaeaadaqadaqaaiaadIhadaWgaaWcbaGaamyAaaqabaGccqGHsi sldaWcgaqaaiqadIfagaqcamaaBaaaleaacaWGObGaamOAaaqabaaa keaaceWGobGbaKaadaWgaaWcbaGaamiAaiaadQgaaeqaaaaaaOGaay jkaiaawMcaamaalyaabaGaamyEamaaBaaaleaacaWGPbaabeaaaOqa aiabec8aWnaaBaaaleaacaWGPbaabeaaaaaabaGaamyAaiabgIGiol aadofadaWgaaqaaiaadIgacaWGQbaabeaaaeqaniabggHiLdaakeaa daaeqaqaamaalyaabaWaaeWaaeaacaWG4bWaaSbaaSqaaiaadMgaae qaaOGaeyOeI0YaaSGbaeaaceWGybGbaKaadaWgaaWcbaGaamiAaiaa dQgaaeqaaaGcbaGabmOtayaajaWaaSbaaSqaaiaadIgacaWGQbaabe aaaaaakiaawIcacaGLPaaadaahaaWcbeqaaiaaikdaaaaakeaacqaH apaCdaWgaaWcbaGaamyAaaqabaaaaaqaaiaadMgacqGHiiIZcaWGtb WaaSbaaeaacaWGObGaamOAaaqabaaabeqdcqGHris5aaaakiaai6ca aaa@6B10@

Autrement, la combinaison des deux sous-strates S h1 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadofada WgaaWcbaGaamiAaiaaigdaaeqaaaaa@3BA8@  et S h2 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadofada WgaaWcbaGaamiAaiaaikdaaeqaaaaa@3BA9@  donne l’estimateur par la régression suivant. (Un examinateur fait remarquer correctement que Y ^ reg,1 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiqadMfaga qcamaaBaaaleaacaqGYbGaaeyzaiaabEgacaaISaGaaGymaaqabaaa aa@3E4E@  dans (1.3) n’est pas l’estimateur groupé que l’on utiliserait si les droites de régression dans la strate h MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadIgaaa a@39E9@  étaient combinées mais que les deux sous-strates ne l’étaient pas; cependant, il est l’estimateur naturel lorsque non seulement les droites de régression, mais aussi les sous-strates sont combinées.) 

Y ^ reg,1 = h [ N h Y ^ h N ^ h + β ^ h ( X h N h X ^ h N ^ h ) ],      (1.3) MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiqadMfaga qcamaaBaaaleaacaqGYbGaaeyzaiaabEgacaaISaGaaGymaaqabaGc cqGH9aqpdaaeqbqabSqaaiaadIgaaeqaniabggHiLdGcdaWadaqaam aalaaabaGaamOtamaaBaaaleaacaWGObaabeaakiqadMfagaqcamaa BaaaleaacaWGObaabeaaaOqaaiqad6eagaqcamaaBaaaleaacaWGOb aabeaaaaGccqGHRaWkcuaHYoGygaqcamaaBaaaleaacaWGObaabeaa kmaabmaabaGaamiwamaaBaaaleaacaWGObaabeaakiabgkHiTmaala aabaGaamOtamaaBaaaleaacaWGObaabeaakiqadIfagaqcamaaBaaa leaacaWGObaabeaaaOqaaiqad6eagaqcamaaBaaaleaacaWGObaabe aaaaaakiaawIcacaGLPaaaaiaawUfacaGLDbaacaaISaGaaCzcaiaa xMaacaWLjaGaaiikaiaaigdacaGGUaGaaG4maiaacMcaaaa@5EEF@

Y ^ h = j Y ^ hj , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiqadMfaga qcamaaBaaaleaacaWGObaabeaakiabg2da9maaqababeWcbaGaamOA aaqab0GaeyyeIuoakiaaykW7ceWGzbGbaKaadaWgaaWcbaGaamiAai aadQgaaeqaaOGaaiilaaaa@442B@   X ^ h = j X ^ hj , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiqadIfaga qcamaaBaaaleaacaWGObaabeaakiabg2da9maaqababeWcbaGaamOA aaqab0GaeyyeIuoakiaaykW7ceWGybGbaKaadaWgaaWcbaGaamiAai aadQgaaeqaaOGaaiilaaaa@4429@   N ^ h = j N ^ hj , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiqad6eaga qcamaaBaaaleaacaWGObaabeaakiabg2da9maaqababeWcbaGaamOA aaqab0GaeyyeIuoakiaaykW7ceWGobGbaKaadaWgaaWcbaGaamiAai aadQgaaeqaaOGaaiilaaaa@4415@  et

β ^ h = j i S hj ( x i X ^ h / N ^ h ) y i / π i j i S hj ( x i X ^ h / N ^ h ) 2 / π i . MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiqbek7aIz aajaWaaSbaaSqaaiaadIgaaeqaaOGaeyypa0ZaaSaaaeaadaaeqaqa amaaqababaWaaeWaaeaacaWG4bWaaSbaaSqaaiaadMgaaeqaaOGaey OeI0YaaSGbaeaaceWGybGbaKaadaWgaaWcbaGaamiAaaqabaaakeaa ceWGobGbaKaadaWgaaWcbaGaamiAaaqabaaaaaGccaGLOaGaayzkaa WaaSGbaeaacaWG5bWaaSbaaSqaaiaadMgaaeqaaaGcbaGaeqiWda3a aSbaaSqaaiaadMgaaeqaaaaaaeaacaWGPbGaeyicI4Saam4uamaaBa aabaGaamiAaiaadQgaaeqaaaqab0GaeyyeIuoaaSqaaiaadQgaaeqa niabggHiLdaakeaadaaeqaqaamaaqababaWaaSGbaeaadaqadaqaai aadIhadaWgaaWcbaGaamyAaaqabaGccqGHsisldaWcgaqaaiqadIfa gaqcamaaBaaaleaacaWGObaabeaaaOqaaiqad6eagaqcamaaBaaale aacaWGObaabeaaaaaakiaawIcacaGLPaaadaahaaWcbeqaaiaaikda aaaakeaacqaHapaCdaWgaaWcbaGaamyAaaqabaaaaaqaaiaadMgacq GHiiIZcaWGtbWaaSbaaeaacaWGObGaamOAaaqabaaabeqdcqGHris5 aaWcbaGaamOAaaqab0GaeyyeIuoaaaGccaaIUaaaaa@6C09@

Puisque Y ^ reg,1 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiqadMfaga qcamaaBaaaleaacaqGYbGaaeyzaiaabEgacaaISaGaaGymaaqabaaa aa@3E4E@  ainsi que Y ^ reg,2 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiqadMfaga qcamaaBaaaleaacaqGYbGaaeyzaiaabEgacaaISaGaaGOmaaqabaaa aa@3E4F@  sont des estimateurs assistés par modèle, ils sont convergents sous échantillonnage répété, que le modèle de régression soit ou non vérifié. Si les droites de régression par les moindres carrés dans les deux sous-strates U hj MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadwfada WgaaWcbaGaamiAaiaadQgaaeqaaaaa@3BDE@  sont les mêmes, Y ^ reg,1 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiqadMfaga qcamaaBaaaleaacaqGYbGaaeyzaiaabEgacaaISaGaaGymaaqabaaa aa@3E4E@  peut être plus efficace que Y ^ reg,2 . MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiqadMfaga qcamaaBaaaleaacaqGYbGaaeyzaiaabEgacaaISaGaaGOmaaqabaGc caGGUaaaaa@3F0B@  Par ailleurs, si les droites de régression sont différentes, Y ^ reg,2 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiqadMfaga qcamaaBaaaleaacaqGYbGaaeyzaiaabEgacaaISaGaaGOmaaqabaaa aa@3E4F@  peut être plus efficace que Y ^ reg,1 . MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiqadMfaga qcamaaBaaaleaacaqGYbGaaeyzaiaabEgacaaISaGaaGymaaqabaGc caGGUaaaaa@3F0A@

Cheng et coll. (2010) ont proposé une méthode fondée sur un test de décision qui consiste à appliquer un test d’hypothèse pour décider s’il faut combiner S h1 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadofada WgaaWcbaGaamiAaiaaigdaaeqaaaaa@3BA8@  et S h2 . MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadofada WgaaWcbaGaamiAaiaaikdaaeqaaOGaaiOlaaaa@3C65@  À l’intérieur de la strate h, MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadIgaca GGSaaaaa@3A99@  on teste l’hypothèse d’égalité des pentes des droites de régression dans U h1 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadwfada WgaaWcbaGaamiAaiaaigdaaeqaaaaa@3BAA@  et U h2 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadwfada WgaaWcbaGaamiAaiaaikdaaeqaaaaa@3BAB@  . Soit

α ^ hj = Y ^ hj β ^ hj X ^ hj N ^ hj , σ ^ xe,hj 2 = n hj N ^ hj 2 i S hj ( x i X ^ hj N ^ hj ) 2 ( y i α ^ hj β ^ hj x i ) 2 π i 2 , σ ^ xhj 2 = i S hj ( x i X ^ hj / N ^ hj ) 2 π i N ^ hj , t h = n h 4 ( β ^ h1 β ^ h2 )/ n h j=1 2 σ ^ xe,hj 2 n hj σ ^ xhj 4 . MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9 vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=x fr=xb9adbaqaaeGacaGaaiaabeqaamaabaabaaGcbaqbaeaaciWaaa uaaiqbeg7aHzaajaWaaSbaaSqaaiaadIgacaWGQbaabeaaaOqaaiab g2da9aqaamaalaaabaGabmywayaajaWaaSbaaSqaaiaadIgacaWGQb aabeaakiabgkHiTiqbek7aIzaajaWaaSbaaSqaaiaadIgacaWGQbaa beaakiqadIfagaqcamaaBaaaleaacaWGObGaamOAaaqabaaakeaace WGobGbaKaadaWgaaWcbaGaamiAaiaadQgaaeqaaaaakiaaiYcacuaH dpWCgaqcamaaDaaaleaacaWG4bGaamyzaiaaiYcacaWGObGaamOAaa qaaiaaikdaaaGccqGH9aqpdaWcaaqaaiaad6gadaWgaaWcbaGaamiA aiaadQgaaeqaaaGcbaGabmOtayaajaWaa0baaSqaaiaadIgacaWGQb aabaGaaGOmaaaaaaGcdaaeqbqabSqaaiaadMgacqGHiiIZcaWGtbWa aSbaaeaacaWGObGaamOAaaqabaaabeqdcqGHris5aOWaaeWaaeaaca WG4bWaaSbaaSqaaiaadMgaaeqaaOGaeyOeI0YaaSaaaeaaceWGybGb aKaadaWgaaWcbaGaamiAaiaadQgaaeqaaaGcbaGabmOtayaajaWaaS baaSqaaiaadIgacaWGQbaabeaaaaaakiaawIcacaGLPaaadaahaaWc beqaaiaaikdaaaGcdaWcaaqaamaabmaabaGaamyEamaaBaaaleaaca WGPbaabeaakiabgkHiTiqbeg7aHzaajaWaaSbaaSqaaiaadIgacaWG QbaabeaakiabgkHiTiqbek7aIzaajaWaaSbaaSqaaiaadIgacaWGQb aabeaakiaadIhadaWgaaWcbaGaamyAaaqabaaakiaawIcacaGLPaaa daahaaWcbeqaaiaaikdaaaaakeaacqaHapaCdaqhaaWcbaGaamyAaa qaaiaaikdaaaaaaOGaaGilaaqaaiqbeo8aZzaajaWaa0baaSqaaiaa dIhacaWGObGaamOAaaqaaiaaikdaaaaakeaacqGH9aqpaeaadaaeqb qabSqaaiaadMgacqGHiiIZcaWGtbWaaSbaaeaacaWGObGaamOAaaqa baaabeqdcqGHris5aOWaaSaaaeaadaqadaqaaiaadIhadaWgaaWcba GaamyAaaqabaGccqGHsisldaWcgaqaaiqadIfagaqcamaaBaaaleaa caWGObGaamOAaaqabaaakeaaceWGobGbaKaadaWgaaWcbaGaamiAai aadQgaaeqaaaaaaOGaayjkaiaawMcaamaaCaaaleqabaGaaGOmaaaa aOqaaiabec8aWnaaBaaaleaacaWGPbaabeaakiqad6eagaqcamaaBa aaleaacaWGObGaamOAaaqabaaaaOGaaGilaiaadshadaWgaaWcbaGa amiAaaqabaGccqGH9aqpdaWcgaqaamaakaaabaGaamOBamaaBaaale aacaWGObaabeaakiabgkHiTiaaisdaaSqabaGccaaMc8+aaeWaaeaa cuaHYoGygaqcamaaBaaaleaacaWGObGaaGymaaqabaGccqGHsislcu aHYoGygaqcamaaBaaaleaacaWGObGaaGOmaaqabaaakiaawIcacaGL PaaaaeaadaGcaaqaaiaad6gadaWgaaWcbaGaamiAaaqabaGcdaaeWb qabSqaaiaadQgacqGH9aqpcaaIXaaabaGaaGOmaaqdcqGHris5aOWa aSaaaeaacuaHdpWCgaqcamaaDaaaleaacaWG4bGaamyzaiaaiYcaca WGObGaamOAaaqaaiaaikdaaaaakeaacaWGUbWaaSbaaSqaaiaadIga caWGQbaabeaakiqbeo8aZzaajaWaa0baaSqaaiaadIhacaWGObGaam OAaaqaaiaaisdaaaaaaaqabaaaaOGaaGOlaaaaaaa@CB1C@

Si | t h |> t 1τ/ 2, n h 4 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaamaaemaaba GaaGPaVlaadshadaWgaaWcbaGaamiAaaqabaaakiaawEa7caGLiWoa caaMe8UaaeOpaiaaysW7caWG0bWaaSbaaSqaamaalyaabaGaaGymai abgkHiTiabes8a0bqaaiaaikdacaaISaGaamOBamaaBaaabaGaamiA aaqabaGaeyOeI0IaaGinaaaaaeqaaaaa@4D66@  , où t 1τ/ 2,ν MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadshada WgaaWcbaWaaSGbaeaacaaIXaGaeyOeI0IaeqiXdqhabaGaaGOmaiaa iYcacqaH9oGBaaaabeaaaaa@40CE@  est le ( 1τ/2 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaamaalyaaba GaaGymaiabgkHiTiabes8a0bqaaiaaikdaaaaaaa@3D3B@  )e quantile de la distribution t avec ν MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiabe27aUb aa@3AB4@  degrés de liberté, alors nous rejetons l’hypothèse d’une pente commune et nous utilisons β ^ hj MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiqbek7aIz aajaWaaSbaaSqaaiaadIgacaWGQbaabeaaaaa@3CB5@  (et fixons ζ h =1 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiabeA7a6n aaBaaaleaacaWGObaabeaakiabg2da9iaaigdaaaa@3D9D@  ). Ici, τ MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiabes8a0b aa@3AC1@  est un seuil de signification nominal fixé par défaut à 0,05, mais nous considérerons d’autres choix de la valeur de τ MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiabes8a0b aa@3AC1@  à la section consacrée aux simulations. La définition de la statistique de test faisant intervenir n h 4 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaad6gada WgaaWcbaGaamiAaaqabaGccqGHsislcaaI0aaaaa@3CBD@  degrés de liberté est un choix légèrement artificiel conçu afin de rendre les probabilités de rejet d’un échantillon modéré plus proches de la valeur nominale, mais la théorie asymptotique en grand échantillon justifiant ce test est donnée à la partie (c) du théorème 1. Si | t h | t 1τ/ 2, n h 4 , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaamaaemaaba GaaGPaVlaadshadaWgaaWcbaGaamiAaaqabaaakiaawEa7caGLiWoa cqGHKjYOcaWG0bWaaSbaaSqaamaalyaabaGaaGymaiabgkHiTiabes 8a0bqaaiaaikdacaaISaGaamOBamaaBaaabaGaamiAaaqabaGaeyOe I0IaaGinaaaaaeqaaOGaaiilaaaa@4BFA@  alors nous acceptons l’hypothèse d’une pente commune, nous combinons les sous-strates S h1 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadofada WgaaWcbaGaamiAaiaaigdaaeqaaaaa@3BA8@  et S h2 , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadofada WgaaWcbaGaamiAaiaaikdaaeqaaOGaaiilaaaa@3C63@  et nous utilisons β ^ h MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiqbek7aIz aajaWaaSbaaSqaaiaadIgaaeqaaaaa@3BC6@   ( en fixant ζ h =0 ). MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaamaabmaaba Gaaeyzaiaab6gacaqGGaGaaeOzaiaabMgacaqG4bGaaeyyaiaab6ga caqG0bGaaGjbVlabeA7a6naaBaaaleaacaWGObaabeaakiabg2da9i aaicdaaiaawIcacaGLPaaacaGGUaaaaa@497C@  Les tests sont effectués de manière indépendante dans les diverses strates h=1,,H. MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadIgacq GH9aqpcaaIXaGaaiilaiablAciljaaiYcacaWGibGaaiOlaaaa@3FB1@  L’estimateur de Y MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadMfaaa a@39DA@  fondé sur le test de décision est alors

Y ^ dec = h j ζ h [ N hj Y ^ hj N ^ hj + β ^ hj ( X hj N hj X ^ hj N ^ hj ) ]+ h ( 1 ζ h )[ N h Y ^ h N ^ h + β ^ h ( X h N h X ^ h N ^ h ) ]. (1.4) MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiqadMfaga qcamaaBaaaleaacaqGKbGaaeyzaiaabogaaeqaaOGaeyypa0Zaaabu aeqaleaacaWGObaabeqdcqGHris5aOWaaabuaeqaleaacaWGQbaabe qdcqGHris5aOGaeqOTdO3aaSbaaSqaaiaadIgaaeqaaOWaamWaaeaa daWcaaqaaiaad6eadaWgaaWcbaGaamiAaiaadQgaaeqaaOGabmyway aajaWaaSbaaSqaaiaadIgacaWGQbaabeaaaOqaaiqad6eagaqcamaa BaaaleaacaWGObGaamOAaaqabaaaaOGaey4kaSIafqOSdiMbaKaada WgaaWcbaGaamiAaiaadQgaaeqaaOWaaeWaaeaacaWGybWaaSbaaSqa aiaadIgacaWGQbaabeaakiabgkHiTmaalaaabaGaamOtamaaBaaale aacaWGObGaamOAaaqabaGcceWGybGbaKaadaWgaaWcbaGaamiAaiaa dQgaaeqaaaGcbaGabmOtayaajaWaaSbaaSqaaiaadIgacaWGQbaabe aaaaaakiaawIcacaGLPaaaaiaawUfacaGLDbaacqGHRaWkdaaeqbqa bSqaaiaadIgaaeqaniabggHiLdGcdaqadaqaaiaaigdacqGHsislcq aH2oGEdaWgaaWcbaGaamiAaaqabaaakiaawIcacaGLPaaadaWadaqa amaalaaabaGaamOtamaaBaaaleaacaWGObaabeaakiqadMfagaqcam aaBaaaleaacaWGObaabeaaaOqaaiqad6eagaqcamaaBaaaleaacaWG ObaabeaaaaGccqGHRaWkcuaHYoGygaqcamaaBaaaleaacaWGObaabe aakmaabmaabaGaamiwamaaBaaaleaacaWGObaabeaakiabgkHiTmaa laaabaGaamOtamaaBaaaleaacaWGObaabeaakiqadIfagaqcamaaBa aaleaacaWGObaabeaaaOqaaiqad6eagaqcamaaBaaaleaacaWGObaa beaaaaaakiaawIcacaGLPaaaaiaawUfacaGLDbaacaaIUaGaaCzcai aaxMaacaWLjaGaaiikaiaaigdacaGGUaGaaGinaiaacMcaaaa@8B49@

Puisque les deux droites de régression ayant une pente commune peuvent avoir des ordonnées à l’origine différentes, on pourrait tester une hypothèse supplémentaire concernant les ordonnées à l’origine pour décider s’il faut combiner les deux sous-strates. Cependant, des points de population ( x i , y i ) MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaamaabmaaba GaamiEamaaBaaaleaacaWGPbaabeaakiaaiYcacaWG5bWaaSbaaSqa aiaadMgaaeqaaaGccaGLOaGaayzkaaaaaa@3F7E@  se trouvant sur deux droites de régression de sous-strate parallèles, mais non identiques seraient discontinus autour du seuil entre les deux sous-strates U h1 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadwfada WgaaWcbaGaamiAaiaaigdaaeqaaaaa@3BAA@  et U h2 , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadwfada WgaaWcbaGaamiAaiaaikdaaeqaaOGaaiilaaaa@3C65@  ce qui ne semble ne se produire que rarement dans les situations pratiques. Par exemple, dans l’ASPEP, Cheng et coll. (2010) ont étudié les pentes et les ordonnées à l’origine de sous-strates dans les ensembles de donnés des recensements des administrations publiques de 2002 et de 2007, et ont constaté que l’hypothèse d’une ordonnée à l’origine commune ne pouvait jamais être rejetée lorsque l’hypothèse d’une pente commune ne pouvait pas l’être. Donc, l’estimateur fondé sur un test de décision donné dans (1.4) dépend uniquement du test de l’hypothèse d’égalité des pentes des droites de régression des sous-strates.

Les estimateurs à deux degrés étudiés ici sont des cas particuliers de procédures nommées antérieurement estimateurs après un test préliminaire (preliminary test estimators). Il existe une littérature abondante traitant de l’utilisation de ce genre de procédures dans les enquêtes, y compris une bibliographie de Bancroft et Han (1977), un livre publié par Saleh (2006) et un traitement proposé par Fuller (2009, section 6.7). Une idée de Saleh (2006) consiste à estimer les coefficients par une combinaison convexe des coefficients estimés à partir des strates distinctes en faisant dépendre les proportions d’une statistique de test. Les estimateurs lissés de ce genre pourraient être plus efficaces que nos procédures fondées sur un test de décision. Si les ordonnées à l’origine et les pentes propres aux strates étaient considérées comme aléatoires, on pourrait aussi essayer d’appliquer à l’estimation une approche bayésienne empirique fondée sur un modèle.

Les estimateurs fondés sur un test de décision (1.4) sont nouveaux, parce qu’ils sont assistés par modèle et convergents sous le plan dans le contexte des sondages, et utilisent explicitement les tailles de population de sous-strate connues. Dans un esprit à peu près semblable, Rao et Ramachandran (1974) avaient effectué antérieurement une comparaison exacte des estimateurs par le ratio distincts et combinés sous un modèle de ratio similaire au modèle de régression considéré dans le présent article.

L’objectif de l’article est d’illustrer certaines propriétés asymptotiques et empiriques des estimateurs de Y MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaadMfaaa a@39DA@  décrits plus haut et des estimateurs de leur variance. La convergence et la normalité asymptotique de Y ^ reg,1 , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiqadMfaga qcamaaBaaaleaacaqGYbGaaeyzaiaabEgacaaISaGaaGymaaqabaGc caGGSaaaaa@3F08@   Y ^ reg,2 , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiqadMfaga qcamaaBaaaleaacaqGYbGaaeyzaiaabEgacaaISaGaaGOmaaqabaGc caGGSaaaaa@3F09@  et Y ^ dec MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiqadMfaga qcamaaBaaaleaacaqGKbGaaeyzaiaabogaaeqaaaaa@3CCB@  sont établies à la section 2, dans le contexte de la théorie asymptotique fondée sur le plan de sondage ou assistée par modèle. Bien que les résultats asymptotiques d’ordre un favorisent Y ^ reg,2 , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiqadMfaga qcamaaBaaaleaacaqGYbGaaeyzaiaabEgacaaISaGaaGOmaaqabaGc caGGSaaaaa@3F09@   Y ^ reg,1 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiqadMfaga qcamaaBaaaleaacaqGYbGaaeyzaiaabEgacaaISaGaaGymaaqabaaa aa@3E4E@  pourrait être meilleur quand certaines tailles d’échantillon de sous-strate n h2 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiaad6gada WgaaWcbaGaamiAaiaaikdaaeqaaaaa@3BC4@  sont modérées, un effet asymptotique d’ordre deux. L’avantage de l’estimateur fondé sur un test de décision Y ^ dec MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiqadMfaga qcamaaBaaaleaacaqGKbGaaeyzaiaabogaaeqaaaaa@3CCB@  tient à l’adaptation en vue d’être proche de Y ^ reg,1 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiqadMfaga qcamaaBaaaleaacaqGYbGaaeyzaiaabEgacaaISaGaaGymaaqabaaa aa@3E4E@  ou de Y ^ reg,2 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9LqFf0x e9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qqaq=Jf9sr 0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaaiqadMfaga qcamaaBaaaleaacaqGYbGaaeyzaiaabEgacaaISaGaaGOmaaqabaaa aa@3E4F@  selon celui qui est le meilleur. Comme l’indique la discussion du paragraphe (III) de la section 4.4, les simulations montrent que l’avantage de cette adaptabilité est de réduire l’EQM d’une quantité allant jusqu’à quelques pour cent sous des conditions de paramétrisation raisonnables, et de plus grandes quantités sous des conditions plus étranges.

L’estimation de la variance de l’estimateur fondé sur un test de décision est traitée à la section 3. Même si la théorie asymptotique exposée à la section 2 laisse entendre que des estimateurs convergents de variance sont obtenus par substitution des quantités inconnues dans les formules de variance asymptotique, nous étudions aussi les estimateurs bootstrap de la variance proposés dans Cheng et coll. (2010), qui ont généralement de meilleures propriétés en échantillon fini que les estimateurs par substitution. Les résultats empiriques sont présentés à la section 4, les interprétations et les conclusions étant formulées à la sous-section 4.4. Toutes les preuves techniques sont données en annexe.

Précédent | Suivant

Date de modification :