4 Méthode bayésienne linéaire pour données catégoriques

Kelly Cristina M. Gonçalves, Fernando A. S. Moura et Helio S. Migon

Précédent | Suivant

Il arrive souvent que l'on s'intéresse à des cas où la caractéristique observée est celle de savoir si l'unité de population possède ou non un certain attribut d'intérêt. Nous pouvons définir une variable dichotomique y i =1, MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadMhadaWgaaWcbaGaamyAaaqabaGccqGH9aqpcaaIXaGaaiilaaaa @3FAA@  si la i e MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadMgadaahaaWcbeqaaiaabwgaaaaaaa@3D1A@  unité possède cet attribut, ce qui est désigné comme une réussite, et y i =0 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadMhadaWgaaWcbaGaamyAaaqabaGccqGH9aqpcaaIWaaaaa@3EF9@  autrement. Pour le cas binaire lorsque la taille de l'échantillon n'est pas suffisamment grande pour appliquer le théorème central limite, l'approche fondée sur le plan de sondage pourrait faire appel à la randomisation introduite par le plan de sondage pour justifier la distribution des quantités aléatoires binaires. Par exemple, Cochran (1977), sections 3.4 et 3.5, montre comment appliquer les lois hypergéométrique et binomiale pour obtenir les intervalles de confiance pour les proportions de population quand on se sert de plans d'échantillonnages aléatoires simples avec et sans remise, respectivement. Par ailleurs, des approches dépendantes d'un modèle ont également été avancées et appliquées pour prédire les totaux ou les moyennes dans les catégories d'intérêt. Malec, Sedransk, Moriarity et LeClere (1997) ont considéré un modèle hiérarchique logistique à deux niveaux, où les grappes forment le deuxième niveau. Ils ont également comparé les estimations bayésiennes entièrement hiérarchiques aux estimations bayésiennes empiriques et aux méthodes classiques. Moura et Migon (2002) ont présenté une approche basée sur un modèle hiérarchique logistique pour la prédiction de proportions sur petits domaines, en tenant compte des effets spatiaux ainsi que des effets d'hétérogénéité non structurée possibles. Nandram et Choi (2008) ont proposé un modèle multinomial-Dirichlet dépendant du temps pour prédire les résultats d'une élection sous non-réponse ignorable et non ignorable. Ils ont également utilisé une approche bayésienne pour répartir les électeurs indécis entre les candidats.

De nouveau, ici, nous n'avons pas besoin d'utiliser des hypothèses au sujet du modèle complet ni une approche de randomisation, mais nous devons émettre certaines hypothèses au sujet des premier et deuxième moments des quantités aléatoires concernées. L'EBL pour les données binaires a été introduit brièvement par O'Hagan (1985), mais ici, nous le développons d'une manière plus générale pour le cas où nous nous intéressons à l'analyse de plus d'un attribut dans une population. L'objectif est de décrire l'estimation de la proportion de réussites avec des données catégoriques. Soit y ij MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadMhadaWgaaWcbaGaamyAaiaadQgaaeqaaaaa@3E1E@  la variable qui indique que l'unité i, MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadMgacaGGSaaaaa@3CB5@   i=1,,N MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadMgacqGH9aqpcaaIXaGaaGilaiablAciljaaiYcacaWGobaaaa@4127@  se trouve dans la catégorie j, MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadQgacaGGSaaaaa@3CB6@   j=1,,k MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadQgacqGH9aqpcaaIXaGaaGilaiablAciljaaiYcacaWGRbaaaa@4145@  donnée par

y ij ={ 1,sila i e unitépossèdele j e attribut; 0,autrement. MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadMhadaWgaaWcbaGaamyAaiaadQgaaeqaaOGaeyypa0Zaaiqaaeaa faqaaeGabaaabaGaaGymaiaaiYcacaaMi8UaaGjbVlaabohacaqGPb GaaGPaVlaaykW7caqGSbGaaeyyaiaaysW7caWGPbWaaWbaaSqabeaa caqGLbaaaOGaaGjbVlaabwhacaqGUbGaaeyAaiaabshacaqGPdGaaG jbVlaabchacaqGVbGaae4CaiaabohacaqGOdGaaeizaiaabwgacaaM c8UaaGPaVlaabYgacaqGLbGaaGjbVlaadQgadaahaaWcbeqaaiaabw gaaaGccaaMe8UaaeyyaiaabshacaqG0bGaaeOCaiaabMgacaqGIbGa aeyDaiaabshacaaI7aaabaGaaGimaiaaiYcacaaMi8UaaGjbVlaabg gacaqG1bGaaeiDaiaabkhacaqGLbGaaeyBaiaabwgacaqGUbGaaeiD aiaai6cacaaMe8UaaGjcVdaaaiaawUhaaaaa@81DA@

Le but principal est d'estimer un vecteur p=( p 1 ,, p k ), MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aahchacqGH9aqpdaqadaqaaiaadchadaWgaaWcbaGaaGymaaqabaGc caaISaGaeSOjGSKaaGilaiaadchadaWgaaWcbaGaam4Aaaqabaaaki aawIcacaGLPaaaiiaacqWFYaIOcqWFSaalaaa@478E@  où p j = N 1 i=1 N y ij , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadchadaWgaaWcbaGaamOAaaqabaGccqGH9aqpcaWGobWaaWbaaSqa beaacqGHsislcaaIXaaaaOWaaabmaeqaleaacaWGPbGaeyypa0JaaG ymaaqaaiaad6eaa0GaeyyeIuoakiaaykW7caWG5bWaaSbaaSqaaiaa dMgacaaMi8UaamOAaaqabaGccaGGSaaaaa@4D55@   j=1,,k, MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadQgacqGH9aqpcaaIXaGaaGilaiablAciljaaiYcacaWGRbGaaiil aaaa@41F5@  est la proportion d'unités dans la catégorie j, MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadQgacaGGSaaaaa@3CB6@  sachant y s , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aahMhadaWgaaWcbaGaam4CaaqabaGccaGGSaaaaa@3DF7@  un vecteur de dimension nk, MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aad6gacaWGRbGaaiilaaaa@3DAA@  défini comme étant y s =( y 11 , y 21 ,, y n1 ,, y 1k , y 2k ,, y nk ). MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aahMhadaWgaaWcbaGaam4CaaqabaGccqGH9aqpdaqadaqaaiaadMha daWgaaWcbaGaaGymaiaaigdaaeqaaOGaaGilaiaadMhadaWgaaWcba GaaGOmaiaaigdaaeqaaOGaaGilaiablAciljaaiYcacaWG5bWaaSba aSqaaiaad6gacaaIXaaabeaakiaaiYcacqWIMaYscaaISaGaamyEam aaBaaaleaacaaIXaGaam4AaaqabaGccaaISaGaamyEamaaBaaaleaa caaIYaGaam4AaaqabaGccaaISaGaeSOjGSKaaGilaiaadMhadaWgaa WcbaGaamOBaiaadUgaaeqaaaGccaGLOaGaayzkaaaccaGae8NmGiQa e8Nla4caaa@5C5D@  Comme nous avons affaire à des situations dans lesquelles il n'est possible d'associer qu'un seul attribut à chaque unité, nous avons j=1 k p j =1. MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaam aaqadabeWcbaGaamOAaiabg2da9iaaigdaaeaacaWGRbaaniabggHi LdGccaaMc8UaamiCamaaBaaaleaacaWGQbaabeaakiabg2da9iaaig dacaGGUaaaaa@46DC@  Donc, nous ne devons estimer que k1 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadUgacqGHsislcaaIXaaaaa@3DAF@  paramètres, puisqu'il s'ensuit que p ^ k =1 j=1 k1 p ^ j MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai qadchagaqcamaaBaaaleaacaWGRbaabeaakiabg2da9iaaigdacqGH sisldaaeWaqabSqaaiaadQgacqGH9aqpcaaIXaaabaGaam4Aaiabgk HiTiaaigdaa0GaeyyeIuoakiaaykW7ceWGWbGbaKaadaWgaaWcbaGa amOAaaqabaaaaa@4AF0@  et que l'estimation de la variance est également obtenue de manière analogue par V ^ ( p ^ k )= j=1 k1 V ^ ( p ^ j )+ lj=1 k1 C ^ ov( p ^ j , p ^ l ). MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai qadAfagaqcamaabmaabaGabmiCayaajaWaaSbaaSqaaiaadUgaaeqa aaGccaGLOaGaayzkaaGaeyypa0ZaaabmaeqaleaacaWGQbGaeyypa0 JaaGymaaqaaiaadUgacqGHsislcaaIXaaaniabggHiLdGccaaMc8Ua bmOvayaajaWaaeWaaeaaceWGWbGbaKaadaWgaaWcbaGaamOAaaqaba aakiaawIcacaGLPaaacqGHRaWkdaaeWaqabSqaaiaadYgacqGHGjsU caWGQbGaeyypa0JaaGymaaqaaiaadUgacqGHsislcaaIXaaaniabgg HiLdGccaaMc8Uabm4qayaajaGaam4BaiaadAhadaqadaqaaiqadcha gaqcamaaBaaaleaacaWGQbaabeaakiaacYcaceWGWbGbaKaadaWgaa WcbaGaamiBaaqabaaakiaawIcacaGLPaaacaGGUaaaaa@64BC@

En l'absence de toute autre information structurelle, nous supposons que les unités dans une catégorie donnée sont échangeables d'ordre deux, mais nous ne supposons aucune échangeabilité entre les unités de différentes catégories. Nos croyances a priori sont exprimées pour i=1,,N, MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadMgacqGH9aqpcaaIXaGaaGilaiablAciljaaiYcacaWGobGaaiil aaaa@41D7@  

j=1,,k1, MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadQgacqGH9aqpcaaIXaGaaGilaiablAciljaaiYcacaWGRbGaeyOe I0IaaGymaiaaiYcaaaa@43A3@  comme il suit :

m j =E( y ij )=P( y ij =1 ), v j =V( y ij )= m j ( 1 m j )et cov( y ij , y i j ) =P( y i j =1| y ij =1 )P( y ij =1 )P( y ij =1 )P( y i j =1 ) = m j ( m jj m j )= c j ,i i et σ j 2 = v j c j = m j ( 1 m jj ), MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOabaq qabaGaamyBamaaBaaaleaacaWGQbaabeaakiabg2da9iaadweadaqa daqaaiaadMhadaWgaaWcbaGaamyAaiaadQgaaeqaaaGccaGLOaGaay zkaaGaeyypa0JaamiuamaabmaabaGaamyEamaaBaaaleaacaWGPbGa amOAaaqabaGccqGH9aqpcaaIXaaacaGLOaGaayzkaaGaaGilaiaadA hadaWgaaWcbaGaamOAaaqabaGccqGH9aqpcaWGwbWaaeWaaeaacaWG 5bWaaSbaaSqaaiaadMgacaWGQbaabeaaaOGaayjkaiaawMcaaiabg2 da9iaad2gadaWgaaWcbaGaamOAaaqabaGcdaqadaqaaiaaigdacqGH sislcaWGTbWaaSbaaSqaaiaadQgaaeqaaaGccaGLOaGaayzkaaGaaG PaVlaaykW7caqGLbGaaeiDaaqaaiaabogacaqGVbGaaeODamaabmaa baGaamyEamaaBaaaleaacaWGPbGaamOAaaqabaGccaaISaGaamyEam aaBaaaleaaceWGPbGbauaacaWGQbaabeaaaOGaayjkaiaawMcaaiab g2da9iaadcfadaqadaqaamaaeiaabaGaamyEamaaBaaaleaaceWGPb GbauaacaWGQbaabeaakiabg2da9iaaigdacaaMc8oacaGLiWoacaWG 5bWaaSbaaSqaaiaadMgacaWGQbaabeaakiabg2da9iaaigdaaiaawI cacaGLPaaacaWGqbWaaeWaaeaacaWG5bWaaSbaaSqaaiaadMgacaWG Qbaabeaakiabg2da9iaaigdaaiaawIcacaGLPaaacqGHsislcaWGqb WaaeWaaeaacaWG5bWaaSbaaSqaaiaadMgacaWGQbaabeaakiabg2da 9iaaigdaaiaawIcacaGLPaaacaWGqbWaaeWaaeaacaWG5bWaaSbaaS qaaiqadMgagaqbaiaadQgaaeqaaOGaeyypa0JaaGymaaGaayjkaiaa wMcaaaqaaiabg2da9iaad2gadaWgaaWcbaGaamOAaaqabaGcdaqada qaaiaad2gadaWgaaWcbaGaamOAaiaadQgaaeqaaOGaeyOeI0IaamyB amaaBaaaleaacaWGQbaabeaaaOGaayjkaiaawMcaaiabg2da9iaado gadaWgaaWcbaGaamOAaaqabaGccaaISaGaaGPaVlaaykW7cqGHaiIi caWGPbGaeyiyIKRabmyAayaafaGaaGPaVlaaykW7caqGLbGaaeiDai aaykW7caaMc8Uaeq4Wdm3aa0baaSqaaiaadQgaaeaacaaIYaaaaOGa eyypa0JaamODamaaBaaaleaacaWGQbaabeaakiabgkHiTiaadogada WgaaWcbaGaamOAaaqabaGccqGH9aqpcaWGTbWaaSbaaSqaaiaadQga aeqaaOWaaeWaaeaacaaIXaGaeyOeI0IaamyBamaaBaaaleaacaWGQb GaamOAaaqabaaakiaawIcacaGLPaaacaaISaaaaaa@C668@

m jj =P( y i j =1| y ij =1 ), MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aad2gadaWgaaWcbaGaamOAaiaadQgaaeqaaOGaeyypa0Jaamiuamaa bmaabaWaaqGaaeaacaWG5bWaaSbaaSqaaiqadMgagaqbaiaadQgaae qaaOGaeyypa0JaaGymaiaaykW7aiaawIa7aiaadMhadaWgaaWcbaGa amyAaiaadQgaaeqaaOGaeyypa0JaaGymaaGaayjkaiaawMcaaiaacY caaaa@4F02@  pour tout i i . MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadMgacqGHGjsUceWGPbGbauaacaGGUaaaaa@3F78@

Pour j j , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadQgacqGHGjsUceWGQbGbauaacaGGSaaaaa@3F78@  nous obtenons de manière analogue la covariance entre ces catégories sous la forme

cov( y ij , y i j )={ m j ( m j j m j ), sii i , m j m j , sii= i . MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aabogacaqGVbGaaeODamaabmaabaGaamyEamaaBaaaleaacaWGPbGa amOAaaqabaGccaaISaGaamyEamaaBaaaleaaceWGPbGbauaaceWGQb GbauaaaeqaaaGccaGLOaGaayzkaaGaeyypa0ZaaiqaaeaafaqaaeGa caaabaGaamyBamaaBaaaleaacaWGQbaabeaakmaabmaabaGaamyBam aaBaaaleaaceWGQbGbauaacaWGQbaabeaakiabgkHiTiaad2gadaWg aaWcbaGabmOAayaafaaabeaaaOGaayjkaiaawMcaaiaaiYcaaeaaca aMi8UaaGjbVlaabohacaqGPbGaaGjbVlaayIW7caWGPbGaeyiyIKRa bmyAayaafaGaaGilaaqaaiabgkHiTiaad2gadaWgaaWcbaGaamOAaa qabaGccaWGTbWaaSbaaSqaaiqadQgagaqbaaqabaGccaaISaaabaGa aGjcVlaaysW7caqGZbGaaeyAaiaaysW7caaMi8UaamyAaiabg2da9i qadMgagaqbaiaai6caaaaacaGL7baaaaa@712C@

Souvent, nous ne possédons pas toutes les données y s , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aahMhadaWgaaWcbaGaam4CaaqabaGccaGGSaaaaa@3DF7@  mais seulement une statistique exhaustive, comme la proportion dans l'échantillon pour chaque catégorie, y ¯ s . MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai qahMhagaqeamaaBaaaleaacaWGZbaabeaakiaac6caaaa@3E11@  Soit y ¯ s MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai qahMhagaqeamaaBaaaleaacaWGZbaabeaaaaa@3D55@  le vecteur de dimension k1­ MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadUgacqGHsislcaaIXaqefCuzVj3zPfgaiuaacaWFTcaaaa@41BE@  dont la j e MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadQgadaahaaWcbeqaaiaabwgaaaaaaa@3D1B@  position est donnée par la moyenne d'échantillon pour la catégorie j. MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadQgacaGGUaaaaa@3CB8@  En utilisant le modèle général donné par (2.4), nous obtenons :

E( y ¯ s )=E( E( y ¯ s |β ) )=aetVar( y ¯ s )=E( V( y ¯ s |β ) )+V( E( y ¯ s |β ) )= V s +R. MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadweadaqadaqaaiqahMhagaqeamaaBaaaleaacaWGZbaabeaaaOGa ayjkaiaawMcaaiabg2da9iaadweadaqadaqaaiaadweadaqadaqaai qahMhagaqeamaaBaaaleaacaWGZbaabeaakiaacYhacaWHYoaacaGL OaGaayzkaaaacaGLOaGaayzkaaGaeyypa0JaaCyyaiaaykW7caaMc8 UaaeyzaiaabshacaaMc8UaaGPaVlaabAfacaqGHbGaaeOCamaabmaa baGabCyEayaaraWaaSbaaSqaaiaadohaaeqaaaGccaGLOaGaayzkaa Gaeyypa0JaamyramaabmaabaGaamOvamaabmaabaGabCyEayaaraWa aSbaaSqaaiaadohaaeqaaOGaaiiFaiaahk7aaiaawIcacaGLPaaaai aawIcacaGLPaaacqGHRaWkcaWGwbWaaeWaaeaacaWGfbWaaeWaaeaa ceWH5bGbaebadaWgaaWcbaGaam4CaaqabaGccaGG8bGaaCOSdaGaay jkaiaawMcaaaGaayjkaiaawMcaaiabg2da9iaahAfadaWgaaWcbaGa am4CaaqabaGccqGHRaWkcaWHsbGaaGOlaaaa@744C@

En appliquant le modèle général donné dans (2.4), où la variable de réponse est donnée par y ¯ s , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai qahMhagaqeamaaBaaaleaacaWGZbaabeaakiaacYcaaaa@3E0F@  le vecteur β MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aahk7aaaa@3C55@  est de dimension k1, MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadUgacqGHsislcaaIXaGaaiilaaaa@3E5F@   X s = I s MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aahIfadaWgaaWcbaGaam4CaaqabaGccqGH9aqpcaWHjbWaaSbaaSqa aiaadohaaeqaaaaa@4022@  et V=diag( V s ¯ , V s ), MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aahAfacqGH9aqpcaqGKbGaaeyAaiaabggacaqGNbWaaeWaaeaacaWH wbWaaSbaaSqaaiqadohagaqeaaqabaGccaaISaGaaCOvamaaBaaale aacaWGZbaabeaaaOGaayjkaiaawMcaaiaacYcaaaa@47BE@  nous obtenons à partir de (2.10) :

p ^ = n y ¯ s +( Nn ) β ^ N et V ^ ( p ^ )= ( Nn ) 2 [ V s ¯ +C ] N 2 ,         (4.1) MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai qahchagaqcaiabg2da9maalaaabaGaamOBaiqahMhagaqeamaaBaaa leaacaWGZbaabeaakiabgUcaRmaabmaabaGaamOtaiabgkHiTiaad6 gaaiaawIcacaGLPaaaceWHYoGbaKaaaeaacaWGobaaaiaaykW7caaM c8UaaeyzaiaabshacaaMc8UaaGPaVlqadAfagaqcamaabmaabaGabC iCayaajaaacaGLOaGaayzkaaGaeyypa0ZaaSaaaeaadaqadaqaaiaa d6eacqGHsislcaWGUbaacaGLOaGaayzkaaWaaWbaaSqabeaacaaIYa aaaOWaamWaaeaacaWHwbWaaSbaaSqaaiqadohagaqeaaqabaGccqGH RaWkcaWHdbaacaGLBbGaayzxaaaabaGaamOtamaaCaaaleqabaGaaG OmaaaaaaGccaaISaGaaeiiaiaabccacaqGGaGaaeiiaiaabccacaqG GaGaaeiiaiaabccacaqGGaGaaeikaiaabsdacaGGUaGaaGymaiaacM caaaa@6AF5@

C 1 = R 1 + V s MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aahoeadaahaaWcbeqaaiabgkHiTiaaigdaaaGccqGH9aqpcaWHsbWa aWbaaSqabeaacqGHsislcaaIXaaaaOGaey4kaSIaaCOvamaaBaaale aacaWGZbaabeaaaaa@4467@  et β ^ =C( V s 1 y ¯ s + R 1 a ), MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai qahk7agaqcaiabg2da9iaahoeadaqadaqaaiaahAfadaqhaaWcbaGa am4CaaqaaiabgkHiTiaaigdaaaGcceWH5bGbaebadaWgaaWcbaGaam 4CaaqabaGccqGHRaWkcaWHsbWaaWbaaSqabeaacqGHsislcaaIXaaa aOGaaCyyaaGaayjkaiaawMcaaiaacYcaaaa@4AF4@  comme il est énoncé en (2.6).

Soit Q= V s +R. MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aahgfacqGH9aqpcaWHwbWaaSbaaSqaaiaadohaaeqaaOGaey4kaSIa aCOuaiaac6caaaa@4173@  L'EBL de p MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aahchaaaa@3C10@  et sa variance associée donnés par (4.1) peuvent s'écrire en fonction des quantités a priori m j , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aad2gadaWgaaWcbaGaamOAaaqabaGccaGGSaaaaa@3DDE@   m j j MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aad2gadaWgaaWcbaGaamOAaiqadQgagaqbaaqabaaaaa@3E1F@  et j=1,,k1, MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadQgacqGH9aqpcaaIXaGaaGilaiablAciljaaiYcacaWGRbGaeyOe I0IaaGymaiaacYcaaaa@439D@  en notant que a=( m 1 ,, m k1 ), MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aahggacqGH9aqpdaqadaqaaiaad2gadaWgaaWcbaGaaGymaaqabaGc caaISaGaeSOjGSKaaGilaiaad2gadaWgaaWcbaGaam4AaiabgkHiTi aaigdaaeqaaaGccaGLOaGaayzkaaaccaGae8NmGiQae8hlaWcaaa@4921@   Q jj = c j + σ j 2 /n MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadgfadaWgaaWcbaGaamOAaiaadQgaaeqaaOGaeyypa0Jaam4yamaa BaaaleaacaWGQbaabeaakiabgUcaRmaalyaabaGaeq4Wdm3aa0baaS qaaiaadQgaaeaacaaIYaaaaaGcbaGaamOBaaaaaaa@46A4@  et Q j j = m j ( m j j m j ) m j m j j /n . MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadgfadaWgaaWcbaGaamOAaiqadQgagaqbaaqabaGccqGH9aqpcaWG TbWaaSbaaSqaaiaadQgaaeqaaOWaaeWaaeaacaWGTbWaaSbaaSqaai qadQgagaqbaiaadQgaaeqaaOGaeyOeI0IaamyBamaaBaaaleaaceWG QbGbauaaaeqaaaGccaGLOaGaayzkaaGaeyOeI0YaaSGbaeaacaWGTb WaaSbaaSqaaiaadQgaaeqaaOGaamyBamaaBaaaleaaceWGQbGbauaa caWGQbaabeaaaOqaaiaad6gaaaGaaiOlaaaa@50A6@  Par conséquent, la matrice R={ r j j },j, j =1,..,k1 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aahkfacqGH9aqpdaGadeqaaiaadkhadaWgaaWcbaGaamOAaiqadQga gaqbaaqabaaakiaawUhacaGL9baacaaISaGaamOAaiaaiYcaceWGQb GbauaacqGH9aqpcaaIXaGaaGilaiaai6cacaaIUaGaaGilaiaadUga cqGHsislcaaIXaaaaa@4CCC@  avec r jj = c j MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadkhadaWgaaWcbaGaamOAaiaadQgaaeqaaOGaeyypa0Jaam4yamaa BaaaleaacaWGQbaabeaaaaa@412B@  et r j j = m j ( m j j m j ) MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadkhadaWgaaWcbaGaamOAaiqadQgagaqbaaqabaGccqGH9aqpcaWG TbWaaSbaaSqaaiaadQgaaeqaaOWaaeWaaeaacaWGTbWaaSbaaSqaai qadQgagaqbaiaadQgaaeqaaOGaeyOeI0IaamyBamaaBaaaleaaceWG QbGbauaaaeqaaaGccaGLOaGaayzkaaaaaa@48F6@  et V s =1/n { v j j },j, j =1,..,k1 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aahAfadaWgaaWcbaGaam4CaaqabaGccqGH9aqpdaWcgaqaaiaaigda aeaacaWGUbaaaiaaykW7caaMc8+aaiWaaeaacaWG2bWaaSbaaSqaai aadQgaceWGQbGbauaaaeqaaaGccaGL7bGaayzFaaGaaGilaiaaykW7 caaMc8UaamOAaiaaiYcaceWGQbGbauaacqGH9aqpcaaIXaGaaGilai aai6cacaaIUaGaaGilaiaadUgacqGHsislcaaIXaaaaa@55F1@  avec v jj = σ j 2 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadAhadaWgaaWcbaGaamOAaiaadQgaaeqaaOGaeyypa0Jaeq4Wdm3a a0baaSqaaiaadQgaaeaacaaIYaaaaaaa@42C7@  et v j j = m j m j j . MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadAhadaWgaaWcbaGaamOAaiqadQgagaqbaaqabaGccqGH9aqpcqGH sislcaWGTbWaaSbaaSqaaiaadQgaaeqaaOGaamyBamaaBaaaleaace WGQbGbauaacaWGQbaabeaakiaac6caaaa@4600@  De manière analogue, nous obtenons V s ¯ =n/ ( Nn ) V s . MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aahAfadaWgaaWcbaGabm4Cayaaraaabeaakiabg2da9maalyaabaGa amOBaaqaamaabmaabaGaamOtaiabgkHiTiaad6gaaiaawIcacaGLPa aacaWHwbWaaSbaaSqaaiaadohaaeqaaaaakiaac6caaaa@4646@

4.1 Obtention des priors

L'obtention des priors est le processus consistant à formuler les connaissances et les croyances d'une personne au sujet d'une ou de plusieurs quantités incertaines sous forme d'une loi de probabilité pour ces quantités. Selon Garthwaite, Kadane et O'Hagan (2005), il est commode de concevoir la tâche d'obtention des priors comme faisant intervenir un facilitateur qui aide l'expert à formuler ses connaissances spécialisées sous forme probabiliste. Dans le contexte de l'obtention d'une loi a priori pour une analyse bayésienne, ce sont les connaissances a priori de l'expert qui sont tirées au clair, mais en général, l'objectif est d'exprimer les connaissances courantes de l'expert sous forme probabiliste. Si l'expert est un statisticien ou s'il connaît très bien les concepts statistiques, l'intervention d'un facilitateur pourrait ne pas être formellement nécessaire, mais cela est rare en pratique. O'Hagan (1998) a illustré au moyen d'un exemple pratique comment obtenir les premier et deuxième moments. En particulier, il a adopté l'approche bayésienne linéaire parce qu'elle permet aux ingénieurs d'appliquer facilement une procédure d'obtention des priors.

À la présente section, nous présentons certaines contraintes concernant les quantités a priori et une solution de rechange pour faciliter le processus d'obtention des priors en vue d'obtenir l'EBL pour des données catégoriques. Comme m j MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aad2gadaWgaaWcbaGaamOAaaqabaaaaa@3D24@  et m j j MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aad2gadaWgaaWcbaGaamOAaiqadQgagaqbaaqabaaaaa@3E1F@  sont des probabilités et que R MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aahkfaaaa@3BF2@  et V s MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aahAfadaWgaaWcbaGaam4Caaqabaaaaa@3D1A@  sont les matrices de covariance dans le modèle (2.4), les contraintes qui suivent doivent être satisfaites :

1.  0< m j <1 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aaicdacqGH8aapcaWGTbWaaSbaaSqaaiaadQgaaeqaaOGaeyipaWJa aGymaaaa@40AB@  et 0 m j j 1, MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aaicdacqGHKjYOcaWGTbWaaSbaaSqaaiaadQgaceWGQbGbauaaaeqa aOGaeyizImQaaGymaiaacYcaaaa@43B8@   j, j =1,,k1; MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadQgacaaISaGabmOAayaafaGaeyypa0JaaGymaiaaiYcacqWIMaYs caaISaGaam4AaiabgkHiTiaaigdacaGG7aaaaa@455D@

2.  R MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aahkfaaaa@3BF2@  et V s MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aahAfadaWgaaWcbaGaam4Caaqabaaaaa@3D1A@  sont des matrices symétriques définies positives.

Afin de vérifier si la condition (2.2) est satisfaite, on peut exécuter les étapes suivantes :

i. vérifier si R MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aahkfaaaa@3BF2@  et V s MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aahAfadaWgaaWcbaGaam4Caaqabaaaaa@3D1A@  sont symétriques en vérifiant que m j m j j = m j m j j ; MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aad2gadaWgaaWcbaGaamOAaaqabaGccaWGTbWaaSbaaSqaaiaadQga ceWGQbGbauaaaeqaaOGaeyypa0JaamyBamaaBaaaleaaceWGQbGbau aaaeqaaOGaamyBamaaBaaaleaaceWGQbGbauaacaWGQbaabeaakiaa cUdaaaa@473A@

ii. vérifier si R MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aahkfaaaa@3BF2@  et V s MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aahAfadaWgaaWcbaGaam4Caaqabaaaaa@3D1A@  sont des matrices définies positives en trouvant les valeurs propres de R MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aahkfaaaa@3BF2@  et V s . MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aahAfadaWgaaWcbaGaam4CaaqabaGccaGGUaaaaa@3DD6@  Si les valeurs propres sont positives, alors les matrices sont définies positives.

Il convient de mentionner que les valeurs propres sont les racines du polynôme caractéristique et que si ce polynôme est de degré n,n4, MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aad6gacaGGSaGaamOBaiabgsMiJkaaisdacaGGSaaaaa@40D0@  il est possible d'obtenir analytiquement ses racines en appliquant Bhaskara, Cardan ou Ferrari; voir Jacobson (2009), chapitre 4, pour les formules. Cependant, si n5, MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aad6gacqGHLjYScaaI1aGaaiilaaaa@3F3F@  il est habituellement nécessaire d'appliquer une méthode itérative pour les obtenir. Néanmoins, pour les matrices de dimensions supérieures à 2×2, MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aaikdacqGHxdaTcaaIYaGaaiilaaaa@3F56@  il n'est pas simple d'obtenir analytiquement ces contraintes en se basant sur les valeurs propres. La proposition qui suit présente les conditions que m j MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aad2gadaWgaaWcbaGaamOAaaqabaaaaa@3D24@  et m j j , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aad2gadaWgaaWcbaGaamOAaiqadQgagaqbaaqabaGccaGGSaaaaa@3ED9@   j=1,,k1, MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadQgacqGH9aqpcaaIXaGaaGilaiablAciljaaiYcacaWGRbGaeyOe I0IaaGymaiaacYcaaaa@439D@  doivent satisfaire afin d'obtenir un prior convenable pour un modèle multinomial comprenant trois catégories en utilisant l'approche d'estimation bayésienne linéaire.

Proposition 1 Supposons que nous obtenons m j , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aad2gadaWgaaWcbaGaamOAaaqabaGccaGGSaaaaa@3DDE@  tel que 0< m j <1, MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aaicdacqGH8aapcaWGTbWaaSbaaSqaaiaadQgaaeqaaOGaeyipaWJa aGymaiaacYcaaaa@415B@   j=1,2. MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadQgacqGH9aqpcaaIXaGaaGilaiaaikdacaGGUaaaaa@3FEB@  Alors, sachant ρ 11 , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai abeg8aYnaaBaaaleaacaaIXaGaaGymaaqabaGccaGGSaaaaa@3F33@   ρ 12 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai abeg8aYnaaBaaaleaacaaIXaGaaGOmaaqabaaaaa@3E7A@  et ρ 22 , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai abeg8aYnaaBaaaleaacaaIYaGaaGOmaaqabaGccaGGSaaaaa@3F35@  nous obtenons m 11 , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aad2gadaWgaaWcbaGaaGymaiaaigdaaeqaaOGaaiilaaaa@3E65@   m 12 , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aad2gadaWgaaWcbaGaaGymaiaaikdaaeqaaOGaaiilaaaa@3E66@   m 21 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aad2gadaWgaaWcbaGaaGOmaiaaigdaaeqaaaaa@3DAC@  et m 22 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aad2gadaWgaaWcbaGaaGOmaiaaikdaaeqaaaaa@3DAD@  au moyen de (4.2). Les quantités a priori m j MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aad2gadaWgaaWcbaGaamOAaaqabaaaaa@3D24@  et m j j , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aad2gadaWgaaWcbaGaamOAaiqadQgagaqbaaqabaGccaGGSaaaaa@3ED9@  pour j, j =1,2, MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadQgacaaISaGabmOAayaafaGaeyypa0JaaGymaiaaiYcacaaIYaGa aiilaaaa@419A@  doivent satisfaire les contraintes qui suivent pour que les matrices R MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aahkfaaaa@3BF2@  et V s MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aahAfadaWgaaWcbaGaam4Caaqabaaaaa@3D1A@  soient définies positives :

m 11 > m 1 et m 22 > m 2 , m 11 m 22 m 11 m 22 +1> m 12 m 21 et m 11 m 22 m 11 m 2 m 1 m 22 > m 12 m 21 2 m 2 m 12 . MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOabae qabaGaamyBamaaBaaaleaacaaIXaGaaGymaaqabaGcqaaaaaaaaaWd biabg6da+8aacaWGTbWaaSbaaSqaaiaaigdaaeqaaOGaaGPaVlaayk W7caqGHbGaaeOBaiaabsgacaaMc8UaaGPaVlaad2gadaWgaaWcbaGa aGOmaiaaikdaaeqaaOWdbiabg6da+8aacaWGTbWaaSbaaSqaaiaaik daaeqaaOGaaGilaiaad2gadaWgaaWcbaGaaGymaiaaigdaaeqaaOGa amyBamaaBaaaleaacaaIYaGaaGOmaaqabaGccqGHsislcaWGTbWaaS baaSqaaiaaigdacaaIXaaabeaakiabgkHiTiaad2gadaWgaaWcbaGa aGOmaiaaikdaaeqaaOGaey4kaSIaaGymaiabg6da+iaad2gadaWgaa WcbaGaaGymaiaaikdaaeqaaOGaamyBamaaBaaaleaacaaIYaGaaGym aaqabaGccaaMc8UaaGPaVlaabwgacaqG0baabaGaamyBamaaBaaale aacaaIXaGaaGymaaqabaGccaWGTbWaaSbaaSqaaiaaikdacaaIYaaa beaakiabgkHiTiaad2gadaWgaaWcbaGaaGymaiaaigdaaeqaaOGaam yBamaaBaaaleaacaaIYaaabeaakiabgkHiTiaad2gadaWgaaWcbaGa aGymaaqabaGccaWGTbWaaSbaaSqaaiaaikdacaaIYaaabeaakiabg6 da+iaad2gadaWgaaWcbaGaaGymaiaaikdaaeqaaOGaamyBamaaBaaa leaacaaIYaGaaGymaaqabaGccqGHsislcaaIYaGaamyBamaaBaaale aacaaIYaaabeaakiaad2gadaWgaaWcbaGaaGymaiaaikdaaeqaaOGa aGOlaaaaaa@8694@

La vérification de la proposition 1 nécessite certaines opérations algébriques. Nous vérifions que les matrices R MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aahkfaaaa@3BF2@  et V s MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aahAfadaWgaaWcbaGaam4Caaqabaaaaa@3D1A@  sont définies positives en utilisant (i) et (ii) susmentionnés. Nous faisons appel au fait que les valeurs propres d'une matrice de dimensions 2×2 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aaikdacqGHxdaTcaaIYaaaaa@3EA6@  sont positives si et seulement si son déterminant est positif et nous obtenons alors m j j , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aad2gadaWgaaWcbaGaamOAaiqadQgagaqbaaqabaGccaGGSaaaaa@3ED9@   j, j =1,2 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadQgacaaISaGabmOAayaafaGaeyypa0JaaGymaiaaiYcacaaIYaaa aa@40EA@  qui satisfait cette contrainte pour les deux matrices. Pour les cas comprenant plus de trois catégories, nous devons vérifier numériquement si les matrices R MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aahkfaaaa@3BF2@  et V s MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aahAfadaWgaaWcbaGaam4Caaqabaaaaa@3D1A@  sont définies positives en remplaçant m j MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aad2gadaWgaaWcbaGaamOAaaqabaaaaa@3D24@  et m j j , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aad2gadaWgaaWcbaGaamOAaiqadQgagaqbaaqabaGccaGGSaaaaa@3ED9@   j=1,,k1 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadQgacqGH9aqpcaaIXaGaaGilaiablAciljaaiYcacaWGRbGaeyOe I0IaaGymaaaa@42ED@  par leur valeur numérique.

Par ailleurs, si un expert a de la difficulté à spécifier certaines de ces probabilités conditionnelles m j j , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aad2gadaWgaaWcbaGaamOAaiqadQgagaqbaaqabaGccaGGSaaaaa@3ED9@  il pourrait être plus simple d'attribuer un prior au coefficient de corrélation. Définissons ρ j j MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai abeg8aYnaaBaaaleaacaWGQbGabmOAayaafaaabeaaaaa@3EED@  comme étant le prior du coefficient de corrélation entre deux unités différentes dans les catégories j MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadQgaaaa@3C06@  et j , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai qadQgagaqbaiaacYcaaaa@3CC2@  c'est-à-dire :

ρ j j =corr( y ij , y i j )={ m jj m j 1 m j , j= j , m j ( m j j m j ) m j ( 1 m j ) m j ( 1 m j ) , j j , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai abeg8aYnaaBaaaleaacaWGQbGabmOAayaafaaabeaakiabg2da9iaa bogacaqGVbGaaeOCaiaabkhadaqadaqaaiaadMhadaWgaaWcbaGaam yAaiaadQgaaeqaaOGaaGilaiaadMhadaWgaaWcbaGabmyAayaafaGa bmOAayaafaaabeaaaOGaayjkaiaawMcaaiabg2da9maaceaabaqbae aabiGaaaqaamaalaaabaGaamyBamaaBaaaleaacaWGQbGaamOAaaqa baGccqGHsislcaWGTbWaaSbaaSqaaiaadQgaaeqaaaGcbaGaaGymai abgkHiTiaad2gadaWgaaWcbaGaamOAaaqabaaaaOGaaGilaaqaaiaa dQgacqGH9aqpceWGQbGbauaacaaISaaabaWaaSaaaeaacaWGTbWaaS baaSqaaiaadQgaaeqaaOWaaeWaaeaacaWGTbWaaSbaaSqaaiqadQga gaqbaiaadQgaaeqaaOGaeyOeI0IaamyBamaaBaaaleaaceWGQbGbau aaaeqaaaGccaGLOaGaayzkaaaabaWaaOaaaeaacaWGTbWaaSbaaSqa aiaadQgaaeqaaOWaaeWaaeaacaaIXaGaeyOeI0IaamyBamaaBaaale aacaWGQbaabeaaaOGaayjkaiaawMcaaiaad2gadaWgaaWcbaGabmOA ayaafaaabeaakmaabmaabaGaaGymaiabgkHiTiaad2gadaWgaaWcba GabmOAayaafaaabeaaaOGaayjkaiaawMcaaaWcbeaaaaGccaaISaaa baGaamOAaiabgcMi5kqadQgagaqbaiaaiYcaaaaacaGL7baaaaa@7A86@

pour i, i =1,,n, MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadMgacaaISaGabmyAayaafaGaeyypa0JaaGymaiaaiYcacqWIMaYs caaISaGaamOBaiaacYcaaaa@43A7@   i i , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadMgacqGHGjsUceWGPbGbauaacaGGSaaaaa@3F76@   j, j =1,,k1. MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadQgacaaISaGabmOAayaafaGaeyypa0JaaGymaiaaiYcacqWIMaYs caaISaGaam4AaiabgkHiTiaaigdacaGGUaaaaa@4550@

Par conséquent, sachant ρ j j , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai abeg8aYnaaBaaaleaacaWGQbGabmOAayaafaaabeaakiaacYcaaaa@3FA7@   j, j =1,,k1, MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadQgacaaISaGabmOAayaafaGaeyypa0JaaGymaiaaiYcacqWIMaYs caaISaGaam4AaiabgkHiTiaaigdacaGGSaaaaa@454E@  nous obtenons

m j j ={ m j + ρ jj ( 1 m j ) j= j , m j m j + ρ j j m j ( 1 m j ) m j ( 1 m j ) m j , j j .          (4.2) MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aad2gadaWgaaWcbaGaamOAaiqadQgagaqbaaqabaGccqGH9aqpdaGa baqaauaabaqaciaaaeaacaWGTbWaaSbaaSqaaiaadQgaaeqaaOGaey 4kaSIaeqyWdi3aaSbaaSqaaiaadQgacaWGQbaabeaakmaabmaabaGa aGymaiabgkHiTiaad2gadaWgaaWcbaGaamOAaaqabaaakiaawIcaca GLPaaaaeaacaWGQbGaeyypa0JabmOAayaafaGaaGilaaqaamaalaaa baGaamyBamaaBaaaleaacaWGQbaabeaakiaad2gadaWgaaWcbaGabm OAayaafaaabeaakiabgUcaRiabeg8aYnaaBaaaleaaceWGQbGbauaa caWGQbaabeaakmaakaaabaGaamyBamaaBaaaleaacaWGQbaabeaakm aabmaabaGaaGymaiabgkHiTiaad2gadaWgaaWcbaGaamOAaaqabaaa kiaawIcacaGLPaaacaWGTbWaaSbaaSqaaiqadQgagaqbaaqabaGcda qadaqaaiaaigdacqGHsislcaWGTbWaaSbaaSqaaiqadQgagaqbaaqa baaakiaawIcacaGLPaaaaSqabaaakeaacaWGTbWaaSbaaSqaaiqadQ gagaqbaaqabaaaaOGaaGilaaqaaiaadQgacqGHGjsUceWGQbGbauaa caaIUaaaaaGaay5EaaGaaeiiaiaabccacaqGGaGaaeiiaiaabccaca qGGaGaaeiiaiaabccacaqGGaGaaeikaiaaisdacaGGUaGaaGOmaiaa cMcaaaa@789B@

Il convient de mentionner que, si l'on dispose de données provenant d'une enquête antérieure, il est possible qu'un expert utilise cette information. Par exemple, m j MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aad2gadaWgaaWcbaGaamOAaaqabaaaaa@3D24@  peut être obtenu en estimant la proportion d'unités dans la catégorie j, MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadQgacaGGSaaaaa@3CB6@   j=1,,k1 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadQgacqGH9aqpcaaIXaGaaGilaiablAciljaaiYcacaWGRbGaeyOe I0IaaGymaaaa@42ED@  à partir de l'enquête antérieure. De façon analogue, ρ j j MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai abeg8aYnaaBaaaleaacaWGQbGabmOAayaafaaabeaaaaa@3EED@  peut être obtenu en utilisant les données d'une enquête antérieure. Comme l'indique la contrainte (2.1), m j MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aad2gadaWgaaWcbaGaamOAaaqabaaaaa@3D24@  ne peut pas prendre les valeurs 0 et 1, sinon les corrélations ne seraient pas définies.

4.2  Analyse de la sensibilité aux priors

Il est utile de vérifier si l'estimateur et sa variance associée dépendent des priors attribués. Nous traitons le cas simple ne comprenant que deux catégories. Soulignons que, dans le cas où il y a plus de deux catégories, le nombre de quantités a priori qu'il faut obtenir augmente rapidement, mais que l'on peut étendre les conclusions obtenues. Par ailleurs, en l'absence d'information a priori, nous pouvons utiliser des priors non informatifs et, comme il est décrit à la section 2.2, on retrouve alors les estimateurs de l'approche fondée sur le plan de sondage.

L'EBL pour la proportion en cas de données binaires peut être obtenu en tant que cas particulier de l'estimateur (4.1),

p ^ 1 = n y ¯ 1 +( Nn ) μ ^ N , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai qadchagaqcamaaBaaaleaacaaIXaaabeaakiabg2da9maalaaabaGa amOBaiqadMhagaqeamaaBaaaleaacaaIXaaabeaakiabgUcaRmaabm aabaGaamOtaiabgkHiTiaad6gaaiaawIcacaGLPaaacuaH8oqBgaqc aaqaaiaad6eaaaGaaGilaaaa@498A@

μ ^ =ω y ¯ 1 +( 1ω ) m 1 estlavaleurprévuedesvaleursnonobservéesdanslacatégorie1, ω= n σ 1 2 n σ 1 2 + c 1 1 , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOabae qabaGafqiVd0MbaKaacqGH9aqpcqaHjpWDceWG5bGbaebadaWgaaWc baGaaGymaaqabaGccqGHRaWkdaqadaqaaiaaigdacqGHsislcqaHjp WDaiaawIcacaGLPaaacaWGTbWaaSbaaSqaaiaaigdaaeqaaOGaaGjc VlaaysW7caqGLbGaae4CaiaabshacaaMe8UaaeiBaiaabggacaaMe8 UaaeODaiaabggacaqGSbGaaeyzaiaabwhacaqGYbGaaGjbVlaabcha caqGYbGaaey6aiaabAhacaqG1bGaaeyzaiaaysW7caqGKbGaaeyzai aabohacaaMe8UaaeODaiaabggacaqGSbGaaeyzaiaabwhacaqGYbGa ae4CaiaaysW7caqGUbGaae4Baiaab6gacaaMe8Uaae4Baiaabkgaca qGZbGaaeyzaiaabkhacaqG2bGaaey6aiaabwgacaqGZbGaaGjbVlaa bsgacaqGHbGaaeOBaiaabohacaaMe8UaaeiBaiaabggacaaMe8Uaae 4yaiaabggacaqG0bGaaey6aiaabEgacaqGVbGaaeOCaiaabMgacaqG LbGaaGjbVlaaigdacaaMi8UaaGilaaqaaiabeM8a3jabg2da9maala aabaGaamOBaiabeo8aZnaaDaaaleaacaaIXaaabaGaeyOeI0IaaGOm aaaaaOqaaiaad6gacqaHdpWCdaWgaaWcbaGaaGymaaqabaGcdaahaa WcbeqaaiabgkHiTiaaikdaaaGccqGHRaWkcaWGJbWaaSbaaSqaaiaa igdaaeqaaOWaaWbaaSqabeaacqGHsislcaaIXaaaaaaakiaaiYcaaa aa@A7A5@

et p ^ 2 =1 p ^ 1 . MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai qadchagaqcamaaBaaaleaacaaIYaaabeaakiabg2da9iaaigdacqGH sislceWGWbGbaKaadaWgaaWcbaGaaGymaaqabaGccaGGUaaaaa@4264@  Notons que σ 1 2 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai abeo8aZnaaDaaaleaacaaIXaaabaGaaGOmaaaaaaa@3E7E@  et c 1 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadogadaWgaaWcbaGaaGymaaqabaaaaa@3CE6@  dépendent de m 11 = m 1 + ρ 11 ( 1 m 1 ), MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aad2gadaWgaaWcbaGaaGymaiaaigdaaeqaaOGaeyypa0JaamyBamaa BaaaleaacaaIXaaabeaakiabgUcaRiabeg8aYnaaBaaaleaacaaIXa GaaGymaaqabaGcdaqadaqaaiaaigdacqGHsislcaWGTbWaaSbaaSqa aiaaigdaaeqaaaGccaGLOaGaayzkaaGaaiilaaaa@4AB0@  voir page  13 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKk Fr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabiabaaGcbaGaaG ymaiaaiodaaaa@398B@ . Nous analysons comment les estimations sont affectées par ρ 11 . MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai abeg8aYnaaBaaaleaacaaIXaGaaGymaaqabaGccaGGUaaaaa@3F35@

1. Si ρ 11 0, MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai abeg8aYnaaBaaaleaacaaIXaGaaGymaaqabaGccqGHsgIRcaaIWaGa aiilaaaa@41DA@  alors ω0 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai abeM8a3jabgkziUkaaicdaaaa@3F8B@  et μ ^ m 1 . MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai qbeY7aTzaajaGaeyOKH4QaamyBamaaBaaaleaacaaIXaaabeaakiaa c6caaaa@415F@  Donc, l'estimateur pour les valeurs non observées dépend en grande partie de la valeur du prior.

2. Si ρ 11 1, MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai abeg8aYnaaBaaaleaacaaIXaGaaGymaaqabaGccqGHsgIRcaaIXaGa aiilaaaa@41DB@  alors ω1 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai abeM8a3jabgkziUkaaigdaaaa@3F8C@  et μ ^ y ¯ 1 . MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai qbeY7aTzaajaGaeyOKH4QabmyEayaaraWaaSbaaSqaaiaaigdaaeqa aOGaaiOlaaaa@4183@  Donc, l'estimateur pour les valeurs non observées ne dépend pas de la valeur du prior.

En outre, il est facile de voir que n/N 1, MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaam aalyaabaGaamOBaaqaaiaad6eaaaGaeyOKH4QaaGymaiaacYcaaaa@404B@   p ^ 1 y ¯ 1 . MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai qadchagaqcamaaBaaaleaacaaIXaaabeaakiabgkziUkqadMhagaqe amaaBaaaleaacaaIXaaabeaakiaac6caaaa@41B3@  Pour illustrer ces résultats, nous avons créé un jeu de données artificielles en fixant la proportion réelle à p=( 0,2380;0,7620 ) MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aahchacqGH9aqpdaqadaqaaiaabcdacaqGSaGaaeOmaiaabodacaqG 4aGaaeimaiaacUdacaqGWaGaaeilaiaabEdacaqG2aGaaeOmaiaabc daaiaawIcacaGLPaaaiiaacqWFYaIOaaa@4959@  et la moyenne d'échantillon à y ¯ s =( 0,2614;0,7386 ). MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai qahMhagaqeamaaBaaaleaacaWGZbaabeaakiabg2da9maabmaabaGa aeimaiaabYcacaqGYaGaaeOnaiaabgdacaqG0aGaai4oaiaabcdaca qGSaGaae4naiaabodacaqG4aGaaeOnaaGaayjkaiaawMcaaGGaaiab =jdiIkab=5caUaaa@4B92@  Ces valeurs ont été tirées de Moura et Migon (2002). Puis, nous avons déterminé comment les valeurs de m 1 , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aad2gadaWgaaWcbaGaaGymaaqabaGccaGGSaaaaa@3DAA@   N, MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aad6eacaGGSaaaaa@3C9A@   f=n/N MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadAgacqGH9aqpdaWcgaqaaiaad6gaaeaacaWGobaaaaaa@3EE4@  et ρ 11 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai abeg8aYnaaBaaaleaacaaIXaGaaGymaaqabaaaaa@3E79@  affectent l'estimateur p ^ 1 . MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai qadchagaqcamaaBaaaleaacaaIXaaabeaakiaac6caaaa@3DBF@  La figure 4.1 donne la représentation graphique en deux dimensions de l'erreur absolue de p ^ 1 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai qadchagaqcamaaBaaaleaacaaIXaaabeaaaaa@3D03@  en fonction de ρ 11 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai abeg8aYnaaBaaaleaacaaIXaGaaGymaaqabaaaaa@3E79@  pour certains cas particuliers. La courbe grise représente l'erreur absolue entre la proportion d'échantillon y ¯ 1 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai qadMhagaqeamaaBaaaleaacaaIXaaabeaaaaa@3D14@  et la proportion réelle p 1 . MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadchadaWgaaWcbaGaaGymaaqabaGccaGGUaaaaa@3DAF@

Il faut souligner que, à mesure que f MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadAgaaaa@3C02@  ou N MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aad6eaaaa@3BEA@  augmente, l'erreur absolue diminue pour toute valeur du prior. De surcroît, quand ρ 11 0, MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai abeg8aYnaaBaaaleaacaaIXaGaaGymaaqabaGccqGHsgIRcaaIWaGa aiilaaaa@41DA@  l'erreur absolue augmente quand m 1 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aad2gadaWgaaWcbaGaaGymaaqabaaaaa@3CF0@  diffère considérablement de la proportion réelle p 1 , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aadchadaWgaaWcbaGaaGymaaqabaGccaGGSaaaaa@3DAD@  mais elle diminue à mesure que la taille de l'échantillon augmente. Enfin, quand ρ 11 1, MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai abeg8aYnaaBaaaleaacaaIXaGaaGymaaqabaGccqGHsgIRcaaIXaGa aiilaaaa@41DB@  nous observons que l'erreur absolue de p ^ 1 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai qadchagaqcamaaBaaaleaacaaIXaaabeaaaaa@3D03@  tend vers l'erreur absolue de la proportion d'échantillon y ¯ 1 . MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai qadMhagaqeamaaBaaaleaacaaIXaaabeaakiaac6caaaa@3DD0@  Donc, si nous avons une bonne information a priori, en ce qui concerne m 1 , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai aad2gadaWgaaWcbaGaaGymaaqabaGccaGGSaaaaa@3DAA@  l'estimateur proposé donne de bons résultats pour toutes les valeurs de ρ 11 . MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai abeg8aYnaaBaaaleaacaaIXaGaaGymaaqabaGccaGGUaaaaa@3F35@  Cependant, si aucune information a priori n'est disponible, des priors non informatifs caractérisés par ρ 11 1 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai abeg8aYnaaBaaaleaacaaIXaGaaGymaaqabaGccqGHsgIRcaaIXaaa aa@412B@  peuvent être utilisés et nous obtenons des résultats similaires à ceux de l'approche fondée sur le plan de sondage.

Figure 4.1 Représentation graphique en deux dimensions de l'erreur absolue de p ^ 1 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai qadchagaqcamaaBaaaleaacaaIXaaabeaaaaa@3D03@ en fonction de ρ 11 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqaaeaabiGaaiaacaqabeaadaqaaqaaaOqaai abeg8aYnaaBaaaleaacaaIXaGaaGymaaqabaaaaa@3E79@ pour certains cas particuliers.

Description pour figure 4.1

Description pour figure 4.1

Note: Erreur absolue pour m 1 { 0,1 ; 0,4 ; 0,7 ; 0,9 } , MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqipu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqabeqabmGadiqaceqabeqadeqadqaaaOqaai aad2gadaWgaaWcbaGaaGymaaqabaGccqGHiiIZdaGadaqaaiaabcda caqGSaGaaeymaiaacUdacaqGWaGaaeilaiaabsdacaGG7aGaaeimai aabYcacaqG3aGaai4oaiaabcdacaqGSaGaaeyoaaGaay5Eaiaaw2ha aiaacYcaaaa@4C02@   N { 1 500,15 288 } MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqipu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqabeqabmGadiqaceqabeqadeqadqaaaOqaai aad6eacqGHiiIZdaGadaqaaiaaigdacaaMe8UaaGynaiaaicdacaaI WaGaaGilaiaaigdacaaI1aGaaGjbVlaaikdacaaI4aGaaGioaaGaay 5Eaiaaw2haaaaa@4A14@  et f { 1 % ,10 % } MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqipu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqabeqabmGadiqaceqabeqadeqadqaaaOqaai aadAgacqGHiiIZdaGadaqaaiaaigdacaaMe8UaaeyjaiaaiYcacaaI XaGaaGimaiaaysW7caqGLaaacaGL7bGaayzFaaaaaa@4704@  fixes et ρ 11 { 0,01 ; 0,25 ; 0,5 ; 0,75 ; 0,9 } MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqipu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqabeqabmGadiqaceqabeqadeqadqaaaOqaai abeg8aYnaaBaaaleaacaaIXaGaaGymaaqabaGccqGHiiIZdaGadaqa aiaabcdacaqGSaGaaeimaiaabgdacaGG7aGaaeimaiaabYcacaqGYa GaaeynaiaacUdacaqGWaGaaeilaiaabwdacaGG7aGaaeimaiaabYca caqG3aGaaeynaiaacUdacaqGWaGaaeilaiaabMdaaiaawUhacaGL9b aaaaa@51D5@  variable. La courbe grise représente l'erreur absolue de la proportion d'échantillon y ¯ 1 MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbcvPDwzYbGeaGqipu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0x e9LqFf0xe9q8qqvqFr0dXdbrVc=b0P0xb9peuD0xXdbvk9qq=xd9qq aq=Jf9sr0=vr0=vrWZqabeqabmGadiqaceqabeqadeqadqaaaOqaai qadMhagaqeamaaBaaaleaacaaIXaaabeaaaaa@3D11@

Précédent | Suivant

Date de modification :