4. Effet de l’utilisation de paramètres de coût estimés

David G. Steel et Robert Graham Clark

Précédent | Suivant

En pratique, les coûts c i MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9 vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=x fr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGaam4yamaaBa aaleaacaWGPbaabeaaaaa@37E9@ ne sont pas connus précisément. Supposons qu’ils sont remplacés par les estimations c ^ i = b i c i MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9 vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=x fr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGabm4yayaaja WaaSbaaSqaaiaadMgaaeqaaOGaeyypa0JaamOyamaaBaaaleaacaWG PbaabeaakiaadogadaWgaaWcbaGaamyAaaqabaaaaa@3D16@ . L’utilisation de la variable auxiliaire et des coûts estimés dans les probabilités optimales implique que π i z i 1 / 2 c ^ i 1 / 2 . MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9 vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=x fr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGaeqiWda3aaS baaSqaaiaadMgaaeqaaOGaeyyhIuRaamOEamaaDaaaleaacaWGPbaa baGaaGymaiaac+cacaaIYaaaaOGabm4yayaajaWaa0baaSqaaiaadM gaaeaacqGHsislcaaIXaGaai4laiaaikdaaaGccaGGUaaaaa@447C@ Pour faire des comparaisons pour les mêmes coûts prévus,

π i = C f z i 1 / 2 c ^ i 1 / 2 j U z j 1 / 2 c ^ j 1 / 2 c j . MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9 vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=x fr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGaeqiWda3aaS baaSqaaiaadMgaaeqaaOGaeyypa0Jaam4qamaaBaaaleaacaWGMbaa beaakmaalaaabaGaamOEamaaDaaaleaacaWGPbaabaGaaGymaiaac+ cacaaIYaaaaOGabm4yayaajaWaa0baaSqaaiaadMgaaeaacqGHsisl caaIXaGaai4laiaaikdaaaaakeaadaaeqaqaaiaadQhadaqhaaWcba GaamOAaaqaaiaaigdacaGGVaGaaGOmaaaakiqadogagaqcamaaDaaa leaacaWGQbaabaGaeyOeI0IaaGymaiaac+cacaaIYaaaaOGaam4yam aaBaaaleaacaWGQbaabeaaaeaacaWGQbGaeyicI4Saamyvaaqab0Ga eyyeIuoaaaGccaaIUaaaaa@56B6@

La variance anticipée résultante est

A V e s t s = σ 2 C f 1 ( i U c ^ i 1 / 2 z i 1 / 2 ) ( j U z j 1 / 2 c ^ j 1 / 2 c j ) σ 2 i U z i . ( 4.1 ) MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9 vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=x fr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGaamyqaiaadA fadaWgaaWcbaGaamyzaiaadohacaWG0bGaam4CaaqabaGccqGH9aqp cqaHdpWCdaahaaWcbeqaaiaaikdaaaGccaWGdbWaa0baaSqaaiaadA gaaeaacqGHsislcaaIXaaaaOWaaeWaaeaadaaeqbqabSqaaiaadMga cqGHiiIZcaWGvbaabeqdcqGHris5aOGabm4yayaajaWaa0baaSqaai aadMgaaeaacaaIXaGaai4laiaaikdaaaGccaWG6bWaa0baaSqaaiaa dMgaaeaacaaIXaGaai4laiaaikdaaaaakiaawIcacaGLPaaadaqada qaamaaqafabeWcbaGaamOAaiabgIGiolaadwfaaeqaniabggHiLdGc caWG6bWaa0baaSqaaiaadQgaaeaacaaIXaGaai4laiaaikdaaaGcce WGJbGbaKaadaqhaaWcbaGaamOAaaqaaiabgkHiTiaaigdacaGGVaGa aGOmaaaakiaadogadaWgaaWcbaGaamOAaaqabaaakiaawIcacaGLPa aacqGHsislcqaHdpWCdaahaaWcbeqaaiaaikdaaaGcdaaeqbqabSqa aiaadMgacqGHiiIZcaWGvbaabeqdcqGHris5aOGaamOEamaaBaaale aacaWGPbaabeaakiaai6cacaaMf8UaaGzbVlaaywW7caaMf8UaaGzb VlaaywW7caaMf8UaaGzbVlaacIcacaaI0aGaaiOlaiaaigdacaGGPa aaaa@80F7@

Si nous supposons que les valeurs de b i MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9 vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=x fr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGaamOyamaaBa aaleaacaWGPbaabeaaaaa@37E8@ ne sont pas reliées aux valeurs de c i MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9 vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=x fr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGaam4yamaaBa aaleaacaWGPbaabeaaaaa@37E9@ et z i , MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9 vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=x fr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGaamOEamaaBa aaleaacaWGPbaabeaakiaacYcaaaa@38BA@ alors

A V e s t s = σ 2 C f 1 ( i U c i 1 / 2 z i 1 / 2 ) 2 N 2 ( i U b i 1 / 2 ) ( i U b i 1 / 2 ) σ 2 i U z i , ( 4.2 ) MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9 vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=x fr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGaamyqaiaadA fadaWgaaWcbaGaamyzaiaadohacaWG0bGaam4CaaqabaGccqGH9aqp cqaHdpWCdaahaaWcbeqaaiaaikdaaaGccaWGdbWaa0baaSqaaiaadA gaaeaacqGHsislcaaIXaaaaOWaaeWaaeaadaaeqbqabSqaaiaadMga cqGHiiIZcaWGvbaabeqdcqGHris5aOGaam4yamaaDaaaleaacaWGPb aabaGaaGymaiaac+cacaaIYaaaaOGaamOEamaaDaaaleaacaWGPbaa baGaaGymaiaac+cacaaIYaaaaaGccaGLOaGaayzkaaWaaWbaaSqabe aacaaIYaaaaOGaamOtamaaCaaaleqabaGaeyOeI0IaaGOmaaaakmaa bmaabaWaaabuaeqaleaacaWGPbGaeyicI4Saamyvaaqab0GaeyyeIu oakiaadkgadaqhaaWcbaGaamyAaaqaaiabgkHiTiaaigdacaGGVaGa aGOmaaaaaOGaayjkaiaawMcaamaabmaabaWaaabuaeqaleaacaWGPb GaeyicI4Saamyvaaqab0GaeyyeIuoakiaadkgadaqhaaWcbaGaamyA aaqaaiaaigdacaGGVaGaaGOmaaaaaOGaayjkaiaawMcaaiabgkHiTi abeo8aZnaaCaaaleqabaGaaGOmaaaakmaaqafabeWcbaGaamyAaiab gIGiolaadwfaaeqaniabggHiLdGccaWG6bWaaSbaaSqaaiaadMgaae qaaOGaaiilaiaaywW7caaMf8UaaGzbVlaaywW7caaMf8Uaaiikaiaa isdacaGGUaGaaGOmaiaacMcaaaa@84A6@

Voir l’annexe pour des renseignements détaillés. Si le coefficient de variation de b i MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9 vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=x fr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGaamOyamaaBa aaleaacaWGPbaabeaaaaa@37E8@ est faible, l’approximation par développement en série de Taylor donne N 2 b i 1 / 2 b i 1 / 2 1 + ( 1 / 4 ) C b 2 . MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9 vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=x fr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGaamOtamaaCa aaleqabaGaeyOeI0IaaGOmaaaakmaaqaeabeWcbeqab0GaeyyeIuoa kiaadkgadaqhaaWcbaGaamyAaaqaaiabgkHiTiaaigdacaGGVaGaaG OmaaaakmaaqaeabeWcbeqab0GaeyyeIuoakiaadkgadaqhaaWcbaGa amyAaaqaaiaaigdacaGGVaGaaGOmaaaakiabgIKi7kaaigdacqGHRa WkdaqadaqaamaalyaabaGaaGymaaqaaiaaisdaaaaacaGLOaGaayzk aaGaam4qamaaDaaaleaacaWGIbaabaGaaGOmaaaakiaac6caaaa@4FEB@ En appliquant cela, ainsi que les mêmes approximations qu’à la sous-section 3.1, (4.2) devient

A V e s t s = σ 2 C f 1 N 2 c ¯   z ¯ ( 1 + 1 4 C b 2 ) ( 1 + 1 4 C c 2 ) ( 1 + 1 4 C z 2 ) . ( 4.3 ) MathType@MTEF@5@5@+= feaagKart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9 vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=x fr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGaamyqaiaadA fadaWgaaWcbaGaamyzaiaadohacaWG0bGaam4CaaqabaGccqGH9aqp daWcaaqaaiabeo8aZnaaCaaaleqabaGaaGOmaaaakiaadoeadaqhaa WcbaGaamOzaaqaaiabgkHiTiaaigdaaaGccaWGobWaaWbaaSqabeaa caaIYaaaaOGabm4yayaaraqcLbeacaqGGaGcceWG6bGbaebadaqada qaaiaaigdacqGHRaWkdaWcaaqaaiaaigdaaeaacaaI0aaaaiaadoea daqhaaWcbaGaamOyaaqaaiaaikdaaaaakiaawIcacaGLPaaaaeaada qadaqaaiaaigdacqGHRaWkdaWcaaqaaiaaigdaaeaacaaI0aaaaiaa doeadaqhaaWcbaGaam4yaaqaaiaaikdaaaaakiaawIcacaGLPaaada qadaqaaiaaigdacqGHRaWkdaWcaaqaaiaaigdaaeaacaaI0aaaaiaa doeadaqhaaWcbaGaamOEaaqaaiaaikdaaaaakiaawIcacaGLPaaaaa GaaiOlaiaaywW7caaMf8UaaGzbVlaaywW7caaMf8Uaaiikaiaaisda caGGUaGaaG4maiaacMcaaaa@6A30@

Voir l’annexe pour des renseignements détaillés.

La comparaison de (4.3) et (3.7) montre qu’utiliser des paramètres de coûts estimés au lieu de faire totalement abstraction des coûts a pour effet de multiplier la variance anticipée par [ 1 + ( 1 / 4 ) C b 2 ] / [ 1 + ( 1 / 4 ) C c 2 ] . MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9 vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=x fr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaWaaSGbaeaada WadaqaaiaaigdacqGHRaWkdaqadaqaamaalyaabaGaaGymaaqaaiaa isdaaaaacaGLOaGaayzkaaGaam4qamaaDaaaleaacaWGIbaabaGaaG OmaaaaaOGaay5waiaaw2faaaqaamaadmaabaGaaGymaiabgUcaRmaa bmaabaWaaSGbaeaacaaIXaaabaGaaGinaaaaaiaawIcacaGLPaaaca WGdbWaa0baaSqaaiaadogaaeaacaaIYaaaaaGccaGLBbGaayzxaaaa aiaac6caaaa@4942@ Par conséquent, l’utilisation de l’information sur les coûts est valable à condition que C b < C c . MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9 vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=x fr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGaam4qamaaBa aaleaacaWGIbaabeaakiabgYda8iaadoeadaWgaaWcbaGaam4yaaqa baGccaGGUaaaaa@3B68@ Le coefficient de variation des facteurs d’erreur doit être plus faible que celui des coûts par unité réels sur l’ensemble de la population.

Précédent | Suivant

Date de modification :