Nouvel estimateur de la variance de l’estimateur par le ratio avec corrections de petit échantillon
Section 4. Conclusion

Dans cet article, nous avons dégagé une nouvelle formule d’approximation de EQM ( Y ¯ ^ R ) MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrpgpC0xc9LqFf0xc9 qqpeuf0xe9q8qiYRWFGCk9vi=dbbf9v8Gq0db9qqpm0dXdHqpq0=vr 0=vr0=edbaqaaeGaciGaaiaabeqaamaabaabaaGcbaaeaaaaaaaaa8 qacaqGfbGaaeyuaiaab2eacaaMc8Uaaeika8aadaqiaaqaa8qaceWG zbWdayaaraaacaGLcmaadaWgaaWcbaWdbiaadkfaa8aabeaak8qaca qGPaaaaa@3E87@ de l’ordre 1 / n 2 , MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrpgpC0xc9LqFf0xc9 qqpeuf0xe9q8qiYRWFGCk9vi=dbbf9v8Gq0db9qqpm0dXdHqpq0=vr 0=vr0=edbaqaaeGaciGaaiaabeqaamaabaabaaGcbaaeaaaaaaaaa8 qadaWcgaqaaiaaigdaaeaacaWGUbWdamaaCaaaleqabaWdbiaaikda aaaaaOGaaiilaaaa@399D@ ainsi qu’une nouvelle formule pour le biais de s e ^ 2 MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrpgpC0xc9LqFf0xc9 qqpeuf0xe9q8qiYRWFGCk9vi=dbbf9v8Gq0db9qqpm0dXdHqpq0=vr 0=vr0=edbaqaaeGaciGaaiaabeqaamaabaabaaGcbaaeaaaaaaaaa8 qacaWGZbWdamaaDaaaleaapeGabmyza8aagaqcaaqaa8qacaaIYaaa aaaa@3930@ de l’ordre 1 / n . MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrpgpC0xc9LqFf0xc9 qqpeuf0xe9q8qiYRWFGCk9vi=dbbf9v8Gq0db9qqpm0dXdHqpq0=vr 0=vr0=edbaqaaeGaciGaaiaabeqaamaabaabaaGcbaaeaaaaaaaaa8 qadaWcgaqaaiaaigdaaeaacaWGUbaaaiaac6caaaa@388D@ Le nouvel estimateur EQM ^ 2 ( Y ¯ ^ R ), MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrpgpC0xc9LqFf0xc9 qqpeuf0xe9q8qiYRWFGCk9vi=dbbf9v8Gq0db9qqpm0dXdHqpq0=vr 0=vr0=edbaqaaeGaciGaaiaabeqaamaabaabaaGcbaWaaecaaeaaqa aaaaaaaaWdbiaabweacaqGrbGaaeytaaWdaiaawkWaamaaBaaaleaa peGaaGOmaaWdaeqaaOWdbiaaykW7caqGOaWdamaaHaaabaWdbiqadM fapaGbaebaaiaawkWaamaaBaaaleaapeGaamOuaaWdaeqaaOWdbiaa bMcacaqGSaaaaa@4128@ qui tient compte du biais de s e ^ 2 , MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrpgpC0xc9LqFf0xc9 qqpeuf0xe9q8qiYRWFGCk9vi=dbbf9v8Gq0db9qqpm0dXdHqpq0=vr 0=vr0=edbaqaaeGaciGaaiaabeqaamaabaabaaGcbaaeaaaaaaaaa8 qacaWGZbWdamaaDaaaleaapeGabmyza8aagaqcaaqaa8qacaaIYaaa aOWdaiaacYcaaaa@39F9@ paraît moins entaché d’un biais que EQM ^ 0 ( Y ¯ ^ R )= Var ^ ( Y ¯ ^ R ) MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrpgpC0xc9LqFf0xc9 qqpeuf0xe9q8qiYRWFGCk9vi=dbbf9v8Gq0db9qqpm0dXdHqpq0=vr 0=vr0=edbaqaaeGaciGaaiaabeqaamaabaabaaGcbaWaaecaaeaaqa aaaaaaaaWdbiaabweacaqGrbGaaeytaaWdaiaawkWaamaaBaaaleaa peGaaGimaaWdaeqaaOWdbiaaykW7caqGOaWdamaaHaaabaWdbiqadM fapaGbaebaaiaawkWaamaaBaaaleaapeGaamOuaaWdaeqaaOWdbiaa bMcacqGH9aqppaWaaecaaeaapeGaaeOvaiaabggacaqGYbaapaGaay PadaWdbiaaykW7caqGOaWdamaaHaaabaWdbiqadMfapaGbaebaaiaa wkWaamaaBaaaleaapeGaamOuaaWdaeqaaOWdbiaabMcaaaa@4B33@ et EQM ^ 1 ( Y ¯ ^ R ). MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrpgpC0xc9LqFf0xc9 qqpeuf0xe9q8qiYRWFGCk9vi=dbbf9v8Gq0db9qqpm0dXdHqpq0=vr 0=vr0=edbaqaaeGaciGaaiaabeqaamaabaabaaGcbaWaaecaaeaaqa aaaaaaaaWdbiaabweacaqGrbGaaeytaaWdaiaawkWaamaaBaaaleaa peGaaGymaaWdaeqaaOWdbiaaykW7caqGOaWdamaaHaaabaWdbiqadM fapaGbaebaaiaawkWaamaaBaaaleaapeGaamOuaaWdaeqaaOWdbiaa bMcacaGGUaaaaa@412A@ Pour n8, MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrpgpC0xc9LqFf0xc9 qqpeuf0xe9q8qiYRWFGCk9vi=dbbf9v8Gq0db9qqpm0dXdHqpq0=vr 0=vr0=edbaqaaeGaciGaaiaabeqaamaabaabaaGcbaaeaaaaaaaaa8 qacaWGUbGaeyyzImRaaGioaiaacYcaaaa@3A42@ le biais de EQM ^ 2 ( Y ¯ ^ R ) MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrpgpC0xc9LqFf0xc9 qqpeuf0xe9q8qiYRWFGCk9vi=dbbf9v8Gq0db9qqpm0dXdHqpq0=vr 0=vr0=edbaqaaeGaciGaaiaabeqaamaabaabaaGcbaWaaecaaeaaqa aaaaaaaaWdbiaabweacaqGrbGaaeytaaWdaiaawkWaamaaBaaaleaa peGaaGOmaaWdaeqaaOWdbiaaykW7caqGOaWdamaaHaaabaWdbiqadM fapaGbaebaaiaawkWaamaaBaaaleaapeGaamOuaaWdaeqaaOWdbiaa bMcaaaa@4079@ était de moins de 7 % dans tous les cas de cette simulation, ce qui est bien mieux que pour l’estimateur de variance standard; le plus souvent, ce résultat s’obtient même pour n4. MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrpgpC0xc9LqFf0xc9 qqpeuf0xe9q8qiYRWFGCk9vi=dbbf9v8Gq0db9qqpm0dXdHqpq0=vr 0=vr0=edbaqaaeGaciGaaiaabeqaamaabaabaaGcbaaeaaaaaaaaa8 qacaWGUbGaeyyzImRaaGinaiaac6caaaa@3A40@ Pour un n MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrpgpC0xc9LqFf0xc9 qqpeuf0xe9q8qiYRWFGCk9vi=dbbf9v8Gq0db9qqpm0dXdHqpq0=vr 0=vr0=edbaqaaeGaciGaaiaabeqaamaabaabaaGcbaaeaaaaaaaaa8 qacaWGUbaaaa@370A@ très petit, EQM ^ 2 ( Y ¯ ^ R ) MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrpgpC0xc9LqFf0xc9 qqpeuf0xe9q8qiYRWFGCk9vi=dbbf9v8Gq0db9qqpm0dXdHqpq0=vr 0=vr0=edbaqaaeGaciGaaiaabeqaamaabaabaaGcbaWaaecaaeaaqa aaaaaaaaWdbiaabweacaqGrbGaaeytaaWdaiaawkWaamaaBaaaleaa peGaaGOmaaWdaeqaaOWdbiaaykW7caqGOaWdamaaHaaabaWdbiqadM fapaGbaebaaiaawkWaamaaBaaaleaapeGaamOuaaWdaeqaaOWdbiaa bMcaaaa@4079@ peut être entaché d’un important biais négatif si la population présente un grand coefficient de variation C x . MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrpgpC0xc9LqFf0xc9 qqpeuf0xe9q8qiYRWFGCk9vi=dbbf9v8Gq0db9qqpm0dXdHqpq0=vr 0=vr0=edbaqaaeGaciGaaiaabeqaamaabaabaaGcbaaeaaaaaaaaa8 qacaWGdbWdamaaBaaaleaapeGaamiEaaWdaeqaaOGaaiOlaaaa@38F2@ À en juger par notre étude de simulation, ce problème semble improbable dans la mesure où C x <0,8. MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrpgpC0xc9LqFf0xc9 qqpeuf0xe9q8qiYRWFGCk9vi=dbbf9v8Gq0db9qqpm0dXdHqpq0=vr 0=vr0=edbaqaaeGaciGaaiaabeqaamaabaabaaGcbaaeaaaaaaaaa8 qacaWGdbWdamaaBaaaleaapeGaamiEaaWdaeqaaOWdbiabgYda8iaa bcdacaqGSaGaaeioaiaac6caaaa@3C23@

Rappelons enfin que, pour les populations de cette étude, le biais de l’estimateur par le ratio même était considérablement petit même pour n=4. MathType@MTEF@5@5@+= feaagKart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr 4rNCHbGeaGqiFu0Je9sqqrpepC0xbbL8F4rqqrpgpC0xc9LqFf0xc9 qqpeuf0xe9q8qiYRWFGCk9vi=dbbf9v8Gq0db9qqpm0dXdHqpq0=vr 0=vr0=edbaqaaeGaciGaaiaabeqaamaabaabaaGcbaaeaaaaaaaaa8 qacaWGUbGaeyypa0JaaGinaiaac6caaaa@3980@ En règle générale, ce même biais pourrait ne pas être négligeable pour d’autres populations. Cochran (1977, pages 174 et 175) examine plusieurs autres estimateurs par le ratio sans biais.

Bibliographie

Cochran, W.G. (1977). Sampling Techniques. New York : John Wiley & Sons, Inc.

David, I.P., et Sukhatme, B.V. (1974). On the bias and mean square error of the ratio estimator. Journal of the American Statistical Association, 69, 464-466.

Kendall, M.G., et Stuart, A. (1958). The Advanced Theory of Statistics, Volume I. Londres : Charles Griffin and Company.

Kish , L. (1995). Survey Sampling. New York : John Wiley & Sons, Inc.

Koop, J.C. (1968). An exercise in ratio estimation. The American Statistician, 22, 29-30.

Nath, S.N. (1968). On product moments from a finite universe. Journal of the American Statistical Association, 63, 535-541.

Rao, J.N.K. (1969). Ratio and regression estimators. Dans New Developments in Survey Sampling, (Éds., N.L. Johnson et H. Smith), New York : John Wiley & Sons, Inc., 213-234.

Sukhatme, P.V. (1954). Sampling Theory of Surveys with Applications, Iowa State College Press, Ames , IA.

Tin, M. (1965). Comparison of some ratio estimators. Journal of the American Statistical Association, 60, 294-307.


Date de modification :