Bayes and EB benchmarkig for SAE. Dissertation 2012.pdf

# Note that λ ij need not be the same as w ij

• 86

This preview shows page 47 - 49 out of 86 pages.

Note that Λ ij need not be the same as W ij . Similarly, Ω i need not be the same as Γ i . Also, p is given. 47

Subscribe to view the full document.

Proof. By standard results, the problem reduces to minimization of g = summationdisplay i summationdisplay j ( ˆ θ ij ˆ θ B ij ) T Λ ij ( ˆ θ ij ˆ θ B ij ) + summationdisplay i ( e i ¯ ˆ θ B iw ) T Ω i ( e i ¯ ˆ θ B iw ) 2 summationdisplay i λ 1i T ( summationdisplay j W ij ˆ θ ij e i ) 2 λ 2 T ( summationdisplay i Γ i e i p ), where λ 1i and λ 2 are the Lagrange multipliers. We solve 0 = g ˆ θ ij = 2Λ ij ( ˆ θ ij ˆ θ B ij ) 2 W ij λ 1i i , j (4.1.16) 0 = g e i = 2Ω i ( e i ¯ ˆ θ B iw ) + 2 λ 1i i λ 2 i . (4.1.17) From ( 4.1.16 ), ˆ θ ij = ˆ θ B ij + Λ 1 ij W ij λ 1i = e i = ¯ ˆ θ B iw + s i λ 1i ⇐⇒ λ 1i = s 1 i ( e i ¯ ˆ θ B iw ) i . (4.1.18) From ( 4.1.17 ), e i = ¯ ˆ θ B iw + Ω 1 i Γ i λ 2 Ω 1 i λ 1i ⇐⇒ e i = ¯ ˆ θ B iw + Ω 1 i Γ i λ 2 Ω 1 i s 1 i ( e i ¯ ˆ θ B iw ) ⇐⇒ Ω 1 i Γ i λ 2 = ( I + Ω 1 i s 1 i )( e i ¯ ˆ θ B iw ) = e i = ¯ ˆ θ B iw + (Ω i + s 1 i ) 1 Γ i λ 2 (4.1.19) Applying constraint (ii), we find p = ¯ ˆ θ B w + i Γ i i + s 1 i ) 1 Γ i λ 2 = ¯ ˆ θ B w + R λ 2 , where R := i Γ i i + s 1 i ) 1 Γ i . This implies that λ 2 = R 1 ( p ¯ ˆ θ B w ). From ( 4.1.19 ), e i = ¯ ˆ θ B iw + (Ω i + s 1 i ) 1 Γ i R 1 ( p ¯ ˆ θ B w ). (4.1.20) Combining ( 4.1.18 ) and ( 4.1.20 ), ˆ θ ij = ˆ θ B ij + Λ 1 ij W ij λ 1i ⇐⇒ ˆ θ ij = ˆ θ B ij + Λ 1 ij W ij s 1 i ( e i ¯ ˆ θ B iw ) = ˆ θ ij = ˆ θ B ij + Λ 1 ij W ij s 1 i i + s 1 i ) 1 Γ i R 1 ( p ¯ ˆ θ B w ). (4.1.21) 48
The result follows from ( 4.1.20 ) and ( 4.1.21 ). 4.2 An Example This section considers small area/domain estimation of the proportion of persons without health insurance for several domains of the Asian subpopulation. Our goal is to benchmark the aggregated probabilities that a person does not have health insurance to the corresponding domain values. We also benchmark the domain estimates to match the overall population estimates. The small domains were constructed on the basis of age, sex, race, and the region where each person lives. The National Health Interview Survey (NHIS) data provides the individual-level binary response data as well as the individual-level covariates ( Ghosh, Kim, Sinha, Maiti, Katzoff and Parsons ( 2009 )). We have information on the main response variable of interest, whether or not a person has health insurance. More information on this data is described in Ghosh, Kim, Sinha, Maiti, Katzoff and Parsons ( 2009 ). Moreover, when targeting specific subpopulations crossclassified by demo- graphic characteristics, direct estimates are usually accompanied with large standard errors and coefficients of variation. Hence, a procedure such as the one proposed in Section 2 is appropriate.
You've reached the end of this preview.
• Spring '16
• Yessi
• The Land, Estimation theory, Mean squared error, Bayes estimator, Empirical Bayes method

{[ snackBarMessage ]}

### What students are saying

• As a current student on this bumpy collegiate pathway, I stumbled upon Course Hero, where I can find study resources for nearly all my courses, get online help from tutors 24/7, and even share my old projects, papers, and lecture notes with other students.

Kiran Temple University Fox School of Business ‘17, Course Hero Intern

• I cannot even describe how much Course Hero helped me this summer. It’s truly become something I can always rely on and help me. In the end, I was not only able to survive summer classes, but I was able to thrive thanks to Course Hero.

Dana University of Pennsylvania ‘17, Course Hero Intern

• The ability to access any university’s resources through Course Hero proved invaluable in my case. I was behind on Tulane coursework and actually used UCLA’s materials to help me move forward and get everything together on time.

Jill Tulane University ‘16, Course Hero Intern