Suppose that X is a discrete random variable with: P(X=0) = (2/3)theta P(X=1) = (1/3)theta P(X=2) = (2/3)(1-theta) P(X=3) = (1/3)(1-theta) where theta: [0,1] is a parameter. The following 10 independent observations were taken from such a distribution: (3,0,2,1,3,2,1,0,2,1) a. Find the moment of methods estimate of theta b. Find an approximate standard error for your estimate c. What is the maximum likelihood estimate of theta? d. What is an approximate standard error of the maximum likelihood estimate? ๐(๐ = 0) = 2 ๐ 3 ๐(๐ = 1) = 1 ๐ 3 ๐(๐ = 2) = 2 (1 − ๐) 3 1 (1 − ๐) 3 ๐ธ(๐) = 0 × ๐(๐ = 0) + 1 × ๐(๐ = 1) + 2 × ๐(๐ = 2) + 3 × ๐(๐ = 3) ๐(๐ = 3) = 2 1 2 1 = 0 × ๐ + 1 × ๐ + 2 × (1 − ๐) + 3 × (1 − ๐) 3 3 3 3 = ๐(1/3 − 4/3 − 3/3) + 4/3 + 3/3 = −2๐ + 7/3 ๐ธ (๐ 2 ) = 02 × ๐(๐ = 0) + 12 × ๐(๐ = 1) + 22 × ๐(๐ = 2) + 32 × ๐(๐ = 3) 2 1 2 1 = 0 × ๐ + 1 × ๐ + 4 × (1 − ๐) + 9 × (1 − ๐) 3 3 3 3 16 17 = ๐(1/3 − 8/3 − 9/3) + 8/3 + 9/3 = − ๐ + 3 3 16 17 2 2 2 ๐ = ๐๐๐(๐) = ๐ธ (๐ ) − [๐ธ(๐)] = − ๐ + − (−2๐ + 7/3)2 3 3 16 17 28 49 2 2 =− ๐+ − (4๐ − ๐ + ) = −4๐ 2 + 4๐ + 3 3 3 9 9 3,0,2,1,3,2,1,0,2,1 Sample size = ๐ = 10 3+0+2+1+3+2+1+0+2+1 15 3 Sample mean ๐ฅฬ = = = 10 10 2 To find moment estimate of ๐, equate population mean and sample mean. ๐ธ(๐) = ๐ฅฬ −2๐ + 7/3 = 3/2 7 3 14 − 9 2๐ = − = = 5/6 3 2 6 5 ๐= 12 5 The moment estimate of ๐ is ๐ฬ = 12 7 In general −2๐ + = ๐ฅฬ 3 7 2๐ = − ๐ฅฬ 3 7 1 ๐ฬ = − ๐ฅฬ 6 2 3 When ๐ฅฬ = , 2 ๐ฬ = The variance of ๐ฬ is 7 3 5 − = 6 4 12 7 1 1 2 1 ๐2 ฬ ( ) ๐๐๐(๐ ) = ๐ฃ๐๐( − ๐ฅฬ ) = ๐๐๐(๐ฅฬ ) = × 6 2 2 4 ๐ 1 2 (−4๐ 2 + 4๐ + ) = 40 9 Substituting ๐ by its moment estimate ๐ฬ, The estimated variance of ๐ฬ is 1 2 1 25 5 2 (−4๐ฬ 2 + 4๐ฬ + ) = (−4 × +4× + ) 40 9 40 144 12 9 1 −100 + 240 + 32 1 172 43 ( )= ( )= = = 0.029861 40 144 40 144 1440 ฬ (๐ฬ) = ๐๐๐ The estimated standard error of ๐ฬ is √0.029861 = 0.1728 Maximum likelihood estimation The likelihood function is ๐2 ๐3 2 ๐0 1 ๐1 2 1 ๐ฟ(๐/๐ฅ1 , … , ๐ฅ๐ ) = ๐ (๐ = ๐ฅ1 ) … ๐(๐ = ๐ฅ๐ ) = ( ๐) ( ๐) ( (1 − ๐)) ( (1 − ๐)) 3 3 3 3 ๐0+๐1 ( ๐2+๐3 = ๐๐๐๐ ๐ก๐๐๐ก × ๐ 1 − ๐) (f0, f1, f2, f3 are respectively the frequencies of 0,1,2, and 3.) ๐๐๐ ๐ฟ = ๐๐๐๐ ๐ก๐๐๐ก + (๐0 + ๐1)๐๐๐ ๐ + (๐2 + ๐3)๐๐๐(1 − ๐) ๐ ๐๐๐๐ฟ ๐0 + ๐1 ๐2 + ๐3 = − =0 ๐๐ ๐ 1−๐ (๐0 + ๐1)(1 − ๐) − (๐2 + ๐3)๐ = 0 (๐0 + ๐1) = (๐0 + ๐1 + ๐2 + ๐3)๐ = ๐๐ ๐0 + ๐1 ๐= ๐ The maximum likelihood estimate of ๐ is ๐0 + ๐1 ๐ฬ = ๐ ๐ is the number of observations and f0 and f1 are respectively the number of 0’s and number of 1’s. 3,0,2,1,3,2,1,0,2,1 In this sample, f0 = 2, f1 =3, f2 = 3, f3 = 2, n = 10. ๐ฬ = ๐0 + ๐1 2 + 3 5 = = = 0.5 ๐ 10 10 Consider the event ๐ = 0 ๐๐ 1 The frequency of this event is f0+f1. 2 1 3 3 The probability of this event is ๐ + ๐ = ๐ The random variable f0+f1 follows binomial distribution with parameters ๐ and ๐. ๐ธ(๐0 + ๐1) = ๐๐ ๐๐๐(๐0 + ๐1) = ๐๐(1 − ๐) ๐๐๐(๐ฬ) = ฬ 1 ๐(1 − ๐) ๐๐๐(๐0 + ๐1) = 2 ๐ ๐ ฬ ๐(1−๐) 0.5(1−0.5) 0.25 Estimated standard error of ๐ฬ = √ ๐ = √ 10 = √ 10 = √0.025 = 0.1581 SUMMARY 5 Moment estimate of ๐ = ๐ฬ = 12 = 0.4167 Standard error of ๐ฬ = 0.1728 Maximum likelihood estimate of ๐ = ๐ฬ = 0.5 Standard error of ๐ฬ = 0.1581