Chapter 6 Parameter estimation The previous chapter has pointed out that the basic problem of mathematical statistics is inferring the distribution of totality through sample, which is called statistical inference. The main elements of statistical inference including parameter estimation and statistical hypothesis testing, they constitute the core of mathematical statistics. This chapter focuses on parameter estimation Methods and criteria for evaluation of the estimated amount, and focus on the classical methods of point estimation as well as the normal population parameter interval estimation. § 6.1 Point estimation Population is characterized by population distribution. In practical problems, we can often Determine the type of population distribution through relevant expertise and previous experience, but the parameters are unknown. We often estimate the parameters by the sample, which is known as parameter estimation. Set the distribution function of population X is F ( x) , which contains the unknown parameters ,we call any function ( X 1 ,..., X n ) of a sample point estimator. 1 Moment estimate The method of moments is the oldest method of finding point estimators. If population distribution contains k unknown parameters, method of moments estimator are found by equating the first k sample moments to the corresponding k population moments, and solving the resulting system of simultaneous equations. Example 6.1 suppose ( X1 , X 2 ,..., X n ) are iid P( ) ,We have 1 X , 1 , according to the method of moments, we yield the moments estimator of is X . Example 6.2 suppose ( X1 , X 2 ,..., X n ) are iid N ( , ) , 2 We have 1 X , 2 1 n 2 X i , 1 , 2 2 2 ,hence we must solve the equation n i 1 X , 2 1 n 2 2 Xi , n i 1 Solving for , 2 yields the moments estimator of , 2 is X , 2 n 1 2 1 n S ( X i X )2 , n n i 1 Example 6.3 suppose ( X1 , X 2 ,..., X n ) are iid,and the common pdf is 1 1 (1 1 ) x , x c, c f ( x; ) 0, 其它, 0 1, c is a known positive constant,we have 1 EX xf ( x; )dx 1 xc c Let X ,we yield the 1 x 1 (1 ) dx c , 1 moment estimator of is 1 c . X 2 Maximum likelihood estimates Consider a random sample ( X1 , X 2 ,..., X n ) from a distribution having pdf or probability distribution ( x; ) , (1 ,..., m ) are unknown parameters, The joint n pdf of ( X1 , X 2 ,..., X n ) is ( x ; ) ,this may be regard as a function of i i 1 ,when so regarded, it is called the likelihood function L of the random sample, and we write n L( x1 ,..., xn ; ) ( xi ; ) , i 1 Suppose that we can find a nontrivial function of ( x1 , x2 ,..., xn ) ,say u( x1 , x2 ,..., xn ) ,when is replaced by u( x1 , x2 ,..., xn ) ,the likelihood function L is a maximum, then the statistic u ( X1 , X 2 ,..., X n ) will be called a maximum likelihood estimator of ,and will be denoted by the symbol u ( X 1 , X 2 ,..., X n ) . The function L can be maximized by setting the first partial derivation of ln L ,with respect to ,equal to zero, that is to say ln L 0, 1 ln L 0. m and solving the resulting equation for of . ,which is the maximum likelihood estimator If the first partial derivation of ln L is not exist, it will be solved by the definition of maximum likelihood estimator. Example 6.4 Let ( X1 , X 2 ,..., X n ) denote a random sample from a distribution that is P( ) ,we shall find , the maximum likelihood estimator of .The probability distribution of X is p( x; ) P( x) e x x! , x 0,1, 2,..., the likelihood function L is n n L e i 1 xi xi ! e n xi i 1 n x ! i 1 , i so n n i 1 i 1 ln L ( xi ) ln ( ln xi !) n , setting d ln L 1 n xi n 0 , d i 1 Solving the equation, we can obtain that the maximum likelihood estimator of is 1 n xi x . n i 1 . Example 6.5 Let ( X1 , X 2 ,..., X n ) denote a random sample from a distribution that is N ( , 2 ), , 2 are unknown parameters, we shall find ˆ , ˆ 2 , the maximum likelihood estimators of n L i 1 , 2 .we can easily obtain the likelihood function L is ( xi )2 1 exp[ ] (2 )n 2 ( 2 )n 2 exp[ 2 2 2 2 2 n 1 (x i 1 i )2 ] , So n n 1 LnL ln(2 ) ln 2 2 2 2 2 Setting n (x i 1 i )2 , 1 n LnL 2 ( xi n ) 0, i 1 n LnL n 1 ( xi ) 2 0 , 2 2 2 2 2 2( ) i 1 Solving the equation, we can obtain that the maximum likelihood estimator of , 2 ,is 1 n Xi X , n i 1 n 2 1 ( X X ) 2 . i n i 1 Example 6.6 If the probability distribution of X is 0 2 1 2 3 , 2 2 (1 ) 1 2 1 2 ( 0 ) is unknown parameter, we observed the following eight values of X , 3, 1, 3, 0, 3, 1, 2, 3, Find the moment estimate and the maximum likelihood estimate of . From the distribution of X ,we can easily obtain that EX 3 4 ,setting 3 4 x ,we have that the moment estimator of 1 4 is (3 X ) . According to the sample values, we have x 2 ,so the moment estimate of is 1 . 4 the likelihood function L is 8 L( ) P( X xi ) 4 6 (1 ) 2 (1 2 ) 4 , i 1 so ln L( ) ln 4 6 ln 2 ln(1 ) 4 ln(1 2 ) , Setting d ln L( ) 6 2 8 0, d 1 1 2 Solving the equation, we have estimate of is 7 13 7 13 1 , so the maximum likelihood .because 12 12 2 7 13 . 12 Example 6.7 Let ( X1 , X 2 ,..., X n ) denote a random sample from a distribution that is U [a, b] , Find the maximum likelihood estimators of a , b .The pdf of population is 1 , a x b, p( x) b a 0, 其它, So we can obtained the likelihood function L is 1 , a xi b, i 1, 2,..., n, L (b a) n 0, 其它. Which is an ever-decreasing function of b and an ever-increasing function of a ,The maximum of such functions cannot be found by differentiation but by selecting b as small as possible and a as large as possible. Now a each xi and b each xi ,so a min( xi ) and b max( xi ) .so the maximum likelihood estimators of a , b is a min( X 1 ,..., X n ), b max( X 1 ,..., X n ). 3, Methods of evaluating estimators The methods discussed in the previous section have outlined reasonable techniques for finding point estimators of parameters. A difficulty that arises,however,is that since we can usually apply more than one of these methods in a particular situation, we are often faced with the task of choosing between estimators. of course, it is possible that different methods of finding estimators will yield the same answer, which makes evaluation a bit easier. but in many cases, different methods will lead to different estimators. 3.1 Unbiased estimator If an estimator ( X 1 ,..., X n ) satisfies that E ( X 1 ,..., X n ) ,the estimator is called unbiased. Because estimator is a random variable, different sample lead to different estimate, the character of unbiased mean that a good estimator should equal to the parameter in mean meaning. Example 6.8 Let ( X1 , X 2 ,..., X n ) X , EX , DX 2 ,Prove that X and S estimator of denote 2 , 2 respect. Proof EX E ( 1 n 1 n 1 X ) EX i .n , i n i 1 n i 1 n a random sample from 1 n ( X i X ) 2 are the unbiased n 1 i 1 DX D( ES 2 1 n 1 Xi ) 2 n i 1 n E[ n DX i 1 i 1 2, n 1 n ( X i X )2 ] n 1 i 1 n 1 E{ [( X i ) ( X )]2 } n 1 i 1 1 n [ E ( X i )2 2 E ( X i )( X ) E ( X )2 ] n 1 i 1 1 n 2 1 ( 2 2 2 ) n 1 i 1 n n 2. In Example 6.7, we can easily know that 1 1 ( X 1 X 2 ) and ( X 1 X 2 X 3 ) are 2 3 both the unbiased estimators of ,which is better? So we should search another standard to evaluate estimators. 3.2 Efficient estimator Let 1 and 2 are both the unbiased estimator of ,if D 1 D 2 , We say 1 is more efficient than 2 .If the number of sample n is fixed, and the variance of is less than or equal to the variance of every other unbiased estimator of ,we say is the efficient estimator of 1 2 . Because D[ ( X 1 X 2 )] more efficient than 1 2 1 1 1 D[ ( X 1 X 2 X 3 )] 2 ,so ( X 1 X 2 ) is 2 3 3 2 1 ( X1 X 2 X 3 ) . 3 Example 6.9 Consider a random sample ( X1 , X 2 ,..., X n ) from a distribution having pdf 1 x e , x 0, p( x) 0 , ot her s, Z min( X1 , , X n ) , Prove that X and nZ ,and when n 1, X is more efficient than nZ . are both the unbiased estimators of Proof we can easily calculate that EX ,so EX ,that is to say X is the unbiased estimator of .On the other side, n FZ ( z ) P( Z z ) 1 P( Z z ) 1 P( X i z ) 1 e nz ,z 0, i 1 n n We have Z ~ E ( ) ,so EZ of , E (nZ ) , that is to say nZ is the unbiased estimator too. In order to compare the efficiency of X and nZ ,we can also easily calculate that D X 2 , D( X ) 2 , D(nZ ) n 2 D ( Z ) n 2 .( ) 2 2 , when n 1, n n D( X ) 2 2 D(nZ ) , n So X is more efficient than nZ .. 3.3 Uniform estimator If for any 0 ,we have lim P(| n | ) 1 , n We call n is the uniform estimator of . §6.2 Interval estimation Generally we have two kinds of forms to estimate a parameter, One is by a point, which is introduced in§6.1,the other is by a interval. for example, we estimate the age of a person is between 35 and 40;we called this kind of estimator interval estimator. An interval estimator of a real-valued parameter is any pair of function 1 ( X 1 ,..., X n ) 2 ( X 1 ,..., X n ) and of a sample that satisfy 1 ( X 1 ,..., X n ) 2 ( X 1 ,..., X n ) for all sample x. if the sample x is observed, the inference 1 ( X 1 ,..., X n ) 2 ( X 1 ,..., X n ) is made. The random interval (1 , 2 ) is called an interval estimator; If P(1 ( X 1 ,..., X n ) 2 ( X 1 ,..., X n )) 1 ,we call the confidence coefficient of (1 , 2 ) is 1 .we often set =0.01,0.05,0.1 etc. 1, Interval estimator for mean in normal population 1.1 Let ( X1 , X 2 ,..., X n ) denote a random sample from a distribution that is N ( , 2 ), 2 is known, is unknown parameter, What is the interval estimator of with the confidence coefficient 1 ? Set U We have U X / n , N (0,1), By the definition of u /2 we have P(| U | u /2 ) 1 , P(| X / n | u /2 ) 1 , P( X u /2 . / n X u /2 . / n ) 1 , the interval estimator of with the confidence coefficient 1 is ( X u /2 . / n , X u /2 . / n ) . Example 6.10 Let a random sample of size 10 from the normal distribution X N ( ,0.22 ), They are 18.3, 17.5, 18.1, 17.7, 17.9, 18.5, 18, 18.1, 17.8, 17.9, Determine a 95 per cent confidence interval for . Solution 1 0.95 , 0.05 , n 10, 0.2, u /2 =1.96, u /2 . / sample, we have n =0.124,By the x 17.98 ,so the 95 per cent confidence interval for is (17.98-0.124, 17.98+0.124)= (17.856, 18.104). 1.2 Let ( X1 , X 2 ,..., X n ) denote a random sample from a distribution that is N ( , 2 ), and 2 are unknown parameters, What is the interval estimator of with the confidence coefficient 1 ? Set T X , S / n We have T ~ t (n 1) ,so P(| T | t /2 (n 1)) 1 , Hence the interval estimator of with the confidence coefficient 1 is ( X t /2 (n 1)S / n , X t /2 (n 1)S / n ) . Example 6.11 Let a random sample of size 9 from the normal distribution N ( , 2 ) ,They are 0.497,0.506,0.518,0.524,0.488,0.510,0.510,0.515,0.512, Determine a 99 per cent confidence interval for . Solution 1 0.99 , 0.01, n 9 , t /2 (8) =3.3554,From the sample data ,we have x 0.5089 , S 0.0109 , t /2 (n 1)S / n =0.0122, so the 99 per cent confidence interval for is (0.5089-0.0122, 0.5089+0.0122)=(0.4967,0.5211). 2, Interval estimator for variance in normal population Let ( X1 , X 2 ,..., X n ) denote a random sample from a distribution that is N ( , 2 ), and 2 are unknown parameters, What is the interval estimator of 2 with the confidence coefficient 1 ? Set 2 We have 1 2 n ( X i 1 X) 2 i (n 1) S 2 2 , 2 ~ 2 (n 1) ,so P( 12 /2 (n 1) 2 2 /2 (n 1)) 1 , n n 2 ( X i X )2 (Xi X ) P i 1 2 2 i 12 1 /2 (n 1) /2 (n 1) So the interval estimator of 2 1 , with the confidence coefficient 1 n n 2 ( X X ) ( X i X )2 i i 1 2 , i 12 ( n 1) 1 /2 (n 1) /2 is . Example 6.12 Let a random sample of size 9 from the normal distribution N ( , 2 ) ,They are 600, 612, 598, 583, 609, 607, 592, 588, 593, Determine a 95 per cent confidence interval for . 2 Solution From the sample data, we can easily calculate that x 598 , S 98.5 , 2 n ( x x) 2 i i 1 ( n 1) S 2 =788, 0.975 (8) 2.18, 0.025 (8) 17.5 ,so the 95 per cent 2 confidence interval for 2 788 788 , ) ,that is to say (45.03,361.47). 17.5 2.18 2 is ( 3, Interval estimator for difference of means and ratio of variance in two normal populations Let ( X 1 , X 2 ,..., X n1 ) denote a random sample of size n1 from a distribution that 2 is N ( 1 , 1 ), (Y1 , Y2 ,..., Yn2 ) denote a random sample of size n2 from a distribution that is N ( 2 , 2 ), where 1 , 1 , 2 , 2 are 2 2 2 unknown parameters, the two random samples are independent, What is the interval estimator of 1 2 and 1 / 2 with the 2 2 confidence coefficient 1 ? 3.1 Difference of means If 1 2 ,but 2 2 2 T 2 is unknown, set ( X Y ) ( 1 2 ) S , 1 1 n1 n2 So T ~ t (n1 n2 2) ,hence P(| T | t /2 (n1 n2 2)) 1 , So the interval estimator of 1 2 with the confidence coefficient 1 is ( X Y t /2 (n1 n2 2) S 1 1 1 1 , X Y t /2 (n1 n2 2) S ). n1 n2 n1 n2 3.2 ratio of variance Set F So F S12 S22 12 22 , F (n1 1, n2 1) .hence P( F1 2 (n1 1, n2 1) F F 2 (n1 1, n2 1) 1 , So the interval estimator of 1 / 2 with the confidence coefficient 1 is 2 2 ( S1 2 S2 2 S1 2 S2 2 , ) F 2 (n1 1, n2 1) F1 2 (n1 1, n2 1) Example 6.13 Let a random sample of size 4 from the normal distribution N ( 1 , 2 ) ,They are 0.143 0.142 0.143 0.137 Another random sample of size 5 from the normal distribution N ( 2 , ) ,They are 2 0.140 0.142 0.136 0.138 0.140 The two random samples are independent; determine the interval estimator of 1 2 and 12 / 22 with the confidence coefficient 0.95. Solution From the sample data, we can easily calculate that X 0.14125, S1 0.00000825, Y 0.1392, S2 0.0000052 , t0.025 (7) 2.36 , 2 2 F0.025 (3, 4) 9.98, F0.975 (3, 4) 1 1 0.0662 , F0.025 (4,3) 15.1 so S 3 0.00000825 4 0.0000052 0.00255091 , 452 Hence the interval estimator of 1 2 with the confidence coefficient 0.95 is (-0.002,0.006) and the interval estimator of 1 / 2 with the confidence coefficient 2 2 0.95 is (0.159,23.957). Exercises 1. Let ( X 1 , , X n ) be a random sample from the distribution having pdf ( 1) x ,0 x 1, , f ( x) 0, ot her s, Find the moment estimator and the maximum likelihood estimator of 2. Let ( X 1 , . , X n ) be a random sample from the distribution having pdf e x , x 0, ( 0) , f ( x; ) 0, x 0 Find the moment estimator and the maximum likelihood estimator of . 3. Let ( X 1 , , X n ) be a random sample from the distribution having pdf 3x 2 , 0 x , f ( x) 3 0, 0, 其它 Find the moment estimator of . 4. Let ( X 1 , , X n ) be a random sample from the Geometry distribution having probability distribution P( X k ) p(1 p) k 1 , k 1,2,,0 p 1, Find the maximum likelihood estimator of p . 5. Let ( X 1 , , X n ) be a random sample from the distribution having pdf 2e2( x ) , x , f ( x; ) x , 0, 0 is unknown parameter. Setting min( X 1 , , X n ) , where (1) Determine the distribute function of X ; (2) Determine the distribute function of (3) is the unbiased estimator of ; or not? 6. Let be the mean of population, ( X 1 , X 2 , X 3 ) is a random sample from the population distribution, show 2 1 2 X1 X 2 X 3 , 5 5 5 1 1 1 2 X 1 X 2 X 3 , 6 3 2 1 3 9 3 X 1 X 2 X 3 , 7 14 14 The unbiased estimators of ; (1) 1 (2)Compare the efficiency of 1 , 2 , 3 . 7. X Let a random sample of size 6 from the normal distribution N ( , 0.25), they are 14.70, 15.21, 14.90, 14.91, 15.32, 15.31, Determine a 95 per cent confidence interval for . 8. Let a random sample of size 9 from the normal distribution N ( , ) ,They 2 are 21.1, 21.3, 21.4, 21.5, 21.3, 21.7, 21.4, 21.3, 21.6, Determine a 99 per cent confidence interval for . 2 9. Let a random sample of size 8 from the normal distribution N ( 1 , ) ,They 2 are 0.143 0.142 0.143 0.137 Another random sample of size 9 from the normal distribution N ( 2 , ) ,They are 2 0.140 0.142 0.136 0.138 0.140 The two random samples are independent; determine the interval estimator of 1 2 and 1 / 2 with the confidence coefficient 0.90. 2 2 10. Let ( X1 , X 2 ,..., X n ) denote a random sample from a distribution that is N ( ,9) ,If the length of the interval estimator of with the confidence coefficient 1 is no longer than 2.How many sizes of the sample at least should be in following two situations (1) 0.1时; (2) 0.01时.