8 ESTIMATION Basic Concepts Random samples Definition 8.1. A random sample is any sequence, X1 , X2 , . . . , Xn of independent, identically distributed random variables. If a continuous random variable X has an unknown parameter θ, we often write it’s probability density function as fX (x; θ). Example: Let X have an exponential distribution with probability density function ( fX (x; θ) = 1 − θ1 x θe 0 x>0 otherwise where θ is unknown. Then a random sample, X1 , X2 , . . . , Xn representing n independent observations from this random variable has a joint probability density function given by ( fX1 ,X2 ,...,Xn (x1 , x2 , . . . , xn ; θ) = 1 − θ1 (x1 +x2 +···+xn ) θe 0 xi > 0 for all i otherwise Statistics Definition 8.2. A statistic is any function, h(X1 , X2 , . . . , Xn ), of independent, identically distributed random variables. If a statistic is being used to establish statistical inferences about a particular unknown parameter θ, then it is often written as θ̂(X1 , X2 , . . . , Xn ). Note: Any function of a random sample can be called a statistic. 225 226 Point Estimation Example: Let X1 , X2 , . . . , Xn be a random sample. You can think of the random variables X1 , X2 , . . . , Xn as representing the future data values that will be obtained by making n independent observations of our experiment. Then the following are statistics: 1. θ̂(X1 , X2 , . . . , Xn ) = n 1X Xi n i=1 The random variable θ̂ is the average of the observations. 2. θ̂(X1 , X2 , . . . , Xn ) = n Y Xi i=1 The random variable θ̂ is the product of the observations. 3. θ̂(X1 , X2 , . . . , Xn ) = max(X1 , X2 , . . . , Xn ) The random variable θ̂ is the largest of the observations. 4. θ̂(X1 , X2 , . . . , Xn ) = X1 The random variable θ̂ is the first observation, with the others discarded. There are two important statistics that arise so frequently that they are given special symbols: Definition 8.3. Let X1 , X2 , . . . , Xn be a random sample. The sample mean is the statistic n 1X X≡ Xi n i=1 and the sample variance is the statistic s2X ≡ n 1 X (Xi − X)2 . n − 1 i=1 Point Estimation The problem Let X be a random variable with cumulative distribution function FX (x; θ) that is completely determined except for the unknown constant θ. A random sample Finding Estimators 227 X1 , X2 , . . . , Xn is to be drawn from this distribution. Suppose that a statistic θ̂(X1 , X2 , . . . , Xn ) has been already been selected for use. Let x1 , x2 , . . . , xn denote the actual observed values taken on by X1 , X2 , . . . , Xn . Applying the function θ̂ to these observations yields the number θ̂(x1 , x2 , . . . , xn ). This number can be thought of as an actual observation of the random variable θ̂(X1 , X2 , . . . , Xn ). We will call the number θ̂(x1 , x2 , . . . , xn ) our estimate for the value of θ. We call the random variable θ̂(X1 , X2 , . . . , Xn ) an estimator for θ. Since any statistic can be called an estimator of any parameter, we will have to find ways to distinguish good estimators from poor estimators. Unbiased estimators Definition 8.4. The estimator θ̂ is an unbiased estimator of θ if its expected value equals θ, i.e., E(θ̂) = θ for all θ. Example: Let X be a discrete random variable with probability mass function given by pX (0) = 1 − θ and pX (1) = θ, where θ is unknown. If X1 , X2 , . . . , Xn is a random sample from X then θ̂(X1 , X2 , . . . , Xn ) = n 1X Xi n i=1 is an unbiased estimator for θ. Finding Estimators Method of maximum likelihood Definition 8.5. Let X1 , X2 , . . . , Xn be a random sample from a discrete random variable X with probability mass function pX (x; θ). The likelihood function for X1 , X2 , . . . , Xn is given by L(x1 , x2 , . . . , xn ; θ) = n Y i=1 pX (xi ; θ). 228 Self-Test Exercises for Chapter 8 If, instead, X is a continuous random variable with probability density function fX (x; θ), then the likelihood function is given by L(x1 , x2 , . . . , xn ; θ) = n Y fX (xi ; θ). i=1 Maximum likelihood estimators are obtained by finding that value of θ that maximizes L for a given set of observations x1 , x2 , . . . , xn . Since the value of θ that does this will usually vary with x1 , x2 , . . . , xn , θ can be thought of as a function of x1 , x2 , . . . , xn , namely θ̂(x1 , x2 , . . . , xn ). To evaluate the properties of θ̂, we can look at its performance prior to actually making the observations x1 , x2 , . . . , xn . That is we can substitute Xi for xi in the specification for θ̂ and look at its properties as the statistic θ̂(X1 , X2 , . . . , Xn ). For example, one of the properties that we might like to check for is whether θ̂ is an unbiased estimator for θ (i.e., check to see if E(θ̂) = θ). Self-Test Exercises for Chapter 8 For the following multiple-choice question, choose the best response among those provided. The answer can be found in Appendix B. S8.1 Suppose that X1 , X2 , . . . , Xn are independent identically distributed random variables each with marginal probability density function ¡ ¢ 1 x−µ 2 1 fXi (x) = √ e− 2 σ σ 2π for −∞ < x < +∞, where σ > 0. Then an unbiased estimator for µ is (A) (X1 )(X2 ) · · · (Xn ) (B) (X1 + X2 )2 /2 (C) 1 n Pn i=1 Xi (D) σ (E) none of the above. Questions for Chapter 8 229 Questions for Chapter 8 8.1 Let X be a random variable with a binomial distribution, i.e., Ã ! pX (k ; θ) = n k θ (1 − θ)n−k k for k = 0, 1, . . . , n. Let X1 be a random sample of size 1 from X . (a) Show that θ̂ = X1 /n is an unbiased estimator for θ. (b) Show that θ̂ = X1 /n is the maximum likelihood estimator for θ. 8.2 Let Y be an estimator for θ based on the random sample X1 , X2 , . . . , Xn . P Suppose that E(Xi ) = θ and Y = ni=1 ai Xi where a1 , a2 , . . . , an are constants. What constraint must be placed on a1 , a2 , . . . , an in order for Y to be an unbiased estimator for θ? 8.3 The life of a light bulb is a random variable X which has probability density function ( 1 − θ1 x x>0 e θ fX (x; θ) = 0 otherwise Let X1 , X2 , . . . , Xn be a random sample from X . (a) Find an estimator for θ using the method of maximum likelihood. (b) Is the estimator for part (a) unbiased? Justify your answer. (c) Find the maximum likelihood estimator for η = 1/θ. 8.4 The number of typographical errors, X , on a page of text has a Poisson distribution with parameter λ, i.e., pX (k ; λ) = λk e−λ , k! for k = 0, 1, 2, . . .. A random sample of n pages are observed (a) Find an estimator for λ using the method of maximum likelihood. (b) Is the estimator for part (a) unbiased? Justify your answer. 230 Questions for Chapter 8 8.5 Given the random sample X1 , X2 , . . . , Xn consider the statistic d2 formed by averaging the squared differences of all possible pairings of {Xi , Xj }. ¡ ¢ There are n2 such pairs. That statistic can be represented as 1 X d2X ≡ ¡n¢ (Xi − Xj )2 2 Prove that d2X = 2s2X . i>j ANSWERS TO SELFTEST EXERCISES Chapter 8 S8.1 C