Statistics 512 Notes 2 Confidence Intervals Definition: For a sample X 1 , , X n from a model {P , } , a (1 ) confidence interval for a parameter g ( ) is an interval Cn [a( X1 ,..., X n ), b( X1 ,..., X n )] such that P ( g ( ) Cn ) 1 for all . In words, Cn is a function of the random sample that traps the parameter g ( ) with probability at least (1 ) . Commonly, people use 95% confidence intervals which corresponds to choosing 0.05 . Example: Suppose X1 , X 2 , X 3 , X 4 are iid N ( ,1) . The interval C [ X 1, X 1] is a 0.9544 confidence interval for : P ( [ X 1, X 1]) P( X 1 X 1) =P( 1 X 1) X 2) 1/ 4 =P( 2 Z 2) =0.9544 =P( 2 Motivation for confidence intervals: A confidence interval can be thought of as an estimate of the parameter, i.e., we estimate by [ X 1, X 1] rather than the point estimate X . What is gained by the using interval rather than the point estimate since the interval is less precise? We gain confidence. We have the assurance that in 95.44% of repeated samples, the confidence interval will contain . In practice, confidence intervals are usually used along with point estimates to give a sense of the accuracy of the point estimate. Interpretation of confidence intervals A confidence interval is not a probability statement about g ( ) since is a fixed parameter, not a random variable. Common textbook interpretation: If we repeat the experiment over and over, a 95% confidence interval will contain the parameter 95% of the time. This is correct but not particularly useful since we rarely repeat the same experiment over and over. More useful interpretation (Wasserman, All of Statistics) : On day 1, you collect data and construct a 95 percent confidence interval for a parameter 1 . On day 2, you collect new data and construct a 95 percent confidence interval for an unrelated parameter 2 . On day 3, you collect new data and construct a 95 percent confidence interval for an unrelated parameter 3 . You continue this way constructing 95 percent confidence intervals for a sequence of unrelated parameters 1 , 2 , Then 95 percent of your intervals will trap the true parameter value. Confidence interval is not a probability statement about : The fact that a confidence interval is not a probability statement about is confusing. Let be a fixed, known real number and let X1 , X 2 be iid random variables such that P( X i 1) P( X i 1) 1/ 2 . Now define Yi X i and suppose we only observe Y1 , Y2 . Define the following “confidence interval” which actually contains only one point: {Y1 1} if Y1 Y2 C {(Y1 Y2 ) / 2} if Y1 Y2 No matter what is, we have P ( C ) 3/ 4 so this is a 75 percent confidence interval. Suppose we now do the experiment and we get Y1 15 and Y2 17 . Then our 75 percent confidence interval is {16}. However, we are certain that is 16. Some common confidence intervals 1. CI for mean of normal distribution with known variance: X 1 , , X n iid N ( , 2 ) where 2 known. X Then ~ N (0,1) n 1 Let z ( ) where is the CDF of a standard normal random variable, e.g., z.975 1.96 . We have X 1 P z z 1 1 2 2 n P z X z X 1 n 2 1 2 n P X z X z 1 1 n n 2 2 Thus, X z1 2 (1 ) CI for n is a 2. CI for mean of normal distribution with unknown variance. X 1 , , X n iid N ( , 2 ) where 2 unknown. X T S Key fact: The random variable , where n 1 n 2 S2 ( X X ) i , has a Student’s t-distribution n 1 i 1 with n-1 degrees of freedom. (Section 3.6.3, page 186) Let t ,n be the inverse of the CDF of the Student’s tdistribution with n degrees of freedomevaluated at . Note t t1 Following the same steps as above, we have X 1 P t t 1 , n 2 1 2 ,n S n S S P t X t X 1 , n n 2 1 2 ,n n S S P X t X t 1 , n 1 , n n n 2 2 S X t (1 ) CI for Thus, 1 , n n is a 2 Note: t1 ,n z1 so we pay a price for not knowing the 2 2 variance but as n , t1 ,n z1 . 2 2 3. CI for mean of iid sample from unknown distribution: Central Limit Theorem (Theorem 4.4.1): For an iid sample from a distribution that has mean and positive variance X Yn n 2 , the random variable converges in n distribution to a standard normal random variable. Slutsky’s Theorem (Theorem 4.3.5): D P P D X n X , An a, Bn b, then An Bn X n a bX . 4 From the weak law of large numbers, if E ( X ) P 1 n 2 2 2 S ( X X ) i . n 1 i 1 Thus, combining Slutsky’s Theorem and the central limit theorem, Xn D N (0,1) S n S X z An approximate (1 ) CI for is n 1 / 2 n because X 1 P z z 1 S 2 1 2 n S S P z X z X 1 n 2 1 2 n S S P X z X z 1 1 n n 2 2 Application: A food-processing company is considering marketing a new spice mix for Creole and Cajun cooking. They interview 200 consumers and find that 37 would purchase such a product. Find an approximate 95% confidence interval for p, the true proportion of buyers.