Lecture 11. Central Limit Theorem Juwon Seo Juwon Seo Lecture 11. Central Limit Theorem 1 / 24 Announcement Today’s lecture is the last lecture of this course. Well done guys! Juwon Seo Lecture 11. Central Limit Theorem 2 / 24 Outline Today we will learn I Central limit theorem I Interval estimation (confidence intervals) I Hypothesis testing Juwon Seo Lecture 11. Central Limit Theorem 3 / 24 Estimation -continued Suppose we have an iid sequence of random variables {X1 , X2 , X3 , . . . Xn }. n is the sample size. I We know a good (unbiased, efficient and consistent) estimator for E(X) : n 1X X̄n = Xi . n i=1 I Although we know that X̄n →p E(X) and E(X̄n ) = E(X), we have no clue on how X̄n is close to E(X). If you are unlucky, it is possible for us to have X̄n far away from the true mean E(X). I Although we can calculate X̄n from our observations, we CANNOT say X̄n = E(X). Juwon Seo Lecture 11. Central Limit Theorem 4 / 24 Estimation -continued Maybe instead, can we say something like...? I The probability that E(X) is between a and b is 0.95 or, I We are 95% confident that E(X) is between a and b. We can find such an interval (a, b)! This is called an interval estimator. Juwon Seo Lecture 11. Central Limit Theorem 5 / 24 Central Limit Theorem Assuming we have iid random variables {X1 , X2 , X3 , . . . Xn } and each of them follows N(µ, σ 2 ). Since we know that 1 X̄n has a normal distribution, 2 E(X̄n ) = µ and, 2 3 Var(X̄n ) = σn , the distribution of the random variable X̄n − µ σ √ n is N(0, 1). Juwon Seo Lecture 11. Central Limit Theorem 6 / 24 Central Limit Theorem From the Standard Normal Table, we can find a constant c such that X̄n − µ √ ≤ c = 0.95 P −c ≤ σ/ n Juwon Seo Lecture 11. Central Limit Theorem 7 / 24 Central Limit Theorem From the Standard Normal Table, we can find a constant c such that X̄n − µ √ ≤ c = 0.95 P −c ≤ σ/ n c is 1.96: X̄n − µ P −1.96 ≤ √ ≤ 1.96 = 0.95 σ n Juwon Seo Lecture 11. Central Limit Theorem 8 / 24 Central Limit Theorem X̄n − µ √ ≤ 1.96 = P X̄n − 1.96n−1/2 σ ≤ µ ≤ X̄n + 1.96n−1/2 σ P −1.96 ≤ σ/ n If σ is known, we can say I µ is between X̄n − 1.96n−1/2 σ and X̄n + 1.96n−1/2 σ with the probability 0.95 or, I we are 95% sure that µ is between X̄n − 1.96n−1/2 σ and X̄n + 1.96n−1/2 σ. (X̄n − 1.96n−1/2 σ, X̄n + 1.96n−1/2 σ) is called the 95% confidence interval. Juwon Seo Lecture 11. Central Limit Theorem 9 / 24 Central Limit Theorem We may also define the 90% confidence interval and the 80% confidence interval for µ: I (X̄n − 1.645n−1/2 σ, X̄n + 1.645n−1/2 σ) is the 90% confidence interval. I (X̄n − 1.28n−1/2 σ, X̄n + 1.28n−1/2 σ) is the 80% confidence interval. Juwon Seo Lecture 11. Central Limit Theorem 10 / 24 Central Limit Theorem How do we assume that we know σ 2 ? In practice, we do not know σ 2 but we can estimate it by the sample variance! As the sample size gets larger, the sample variance σ̂ 2 converges in probability to σ 2 . Therefore X̄n − µ σ̂ √ n also has the same limit distribution N(0, 1). If σ 2 is unknown, we find the 95% confidence interval as (X̄n − 1.96n−1/2 σ̂, X̄n + 1.96n−1/2 σ̂). Juwon Seo Lecture 11. Central Limit Theorem 11 / 24 Central Limit Theorem Can we assume that each of {X1 , X2 , X3 , . . . Xn } follows a normal distribution? Actually, we do not know that each of {Xi } has a normal distribution. Sometimes it is not normal! However, the Central Limit Theorem justifies that we can use the same confidence interval (X̄n − 1.96n−1/2 σ̂, X̄n + 1.96n−1/2 σ̂) for any iid random variables {X1 , X2 , X3 , . . . Xn } (with any distribution), when the sample size is large enough. Juwon Seo Lecture 11. Central Limit Theorem 12 / 24 Central Limit Theorem Theorem Suppose we have an iid sequence of random variables {X1 , X2 , X3 , . . . Xn }. We denote the common expected value by µ and the common variance by σ 2 . We define the sample mean X̄n by: n 1X Xi . X̄n = n i=1 The Central Limit Theorem says √ X̄n − µ n(X̄n − µ) √ = →d N(0, 1), σ σ/ n as n → ∞. Juwon Seo Lecture 11. Central Limit Theorem 13 / 24 Central Limit Theorem (Another version) Theorem Suppose we have an iid sequence of random variables {X1 , X2 , X3 , . . . Xn }. We denote the common expected value by µ and the common variance by σ 2 . We define the sample mean X̄n by: n 1X Xi . X̄n = n i=1 The Central Limit Theorem says n 1 X √ n i=1 Xi − µ σ →d N(0, 1), as n → ∞. Juwon Seo Lecture 11. Central Limit Theorem 14 / 24 Central Limit Theorem (Another version) I Here I randomly drew the samples {Xi } from the Bernoulli( 13 ) distribution the sample size n. As n increases, we can see P fixing −(1/3) that √1n ni=1 Xi√ approaches N(0, 1). 2/3 Juwon Seo Lecture 11. Central Limit Theorem 15 / 24 Central Limit Theorem A restaurant wants to examine the mean of a dinner check. A random sample of 50 dinners showed an average bill of $36 and the sample variance of $122 . We want to find the confidence interval for the true mean. You may assume that the sample size 50 is large enough to apply the Central Limit Theorem. Juwon Seo Lecture 11. Central Limit Theorem 16 / 24 Central Limit Theorem Find 95% confidence interval for the population mean of the dinner check for all costumers. Note that n−1/2 σ̂ = √12 50 ≈ 1.697, thus, (X̄n −1.96n−1/2 σ̂, X̄n +1.96n−1/2 σ̂) = (36−1.96×1.697, 36+1.96×1.697) Interpretation:µ is between 32.67 and 39.33 with probability 0.95. Juwon Seo Lecture 11. Central Limit Theorem 17 / 24 Central Limit Theorem Find 90% confidence interval for the population mean of the dinner check for all costumers. (X̄n −1.645n−1/2 σ̂, X̄n +1.645n−1/2 σ̂) = (36−1.645×1.697, 36+1.645×1.697) Interpretation:µ is between 33.21 and 38.79 with probability 0.9. Juwon Seo Lecture 11. Central Limit Theorem 18 / 24 Hypothesis testing Suppose that you collected iid samples and calculated the sample mean, which is 0.02. Based on what you observe, you want determine if you can say µ = 0. H0 : µ = 0. I Is our sample mean close enough to 0 such that we can conclude that µ = 0? I Or is 0.02 far away from 0? Juwon Seo Lecture 11. Central Limit Theorem 19 / 24 Hypothesis testing Statistical test: I H0 : µ = 0 is called the null hypothesis. I Idea: If µ̂ is close enough to 0, we accept the null hypothesis (some textbook says, “we cannot reject the null hypothesis") and conclude that µ = 0. Otherwise, we reject the null hypothesis and conclude that µ 6= 0. I How do we define being close enough? If the confidence interval of µ (which has µ̂ as the center) includes 0, we can conclude that µ̂ is close to 0. Juwon Seo Lecture 11. Central Limit Theorem 20 / 24 Hypothesis testing In the previous example, let us test if H0 : µ = 35. I Since 95% confidence interval does include 35, we accept the null hypothesis (or we cannot reject the null hypothesis) and say (µ = 35) at the 95% confidence level. I Since 90% confidence interval does include 35, we accept the null hypothesis (or we cannot reject the null hypothesis) and say (µ = 35) at the 90% confidence level. Juwon Seo Lecture 11. Central Limit Theorem 21 / 24 Hypothesis testing What if we test for the null hypothesis H0 : µ = 33? I Since 95% confidence interval does include 33, we accept the null hypothesis and say (µ = 33) at the 95% confidence level. I Since 90% confidence interval does not include 33, we reject the null hypothesis and say (µ 6= 33) at the 90% confidence level. Juwon Seo Lecture 11. Central Limit Theorem 22 / 24 Hypothesis testing Assuming that the random sample has 150 diners now, and that the average bill and sample variance remain unchanged: the sample mean is 36 and the sample variance of 122 . Verify the following hypotheses at the 95% confidence level. I H0 : µ = 34 I H0 : µ = 35 Juwon Seo Lecture 11. Central Limit Theorem 23 / 24 Hypothesis testing I The 95% confidence interval is (36 − 1.96 ∗ 150 − 1/2 ∗ 12, 36 + 1.96 ∗ 150 − 1/2 ∗ 12) = (36 − 1.920, 36 + 1.920) = (34.08, 37.92) I Reject (H0 : µ = 34) at the 95% confidence level. We cannot conclude that µ = 34 at the 95% confidence level. I Accept (H0 : µ = 35) at the 95% confidence level. We can conclude that µ = 35 at the 95% confidence level. Juwon Seo Lecture 11. Central Limit Theorem 24 / 24