S OME E LEMENTARY S TATISTICAL I NFERENCES Chapter 5: Some Elementary Statistical Inferences • Sampling Statistics • Order Statistics • More on Confidence Intervals • Introduction to Hypothesis Tests S OME E LEMENTARY S TATISTICAL I NFERENCES Sampling and Statistics • In a typical statistical problem, we have a random variable X of interest but its pdf f (x) or pmf p(x) is not known. But the form of f (x) or p(x) is known down to a parameter θ (maybe a vector). – X has an exponential distribution, exp(θ). – X has b(n, p) distribution, where n is known but p is unknown. – X has Gamma distribution Γ(α, β), where both α and β are unknown. • We want to estimate θ from a random sample. S OME E LEMENTARY S TATISTICAL I NFERENCES An Motivation Example • An urn contains m balls labeled from 1 to m. The experiment is to choose a ball at random and record the number. Let X denote the number. Then the distribution of X is given by 1 P (X = x) = , x = 1, . . . , m. m We do not know m and want to estimate m. So, θ = m. We take a sample of n balls denoted by X = {X1 , . . . , Xm }. S OME E LEMENTARY S TATISTICAL I NFERENCES • Sampling with Replacement: Here a ball is selected at random, its number is recorded, the ball is replaced in the urn, the balls are then remixed. • Sampling without Replacement: Here a ball is selected at random. If the balls are selected one-at-a-time, they are not replaced after each draw. • When m is much larger that n, the sampling schemes are practically the same. S OME E LEMENTARY S TATISTICAL I NFERENCES • To estimate m we draw X1 , . . . , Xn with replacement. The distribution of Xi is P (Xi = x) = 1/m for x = 1, . . . , m. An intuitive estimator of m is the statistics T = max{X1 , . . . , Xn }. How far is T from m? Prove that T is consistent estimator of m, T →P m. • In the above problem, E(X) = (m + 1)/2. Hence, E(2X̄ − 1) = m. Perhaps, 2X̄ − 1 is also a good estimator of m as it is unbiased. We will show that T is a better estimator in this case. S OME E LEMENTARY S TATISTICAL I NFERENCES An Example • Suppose X is a random variable with unknown mean θ. Let X1 , . . . , Xn be a random sample from the distribution of X and let X̄ be the sample mean. Since E(X̄) = θ, X̄ is a point estimator of θ. But how far is X̄ to θ? We will study the general case later. Now for a specific case where X ∼ N (θ, σ 2 ). So, X̄ ∼ N (θ, σ 2 /n). √ (X̄ − θ)/(σ/ n) has a standard normal distribution. So, X̄ − θ √ < 2) 0.954 = P (−2 < σ/ n σ σ = P (X̄ − 2 √ < θ < X̄ + 2 √ ) n n It says that before the sample is drawn the probability that the random S OME E LEMENTARY S TATISTICAL I NFERENCES − 2 √σn , X̄ + 2 √σn ) traps θ is 0.954. After the sample is drawn the realized interval has either traps θ or it has not. But because interval (X̄ of the high probability of success, namely 0.954, before the sample is drawn, we call this interval a 95.4% confidence interval for θ .