Dongfeng Li. Autumn 2010

Transcription

1 Autumn 2010

2 Chapter Contents Some statistics background; ; Comparing means and proportions; variance. Students should master the basic concepts, descriptive statistics measures and graphs, basic hypothesis testing, basic analysis of variance.

3 Chapter Contents Some statistics background; ; Comparing means and proportions; variance. Students should master the basic concepts, descriptive statistics measures and graphs, basic hypothesis testing, basic analysis of variance.

4 Chapter Contents Some statistics background; ; Comparing means and proportions; variance. Students should master the basic concepts, descriptive statistics measures and graphs, basic hypothesis testing, basic analysis of variance.

5 Chapter Contents Some statistics background; ; Comparing means and proportions; variance. Students should master the basic concepts, descriptive statistics measures and graphs, basic hypothesis testing, basic analysis of variance.

6 Chapter Contents Some statistics background; ; Comparing means and proportions; variance. Students should master the basic concepts, descriptive statistics measures and graphs, basic hypothesis testing, basic analysis of variance.

7 Section Contents Review statistics concepts: Distribution, discrete distribution, continuouse distribution, PDF, CDF, quantile. Normal distribution., median, variance, standard deviation, interquantile range, skewness, kurtosis. Population, parameter, sample, statistics, estimates, sampling distribution. MLE, standard error. Hypothesis tests, two types of errors, p-value.

8 Section Contents Review statistics concepts: Distribution, discrete distribution, continuouse distribution, PDF, CDF, quantile. Normal distribution., median, variance, standard deviation, interquantile range, skewness, kurtosis. Population, parameter, sample, statistics, estimates, sampling distribution. MLE, standard error. Hypothesis tests, two types of errors, p-value.

9 Section Contents Review statistics concepts: Distribution, discrete distribution, continuouse distribution, PDF, CDF, quantile. Normal distribution., median, variance, standard deviation, interquantile range, skewness, kurtosis. Population, parameter, sample, statistics, estimates, sampling distribution. MLE, standard error. Hypothesis tests, two types of errors, p-value.

10 Section Contents Review statistics concepts: Distribution, discrete distribution, continuouse distribution, PDF, CDF, quantile. Normal distribution., median, variance, standard deviation, interquantile range, skewness, kurtosis. Population, parameter, sample, statistics, estimates, sampling distribution. MLE, standard error. Hypothesis tests, two types of errors, p-value.

11 Section Contents Review statistics concepts: Distribution, discrete distribution, continuouse distribution, PDF, CDF, quantile. Normal distribution., median, variance, standard deviation, interquantile range, skewness, kurtosis. Population, parameter, sample, statistics, estimates, sampling distribution. MLE, standard error. Hypothesis tests, two types of errors, p-value.

12 Section Contents Review statistics concepts: Distribution, discrete distribution, continuouse distribution, PDF, CDF, quantile. Normal distribution., median, variance, standard deviation, interquantile range, skewness, kurtosis. Population, parameter, sample, statistics, estimates, sampling distribution. MLE, standard error. Hypothesis tests, two types of errors, p-value.

13 Distribution Random Variable: Discrete, such as sex, patient/control, age group. Continuous, such as weight, blood pressure. Distribution: used to describe the relative chance of taking some value. For discrete variable X, use P(X = xi ), where {x i } are the value set of X. Called probability mass function(pmf). For continuous variable X, use the probability density function(pdf) f (x), where P(X (x ɛ, x + ɛ) f (x)(2ɛ).

14 Distribution Random Variable: Discrete, such as sex, patient/control, age group. Continuous, such as weight, blood pressure. Distribution: used to describe the relative chance of taking some value. For discrete variable X, use P(X = xi ), where {x i } are the value set of X. Called probability mass function(pmf). For continuous variable X, use the probability density function(pdf) f (x), where P(X (x ɛ, x + ɛ) f (x)(2ɛ).

15 Distribution Random Variable: Discrete, such as sex, patient/control, age group. Continuous, such as weight, blood pressure. Distribution: used to describe the relative chance of taking some value. For discrete variable X, use P(X = xi ), where {x i } are the value set of X. Called probability mass function(pmf). For continuous variable X, use the probability density function(pdf) f (x), where P(X (x ɛ, x + ɛ) f (x)(2ɛ).

16 Distribution Random Variable: Discrete, such as sex, patient/control, age group. Continuous, such as weight, blood pressure. Distribution: used to describe the relative chance of taking some value. For discrete variable X, use P(X = xi ), where {x i } are the value set of X. Called probability mass function(pmf). For continuous variable X, use the probability density function(pdf) f (x), where P(X (x ɛ, x + ɛ) f (x)(2ɛ).

17 Distribution Random Variable: Discrete, such as sex, patient/control, age group. Continuous, such as weight, blood pressure. Distribution: used to describe the relative chance of taking some value. For discrete variable X, use P(X = xi ), where {x i } are the value set of X. Called probability mass function(pmf). For continuous variable X, use the probability density function(pdf) f (x), where P(X (x ɛ, x + ɛ) f (x)(2ɛ).

18 Distribution Random Variable: Discrete, such as sex, patient/control, age group. Continuous, such as weight, blood pressure. Distribution: used to describe the relative chance of taking some value. For discrete variable X, use P(X = xi ), where {x i } are the value set of X. Called probability mass function(pmf). For continuous variable X, use the probability density function(pdf) f (x), where P(X (x ɛ, x + ɛ) f (x)(2ɛ).

19 CDF Cumulative distribution function(cdf) F(x): F(x) = P(X x) P(x (a, b]) = F(b) F(a) For discrete distribution, F(x) = P(X = x i ) x i x For continuous distribution, F(x) = x f (t)dt

20 CDF Cumulative distribution function(cdf) F(x): F(x) = P(X x) P(x (a, b]) = F(b) F(a) For discrete distribution, F(x) = P(X = x i ) x i x For continuous distribution, F(x) = x f (t)dt

21 CDF Cumulative distribution function(cdf) F(x): F(x) = P(X x) P(x (a, b]) = F(b) F(a) For discrete distribution, F(x) = P(X = x i ) x i x For continuous distribution, F(x) = x f (t)dt

22 Quantile function Quantile function: the inverse of CDF q(p) = F 1 (p), p (0, 1) if F (x) is 1-1 mapping. Generally, q(p) = x p where P(X x p ) p, P(X x p ) 1 p (x p can be non-unique.)

23 Quantile function Quantile function: the inverse of CDF q(p) = F 1 (p), p (0, 1) if F (x) is 1-1 mapping. Generally, q(p) = x p where P(X x p ) p, P(X x p ) 1 p (x p can be non-unique.)

24 The normal distribution Standard normal distribution(n(0,1)), PDF ϕ(x) = 1 2π e x2 2 Normal distribution N(µ, σ 2 ), PDF f (x) = ϕ( x µ σ ) = 1 2πσ exp{ (x µ)2 2σ 2 }

25 The normal distribution Standard normal distribution(n(0,1)), PDF ϕ(x) = 1 2π e x2 2 Normal distribution N(µ, σ 2 ), PDF f (x) = ϕ( x µ σ ) = 1 2πσ exp{ (x µ)2 2σ 2 }

26 Numerical charasteristics PDF is a curve with infinite number of points. We can use some numbers to describe the key part of a distribution. Firstly, the location measurement. The mean EX. The meadian, x0.5 = q(0.5) where (can be non-unique). Variability measurement. P(x x 0.5 ) 0.5, P(x x 0.5 ) 0.5 The standard deviation σ X = E(X EX) 2. Interquantile range q(0.75) q(0.25).

27 Numerical charasteristics PDF is a curve with infinite number of points. We can use some numbers to describe the key part of a distribution. Firstly, the location measurement. The mean EX. The meadian, x0.5 = q(0.5) where (can be non-unique). Variability measurement. P(x x 0.5 ) 0.5, P(x x 0.5 ) 0.5 The standard deviation σ X = E(X EX) 2. Interquantile range q(0.75) q(0.25).

28 Numerical charasteristics PDF is a curve with infinite number of points. We can use some numbers to describe the key part of a distribution. Firstly, the location measurement. The mean EX. The meadian, x0.5 = q(0.5) where (can be non-unique). Variability measurement. P(x x 0.5 ) 0.5, P(x x 0.5 ) 0.5 The standard deviation σ X = E(X EX) 2. Interquantile range q(0.75) q(0.25).

29 Numerical charasteristics PDF is a curve with infinite number of points. We can use some numbers to describe the key part of a distribution. Firstly, the location measurement. The mean EX. The meadian, x0.5 = q(0.5) where (can be non-unique). Variability measurement. P(x x 0.5 ) 0.5, P(x x 0.5 ) 0.5 The standard deviation σ X = E(X EX) 2. Interquantile range q(0.75) q(0.25).

30 Numerical charasteristics PDF is a curve with infinite number of points. We can use some numbers to describe the key part of a distribution. Firstly, the location measurement. The mean EX. The meadian, x0.5 = q(0.5) where (can be non-unique). Variability measurement. P(x x 0.5 ) 0.5, P(x x 0.5 ) 0.5 The standard deviation σ X = E(X EX) 2. Interquantile range q(0.75) q(0.25).

31 Numerical charasteristics PDF is a curve with infinite number of points. We can use some numbers to describe the key part of a distribution. Firstly, the location measurement. The mean EX. The meadian, x0.5 = q(0.5) where (can be non-unique). Variability measurement. P(x x 0.5 ) 0.5, P(x x 0.5 ) 0.5 The standard deviation σ X = E(X EX) 2. Interquantile range q(0.75) q(0.25).

32 Numerical charasteristics PDF is a curve with infinite number of points. We can use some numbers to describe the key part of a distribution. Firstly, the location measurement. The mean EX. The meadian, x0.5 = q(0.5) where (can be non-unique). Variability measurement. P(x x 0.5 ) 0.5, P(x x 0.5 ) 0.5 The standard deviation σ X = E(X EX) 2. Interquantile range q(0.75) q(0.25).

33 Numerical charasteristics PDF is a curve with infinite number of points. We can use some numbers to describe the key part of a distribution. Firstly, the location measurement. The mean EX. The meadian, x0.5 = q(0.5) where (can be non-unique). Variability measurement. P(x x 0.5 ) 0.5, P(x x 0.5 ) 0.5 The standard deviation σ X = E(X EX) 2. Interquantile range q(0.75) q(0.25).

34 Shape characteristics. Skewness ( X EX E σ X ) 3 Symmetric, left-skewed, right-skewed density. Shape characteristics. Kurtosis ( ) X EX 4 E 3 σ X Heavy tail problem. Multimodel distribution. E.g., the weight of 10 year old and 20 year old mixed together.

35 Shape characteristics. Skewness ( X EX E σ X ) 3 Symmetric, left-skewed, right-skewed density. Shape characteristics. Kurtosis ( ) X EX 4 E 3 σ X Heavy tail problem. Multimodel distribution. E.g., the weight of 10 year old and 20 year old mixed together.

36 Shape characteristics. Skewness ( X EX E σ X ) 3 Symmetric, left-skewed, right-skewed density. Shape characteristics. Kurtosis ( ) X EX 4 E 3 σ X Heavy tail problem. Multimodel distribution. E.g., the weight of 10 year old and 20 year old mixed together.

37 Shape characteristics. Skewness ( X EX E σ X ) 3 Symmetric, left-skewed, right-skewed density. Shape characteristics. Kurtosis ( ) X EX 4 E 3 σ X Heavy tail problem. Multimodel distribution. E.g., the weight of 10 year old and 20 year old mixed together.

38 Shape characteristics. Skewness ( X EX E σ X ) 3 Symmetric, left-skewed, right-skewed density. Shape characteristics. Kurtosis ( ) X EX 4 E 3 σ X Heavy tail problem. Multimodel distribution. E.g., the weight of 10 year old and 20 year old mixed together.

39 Population A population is modeled by a random variable or a distribution. Population parameters: unknown numbers which could decide the distribution. E.g., for N(µ, σ 2 ) population, (µ, σ 2 ) are the two unknown population parameters.

40 Population A population is modeled by a random variable or a distribution. Population parameters: unknown numbers which could decide the distribution. E.g., for N(µ, σ 2 ) population, (µ, σ 2 ) are the two unknown population parameters.

41 Sample A sample(typically) is n observations draw independently from the population, X 1, X 2,..., X n. A sample can be regarded as n numbers, or n iid random variables. Statistics: calculate some values to estimate the distribution of the population. Can be regarded as a random variable. Such as sample mean, sample standard deviation. Can be used to estimate unknown population parameters, called estimates. Sampling distribution: The distribution of an estimator or other statistic. E.g., if X 1,..., X n is a sample from N(µ, σ 2 ), then X N(µ, σ 2 /n).

42 Sample A sample(typically) is n observations draw independently from the population, X 1, X 2,..., X n. A sample can be regarded as n numbers, or n iid random variables. Statistics: calculate some values to estimate the distribution of the population. Can be regarded as a random variable. Such as sample mean, sample standard deviation. Can be used to estimate unknown population parameters, called estimates. Sampling distribution: The distribution of an estimator or other statistic. E.g., if X 1,..., X n is a sample from N(µ, σ 2 ), then X N(µ, σ 2 /n).

43 Sample A sample(typically) is n observations draw independently from the population, X 1, X 2,..., X n. A sample can be regarded as n numbers, or n iid random variables. Statistics: calculate some values to estimate the distribution of the population. Can be regarded as a random variable. Such as sample mean, sample standard deviation. Can be used to estimate unknown population parameters, called estimates. Sampling distribution: The distribution of an estimator or other statistic. E.g., if X 1,..., X n is a sample from N(µ, σ 2 ), then X N(µ, σ 2 /n).

44 Sample A sample(typically) is n observations draw independently from the population, X 1, X 2,..., X n. A sample can be regarded as n numbers, or n iid random variables. Statistics: calculate some values to estimate the distribution of the population. Can be regarded as a random variable. Such as sample mean, sample standard deviation. Can be used to estimate unknown population parameters, called estimates. Sampling distribution: The distribution of an estimator or other statistic. E.g., if X 1,..., X n is a sample from N(µ, σ 2 ), then X N(µ, σ 2 /n).

45 MLE(Maximimum Likelihood Estimate) MLE is a commonly used parameter estimation method. For random sample X 1, X 2,..., X n from population X, if PDF of PMF of X is f (x; β), then ˆβ = arg max β L(β) = arg max β n f (X i ; β) i=1 is called the MLE of the unknow parameter β. Under some conditions, when the sample size n, ˆβ has a limiting(approximate) distribution N(β, σ 2 β ).

46 MLE(Maximimum Likelihood Estimate) MLE is a commonly used parameter estimation method. For random sample X 1, X 2,..., X n from population X, if PDF of PMF of X is f (x; β), then ˆβ = arg max β L(β) = arg max β n f (X i ; β) i=1 is called the MLE of the unknow parameter β. Under some conditions, when the sample size n, ˆβ has a limiting(approximate) distribution N(β, σ 2 β ).

47 Standard Error If ˆβ = ψ(x 1,..., X n ) is a estimate of an population parameter β, it has a sampling distribution F ˆβ(x). The standard deviation of the sampling distribution is σ ˆβ. An estimate of σ ˆβ is called the standard error(se) of ˆβ. If ˆβ has a limiting normal distribution, then ˆβ is approximately distributed N(β, SE( ˆβ)). SE can measure the precision of estimation. SE can be used to construct (approximate) confidence intervals.

48 Standard Error If ˆβ = ψ(x 1,..., X n ) is a estimate of an population parameter β, it has a sampling distribution F ˆβ(x). The standard deviation of the sampling distribution is σ ˆβ. An estimate of σ ˆβ is called the standard error(se) of ˆβ. If ˆβ has a limiting normal distribution, then ˆβ is approximately distributed N(β, SE( ˆβ)). SE can measure the precision of estimation. SE can be used to construct (approximate) confidence intervals.

49 Standard Error If ˆβ = ψ(x 1,..., X n ) is a estimate of an population parameter β, it has a sampling distribution F ˆβ(x). The standard deviation of the sampling distribution is σ ˆβ. An estimate of σ ˆβ is called the standard error(se) of ˆβ. If ˆβ has a limiting normal distribution, then ˆβ is approximately distributed N(β, SE( ˆβ)). SE can measure the precision of estimation. SE can be used to construct (approximate) confidence intervals.

50 Standard Error If ˆβ = ψ(x 1,..., X n ) is a estimate of an population parameter β, it has a sampling distribution F ˆβ(x). The standard deviation of the sampling distribution is σ ˆβ. An estimate of σ ˆβ is called the standard error(se) of ˆβ. If ˆβ has a limiting normal distribution, then ˆβ is approximately distributed N(β, SE( ˆβ)). SE can measure the precision of estimation. SE can be used to construct (approximate) confidence intervals.

51 Standard Error If ˆβ = ψ(x 1,..., X n ) is a estimate of an population parameter β, it has a sampling distribution F ˆβ(x). The standard deviation of the sampling distribution is σ ˆβ. An estimate of σ ˆβ is called the standard error(se) of ˆβ. If ˆβ has a limiting normal distribution, then ˆβ is approximately distributed N(β, SE( ˆβ)). SE can measure the precision of estimation. SE can be used to construct (approximate) confidence intervals.

52 Standard Error If ˆβ = ψ(x 1,..., X n ) is a estimate of an population parameter β, it has a sampling distribution F ˆβ(x). The standard deviation of the sampling distribution is σ ˆβ. An estimate of σ ˆβ is called the standard error(se) of ˆβ. If ˆβ has a limiting normal distribution, then ˆβ is approximately distributed N(β, SE( ˆβ)). SE can measure the precision of estimation. SE can be used to construct (approximate) confidence intervals.

53 Hypothesis Tests To test the null hypothesis H 0 agaist the alternative hypothesis H a on the population; Given sample X 1, X 2,..., X n from population F(x; θ), construct some statistic ξ, whose distribution does not depend on θ, but its value can indicate the possible choice of H 0 or H a. Traditionally, we first choose an significance level α, then we find a rejection area W, where sup H0 P(ξ W ) α. We reject H 0 when ξ W.

54 Hypothesis Tests To test the null hypothesis H 0 agaist the alternative hypothesis H a on the population; Given sample X 1, X 2,..., X n from population F(x; θ), construct some statistic ξ, whose distribution does not depend on θ, but its value can indicate the possible choice of H 0 or H a. Traditionally, we first choose an significance level α, then we find a rejection area W, where sup H0 P(ξ W ) α. We reject H 0 when ξ W.

55 Hypothesis Tests To test the null hypothesis H 0 agaist the alternative hypothesis H a on the population; Given sample X 1, X 2,..., X n from population F(x; θ), construct some statistic ξ, whose distribution does not depend on θ, but its value can indicate the possible choice of H 0 or H a. Traditionally, we first choose an significance level α, then we find a rejection area W, where sup H0 P(ξ W ) α. We reject H 0 when ξ W.

56 Errors and P-Values Two types of possible errors in a hypothesis test: Type I error, H 0 true but rejected. Error rate is at most the significance level α. Type II error, H0 false but accepted. Error rate can be as large as 1 α. To reduce type II error: Construct theoretically good tests. Don t choose α too small. Choose a big enough sample size n. p-value: the minimum α we can use if we want to reject H 0, after the test statistic value is known. The smaller the p-value is, the more confidence we have when we reject H 0. Reject H 0 if and only if the p-value is less than α.

57 Errors and P-Values Two types of possible errors in a hypothesis test: Type I error, H 0 true but rejected. Error rate is at most the significance level α. Type II error, H0 false but accepted. Error rate can be as large as 1 α. To reduce type II error: Construct theoretically good tests. Don t choose α too small. Choose a big enough sample size n. p-value: the minimum α we can use if we want to reject H 0, after the test statistic value is known. The smaller the p-value is, the more confidence we have when we reject H 0. Reject H 0 if and only if the p-value is less than α.

58 Errors and P-Values Two types of possible errors in a hypothesis test: Type I error, H 0 true but rejected. Error rate is at most the significance level α. Type II error, H0 false but accepted. Error rate can be as large as 1 α. To reduce type II error: Construct theoretically good tests. Don t choose α too small. Choose a big enough sample size n. p-value: the minimum α we can use if we want to reject H 0, after the test statistic value is known. The smaller the p-value is, the more confidence we have when we reject H 0. Reject H 0 if and only if the p-value is less than α.

59 Errors and P-Values Two types of possible errors in a hypothesis test: Type I error, H 0 true but rejected. Error rate is at most the significance level α. Type II error, H0 false but accepted. Error rate can be as large as 1 α. To reduce type II error: Construct theoretically good tests. Don t choose α too small. Choose a big enough sample size n. p-value: the minimum α we can use if we want to reject H 0, after the test statistic value is known. The smaller the p-value is, the more confidence we have when we reject H 0. Reject H 0 if and only if the p-value is less than α.

60 Errors and P-Values Two types of possible errors in a hypothesis test: Type I error, H 0 true but rejected. Error rate is at most the significance level α. Type II error, H0 false but accepted. Error rate can be as large as 1 α. To reduce type II error: Construct theoretically good tests. Don t choose α too small. Choose a big enough sample size n. p-value: the minimum α we can use if we want to reject H 0, after the test statistic value is known. The smaller the p-value is, the more confidence we have when we reject H 0. Reject H 0 if and only if the p-value is less than α.

61 Errors and P-Values Two types of possible errors in a hypothesis test: Type I error, H 0 true but rejected. Error rate is at most the significance level α. Type II error, H0 false but accepted. Error rate can be as large as 1 α. To reduce type II error: Construct theoretically good tests. Don t choose α too small. Choose a big enough sample size n. p-value: the minimum α we can use if we want to reject H 0, after the test statistic value is known. The smaller the p-value is, the more confidence we have when we reject H 0. Reject H 0 if and only if the p-value is less than α.

62 Errors and P-Values Two types of possible errors in a hypothesis test: Type I error, H 0 true but rejected. Error rate is at most the significance level α. Type II error, H0 false but accepted. Error rate can be as large as 1 α. To reduce type II error: Construct theoretically good tests. Don t choose α too small. Choose a big enough sample size n. p-value: the minimum α we can use if we want to reject H 0, after the test statistic value is known. The smaller the p-value is, the more confidence we have when we reject H 0. Reject H 0 if and only if the p-value is less than α.

63 Errors and P-Values Two types of possible errors in a hypothesis test: Type I error, H 0 true but rejected. Error rate is at most the significance level α. Type II error, H0 false but accepted. Error rate can be as large as 1 α. To reduce type II error: Construct theoretically good tests. Don t choose α too small. Choose a big enough sample size n. p-value: the minimum α we can use if we want to reject H 0, after the test statistic value is known. The smaller the p-value is, the more confidence we have when we reject H 0. Reject H 0 if and only if the p-value is less than α.

64 Errors and P-Values Two types of possible errors in a hypothesis test: Type I error, H 0 true but rejected. Error rate is at most the significance level α. Type II error, H0 false but accepted. Error rate can be as large as 1 α. To reduce type II error: Construct theoretically good tests. Don t choose α too small. Choose a big enough sample size n. p-value: the minimum α we can use if we want to reject H 0, after the test statistic value is known. The smaller the p-value is, the more confidence we have when we reject H 0. Reject H 0 if and only if the p-value is less than α.

65 Example hypothesis test X N(µ, σ 2 ). Sample is X 1,..., X n. H 0 : µ µ 0 H a : µ > µ 0. Test statistic T = X µ 0 SE( X) where SE( X) = S/ n, S is the sample standard deviation. Rejection area: W = {ξ > λ}, λ = F 1 (1 α, n 1), F 1 (p, n) is the quantile function of the t(n 1) distribution. P-value is 1 F(T ; n 1), where F(x; n) is the CDF of the t(n) distribution. The larger T is, the smaller the p-value is, the more confidence we have when we reject H 0.

66 Example hypothesis test X N(µ, σ 2 ). Sample is X 1,..., X n. H 0 : µ µ 0 H a : µ > µ 0. Test statistic T = X µ 0 SE( X) where SE( X) = S/ n, S is the sample standard deviation. Rejection area: W = {ξ > λ}, λ = F 1 (1 α, n 1), F 1 (p, n) is the quantile function of the t(n 1) distribution. P-value is 1 F(T ; n 1), where F(x; n) is the CDF of the t(n) distribution. The larger T is, the smaller the p-value is, the more confidence we have when we reject H 0.

67 Example hypothesis test X N(µ, σ 2 ). Sample is X 1,..., X n. H 0 : µ µ 0 H a : µ > µ 0. Test statistic T = X µ 0 SE( X) where SE( X) = S/ n, S is the sample standard deviation. Rejection area: W = {ξ > λ}, λ = F 1 (1 α, n 1), F 1 (p, n) is the quantile function of the t(n 1) distribution. P-value is 1 F(T ; n 1), where F(x; n) is the CDF of the t(n) distribution. The larger T is, the smaller the p-value is, the more confidence we have when we reject H 0.

68 Example hypothesis test X N(µ, σ 2 ). Sample is X 1,..., X n. H 0 : µ µ 0 H a : µ > µ 0. Test statistic T = X µ 0 SE( X) where SE( X) = S/ n, S is the sample standard deviation. Rejection area: W = {ξ > λ}, λ = F 1 (1 α, n 1), F 1 (p, n) is the quantile function of the t(n 1) distribution. P-value is 1 F(T ; n 1), where F(x; n) is the CDF of the t(n) distribution. The larger T is, the smaller the p-value is, the more confidence we have when we reject H 0.

69 Example hypothesis test X N(µ, σ 2 ). Sample is X 1,..., X n. H 0 : µ µ 0 H a : µ > µ 0. Test statistic T = X µ 0 SE( X) where SE( X) = S/ n, S is the sample standard deviation. Rejection area: W = {ξ > λ}, λ = F 1 (1 α, n 1), F 1 (p, n) is the quantile function of the t(n 1) distribution. P-value is 1 F(T ; n 1), where F(x; n) is the CDF of the t(n) distribution. The larger T is, the smaller the p-value is, the more confidence we have when we reject H 0.

70 Example hypothesis test X N(µ, σ 2 ). Sample is X 1,..., X n. H 0 : µ µ 0 H a : µ > µ 0. Test statistic T = X µ 0 SE( X) where SE( X) = S/ n, S is the sample standard deviation. Rejection area: W = {ξ > λ}, λ = F 1 (1 α, n 1), F 1 (p, n) is the quantile function of the t(n 1) distribution. P-value is 1 F(T ; n 1), where F(x; n) is the CDF of the t(n) distribution. The larger T is, the smaller the p-value is, the more confidence we have when we reject H 0.

71 Section Contents Measurment level of variables. statistics for nominal variables. statistics for interval variables. Histograms, boxplots, QQ plots, probability plots, stem-leaf plots. Using PROC FREQ, PROC MEANS, PROC UNIVARIATE. statistics Summary statistics

72 Section Contents Measurment level of variables. statistics for nominal variables. statistics for interval variables. Histograms, boxplots, QQ plots, probability plots, stem-leaf plots. Using PROC FREQ, PROC MEANS, PROC UNIVARIATE. statistics Summary statistics

73 Section Contents Measurment level of variables. statistics for nominal variables. statistics for interval variables. Histograms, boxplots, QQ plots, probability plots, stem-leaf plots. Using PROC FREQ, PROC MEANS, PROC UNIVARIATE. statistics Summary statistics

74 Section Contents Measurment level of variables. statistics for nominal variables. statistics for interval variables. Histograms, boxplots, QQ plots, probability plots, stem-leaf plots. Using PROC FREQ, PROC MEANS, PROC UNIVARIATE. statistics Summary statistics

75 Section Contents Measurment level of variables. statistics for nominal variables. statistics for interval variables. Histograms, boxplots, QQ plots, probability plots, stem-leaf plots. Using PROC FREQ, PROC MEANS, PROC UNIVARIATE. statistics Summary statistics

76 Measurement levels Nominal, such as sex, job type, race. Ordinal, such as age group(child, youth, medium, old), dose level(none, low, high). Interval, such as weight, blood pressure, heart rate. statistics Summary statistics

77 Measurement levels Nominal, such as sex, job type, race. Ordinal, such as age group(child, youth, medium, old), dose level(none, low, high). Interval, such as weight, blood pressure, heart rate. statistics Summary statistics

78 Measurement levels Nominal, such as sex, job type, race. Ordinal, such as age group(child, youth, medium, old), dose level(none, low, high). Interval, such as weight, blood pressure, heart rate. statistics Summary statistics

79 Statistics for nominal variables The value set. The count of each value(called frequency), and the percentage. Use a bar chart to show the distribution. statistics Summary statistics

80 Statistics for nominal variables The value set. The count of each value(called frequency), and the percentage. Use a bar chart to show the distribution. statistics Summary statistics

81 Statistics for nominal variables The value set. The count of each value(called frequency), and the percentage. Use a bar chart to show the distribution. statistics Summary statistics

82 Example: frequency table Use PROC FREQ to list the value set and the frequency counts, percents. proc freq data=sasuser.class; tables sex; run; Gender sex Frequency Percent Cumulative Cumulative Frequency Percent F M statistics Summary statistics

83 Example: bar chart Use PROC GCHART to make a bar chart. hbar could be replaced with vbar, hbar3d, vbar3d. proc gchart data=sasuser.class; hbar sex; run; statistics Summary statistics

84 statistics for interval variables Location statistics: sample mean, sample median. Variability statistics: sample standard deviation, sample interquantile range, range, coefficient of variation. Sample skewness Sample kurtosis Moments, quantiles. n (n 1)(n 2) n(n+1) (n 1)(n 2)(n 3) ( ) y i y 3. s (yi y) 4 3(n 1)2 s 4 (n 2)(n 3) statistics Summary statistics

85 statistics for interval variables Location statistics: sample mean, sample median. Variability statistics: sample standard deviation, sample interquantile range, range, coefficient of variation. Sample skewness Sample kurtosis Moments, quantiles. n (n 1)(n 2) n(n+1) (n 1)(n 2)(n 3) ( ) y i y 3. s (yi y) 4 3(n 1)2 s 4 (n 2)(n 3) statistics Summary statistics

86 statistics for interval variables Location statistics: sample mean, sample median. Variability statistics: sample standard deviation, sample interquantile range, range, coefficient of variation. Sample skewness Sample kurtosis Moments, quantiles. n (n 1)(n 2) n(n+1) (n 1)(n 2)(n 3) ( ) y i y 3. s (yi y) 4 3(n 1)2 s 4 (n 2)(n 3) statistics Summary statistics

87 statistics for interval variables Location statistics: sample mean, sample median. Variability statistics: sample standard deviation, sample interquantile range, range, coefficient of variation. Sample skewness Sample kurtosis Moments, quantiles. n (n 1)(n 2) n(n+1) (n 1)(n 2)(n 3) ( ) y i y 3. s (yi y) 4 3(n 1)2 s 4 (n 2)(n 3) statistics Summary statistics

88 statistics for interval variables Location statistics: sample mean, sample median. Variability statistics: sample standard deviation, sample interquantile range, range, coefficient of variation. Sample skewness Sample kurtosis Moments, quantiles. n (n 1)(n 2) n(n+1) (n 1)(n 2)(n 3) ( ) y i y 3. s (yi y) 4 3(n 1)2 s 4 (n 2)(n 3) statistics Summary statistics

89 PROC UNIVARIATE Use PROC UNIVARIATE to compute sample statistics for an interval variable. Example: proc univariate data=sasuser.class; var height; run; statistics Summary statistics

90 PROC UNIVARIATE Use PROC UNIVARIATE to compute sample statistics for an interval variable. Example: proc univariate data=sasuser.class; var height; run; statistics Summary statistics

91 Output Moments N 19 Sum Weights Sum Observations Std Deviation Skewness Kurtosis Uncorrected SS Corrected SS Coeff Variation Std Error Basic Statistical Measures Location Variability Std Deviation Median Mode Range Interquartile Range statistics Summary statistics Note The mode displayed is the smallest of 2 modes with a count of 2.

92 Tests for Location: Mu0=0 Test Statistic p Value Student s t t Pr > t <.0001 Sign M 9.5 Pr >= M <.0001 Signed Rank S 95 Pr >= S <.0001 statistics Summary statistics

93 Quantiles (Definition 5) Quantile Estimate 100% Max % % % % Q % Median % Q % % % % Min 51.3 statistics Summary statistics

94 Extreme Observations Lowest Highest Value Obs Value Obs statistics Summary statistics

95 Histograms The histogram is a nonparametric estimate of the population density function. It divide the value set into intervals, and count the percent of observations in each interval. Example: proc univariate data=sasuser.class noprint; var height; histogram; run; statistics Summary statistics

96 Histograms The histogram is a nonparametric estimate of the population density function. It divide the value set into intervals, and count the percent of observations in each interval. Example: proc univariate data=sasuser.class noprint; var height; histogram; run; statistics Summary statistics

97 Histograms The histogram is a nonparametric estimate of the population density function. It divide the value set into intervals, and count the percent of observations in each interval. Example: proc univariate data=sasuser.class noprint; var height; histogram; run; statistics Summary statistics

98 Examples of histograms Skewness shown in histograms. statistics Summary statistics

99 Box plot Quantiles could be used to describe key distribution charasteristics. The box plot shows the median, 1/4 and 3/4 quantile, and the minimum and maximum of a variable. The box holds the middle 50% range. Length is the IQR(inter quantile range). The wiskers extend to the minimum and the maximum; but the length of the wisker is not allowed to exceed 1.5 times the IQR. Points beyond wiskers are plotted as individual points. They are candidate for outliers. statistics Summary statistics

100 Box plot Quantiles could be used to describe key distribution charasteristics. The box plot shows the median, 1/4 and 3/4 quantile, and the minimum and maximum of a variable. The box holds the middle 50% range. Length is the IQR(inter quantile range). The wiskers extend to the minimum and the maximum; but the length of the wisker is not allowed to exceed 1.5 times the IQR. Points beyond wiskers are plotted as individual points. They are candidate for outliers. statistics Summary statistics

101 Box plot Quantiles could be used to describe key distribution charasteristics. The box plot shows the median, 1/4 and 3/4 quantile, and the minimum and maximum of a variable. The box holds the middle 50% range. Length is the IQR(inter quantile range). The wiskers extend to the minimum and the maximum; but the length of the wisker is not allowed to exceed 1.5 times the IQR. Points beyond wiskers are plotted as individual points. They are candidate for outliers. statistics Summary statistics

102 Box plot Quantiles could be used to describe key distribution charasteristics. The box plot shows the median, 1/4 and 3/4 quantile, and the minimum and maximum of a variable. The box holds the middle 50% range. Length is the IQR(inter quantile range). The wiskers extend to the minimum and the maximum; but the length of the wisker is not allowed to exceed 1.5 times the IQR. Points beyond wiskers are plotted as individual points. They are candidate for outliers. statistics Summary statistics

103 Box plot Quantiles could be used to describe key distribution charasteristics. The box plot shows the median, 1/4 and 3/4 quantile, and the minimum and maximum of a variable. The box holds the middle 50% range. Length is the IQR(inter quantile range). The wiskers extend to the minimum and the maximum; but the length of the wisker is not allowed to exceed 1.5 times the IQR. Points beyond wiskers are plotted as individual points. They are candidate for outliers. statistics Summary statistics

104 Example: boxplot of height statistics Summary statistics

105 proc sql; create view work._tmp_0 as select height, 1 as _dummy_ from sasuser.class; title ; axis1 major=none value=none label=none; proc boxplot data=_tmp_0; plot height*_dummy_ / boxstyle=skematic cboxfill=blue haxis=axis1; run; statistics Summary statistics

106 Example: boxplot of GPA and CO Skewness and possible outliers shown in boxplot. statistics Summary statistics

107 Grouped boxplot statistics Summary statistics

108 proc sort data=sasuser.gpa out=_tmp_1; by sex; proc boxplot data=_tmp_1; plot gpa*sex / boxstyle=skematic cboxfill=blue; run; statistics Summary statistics

109 QQ Plot The normal distribution is the most commonly used distribution. Graphs are designed to show discrepency from the normal distribution. QQ plot is one of them. Suppose Y 1, Y 2,..., Y n is from N(µ,σ 2 ) and sorted ascendingly. CDF F(x) = Φ( x µ σ ), F(Y i) i n, Φ( Y i µ σ ) i n, Y i µ + σφ 1 ( i n ). Let x i = Φ 1 ( i n ), plot (x i, Y i ),i = 1,..., n. The points should lie around the line y = µ + σx. Continuity adjustment: x i = Φ 1 ( i n+0.25 ). statistics Summary statistics

110 QQ Plot The normal distribution is the most commonly used distribution. Graphs are designed to show discrepency from the normal distribution. QQ plot is one of them. Suppose Y 1, Y 2,..., Y n is from N(µ,σ 2 ) and sorted ascendingly. CDF F(x) = Φ( x µ σ ), F(Y i) i n, Φ( Y i µ σ ) i n, Y i µ + σφ 1 ( i n ). Let x i = Φ 1 ( i n ), plot (x i, Y i ),i = 1,..., n. The points should lie around the line y = µ + σx. Continuity adjustment: x i = Φ 1 ( i n+0.25 ). statistics Summary statistics

111 QQ Plot The normal distribution is the most commonly used distribution. Graphs are designed to show discrepency from the normal distribution. QQ plot is one of them. Suppose Y 1, Y 2,..., Y n is from N(µ,σ 2 ) and sorted ascendingly. CDF F(x) = Φ( x µ σ ), F(Y i) i n, Φ( Y i µ σ ) i n, Y i µ + σφ 1 ( i n ). Let x i = Φ 1 ( i n ), plot (x i, Y i ),i = 1,..., n. The points should lie around the line y = µ + σx. Continuity adjustment: x i = Φ 1 ( i n+0.25 ). statistics Summary statistics

112 QQ Plot The normal distribution is the most commonly used distribution. Graphs are designed to show discrepency from the normal distribution. QQ plot is one of them. Suppose Y 1, Y 2,..., Y n is from N(µ,σ 2 ) and sorted ascendingly. CDF F(x) = Φ( x µ σ ), F(Y i) i n, Φ( Y i µ σ ) i n, Y i µ + σφ 1 ( i n ). Let x i = Φ 1 ( i n ), plot (x i, Y i ),i = 1,..., n. The points should lie around the line y = µ + σx. Continuity adjustment: x i = Φ 1 ( i n+0.25 ). statistics Summary statistics

113 QQ Plot The normal distribution is the most commonly used distribution. Graphs are designed to show discrepency from the normal distribution. QQ plot is one of them. Suppose Y 1, Y 2,..., Y n is from N(µ,σ 2 ) and sorted ascendingly. CDF F(x) = Φ( x µ σ ), F(Y i) i n, Φ( Y i µ σ ) i n, Y i µ + σφ 1 ( i n ). Let x i = Φ 1 ( i n ), plot (x i, Y i ),i = 1,..., n. The points should lie around the line y = µ + σx. Continuity adjustment: x i = Φ 1 ( i n+0.25 ). statistics Summary statistics

114 QQ Plot The normal distribution is the most commonly used distribution. Graphs are designed to show discrepency from the normal distribution. QQ plot is one of them. Suppose Y 1, Y 2,..., Y n is from N(µ,σ 2 ) and sorted ascendingly. CDF F(x) = Φ( x µ σ ), F(Y i) i n, Φ( Y i µ σ ) i n, Y i µ + σφ 1 ( i n ). Let x i = Φ 1 ( i n ), plot (x i, Y i ),i = 1,..., n. The points should lie around the line y = µ + σx. Continuity adjustment: x i = Φ 1 ( i n+0.25 ). statistics Summary statistics

115 Example: QQ Plot of HEIGHT proc univariate data=sasuser.class noprint; qqplot height / normal(mu=est sigma=est); run; statistics Summary statistics

116 Example: Skewness in QQ Plot statistics Summary statistics

117 Probability Plot Probability plot is the same plot as a QQ plot, exept that the label of the x axis Φ(x i ) instead of x i. Probability plots are preferable for graphical estimation of percentiles, whereas Q-Q plots are preferable for graphical estimation of distribution parameters. statistics Summary statistics

118 Probability Plot Probability plot is the same plot as a QQ plot, exept that the label of the x axis Φ(x i ) instead of x i. Probability plots are preferable for graphical estimation of percentiles, whereas Q-Q plots are preferable for graphical estimation of distribution parameters. statistics Summary statistics

119 Example: Probability Plot of HEIGHT proc univariate data=sasuser.class noprint; probplot height / normal(mu=est sigma=est); run; statistics Summary statistics

120 The Stem-leaf Plot The stem-leaf plot is a text-based plot, which display information like the histogram, but with detail on each data value. Each leaf corresponds to one data value. PROC UNIVARIATE has an option PLOT, which generates text-based stem-leaf plot, boxplot and QQ plot. statistics Summary statistics

121 The Stem-leaf Plot The stem-leaf plot is a text-based plot, which display information like the histogram, but with detail on each data value. Each leaf corresponds to one data value. PROC UNIVARIATE has an option PLOT, which generates text-based stem-leaf plot, boxplot and QQ plot. statistics Summary statistics

122 proc univariate data=sasuser.class plot; var height; run; Stem Leaf # Multiply Stem.Leaf by 10**+1 statistics Summary statistics

123 PROC MEANS and PROC SUMMARY PROC MEANS and PROC SUMMARY are used to produce summary statistics. They can display overall summary statistics and classified summary statistics. PROC MEANS displays result by default; PROC SUMMARY need the PRINT option to display result. They can output data sets with the summary statistics. PROC SUMMARY is designed to do this, although PROC MEANS can do the same. statistics Summary statistics

124 PROC MEANS and PROC SUMMARY PROC MEANS and PROC SUMMARY are used to produce summary statistics. They can display overall summary statistics and classified summary statistics. PROC MEANS displays result by default; PROC SUMMARY need the PRINT option to display result. They can output data sets with the summary statistics. PROC SUMMARY is designed to do this, although PROC MEANS can do the same. statistics Summary statistics

125 PROC MEANS and PROC SUMMARY PROC MEANS and PROC SUMMARY are used to produce summary statistics. They can display overall summary statistics and classified summary statistics. PROC MEANS displays result by default; PROC SUMMARY need the PRINT option to display result. They can output data sets with the summary statistics. PROC SUMMARY is designed to do this, although PROC MEANS can do the same. statistics Summary statistics

126 Example of overall statistics Use the VAR statement to specify which varibles to summarize. Example of overall statistics: proc means data=sasuser.class; var height weight; run; proc summary data=sasuser.class print; var height weight; run; statistics Summary statistics

127 Example of overall statistics Use the VAR statement to specify which varibles to summarize. Example of overall statistics: proc means data=sasuser.class; var height weight; run; proc summary data=sasuser.class print; var height weight; run; statistics Summary statistics

128 Classified summary Use the CLASS statement to specify one or more class variables. Example: proc means data=sasuser.class ; var height weight; class sex; run; statistics Summary statistics

129 Classified summary Use the CLASS statement to specify one or more class variables. Example: proc means data=sasuser.class ; var height weight; class sex; run; statistics Summary statistics

130 Other statistical measures PROC MEANS calculate summaries N,, Standard Deviation, Minimum, Maximum by default. Use PROC MEANS options to specify other statistical measures. Example: proc means data=sasuser.class MEAN VAR CV; var height weight; class sex; run; statistics Summary statistics

131 Other statistical measures PROC MEANS calculate summaries N,, Standard Deviation, Minimum, Maximum by default. Use PROC MEANS options to specify other statistical measures. Example: proc means data=sasuser.class MEAN VAR CV; var height weight; class sex; run; statistics Summary statistics

132 Other statistical measures PROC MEANS calculate summaries N,, Standard Deviation, Minimum, Maximum by default. Use PROC MEANS options to specify other statistical measures. Example: proc means data=sasuser.class MEAN VAR CV; var height weight; class sex; run; statistics Summary statistics

133 Producing output datasets Use the OUTPUT statement to save the summary statistics to an output dataset. Syntax: OUTPUT OUT=output-dataset keyword= keyword= / AUTONAME; Where keyword is a name of some statistical measure, like MEAN, STD, N, MIN, MAX, etc. Example: statistics Summary statistics proc means data=sasuser.class MEAN VAR CV; var height weight; class sex; output out=res N= CV= / AUTONAME; run; proc print data=res;run;

134 Producing output datasets Use the OUTPUT statement to save the summary statistics to an output dataset. Syntax: OUTPUT OUT=output-dataset keyword= keyword= / AUTONAME; Where keyword is a name of some statistical measure, like MEAN, STD, N, MIN, MAX, etc. Example: statistics Summary statistics proc means data=sasuser.class MEAN VAR CV; var height weight; class sex; output out=res N= CV= / AUTONAME; run; proc print data=res;run;

135 Producing output datasets Use the OUTPUT statement to save the summary statistics to an output dataset. Syntax: OUTPUT OUT=output-dataset keyword= keyword= / AUTONAME; Where keyword is a name of some statistical measure, like MEAN, STD, N, MIN, MAX, etc. Example: statistics Summary statistics proc means data=sasuser.class MEAN VAR CV; var height weight; class sex; output out=res N= CV= / AUTONAME; run; proc print data=res;run;

136 Producing output datasets Use the OUTPUT statement to save the summary statistics to an output dataset. Syntax: OUTPUT OUT=output-dataset keyword= keyword= / AUTONAME; Where keyword is a name of some statistical measure, like MEAN, STD, N, MIN, MAX, etc. Example: statistics Summary statistics proc means data=sasuser.class MEAN VAR CV; var height weight; class sex; output out=res N= CV= / AUTONAME; run; proc print data=res;run;

137 One sample Z tests, t tests. Two sample t tests. The Wilcoxon rank sum test. Paired t tests. Comparing proportions: one sample and two sample. One Sample Test of the Comparing Two Groups Paired Comparison Comparing Proportions

138 One sample Z tests, t tests. Two sample t tests. The Wilcoxon rank sum test. Paired t tests. Comparing proportions: one sample and two sample. One Sample Test of the Comparing Two Groups Paired Comparison Comparing Proportions

139 One sample Z tests, t tests. Two sample t tests. The Wilcoxon rank sum test. Paired t tests. Comparing proportions: one sample and two sample. One Sample Test of the Comparing Two Groups Paired Comparison Comparing Proportions

140 One sample Z tests, t tests. Two sample t tests. The Wilcoxon rank sum test. Paired t tests. Comparing proportions: one sample and two sample. One Sample Test of the Comparing Two Groups Paired Comparison Comparing Proportions

141 One sample Z tests, t tests. Two sample t tests. The Wilcoxon rank sum test. Paired t tests. Comparing proportions: one sample and two sample. One Sample Test of the Comparing Two Groups Paired Comparison Comparing Proportions

142 One Sample Z Test Two-sided X N(µ, σ 2 0 ), σ 0 known. H 0 : µ = µ 0 H a : µ µ 0 Z = X µ 0 σ 0 / H 0 n N(0, 1). W = { Z > λ}, λ = Φ 1 (1 α/2). p-value=2(1 Φ( Z ). Wald test: Based on the central limit theorem, to test H 0 : θ = θ 0 H a : θ θ 0, use W = ˆθ θ 0 as the SE(ˆθ) test statistic, if W N(0, 1). One Sample Test of the Comparing Two Groups Paired Comparison Comparing Proportions

143 One Sample Z Test Two-sided X N(µ, σ 2 0 ), σ 0 known. H 0 : µ = µ 0 H a : µ µ 0 Z = X µ 0 σ 0 / H 0 n N(0, 1). W = { Z > λ}, λ = Φ 1 (1 α/2). p-value=2(1 Φ( Z ). Wald test: Based on the central limit theorem, to test H 0 : θ = θ 0 H a : θ θ 0, use W = ˆθ θ 0 as the SE(ˆθ) test statistic, if W N(0, 1). One Sample Test of the Comparing Two Groups Paired Comparison Comparing Proportions

144 One Sample Z Test Two-sided X N(µ, σ 2 0 ), σ 0 known. H 0 : µ = µ 0 H a : µ µ 0 Z = X µ 0 σ 0 / H 0 n N(0, 1). W = { Z > λ}, λ = Φ 1 (1 α/2). p-value=2(1 Φ( Z ). Wald test: Based on the central limit theorem, to test H 0 : θ = θ 0 H a : θ θ 0, use W = ˆθ θ 0 as the SE(ˆθ) test statistic, if W N(0, 1). One Sample Test of the Comparing Two Groups Paired Comparison Comparing Proportions

145 One Sample Z Test Two-sided X N(µ, σ 2 0 ), σ 0 known. H 0 : µ = µ 0 H a : µ µ 0 Z = X µ 0 σ 0 / H 0 n N(0, 1). W = { Z > λ}, λ = Φ 1 (1 α/2). p-value=2(1 Φ( Z ). Wald test: Based on the central limit theorem, to test H 0 : θ = θ 0 H a : θ θ 0, use W = ˆθ θ 0 as the SE(ˆθ) test statistic, if W N(0, 1). One Sample Test of the Comparing Two Groups Paired Comparison Comparing Proportions

146 One Sample Z Test Two-sided X N(µ, σ 2 0 ), σ 0 known. H 0 : µ = µ 0 H a : µ µ 0 Z = X µ 0 σ 0 / H 0 n N(0, 1). W = { Z > λ}, λ = Φ 1 (1 α/2). p-value=2(1 Φ( Z ). Wald test: Based on the central limit theorem, to test H 0 : θ = θ 0 H a : θ θ 0, use W = ˆθ θ 0 as the SE(ˆθ) test statistic, if W N(0, 1). One Sample Test of the Comparing Two Groups Paired Comparison Comparing Proportions

147 One Sample Z Test Two-sided X N(µ, σ 2 0 ), σ 0 known. H 0 : µ = µ 0 H a : µ µ 0 Z = X µ 0 σ 0 / H 0 n N(0, 1). W = { Z > λ}, λ = Φ 1 (1 α/2). p-value=2(1 Φ( Z ). Wald test: Based on the central limit theorem, to test H 0 : θ = θ 0 H a : θ θ 0, use W = ˆθ θ 0 as the SE(ˆθ) test statistic, if W N(0, 1). One Sample Test of the Comparing Two Groups Paired Comparison Comparing Proportions