Preliminary Statistics. Lecture 3: Probability Models and Distributions

Transcription

1 Preliminary Statistics Lecture 3: Probability Models and Distributions Rory Macqueen September 2015

2 Outline Revision of Lecture 2 Probability Density Functions Cumulative Distribution Functions Properties of random variables Properties Variance Covariance Common Continuous Models Normal Distribution T-Distribution Chi-Square Distribution F-Distribution Sampling Distributions

3 Probability Density Function f(x) A formula defining the curve of a continuous probability model. The area under the curve between two points gives the probability that a value between the two points will arise. This area is obtained by integrating the PDF between the two points. The sum of the area under the curve is equal to one.

4 Probability Density Function f(x) PDF of a continuous random variable measures the probability of the random variable over a certain range or interval. P a X b = f x dx a For example, the probability that the height of an individual lies in the interval 60 and 75 inches is given by the area between 60 and 75. b

5 Cumulative Distribution Function F(x) x i P X x i = f x dx Geometrically, the CDF of a continuous random variable is a continuous curve

6 Expected Values The expected value of a random variable (often denoted as μ) is the sum/integral of each value it can take, x i, multiplied by the probability of taking that value, f(x i ) (the pdf). It is the population mean of the variable For discrete random variables: For continuous random variables: N E X = x i f x i = μ i=1 E X = xf x dx = μ The expected value of random variables is also known as the population mean. Eg. In the example with the dice: 1 7 = If all values are equally likely, f(x i )=1/N, and the expected value is the arithmetic mean.

7 Expected Values Properties Properties of Expected Value 1. E(b) = b Where b is a constant. 2. E(X+Y) = E(X) + E(Y) Where X and Y are random variables. 3. E(XY) E(X) E(Y) Where X and Y are non independent random variables. 4. E(XY) = E(X) E(Y) If X and Y are independent random variables. 5. E(aX) = ae(x) Where a is a constant. 6. E(aX+b) = ae(x) + E(b) = ae(x) + b Where a and b are constant. 7. E(g x ) = g x f(x) x Where g(x) is a function of X

8 Variance Let X be a random variable with E(X) = μ. The dispersion of X around its mean (expected value) can be described by the (population) variance denoted by ς 2 : Var X = ς 2 = = 2 (( )ܧ )ܧ X )ܧ 2 ) μ 2

9 Variance Let X be a discrete random variable with E(X) = μ: Var X = )ܧ μ) 2 = (x i μ) 2 f(x i ) N i=1 If each outcome is equally likely, f(x i ) = 1 N Let X be a continuous random variable with E(X) = μ: Var X = )ܧ μ) 2 = (x μ) 2 f x dx

10 Variance 1. Var(x) = E(X-μ) 2 = E(X 2 ) - μ 2 Properties of Variance 2. Var(b) = 0 With b being a constant. 3. Var(X+b) = Var(X) With b being a constant. 4. Var(aX) = a 2 Var(X) With a being a constant 5. Var(aX+b) = a 2 Var(X) With a and b being constants 6. Var(X+Y) = Var (X) + Var(Y) Var (X-Y) = Var (X) + Var(Y) Var (ax+by) = a 2 Var(X) + b 2 Var(Y) If X and Y are independent random variables. Otherwise see properties of covariance (later).

11 Covariance Let X and Y be two random variables with means E(X) = μ x and E(Y) = μ y. Then the covariance between the two is: Cov X, Y = E[ X μ x Y μ y ] = E XY μ x μ y

12 Covariance For a discrete random variable this translates to Cov X, Y = x y x i y i f x i, y i μ x μ y For a continuous random variable Cov X, Y = XYf x i, y i μ x μ y dxdy

13 Covariance Properties of Covariance 1. cov(x,y) = E(XY) E(X)E(Y) If X and Y are two random variables 2. cov(x,x) = var(x) 3. cov(x,a) = 0 With a being a constant. 4. cov(a+bx, c+dy) = bd cov(x,y) With a,b,c, and d being constants 5. cov(x,y) = 0 If X and Y are independent, since E(XY)=E(X)E(Y)=μ x μ y 6. var(x+y) = var(x) + var(y) + 2cov(X,Y) var(x-y) = var(x) + var(y) 2cov(X,Y) If X and Y are two random variables The opposite does not hold: if the covariance between X and Y is zero, we can not infer that they are independent.

14 Correlation coefficient As we saw in Lecture 1, Cov(x, y) ρ = Var X Var(Y) = Cov(x, y) ς(x)ς(y)

15 Conditional moments We can define the conditional expectation of X, given that Y=y i, E X Y = y i = x x f(x Y = y i ) for discrete = x f(x Y = y i ) dx for continuous And similar for conditional variance and higher moments.

16 Normal distribution t (or Student s t ) distribution Chi-squared distribution F distribution

17 Normal Distribution Normality may arise: When a random variable is the result of many independent, random influences, none of which is dominant, From a Central Limit Theorem, When a random variable is logged.

18 Normal Distribution Probability Density Function of a Normal Distribution f y i = 2πς exp( 1 y i μ 2 ) 2 ς Two parameters: mean (μ) and variance (ς 2 ) The normal distribution is the exponential of a quadratic. X ~ N(μ,σ 2 ): The random variable X is normally distributed with mean μ and variance σ 2.

19 Normal Distribution Properties: Bell shaped and symmetric. Mean = Median = Mode. Skewness and Excess Kurtosis are equal to zero.

20 Normal Distribution Linear Transformations Any linear transformation of a normally distributed random variable is also normally distributed. Y ~ N(μ,σ 2 ) W = a + by ~ N(a+bμ,b 2 σ 2 ) A very useful linear transformation is: Z i = y i μ ς and Z is distributed Standard Normal.

21 Standard Normal Distribution The standard normal distribution has mean 0 and variance (and standard deviation) 1. Z ~ N(0,1) The SND is the reference distribution for the normal tables.

22 Standard Normal Distribution Areas Under a Normal and Standard Normal Distribution Z Y = μ + σz Probabilities What it means P(-1 < Z < 1) P(μ σ < Y < μ + σ) = Prob. of being within 1 sd From mean is 68.26%. P(-1.96 < Z < 1.96) P(μ 1.96σ < Y < μ σ) = % of the normal distribution lies within 1.96 sds form the mean. P(-2 < Z < 2) P(μ 2σ < Y < μ + 2σ) = Prob. of being within 2 sd from the mean is 95.44%. P(-3 < Z < 3) P(μ 3σ < Y < μ + 3σ) = Prob. of being within 3 sd from the mean is 99.7%.

23 Standard Normal Distribution

24 Chi-squared Distribution The sum of n independent squared standard normal variables, A n A = Z i 2 i=1 ~χ 2 (n) follows a chi-squared distribution with n degrees of freedom The chi-squared distribution takes only positive values and ranges from 0 to infinity Estimates of variance are distributed χ 2

25 Chi-squared Distribution χ 2 is skewed. For relatively few degrees of freedom (d.f.) the distribution is highly skewed to the right, but as the d.f. increase, the distribution approaches the normal. The mean of the chi-squared random variable is n and its variance is 2n, where n is the d.f.

26 Student s t Distribution Consider two independent variables: one standard normal (Z~N 0,1 ) and one Chi-squared (X~χ 2 (n) ). G = Z X ~t(n) n i.e. G follows a Student s t distribution with n degrees of freedom. This is used regularly in hypothesis testing (Lecture 5) when we divide an estimate of a coefficient (~N) by its standard error ( ~χ 2 n )

27 Student s t Distribution The Student s t-distribution is a bell shaped, symmetric distribution, similar to the standard normal distribution. It has mean 0 and variance n/(n-2). For low d.f., it is flatter and with fatter tails than the normal Used to compensate for the extra uncertainty when the sample is used to calculate the parameters of a normal distribution.

28 F Distribution If we have two independent Chi-squared variables X 1 (n) and X 2 (m). The ratio of these two chi-squared variables, each divided by its degrees of freedom follows an F- distribution. X 1 n B = ~F X n,m 2 m Two indexing parameters: the degrees of freedom in the numerator and the degrees of freedom in the denominator.

29 F Distribution F resembles the χ 2 : is always non negative and is skewed to the right.

30 F Distribution Use the identity E.g. Left-Hand Tail of an F-Distribution F a b,1 α = 1/Fb a,α For F 5,10 the 0.05 right-hand critical value F 5 10,0.05 = 3.33 For F 10,5 the 0.05 right-hand critical value F 10 5,0.05 = 4.74 For F 5,10 the 0.05 left-hand critical value F 5 10,0.95 =

31 Sampling Distributions Given random sampling Statistics (eg. sample mean), being calculated from a sample, are random variables. The value of the statistic is the outcome of a random process, the random sampling process. The value of the statistic will differ depending on the sample drawn. As such it will have a spread of values, each with an associated probability, i.e. a probability distribution. Probability distributions for statistics are commonly referred to as sampling distributions.