Continuous Probability Distributions Continuous Random Variables and Probability Distributions • • • • Random Variable: Y Cumulative Distribution Function (CDF): F(y)=P(Y≤y) Probability Density Function (pdf): f(y)=dF(y)/dy Rules governing continuous distributions: f(y) ≥ 0 y f ( y )dy 1 P(a≤Y≤b) = F(b)-F(a) = P(Y=a) = 0 a b a f ( y )dy Expected Values of Continuous RVs Expected Value : E (Y ) yf ( y )dy (assuming absolute convergenc e) E g (Y ) g ( y ) f ( y )dy y 2 y f ( y )dy y f ( y )dy 2 E Y 2 ( ) (1) E Y Variance : V (Y ) E (Y E (Y )) ( y ) 2 f ( y )dy 2 2 2 2 2 2 2 2 yf ( y )dy 2 f ( y )dy 2 E aY b (ay b) f ( y )dy a yf ( y )dy b f ( y )dy a ( ) b(1) a b V aY b E (aY b) E (aY b) 2 2 (ay b) (a b) f ( y )dy (ay a ) 2 f ( y )dy a 2 ( y ) 2 f ( y )dy a 2V (Y ) a 2 2 aY b a Example – Cost/Benefit Analysis of Sprewell-Bluff Project (I) • Subjective Analysis of Annual Benefits/Costs of Project (U.S. Army Corps of Engineers assessments) • Y = Actual Benefit is Random Variable taken from a triangular distribution with 3 parameters: A=Lower Bound (Pessimistic Outcome) B=Peak (Most Likely Outcome) C=Upper Bound (Optimistic Outcome) 6 Benefit Variables 3 Cost Variables Source: B.W. Taylor, R.M. North(1976). “The Measurement of Uncertainty in Public Water Resource Development,” American Journal of Agricultural Economics, Vol. 58, #4, Pt.1, pp.636-643 Example – Cost/Benefit Analysis of Sprewell-Bluff Project (II) ($1000s, rounded) Benefit/Cost Pessimistic (A) Most Likely (B) Optimistic (C) Flood Control (+) 850 1200 1500 Hydroelec Pwr (+) 5000 6000 6000 Navigation (+) 25 28 30 Recreation (+) 4200 5400 7800 Fish/Wildlife (+) 57 127 173 Area Redvlp (+) 0 830 1192 Capital Cost (-) -193K -180K -162K Annual Cost (-) -7000 -6600 -6000 Operation/Maint(-) -2192 -2049 -1742 Example – Cost/Benefit Analysis of SprewellBluff Project (III) (Flood Control, in $100K) Triangular Distribution with: lower bound=8.5 Peak=12.0 8.5 y 12.0 k ( y 8.5) 3.5 f ( y ) k (15.0 y ) / 3.0 12.0 y 15.0 0 elsewhere ( y 8.5, y 15.0) upper bound=15.0 Triangular Distribution (Not Scaled) 1 0.9 Choose k area under density curve is 1: 0.8 0.7 Area above 12.0 is 0.5((15.0-12.0)k) = 1.50k Total Area is 3.25k k=1/3.25 Probability Density Area below 12.0 is: 0.5((12.0-8.5)k) = 1.75k 0.6 0.5 0.4 0.3 0.2 0.1 0 8 8.5 9 9.5 10 10.5 11 11.5 12 12.5 Flood Control Benefits ($100K) 13 13.5 14 14.5 15 15.5 16 Example – Cost/Benefit Analysis of Sprewell-Bluff Project (IV) (Flood Control) ( y 8.5) 11.375 8.5 y 12.0 f ( y ) (15.0 y ) / 9.75 12.0 y 15.0 0 elsewhere y 8.5 F ( y ) 0 8.5 y 12 F ( y ) (t 8.50) 11.375 dt (1/11.375) t 2 8.5t y 8.5 y 2 8.5 y 8.5 2 8.5 11.375 y 2 2 2 2 2 y 8.5 17 y 8.52 22.75 12 y 15 F ( y ) F (12) (15 t ) 9.75 dt .5385 (1/ 9.75) 15t t 2 y 12 .5385 15 y y 2 2 15(12) 12 2 2 9.75 .5385 216 30 y y 2 19.5 12 y 15 y 15 F ( y ) 1 2 y 12 Example – Cost/Benefit Analysis of Sprewell-Bluff Project (V) (Flood Control) ( y 8.5) 11.375 8.5 y 12.0 f ( y ) (15.0 y ) / 9.75 12.0 y 15.0 0 elsewhere y 8.5 0 3.175824 0.747253 y 0.043956 y 2 8.5 y 12 F ( y) 2 10 . 538462 1 . 538462 y 0 . 051282 y 12 y 15 y 15 1 Example – Cost/Benefit Analysis of Sprewell-Bluff Project (VI) (Flood Control) Cumulative Distribution Function 1 0.9 0.8 0.7 F(y) 0.6 0.5 0.4 0.3 0.2 0.1 0 8 8.5 9 9.5 10 10.5 11 11.5 12 y 12.5 13 13.5 14 14.5 15 15.5 16 Example – Cost/Benefit Analysis of Sprewell-Bluff Project (VII) (Flood Control) 12 15 15 15 y y3 15 y 2 8.5 y 2 y3 y 8.5 E (Y ) yf ( y )dy y dy y dy 8. 5 12 9 . 75 11 . 375 29 . 25 19 . 5 22 . 75 34 . 125 8.5 12 12 123 8.5(12) 2 8.53 8.53 153 153 15(12) 2 123 29 . 25 19 . 5 29 . 25 19 . 5 22 . 75 34 . 125 22 . 75 34 . 125 59.08 62.77 21.00 31.49 148.35 98.90 94.95 50.64 3.69 10.49 49.45 44.41 11.84 E Y2 12 15 15 y 4 8.5 y 3 15 y 3 y4 2 2 y 8.5 2 15 y y f ( y )dy y dy y dy 8.5 12 9 . 75 11 . 375 39 29 . 25 34 . 125 45 . 5 8. 5 12 12 12 4 8.5(12)3 8.54 8.54 154 154 15(12) 3 12 4 39 29 . 25 39 29 . 25 34 . 125 45 . 5 34 . 125 45 . 5 531.69 502.15 133.85 178.46 1483.52 1112.64 759.56 455.74 29.54 44.61 370.88 303.82 141.21 V (Y ) E Y 2 E (Y ) 141.21 11.84 2 141.21 140.19 1.02 1.02 1.01 2 Uniform Distribution • Used to model random variables that tend to occur “evenly” over a range of values • Probability of any interval of values proportional to its width • Used to generate (simulate) random variables from virtually any distribution • Used as “non-informative prior” in many Bayesian analyses 1 b a f ( y) 0 a yb elsewhere 0 ya F ( y) b a 1 ya a yb yb Uniform Distribution - Expectations E (Y ) b a E Y2 1 1 y y dy ba ba 2 b a 2 b a b 2 a 2 (b a )(b a ) b a 2(b a ) 2(b a ) 2 3 b 1 1 y y2 dy ba ba 3 a b 3 a 3 (b a )( a 2 b 2 ab) 3(b a ) 3(b a ) (a 2 b 2 ab) 3 (a 2 b 2 ab) b a 2 2 V (Y ) E Y E (Y ) 3 2 2 4(a 2 b 2 ab) 3(b 2 a 2 2ab) a 2 b 2 2ab (b a ) 2 12 12 12 (b a ) 2 b a 0.2887(b a ) 12 12 Exponential Distribution • Right-Skewed distribution with maximum at y=0 • Random variable can only take on positive values • Used to model inter-arrival times/distances for a Poisson process 1 y / e f ( y) 0 F ( y) y 0 1 e t y0 elsewhere 1 dt 1 1 t e y 0 e y e 0 1 e y y0 Exponential Density Functions (pdf) Exponential pdf's 1.2 1 f(y) 0.8 f(y|th=1) f(y|th=2) f(y|th=5) f(y|th=10) 0.6 0.4 0.2 0 0 1 2 3 4 5 y 6 7 8 9 10 Exponential Cumulative Distribution Functions (CDF) Exponential CDF 1 0.9 0.8 0.7 F(y) 0.6 F(y|th=1) F(y|th=2) F(y|th=5) F(y|th=10) 0.5 0.4 0.3 0.2 0.1 0 0 3 6 9 y 12 15 Gamma Function ( ) y 1e y dy 0 ( 1) y e y dy Integratin g by Parts : 0 u y du y 1dy dv e y dy v e y y ( 1) y e dy uv vdu y e y 0 0 y 1e y dy 0 0 (0) y 1e y dy ( ) (Recursive Property) 0 Note that if is an integer, ( ) ( 1)! Consider t he integral : 0 y 1e y dy Letting x y : y x dy dx y 0 1 y e EXCEL Function: dy ( x ) 0 1 x e dx =EXP(GAMMALN( 0 x 1e x dx ( ) Exponential Distribution - Expectations E (Y ) 0 E Y 2 1 (2) 2 (2 1)! 0 1 1 y 1 y 2 1 y y e dy ye dy y e dy 0 0 1 2 y 1 y 31 y y e dy y e dy y e dy 0 0 2 (3) 3 2 (3 1)! 2 2 V (Y ) E Y 2 E (Y ) 2 2 ( ) 2 2 2 Exponential Distribution - MGF M (t ) E etY 0 1 0 e 1t y 1 1 y t ty 1 y e e dy dy 0 e dy 1 0 e y * 1 1 y * M (t ) e * 0 1 dy where 1 t * * 1 (0 1) (1 t ) 1 1 t M ' (t ) 1(1 t ) 2 ( ) (1 t ) 2 M ' ' (t ) 2 (1 t ) 3 ( ) 2 2 (1 t ) 3 E (Y ) M ' (0) V (Y ) M ' ' (0) M ' (0) 2 2 2 2 2 * Exponential/Poisson Connection • Consider a Poisson process with random variable X being the number of occurences of an event in a fixed time/space X(t)~Poisson(lt) • Let Y be the distance in time/space between two such events • Then if Y > y, no events have occurred in the space of y Exponentia l " Survival ": P (Y y ) e y e ly ( l y ) 0 Poisson Probabilit y : P ( X ( y ) 0) e ly 0! l 1 Inter - arrivals distances in Poisson Process are Exponentia l with mean 1 l Gamma Distribution • • • • Family of Right-Skewed Distributions Random Variable can take on positive values only Used to model many biological and economic characteristics Can take on many different shapes to match empirical data 1 1 y ( ) y e f ( y) 0 y 0, , 0 otherwise Obtaining Probabilities in EXCEL: To obtain: F(y)=P(Y≤y) Use Function: =GAMMADIST(y,,,1) Gamma/Exponential Densities (pdf) Exponential and Gamma density functions 0.5 0.4 0.3 exp(2.0) f(y) exp(5.0) gam(2,2) gam(2,3) gam(3,2) 0.2 0.1 0 0 2 4 6 y 8 10 Gamma Distribution - Expectations 1 1 1 y y E (Y ) y y e dy y e dy 0 0 ( ) ( ) 1 1 ( 1) ( 1) 1 y 1 y e dy ( 1) 0 ( ) ( ) ( ) ( ) ( ) 1 1 1 y 1 y E Y y y e dy y e dy 0 0 ( ) ( ) 1 1 ( 2) 2 ( 2 ) 1 y 2 y e dy ( 2) 0 ( ) ( ) ( ) 2 2 ( 1)( 1) 2 ( 1)( ) 2 ( 1) ( ) ( ) V (Y ) E Y 2 E (Y ) ( 1) ( ) 2 2 2 2 2 2 2 2 Gamma Distribution - MGF M (t ) E e tY 0 1 1 y dy ety y e ( ) 1 y t 1 1 t y 1 1 1 y e dy y e 0 0 ( ) ( ) 1 1 y * * y e dy where 0 ( ) 1 t 1 * M (t ) ( ) ( 1 t ) ( ) dy M ' (t ) (1 t ) 1 ( ) (1 t ) 1 M ' ' (t ) ( 1) (1 t ) 2 ( ) ( 1) 2 (1 t ) 2 E (Y ) M ' (0) V (Y ) M ' ' (0) M ' (0) ( 1) 2 ( ) 2 2 2 Gamma Distribution – Special Cases • Exponential Distribution – 1 • Chi-Square Distribution – n/2, 2 (n ≡ integer) – – – – E(Y)=n V(Y)=2n M(t)=(1-2t)-n/2 Distribution is widely used for statistical inference Notation: Chi-Square with n degrees of freedom: Y ~ n 2 Normal (Gaussian) Distribution • Bell-shaped distribution with tendency for individuals to clump around the group median/mean • Used to model many biological phenomena • Many estimators have approximate normal sampling distributions (see Central Limit Theorem) f ( y) 1 2 2 e 1 ( y )2 2 2 y , , 0 Obtaining Probabilities in EXCEL: To obtain: F(y)=P(Y≤y) Use Function: =NORMDIST(y,,,1) Normal Distribution – Density Functions (pdf) Normal Densities 0.045 0.04 0.035 0.03 N(100,400) 0.025 f(y) N(100,100) N(100,900) N(75,400) 0.02 N(125,400) 0.015 0.01 0.005 0 0 20 40 60 80 100 y 120 140 160 180 200 Normal Distribution – Normalizing Constant Consider t he integral : e ( y )2 2 2 Changing variables : z k e ( y )2 2 2 y dy e z2 2 dy k (we want to solve for k ) dz 1 dy dz dy dz k e z2 2 dz 1 2 z12 z 22 k 2 2 e dz1 e dz 2 e 2 dz1dz 2 Changing to Polar Co - Ordinates : z1 r cos , z 2 r sin with domains : r (0, ), [0,2 ) and dz1dz 2 rdrd z2 2 z2 1 z12 z 22 2 r 2 cos 2 sin2 2 r 2 k 2 2 dz1dz 2 e rdrd e 2 rdrd e 0 0 0 0 2 2 1 e 1 r2 2 0 2 r 0 1 1 2 2 2 0 0 0 d (0 (1)) d d k 2 k 2 2 2 k 2 2 2 (cos 2 sin 2 1) Obtaining Value of 1/2 From Previous slide, we get : e z2 2 dz 2 1 e dz 2 0 2 1 1 2 1 u Now, Consider : u e du u 1 2 e u du 0 2 0 z2 2 z2 Changing Variables : u du zdz 2 1 z 2 0 2 2 2 e 0 z2 2 1 2 e z2 2 zdz 0 1 z2 2 2 e zdz z 1 dz 2 2 2 Normal Distribution - Expectations 1 12 z 2 Z ~ N (0,1) f ( z ) e 2 1 E ( Z ) z e 2 1 z2 2 1 dz 2 ze 1 z2 2 dz 1 2 e 1 z2 2 0 (0) 0 1 1 12 z 2 1 2 2 z2 dz 2 E Z z e z e dz 2 0 2 1 Changing Variables : u z 2 du 2 zdz du zdz 2 1 2 12 z 2 2 1 2 2 z2 u 2 1 2 ze zdz u e z e dz du 0 0 0 2 2 2 2 2 1 2 2 0 u 3 2 1 e Note : If Y ~ N , u 2 1 1 3 du 2 3 2 2 2 2 V ( Z ) E Z 2 E ( Z ) 1 0 2 1 1 2 then Y Z E (Y ) E ( Z ) E ( Z ) (0) 2 V (Y ) V ( Z ) 2V ( Z ) 2 (1) 2 32 1 1 3/ 2 2 2 1 2 2 2 2 Normal Distribution - MGF 1 1 y 2 1 dy M (t ) E e e exp 2 2 2 2 2 2 y2 1 y ( t 2 ) 2 exp 2 2 dy 2 2 2 2 2 tY ty y2 y 2 exp ty 2 2 2 2 2 dy Completing the square : ( t 2 ) 2 2 2 t 2 t 2 2 2 2 y 2 y ( t 2 ) 2 2 t 2 t 2 2 t 2 t 2 M (t ) exp 2 2 dy 2 2 2 2 2 2 2 2 2 2 2 y 2 1 y ( t 2 ) 2 2 t 2 t 2 2 t 2 t 2 exp 2 dy 2 2 2 2 2 2 2 2 1 y ( t 2 ) 2 2 t 2 t 2 2 1 exp dy 2 2 2 2 2 2 1 y ( t 2 ) 2 2 t 2 t 2 2 1 dy exp exp 2 2 2 2 2 2 The last integral being 1, since it is integratin g over the density of a normal R.V. : Y ~ N t 2 , 2 1 exp t t 2 t 2 t 2 M (t ) exp 2 2 2 2 2 2 Normal(0,1) – Distribution of Z2 1 z M Z 2 (t ) E etZ etz e 2 2 2 2 0 e 1 2 t z2 2 2 dz 2 0 2 2 e 1 dz 2 2 z2 1 2 t e Changing Variables : u z z u and dz M Z 2 (t ) 2 0 e 2 u 1 2 t 12 1 1 2 2 2 1 2t Z 2 ~ 12 1 1 du 2 u 2 dz dz 1 2 1 z 2 t 2 0 u 2 u du 2 1 1 u 1 2 t 2 e du 2 (1 2t ) 1 2 (1 2t ) 1 2 2 Beta Distribution • Used to model probabilities (can be generalized to any finite, positive range) • Parameters allow a wide range of shapes to model empirical data ( ) 1 1 y ( 1 y ) 0 y 1, , 0 ( )( ) f ( y) 0 otherwise Obtaining Probabilities in EXCEL: To obtain: F(y)=P(Y≤y) Use Function: =BETADIST(y,a,b) Beta Density Functions (pdf) Beta Density Functions 4.5 4 3.5 3 Beta(1,1) 2.5 f(y) Beta(2,2) Beta(4,1) Beta(1,3) 2 Beta(5,5) 1.5 1 0.5 0 0 0.1 0.2 0.3 0.4 0.5 y 0.6 0.7 0.8 0.9 1 Weibull Distribution 0 F ( y) y 1 exp y0 y 0 ( , 0) y dF ( y ) 1 f ( y) y exp dy 1 y y exp for y 0 1 y E (Y ) y y exp dy 0 y y 1 Changing variables : u du dy and y (u )1 1 y 1 E (Y ) y y exp dy (u )1 e u du 1 u1 e u du 1 1 0 0 0 y 2 2 2 1 E Y y y exp dy (u ) 2 e u du 2 u 2 e u du 2 1 0 0 0 1 2 2 V (Y ) E Y 2 E (Y ) 2 1 2 1 Note: The EXCEL function WEIBULL(y,*,* uses parameterization: *=, * Weibull Density Functions (pdf) Weibull pdf's 1.2 1 f(y) 0.8 W(1,1) W(1,2) W(2,1) W(2,2) 0.6 0.4 0.2 0 0 1 2 3 4 5 y 6 7 8 9 10 Lognormal Distribution 1 log y 1 2 e 2y 2 2 f ( y) 0 Note : Y * ln( Y ) ~ N , 2 2 y 0, , 0 otherwise 2 12 2 2 e E (Y ) E e M Y * (t 1) exp (1) 2 2 22 2 Y* 2 2Y * EY E e Ee M Y * (t 2) exp (2) 2 Y* e V (Y ) E Y E (Y ) e 2 2 2 2 e 2 Obtaining Probabilities in EXCEL: To obtain: F(y)=P(Y≤y) Use Function: 2 2 2 =LOGNORMDIST(y,,) Lognormal pdf’s Lognormal pdf's 1.2 1 f(y) 0.8 LN(0,1) LN(0,4) LN(1,1) LN(1,4) 0.6 0.4 0.2 0 0 1 2 3 4 y 5 6 7 8