Vous êtes sur la page 1sur 5

# Gregg Lind, 09/26/2003, Stat 8101, Page 1

2.14

FINAL VERSION

2.14

Part A: Let X be a continuous, nonnegative random variable [f (x) = 0 for x < 0]. Show that EX =
0 [1

FX (x)]dx

## where FX (x) is the cdf of X . Answer: EX = Solving

xfX (x)dx d dx FX (x) 0 xfX (x)dx,

0 x

## = fx (x) for fX (x)dx yeilds d[FX (x)] and EX =

d[FX (x)].

After integrating by parts (u = x, du = dx, v = F X (x), dv = dFX (x)) and inserting some convenient terms: EX = xFX (x)| 0 EX = [x(1 EX = 0 0 +
0 FX (x)dx + x|0 x|0 . FX (x)]| 0 + 0 (dx) 0 FX (x)dx. 0 (dx) 0 [1

Which when evaluated, and simplied, yeilds: FX (x)dx = FX (x)]dx the expression of interest.

Part B: Let X be a discrete random variable whose range is the nonnegative integers. Show that EX =
0 [1

FX (x)]dx

## where FX (k ) = P (X k ). Compare this with part (a). Answer: EX = EX =

xX

0 x

x fX (x) =

0 x

fX (x) =

0 x

P (X = x)

(FX (x) FX (x 1)) based on the properties of cdfs. ((FX (x)) FX (x 1) + 1 1) ((1 FX (x)) + (1 FX (x 1))) = FX (x 1))
x=0 x(1 0 x

0 x

0 x

## The summand term for x = 0 is 0 so, splitting the summation: EX =

x=1 x(1

FX (x)) FX (x))

Reindexing: EX = EX =
x=0 (x 0 [1

## + 1)(1 FX (x)) FX (x)]

x=0 x(1

An intergral is a Summand taken over innitely small intervals, and since 1-Fx(x) is monotonic, it is intuitively clear that the limit of the summand would be an integral, linking parts A and B.

## Gregg Lind, 09/26/2003, Stat 8101, Page 2

2.24

2.24

Compute EX and VarX for each of the following probability distributions. (a) fX (x) = axa1 , 0 < x < 1, a > 0 EX = EX 2 =
a1 dx xax

1 a 0 ax dx

axa+1 a+1

a1 |1 0 = ( a+1

a+1

a0a+1 a+1 )

a a+1

2 a1 dx x ax

1 a+1 dx 0 ax a a+2

axa+2 a+2

a1 |1 0 = ( a+2

a+2

a0a+2 a+2 )

a a+2

a2 (a+1)2

## (b) fX (x) = 1/n, x = 1, 2, . . . , n, n > 0 an integer EX = EX 2 =

n 1 i=1 i n

1 n

+
12 n

2 n

+ +
22 n

n n

1+2++n n

n(n+1) 2n

(n+1) 2 .

n 21 i=1 i n

+ +

n2 n

12 +22 ++n2 n

n(n+1)(n+2) 6n (n+1)(n1) 12

(n+1)(n+2) . 6

(n+1)(n+2) 6

(n+1)2 22

n2 1 12

EX =

x(3/2)(x

1)2 dx = (3/2)

2 3 0 x

2x2 + x dx
2 0

## 3 2 2 2 EX = (3/2)(x4 /4) |2 0 (3/2)(2x /3) |0 + (3/2)(x /2) |0 = 6 8 + 1 = 1.

Similarly, EX 2 =

2 x (3/2)(x

1)2 dx = (3/2)

x3 2x2 + x dx

## Gregg Lind, 09/26/2003, Stat 8101, Page 3

2.33A

2.33a

Verify the expression given for the mgf and use the mgf to calculate EX and VarX . (a) P (X = x) = Answer: Fact: For any R, MX (t) =Eetx = MX (t) = e
xX (et )x 0 x! x x=0 x! e x x! ,

MX (t) = e(e

t 1)

, x = 0, 1, . . . ; > 0.

= e .
tx 0 e
t 1)

g (x) P (X = x) = = e ee = e(e
t

e x x! .

. t (, ) so mgf exists.

1 (0) = EX 1 = MX

d1 dt1 MX (t)|t=0

e(e

t 1)

et |t=0 = e(11) e0 = .
d2 dt2 MX (t)|t=0 d dt |t=0
t 1)+t

2 (0) = EX 2 = MX

EX 2 = e(e EX 2 = e(e

= e(e

t 1)+t

d ((et 1) + t dt ) |t=0 .

t 1)+t

(et + 1 |t=0 .

## Gregg Lind, 09/26/2003, Stat 8101, Page 4

2.34

2.34

A distribution cannot be uniquely dened by a nite collection of moments... Let X have the normal distribution, that is X has pdf fX (x) =
2 1 ex /2 , 2

## < x < . P (Y = 0) = 2/3.

Dene a discrete rv Y by P (Y = 3) = P (Y = 3) = 1/6, Show that EX r =EY r for r = 1, 2, 3, 4, 5. Answer: MX (t) =Eetx = MX (t) =
tx e

2 1 ex /2 , 2

< x < .
2

txx2 /2 1 2 e

2txx 1 2 2 e

Which, after inserting some convenient terms and rearranging becomes: MX (t) = MX (t) =
x 1 2 e 2 +2txt2 +t2 2

(x 1 2 e

2 2tx+t2 )+t2 2 2

(xt) 1 2 2 e 2

2 +t2

= et

2 /2

(xt) 1 2 2 e

(xt) 1 2 2 e

MX (t) = et

2 /2

## MY (t) =EY r = MY (t) = MY (t) =

ty y Y e P (Y = 1 t 3 1 t 3 t0 e + e +2 6 6 3e 1 t 3t 3 ( e )+ 2 6 3

y ),

The following table shows rst 5 derivatives of the the MGFs for X and Y and what they evaluate to: r 1 2 3 4 5
r (t) MX 2 tet /2 2 2 t2 et /2 + et /2 2 2 t3 et /2 + 3tet /2 2 2 2 t4 et /2 + 6t2 et /2 + 3et /2 2 2 2 t5 et /2 + 10t3 et /2 + 15tet /2 r (t) MY 1 t3 1 t3 1 [ 3 e + ( 3) e ] 6 2 t3 2 t3 1 6 [3 e + (3) e ] 3 t 3 1 3 t 3 6 [3 e + (3) e ] 4 1 t 3 4 t 3 6 [3 e + (3) e ] 5 1 t 3 + ( 3)5 et 3 ] 6[ 3 e r (0) MX 0 1 0 3 0 r (0) MY 0 1 0 3 0

## Gregg Lind, 09/26/2003, Stat 8101, Page 5

3.5

3.5

A standard drug is known to be eective in 80% of cases in which it is used. A new drug is tested on 100 patients and found to be eective in 85 cases. Is the new drug superior? (Hint: Evaluate the probability of observing 85 or more successes assuming that the new and old drugs are equally eective.) Answer: Let S be a drug success. P (S 85) =
100 100 i=85 i

## .8i .2100i = 0.1285.

While promising this is not enough to choose the new drug over the old. A larger sample size with the same proportion would be more convincing. For example, if there were 850 successes out of 1000 patients, the probability drops to 2.64e05.