24 SOME PROPERTIES OF EXPECTED VALUES a
Freure 2.9 The COF of amixed distribution
Po
minutes is
PLX < 05] =04 + 0.6(1 —e°4} = 0.636
The distribution of X given 0 < X corresponds to
PlO
ena)
An important special expected value Is obtained if we consider the funetion
wx) =e Example 24.9
24. SOME PROPERTIES OF EXPECTED VALUES 7
Definition 2.8.9
“The vatinnes of random variable X is piven by
[ Var) = 2 = 8) eas)
Other common notations for the variance are 62, a}, or V(X), and a related
quantity, called the standard deviation of X, is the positive square root of the
variance, ¢ = oy = /Var(®).
The variance provides @ measurc of the variability or amount of “spread” in
the distribution of a random variable.
In the experiment of Example 2.2.2, E(X?) = 23(1/2) + 4°(1/4) + 8%(1/4) = 22, and
thus Var(X) = 22—4? = 6 and oy = /6 = 245. For comparison, consider a
slightly different experiment where two chips are labeled with zeros, one with a4,
and one with a 12. If one chip is selected at random, and Y is its number,
then E(¥)=4 as in the original example, However, Var(¥) = 24 and o
= 276 > ox, which reflects the fact that the probability distribution of ¥
‘more spread than that of X.
Certain special expected values, called moments, arc useful in characterizing
some features of the disttibution.
——
inition 2.4.2
The Ath moment about the arigin ofa candor variable X is
= BO) 288
‘and the eh moment abou the enon is
p= BLY ~ BUX) = 20 at aan
‘Thus EQ") may be considered as the kth moment of X or as the first moment
of X*, The first moment is the mean, and the simpler notation 4, rather than 4,
generally is preferred. The first moment about the mean is zero,
Hy = BLX — E(X)] = EX) ~ EX) = 0
‘The second moment about the mean is the variance,
By = EEX — =o? cpgenaereeamaten nen
74 HAPTER 2. RANDOM VARIABLES AND THEE rsra/oUTIONS
rand the second momeat about the origin # involved in the following theorem
aabont the variance.
Freorem 243 CX isarandom variable, shen
vec(X) = BX?) - 2
Proof
yas( i) = BX? = 2X +7")
= BUX?) — HBO) tH?
RX) e+
which yields the theorem.
‘also follows immediately that
aso te
as noted previously, the variance provides a measure of the ‘amount of spread
jae distripation ot the variability moos ‘members of a populati
ve ane example of this occurs when assumes only ome
PLEX = 6) = 1 In this caso BU) = 6 snd Var(X) = ©
Ae cask following Theosets 2.42 dealt rit the expected valuc of a linear
fonction of a random variable. The followin theorem deas with the va
pheorem 2.44 s6X is random variable ‘and g and bare constants, ther
var(ax + B= @ Vat()
Proot
arta +8) = BUX +b — Fx ol
= Lee —
=o? VarlX)
Fis means thatthe variance is affected PY ® dxange of sale, but not by a (ATS
tation.
reser natural measure of variability would be the mean absolute
B\X —p, but the variance 1s generally a more convenient quantity vit
to work. Theorere 2.4.5
Theorem 2.4.8
2 SOME PROPERTIES OF EXPECTED VALUES 5
The mean and variance provide a good deal of information about a population
distribution, but higher moments and other quantities also may be useful. For
example, the thicd moment about the mean, 25, is a measure of asymmetry or
“skewness” of a distribution.
Hf the distribution of X is symmetric about the mean v= £(X), then the third
moment about jis 22r0, us = 0.
Proot
See Exercise 28, a
‘We can conclude that if ., #0, then the distribution is not symmetric, but not
conversely, because distributions exist that are not symmetric but which do have
Ms = 0 (see Exercise 29).
SOUNDS ON PROBABILITY
tis possible, in some cases, to find bounds on probabilities based on moments.
TCX isa random variable and u(x) is a nonnegative real-valued function, then for
any positive constant ¢ > 0,
pray ><) < SX. ett
Proof
HA = {x[u) > c), then for a continuous random variable,
Ba) = £ 93) be
- fuore are f wif) ax
> fave ae
5 [erie
= cP[X € A]
= cP(MX)> 76
Yaeorem 24.7
quapren 2. RANDOM VARIABLES AND THER pisTAIBUTIONS
‘A similar proof holds for disoete variables a
‘A special case, knovin as the Marker inequality, is obtained if xf) = 1s for
+> O,namely
puxiza sO si)
“Another well-known result the Chebycher Ineqoality, is given by the following
theorem.
if.X isa random variable with mean 1 and vari2ne then for any k >
pax -m> eSB ata
Praot
wasp =(X Wee = (202, then using equation (2+ AA,
so DE oS
_ tone a 1x any
and if we let 6 ko, then
2
\x—wl — | 1 for some integer 1% 1, and
conversely. Thus,
fh 1
(¥#e)=U [ix ~#l >]
and using Boole’s inequality, equation (1.4.5), we have
PIX #41 < Safix -ni>f]
and using equation (2.4.16) we obtain
PU 4x Zot =o
which implies that PX = 42] ~ 1. a
APPROXIMATE MEAN AND VARIANCE
if a function of a random variable, say H(X), can be expanded in a Taylor series,
thea an expression for the approximate mean and variance of HX) can be
obyained in terms of the mean and variance of X.
For example, suppose that H(x) has derivatives HG), °C), ... in an open
interval containing jz = £(X). The function H() has a Taylor approximation
about 4,
Rl
Hy) + als — a) + SHON — a (2437) cunaren 2. RANDOM VARIABLES AND THE OISTROUTIONS
which suggests the approximation
BTHOO) = Hu) + we
and, using the first two terms,
‘Var HX] = Le wit?
where o? = VarlX),
“The accuracy of
function H(x) as well as oa the amount
(2.4.18)
2.449)
these approximations depends primarily 09 the nature of the
‘of variability in the distribution of
ee
| exannple 2.4.5 Let ¥ bo a positive-valued random variable, and et HG) =In %, 80 that
BG) = Lx and H%@) = — Us? Iefollows that
Afi XJ sin a+ G- a)
and
2.5
Bh
BIOMENT GENERATING FUNCTIONS
1h special expected value that is quite usefl is Known as the moment generating
function.
Definition 2.8.1
1.x isa random vatiabl, then the expected value
Mp = Be) sn)
taneion (AGE) ot X sf shi expected ven ents For
js called the moment quoessting
the form —h << ffor some h > O-
fil values of ria some intorvel of
ju desirable 10 suppress the sabscrig and use the simplet
In some situations it
notation M(O. Exanpie 2.5.4
Theorem 2.5.1
28 MOMENT GENERATING FUNCTIONS 79
‘Assume that X is 2 discrete finite-valued random variable with possible values x).
syne The MGE is
Md)
‘which i differentiable function of 1, with derivative
mse = Series)
and in goncral, for any poshve integer r
mee) = Senha)
Notice that if we evaluate Mp1) at ¢ = 0 we obtain
MPO) = Yi xp feled = BO)
the rth moment about the origin. Tt
ing in a power series about ¢—
= Ar.
These properties hold for any random variable for which an MGF exists
although a general proof is somewhat harder.
is also suggests the possibility of expand-
Mx) = cot etter bo, where ¢,
ithe MGF of X exists, then
AX) = MYO) for ally = 1, 2,... (2.6.2)
myo=1+ § 08 ass
a
We will consider the case of a continuous random yariable X, The MGF for a
continuous random variable is
mat) = | ego ae
When the MGF exists, it can be shown that the rth derivative exists, and it can
bbe obtained by differentiating under tho integra! sign,
mei [ee de HAPTER 2 RANDOM VARIABLES AND THEIR DISTRIBUTIONS
fom which it follows that for all r= 4,2,
ee) = f xfre) dx = f xoPMfa) dx = MEO)
when the MGF exit it also ca be dhown that a power snes ORR about
Wher possible, and from standard results about power Site the coefficients
zero i lor (oy. We combine this wth the above result to obtain
= MpO;! 2 Bt
seqo=ie § OOF mee
‘The discrete casc is similar. a
Consider 2 continuous random variable X with pats)
otherwise. The MGF is
H* if x > 0, and 22r0
5 Ot
‘The first derivative is My@)=el(@—e)~4, and thus EU) = My(0)
26(2 — 29}? = 1. It is possible to obtain higher derivatives, but the complox-
ity increases with the order of the derivative,
PHOPERTIES OF MOMENT GENERATING FUNCTIONS
MEY = aX + b, then Mf {9 = eM, (a,
Proof
Md = Be)
Rett
Betti)
Ae)
= eM tap 5
One possible application is in computing the rth moment about the mean,
FLX — ji], Because Mz._,(0) = eM (3),
eux — w= Ste MO ne (254)
Ie can be shown that MGFs uniquely determine a distribution.
Suppose, for example, that X and ¥ ate both integer-valued with the same set of
possible values—say 0, 1, and 2—and that % and ¥ have the same MGF,
MO = eho = EHH)
if we let s=e and co, =f4) — fd) for i= 0, 1, 2, then we have cote,s
+c zs*=0 for all s>0. The only possible coefficients are co = 0,
which implies that (el) = Ji) for = 0, 1,2, and consequently Xvand ¥ have the
same distribution,
In other words, X and Y cannot have the same MGF but different pas. Thus,
the form of the MGF determines the form of the pat
‘This is true in general, although harder (o prove in general uarren 2. RANDOH VARIABLES AND THE DiSTAIBUTIONS
Uniqueness 16%, and X, have respect CDF Fy) and F,{x), and MGFs
May(f) and M,{0, then FC) = #68) for veal x stand only if M,G) = M6) foe
MA Seome interval ~h << for some h > 0 a
For nonnegative integer-valued randort variables, the derivation of moments
often is made more tractable by Bist concidering another type of expectation
‘gnown as a factorial moment
FACTORIAL MOMENTS
Definition 2.8.2
“The rth fastortal moment of X is
apex — ee or (258)
‘and the factorial mornent generating function (PMGE) of X is
(25.8)
in gome interval ofthe form 1— her 31
{@) Find the probably thatthe sore an 068 integst
a bag eontins tres coins one of which has 0 Bead ‘poth sides while the other te
ae oral A cia is chosen at rancor fom the PAB ‘and tossed three times: The
carbon of heads is condom vatiable, say %.
(a) Find the discrete pot X. (Hint: Use the Last of Totat Probability with B, = &
eoemat coin and , = two-neaded coin)
(py sketch the dizerete pal andthe CDF of %
se box contains five colored bal, wo blac and then SS alls are dravin successively
oo replacement FX isthe nurber of draws unt the fast black ball is obtained, Bnd
the discrete pal ft)
ac digorate candor variable has pat 4%)
(o) Wyte) = MUA for xm 1, 2,3;and ete oe find k.
{b) tnx function of the form 7) = MOAI yD] for x= 0, 12a pat or any B?
‘Denote by [x] the arestest integer aot excoodion Tt the palin Example 22.1, show
Donat WEDE can be represented 18 Fx) = (EXV/U2F fr Dex < 132070 ifx < 02nd
one it 1
Ap diveete rundown variable Xba apa of he Forma) = ag — 2) for ¥= 012245,
tan 720 otherwise
{@) Find the eonslant
() Find the ODF, Fle
(g Find PLX > 21
(a) Find ECO.
nonnegative integet-valued random variable Has A CDF of the form
SU" for ¥= 0.1 2e--and zeros < 0
() Find the pat of 8
( Find P10< x < 20h
fe) Find PLY is even
Somedmes i i desirable to asian aumorical en values to experimental responses (hat
Som basally of nureroal type: For exam7pe i ATS the color preferences of
are ctental bie! suppose that so colors Bs SEH: ‘and red oceut with probabilities exercises a5
1/4, 1/4, and 1/2, respectively, different integer value is assigned to each color, and this
corresporids to a random vaciable X that can take on one of these three integer values.
fa) Can fs) = (18) 172" 1 for x= =, 10 be used 8 pes for thi ceperimont?
() Can fis) = (e (2) for x = 0,1, 2 be used?
(©) Cen fe) = 0/4 for = =1,0,2 be wed?
10, Let X be a discrete candom variable such that P[X
4 and PLX ~ x] = O otherwise. Suppose the CDI
x= 12 3,0r4,
(@) Skoteh the graph of the CDF,
(b) Sketeh the graph of the disereve pal /(x)
(6) Find BX).
J >0 if x= 1,230
i F(x) = OSx(1 +x) atthe values
47. A player rolls a sixsided die und receives a number of dollars corresponding to the
hhumber of dots on the face that turns up. What amount shoulé the player pay for rolling
to make this a fale” game?
472. A continuous random variable X has pa given by f(x) = ofl — xh? if 0- 1/2}.
(@) Find P(X < 1.25),
(© Whatis PLX = 1251?
48. Acontinuows random variable X/has apa ofthe form f(x) = 23/9 for O-< > <3, and
aro otherwise.
(@) Find the CDF of X
b) Find PLY < 2.
(@) Find P[-1 mi
fe) Find £(3),
49, A candom variable X bas the pal
x fOcx<1
fayaraa iti fhs fegult extends to
aux Th Peal de
= [Fay aet
ta) Use Chebyee's inequality to obtain alower bond on ppsg
uppone atx is atandom varale wth MGR MA) = CIR + cyme + GB
{@) Whats the distribution of X7
{o) Wat is POX = 22?
‘prove Theorem 24 fora sonnegatveinteger-vlued random variable
“Assome thay X sa continuous andor: variable wth po
“fs) exp Ele +2) Hf —2.< x < co: and ao OMENS
(Gq) Finé the moment generating fonction of ©
{hy Use the MGP of et ind BOX) and EOS
ee the FMGF of Example 2550 find FLX — NK — 2h an then find BP),
tn Eveecie 26, sppoe instead of ordering our copies of PS software package, the store
Ue eae copes 0 << ah Then he number sold S47 5 he smaller of¢ oF
(a) Express the nt prof ¥ as. lincar function ofS
{b) Find B(Y for cach value of cand india the soTavon © ‘that maximizes the
expected profit.
Show that of = £00 — D]~ ae 3 exERcises ae
Let ya(@)— In LMC), where M0 is] MOF. The function fis calle the cumulant
ageverating fonction of X, and he value ofthe rth derivative evaluated at 1 = 0, x, = 10),
i ealled the rth cumlant of X.
(a) Show mat w= ¥(0)
{b) Show that a? ~ 40)
(Q, Use y(t) to find ys and c? for the random variable of Exercise 26
(2) Use yafe to finds and o? for the random variable of Example 25.5,
