CONVERGENCE OF RANDOM VARIABLES 29
it is not comparable with the almost sure convergence.
tees the same kind of convergence of
h(X,,) to h(X). For example. if X,
and h(z) is continuous. then h(X,)
h(X,) 5 h(X) and h(X,) + h(X).
If h(z) is a continuous mapping, then the convergence of X, to X guaran-
X
h(X). which further implies that
Laws of Large Numbers (LLN). For i.i.d. random variables XI. X2, . . .with
finite expectation
EXl = p. the sample mean converges to p in the almost-sure
sense. that is,
Sn/n - p, for S, = XI - . . . + X,. This is termed the strong
law
of large numbers (SLLN). Finite variance makes the proof easier, but it
is not
a necessary condition for the SLLN to hold. If. under more general
conditions.
Sn/n = X converges to p in probability. we say that the weak
law
of large numbers (IYLLK) holds. Laws of large numbers are important in
statistics for investigating the consistency of estimators.
as
Slutsky's Theorem. Let {X,} and {Y,} be two sequences of random variables
on some probability space. If
X, -Y, --+ 0. and Y, + X. then X, ==+ X.
P
Corollary to Slutsky's Theorem. In some texts. this is sometimes called Slut-
sky's Theorem.
If X, --r. X. Y, 5 a. and 2, + b, then X,Y, + 2, ==+
aX + b.
P
Delta Method. If EX, = p and VarX, = c2. and if h is a differentiable function
in the neighborhood
of /-1 with h'(p) # 0. then fi(h(X,) - h(p)) ==+ W.
where W - N(0. [h'(p)I2a2).
Central Lzmzt Theorem (CLT). Let XI, X2. . . , be i.i.d. random variables with
EX1 = p and VarXl = a2 < m. Let S, = XI + . . . + X,. Then
=* 2,
S, - np
42
where 2 - N(0. 1). For example, if XI,. . . , X, is a sample from population
with the mean
/L and finite variance u2. by the CLT. the sample mean X =
(XI + . . 1 X,)/n is approximately normally distributed, x "z' N(p. 02/n),
or equivalently. (+(X - p))/o - hr(0. 1). In many cases, usable approxi-
mations are achieved for
n as low as 20 or 30.
wpr
Example 2.4 Iz'e illustrate the CLT by LIATLAB simulations. A single
sample of size
n = 300 from Poissoii P(1/2) distribution is generated as
sample = poissrnd(l/2, [I, 3001 ) ; According to the CLT. the sum ,9300 =