SIMPLE RANDOM SAMPLING

√
ESTIMATION OF A POPULATION MEAN AND TOTAL
E ( y )=μ 2 √ V^ ( y ) =2
σ 2 N −n
n N (
=B ) (11)

( )
2
σ N −n --------------------------------------------------------------------------------------
V ( y )= (1)
n N−1 Sample size required to estimate μ with a bound on the error of
estimation B :
2
Estimator of the population mean : Nσ
n= (12)
n
( N −1 ) D+ σ 2
∑ xi (2) B
2
^μ= y= i=1 where D=
n 4

Estimated variance of y : 2 √ V^ ( N y ) = B or, equivalently 2 N √ V^ ( y )

( )
2
^ ( y )= s N −n
V (3)
(13)
n N
Sample size required to estimate τ with a bound on the error of
where estimation B :
n n 2
∑ ( y i− y ) ∑ y 2i −ny 2
2 Nσ
n= (14)
s2= i=1 = i=1 ( N −1 ) D+ σ 2
n−1 n−1 B
2
where D= 2
Bound on the error on estimation : 4N

√
==================================================
2 √ V^ ( y ) =2
n (
s 2 N−n
N )
(4) ESTIMATION OF A POPULATION PROPORTION
-------------------------------------------------------------------------------------- Estimator of the population proportion p :
n
Estimator of the population total τ :
n ∑ xi (15)
N ∑ xi ^p= y= i=1
i =1 (5) n
τ^ =N y=
n Estimated variance of y :

( )
Estimated variance of τ^ :
^ ( ^p )= p^ q^ N−n
V (16)

( )( )
2
^ ( N y )=N 2 s
^ ( τ^ ) =V N −n n−1 N
V
n N Where q^ =1− ^p
(6) Bound on the error od estimation :

√
where
n

∑ ( y i− y ) 2
2 √ V^ ( ^p ) =2
n−1 N (
^p q^ N−n
)
s2= i=1 (17)
n−1
Bound on the error od estimation : You will recall that the required sample size for estimating μ is given

√
by,

( )( )
2
s N−n 2
2 √ V^ ( N y )=2 N Nσ
2
n N n= (18)
( N −1 ) D+ σ 2
(7) where D=B2 / 4
==================================================
SELECTING THE SAMPLE SIZE FOR ESTIMATING
Sample size required to estimate μ with a bound on the error of
POPULATION MEANS AND TOTALS
estimation B :
Npq
2 √ V^ ( y ) =B (8) n= (19)
( N −1 ) D+ pq
2
^ ( y ), is given by :
You will recall that the estimated variance of y , V where D=
B
4
( )
2
^ ( y )= s N −n
V (9) ==================================================
n N SAMPLING WITH PROBABILITIES PROPORTIONAL TO
also SIZE

( )
2
σ N −n Estimator of the population total τ
V ( y )= (10)
n N−1 Let π i = probability that y i appears in the sample,

( )
n
1 yi
The required sample size can now be found by solving the following τ^ pps= ∑ (20)
equation for n : n i=1 π i
 √
Estimated variance of τ^ :

( )
n 2
1 yi
2 √ V^ ( μ^ pps )=2 ∑ − τ^ pps
( )
yi n 2 (25)
1
∑
2
^ ( τ^ pps )=
V − τ^ pps (21) N n (n−1) i=1 π i
n(n−1) i=1 π i ==================================================
Bound on the error od estimation :

√ √ ( )
n 2
1 yi
2 V^ ( τ^ pps ) =¿=2 ∑ −τ pps ¿
n(n−1) i=1 π i
^ (22)

--------------------------------------------------------------------------------------
Estimator of the population mean μ :

( )
n
1 1 yi
^μ pps = τ^ pps =
N
∑
Nn i=1 π i
(23)

Estimated variance of y :

( )
n 2
1 yi
^ ( ^μ pps )=
V 2 ∑
N n (n−1) i=1 π i
− τ^ pps (24)

Bound on the error on estimation :

STRATIFIED RANDOM SAMPLING

ESTIMATION OF A POPULATION MEAN AND TOTAL 2
Where wi is the fraction of observations allocated to stratum i, σ i is
Estimator of the population mean : the population variance for stratum i, and
L 2
1 B
y st = [ N y + N 2 y 2+ …+ N L y L ]=∑ N i y i
N 1 1
(1) D= when estimating μ
i=1 4
2
Estimated variance of y st : B
D= 2 when estimating τ
4N
^ ( y st ) = 1 [ N 12 V^ ( y 1 ) + N 22 V
V ^ ( y L) ]
^ ( y 2) + …+ N L2 V
2
N
Approximate allocation that minimizes cost for a fixed value of

( )
L 2
1 2 N i −ni s i ^ ( y st ) or minimizes V
^ ( y st ) for a fixed cost :
¿ 2 ∑ Ni (2) V
N i=1 Ni ni
Bound on the error on estimation : ni =n
( N i σ i / √ ci
N 1 σ 1 / √ c1 + N 2 σ 2 / √ c 2 +…+ N L σ L / √ c L )
√
L
1
2 √ V^ ( y st )=2 ∑ ¿¿¿ (3) (9)
N 2 i=1

( )
-------------------------------------------------------------------------------------- N i σi / √ ci
¿n L
Estimator of the population total τ :
L ∑ N k σ k / √ ck
N y st =N 1 y 1 + N 2 y 2+ …+ N L y L =∑ N i y i (4)
k=1

i=1 ci denotes the cost obtaining of a single observation from the ith stratum.
Estimated variance of N y st : Substituting the ni/n given by Equation (9) for wi in Equation (8) gives

( )
( )
2

( )(∑ )
L
^ ( N y st ) =N 2 V^ ( y st )=∑ N i2 N i−ni s i
L L
V
i=1 Ni ni
(5) ∑ N k σk / √ ck N i σi / √ ci
k=1 i=1
Bound on the error od estimation : n= L
(10)
N D+ ∑ N i σ i

√∑
2 2

( N i −ni s i2
)
L
2 √ V^ ( N y st )=2
2
Ni (6) i=1

i=1 Ni ni for optimal allocation with the variance of y st fixed at D.

==================================================
SELECTING THE SAMPLE SIZE FOR ESTIMATING If c1 = c2 = … = cL, then the cost terms cancel in Equation (9) and

( )
POPULATION MEANS AND TOTALS
Ni σ i
ni = nwi i = 1, 2, …, L (7) ni =n L
(11)
Approximate sample size required to estimate μ or τ with a bound ∑ Ni σi
i=1
B on the error of estimation :
L
This method selecting n1, n2, …, nL is called Neyman allocation.
∑ N i2 σ i2 /wi Under Neyman allocation Equation (10) for the total sample size n
i=1
n= L (8) becomes
N D+ ∑ N i σ i
2 2

i=1
 (∑ )
L 2 ci denotes the cost obtaining of a single observation from the ith stratum.
2 2
N i σi
i =1 Recall that the allocation formula (9) assumes a very simple form when
n= L
(12) the variances as well as costs are equal for all strata. Equation (19)
N 2 D+ ∑ N i σ i2 simplifies in the same way, provided all stratum proportions p i are
i=1 equal and all cost ci are equal. Then Equation (19) becomes
In addition to encountering equal costs, we sometimes encounter
2 2 2
approximately equal variances, σ 1 , σ 2 , … , σ L . In that case the σ i
❑ ni =n ( )
Ni
N
i=1 ,2 , … , L (20)

cancel in Equation (11) and As previously noted, this method for assignment of sample sizes to the

( ) ( NN )
Ni i strata is called proportional allocation.
ni =n L
=n --------------------------------------------------------------------------------------
(13)
∑ Ni
( )
L 2 L 2 L
2 si N i−ni 2 si 1
i=1 V ( y st ) =∑ W i
^ =∑ W i − ∑ W i s i2
i=1 ni Ni i=1 ni N i=1
Under proportional allocation Equation (8) for the value n, which
yields V( y st ¿ = D, becomes (21)
L

∑ N i2 σ i2
n= i=1

1
L (14)
E
( ) 1
≈
1 1−W i
+
n i nW i n i W i2
(22)

ND +
N
∑ N i σ i2
i=1 Subtitution of the expression in (22) for 1/ni Equation (21) yields
------------------------------------------------------------------------------------- L
Estimator of the population proportion p : ^ p ( y st ) = N −n ∑ W i2 s i2 + 1 ( 1−W i ) s i2
V (23)
L Nn i=1 n2
1
^pst =
N 1 1
[ N ^p + N 2 ^p 2+ …+ N L ^p L ]=∑ N i ^pi (15)
i=1 Suppose the first phase sample of n’, is used to determine which
Estimated variance of ^pst : elements fall into the various strata. Let
ni '
^ ( ^pst ) = 1 [ N 12 V^ ( ^p1 ) + N 22 V
V ^ ( ^p 2) + …+ N L2 V ^ ( ^p L ) ] w i=
'
i=1 , … , L (24)
N
2 n'
wi' is an unbiased estimator of W i, assuming randomness of the first
2 N i −ni ^pi q^ i
( )( )
L L
1 1
¿ 2 ∑ N i V^ ( ^pi ) = 2 ∑ N i phase sample.
2

N i=1 N i=1 Ni ni−1

(16)
Bound on the error of estimation :

√ ( )( ) ^pi q^ i
L
1 N −n
2 √V^ ( ^p st )=2 2∑
N i2 i i
N i=1 Ni ni −1
(17)
--------------------------------------------------------------------------------------
Approximate sample size required to estimate μ or τ with a bound
B on the error of estimation :
L

∑ N i2 p i q i /wi
i=1
n= L (8)
N D+ ∑ N i p i qi
2

i=1

Where wi is the fraction of observations allocated to stratum i, pi is the

population proportion for stratum i, and
2
B
D=
4

Approximate allocation that minimizes cost for a fixed value of

^ ( y st ) or minimizes V
V ^ ( y st ) for a fixed cost :

ni =n
( N i √ pi qi / ci
N 1 √ p1 q1 /c 1 + N 2 √ p2 q2 /c 2 +…+ N L √ p L q L /c L )
(9)

( )
N i √ p i q i /c i
¿n L

∑ N k √ pk q k /c k
k=1

(19)