SURV 440

SURVMETH 617/STATISTICS 580/BIOSTATISTICS 617/SOCIOLOGY 717

Methods And Theory Of Sample Design

Design Summary

Design Estimated mean Estimated sampling variance Design effect

SRS (without replacement) y 1− f 2

• f = n/N y= =p varsrs ( y ) = s
n n
1.0
⎛1⎞ n 1 ⎛ n 2 y2 ⎞
= ⎜ ⎟ ∑ yi s =
2
⎜ ∑ yi − ⎟
⎝ n ⎠ i =1 n − 1 ⎝ i =1 n ⎠
1− f
varsrs ( p ) = p (1 − p )
n −1
Stratified random sampling
H H
1 − fh 2 var ( yw )
• f = n/N
yw = ∑ Wh yh var ( y ) = ∑ Wh2 sh
• H strata nh varsrs ( y )
h =1
• n = ∑ nh h =1
• N =∑Nh 1 ⎛ nh 2 yh2 ⎞
s = ⎜ ∑ yhi − ⎟
H 2
• f h = nh / N h pw = ∑ Wh ph h
nh − 1 ⎝ i =1 nh ⎠
• W h = N h /N h =1 H
⎛ 1 − fh ⎞
• y h = ( 1/ n h ) ∑ y hi var ( p ) = ∑ Wh2 ⎜ ⎟ ph (1 − ph )
i h =1 ⎝ h ⎠
n − 1
Proportionately allocated
y 1− f H
1− f 2 sw2
stratified random sampling
• f h= f y= =p var ( y ) =
n
∑ Ws =2
h h
n
sw s2
< 1.0

• n h /n = N h /N
n h =1

⎛1⎞ n 1− f H
⎛ nh ⎞
= ⎜ ⎟ ∑ yi var ( p ) = ∑W h ⎜
−
⎟ ph (1 − ph )
⎝ n ⎠ i =1
n h =1 ⎝ h ⎠
n 1

1
Other allocations
H H
1 − fh 2
yw = ∑ Wh yh var ( y ) = ∑ Wh2
• Equal:
sh
n h = n /H nh
• Neyman: h =1
h =1 var ( yw )
H
1 − f h ⎛ nh ⎞ varsrs ( y )
W hS h
•n H
var ( p ) = ∑ Wh2 ⎜ ⎟ ph (1 − ph )
pw = ∑ Wh ph
nh =
∑W hS h h =1 nh ⎝ nh − 1 ⎠
h =1
Systematic sampling
y 1− f 2 1.0
var(1) ( y ) =
• k = N /n
• Random start from 1 to k y= =p s
• Fractional intervals n n
⎛1⎞ n 1− f H sw2
∑
• Proportionate allocation
across zones so that = ⎜ ⎟ ∑ yi var(2) ( y ) = Wh sh2 s2
< 1.0

W h = 1 /k ⎝ n ⎠ i =1 n h =1
• Not measurable unless
1− f n/2
var(3) ( y ) = 2 ∑ ( yh1 − yh 2 )
2
more than one random start < 1.0
is used n h =1

1 − f n −1
var(4) ( y ) = ∑ ( yhg − yh , g +1 )
< 1.0
2

2n ( n − 1) g =1
[Degrees of freedom: (1) SRS, (n-1); (2) Stratified random,
(n-H); (3) Paired differences, (n/2); (4) Successive
differences, from (n/2) to (n-1).]
Cluster sampling
1− f 2
⎛1⎞ a B var ( y ) =
y = ⎜ ⎟ ∑∑ yαβ
(equal-sized "take all" sa
selection) a
a B a
• f= • = ⎝ n ⎠ α =1 β =1 1 ⎛ a 2 y2 ⎞ 1 + ( B − 1)roh
sa = 2 ⎜ ∑ ya − ⎟
2
b B A
• n= a • b B ( a − 1) ⎝ α =1 a ⎠
1 ⎛ a 2 2⎞ 1 ⎛ a 2 ⎞
= ⎜ ∑
( a − 1) ⎝ α =1
y a − ay ⎟ = ⎜ ∑
⎠ ( a − 1) ⎝ α =1
pa − ap 2 ⎟
⎠

2
Cluster sampling
⎛1⎞ a b 1− f 2
var ( y ) =
y = ⎜ ⎟ ∑∑ yαβ
(equal-sized subsamples) sa
• First stage: with a
replacement ⎝ n ⎠ α =1 β =1 1 ⎛ a 2 y2 ⎞ 1 + (b − 1)roh
sa = 2 ⎜ ∑ ya − ⎟
• Second stage: without 2
replacement b ( a − 1) ⎝ α =1 a ⎠
a b n
• f= • = 1 ⎛ a 2 2⎞ 1 ⎛ a 2 ⎞
A B N = ⎜ ∑
( a − 1) ⎝ α =1
y a − ay ⎟ = ⎜ ∑
⎠ ( a − 1) ⎝ α =1
pa − ap 2 ⎟
⎠
Cluster sampling
a b Taylor Series approximation:
∑∑
(unequal-sized clusters)
• PPS selection yαβ var ( r ) ≈
1
⎡ var ( y ) + r 2 var ( x ) − 2r cov ( y, x ) ⎤⎦
• First and second stages α β
=1 =1 2 ⎣
with and without y =r= a b
x
With replacement selection of clusters:
replacement,
respectively ∑∑
α β
=1
xαβ
=1
⎛ 1 ⎞⎛ a ⎞ ⎡ a 2 2 a 2
var ( r ) ≈ ⎜ 2 ⎟⎜ ⎟ ⎢∑ a y + r ∑ x − 2 r ∑
a
⎤
yα xα ⎥
a M α b* ⎝ x ⎠⎝ a − 1 ⎠ ⎣ α =1
a
• f= • α =1 α =1 ⎦ 1 + (b − 1)roh
∑Mα Mα
Systematic selection of clusters:
• x = sample size
⎡ a −1 a −1
2⎤
∑ ( ) ∑ ( )
2
• b = x /a ⎢ y − y g +1 + r 2
x − x g +1 ⎥
⎛ 1 ⎞ ⎛ a ⎞ ⎢ g =1
g g

var ( r ) ≈ ⎜ 2 ⎟ ⎜⎜ ⎥
α =1
⎟⎟
⎝ ⎠⎝ ( − ) ⎢ a −1 ⎥
( yg − yg +1 )( xg − xg +1 ) ⎥
x 2 a 1 ⎠ −2r
⎢
⎣
∑
α =1 ⎦

3
Stratified cluster sampling
H a b Multiple selections:
∑∑∑
• Ultimate cluster sampling
• Implicitly weighted y αβ ⎡ H ⎛ ah ⎞ ⎛ a 2 yh2 ⎞ ⎤
⎢∑ ⎜ ⎟ ⎜ ∑ yha − ⎟
h
quantities y hα , x hα α β
⎥
h =1 =1 =1 −
r= ⎢ h =1 ⎝ a h 1 ⎠ ⎝ α =1 ah ⎠ ⎥
H a b ⎢ ⎥
∑∑∑ ⎛ 1 ⎞ ⎛ a ⎞⎛ a 2 xh2 ⎞
H
x αβ var ( r ) ≈ ⎜ 2 ⎟ ⎢ + r 2 ∑ ⎜ h ⎟⎜ ∑ xha − ⎟ ⎥ 1 + ( b − 1) roh
⎝ x ⎠⎢ − ⎥
h
h =1 ⎝ h
a 1 ⎠⎝ α =1 ah ⎠
α β
h =1 =1 =1 ⎢ ⎥
⎢ H
⎛ ah ⎞ ⎛ a yh xh ⎞⎥
⎢
⎢⎣
− 2 r ∑ ⎜ ⎟ ⎜ ∑ yha xhα −
h =1 ⎝ ah − 1 ⎠ ⎝ α =1
⎟⎥
ah ⎠ ⎥⎦
Paired selections:
⎡H ⎤
⎢ ∑ ( yh1 − yh 2 )
2
⎥
⎢ h =1
⎥
⎛ 1 ⎞⎢ 2 H ⎥
var ( r ) ≈ ⎜ 2 ⎟ ⎢ + r ∑ ( xh1 − xh 2 )
2
⎥
⎝ x ⎠⎢ h =1
⎥
⎢ H
⎥
⎢
⎣
− 2 r ∑
h =1
( y h1 − y h2 )( xh1 − x )
h2 ⎥
⎦
Successive differences:
⎡H ⎛ ⎞ ah −1 ⎤
⎟⎟ ∑ ( yhg − yh , g +1 )
ah
⎢ ∑ ⎜⎜
2
⎥
⎢ h =1 ⎝ 2 ( ah − 1) ⎠ g =1 ⎥
⎢ ⎥
⎛ 1 ⎞⎢ 2 H ⎛ ⎞ ah −1
⎟⎟ ∑ ( xhg − xh , g +1 )
var ( r ) ≈ ⎜ 2 ⎟ + r ∑ ⎜⎜
ah 2
⎥
⎝ ⎠
x ⎢ h =1 ⎝ ( h
2 a − 1) ⎠ g =1 ⎥
⎢ ⎥
⎢ H ⎛ ⎞ ah −1 ⎥
⎟⎟ ∑ ( hg h , g +1 )( hg h , g +1 ) ⎥
ah
⎢ −2r ∑ ⎜⎜ y − y x − x
⎢⎣ h =1 ⎝ 2 ( ah − 1) ⎠ g =1 ⎥⎦