MAS 201: SAMPLE SURVEYS

PROBLEM SET 2

1 A population of size 1 000 is divided into 3 strata. Their sizes and sd are given below:

Strata 1 2 3
Size 200 300 500
SD 6 8 12

A stratified sample of size 120 is to be drawn from the population. Determine the sample sizes in case
of:

) proportional allocation {Answer:  =24,  =36, =60}

(

) optimum allocation. {Answer:  =15,  =30,  =75.}

(

2. Let a population consist of 8 numerals {1, 2, 3, . . ., 8}. Suppose a sample of size 3 is selected from
it. Find the:
 
) inclusion probability {Ans: Inclusion prob = = }
(

( )
) joint inclusion probability {Ans: Joint inclusion prob =(
( )
=…… Note n=3, N=8}

3 Explain the concept ‘’simple random sampling’’ { srs is a method of selecting n units out of N in the
target population in such a way that each combination of n units has the same prob of being selected.}

4 Show that  is an unbiased estimator of .

In case the observations take the values 0 or 1, does this imply that ‘’pq’’ is an unbiased estimator of

‘’PQ’’ ? {1ST Part check your Lecture Notes. 2nd part: we know that   = and that E(  ) =   .
 

 
Thus E( ) =  

 
Or E(pq)=
   

( )
From which we have: E(pq)= ( )
≠ .  ′′ ! "" # $% & '(#&) %#*&%$+ $, ′′′′

5 (-
-) The households in a town are to be sampled in order to estimate the average amount of assets
that are readily convertible into cash. The households are stratified into a high and low rent stratum.
A household in the high rent stratum is thought to have about 9 times as much assets as are in the low
rent stratum and the th stratum variance is expected to be proportional to the square root of the
mean of the th stratum. There are 4 000 households in the high rent stratum and 20 000 in the low
rent stratum. How would you distribute a sample of 1 000 households between the two strata?
 .) If the cost function is of the form C=/0 +∑ /2 32 where /2 is the cost per unit and /0 is the initial
(.
cost, by minimizing var(4 5 6 ), find the value of 32

789:;<8=

(-) Let the high rent stratum be represented by N and low rent stratum by N

A household in the high rent stratum is thought to have about 9 times as much assets as are in the
low rent stratum ⇒ WX 9W

The th stratum variance is expected to be proportional to the square root of the mean of the th
stratum ⇒] ∝ _Ẁ] (a is proportional to the square root of the mean of the th stratum)
b f
5g
⇒] ∝ (Ẁ]c ) ⇒] = de 

We’re given N = 4 000 and N =20 000, n=1 000.

b b
] q]
We know that ] = ∑ , where ∑ N] ] =N  + N  =4 000dẀc +20 000dẀc
] q]

b
 q  ttt(u ttt)vẁbc
 =∑ b b = b b
s qs
u tttvẁbc xt tttvẁyc

b b b
u tttv(zẁy )c
= b b (after replacing Ẁ by (9Ẁ )c as from above and dividing through
c

uv(zẁy )c xtv(ẁyc )
numerator/denominator by 1 000)
b
 ttt√
= (Note that 9c =√3 )
√x}

=257.3
b b
 q  ttt(t ttt)vẁyc  ttt(t ttt)vẁyc t ttt
 = ∑ y y = b b = b b = =742.7
s qs √x}
u ttt v ẁbc xt tttvẁyc u tttv(zẁy )c xt tttvẁyc

Thus the sample of 1 000 households is to be distributed in the following manner:

High rent stratum=  =257.3≅ 257

Low rent stratum= =742.7≅ 743

(.) The cost function is given by C=†t +∑ †a a . We wish to min Var(ˆ`‰Š ) wrt a subject to the
condition C=†t +∑ †a a . Using Lagrange’s Multipliers, we’ve:

 qsy
F(a , )= ∑ Na (Na − a ) +λ(†t + ∑ †a a − †) . . . . . . . . . . . .(*)
y s
  sy qsy  s s q“y
=y ∑ s
− y ∑ s
+λ†t –λC +λ∑ †a a [after opening up brackets in (*) ]

 sy qsy 
= ∑ - ∑ Na a +λ†t –λC + λ∑ †a a (after simplifying)
y s y

˜™
For min, ˜ = 0 ⇒
s

s qsy s qs
- +λ †a = 0 ⇒ a = . . . . . . . . . . . . . . . . . . .(**)
y sy _š ›s

To obtain the value of λ, we multiply (**) by †a and sum wrt :

 ∑ s qs ›s
∑ †a a = =C - †t from which we get the value of λ:
√š _›s

s s∑  q _›
s s ∑  qs _›s
√=  ∑ ›  = (› ›œ )
. Going back to equation (**), we have:
s s

s qs s qs/_ž s qs /_›s ¢ 2 /_/2

a = = s
= ∑ Ÿ _ž
=∑ 2 (C-
(C- /0 ) (after replacing λ with the value obtained above
√š _›s √š  s s s ¢2 2 _/2
Ÿ(ž¡žœ )

and simplifying).

66 In a population
population with N=6, and L=2, the values of e2¥ are 0, 1, 2 in stratum 1 and 4, 6, 11 in stratum
2. A sample with n =4 is to be taken. Show that this sample can be distributed as follows: 3f =1 and
3 =3 (approximate values)

789:;<8=

N=6, L=2, n=4 Stratum 1:Wa§ =0, 1, 2, Stratum 2: Wa§ =4, 6, 11

©wsª ẁs « wsª

Now, a =∑
s qs
. But a =∑§X
s 
, whereẀa =∑§X
s  5 f=∑¬¥Xf e2¥ =0xfx = f;
⇒e
s qs s  s ¢f ¬

5  =∑¬¥Xf e¥=gx®xff =7
e
¢ ¬

(wsª ẁs )y (wbª ẁb )y (wbb ẁb )y x(wby ẁb )y x(wb¯ ẁb )y (t )y x( )y x( )y
a=∑ ⇒  =∑ = = =1 ⇒ =1 (Note
s  b     
that from above Ẁ =1)

(wyª ẁy )y (wyb ẁy )y x(wyy ẁy )y x (wy¯ ẁy )y (u °)y x(± °)y x( ) °y
Similarly,  =∑ = = =13 ⇒ =3.6
y    

s qs b qb (g)(¬)(f) 3¢  g(¬)(¬.®)

From a =∑ , we’ve  = ,viz., 3f =(¬)(f)x(¬)(¬.®) ≅ f,, 3 = =(¬)(f)x(¬)(¬.®) ≅ ¬
s qs b qb xy qy ¢f f x¢ 

Hence  = 1,  = 3 (& +$³#*&% ´&µ').

 C= 0 + ∑ 62 _32, where /0 and 62 are known members, show

7 If the cost function is of the form C=/
¶2 2 
var( 5 6 ) for total fixed cost, 32 must be proportional to [
that in order to minimize var(4 62
]¬ (raised to
 ¢
the power ) where ¶2 = 2 . Find 32 for a sample of 1 000 under the following conditions:
¬ ¢

Stratum ¶2 2 62
1 .4 4 1
2 .3 5 2
3 .2 6 4

789:;<8=

 qy
Var((Ẁ‰Š ) = ∑ Na (Na − a ) s ; C=†t + ∑ %a _a
y s

Using Langrages’ Multipliers, we’ve:

 qsy
F(a , )= ∑ Na (Na − a ) +λ(∑ %a _a +†t –C)
y s

¡b
˜™(a,š) sy qsy 
= + %a ay =0
˜s y sy 

¡b ¯
sy qsy sy qsy
=λ%a ay ⇒ =ay
y sy y šŠs

sy qsy y sy qsy y  ·sy qsy y ·sy qsy y s

⇒a =( )¯ =( )¯ =( )¯ ⇒a ∝ ( )¯ where ¸a =
šy Š s y šŠs š Šs Šs 

·sy qsy y  ¹
or a =d( )¯ , where d=( ) 
Šs š

Part ()

 ·sy qsy ¹  ·sy qsy ¹

From above, we have a = y ( )  . . . . (*) ⇒∑ a = y ∑( )  =n
š ¹¯ Šs š ¹¯ Šs

¹  ·sy qsy ¹
Thus  = ∑( )  Substituting this value in (*), we’ve:
 Šs

y
·sy qsy ¹  (·sy qsy /Šs ) ¹¯
a =( )  y»y y = y
Šs
∑( s s )y¹¯ ∑(·ys qsy /Šs ) ¹¯
¼s

·sy qsy ¹ ·y q y  ·yy qyy ¹ ·¯y q¯¯ ¹ (.±)(±) ¹ (.tz)(}) ¹ (.tu)(±) ¹

Clearly ∑( ) =( b b ) ¹ +( )  +( )  =( )  +( )  +( ) 
Šs Šb Šy Š¯   u

=1.87 +1.08 + .51=3.46

y
(·by qby ) ¹¯  ttt(. °)
 = .u±
= .u±
=541
 y
(·yy qyy ) ¹¯  ttt(.t )
 = = =312
.u± .u±

y
(·¯y q¯y ) ¹¯  ttt(.})
 = = =147
.u± .u±

88 Define proportional allocation and optimum allocation and obobtain

tain an expression for the variance
3
of a sample mean under these allocations. If 2 is negligible, compare the variance of estimates of the
¢ 2
population mean under:

() simple random sampling

() stratified sampling with proportional allocation,

() stratified sampling with optimum allocation.

789:;<8=:
s s
Proportional allocation:
allocation if the sample sizes in the strata are chosen such that you’ve =
 
=constant, then stratification is defined as stratification with proportional allocation of the a ,
(=1, 2, 3, . . . , L)

Optimum allocation: when the problem of allocating the sample sizes a is such that

() the variance of the estimated mean is minimal for fixed cost

() the sampling cost is minimal for a given precision’s sample estimate, then stratification is
defined as stratification with optimum allocation.

 qsy s 
Now var(ˆ`‰Š )= ∑ÀÁX Na (N-n) Using proportional allocation = s, we’ve:
y   

y
`  q  
Var(ˆ`‰Š )= y ∑ÀaX Na (Na − s ) s (since a = s)
  s 

 s qsy   qy 2 ¢2 ¢ 3

= ∑ÀaX(Na − ) = ∑ÀaX Na (1 − ) s =∑Ã2Xf ( )
      3¢ ¢

s qs
If we take ∑ a =n (fixed), then we’ve a =∑ . . . . . . . . . . . (*)
s qs

y y
  q 
Thus, var(ˆ`‰Š ) = y ∑ s s − ∑ Na a . . . . . . . . . . . . . . . (**)
 s y

Replacing a in (**) by (*), we get:

y y
  q   
var(ˆ`‰Š )= y ∑ s s .∑ Na a − ∑ Na a = ∑ Na a ∑ Na a − ∑ Na a
 s qs y y y
  
= ((∑ Na a ) − ∑ Na a
y y

which is the variance under optimum allocation.

Last part:
 
(1) for SRS, var(ˆ`)= 

   qy   qy qy qy qy
If is ignored, then var(ˆ` )=( ) =( − ) =(1 − 0) = ⇒var(ˆ` ) = ÅÆÇÈ = .
        
 qsy   qy
(2) For stratified RS, var(ˆ`)= y ∑ Na (Na − a ) = y ∑ Na(1- s ) s
 s  s s
y y
s   q
If s
is ignored, then var(ˆ`)=y ∑ s s . . . . . . . . . . . .(*)
s
s s s
(3) For proportional allocation, we’ve 
= 
, from which a = 
. Replacing a in (*) above by
∑ s qsy
this value, we’ve: ÅÆÉ = 
.
s qs
(4) For optimum allocation with ∑ a =  ,#³), we’ve a =∑ . Replacing a in (*) with this
s qs
value, we get

(∑ s qs )y
´ÉŠ =
y

(Êsª Ề)
(5) Again for stratified RS,   = ∑a À
§X ∑aX  
 
or (N − 1) =∑ÀaX ∑§X
s
(ˆa§ − ˆ`)=∑ÀaX ∑§X
s
(ˆa§ − ˆ`a + ˆ`a − ˆ`)
s
=∑ÀaX ∑§X (ˆa§ − ˆ`a ) + ∑ÀaX Na (ˆa − ˆ`)
Ÿ
s (Ê y
∑ªËb sª Ềs ) s
But a = ⇒∑§X (ˆa§ − ˆ`a ) =(Na − 1)a
s 
Thus, (N-1)  =∑ÀaX(Na − 1) a +∑ÀÁX Na (ˆa − ˆ`)
N  =∑ÀaX Na a + ∑ÀaX Na (ˆa − ˆ`)
q y ∑ s qsy ∑ s (Ês Ề)y
= +
  
∑ s (Ês Ề)y
Clearly: ÅÆÇÈ =ÅÆÉ +

⇒ ÅÆÇÈ ≥ ÅÆÉ
Now from the definition of optimum allocation, we must have ÅÆÉ ≥ ÅÉŠ, viz.,
ÅÆÉ − ÅÉŠ ≥ 0
∑ s qsy (∑ s qsy )y
or ÅÆÉ − ÅÉŠ = − ≥0
 y
 (∑ s qs )y
ÅÆÉ =ÅÉŠ + [∑ Na a − ]
 
∑ s (Ềs Ề) y  (∑ s qs )y
ÅÆÇÈ − =ÅÉŠ + [∑ Na a − ]
  
 (∑ s qs )y ∑ s (Ềs Ề)y
ÅÆÇÈ =ÅÉŠ + [∑ Na a − ]+
  
,+$* Íℎ#ℎ #% ,$µµ$Í %ℎ&% ÅÆÇÈ ≥ ÅÉŠ
 : ÅÆÇÈ ≥ ÅÆÉ ≥ ÅÉŠ.