Double Sampling for stratification

Experiment no: 01: From the given data on two variables x and y in double sampling for stratification-
(i) Draw a SRS of variable x of size 30.
(ii) Divide the data into 4 strata of equal size and allocate the sample into the strata.
(iii) For each stratum select a random sub-sample of size 4 from the 1st sample and consider the
corresponding values of main variable y in the 2nd sample.
(iv) Calculate an unbiased estimate of population mean using-
(a) Weight estimated from 1st sample;
(b) Actual weight;
(v) Estimate corresponding two variances from sample and comment.
(vi) Assuming c  1.0, ch  2.0, c   10.0 , estimate v h and variance minimum Vmin  .
Village x y Village x y
1 159 81 31 716 305
2 523 292 32 230 277
3 569 275 33 349 245
4 411 118 34 379 270
5 240 44 35 135 64
6 121 56 36 873 445
7 106 33 37 248 69
8 397 147 38 1034 401
9 453 194 39 503 164
10 710 282 40 206 46
11 176 65 41 682 166
12 730 288 42 653 129
13 627 115 43 166 44
14 280 161 44 445 166
15 287 179 45 1495 164
16 404 273 46 501 96
17 124 58 47 1473 373
18 44 58 48 1117 261
19 370 98 49 389 79
20 541 155 50 716 191
21 153 41 51 276 51
22 658 230 52 570 106
23 805 239 53 251 56
24 435 158 54 691 83
25 418 130 55 147 13
26 235 114 56 864 146
27 364 170 57 808 242
28 515 272 58 924 340
29 586 418 59 601 189
30 246 97 60 396 90
Solution: (i) From the given data we draw an initial sample of size 30 for x variable on SRS basis using random
number table. Since the population size N  60 , which is of two digits, we draw 30 random numbers from
random number table each of which is two digits.

Serial Random Selected Value of the Serial Random Selected Value of the
no. number unit no. selected no. number unit no. selected
unit  xi  unit  xi 

1 02 02 523 16 05 05 240

2 11 11 176 17 46 46 501

3 40 40 206 18 24 24 435

4 15 15 287 19 23 23 805

5 50 50 716 20 41 41 682

6 06 06 121 21 49 49 379

7 27 27 364 22 58 58 924

8 55 55 147 23 29 29 586

9 22 22 658 24 39 39 503

10 20 20 541 25 45 45 1495

11 19 19 370 26 36 36 873

12 51 51 274 27 14 14 280

13 60 60 396 28 43 43 160

14 01 01 159 29 33 33 349

15 47 47 1473 30 12 12 730
(ii) We divide the data into 4 strata of equal size and allocate the 1st sample into strata.
Stratum Village xi yi n h Stratum Village xi yi n h
1 159 81 (1) 31
2 523 292 (2) 32
3 33 349 245 (1)
4 34
5 240 44 (3) 35
6 121 56 (4) 36 873 445 (2)
7 37
8 38
1 9 8 3 39 503 164 (3) 7
10 40 206 46 (4)
11 176 65 (5) 41 682 166 (5)
12 730 288 (6) 42
13 43 166 44 (6)
14 280 161 (7) 44
15 287 179 (8) 45 495 164 (7)

16 46 501 96 (1)
17 47 1473 373 (2)
18 48
19 370 98 (1) 49 389 79 (3)
20 541 155 (2) 50 716 191 (4)
21 51 276 51 (5)
22 658 230 (3) 52
23 805 239 (4) 53
2 24 435 158 (5) 7 4 54 8
25 55 147 13 (6)
26 56
27 364 170 (6) 57
28 58 924 340 (7)
29 586 418 (7) 59
30 60 396 90 (8)

(iii) For each stratum we select a sub-sample of sample of size 4 from 1st sample on SRS basis and consider the
corresponding values on the y variable in the 2nd sample.

Now, we draw 4 random numbers each of which is one digit from random number table.]

Stratum – 1
Random Selected unit Serial Value of the y variable
number no. no. corresponding to selected
unit no.
5 5 (5) 65
6 6 (6) 288
8 8 (8) 179
4 4 (4) 56
We draw 4 random numbers each of which is one digit from random number table.

Stratum – 2
Random Selected unit Serial Value of the y variable
number no. no. corresponding to selected
unit no.
7 7 (7) 418
5 5 (5) 158
6 6 (6) 170
3 3 (3) 230

We draw 4 random numbers each of which is one digit from random number table.

Stratum – 3
Random Selected unit Serial Value of the y variable
number no. no. corresponding to selected
unit no.
6 6 (6) 44
3 3 (3) 164
2 2 (2) 445
1 1 (1) 245

We draw 4 random numbers each of which is one digit from random number table.

Stratum – 4
Random Selected unit Serial Value of the y variable
number no. no. corresponding to selected
unit no.
3 3 (3) 79
5 5 (5) 51
6 6 (6) 13
4 4 (4) 191
 (iv)
Table for necessary calculation:

N h n h Elements y ij   y n n  y  yh 
2
wh y h N h y
Stratum

yh 
hi
wh  h  h  h s h2  hi

n nh
h
n nh  1 N

1 15 8 65 288 179 56 147 0.26666 0.50 11970 39.20 36.75

2 15 7 418 158 170 230 244 0.23333 0.57 14448 56.93 61.00
3 15 7 44 164 445 245 224.5 0.23333 0.57 28427 52.38 56.13
4 15 8 79 51 13 191 83.5 0.26666 0.50 5867.6667 22.27 20.88
30 60712.6667 170.78 174.76

Here
4
n    nh  30,
h 1

nh  n1  n2  n3  n4  4

(a) An unbiased estimate of population mean by using weight estimate from 1st sample is given by,
4
y st   wh y h  170 .78
h 1
(b) An unbiased estimate of population mean by using actual weight is given by,
4
Nh
y st   y h  174 .76
h 1 N

(v) The estimated variance of (a) is given by,

nN  1  2 1 1  g w 1  g 
v y st    wh sh      sh2  h     wh  yh  y st 2 
n  1N   n h N  n  N n h  n 

N  n 60  30
Now, g     0.50847
N 1 60  1
Table for necessary calculation:

1 1  1 1 w 1  wh  yh  yst 
2
wh s h2  wh s h2    sh2  h  
n h N  n  h N   N n h 
3191.9202 0.05 159.5999601 -744.80133 150.79653
3371.19584 0.0418128 140.9595512 -898.988272 1250.9391
6632.957191 .0418128 277.3433035 -1768.79427 673.36133
1564.529422 0.05 78.22664711 -365.0999133 2031.4078
14760.602653 656.12946191 -3777.6837853 4106.50476

30  60  1  
 v y st  
0.508474576 0.508474576
 656 .12946191   3777 .6837853   4106 .50476 
30  1  60  30 30 
 1.017241379  656 .12946191  64.02853869  69.60177556   673 .0881839
The estimated variance of (b) is given by,
nh 4
fh   ; nh  4; N h  15; N  60
N h2  1  f h  2 N h 15
v y st    2  sh
N  nh 
N12  1  f1  2 N 22  1  f 2  2 N 32  1  f 3  2 N 42  1  f 4  2
 2   s1  2   s2  2   sh  2   s4
N  n1  N  n2  N  n3  N  nh 
15 2  1  0.26667  15 2  1  0.26667 
 2  11970  2   14448
60  4  60  4 
15 2  1  0.26667  15 2  1  0.26667 
 2   28427  2    5867 .6667
60  4  60  4 
 137 .1556266  165 .5492475  325 .7245611  67.23337533  695 .6628105
Comment: Since the actual weighted variance is greater than the weighted estimated variance, so weighted
estimated variance is more efficient.

(vi) An estimated h is given by,

1 c  1.0, ch  2.0,
 c   2
ˆh  sh   2 1  w s 2
 
 ch  s   wh sh
 
h h
2
 

Where, s
2

  y hi  yh 
2 4
  s h2  60712 .6667
nh  1 h 1
1
 1.0  1  2
ˆ1  109 .4075    
 2.0  60712 .6667  14760 .602653 
 109 .4075  0.003298621  0.3608939936
1
 1.0  1  2
ˆ2  120 .19983    
 2.0  60712 .6667  14760 .602653 
 120 .19983  0.003298621  0.396493695
1
 1.0  1  2
ˆ3  168.60308    
 2.0  60712 .6667  14760 .602653 
 168 .60308  0.003298621  0.0.556157674
1
 1.0  1  2
ˆ4  76.60069677    
 2.0  60712 .6667  14760 .602653 
 76.60069  0.003298621  0.0.252676666
 2
1  
 wh sh ch  s   wh sh  c   N
1
s2
 v min  y st   2 2 2

c  
 
2
1 
  165 .4450676  60712 .6667   2  1.0  
1

10   60

  165 .4450676  214 .3643255   246 .0100442

1 2

10
 14425 .51751  246 .0100442
 14179 .50746
Experiment no-02: The data are given below:
S.N x y S.N x y
1 78 80 21 50 64
2 138 143 22 40 58
3 67 67 23 77 89
4 29 50 24 64 63
5 381 464 25 64 77
6 23 48 26 56 142
7 37 63 27 40 60
8 120 115 28 40 64
9 61 69 29 38 52
10 387 459 30 136 139
11 93 104 31 116 130
12 172 183 32 46 53
13 78 106 33 283 291
14 66 86 34 87 105
15 60 57 35 30 111
16 46 65 36 71 79
17 02 50 37 256 288
18 507 634 38 43 61
19 179 260 39 25 57
20 121 113 40 94 85
(i) Draw a first sample of size 15 for auxiliary variable (x) and draw a sub-sample of size 8 from
main variable (y)
(ii) Calculate ratio and regression estimates of population mean and total and their approximate
variances.
(iii) Compute estimates of these variances from sample and comment.
(iv) How do the precision of ratio an regression estimate with that of sample of y?
n
(v) For regression estimate find the ratio of sample sizes if the ratio of the corresponding cost is
n
Tk. 5.
Solution: (i) From the given data we draw an initial sample of size 15 n by SRS method. Since population
size is N  40 which is of two digits. So, we draw 15 random numbers each of which is two digits from
random number table.
S.N Random number Selected unit No. Selected initial unit  xi  y i
1 48 08 120 115
2 16 16 46 65
3 73 33 283 291
4 47 07 37 63
5 70 30 136 139
6 05 05 381 464
7 44 04 29 50
8 14 14 66 86
9 12 12 172 183
10 13 13 78 106
11 01 01 76 80
12 21 21 50 64
13 03 03 67 67
14 29 29 38 52
15 39 39 25 57
1604
 Here, x  
x i

1606
 107 .07, n  size of 1st sample=15.
n 15

= Mean of the initial sample.

Now from 15 selected initial samples we draw a sub-sample of size 8 (n) by SRS method. Since the size of the
initial sample is 15 which is of two digits, so , we draw 15 random numbers of two digits from number table.

Random Selected unit Selected initial

Number serial no. sample no. xi yi
48 3 33 283 291
16 1 08 120 115
73 13 03 67 67
47 2 16 46 65
70 10 13 78 106
24 9 12 172 183
65 5 30 136 139
44 14 29 38 52
 xi  940  yi  1018

x 
x i

940
 117.5 Here, n  8
n 8 = Size of sub-sample

y
 yi 
1018
 127.25
n 8

xi 2 940 2
SS x    x  2
i  156702 
 46252
n 8
 yi 2 1018 2
SS  y    yi 
2
 173370   43829 .5
n 8

SPxy    xi y i 
 xi  yi  164256  940  1018  44641
n 8

s x2 
 xi  x   6607 .43,
2

sy 
2   yi  y 2
 6261 .36
n 1 n 1

(ii) Ratio estimate of population mean is given by,

y 127.25 Where, x   107.07
yR  x    107.07  115.95 = mean of 1st sample
x 117.5 y
r
x
And Ratio estimate of population total is given by, = ratio estimate of
YˆR  Ny R  40  115.95  4698 population ratio
x, y  mean of sub-sample of x
and y respectively.
Regression estimate of population mean is given by, Where, b = the least square reg.
y lr  y  bx   x  Coeff. y on x of 2nd sample
SPxy  44641
 127 .25  0.97107 .07  117 .5    0.97
SS x  46252
 117 .133
And Regression estimate of population total is given by,
YR  Nylr  40  117.133  4685.32

An approximate variance of ratio estimate is given by,

S y2  R 2 S x2  2 RS yx 2 RS yx  R 2 S x2 S y2
V  yR    
n n  N
Where, n  initial sample size= 15, N = 40
n = sub-sample size = 8
N N

 y i  y
2
 x i  x
2

 S y2  i 1
 16604 .45, S x2  i 1
 12094 .92
N 1 N 1
Y  132 .1, X  107 .475

Y 132 .1
Population ratio, R    1.23
X 107 .475
SPxy   XY  NXY 1109703  40  107 .475  132 .1
S yx     13892 .44
N 1 N 1 40  1

16604 .45  1.23 2  12094 .92  2  1.23  13892 .44

V yR  
8
2  1.23  13892 .44  1.23  12094 .92 16604 .45
2
 
15 40
 1564 .5

The variance of regression estimate is given by,

V  y lr  

S y2 1   2

 S 2
y 2 S y2
n n N
16604 .45  1  0.961 16604 .45  0.961 16604 .45
  
8 15 40
 729 .63
(iii) The estimated variance of ratio estimate of population mean is given by,
s y2  r  2 s x2  2r s yx 2r s yx  r  2 s x2 s y2
v y R    
n n N
y 127.25
Where, r  = sample ratio =   1.083
x 117.5
SPxy  44641
And, s xy    6377 .29
n 1 7
6261 .36  1.083 2  6607 .73  2  1.083  6377 .29
 v y R  
8
2  1.083  6377 .29  1.083  6607 .73 6261 .36
2
   272 .457
15 40

An estimated variance of regression estimate of population mean is given by,

v ylr  

s y2 1  r 2  s r
2 2

r  0.99
y
Here,
n n
6261 .36  1  0.99  6261 .36  0.99 2
   416 .94
8 15
Since V  ylr   V  y R  , so we can conclude that regression method for estimating population mean gives
better estimate than that of ratio method.

(iv) We compare the precision of ratio and regression estimate with sample estimate of y alone.
If we take simple sample of y then estimated variance is given by,
1 f 2
v y   sy n 8
n Here, f    0.2
1  0.2 N 40
  6261 .36  626 .136
8
v y R  272.457
  0.435  1
v y  626.136
For ratio method,

v ylr  260.41
  0.416  1
v y  626.136
For regression method,

Precision of ratio and regression method is greater than that of simple method of y i.e. ratio and regression
methods are more precise than simple method of y.

(v) Ratio of sample size for regression estimate is,

1 Where, c  cost of unit in sub-sample
n  c 1    2 2 c  cost of unit in initial sample
 
n  c  2  , Given,
c c 1
 5, 
1 c c 5
 1 1  0.961  2 and ,   0.98,   0.961
2
   0.09009187
5 0.961 