Lecture #10 Varying Probability Sampling

Objectives of this Lecture:

After reading this unit, you should be able to
• understand about the PPS sampling without replacement.
• understand about the Des-Raj and Horvitz-Thompson estimator.

Cumulative Total Method

The Cumulative Total Method is a systematic approach used in probability proportional to size
(PPS) sampling, particularly when selecting samples where each unit’s probability of selection is
proportional to its size or importance. It involves calculating cumulative totals based on a measure
of size, then using random numbers to select units.

Let us consider a population of size N and we will take a PPS with replacement sample of size n.
N
th
Let the size of the i unit is denoted by X i and X =∑ X i .
i=1

The selection procedure consist of the following steps:

i. Write down the cumulative totals for the sizes X i ; i=1,2,3 , . . . , N
ii. Choose a random number r such that 1≤r≤ N / X .
iii. Select ith population unit if T i−1≤r≤T i , where,
T i−1= X 1+ X 2 +. .+ X i−1 and T i= X i +T i −1
The probability of selecting the ith unit using by this procedure is p i= X i / X . For selecting a
sample of n units with PPS with replacement the above operation is to be repeated n times.

Lahiri’s Method
Lahiri’s Method is a sampling technique used in Probability Proportional to Size (PPS)
Sampling. It ensures that larger units (based on a predefined size measure) have a higher
probability of being included in the sample. The method was developed by D. B. Lahiri and is
particularly suitable when there is no cumulative size list or when computational simplicity is
desired.

Sampling Technique-II | Varying Probability Sampling | Lecture #10 1

 Lectured by
STT366: Sampling Technique-II Md. Kaderi Kibria, STAT, HSTU

Lahiri’s suggested an alternative procedure in which cumulations are avoided completely. Lahiri’s
method involve the following steps for selecting a sample:
i. Select a pair of random number (i, j) such that 1≤i ≤N ,1≤ j≤M . where M is the
maximum size of the N units of the population.
ii. If j≤ X i then the ith unit is selected in the sample otherwise rejected and another pair of
random number is chosen.
iii. To get a sample of size n the procedure is repeated till n units are selected.

In PPS sampling, WR, an unbiased estimator of the population mean, Ȳ is given by

N
1
Ȳ^ pps = ∑ ( y i / pi )
Nn i
In PPS sampling, WR, an unbiased estimator of V ( Y^ PPS ) is given by
n n
V^ ( Ȳ^ pps )=v ( Ȳ^ pps )=
1
∑
n(n−1) i
( yi / N pi −Ȳ^ )2=
1
n(n−1)N 2 [∑
i
( y i / p i)2−n Y^ 2PPS ]
Unbiased estimator of the gain due to PPSWR sampling as compared to SRSWR is
n 2
1 yi
2∑
( N −1/ p i)
n i=1 p i

Example 10.1 A sample survey was conducted to study the yield of wheat in Haryana. A total of 50
farms were included in the study. The total area under wheat cultivation (X) was 484.5 hectares. The
area under cultivation (x) and yield (y) were recorded in hectares and quintals per hectare,
respectively.
Farm Area under crop (x) Yield (y) Farm Area under crop (x) Yield (y)
1 5.0 20.2 26 2.1 17.0
2 3.5 18.9 27 5.4 21.9
3 2.8 17.6 28 3.2 18.2
4 4.2 22.4 29 2.3 17.6
5 6.1 21.1 30 4.4 19.9
6 1.9 15.0 31 1.6 14.5
7 3.3 19.2 32 3.0 18.1
8 5.6 20.7 33 4.1 20.4
9 2.5 17.3 34 5.9 22.7
10 4.0 20.9 35 2.7 17.5
11 2.2 16.5 36 3.4 19.0
12 6.4 23.0 37 4.9 21.8
13 3.1 18.7 38 6.3 23.2
14 4.6 21.3 39 2.9 17.9
15 3.9 19.8 40 1.7 15.1

Sampling Technique-II | Varying Probability Sampling | Lecture #10 2

 Lectured by
STT366: Sampling Technique-II Md. Kaderi Kibria, STAT, HSTU

Farm Area under crop (x) Yield (y) Farm Area under crop (x) Yield (y)
16 5.3 20.5 41 3.8 20.1
17 2.0 16.9 42 5.1 21.6
18 1.8 15.8 43 4.3 19.5
19 3.6 18.4 44 2.0 16.3
20 2.4 17.2 45 6.0 22.9
21 5.7 22.0 46 3.9 18.8
22 6.2 21.5 47 4.5 20.0
23 2.6 16.8 48 2.8 17.4
24 4.8 20.2 49 5.5 21.2
25 3.7 19.4 50 4.7 19.7

a) Draw a sample of 20 farms from a total of 50 using the cumulative total and Lahiri’s

method.

b) Estimate the average yield per farm along with its standard error for these two samples.

c) Estimate the gain in efficiency due to PPS sampling compared to simple random sampling

with replacement.

Example 10.2 The Department of Agricultural Statistics aims to estimate the average wheat yield

(in quintals per hectare) in a region consisting of 18 villages. Due to resource constraints, a sample

of 2 villages is selected using Probability Proportional to Size (PPS) without replacement. The area

under wheat cultivation in hectares is used as the size measure for PPS selection.

Village ID Wheat Yield (yᵢ) (q/ha) Area under Wheat (xᵢ in ha) Selection Probability (pᵢ)
1 22 20 0.020
2 25 25 0.025
3 28 35 0.035
4 30 40 0.040
5 33 50 0.050
6 35 60 0.060
7 37 65 0.065
8 40 70 0.070
9 42 75 0.075
10 44 80 0.080
11 46 85 0.085
12 48 90 0.090

Sampling Technique-II | Varying Probability Sampling | Lecture #10 3

 Lectured by
STT366: Sampling Technique-II Md. Kaderi Kibria, STAT, HSTU

Village ID Wheat Yield (yᵢ) (q/ha) Area under Wheat (xᵢ in ha) Selection Probability (pᵢ)
13 50 95 0.095
14 52 100 0.100
15 55 105 0.105
16 58 110 0.110
17 60 115 0.115
18 62 120 0.120

a) Draw a sample of 2 villages without replacement using the following PPS sampling

methods:

• Sen-Midzuno Method

• Narain’s Scheme

• Brewer’s Selection Procedure

b) Using the selected sample, estimate the population mean yield (quintals/hectare) and 95%

CI using the following estimators:

• Des Raj estimator

• Horvitz-Thompson estimator

Reference Book:

i. Singh D. and Choudhary F.S. (1986), Theory and analysis of sample survey designs, New
Age International (P) Ltd. Publishers, New Delhi.

ii. Islam, M. N. (2009), An Introduction to Sampling Methods, 3rd Ed. Book World, Dhaka.

<><><><><><><><> End <><><><><><><><><>

Sampling Technique-II | Varying Probability Sampling | Lecture #10 4