SAMPLING DESIGN & CONSIDERATIONS

Learning Objectives
Upon completion of this chapter, you will be able to:
Understand the importance of sampling Differentiate between random and non-random sampling

Understand the concept of sampling and non-sampling errors

Understand the concept of sampling distribution and the application of central limit theorem Understand sampling distribution of sample proportion

Sampling
A researcher generally takes a small portion of the population
for study, which is referred to as sample. The process of selecting a sample from the population is called sampling.

Sampling Concepts

Population: Population refers to any group of people or objects that form the subject of study in a particular survey and are similar in one or more ways. Element: An element comprises a single member of the population. Sampling frame: Sampling frame comprises all the elements of a population with proper identification that is available to us for selection at any stage of sampling. Sample: It is a subset of the population. It comprises only some elements of the population. Sampling unit: A sampling unit is a single member of the sample. Sampling: It is a process of selecting an adequate number of elements from the population so that the study of the sample will not only help in understanding the characteristics of the population but will also enable us to generalize the results. Census (or complete enumeration): An examination of each and every element of the population is called census or complete enumeration.

Why Is Sampling Essential?

Sampling saves time. Sampling saves money. The study of a sample instead of complete enumeration may, at times, produce more reliable results. Sampling broadens the scope of the study in light of the scarcity of resources. It has been noticed that sampling provides more accurate results, as compared to census because in sampling, nonsampling errors can be controlled more easily. In most cases complete census is not possible and, hence, sampling is the only option left. A census is appropriate when the population size is small.

Figure 5.1: Steps in the sampling design process

The Sampling Design Process

Step 1: Target population must be defined
Target population is the collection of the objects which possess the information required by the researcher and about which an inference is to be made. A researcher takes a sample from a population list, directory, map, city directory, or any other source used to represent the population. This list possesses the information about the subjects and is called the sampling frame. Sampling is carried out from the sampling frame and not from the target population.

Step 2: Sampling frame must be determined

The Sampling Design Process (Contd.)

Step 3: Appropriate sampling technique must be selected

In sampling with replacement, an element is selected from the frame, required information is obtained, and then the element is placed back in the frame. This way, there is a possibility of the element being selected again in the sample. As compared to this, in sampling without replacement, an element is selected from the frame and not replaced in the frame. This way, the possibility of further inclusion of the element in the sample is eliminated. Sample size refers to the number of elements to be included in the study.

Step 4: Sample size must be determined

Step 5: Sampling process must be executed

Random Versus Non-random Sampling

In random sampling, each unit of the population has the same probability (chance) of being selected as part of the sample. In non-random sampling, members of the sample are not selected by chance. Some other factors like familiarity of the researcher with the subject, convenience, etc. are the basis of selection

Figure 5.2: Random and non-random sampling methods

Random Sampling Methods

Simple Random Sampling

In simple random sampling, each member of the population has an equal chance of being included in the sample. In stratified random sampling, elements in the population are divided into homogeneous groups called strata. Then, researchers use the simple random sampling method to select a sample from each of the strata. Each group is called stratum. In stratified random sampling, stratum should be relatively homogenous and the strata should contrast with each other.

Stratified Random Sampling

Random Sampling Methods (Contd.)

In cases where the percentage of sample taken from each stratum is proportionate to the actual percentage of the stratum within the whole population, stratified sampling is termed as proportionate stratified sampling. In cases where the sample taken from each stratum is disproportionate to the actual percentage of the stratum within the whole population, disproportionate stratified random sampling occurs.

Figure 5.5: Stratified random sampling based on educational levels

Random Sampling Methods (Contd.)

Cluster (or Area) Sampling

In cluster sampling, we divide the population into non-overlapping areas or clusters. In stratified sampling, strata happen to be homogenous but in cluster sampling, clusters are internally heterogeneous. A cluster contains a wide range of elements and is a good representative of the population.

Figure 5.6: Diagram for cluster sampling

Systematic (or Quasi-random) Sampling

In systematic sampling, sample elements are selected from the population at uniform intervals in terms of time, order, or space. A researcher wants to take a sample of size 30 from a population of size 900 and he has decided to use systematic sampling for this purpose.
For obtaining the sample, the first member can be selected randomly and after that every 30th member of the population is included in the sample. Suppose the first element 3 is selected randomly and after this, every 30th element, that is, 33rd, 63rd, element up to a sample size of 30 are included in the sample.

Multi-Stage Sampling

As the name indicates, multistage sampling involves the selection of units in more than one stage.

Figure 5.7: Multi-stage (four stages) sampling

Non-Random Sampling
Sampling techniques where selection of the sampling units is not based on a random selection process are called nonrandom sampling techniques.
Quota Sampling
In quota sampling, certain subclasses, such as age, gender, income group, and education level are used as strata. Stratified random sampling is based on the concept of randomly selecting units from the stratum. However, in case of quota sampling, a researcher uses nonrandom sampling methods to gather data from one stratum until the required quota fixed by the researcher is fulfilled.
In convenience sampling, sample elements are selected based on the convenience of a researcher.

Convenience Sampling

Non-Random Sampling (Contd.)

Judgement Sampling

In judgement sampling, selection of the sampling units is based on the judgement of a researcher. In snowball sampling, survey respondents are selected on the basis of referrals from other survey respondents. A sampling procedure in which initial respondents are selected by probability methods and additional respondents are obtained from information provided by the initial respondents. This technique is used to locate members of rare population by referrals.

Snowball Sampling

Sampling and Non-Sampling Errors

Sampling Error: This error arises when a sample is not representative of the population.
Sampling errors can occur due to some specific reasons: Faulty selection of the sample. Sometimes due to the difficulty in selection a particular sampling unit, researchers try to substitute that sampling unit with another sampling unit which is easy to be surveyed. Sometimes researchers demarcate sampling units wrongly and hence, provide scope for committing sampling errors.

Sampling and Non-sampling Errors (Contd.)

Non-Sampling Errors All errors other than sampling can be included in the category of nonsampling errors. The following are some common non-sampling errors:

Plain lying by the respondent. the spreadsheet on the computer.

The error can arise while transferring the data from the questionnaire to

There can be errors at the time of coding, tabulation and computation. Population of the study is not properly defined Respondent may refuse to be part of the study. There may be a sampling frame error.

Determination of Sample Size

The size of the population does not influence the size of the sample Methods of determining the sample size in practice:
Researchers may arbitrary decide the size of sample

without giving any explicit consideration to the accuracy of the sample results or the cost of sampling.
The total budget for the field survey in a project proposal is

allocated.
Researchers may decide on the sample size based on

what was done by the other researchers in similar studies.

Determination of Sample Size

Confidence interval approach for determining the size of the sample The following points are taken into account for determining the sample size in this approach.
The variability of the population: Higher the variability as

measured by the population standard deviation, larger will be the size of the sample.
The confidence attached to the estimate: Higher the confidence

the researcher wants for the estimate, larger will be sample size.
The allowable error or margin of error: Greater the precision the

research seeks, larger would be the size of the sample.

Determination of Sample Size

Sample size for estimating population mean The formula for determining sample size is given as:
Where

n = Sample size = Population standard deviation e = Margin of error Z = The value for the given confidence interval

Example 1

An economist is interested in estimating the average monthly household expenditure on food items by the households of a town. Based on past data, it is estimated that the standard deviation of the population on the monthly expenditure on food item is Rs.30. With allowable error set at Rs.7, estimate the sample size required at a 90 per cent confidence interval. (z=1.645)
Example 2

It is desired to estimate the mean life time of a certain kind of vacuum cleaner. Given that the population std. dev. is 320 days, how large a sample is needed to be able to assert with a confidence level of 96 per cent that the mean of the sample will differ from the population mean by less than 45 days? (z=2.055)

Case let 1: Systematic Sampling

You have a population of approximately 1500 patients and wish to find out their attitudes to a new voucher scheme. There is insufficient time & money to collect data from all of them using a questionnaire and so you decide to send the questionnaire to a sample. Your calculation of sample size reveals that to obtain acceptable levels of confidence & accuracy you need an actual sample size of approximately 300 patients to whom you will send the questionnaire. You decide to select them using systematic sampling. First you need to work out the sampling fraction: 300/1500=1/5 Sampling fraction of 1/5 means you need to select every 5th patient from the sampling frame.

Cont

You use a random number to decide where to start on the sampling frame. As your sampling fraction is1/5, the starting point must be one of the first five patients. You therefore select a one-digit random number between 1 to 5. Once you have selected your first patient at random you select every fifth patient until you have gone right through your sampling frame. If the random number you selected was3, then you would select following patient numbers. 3 8 13 18 23 28 33 38 and so on until 300 patients had been selected.

Case let 2: Cluster Sampling

Ceri needed to select a sample of firms to undertake an interview based survey about the use of photocopiers. As she had limited resources with which to pay for travel and other associated data collection costs she decide to interview firms in 4 geographical areas selected from a cluster grouping of local administrative areas.

A list of all local administrative areas formed her sampling frame. Each of the local administrative areas (clusters) was given a unique number, the first being 1 and so on.
The four sample clusters were selected from this sampling frame of local administrative areas using simple random sampling. Ceris sample was all firms within the selected clusters. She decided that the appropriate telephone directories would provide a suitable list of all firms in each cluster.

Case let 3: Multi stage Sampling

A market research organisation needs you to interview a sample of 400 households in England & Wales. The electoral register provides a possible sampling frame. Selecting 400 households using either systematic or random sampling would probably result in these 400 households being dispersed throughout England & Wales. The time and cost of travelling to and interviewing your sample would be enormous. By using multi-stage sampling these problems can be overcome. In the first stage the geographical area (England & Wales) is split into discrete sub areas (countries). These form the sampling frame. After numbering, a small no. of countries are selected using simple random sampling. Since each case (household) is located in a country each has an equal chance of being selected for the final sample.

Cont

As the countries selected are still too large the selected countries are subdivided into smaller geographically discrete areas (electoral wards), which form the next sampling frame (Stage 2). Another random sample is selected. A larger no. of wards are selected to allow for likely important variations in households between wards.
A sampling frame is generated for each ward using a combination of the electoral register and UK Royal Mails postcode address file. The cases (households) that will be interviewed are then selected using either random or systematic technique.

Example 1
The distribution of the annual earnings of the employees of a cement factory is normally distributed. This distribution has a mean of Rs 25,000 and standard deviation of Rs 3000. If a researcher draws a random sample of size 50, what is the probability that their average earnings will be more than Rs 26,000?

Example 1 (Contd.)

Figure 5.10: Probability that the average earnings of employees is more than Rs 26,000

Figure 5.11: Corresponding z scores for probability of average earnings more than Rs 26,000

END OF CHAPTER