Understanding Survey Sampling

Please download to get full document.

View again

of 7
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Information Report
Category:

Documents

Published:

Views: 8 | Pages: 7

Extension: PDF | Download: 0

Share
Related documents
Description
Drawing up a sample is an essential step in undertaking most types of survey research. Information gathered from a sample can be used to draw inferences about the whole population being studied, and the more representative a sample is, the more accurate inferences based on it are likely to be. Part of Oxfam’s Research Guidelines series, this guideline provides a concise and easy-to-understand introduction to survey sampling. It explains the key steps in the sampling process, and describes the different kinds of samples and their various advantages and disadvantages. The guideline also outlines the variables that typically determine sample size, and provides links to further resources for readers who want to learn more.
Transcript
    www.oxfam.org.uk/policyandpractice   UNDERSTANDING SURVEY SAMPLING W HY MASTER THIS SKILL ? In social research, sampling is the process of selecting a group of subjects from the larger population that is being studied. You can make a sample from households, organizations, a value chain or any other unit of analysis. Drawing up a sample is an essential step in undertaking most types of survey (Oxfam’s guideline on Planning Survey Research outlines other key steps). Information gathered from a sample can be used to draw inferences about the whole population, a relationship schematized in the diagram below. The more representative a sample is, the more accurate inferences based on it are likely to be. If you do not have a sample and simply collect data from anyone, you will not be able to explain what the data refers to. Sampling is different from the procedure used in administering censuses, when the whole population is surveyed. Unless the population being studied is very small, sampling is much quicker and cheaper, and generally the only feasible way to undertake a survey and generate credible results about the population as a whole. While survey researchers should know how to draw up different kinds of samples, understanding this process is also important for others involved in commissioning surveys and interpreting the results. This guideline therefore provides an introductory overview of sampling and different kinds of samples.    R   E   S   E   A   R   C   H    G   U   I   D   E   L   I   N   E   S    KEY TERMS USED IN SAMPLING   This guide uses key terms from sampling theory and practice to ensure clarity in communication and alignment with the state of the knowledge about sampling. ã   Sampling units : the type of entity on which data are sought (individuals, households, schools, etc.). ã   Population  (universe): the full set of units of analysis about which you want to infer conclusions. ã   Sampling frame : a list of all units of analysis in a given population. ã   Census : the collection of data from all units of analysis in a given population. ã   Sample:  a subset of units drawn from the sampling frame or sometimes directly from the population. ã   Sampling error:  mistakes in estimation that reflect the difference between a sample and the population from which it is drawn. KEY STEPS IN THE SAMPLING PROCESS   The sampling process can be divided into three key steps: definition of the population of interest, identification of the sampling frame, and selection of the sample. 1. Determine the population of interest The first stage is to identify the population of interest for research. This depends on the research question(s) the survey is trying to answer. In some cases the definition of the population is obvious, for example when the aim is to collect descriptive statistics about people in a particular geographic area (with defined boundaries). In other cases the population of interest may be harder to define. For example when evaluating impacts of development projects, it is important to sample both project participants (referred to as the ‘intervention group’) and non-project participants (the ‘comparison group’) whose selection may depend on a range of factors. Time spent carefully defining the population of interest is seldom wasted time, and it often helps to raise questions and ambiguities that become relevant later. 2. Identify the sampling frame This entails finding the most comprehensive list that can be obtained of the sampling units (usually households or individuals) in the population of interest. A list of all the households or residents in a location, for example, can sometimes be obtained from local authorities or a previous census. This must be as accurate and complete as possible; in particular you should make sure that portions of the population with particular characteristics are not systematically excluded. Where it is not possible to obtain a full and complete list of sampling units, innovative techniques are now available that use satellite images in order to map settlements on the ground and so construct sample frames (Shannon et al. 2012). 3. Select the sample This entails selecting the units for data collection. The sample can be selected from the sampling frame using either probability or non-probability sampling techniques:    ã   Probability samples , sometimes called random samples , are those in which every unit in the population has the possibility (a ‘nonzero’ probability) of being selected. It is not necessary for each element to have the same chance of selection, but it must at least have some chance, and this chance needs to be known. ã   Non-probability samples , including quota and accidental samples (defined below), are not representative of the whole population. These sampling strategies can be used for testing questionnaires or for ad hoc research that does not require making inferences or generalizing estimates to the entire population. Only probability samples allow conclusions to be drawn about the whole population, test for statistical significance and compute for confidence intervals, and preclude the possibility of consciously or unconsciously introducing biases in the sample. They are therefore preferable to non-probability samples. DIFFERENT KINDS OF SAMPLES   1. Probability samples The following are the most common kinds of probability sampling: Simple random sampling In simple random sampling (SRS) the units to be surveyed are randomly selected from the sampling frame. This means that each unit (individual, household, etc.) will have the same probability of selection. If the sample frame comprises a list of all the households in a location, then SRS will entail picking a random selection of these households for data collection. SRS is relatively easy and straightforward to implement. It can, however, have some limitations if the researcher is investigating questions specific to particular subgroups of the population. 1  Stratified sampling (discussed next) addresses this weakness of SRS. Stratified random sampling Stratified random sampling (SS) is an approach that makes it easier to compare and contrast different recognized subgroups in a population. Examples of such subgroups are different socio-economic classes or castes, religious denominations and genders. SS entails first dividing the population in the sampling frame into distinct categories called ‘strata’ based on the subpopulations of interest. Each stratum is then sampled as an independent subpopulation in which individuals or other survey units can be randomly selected. For example, suppose you are interested in studying a population of 10,000 which is divided into one group of 8,000 people (A) and a second group of 2,000 (B), while you have resources sufficient for a sample of 100, and your priority objective is to generalize results to all 10,000 people in the population. For a proportional stratified random sample , you would randomly draw 80 people from  A and then 20 from B to guarantee the proportion of people in each stratum is the same as in the population. Note that sampling 100 people using simple random sampling might yield a sample with 88 from A and only 12 from B, or perhaps 75 from A and 25 from B. Proportional stratified random 1  The concern is that with smaller subgroups there might be some random error which can increase the standard errors, reducing power for comparison.    sampling ensures that will not happen. On the other hand, if your primary purpose was to test whether group B has different levels of a certain outcome compared with A, then it would be best to sample 50 observations from A and 50 observations from B. This would increase the precision of inferences made about group B, as the sample would be much larger. This is called a disproportional stratified random sample . SS can increase the precision of inferences made to the full population or increase inferences made to comparison among strata, but it cannot do both at the same time. Increasing precision of inferences to the full population is done by keeping the ratio of sample size and population size equal to all subpopulations (proportional stratified random sample). Increasing precision of inferences to comparisons among strata is done by taking approximately equal-sized samples for each group (disproportional stratified random sample). Multi-stage sampling For practical purposes it is sometimes more cost-effective to choose the sample in a process of two or more successive stages, which is called multi-stage sampling. Primary sampling units  (PSU) are units selected at the first stage, secondary sampling units  (SSU) are those selected at the second stage, and so on. The most common multi-stage sampling strategy is two-  (or three-) stage cluster sampling .  A cluster consists of the units in well-defined geographic areas (such as a village, city block, school, etc.). The two-stage cluster involves selecting the clusters in the first stage. The clusters are usually selected using a probability proportional to size  (PPS) method, where larger clusters have a greater probability of being selected for the sample. Within each selected cluster a list of all SSU is then drawn up. In the second stage a fixed (or proportional) number of SSU is then selected from each selected cluster using simple random sampling. Using two-stage cluster sampling has the advantage of reducing the cost of travel and survey administration, and it guarantees a representative sample of the target population when there is a list of all clusters to be sampled but not a list of all target individuals within each cluster. However, this sampling approach is generally not advisable if there are only a small number of clusters in the population, or only a small number of units within many of the clusters. 2. Non-probability samples There are three common non-probability sampling techniques. Convenience samples  A convenience sample is a non-probabilistic technique that involves selecting a sample on the basis of its accessibility or other factors of convenience. As a result some categories of people are likely to be overrepresented and others underrepresented or excluded, making it impossible to generalize with any accuracy from this kind of survey. This kind of sample should only be used as a last resort. Snowball samples Snowball sampling is used when the boundaries of the population are unknown, and it is impossible to obtain the sample frame. In this type of sample each suitable unit which is identified is included in the sample, and then used to identify other appropriate units that are also included. This method should
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks