Handout #8: - Sampling
The process of sampling involves any technique that uses part of the population to make conclusions regarding the entire population. The purpose is to use statistics (sample values) to estimate some unknown parameters (population values).
A CENSUS is complete enumeration of the population
Sampling:
- cuts cost
- reduces manpower requirements
- gathers vital information quickly
- gives accurate and reliable results since mistakes increase with volume of work
- limits the destruction to test units in quality control testing
1. Define the target population.
The target population is the specific complete group relevant to the study. Answers the question, “to whom do we talk?” For example in a study of housewives, who is really a housewife? One who keeps house? Does that include women in common-law marriage (baby-mothers)?
2. Select a sample frame. (Where necessary)
A sampling frame is a list of study subjects from which the actual sample is chosen. Maps or aerial photographs may be utilised as sampling frames. There may be omission or commission errors which may need to be updated.
3. Determine sampling method.
Factors that determine the appropriate method include:
- resources (money, people)
- degree of accuracy
- advance knowledge of population
- survey scale (local or national?)
4. Workout details (sample size etc)
Technically, the size of the sample depends on the precision the researcher desires in estimating the population parameter at a particular confidence level.
There is no single rule that can be used to determine sample size. Other things being equal, a large sample is much more likely to be representative of the population. If the population is heterogeneous a larger sample will be needed. Sample size over 30 permits the use of large sample statistics.
5. Conduct fieldwork.
There are two broad categories of sampling methods viz., probability and non-probability methods.
All probability samples are based on random selection procedures. Every element in the population has a non-zero probability of selection. They are popular because of their sound theoretical basis. Probability sampling results may be generalised.
Each unit of the population is assigned a number and sample units are selected randomly using a table of random numbers, a computer or some other random selection procedure that guarantees each member of the population has an equal chance of being selected. The “blind draw” is a form of SRS.
SRS begins with a list and to obtain a current and complete listing is often difficult. It is quite useful for small populations and computerised lists. Its calculations are easy.
Using a list of the population, the researcher chooses a random start point. A constant sampling interval is then used to select other sample units.
Sampling interval = population list size ¸ sample size (This ensures that the entire list is covered.)
It is less difficult than SRS because it is not necessary to write all numbers on slips of paper or computer files. Though it is similar, less time consuming and less expensive it is also less representative than SRS. If the list has hidden patterns or periodicities it becomes highly unrepresentative.
If the population has distinguishing factors that affect it such as sex, income, status, education or geographical location, etc, the population would need to be divided into sub-samples or strata. The groupings should be homogenous. A probability sample must be taken from each stratum. Weighting procedures may then be applied to estimate the population values. Allocation is either done by proportional or optimum allocation. In optimum allocation the sample size is proportional to the stratum size and the variability.
A cluster sampling method divides the population into groups any of which can be considered a representative sample. The groupings are heterogeneous as the population. The researcher randomly selects clusters and does either a one-step or two-step sampling. It is better to take many small clusters than a few larger clusters. Only a listing of the clusters is required. Frequently utilised when no lists are available. Hospitals and hi-rise buildings exist as clusters. Cluster sampling is also common in industries where items appear in bales, boxes, containers which are used as primary sampling units.
Non-probability samples are used for those research situations in which probability samples would be extremely expensive and/or when precise representativeness is not essential. Though they save time, money and effort there is no proper basis for measuring their effectiveness since they are not probability based.
Convenience samples (“availability”
or “accidental”)
Only those respondents that are close at hand are selected; for example, when volunteers are used and in street surveys. In effect they talk to whoever is available.
A researcher selects a sample that on the basis of available information can be judged to be representative of the total population. For example, choosing a typical village to represent national rural population.
Few respondents are identified who are used as informants to identify others. Used for hidden or obscure populations; for example, studies of prostitution or homosexuality.
After the population has been classified, interviewers select their respondents within each subgroup using their own judgement.
NOTE: It is possible to combine probability and non-probability sampling techniques in the same design.