Sampling

Types of Sampling

Random sampling (or probability sampling) is a process whereby every sampling unit in a finite population has an equal chance of being selected or not selected for participation in a research study.

Important in random sampling is that the chance of being included can be clearly calculated.

For example, if we choose 500 persons from a 50,000 population the chance of being included is 1 in 100.

Random sampling is statistically sounder than other types of sampling and is widely used by most major research organizations.

Random sampling is analogous to putting everyone's name into a hat and drawing out several names.

Each element in the population has an equal chance of occurring. While this is the preferred way of sampling, it is often difficult to do.

It requires that a complete list of every element in the population be obtained.

Computer generated lists are often used with random sampling.

Simple random sampling--accomplished by the lottery method or by using random tables.

The lottery method--every unit of the population is identified by a number disc or slip. They are well mixed and then the appropriate number of samples are chosen.

Random Tables--these are tables produced for sampling where random numbers are given for populations.

Both methods produce random selection that do not rely on human judgement.

Practically, there use is restricted to small populations.

Systematic or quasi-random sampling differs from random sampling in not giving equal probability of selection to all possible samples which could be taken from a population.

It is also widely used and offers the most practical approximation to random sampling.

This method entails calculating a sample interval as a starting point and then adding that interval to each succeeding number.

It is not strictly random, the initial number is random but the successive ones are not.

Example of systematic sampling:

In a survey covering a population of 10,000 it may be decided to take a sample of 250.

The sampling interval will be 10,000/250=40.

A randomly selected number between 1 and 40 is chosen (in our example we use 4).

The sampling series then becomes 4, 44, 84, etc. until we reach 250 samples.

Stratified sampling divides the population into groups called strata by some characteristic, not geographically.

For instance, the population might be separated into males and females. A sample is taken from each of these strata using either random, systematic, or convenience sampling.

This type of sampling may result in increasing the precision of the sample survey.

Clearly individuals are influenced by characteristics such as age, sex, income group, etc.

By dividing our samples into strata with similar characteristics we may be able assess their opinions more accurately.

Cluster sampling is accomplished by dividing the population into groups -- usually geographically.

These groups are called clusters or blocks. The clusters are randomly selected, and each element in the selected clusters are used.

This type of sampling is particularly useful where the populations under survey are widely dispersed, and it would be impractical to take a simple random sample.

This method is attractive for cost reasons, but it also increases the size of the sampling error.

Multi-stage sampling is used where populations are widely dispersed and interviewing would be difficult.

Interviews can be concentrated in convenient areas.

The selection process takes place in two or more stages until the final number of sampling units is reached.

(see example on p.92)

In replicated sampling what we do is to divide the sample into a number of equal sub-samples.

Each of these sub-samples is selected at random from the population.

Each represents a self-contained miniature population.

We use this type of sampling when the size of the population makes it impractical to get results quickly.

Master samples are normally produced by government organizations because of the large scale and complexity of producing this type of sample.

The idea is to have a "master" random sample from which sub-samples can be chosen.

With multi-phase sampling the same type of sampling unit is involved at each phase, with some units (a sub sample of the original sample) asked more in depth questions than others.

The idea is to gather useful and relevant data on a complex set of questions without having to incur high costs in extensive interviews of all units.

This method has been found to be very accurate. (see p.95)

Non-probability sampling occurs when human judgement is involved in the selection of sampling units.

This method is not encouraged as the results can almost never be applied to the population in general.

The only instances when this might be appropriate is in farm products or parts where you are certain that there is little variance in the whole.

This is a form of judgement sampling where the biases from the non-probability method of selection are controlled somewhat by stratification, weighting, and the setting of quotas for each stratum.

This type of sampling is used much by commercial researchers as it is very cost effective and easy to administrate.

See point on p.97

Interviewers randomly select people at specific points in geographically designated areas.

In Random location sampling interviewers are instructed to work an area until certain well-defined quotas are achieved.

GRID, is another form of this random sampling developed by the British Market Research Bureau.

Instead of starting your research with a predefined number of sample units, you can use sequential sampling.

Sequential sampling involves choosing units until a significant amount of knowledge has been gained about a particular question.

This is often used in trying to determine the ownership of a specific product or the circulation of a specific magazine.

Interviewers tend to be bias in who they choose to interview.

A disproportionate number of highly educated people and certain occupations tend to be more highly represented.

These problems can cause the data to be invalid.

All efforts should be made to minimize these biases by closing controlling and supervising the research staff.

Using random techniques, eliminates these problems.

Speed, economy, and administrative simplicity.

Interviews can obtain sample units without unnecessary traveling.

Administratively simple because it is independent of sampling frames.