Sampling

The research strategy must include a sampling plan because the research usually does not include the entire population but only a part of it. Research on the whole population is often impossible, as it is too expensive and time-consuming (namely, results are obtained much faster on a sample basis).

The researchers must specify the following in a sampling plan:

• Sampling frame. It is the basis for including units into the sample. The sampling frame is crucial for research quality because it is supposed to identify all units of the target population;
• Sampling unit. The researchers must specify who or what they are studying;
• Sampling method. How to select the individuals whose characteristics will be the studied-sample size;
• Sample size. It is essential to determine how many persons must we include in research to obtain reliable results and at the same time not get over budget and miss deadlines.

There are two basic sample types:

• Probability samples, where units of the population have an a priori known and nonzero probability of being included in the sample. When discussing web surveys, probability samples are strongly influenced by noncoverage and sampling frame problems. Noncoverage error arises because all elements of the target population do not have access to the Internet. Sampling frame error arises because invitations to web surveys are usually shared through email and there are no complete records of email addresses of all internet users. An increase in sample size helps decrease sampling error;
• Nonprobability samples, where units of the population do not have an a priori known and nonzero probability of being included in the sample. Lower costs are the prevalent reason that most web surveys are based on nonprobability panels.

Statistical inference (i.e. drawing conclusions about the population based on a sample) is only possible by using a probability sample.

Web surveys seem less intrusive and easily traceable when compared to traditional surveys, but companies can't acquire opinions of people without computers, computer illiterate people or older population not yet touched by the computer revolution without the use of classic paper questionnaires and telephone surveys. But more and more people have access to the world wide web.

Web surveys can be carried out in three different ways:

1. General invitations:
   • There is no list of individuals who were invited to participate in the survey;
   • Nonprobability samples are employed.
2. Individual invitations:
   • There is a list of individuals who were invited to participate in the survey;
   • Such surveys can be based on either probability or nonprobability samples.
3. Intercept surveys:
   • There is no list of individuals that were invited to participate in the survey before it was carried out. Instead, such a list is formed while the survey is in progress by using some form of probability sampling;
   • Intercept surveys are a special category of probability web surveys.

We know multiple ways of surveying:

• Respondents can volunteer themselves without any encouragement and any individual urging. From the point of sampling we can't talk about probability sampling, where the value of inclusion of each unit from the population is known before and where each unit has a positive value of probability to be included, in this case;
• When researching users of, for example, the Internet, lists of email addresses represent sample frames. On this basis, we can then choose units into the sample and then invite them into the survey.

1KA users can easily disseminate survey invitations. An invitation is created in two simple steps, whereas the questionnaire link can be automatically sent via e-mail. If a list of respondents is not available, it is possible to embed the questionnaire link into an existing web page (e.g. banner ad).

<< Back to Web Surveys