CHAPTER 4: RESEARCH DESIGN AND METHODOLOGY
4.5 Sample Design
4.5.2 Sample Selection
The next step in determining the sample is to specify the process used to select the population under study. There are two broad sampling methods: probability and nonprobability sampling. In probability samples, each element of the population has a known chance (or probability) of being selected for the sample. This implies that the sampling operation is controlled objectively and that the items are chosen randomly. So, the person undertaking the study does not influence the selection of sample items. Another advantage of probability sampling is that the “sampling error” can be calculated because a sample rather
than a census is employed. Sampling error is the degree to which a sample might differ from the population. So, when inferring to the population, results are reported plus or minus the sampling error. Probability sampling is time consuming and expensive. It is used when researchers want accurate descriptions of the population and in large-scale surveys. Probability methods include random sampling, systematic sampling, and stratified sampling (Babbie 2001; Churchill and Iacobucci 2002).
In nonprobability sampling, not all elements of the population have a known chance of being included in the sample; they are selected from the population in some random manner. So, to a large extend the selection process is subjective and it relies on either the researchers or the interviewers judgment. In nonprobability sampling, the degree to which the sample differs from the population cannot be measured. So, we cannot evaluate the adequacy of the sample. An advantage of this method is that sampling tends to be less
complicated and less time consuming than probability sampling. In our study, as it was explained in section 4.2.3, we have selected to collect our data through telephone interviewing. A major problem that researchers face with telephone interviewing is that of obtaining representative probability sample due to the nonresponse rate. So, even if we use a probability sample and select some elements of the population, we might fail to obtain information from them due to nonresponse. In telephone interviewing, nonresponse might be attributed either to the percentage of the population that do not have a telephone line in their home or to the percentage of those that do not list their number in the directory or to the increased usage of cell phones to the detriment of household telephone lines (Smith 1983; Fowler 2009). The above issues add a non-random selection to random sampling and destroy any attempt for randomization. Thus, a nonprobability sample was selected.
Nonprobability methods include convenience, judgment, and quota sampling (Babbie 2001; Churchill and Iacobucci 2002). The most widely used method of sampling in marketing surveys is the nonprobability quota sampling (Hauser and Hansen 1944; May 2001). A quota sample attempts “…to ensure that the sample is representative by selecting sample elements in such a way that the proportion of the sample elements processing a certain characteristic is approximately the same as the proportion of the elements with the characteristic in the population” (Churchill and Iacobucci 2002). Quota sampling is often confused with stratified and cluster sampling, two probability sampling methodologies. All of these methods sample a population that has been
subdivided into classes or categories. The primary differences between them is that with stratified and cluster sampling the classes are mutually exclusive and are isolated prior to sampling. Thus, the probability of being selected is known, and the members of the population are chosen at random. In quota sampling, sampling within each category is non-random and members of the population are arbitrarily disqualified from being selected (Moser and Kalton 1972; Churchill and Iacobucci 2002). Quota sampling can be either proportionate or non- proportionate. Proportionate quota sampling is based on population proportions and the number of observations is allocated accordingly. For instance, if we know that a certain grocery retailer has a market share of two percent and we have a total sample of 900 respondents, with the proportionate quota sample, we need to select 18 respondents that purchase from that grocery retailer. Therefore, the proportionate quota ensures that the composition of sample is the same as the composition of the population with respect to the market share of each retailer. In non-proportionate sampling, we identify the sub-groups from which we want to ensure sufficient coverage and then specify the sample size for each sub-group
In our study, a non-proportionate quota sample was selected due to the specific nature of our research objectives. Our study is focusing on those retailers that offer SB, and we want to compare results across the different retailers that offer SB. So, it is important that these nine supermarkets - the control characteristic - were represented in our sample. The interviewer was instructed to select respondents from each retailer based on a specified
proportion. When the quota sample was reached, the data collection for this retailer was completed and additional respondents were discarded from the results. Overall, the non-proportionate quota ensures that shoppers from the nine major grocery retailers that sell SBs would be included in the sample and that we will have enough responses per retailer to ensure the validity of our analysis at the level of the selected retailer.