Sampling Fundamentals - Summary
Sampling Fundamentals - Summary
There are four main considerations in developing a probability sample. First, the target population must be defined. In doing so, the researcher should look to the research objectives for guidance and consider alternative definitions.
Second, the mechanism for selecting the sample needs to be determined The simple random sampling, cluster sampling, stratified sampling, and multistage designs are among the available choices. It is important to consider the differences that may exist between the population list or sampling frame from which the sample is drawn and the target population. Potential biases should be Identified. For example, in telephone interviewing, the telephone directory will not Include those with unlisted numbers.
The third consideration is sample size. Several ad hoc methods are available, such as ensuring that there are at least 100 sample members for each group within the population that is of interest. In the next chapter we will examine analytical approaches to determining sample size. Regardless of the method chosen, four factory must be considered: the number of subgroups to be analyzed, the accuracy desired, the cost of sampling, and the amount of variation within the population.
The fourth consideration is nonresponse bias. Nonresponse bias can be reduced by improving the research design to reduce refusals and by using callbacks. Sometimes the best approach is to estimate the amount of bias and adjust the interpretation accordingly.
Nonprobability sampling methods, such as judgmental sampling, snowball sampling, and quota sampling, are appropriate in the right context, even though they can be biased and lack precise estimates of sampling
16Sudman, op. cit., 430. 17Ibid., 430.
variation. Shopping-center sampling is widely used, in part because it :s relatively inexpensive. Biases in shopping-center samples can be reduced b* adjusting the sample to reflect shopping-center characteristics, the locatitr of the shoppers within the shopping center, the time period of the inter viewing, and the frequency of shopping.
It is true that judgmental sampling does contain potential biases; however, the reality is that there are many sources of bias in research, and the biases associated with judgmental sampling may be small in terms of the total overall design. In particular, as the previous chapters have indicated bias and uncertainty can be caused by:
1. Nonresponse bias in probability sampling. In fact, many of the biases mentioned with respect to judgmental sampling also occur when there is substantial nonresponse in a probability sampling scheme.
2. Wording of questions. Questions can be ambiguous, hard to understand, or biased.
3. Questionnaire structure. The questionnaire can be too long or badly organized and motivated; the result can be fatigue and resentment.
4. Interviewer bias. The way the interviewer follows the sampling plan, introduces the survey, asks the questions, and interprets the answers all can create both bias and uncertainty in the data
5. Data analysis. The data-analysis phase, to be discussed in upcoming chapters, can involve coding errors and errors in interpretation.
Thus, sampling errors need to be kept in perspective. It is silly but common to create a sampling design that has less than a one percent sampling error but may have a 30 percent incidence of errors from other sources.
There are four main considerations in developing a probability sample. First, the target population must be defined. In doing so, the researcher should look to the research objectives for guidance and consider alternative definitions.
Second, the mechanism for selecting the sample needs to be determined The simple random sampling, cluster sampling, stratified sampling, and multistage designs are among the available choices. It is important to consider the differences that may exist between the population list or sampling frame from which the sample is drawn and the target population. Potential biases should be Identified. For example, in telephone interviewing, the telephone directory will not Include those with unlisted numbers.
The third consideration is sample size. Several ad hoc methods are available, such as ensuring that there are at least 100 sample members for each group within the population that is of interest. In the next chapter we will examine analytical approaches to determining sample size. Regardless of the method chosen, four factory must be considered: the number of subgroups to be analyzed, the accuracy desired, the cost of sampling, and the amount of variation within the population.
The fourth consideration is nonresponse bias. Nonresponse bias can be reduced by improving the research design to reduce refusals and by using callbacks. Sometimes the best approach is to estimate the amount of bias and adjust the interpretation accordingly.
Nonprobability sampling methods, such as judgmental sampling, snowball sampling, and quota sampling, are appropriate in the right context, even though they can be biased and lack precise estimates of sampling
16Sudman, op. cit., 430. 17Ibid., 430.
variation. Shopping-center sampling is widely used, in part because it :s relatively inexpensive. Biases in shopping-center samples can be reduced b* adjusting the sample to reflect shopping-center characteristics, the locatitr of the shoppers within the shopping center, the time period of the inter viewing, and the frequency of shopping.
It is true that judgmental sampling does contain potential biases; however, the reality is that there are many sources of bias in research, and the biases associated with judgmental sampling may be small in terms of the total overall design. In particular, as the previous chapters have indicated bias and uncertainty can be caused by:
1. Nonresponse bias in probability sampling. In fact, many of the biases mentioned with respect to judgmental sampling also occur when there is substantial nonresponse in a probability sampling scheme.
2. Wording of questions. Questions can be ambiguous, hard to understand, or biased.
3. Questionnaire structure. The questionnaire can be too long or badly organized and motivated; the result can be fatigue and resentment.
4. Interviewer bias. The way the interviewer follows the sampling plan, introduces the survey, asks the questions, and interprets the answers all can create both bias and uncertainty in the data
5. Data analysis. The data-analysis phase, to be discussed in upcoming chapters, can involve coding errors and errors in interpretation.
Thus, sampling errors need to be kept in perspective. It is silly but common to create a sampling design that has less than a one percent sampling error but may have a 30 percent incidence of errors from other sources.
Comments
Post a Comment