Sampling Fundamentals - Nonprobability Sampling
Sampling Fundamentals - Nonprobability Sampling
In probability sampling, the theory of probability allows the researcher to calculate the nature and extent of any biases in the estimate and to determine what variation in the estimate is due to the sampling procedure. It requires a sampling frame—a list of sampling units or a procedure to reach respondents with a known probability. In nonprobability sampling, the costs and trouble of developing a sampling frame are eliminated, but so is the precision with which the resulting information can be presented. In fact, the results can contain hidden biases and uncertainties that make them worse than no information at all. These problems, it should be noted, are not alleviated by increasing the sample size. For this reason, statisticians prefer to avoid nonprobability sampling designs; however, they often are used legitimately and effectively.
It is worthwhile to distinguish among four types of nonprobability
sampling procedures: judgmental samples, snowball designs, convenience samples, and quota samples.
Judgmental Sampling
In judgmental sampling, an "expert" uses judgment tojdentify representa tive samples. For example, patrons of a shopping center might serve tc represent the residents of a city, or several cities might be selected to repre-sent a country.
Judgmental sampling usually is associated with a variety of obvious and not-so-obvious biases. For example, shopping center intercept interviewing can oversample those who shop frequently, who appear friendly and who have extra time. Worse, there is no way of really quantifying the resulting bias and uncertainty, because the sampling frame is unknown and the sampling procedure is not well specified.
There are situations where judgmental sampling is useful and even advisable. First, there are times when probability sampling is either non-feasible or prohibitively expensive. For example, a list of sidewalk vender; might be impossible to obtain, and a judgmental sample might be appropriate in that case.
Second, if the sample size is to be very small—say, under 10—a judgmental sample usually will be more reliable and representative than a probability sample. Suppose one or two cities of medium size were to be used tc represent 200 such cities. Then it would be appropriate to pick judgmen-tally two cities that appeared to be the most representative with respect tc such external criteria as demographics, media habits, and shopping characteristics. The process of randomly selecting two cities could very well generate a highly nonrepresentative set. If a focus-group interview of eight or nine people were needed, again, a judgmental sample might be a highly-appropriate way to proceed.
Third, it sometimes is useful to obtain a deliberately biased sample. If for example, a product or service modification were to be evaluated, it might be possible to identify a group that, by its very nature, should be disposed toward the modification. If it were found that they did not like it, then it could be assumed that the rest of the population would be at least as negative. If they liked it, of course, more research probably would be required.
Snowball Design
A snowball design is a form of judgmental sampling that is very appropriate when it is necessary to reach small, specialized populations. Suppose a long-range planning group wanted to sample people who were very knowledgeable about a new specialized technology, such as the use of lasers in construction. Even specialized magazines would have a small percentage of readers in this category. Further, the target group may be employed by diverse organizations, like the government, universities, research organizations, and industrial firms. Under a snowball design, each respondent, after being interviewed, is asked to identify one or more others in the field. The result can be a very useful sample. This design can be used to reach any small population, such as deep-sea divers, people confined to wheelchairs, owners of dunebuggies, families with triplets, and so on. One problem is that those who are socially visible are more likely to be selected.
Convenience Sampling
To obtain information quickly and inexpensively, a convenience sample
can be employed. The procedure is simply to contact sampling units that are convenient—a church activity group, a classroom of students, women at a shopping center on a particular day, the first 50 recipients of mail questionnaires, or a few friends and neighbors. Such procedures seem indefensible, and, in an absolute sense, they are. The reader should recall, however, that information must be evaluated, not "absolutely," but in the context of a decision. If a quick reaction to a preliminary service concept is desired to determine if it is worthwhile to develop it further, a convenience sample may be appropriate. It obviously would be foolish to rely on it in any context where a biased result could have serious economic consequences, unless the biases could be identified. A convenience sample often is used to pretest a questionnaire.
Quota Sampling
Quota sampling is judgmental sampling with the constraint that the sample include a minimum number from each specified subgroup in the population. Suppose a 1000-person sample of a city is desired and it is known how the population of the city is distributed geographically. The sample could be dispersed in the same manner, as shown in Table 11-6. Thus, interviewers might be asked to obtain 100 interviews on the east side, 300 on the north side, and so on.
Quota sampling often is based on such demographic data as geographic location, age, sex, education, and income. As a result, the researcher knows that the sample "matches" the population with respect to these demographic characteristics. This fact is reassuring and does eliminate some gross biases that could be part of a judgmental sample; however, there are often serious biases that are not controlled by the quota sampling approach.
The interviewers will contact those most accessible, at home, with time with acceptable appearance, and so forth. Biases will result. Of course, a random sample with a 15-to-25-percent or more nonresponse rate will have many of the same biases. Thus, quota sampling and other judgmental approaches, which are faster and cheaper, should not always be discarded as inferior.
In probability sampling, the theory of probability allows the researcher to calculate the nature and extent of any biases in the estimate and to determine what variation in the estimate is due to the sampling procedure. It requires a sampling frame—a list of sampling units or a procedure to reach respondents with a known probability. In nonprobability sampling, the costs and trouble of developing a sampling frame are eliminated, but so is the precision with which the resulting information can be presented. In fact, the results can contain hidden biases and uncertainties that make them worse than no information at all. These problems, it should be noted, are not alleviated by increasing the sample size. For this reason, statisticians prefer to avoid nonprobability sampling designs; however, they often are used legitimately and effectively.
It is worthwhile to distinguish among four types of nonprobability
sampling procedures: judgmental samples, snowball designs, convenience samples, and quota samples.
Judgmental Sampling
In judgmental sampling, an "expert" uses judgment tojdentify representa tive samples. For example, patrons of a shopping center might serve tc represent the residents of a city, or several cities might be selected to repre-sent a country.
Judgmental sampling usually is associated with a variety of obvious and not-so-obvious biases. For example, shopping center intercept interviewing can oversample those who shop frequently, who appear friendly and who have extra time. Worse, there is no way of really quantifying the resulting bias and uncertainty, because the sampling frame is unknown and the sampling procedure is not well specified.
There are situations where judgmental sampling is useful and even advisable. First, there are times when probability sampling is either non-feasible or prohibitively expensive. For example, a list of sidewalk vender; might be impossible to obtain, and a judgmental sample might be appropriate in that case.
Second, if the sample size is to be very small—say, under 10—a judgmental sample usually will be more reliable and representative than a probability sample. Suppose one or two cities of medium size were to be used tc represent 200 such cities. Then it would be appropriate to pick judgmen-tally two cities that appeared to be the most representative with respect tc such external criteria as demographics, media habits, and shopping characteristics. The process of randomly selecting two cities could very well generate a highly nonrepresentative set. If a focus-group interview of eight or nine people were needed, again, a judgmental sample might be a highly-appropriate way to proceed.
Third, it sometimes is useful to obtain a deliberately biased sample. If for example, a product or service modification were to be evaluated, it might be possible to identify a group that, by its very nature, should be disposed toward the modification. If it were found that they did not like it, then it could be assumed that the rest of the population would be at least as negative. If they liked it, of course, more research probably would be required.
Snowball Design
A snowball design is a form of judgmental sampling that is very appropriate when it is necessary to reach small, specialized populations. Suppose a long-range planning group wanted to sample people who were very knowledgeable about a new specialized technology, such as the use of lasers in construction. Even specialized magazines would have a small percentage of readers in this category. Further, the target group may be employed by diverse organizations, like the government, universities, research organizations, and industrial firms. Under a snowball design, each respondent, after being interviewed, is asked to identify one or more others in the field. The result can be a very useful sample. This design can be used to reach any small population, such as deep-sea divers, people confined to wheelchairs, owners of dunebuggies, families with triplets, and so on. One problem is that those who are socially visible are more likely to be selected.
Convenience Sampling
To obtain information quickly and inexpensively, a convenience sample
can be employed. The procedure is simply to contact sampling units that are convenient—a church activity group, a classroom of students, women at a shopping center on a particular day, the first 50 recipients of mail questionnaires, or a few friends and neighbors. Such procedures seem indefensible, and, in an absolute sense, they are. The reader should recall, however, that information must be evaluated, not "absolutely," but in the context of a decision. If a quick reaction to a preliminary service concept is desired to determine if it is worthwhile to develop it further, a convenience sample may be appropriate. It obviously would be foolish to rely on it in any context where a biased result could have serious economic consequences, unless the biases could be identified. A convenience sample often is used to pretest a questionnaire.
Quota Sampling
Quota sampling is judgmental sampling with the constraint that the sample include a minimum number from each specified subgroup in the population. Suppose a 1000-person sample of a city is desired and it is known how the population of the city is distributed geographically. The sample could be dispersed in the same manner, as shown in Table 11-6. Thus, interviewers might be asked to obtain 100 interviews on the east side, 300 on the north side, and so on.
Quota sampling often is based on such demographic data as geographic location, age, sex, education, and income. As a result, the researcher knows that the sample "matches" the population with respect to these demographic characteristics. This fact is reassuring and does eliminate some gross biases that could be part of a judgmental sample; however, there are often serious biases that are not controlled by the quota sampling approach.
The interviewers will contact those most accessible, at home, with time with acceptable appearance, and so forth. Biases will result. Of course, a random sample with a 15-to-25-percent or more nonresponse rate will have many of the same biases. Thus, quota sampling and other judgmental approaches, which are faster and cheaper, should not always be discarded as inferior.
Comments
Post a Comment