Sampling: What is it? Quantitative Research Methods ENGL 5377 Spring 2007

Sampling: What is it? Quantitative Research Methods ENGL 5377 Spring 2007 Bobbie Latham March 8, 2007

Introduction In any research conducted, people, places, and things are studied. The opportunity to study the entire population of those people, places, and things is an endeavor that most researchers do not have the time and/or money to undertake. The idea of gathering data from a population is one that has been used successfully over the years and is called a census. This method is mentioned several times in the bible (Wikipedia). It was also used by the Ancient Egyptians to obtain empirical data describing their subjects (Babbie 37). In past years, the idea of collecting data from the entire population was used by political entities to collect opinions about potential political candidates. Census data collection is still very popular for collecting public opinion for political endeavors. For most researchers, however, collecting data from an entire population is almost impossible because of the amount of people, places, or things within the population. Taking a census involves much time and money; something to which most researchers are not accustomed. To collect data on a smaller scale, researchers gather data from a portion or sample of the population. The purpose of this paper is to describe sampling as a method of data collection. Probability and non-probability sampling as well as the surrounding validity issues will be discussed. Sampling theory may be adapted for content analysis, laboratory experiments, and participant observation (Babbie 100). However, this paper will focus on sampling as a method to select participants for surveys; more specifically interviewing and self-administered questionnaires. Sampling Definitions The sample method involves taking a representative selection of the population and using the data collected as research information. A sample is a subgroup of a population (Frey et al. 125). It has also been described as a representative taste of a group (Berinstein 17). The sample should be representative in the sense that each sampled unit will represent the characteristics of a known number of units in the population (Lohr 3). All disciplines conduct research using sampling of the population as a method, and the definition is standard across these disciplines. Only the creative description of sampling changes for purposes of creating understanding. The standard definition always includes the ability of the research to select a portion of the population that is truly representative of said population. Sampling theory is important to understand in regards to selecting a sampling method because it seeks to make sampling more efficient (Cochran 5). Cochran posits that using correct sampling methods allows researchers the ability to reduce research costs, conduct research more efficiently (speed), have greater flexibility, and provides for greater accuracy (2). Two standard categories of the sampling method exist. These two categories are called probability sampling and non-probability sampling. Probability sampling is sometimes

called random sampling as non-probability sampling is sometimes called non-random sampling. These terms are interchangeable. For the purpose of this paper, I will use probability and non-probability as the naming conventions for the two sampling method categories. It is important to note that all sample selection methods described are selection without replacement, that is once a unit is selected in the sampling process, it is removed from the pool eligible for future selection (Henry 27). All other texts referenced in this paper assume selection without replacement. Gary Henry defines selection with replacement as a process where the selected unit is returned to the pool eligible for selection (28); however, no other references were found to this type of selection method. The choice to use probability or non-probability sampling depends on the goal of the research. When a researcher needs to have a certain level of confidence in the data collection, probability sampling should be used (MacNealy 125). Frey, et al. indicates that the two sampling methods differ in terms of how confident we are about the ability of the selected sample to represent the population from which it is drawn (126). Probability samples can be rigorously analyzed to determine possible bias and likely error (Henry 17). Non-probability sampling does not provide this advantage but is useful for researchers to achieve particular objectives of the research at hand (Henry 17). These objectives may allow for selection of the sample acquired by accident, because the sample knows the most, or because the sample is the most typical (Fink & Kosecoff 53). Probability and non-probability sampling have advantages and disadvantages and the use of each is determined by the researcher s goals in relation to data collection and validity. Each sampling category includes various methods for the selection process. Probability Sampling Probability sampling provides an advantage because of researcher s ability to calculate specific bias and error in regards to the data collected. Probability sampling is defined as having the distinguishing characteristic that each unit in the population has a known, nonzero probability of being included in the sample (Henry 25). It is described more clearly as every subject or unit has an equal chance of being selected from the population (Fink 10). It is important to give everyone an equal chance of being selected because it eliminates the danger of researchers biasing the selection process because of their own opinions or desires (Frey, et al. 126). When bias is eliminated, the results of the research may be generalized from the sample to the whole of the population because the sample represents the population (Frey, et al. 126). There are four types of probability sampling that are standard across disciplines. These four include simple random sampling, systematic random sampling, stratified random sampling, and cluster sampling (Table 1). Simple Random Sampling Simple random sampling is often called straight random sampling. The naming convention of this type of probability sampling method is not indicative of the discipline

but reliant upon the researcher or author of the various books and articles referenced. That is to say that these two terms are interchangeable and is not interdependent on a specific discipline within academia. Simple random sampling requires that each member of the population have an equal chance of being selected (as is the main goal of probability sampling). A simple random sample is selected by assigning a number to each member in the population list and then use a random number table to draw out the members of the sample (MacNealy 155). Sharon Lohr explains that by using simple random sampling, the researcher is in effect mixing up the population before grabbing n units (24). Another way of viewing simple random sampling precludes that all members of the study population are either physically present or listed, and the members are selected at random until a previously specified number of members or units has been selected (Henry 27). Each member of the population is selected one at a time, independent of one another and without replacement; once a unit is selected, it has no further chance to be selected (Fowler 14). Regardless of the process used for simple random sampling, the process can be laborious if the list of the population is long or it is completed manually without the aid of a computer (Babbie 84; Fowler, Jr. 14). An example of simple random sampling may include writing each member of the population on a piece of paper and putting in a hat. Selecting the sample from the hat is random and each member of the population has an equal chance of being selected. This example is not feasible for large population, but can be completed easily if the population is very small. Researchers who choose simple random sampling must be cognizant of the numbers that they choose. Researcher bias in regards to preferred numbers can be a problem for the end results in regards to sample selection (Frey, et al. 126). It is best to ask other researchers to aid in the selection of the numbers to be used in the selection process. It is also important to note that by using simple random sampling, the sample selected may not include all elements in the population that are of interest (Fink 11). Systematic Random Sampling Systematic random sampling is usually preferred over simple random sampling in so far as it is more convenient for the researcher. This type of probability sampling is also called ordinal sampling and pseudo-simple random samples (Frey, et al. 128; Henry 28). Systematic random sampling includes selection of sampling units in sequences separated on lists by the interval of selection (Kish 21). The selection of the sample from the population list is made by randomly selecting a beginning and choosing every nth name (MacNealy 155). Frey (et al.) calls the interval used to select every nth name the sampling rate (28). Earl Babbie defines the same as sampling interval (84). Before selecting from the population list, determine the number of entries on the list and the number of elements from the list that are to be selected (Fowler, Jr. 14). For example, if there are 129 people on the population list select a beginning or starting point

at random and choose every tenth name that appears on the list. If you randomly choose to begin on the name that appears on line 24, you will select for the sample the names that appear on lines 24, 34, 44, and so on. The most important element of systematic random sampling is that the selection starting point is random. (Babbie 84; Fowler Jr. 14; Henry 28; MacNealy 155). One inherent disadvantage to systematic random sampling that researchers face is that the population list should be carefully examined for arrangement order (Babbie 85). Babbie goes on to explain that if the elements are arranged in any particular order, you should ascertain whether that order will bias the sample to be selected and should take steps to counteract any possible bias (85). Henry expounds by describing that this issue arises when the population listing is arranged in cyclical fashion and the cycle coincides with the selection interval (28). This problem can be remedied by examining the list and making sure that the list of names is not arranged in any type of order. Table 1 Probability Sampling Methods Type of Sampling Simple Selection Strategy Each member of the study population has an equal probability of being selected. Systematic Stratified Cluster Each member of the study population is either assembled or listed, a random start is designated, then members of the population are selected at equal intervals Each member of the study population is assigned to a group or stratum, then a simple random sample is selected from each stratum. Each member of the study population is assigned to a group or cluster, then clusters are selected at random and all members of a selected cluster are included in the sample. (Henry 27) Stratified Random Sampling Stratified random sampling is one in which the population is divided into subgroups or strata, and a random sample is then selected from each subgroup (Fink 11). When a few characteristics are know about a population, stratified random sampling is preferable because the population may be arranged in subgroups and then a random sample may be

selected from each of these subgroups (Babbie 85; Cochran 65; Fowler, Jr. 15; Henry 28; Kish 21). MacNealy further advises arranging the original unit into categories so that the distribution of a particular group in the population of interest will be closely replicated in the sample (156). These subgroups can exhibit characteristics including but not limited to gender, race, ethnicity, religion, and age groups. Two types of stratified random sampling include proportionate and disproportionate. Proportionate stratification is often done to insure representation of groups that have importance to the research and disproportionate is done to allow analysis of some particular strata members or to increase the overall precision of the sample estimates (Henry 29). The big difference between the two stems from the use of a fraction. Proportionate stratified uses the same fraction for each subgroup and disproportionate uses different fractions for each subgroup. To choose which is right for a research project, the researcher must be aware of the various numbers of members in each subgroup. Take for instance a population of churches in Lubbock, Texas. Whereas the First Baptist Church may have 700 members in the subgroup, the Assembly of God may only have 130 members. This is yet another choice the researcher must make A more simple example is if the population being examined is high school students, of which the population is 55% female and 45% male, the population should be listed by gender. The selection process would then include selecting every nth female from the female list until 55% of the list is of the female gender. The remaining 45% should be selected from the male list by choosing every nth male. This ensures that the sample is representative of the population in so far as gender is concerned. A concern when using stratified random sample is that the researcher must identify and justify the subgroups (Fink 13). By using stratified random sampling, there is an attempt to control for sampling error (MacNealy 156). To control for sampling error, researchers must not only identify and justify the subgroups but make sure they are truly representative of the population. Cluster Sampling Cluster sampling, on the surface, is very similar to stratified sampling in that survey population members are divided into unique, nonoverlapping groups prior to sampling (Henry 29). These groups are referred to as clusters instead of strata because they are naturally occurring groupings such as schools, households, or geographic units (Henry 29). Where as a stratified sample involves selecting a few members from each group or stratum, cluster sampling involves the selection of a few groups and data are collected from all group members (Henry 29). This sampling method is used when no master list of the population exists but cluster lists are obtainable (Babbie 88; Frey, et al. 130; Henry 29; Lohr 24; MacNealy 156). For example, a researcher wants a list of all special education teachers in the United States. A comprehensive list does not exist; therefore, the researcher must then contact each state and ask for a comprehensive list for that specific state. If each state does not compile a comprehensive list, the researcher would then contact each school in that state

asking for a list of special education teachers. The groups of special education teachers compiled may then be put into groups or clusters depending on the state in which they are teaching. It is important to note that with the method of cluster sampling, an additional sampling method resides. Multistage sampling is used in cluster sampling. At least one reference separated multistage sampling from cluster sampling as a probability sampling method (Henry 30). Another, Fowler, Jr., named only multistage sampling and left the word cluster out all together (18). Henry indicates that multistage sampling is an extension of cluster sampling whereas all others include within the method of multistage sampling as part of cluster sampling. Multistage sampling occurs when a researcher must cluster together certain groups because a master list is not available but encounters a more complex design. It involves two stages: 1) Select clusters randomly from the population and list, and 2) Select individuals randomly from the clusters (Babbie 88; Frey et al. 130). While multistage is a part of cluster sampling in most of the books researched, not all see it as one method. A drawback to using cluster sampling occurs within the precision of the statistics (Babbie 88; Henry 30). While cluster sampling is convenient when a master list of the population does not exist, the researcher will run the risk of inaccurate findings. One way to increase the accuracy of results from cluster sampling is to use many clusters when implementing multistage sampling (Fink 16). Fink goes on to explain as you increase the number of clusters, you can decrease the size of the sample within each (16). Non-probability Sampling The advantage of non-probability sampling is that it a convenient way for researchers to assemble a sample with little or no cost and/or for those research studies that do not require representativeness of the population (Babbie 97). Non-probability sampling is a good method to use when conducting a pilot study, when attempting to question groups who may have sensitivities to the questions being asked and may not want answer those questions honestly, and for those situations when ethical concerns may keep the researcher from speaking to every member of a specific group (Fink 17). In nonprobability sampling, subjective judgments play a specific role (Henry 16). Researchers must be careful not to generalize results based on non-probability sampling to the general population. Non-probability sampling includes various methods. None of the resources agree on all of them. MacNealy indicates three methods where as Frey, et al. and Henry provide five methods under non-probability sampling. Frey, et al. and Henry do not agree on the naming conventions of the five given in each of their books. I attempt to summarize the various non-probability sampling methods in Table 2.

Table 2 Various Non-probability Sampling Methods by Author Author Types of Non-probability Sampling Babbie - Purposive or judgmental sampling - Quota sampling - Reliance of available subjects (Convenience) Fink - Convenience - Snowball sampling - Quota sampling - Focus groups Frey, et al. - Convenience - Volunteer - Purposive - Quota - Network (snowball) Henry - Conveniences samples - Most similar/most dissimilar samples (purposive) - Typical case samples (purposive) - Critical case samples (purposive) - Snowball samples - Quota samples MacNealy - Convenience sampling - Purposeful sampling - Snowball sampling The following non-probability sampling methods will be discussed in this section with reference to the various naming conventions in table 2: Convenience, Purposive, Snowball, and Quota. These four methods of non-probability sampling cover all those listed in Table 2 although the naming conventions are not the same. Table 3 summarizes the four non-probability sampling methods. Convenience Convenience sampling includes participants who are readily available and agree to participate in a study (Fink 18; Frey, et al. 131; Henry 18; MacNealy 156). MacNealy indicates that convenience sampling is often called accidental (156), while Frey, et al. agree with the alternate title of accidental but also include haphazard as an alternate title

(131). Babbie does not use the specific title of convenience, but calls this same type of non-probability sample reliance on available subjects (99). All of these alternate names for convenience non-probability sampling include the same definition. Convenience is just that convenient. This is a relatively easy choice for researchers when a group of people cannot be found to survey or question. For example, convenience sampling may include going to a place of business (mall, restaurant, etc.) and questioning or surveying those people who are available and consent to being questioned. If the researcher is interested in what people think of hair cutting techniques from a consumer perspective, the researcher may go to a hair salon and a barber shop and poll those patrons leaving the establishment after getting their hair cut. While convenience sampling includes only those ready and available, there is no excuse for sloppiness (Babbie 99). Babbie goes on to explain that survey researchers need to find ways of procuring a sample that will represent the population they are interested in learning about (99). In the example above, the interest is in people who have had their hair cut recently. The researcher would get far less results from those people exiting a restaurant. While some of those people may have had their haircut that day, the better selection is to go to a place where haircuts take place. Purposive Purposive non-probability sample is also known as judgment or judgmental (Babbie 97; Jones 766). It is referred to as purposeful by MacNealy (157). Gary Henry breaks purposive down into three different methods: Most similar/dissimilar cases, typical cases, and critical cases. No matter the naming convention used, all authors agree on the definition of this non-probability sampling method. Purposive sampling is selecting a sample on the basis of your own knowledge of the population, its elements, and the nature of your research aims (Babbie 97). That is the population is non-randomly selected based on a particular characteristic (Frey, et al. 132). The individual characteristics are selected to answer necessary questions about a certain matter or product (MacNealy 157). The researcher is then able to select participants based on internal knowledge of said characteristic. This method is useful if a researcher wants to study a small subset of a larger population in which many members of the subset are easily identified but the enumeration of all is nearly impossible (Babbie 97). Pilot studies are well suited to this type of non-probability sampling method. For example, if a researcher wants to know student thoughts on using an online registration system, those students who attempt to use the system would be surveyed. If this survey took place at one institution, the results could not be generalized to every institution utilizing web registration, only the institution where the survey took place. Frey, et al. indicates that purposive non-probability sampling and stratified probability sampling are very similar but warn that there is a crucial difference between the two. Researchers using purposive sampling do not select respondents randomly from each

group within the stratification categories where as stratified sampling includes random sampling at its core (132). All respondents, not only those randomly selected, who possess the characteristic are included (132). It is important to note that purposive sampling precludes that the researcher understand the characteristics clearly and thoroughly enough to choose the sample and relate those findings only to that specific group and not to the population as a whole. Table 3 Non-probability Sampling Methods Type of Sampling Convenience Selection Strategy Select cases based on their availability for the study. Purposive Snowball Quota Select cases that judged to represent similar characteristics. Group members identify additional members to be included in the sample. Interviewers select a sample that yields the same proportions as the population proportions on easily identified variables. (Henry 18) Snowball Frey, et al. call snowball sampling network sampling (133). The definitions are the same. Snowball sampling is used in those rare cases when the population of interest cannot be identified other than by someone who knows that a certain person has the necessary experience or characteristics to be included (MacNealy 157). Snowball sampling also includes relying on previously identified group members to identify others who may share the same characteristics as the group already in place (Henry 21). For example, a researcher wants to find usability engineers who have lost their job due to company downsizing. A list of these types of people does not exist, but if the researcher knows someone who has experienced this, that person may know of others and give contact information so that others may be added to the group. MacNealy describes this as one participant leads to another (157). Again, this type of non-probability sampling cannot be generalized to a population but can be generalized to the group who shares the same characteristics.

Quota MacNealy does not include quota sampling in her non-probability section, but others do well to define it as dividing the population group being studied into subgroups. Then based on the proportions of the subgroups needed for the final sample, interviewers are given a number of units from each subgroup that they are to select and interview (Henry 22). Quota sampling is a good method to use to non-randomly select groups based on gender, age, race, and ethnicity, to name a few. Frey, et al. describe quota sampling where respondents are selected non-randomly on the basis of their known proportion to the population (133). Gary Henry describes quota sampling as dividing the population group into subgroups and based on the proportions, interviewers are given a number of units from each subgroup that they are to select and interview (22). Henry compares quota sampling to stratified probability sampling but gives a big difference between the two. Quota non-probability sampling and stratified probability sampling are different in that quota sampling allows the interviewer discretion in the selection of the individuals for the sample (22). There are a number of problems that researchers should be aware of when choosing to use this method of non-probability sampling: The list of subgroups and the proportions identified must be accurate before the sampling begins (Babbie 99). The selection of the sample elements within a given cell (for proportion choice) may include bias although the proportion of the population is estimated correctly (Babbie 99). Non-response is hidden in quota sampling because the interviewer may simply select another household to interview and may under represent the proportion of the population that is difficult to reach (Henry 22). Generalizations to the population cannot be made when using quota sampling (Henry 22). Gary Henry concludes his discussion of quota sampling by indicating that this nonprobability sampling method has fallen out of favor for the reasons stated above as well as indicating that quotas is a biased sampling technique, although the bias is generally small (23). He also indicates that sampling error may higher when using this technique (23). Sample Errors As with all research methods, sampling provides some room for error on the part of the researcher. Being aware of those possible errors is essential in selection of the sampling method used as well as calculation of the data collected. Simply being aware of possible errors is often not enough. Arlene Fink believes that no matter how thorough and proficient the researcher is, sampling bias or error is inevitable (25). Sampling error may be defined as the error that results from taking one sample instead of examining the

whole population (Lohr 15). Lohr simply defines several types of sample errors as undercoverage, nonresponse, and sloppiness in data collection (16). Undercoverage refers to selecting a sample that is not large enough. The error here is that the information gathered from a small sample is not representative of the population and cannot be generalized to that population. Gary Henry indicates that small sample size may contribute to a conservative bias (Type II error) in the application of the statistical test (13). This happens when a null hypothesis is not rejected although in fact it is false (13). Nonresponse is a nonsampling error that precludes that some members of the population who are eligible to be sampled are unwilling to participate or do not answer all questions on the survey(s) (Cochran 292: Fink 26; Henry 124; Lohr 6). Lohr indicates that the main problem caused by nonresponse is potential bias of population estimates (257). Nonsampling errors occurs because of imprecision in the definition of the target and study population and errors in survey design and measurement (Fink 25). Some errors of nonsampling include changes due to historical circumstances, neglecting definitions and inclusion and exclusion of criteria, and instrument or survey process instrument bias (Fink 26). Researchers should keep in mind that an increase and sample size and an increased homogeneity of the elements being sampled allow for the reduction of sampling error (Babbie 89). However, Lohr warns that increasing the sample size without targeting nonresponse does nothing to reduce nonresponse bias; a larger sample size merely provides more observations from the class of persons that would respond to the survey (257). Conclusion Researchers may choose from a variety of sampling methods. The researcher goals inform which sampling method is best for the research to be conducted. The main choice in regards to sample method choice is whether or not the researcher wants to generalize the findings from the sample to the whole of the population being studied. Being aware of possible errors due to the sample method chosen is also very important because giving possible errors within the results section allows the study to be regarded as valid. Many sample method choices are available; the researcher must choose the method that is right for the study.

References Babbie, Earl. Survey Research Methods. Belmont, California: Wadsworth Publishing Company, 2 nd ed., 1990. Berinstein, Paula. Business Statistics on the Web: Find Them Fast At Little or No Cost. New Jersey: CyberAge Books, 2003. Cochran, William G. Sampling Techniques. New York: John Wiley & Sons, Inc., 1953. Fink, Arlene. How to Sample in Surveys. Vol. 6. London: Sage Publications, 1995. Fowler, Jr. Floyd J. Survey Research Methods. 2 nd ed. Vol. 1. London: Sage Publications, 1993. Frey, Lawrence R., Carl H. Botan, and Gary L. Kreps. Investigating Communication: An Introduction to Research Methods. 2 nd ed. Boston: Allyn and Bacon, 2000. Henry, Gary T. Practical Sampling. Vol. 21. London: Sage Publications, 1990. Jones, Howard L. The Application of Sampling Procedures to Business Operations. Journal of the American Statistical Association. 50.271 (1955): 763-774. Kish, Leslie. Survey Sampling. New York: John Wiley & Sons, 1965. Lohr, Sharon L. Sampling: Design and Analysis. Albany: Duxbury Press, 1999. MacNealy, Mary Sue. Strategies for Empirical Research in Writing. New York: Longman, 1999. Wikipedia. The Free Encyclopedia. 7 March 2007. 22 February 2007 http://en.wikipedia.org/wiki/sampling_(statistics)#history_of_sampling