Nonprobability Sampling Convenience Sampling Quota Sampling
Purposive or Judgmental Sampling Snowball Sampling
Deviant Case Sampling Sequential Sampling Probability Sampling
Populations, Elements, and Sampling Frames Why Random?
Types of Probability Samples Hidden Populations
How Large Should a Sample Be?
Drawing Inferences
From Chapter 6 of Basics of Social Research: Qualitative and Quantitative Approaches, Third Edition. W. Lawrence Neuman. Copyright © 2012 by Pearson Education, Inc. All rights reserved.
When you sample, you select some cases to examine in detail, then you use what you learn from them to understand a much larger set of cases. Most, but not all, social research studies use sampling. Depending on the study, how you conduct the sampling can differ. Most books on sampling emphasize its use in quantitative research and contain applied mathematics and quantitative examples. In quantitative studies, the primary purpose of sampling is to create a representative sample, that is, a sample (a selected small collection of cases or units) that closely reproduces features of interest in a larger collection of cases, called the population .
After sampling, you examine data in a sam- ple in detail; if you sampled correctly, you can generalize accuracy to the entire population. To create representative samples in quantitative research you need to use very precise sampling procedures. These procedures rely on the math- ematics of probabilities, hence they are called probability sampling.
In most quantitative studies, you want to see how many cases of a population fall into various categories of interest. For example, you might ask how many in the population of all Chicago’s high school students fit into various categories (e.g., high-income family, single-par- ent family, illegal drug user, arrested for delin- quent behavior, musically talented). The great thing about probability samples in quantitative research is their efficiency. They save a great deal of time and cost for the accuracy they deliver. A properly conducted probability sample may cost 1/1000 the cost and time of gathering informa- tion on an entire population yet yield virtually identical results. Let’s say you wanted to learn about the 18 million people in the United States diagnosed with diabetes. From a well-designed probability sample of 1,800, you can take what you learned and generalize it to all 18 million. It is more efficient to study 1,800 people to learn about 18 million than to study all 18 million people.
Probability samples can be highly accurate.
For large populations, a well-designed, carefully
executed probability sample can be equally if not more accurate than trying to reach every case in the population. This confuses many people. The U.S. Census tries to get data on all approximately 300 million Americans. A careful probability sample of 30,000 has a very tiny and known error rate. Trying to locate every single person of 300 million allows systematic errors to slip in unless extraordinary efforts are under- taken and huge amounts of time and money are expended. By the way, the scientific community for many years recommended against the U.S.
Census practice and recommended high-quality random samples, but political considerations overruled scientific ones.
Sampling proceeds differently and has dif- ferent purposes in qualitative studies. In fact, using the word sampling creates confusion in qualitative research because the term is so closely associated with quantitative research. In qualitative studies, you rarely sample a small set of cases that is a mathematically accurate repro- duction of the entire population so you can make statements about categories in the popu- lation. Instead, you sample to identify relevant categories at work in a few cases.
In quantitative sampling, you select cases/
units that you treat the cases/units as carriers of aspects/features of the social world. A sample of cases/units “stands in” for the much larger population of cases/units. By contrast, the logic of the qualitative sample is to sample aspects/
features of the social world. The aspects/features of your sample highlight or “shine light into”
key dimensions or processes in a complex social life. You pick a few to provide clarity, insight, and understanding about issues or relationships in the social world. In qualitative sampling, your goal is to deepen understanding about a larger process, relationship, or social scene. A sample provides valuable information or aspects, and these aspects accentuate, enhance, or enrich key features or situations.
In a qualitative study, you sample to open up new theoretical insights, reveal distinctive aspects of people or social settings, or deepen
by television programs is an example of a conve- nience sample. Television interviewers go out on the street with camera and microphone to talk to a few people who are convenient to interview.
The people walking past a television studio in the middle of the day do not represent everyone (e.g., homemakers, people in rural areas, etc.).
Likewise, television interviewers often select people who look “normal” to them and avoid people who are unattractive, poor, very old, or inarticulate. Another example of a convenience understanding of complex situations, events, or
relationships. However, we should not overdo the quantitative–qualitative distinction. In a few situations, a study that is primarily quantitative uses the qualitative–sampling strategy and vice versa. Nonetheless, most quantitative studies use probability or probabilitylike samples, and most qualitative studies use a nonprobability method and a nonrepresentative strategy.
NONPROBABILITY SAMPLING Instead of probability sampling, qualitative researchers use nonprobability or nonrandom samples . This means they rarely determine the sample size in advance and have limited knowl- edge about the larger group or population from which the sample is taken. Unlike a quantita- tive researcher who uses a preplanned approach based on mathematical theory, in a qualitative study you select cases gradually, with the specific content of a case determining whether it is cho- sen. Table 1 shows a variety of nonprobability sampling techniques.
Convenience Sampling
In convenience sampling (also called acciden- tal, availability, or haphazard) your primary criteria for selecting cases is that they are easy to reach, convenient, or readily available. This kind of sample may be legitimate for a few exploratory preliminary studies and some qualitative research studies when your purpose is something other than creating a representa- tive sample. Unfortunately, it often produces very nonrepresentative samples. It is not recom- mended if you want to create an accurate sample to represent the population. When you haphaz- ardly select cases that are convenient, you can easily get a sample that seriously misrepresents the population. Such samples are cheap and quick; however, the systematic errors that easily occur make them worse than no sample at all.
The person-on-the-street interview conducted
T A B L E 1 Types of Nonprobability Samples
Type of Sample Principle
Convenience Get any cases in any manner that is convenient.
Quota Get a preset number of cases in each of several predetermined categories that will reflect the diversity of the population, using haphazard methods.
Purposive Get all possible cases that fit particular criteria, using various methods.
Snowball Get cases using referrals from one or a few cases, and then referrals from those cases, and so forth.
Deviant Case Get cases that substantially differ from the dominant pattern (a special type of purposive sample).
Sequential Get cases until there is no additional information or new characteristics (often used with other sampling methods).
relevant categories among the population you are sampling to capture diversity among units (e.g., male and female; under age 30, ages 30–60, over age 60, etc.). Next, you determine how many cases to get for each category—this is your “quota.” Thus, you fix a number of cases in various categories of the sample at the start (see Figure 1 ).
Quota sampling is an improvement because you are ensuring some variety in the sample. In convenience sampling, all those interviewed might be of the same age, sex, or race. But once you fix the categories and number of cases in each quota category, you use convenience sam- pling. For example, you interview the first five males under age 30 you encounter, even if all five just walked out of the campaign head- quarters of a political candidate. Not only is sample is that of a newspaper that asks readers
to clip a questionnaire from the paper and mail it in. Not everyone reads the newspaper, has an interest in the topic, or will take the time to cut out the questionnaire and mail it. Some people will, and the number who do so may seem large (e.g., 5,000), but the sample cannot be used to generalize accurately to the population. Such convenience samples may have entertainment value, but they can give a distorted view and seriously misrepresent the population.
Quota Sampling
For many purposes, a well-designed quota sam- ple is an acceptable nonprobability substitute method for producing a quasi-representative sample. In quota sampling, you first identify
F I G U R E 1 Quota Sampling
Of 32 adults and children in the street scene, select 10 for the sample:
4 Adult Males 4 Adult Females
1 Female Child 1 Male Child
associate, etc.) and experts (e.g., police who work on vice units, other prostitutes, etc.) to identify a “sample” of prostitutes for inclusion in the research project. You use many different methods to identify the cases because your goal is to locate as many cases as possible.
3. You want to identify particular types of cases for in-depth investigation. The purpose is less to generalize to a larger population than it is to gain a deeper understanding of specific types. For example, Gamson (1992) used purposive sampling in a focus group study of what working-class people think about politics. Gamson wanted a total of 188 working-class people to participate in one of 37 focus groups. He sought respon- dents who had not completed college but who were diverse in terms of age, ethnic- ity, religion, interest in politics, and type of occupation. He recruited people from 35 neighborhoods in the Boston area by going to festivals, picnics, fairs, and flea markets and posting notices on many pub- lic bulletin boards. In addition to explain- ing the study, he paid the respondents well so as to attract people who would not tra- ditionally participate in a study.
Snowball Sampling
Snowball sampling (also called network, chain referral, or reputational sampling ) is a method for identifying and sampling (or selecting) the cases in a network. It uses the analogy of a snowball. It begins small but becomes larger as it is rolled on wet snow and picks up additional snow. Snowball sampling is a multistage technique. You begin with one or a few people or cases and spread out on the basis of links to the initial cases.
An important use of snowball sampling is to sample a network. We are often interested in an interconnected network of people or organi- zations. The network could be scientists around the world investigating the same problem, the misrepresentation possible because convenience
sampling is used within the categories, but noth- ing prevents you from selecting people who “act friendly” or who want to be interviewed.
A case from the history of sampling illus- trates the limitations of quota sampling. George Gallup’s American Institute of Public Opinion, using quota sampling, successfully predicted the outcomes of the 1936, 1940, and 1944 U.S. pres- idential elections. But in 1948, Gallup predicted the wrong candidate. The incorrect prediction had several causes (e.g., many voters were unde- cided, interviewing stopped early), but a major reason was that the quota categories did not accurately represent all geographical areas and all people who actually cast a vote.
Purposive or Judgmental Sampling Purposive sampling is a valuable kind of sam- pling for special situations. It is usually used in exploratory research or in field research. In this type of sampling the judgment of an expert or prior knowledge is used to select cases. It is not appropriate if it your goal is to get a representa- tive sample, or if you want to pick the “average”
or the “typical” case. In purposive sampling, the cases selected rarely represent the entire population.
Purposive sampling is most appropriate in the following three situations:
1. You use it to select unique cases that are especially informative. For example, you want to use content analysis to study mag- azines to find cultural themes. You select a specific popular women’s magazine to study because it is trendsetting.
2. You may use purposive sampling to select members of a difficult-to-reach, special- ized population. For example, you want to study prostitutes. It is impossible to list all prostitutes and sample randomly from the list. Instead, you use subjective infor- mation (e.g., locations where prostitutes solicit, social groups with whom prostitutes
You can also use snowball sampling in com- bination with purposive sampling. This is what Kissane (2003) did in a descriptive field research study of low-income women in Philadelphia.
The U.S. policy to provide aid and services to low-income people changed in 1996 to increase assistance (e.g., food pantries, domestic violence shelters, drug rehabilitation services, clothing distribution centers) delivered by nonpublic as opposed to government/public agencies. As frequently occurs, the policy change was made without a study of its consequences in advance.
No one knew whether the affected low-income people would use the assistance provided by nonpublic agencies as much as that provided by public agencies. One year after the new policy, Kissane studied whether low-income women were equally likely to use nonpublic aid. She focused on the Kensington area of Philadelphia.
It had a high (over 30 percent) poverty rate and was a predominately White (85 percent) section of the city.
First, Kissane (2003) identified nonpub- lic service providers by using telephone books, the Internet, referral literature, and walking down every street of the area until she iden- tified 50 nonpublic social service providers.
She observed that a previous study found low- income women in the area distrusted outsiders and intellectuals. Her snowball sample began asking service providers for the names of a few low-income women in the area. She then asked those women to refer her to others in a similar situation, and asked those respondents to refer her to still others. She identified 20 low-income women aged 21 to 50, most who had received public assistance. She conducted in-depth, open- ended interviews about their awareness and experience with nonpublic agencies. She learned that the women were less likely to get nonpublic than public assistance. Compared to public agen- cies, the women were less aware of nonpublic agencies. Nonpublic agencies created more social stigma, generated greater administrative hassles, were in worse locations, and involved more scheduling difficulties because of limited hours.
elites of a medium-sized city, the members of an organized crime family, persons who sit on the boards of directors of major banks and cor- porations, or people on a college campus who have had sexual relations with each other. The crucial feature is that each person or unit is con- nected with another through a direct or indirect linkage. This does not mean that each person directly knows, interacts with, or is influenced by every other person in the network. Rather, it means that, taken as a whole, with direct and indirect links, they are within an interconnected web of linkages.
We can represent such a network by draw- ing a sociogram —a diagram of circles connected with lines. For example, Sally and Tim do not know each other directly, but each has a good friend, Susan, so they have an indirect connec- tion. All three are part of the same friendship network. The circles represent each person or case, and the lines represent friendship or other linkages (see Figure 2 ).
F I G U R E 2 Sociogram of Friendship Relations
Jian
Anne
Maria Chris
Bill Bob Kim
Pat
Paul
Sally Joyce Peter
Tim
Susan Jorge
Larry Dennis
Edith Mie
the marginal utility, or incremental benefit for additional cases, levels off or drops significantly.
It requires that you continuously evaluate all the collected cases. For example, you locate and plan in-depth interviews with 60 widows who are over 75 years old and who have been widowed for 10 or more years. Depending on your purposes, getting an additional 20 widows whose life experiences, social backgrounds, and worldviews differ little from the first 60 widows may be unnecessary.
PROBABILITY SAMPLING
A specialized vocabulary or jargon has devel- oped around terms used in probability sam- pling. Before examining probability sampling, it is important to review its language.
Populations, Elements, and Sampling Frames
You draw a sample from a larger pool of cases, or elements. A sampling element is the unit of analysis or case in a population. It can be a per- son, a group, an organization, a written docu- ment or symbolic message, or even a social action (e.g., an arrest, a divorce, or a kiss) that is being measured. The large pool is the popula- tion, which has an important role in sampling.
Sometimes, the term universe is used inter- changeably with population. To define the popu- lation, you specify the unit being sampled, the geographical location, and the temporal bound- aries of populations. Consider the examples of populations in Example Box 1. All the examples include the elements to be sampled (e.g., people, businesses, hospital admissions, commercials, etc.) and geographical and time boundaries.
When sampling you start with an idea of the population (e.g., all people in a city) but define it more precisely. The term target population refers to the specific pool of cases that you want to study. The ratio of the size of the sample to Deviant Case Sampling
You might use deviant case sampling (also called extreme case sampling ) when you seek cases that differ from the dominant pattern or have features that differ from the predominant features of most other cases. Similar to purpo- sive sampling, you use a variety of techniques to locate cases with specific characteristics.
Deviant case sampling differs from purposive sampling in that your goal is to locate a collec- tion of unusual, different, or peculiar cases that are not representative of the whole. You select the deviant cases because they are unusual, and you hope to learn more about the social life by considering cases that fall outside the general pattern or including what is beyond the main flow of events.
For example, you are interested in study- ing high school dropouts. Let us say that previous research suggested that a majority of dropouts from low-income, single-parent households have been geographically mobile, and they are disproportionately members of racial minority groups. The parent(s) and siblings tend to be high school dropouts. In addition, dropouts are more likely to have engaged in illegal behavior prior to dropping out. If you used deviant case sampling, you would seek majority-group dropouts without any prior evidence of illegal activities who are from stable two-parent, upper-middle-income families who are geographically stable and well educated.
Sequential Sampling
Sequential sampling is similar to purposive sampling with one difference. In purposive sampling, you try to find as many relevant cases as possible, until time, financial resources, or your energy is exhausted. You want to get every possible case. In sequential sampling, you con- tinue to gather cases until the amount of new information or diversity of cases is filled. In economic terms, you gather information until