Aaa PDF
Aaa PDF
Sampling Designs
ampling is a complex and technical topic. researchers are not concerned with issues of gener-
S Yet at the same time, sampling is familiar to
us all. In the course of our daily activities, we get
alizability but rather with a holistic understanding
of the phenomenon of interest. They make sam-
information, make decisions, and develop predic- pling decisions during the course of data collection
tions through sampling. A nursing student may se- based on informational and theoretical needs, and
lect an elective course by sampling two or three typically do not develop a formal sampling plan in
classes on the first day of the semester. Patients advance. This chapter discusses sampling issues
may generalize about nurses’ friendliness in a par- for both quantitative and qualitative research.
ticular hospital based on the care they received
from a sample of nurses. We all come to conclu-
BASIC SAMPLING
sions about phenomena based on exposure to a lim-
CONCEPTS IN
ited portion of those phenomena.
Q U A N T I TAT I V E S T U D I E S
Researchers, too, usually obtain data from sam-
ples. For example, in testing the efficacy of a new Sampling is a critical part of the design of quantita-
asthma medication, researchers reach conclusions tive research. Let us first consider some terms asso-
without administering the drug to all asthmatic pa- ciated with sampling—terms that are used primarily
tients. Researchers, however, cannot afford to draw (but not exclusively) with quantitative studies.
conclusions about the effectiveness of interventions
or the validity of relationships based on a sample of
Populations
only three or four subjects. The consequences of
making erroneous decisions are more momentous in A population is the entire aggregation of cases in
disciplined inquiries than in private decision making. which a researcher is interested. For instance, if a
Quantitative and qualitative researchers have nurse researcher were studying American nurses
different approaches to sampling. Quantitative re- with doctoral degrees, the population could be de-
searchers seek to select samples that will allow fined as all U.S. citizens who are registered nurses
them to generalize their results to broader groups. (RNs) and who have acquired a Ph.D., D.N.Sc.,
They therefore develop a sampling plan that spec- Ed.D., or other doctoral-level degree. Other possi-
ifies in advance how study participants are to be ble populations might be all male patients who un-
selected and how many to include. Qualitative derwent cardiac surgery in St. Peter’s Hospital
290 ■ PART 3 Designs for Nursing Research
during 2002, all women currently in treatment for foreign students enrolled in American nursing pro-
breast cancer in Boston, or all children in Canada grams qualify? Insofar as possible, the researcher
with cystic fibrosis. As this list illustrates, a popu- must consider the exact criteria by which it could
lation may be broadly defined, involving thousands be decided whether an individual would or would
of individuals, or may be narrowly specified to in- not be classified as a member of the population.
clude only several hundred people. The criteria that specify population characteristics
Populations are not restricted to human sub- are referred to as eligibility criteria or inclusion
jects. A population might consist of all the hospital criteria. Sometimes, a population is defined in
records on file in a particular hospital, all the blood terms of characteristics that people must not pos-
samples taken from clients of a health maintenance sess (i.e., stipulating the exclusion criteria). For
organization, or all the high schools in the United example, the population may be defined to exclude
States with a school-based clinic that dispenses people who cannot speak English.
contraceptives. Whatever the basic unit, the popu- Inclusion or exclusion criteria for a study often
lation always comprises the entire aggregate of el- reflect considerations other than substantive or the-
ements in which the researcher is interested. oretical interests. The eligibility criteria may reflect
As noted in Chapter 9, it is sometimes useful to one or more of the following issues:
make a distinction between target and accessible
populations. The accessible population is the aggre- • Costs. Some criteria result from cost constraints.
gate of cases that conform to the designated criteria For example, when non–English-speaking people
and that are accessible as a pool of subjects for a are excluded, this does not necessarily mean
study. The target population is the aggregate of researchers are not interested in non–English
cases about which the researcher would like to make speakers, but may mean that they cannot afford to
generalizations. A target population might consist of hire translators and multilingual data collectors.
all diabetic people in the United States, but the ac- • Practical concerns. Sometimes, there are other
cessible population might consist of all diabetic peo- practical constraints, such as difficulty in in-
ple who are members of a particular health plan. cluding people from rural areas, people who are
Researchers usually sample from an accessible pop- hearing impaired, and so on.
ulation and hope to generalize to a target population. • People’s ability to participate in a study. The
health condition of some people may preclude
TIP: A serious issue for the development their participation. For example, people with
of an evidence-based practice is information mental impairments, who are in a coma, or who
about the populations on whom research has been are in an unstable medical condition may need
conducted. Many quantitative researchers fail to to be excluded.
identify their target population, or discuss the issue • Design considerations. As noted in Chapter 9, it
of the generalizability of the results. The popula- is sometimes advantageous to define a fairly
tion of interest needs to be carefully considered in homogeneous population as a means of control-
planning and reporting a study. ling extraneous variables.
participants were seniors, and could be either nor- samples, elements are selected by nonrandom
motensive or hypertensive. People were excluded methods. There is no way to estimate the probabil-
if they were taking antihypertensive medications ity that each element has of being included in a
and had not taken their medication that day; had a nonprobability sample, and every element usually
diagnosis of peripheral vascular disease; had lower does not have a chance for inclusion.
leg amputations; had had surgery within the 2 prior
weeks; or could not cross their legs.
Strata
Sometimes, it is useful to think of populations as
Samples and Sampling
consisting of two or more subpopulations, or strata.
Sampling is the process of selecting a portion of A stratum is a mutually exclusive segment of a pop-
the population to represent the entire population. A ulation, established by one or more characteristics.
sample, then, is a subset of population elements. For instance, suppose our population was all RNs
An element is the most basic unit about which in- currently employed in the United States. This pop-
formation is collected. In nursing research, the ele- ulation could be divided into two strata based on
ments are usually humans. gender. Alternatively, we could specify three strata
Samples and sampling plans vary in quality. consisting of nurses younger than 30 years of age,
The overriding consideration in assessing a sample nurses aged 30 to 45 years, and nurses 46 years or
in a quantitative study is its representativeness. A older. Strata are often used in the sample selection
representative sample is one whose key character- process to enhance the sample’s representativeness.
istics closely approximate those of the population. If
the population in a study of blood donors is 50%
Sampling Bias
male and 50% female, then a representative sample
would have a similar gender distribution. If the sam- Researchers work with samples rather than with
ple is not representative of the population, the exter- populations because it is more cost-effective to do
nal validity (generalizability) of the study is at risk. so. Researchers typically have neither the time nor
Unfortunately, there is no way to make sure the resources to study all members of a population.
that a sample is representative without obtaining Furthermore, it is unnecessary to gather data from
the information from the population. Certain sam- an entire population; it is usually possible to obtain
pling procedures are less likely to result in biased reasonably accurate information from a sample.
samples than others, but a representative sample Still, data from samples can lead to erroneous
can never be guaranteed. This may sound discour- conclusions. Finding 100 people willing to partici-
aging, but it must be remembered that researchers pate in a study seldom poses difficulty. It is consid-
operate under conditions in which error is possible. erably more problematic to select 100 subjects who
Quantitative researchers strive to minimize those are not a biased subset of the population. Sampling
errors and, if possible, to estimate their magnitude. bias refers to the systematic over-representation
Sampling designs are classified as either or under-representation of some segment of the
probability sampling or nonprobability sampling. population in terms of a characteristic relevant to
Probability sampling involves random selection the research question.
in choosing the elements. The hallmark of a proba- As an example of consciously biased selection,
bility sample is that researchers can specify the suppose we were investigating patients’ responsive-
probability that each element of the population will ness to nurses’ touch and decide to use as our sam-
be included in the sample. Probability sampling is ple the first 50 patients meeting eligibility criteria in
the more respected of the two approaches because a specific hospital unit. We decide to omit Mr. Z
greater confidence can be placed in the representa- from the sample because he has shown hostility to
tiveness of probability samples. In nonprobability nurses. Mrs. X, who has just lost a spouse, is also
292 ■ PART 3 Designs for Nursing Research
excluded from the study because she is under stress. samples. Despite this fact, most research samples
We have made conscious decisions to exclude cer- in nursing and other disciplines are nonprobability
tain individuals, and the decisions do not reflect samples. Three primary methods of nonprobability
bona fide eligibility criteria. This can lead to bias sampling are convenience, quota, and purposive.
because responsiveness to nurses’ touch (the depen-
dent variable) may be affected by patients’ feelings
Convenience Sampling
about nurses or their emotional state.
Sampling bias usually occurs unconsciously, Convenience sampling entails using the most con-
however. If we were studying nursing students and veniently available people as study participants. A
systematically interviewed every 10th student who faculty member who distributes questionnaires to
entered the nursing library, the sample of students nursing students in a class is using a convenience
would be biased in favor of library-goers, even if sample, or an accidental sample, as it is some-
we are conscientious about including every 10th times called. The nurse who conducts an observa-
entrant regardless of the person’s appearance, gen- tional study of women delivering twins at the local
der, or other traits. hospital is also relying on a convenience sample.
Sampling bias is partly a function of population The problem with convenience sampling is that
homogeneity. If population elements were all iden- available subjects might be atypical of the popula-
tical with respect to key attributes, then any sample tion of interest with regard to critical variables.
would be as good as any other. Indeed, if the popu- Convenience samples do not necessarily com-
lation were completely homogeneous, that is, ex- prise individuals known to the researchers.
hibited no variability at all, then a single element Stopping people at a street corner to conduct an in-
would be a sufficient sample to draw conclusions terview is sampling by convenience. Sometimes,
about the population. For many physiologic attrib- researchers seeking people with certain character-
utes, it may be safe to assume a reasonably high de- istics place an advertisement in a newspaper, put up
gree of homogeneity. For example, the blood in a signs in clinics or supermarkets, or post messages
person’s veins is relatively homogeneous and so a in chat rooms on the Internet. These approaches are
single blood sample chosen haphazardly is ade- subject to bias because people select themselves as
quate. For most human attributes, however, homo- pedestrians on certain streets or as volunteers in re-
geneity is the exception rather than the rule. Age, sponse to posted notices.
health condition, stress, attitudes, habits—all these Snowball sampling (also called network
attributes reflect the heterogeneity of humans. sampling or chain sampling) is a variant of conve-
When variation occurs in the population, then simi- nience sampling. With this approach, early sample
lar variation ideally should be reflected in a sample. members are asked to identify and refer other peo-
ple who meet the eligibility criteria. This method of
TIP: One straightforward way to increase
sampling is often used when the research popula-
the generalizability of a study is to select
tion is people with specific traits who might other-
study participants from two or more sites, such as
wise be difficult to identify (e.g., people who are
from different hospitals, nursing homes, communi-
afraid of hospitals). Snowballing begins with a few
ties, and so on. Ideally, the two different sites
eligible study participants and then continues on the
would be sufficiently divergent that broader repre-
basis of referrals from those participants until the
sentation of the population would be obtained.
desired sample size has been obtained.
Convenience sampling is the weakest form of
NONPROBABILITY
sampling. It is also the most commonly used sam-
SAMPLING
pling method in many disciplines. In heterogeneous
Nonprobability sampling is less likely than probabil- populations, there is no other sampling approach in
ity sampling to produce accurate and representative which the risk of sampling bias is greater.
CHAPTER 13 Sampling Designs ■ 293
Example of a convenience sample: the first two columns. In this example, the conve-
Board and Ryan-Wenger (2002) prospec- nience sample over-represents women and under-
tively examined the long-term effects of the pedi- represents men. We can, however, guide the selec-
atric intensive care unit experience on parents and on tion of study participants so that the sample
family adaptation. The researchers used convenience includes the correct number of cases from both
sampling to recruit three groups of parents: those strata. The far-right panel of Table 13-1 shows the
with a hospitalized child in the pediatric intensive number of men and women required for a quota
care unit, those with a child in a general care unit, sample for this example.
and those with nonhospitalized ill children. If we pursue this same example a bit further, you
may better appreciate the dangers of a biased sample.
Suppose that one of the key questions in this study
Quota Sampling
was, “Would you be willing to work on a unit that
A quota sample is one in which the researcher cared exclusively for AIDS patients?” The percent-
identifies population strata and determines how age of students in the population who would respond
many participants are needed from each stratum. “yes” to this inquiry is shown in the first column of
By using information about population characteris- Table 13-2. Of course, we would not know these val-
tics, researchers can ensure that diverse segments ues; they are displayed to illustrate a point. Within the
are represented in the sample, preferably in the pro- population, men are more likely than women to be
portion in which they occur in the population. willing to work on a unit with AIDS patients, yet men
Suppose we were interested in studying nurs- were seriously under-represented in the convenience
ing students’ attitude toward working with AIDS sample. As a result, there is a discrepancy between
patients. The accessible population is a school of the population and sample values on the outcome
nursing with an undergraduate enrollment of 1000 variable: about 27% more students in the population
students; a sample of 200 students is desired. The are favorable toward working with AIDS victims
easiest procedure would be to use a convenience (14.0%) than we would conclude based on results
sample by distributing questionnaires in classrooms from the convenience sample (11.0%). The quota
or catching students as they enter or leave the sample, on the other hand, does a better job of
library. Suppose, however, we suspect that male and reflecting the viewpoint of the population (14.5%).
female students have different attitudes toward work- In actual research situations, the distortions from a
ing with AIDS victims. A convenience sample might convenience sample may be smaller than in this
result in too many men or women. Table 13-1 pre- example, but could be larger as well.
sents fictitious data showing the gender distribution Quota sampling does not require sophisticated
for the population and for a convenience sample in skills or a lot of effort—and it is surprising that so
TABLE 13.1 Numbers and Percentages of Students in Strata of a Population, Convenience Sample,
and Quota Sample
TABLE 13.2 Students Willing to Work on an AIDS Unit, in the Population, Convenience Sample, and
Quota Sample
Male 55 3 12
Female 85 19 17
Total Number of
Willing Students 140 22 29
Total Number of
All Students 1000 200 200
Percentage Willing 14.0% 11.0% 14.5%
few researchers use this strategy. Many researchers lems, quota sampling represents an important im-
who use a convenience sample could probably de- provement over convenience sampling and should
sign a quota sampling plan, and it would be advan- be considered by quantitative researchers whose
tageous to do so. Stratification should be based on resources prevent the use of probability sampling.
one or more variables that would reflect important
Example of a quota sample:
differences in the dependent variable under study.
Williams, Soetjiningsih, and Williams
Such variables as age, gender, ethnicity, educa-
(2000) studied Balinese mothers’ expectations for
tional attainment, and medical diagnosis are often
children’s development. The researchers used
good stratifying variables.
quota sampling to ensure an equal number of urban
Except for identifying the strata and the de-
and rural Balinese mothers, and an equal number of
sired representation for each, quota sampling is
male and female children.
procedurally similar to convenience sampling. The
subjects in any particular cell constitute, in
Purposive Sampling
essence, a convenience sample from that stratum of
the population. Referring back to the example in Purposive sampling or judgmental sampling is
Table 13-1, the initial sample of 200 students con- based on the belief that researchers’ knowledge
stituted a convenience sample from the population about the population can be used to hand-pick sam-
of 1000. In the quota sample, the 40 men constitute ple members. Researchers might decide purposely
a convenience sample of the 200 men in the popu- to select subjects who are judged to be typical of
lation. Because of this fact, quota sampling shares the population or particularly knowledgeable about
many of the same weaknesses as convenience sam- the issues under study. Sampling in this subjective
pling. For instance, if a researcher is required by a manner, however, provides no external, objective
quota sampling plan to interview 10 men between method for assessing the typicalness of the selected
the ages of 65 and 80 years, a trip to a nursing subjects. Nevertheless, this method can be used to
home might be the most convenient method of ob- advantage in certain situations. Newly developed
taining those subjects. Yet this approach would fail instruments can be effectively pretested and evalu-
to represent the many senior citizens who live in- ated with a purposive sample of diverse types of
dependently in the community. Despite its prob- people. Purposive sampling is often used when
CHAPTER 13 Sampling Designs ■ 295
researchers want a sample of experts, as in the case TIP: If you use a convenience sample, you
of a needs assessment using the key informant ap- can still take steps to enhance the sample’s
proach or in Delphi surveys. Also, as discussed representativeness. First, identify important extra-
later in this chapter, purposive sampling is fre- neous variables—factors that affect variation in the
quently used by qualitative researchers. dependent variable. For example, in a study of the
effect of stress on health, family income would be
Example of purposive sampling:
an important extraneous variable because poor peo-
Friedemann, Montgomery, Rice, and Farrell
ple tend to be less healthy (and more stressed) than
(1999) studied family members’ involvement in the
more affluent ones. Then, decide how to account
nursing home. The first stage of their sampling plan
for this source of variation in the sampling design.
involved purposively sampling 24 nursing homes
One solution is to eliminate variation from extrane-
with a diversity of policies related to family in-
ous variables, as discussed in Chapter 9. In the
volvement, based on a survey of 208 nursing
stress and health example, we might restrict the
homes in southern Michigan. In the second stage,
population to middle-class people. Alternatively,
all family members of residents admitted to these
we could select the convenience sample from two
nursing homes during a 20-month window were in-
communities known to differ socioeconomically so
vited to participate.
that our sample would reflect the experiences of
both lower- and middle-class subjects. This ap-
Evaluation of Nonprobability
proaches using a quota sampling method. In other
Sampling
words, if the population is known to be heteroge-
Although a nonprobability sample is often acceptable neous, you should take steps either to make it more
for pilot, exploratory, or in-depth qualitative research, homogeneous (thereby redefining the population)
for most quantitative studies, the use of nonprobabil- or to capture the full variation in the sample.
ity samples is problematic. Nonprobability samples
are rarely representative of the population. When
PROBABILITY SAMPLING
every element in the population does not have a
chance of being included in the sample, it is likely Probability sampling involves the random selection
that some segment of it will be systematically under- of elements from a population. Random selection
represented. should not be (although it often is) confused with
Why, then, are nonprobability samples used in random assignment, which was described in con-
most nursing studies? Clearly, the advantage of nection with experimental designs in Chapter 8.
these sampling designs lies in their convenience Random assignment refers to the process of allo-
and economy. Probability sampling, discussed cating subjects to different treatment conditions at
next, requires skill and resources. There is often no random. Random assignment has no bearing on
option but to use a nonprobability approach or to how subjects in an experiment were selected in the
abandon the project altogether. Even hard-nosed first place. Random sampling involves a selection
research methodologists would hesitate to advocate process in which each element in the population
the abandonment of an idea in the absence of a ran- has an equal, independent chance of being selected.
dom sample. Quantitative researchers using non- The four most commonly used probability sam-
probability samples out of necessity must be cau- pling methods are simple random, stratified ran-
tious about the inferences and conclusions drawn dom, cluster, and systematic sampling.
from the data. With care in the selection of the sam-
ple, a conservative interpretation of the results, and
Simple Random Sampling
replication of the study with new samples, re-
searchers may find that nonprobability samples Simple random sampling is the most basic prob-
work reasonably well. ability sampling design. Because the more complex
296 ■ PART 3 Designs for Nursing Research
Simple random sampling tends to be labori- eligibility for a particular stratum. In stratified sam-
ous. Developing the sampling frame, numbering all pling, however, a person’s status in a stratum must
the elements, and selecting sample elements are be known before random selection.
time-consuming chores, particularly if the popula- The most common procedure for drawing a
tion is large. Imagine enumerating all the telephone stratified sample is to group together elements
subscribers listed in the New York City telephone belonging to a stratum and to select randomly the de-
directory! If the elements can be arranged in sired number of elements. Researchers can either se-
computer-readable form, then the computer can be lect an equal number of elements from each stratum
programmed to select the sample automatically. In or select unequal numbers, for reasons discussed
actual practice, simple random sampling is not later. To illustrate the procedure used in the simplest
used frequently because it is a relatively inefficient case, suppose that the list in Table 13-3 consisted of
procedure. Furthermore, it is not always possible to 25 men (numbers 1 through 25) and 25 women
get a listing of every element in the population, so (numbers 26 through 50). Using gender as the strati-
other methods may be required. fying variable, we could guarantee a sample of 10
men and 10 women by randomly sampling 10 num-
Example of a simple random sample: bers from the first half of the list and 10 from the sec-
Yoon and Horne (2001) studied the use of ond half. As it turns out, our simple random sampling
herbal products for medicinal purposes in a sample did result in 10 elements being chosen from each half
of older women. A random sample of 86 women of the list, but this was purely by chance. It would not
aged 65 or older who lived independently in a have been unusual to draw, say, 8 names from one
Florida County was selected, using a sampling half and 12 from the other. Stratified sampling can
frame compiled from information from the state guarantee the appropriate representation of different
motor vehicle agency. segments of the population.
Stratifying variables usually divide the popula-
tion into unequal subpopulations. For example,
Stratified Random Sampling
if the person’s race were used to stratify the
In stratified random sampling, the population is population of U. S. citizens, the subpopulation of
first divided into two or more strata. As with quota white people would be larger than that of African-
sampling, the aim of stratified sampling is to en- American and other nonwhite people. The researcher
hance representativeness. Stratified sampling de- might decide to select subjects in proportion to the
signs subdivide the population into homogeneous size of the stratum in the population, using pro-
subsets from which an appropriate number of ele- portionate stratified sampling. If the population
ments are selected at random. was students in a nursing school that had 10%
Stratification is often based on such demo- African-American students, 10% Hispanic stu-
graphic attributes as age, gender, and income level. dents, and 80% white students, then a proportion-
One difficulty is that the stratifying attributes must ate stratified sample of 100 students, with racial/
be known in advance and may not be readily dis- ethnic background as the stratifying variable,
cernible. If you were working with a telephone would consist of 10, 10, and 80 students from the
directory, it would be risky to guess a person’s gen- respective strata.
der, and age, ethnicity, or other personal informa- When researchers are interested in understand-
tion could not be used as stratifying variables. ing differences among strata, proportionate sampling
Patient listings, student rosters, or organizational may result in insufficient numbers for making com-
directories might contain information for a mean- parisons. In the previous example, would the re-
ingful stratification. Quota sampling does not have searcher be justified in drawing conclusions about
the same problem because researchers can ask the characteristics of Hispanic nursing students based
prospective subjects questions that determine their on only 10 cases? It would be unwise to do so. For
298 ■ PART 3 Designs for Nursing Research
this reason, researchers often adopt a disproportion- institution. If personal interviews were involved,
ate sampling design when comparisons are sought the interviewers would have to travel to students
between strata of greatly unequal size. In the exam- scattered throughout the country. Large-scale sur-
ple, the sampling proportions might be altered to veys almost never use simple or stratified random
select 20 African-American students, 20 Hispanic sampling; they usually rely on cluster sampling.
students, and 60 white students. This design would In cluster sampling, there is a successive ran-
ensure a more adequate representation of the two dom sampling of units. The first unit is large group-
racial/ethnic minorities. When disproportionate sam- ings, or clusters. In drawing a sample of nursing
pling is used, however, it is necessary to make an ad- students, we might first draw a random sample of
justment to the data to arrive at the best estimate of nursing schools and then draw a sample of students
overall population values. This adjustment process, from the selected schools. The usual procedure for
known as weighting, is a simple mathematic compu- selecting samples from a general population is to
tation described in textbooks on sampling. sample successively such administrative units as
Stratified random sampling enables researchers states, cities, census tracts, and then households.
to sharpen the precision and representativeness of the Because of the successive stages in cluster sam-
final sample. When it is desirable to obtain reliable in- pling, this approach is often called multistage sam-
formation about subpopulations whose memberships pling. The resulting design is usually described
are relatively small, stratification provides a means of in terms of the number of stages (e.g., three-stage
including a sufficient number of cases in the sample cluster sampling).
by oversampling for that stratum. Stratified sampling, The clusters can be selected either by simple
however, may be impossible if information on the or stratified methods. For instance, in selecting
critical variables is unavailable. Furthermore, a strati- clusters of nursing schools, it may be advisable to
fied sample requires even more labor and effort than stratify on program type. The final selection from
simple random sampling because the sample must be within a cluster may also be performed by simple
drawn from multiple enumerated listings. or stratified random sampling.
For a specified number of cases, cluster sam-
Example of stratified random sampling: pling tends to be less accurate than simple or strati-
Bath, Singleton, Strikas, Stevenson, fied random sampling. Despite this disadvantage,
McDonald, and Williams (2000) conducted a sur- cluster sampling is more economical and practical
vey to determine the extent to which hospitals with than other types of probability sampling, particularly
labor and delivery services had policies about when the population is large and widely dispersed.
screening pregnant women for hepatitis B. A strat-
ified random sample of 968 hospitals (stratified by Example of cluster/multistage sampling:
number of beds and affiliation with a medical Trinkoff, Zhou, Storr, and Soeken (2000)
school) was selected. studied nurses’ substance abuse, using data from a
two-stage cluster sample. In the first stage, 10
states in the United States were selected using a
Cluster Sampling complex stratification procedure. In the second
For many populations, it is impossible to obtain a stage, RNs were selected from each state (a total
listing of all elements. For example, the population sample of 3600) by simple random sampling.
of full-time nursing students in the United States
would be difficult to list and enumerate for the pur-
Systematic Sampling
pose of drawing a simple or stratified random sam-
ple. It might also be prohibitively expensive to The final sampling design can be either probability
sample students in this way because the resulting or nonprobability sampling, depending on the exact
sample would include only one or two students per procedure used. Systematic sampling involves the
CHAPTER 13 Sampling Designs ■ 299
selection of every kth case from a list or group, sampling because the same results are obtained in
such as every 10th person on a patient list or every a more efficient manner. Systematic sampling can
100th person in a directory of American Nurses also be applied to lists that have been stratified.
Association members. Systematic sampling is
Example of a systematic sample:
sometimes used to sample every kth person entering
Tolle, Tilden, Rosenfeld, and Hickman
a bookstore, or passing down the street, or leaving a
(2000) explored barriers to optimal care of the
hospital, and so forth. In such situations, unless the
dying by surveying family members of decedents.
population is narrowly defined as all those people
Their sampling frame was 24,074 death certificates
entering, passing by, or leaving, the sampling is
in Oregon, from which they sampled, through sys-
nonprobability in nature.
tematic sampling, 1458 certificates. They then
Systematic sampling can be applied so that an
traced as many family members of the decedents as
essentially random sample is drawn. If we had a list,
possible and conducted telephone interviews.
or sampling frame, the following procedure could be
adopted. The desired sample size is established at
some number (n). The size of the population must be Evaluation of Probability Sampling
known or estimated (N). By dividing N by n, the
Probability sampling is the only viable method of
sampling interval width (k) is established. The sam-
obtaining representative samples. If all the elements
pling interval is the standard distance between ele-
in the population have an equal probability of being
ments chosen for the sample. For instance, if we were
selected, then the resulting sample is likely to do a
seeking a sample of 200 from a population of 40,000,
good job of representing the population. A further
then our sampling interval would be as follows:
advantage is that probability sampling allows re-
40,000 searchers to estimate the magnitude of sampling
k⫽ ⫽ 200
200 error. Sampling error refers to differences between
population values (such as the average age of the
In other words, every 200th element on the list
population) and sample values (such as the average
would be sampled. The first element should be se-
age of the sample). It is a rare sample that is per-
lected randomly, using a table of random numbers.
fectly representative of a population; probability
Let us say that we randomly selected number 73
sampling permits estimates of the degree of error.
from a table. The people corresponding to numbers
Advanced textbooks on sampling elaborate on pro-
73, 273, 473, 673, and so forth would be sampled.
cedures for making such estimates.
Alternatively, we could randomly select a number
The great drawbacks of probability sampling
from 1 to the number of elements listed on a page,
are its inconvenience and complexity. It is usually
and then randomly select every kth unit on all
beyond the scope of most researchers to sample
pages (e.g., number 38 on every page).
using a probability design, unless the population is
Systematic sampling conducted in this manner
narrowly defined—and if it is narrowly defined,
yields essentially the same results as simple ran-
probability sampling may seem like “overkill.”
dom sampling, but involves far less work.
Probability sampling is the preferred and most re-
Problems would arise if the list were arranged in
spected method of obtaining sample elements, but
such a way that a certain type of element is listed at
it may in some cases be impractical.
intervals coinciding with the sampling interval. For
instance, if every 10th nurse listed in a nursing per- TIP: Whenever possible, it is useful to
sonnel roster were a head nurse and the sampling compare sample characteristics with popula-
interval was 10, then head nurses would either tion characteristics. Published information about
always or never be included in the sample. Problems the characteristics of many groups of interest to
of this type are rare, fortunately. In most cases, sys- nurses may be available to help provide a context
tematic sampling is preferable to simple random for evaluating sampling bias. For example, if you
300 ■ PART 3 Designs for Nursing Research
were studying low-income children in Detroit, you students can read about power analysis in Chapter 20
could obtain information on the Internet about or consult a sampling or statistical textbook.
salient characteristics (e.g., race/ethnicity, age dis- There are no simple formulas that can tell you
tribution) of low-income American children from how large a sample is needed in a given quantitative
the U. S. Bureau of the Census. Population charac- study, but we can offer a simple piece of advice:
teristics could then be compared with sample char- You should use the largest sample possible. The
acteristics, and differences taken into account in larger the sample, the more representative of the
interpreting the findings. population it is likely to be. Every time researchers
calculate a percentage or an average based on sam-
ple data, they are estimating a population value.
SAMPLE SIZE IN
Smaller samples tend to produce less accurate esti-
Q U A N T I TAT I V E S T U D I E S
mates than larger ones. In other words, the larger
Quantitative researchers need to pay careful attention the sample, the smaller the sampling error.
to the number of subjects needed to test research hy- Let us illustrate this with an example of monthly
potheses adequately. A sophisticated procedure known aspirin consumption in a nursing home facility (Table
as power analysis can be used to estimate sample size 13-4). The population consists of 15 residents whose
needs, but some statistical knowledge is needed before aspirin consumption averages 16 aspirins per month,
this procedure can be explained. In this section we as shown in the top row of the table. Eight simple
offer guidelines to beginning researchers; advanced random samples—two each with sample sizes of 2,
2 Sample 1A 6, 14 10.0
2 Sample 1B 20, 28 24.0
3, 5, and 10—have been drawn. Each sample average automobile registrations during a depression year
represents an estimate of the population average (16). when only the well-to-do (who preferred Landon)
Under ordinary circumstances, of course, the popula- had a car or telephone. Thus, a large sample cannot
tion value would be unknown, and we would draw correct for a faulty sampling design.
only one sample. With a sample size of two, our esti- Because practical constraints such as time, sub-
mate might have been wrong by as many as eight as- ject availability, and resources often limit sample
pirins (sample 1B, average of 24), which is a 50% size, many nursing studies are based on relatively
error. As the sample size increases, the averages get small samples. In a survey of nursing studies pub-
closer to the true population value, and the differ- lished over four decades (the 1950s to the 1980s),
ences in the estimates between samples A and B get Brown, Tanner and Padrick (1984) found that the av-
smaller as well. As the sample size increases, the erage sample size was under 100 subjects in all four
probability of getting a markedly deviant sample di- decades, and similar results were reported in a more
minishes. Large samples provide an opportunity to recent analysis (Moody, Wilson, Smyth, Schwartz,
counterbalance atypical values. Unless a power Tittle, & VanCott, 1988). In many cases, a small
analysis can be done, the safest procedure is to obtain sample can lead to misleading or inconclusive results.
data from as large a sample as is practically feasible. Below we discuss some considerations that affect
Large samples are no assurance of accuracy, sample size requirements in quantitative studies.
however. When nonprobability sampling methods are
used, even a large sample can harbor extensive bias.
Homogeneity of the Population
The famous example illustrating this point is the 1936
presidential poll conducted by the magazine Literary If there is reason to believe that the population is
Digest, which predicted that Alfred M. Landon would relatively homogeneous, a small sample may be ad-
defeat Franklin D. Roosevelt by a landslide. About equate. Let us demonstrate that this is so. The top
2.5 million individuals participated in this poll—a half of Table 13-5 presents hypothetical population
substantial sample. Biases resulted from the fact that values for three different populations, with only 10
the sample was drawn from telephone directories and people in each population. These values could
of error and lack precision. When measuring tools ity sampling is an option for selecting a sam-
are imprecise and susceptible to errors, larger sam- ple, that option should be exercised. If you are
ples are needed to test hypotheses adequately. not in a position to do either, we recommend
using as large a sample as possible and taking
IMPLEMENTING A steps to build representativeness into the de-
SAMPLING PLAN IN sign (e.g., by using quota sampling).
Q U A N T I TAT I V E S T U D I E S 4. Recruit the sample. Once the sampling design
has been specified, the next step is to recruit
Once decisions are made about the sampling design prospective study participants according to the
and sample size, the plan must be implemented. plan (after any needed institutional permissions
This section provides some practical information have been obtained) and ask for their coopera-
about implementation of a sampling plan. tion. Issues relating to subject recruitment are
discussed next.
Steps in Sampling in
Quantitative Studies Sample Recruitment
The steps to be undertaken in drawing a sample vary Recruiting subjects to participate in a study involves
somewhat from one sampling design to the next, but two major tasks: identifying eligible candidates and
a general outline of procedures can be described. persuading them to participate. Researchers may in
1. Identify the population. It is good to begin with some cases need to spend time early in the project
a clear idea about the target population to deciding the best sources for recruiting potential
which you would ideally like to be able to gen- participants. Researchers must ask such questions
eralize your results. Unless you have extensive as, Where do people with the characteristics I want
resources, you are unlikely to have access to live or obtain care in large numbers? Will I have di-
the entire target population, and so you will rect access to subjects, or will I need administrative
also need to identify the portion of the target approval? Will there be sufficiently large numbers
population that is accessible to you. Researchers in one location, or will multiple sites be necessary?
often begin by identifying an accessible popu- During the recruitment phase, it may be necessary
lation, and then decide how best to define the to develop a screening instrument, which is a brief
target population. interview or form that allows researchers to deter-
2. Specify the eligibility criteria. The criteria for mine whether a prospective subject meets all eligi-
eligibility in the sample should then be spelled bility criteria for the study.
out. The criteria should be as specific as possible The next task involves actually gaining the co-
with respect to characteristics that might exclude operation of people who have been deemed eligible
potential subjects (e.g., extremes of poor health, for the study. It is critical to have an effective
inability to read English). The criteria might lead recruitment strategy. Most people, given the right
you to redefine your target population. circumstances, will agree to cooperate, but some
3. Specify the sampling plan. Once the accessible are hesitant. Researchers should ask themselves,
population has been identified, you must de- What will make this research experience enjoyable,
cide (a) the method of drawing the sample and worthwhile, convenient, pleasant, and nonthreaten-
(b) how large it will be. Sample size specifica- ing for subjects? Factors over which researchers
tions should consider the aspects of the study have control that can influence the rate of coopera-
discussed in the previous section. If you can tion include the following:
perform a power analysis to determine the • Recruitment method. Face-to-face recruitment
desired number of subjects, it is highly recom- is usually more effective than solicitation by a
mended that you do so. Similarly, if probabil- telephone call or a letter.
304 ■ PART 3 Designs for Nursing Research
• Courtesy. Successful recruitment depends on using it is a good idea to develop contingency plans for
recruiters who are pleasant, courteous, respectful, recruiting more subjects, should the initial plan
and nonthreatening. Cooperation sometimes is en- prove overly optimistic. For example, a contin-
hanced if characteristics of recruiters are similar to gency plan might involve relaxing the eligibility
those of prospective subjects—particularly with criteria, identifying another institution through
regard to gender, race, and ethnicity. which participants could be recruited, offering in-
• Persistence. Although high-pressure tactics are centives to make participation more attractive, or
never acceptable, persistence may sometimes be the lengthening the recruitment period. When con-
needed. When prospective subjects are first ap- tingency plans are developed at the outset, it re-
proached, their initial reaction may be to decline duces the likelihood that you will have to settle for
participation, because they might be taken off a less-than-desirable sample size.
guard. If a person hesitates or gives an equivo-
cal answer at the first attempt, recruiters should
ask if they could come back at a later time. Generalizing From Samples
• Incentives. Gifts and monetary incentives have
Ideally, the sample is representative of the accessi-
been found to increase participation rates.
ble population, and the accessible population is
• Research benefits. The benefits of participating
representative of the target population. By using an
to the individual and to society should be care-
appropriate sampling plan, researchers can be rea-
fully explained, without exaggeration or mis-
sonably confident that the first part of this ideal has
leading information.
been realized. The second part of the ideal entails
• Sharing results. Sometimes it is useful to pro-
greater risk. Are the unemployed nurses in Atlanta
vide people with tangible evidence of their con-
representative of all unemployed nurses in the
tribution to the study by offering to send them a
United States? Researchers must exercise judg-
brief summary of the study results.
ment in assessing the degree of similarity.
• Convenience. Every effort should be made to
The best advice is to be realistic and conserv-
collect data at a time and location that is conve-
ative, and to ask challenging questions: Is it rea-
nient for subjects. In some cases, this may mean
sonable to assume that the accessible population is
making arrangements for transportation or for
representative of the target population? In what
the care of young children.
ways might they differ? How would such differ-
• Endorsements. It may be valuable to have the
ences affect the conclusions? If differences are
study endorsed or acknowledged by a person,
great, it would be prudent to specify a more re-
group, or organization that has prospective sub-
stricted target population to which the findings
jects’ confidence, and to communicate this to
could be meaningfully generalized.
them. Endorsements might come from the insti-
tution serving as the research setting, from a TIP: As you recruit your sample, it is wise
funding agency, or from a respected community to document thoroughly. The more informa-
group or person, such as a church leader. Press tion you have about who the sample is and who it
releases in advance of recruitment are some- is not, the better able you will be to identify poten-
times advantageous. tial biases. Biases can occur even in probability
• Assurances. Prospective subjects should be told sampling because the selection of elements that are
who will see the data, what use will be made of the representative of the population does not guarantee
data, and how confidentiality will be maintained. the participation of all those elements, and refusal
to participate in a study is rarely random. Thus, you
TIP: Subject recruitment often proceeds at should calculate a response rate (the number of
a slower pace than researchers anticipate. people participating in the study relative to the
Once you have determined your sample size needs, number of people sampled) and document the
CHAPTER 13 Sampling Designs ■ 305
visits to community agencies. Thirty-five parent— ticipants through snowballing, but most qualitative
adolescent dyads volunteered. studies eventually evolve to a purposive (or pur-
poseful) sampling strategy—that is, hand-picking
Snowball Sampling cases that will most benefit the study.
Qualitative researchers also use snowball sam- Example of a purposive sample:
pling, asking early informants to make referrals to Gebbie, Wakefield, and Kerfoot (2000) pur-
other study participants. This method is sometimes posefully selected 27 American nurses currently
referred to as nominated sampling because it re- active in public health policy to describe their ex-
lies on the nominations of others already in the periences in policy development.
sample. Researchers may use this method to gain
access to people who are difficult to identify. In purposive sampling, several strategies have
Snowball sampling has distinct advantages over been identified (Patton, 2002), only some of which
convenience sampling. The first is that it may be are mentioned here. Note that researchers them-
more cost-efficient and practical. Researchers may selves do not necessarily refer to their sampling
spend less time screening people to determine if plans with Patton’s labels; his classification shows
they are appropriate for the study, for example. the kind of diverse strategies qualitative researchers
Furthermore, with an introduction from the refer- have adopted to meet the theoretical needs of their
ring person, researchers may have an easier time research:
establishing a trusting relationship with new partic- • Maximum variation sampling involves pur-
ipants. Finally, researchers can more readily specify posefully selecting cases with a wide range of
the characteristics that they want new participants variation on dimensions of interest. By selecting
to have. For example, in the study of people with participants with diverse views and perspectives,
nightmares, we could ask early respondents if they researchers invite challenges to preconceived or
knew anyone else who had the same problem and emerging conceptualizations. Maximum varia-
who was articulate. We could also ask for referrals to tion sampling might involve ensuring that people
people who would add other dimensions to the sam- with diverse backgrounds are represented in the
ple, such as people who vary in age, race, socioeco- sample (ensuring that there are men and women,
nomic status, and so on. poor and affluent people, and so on). It might
A weakness of this approach is that the even- also involve deliberate attempts to include peo-
tual sample might be restricted to a rather small ple with different viewpoints about the phenom-
network of acquaintances. Moreover, the quality of enon under study. For example, researchers
the referrals may be affected by whether the refer- might use snowballing to ask early participants
ring sample member trusted the researcher and for referrals to people who hold different points
truly wanted to cooperate. of view.
• Homogeneous sampling deliberately reduces
Example of a snowball sample: variation and permits a more focused inquiry.
Meadows, Thurston, and Berenson (2001) Researchers may use this approach if they wish
studied the messages that rural midlife women get to understand a particular group of people espe-
about preventive health care. Study participants cially well. Homogeneous sampling is often
were recruited through convenience sampling at used to select people for group interviews.
first, and subsequently through snowball sampling. • Extreme (deviant) case sampling provides op-
A sample of 24 midlife women were interviewed. portunities for learning from the most unusual
and extreme informants (e.g., outstanding suc-
Purposive Sampling cesses and notable failures). The assumption
Qualitative sampling may begin with volunteer in- underlying this approach is that extreme cases
formants and may be supplemented with new par- are rich in information because they are special
CHAPTER 13 Sampling Designs ■ 307
in some way. In some cases, more can be learned fit and serve to challenge researchers’ interpre-
by intensively studying extreme cases, but ex- tations. These “negative” cases may offer new
treme cases can also distort understanding of a insights about how the original conceptualiza-
phenomenon. tion needs to be revised or expanded.
• Intensity sampling is similar to extreme case
It is important to note that almost all of these
sampling, but with less emphasis on the extremes.
sampling strategies require that researchers have
Intensity samples involve information-rich cases
some knowledge about the setting in which the
that manifest the phenomenon of interest in-
study is taking place. For example, to choose ex-
tensely, but not as extreme or potentially distort-
treme cases, typical cases, or critical cases, re-
ing manifestations. Thus, the goal in intensity
searchers must have information about the range of
sampling is to select rich cases that offer strong
variation of the phenomenon and how it manifests
examples of the phenomenon.
itself. Early participants may be helpful in imple-
• Typical case sampling involves the selection of
menting these sampling strategies.
participants who illustrate or highlight what is
typical or average. The resulting information
Theoretical Sampling
can be used to create a qualitative profile illus-
The method of sampling used in grounded theory
trating typical manifestations of the phenome-
is called theoretical sampling. Glaser (1978,
non being studied.
p. 36) defined this sampling as “the process of
• Critical case sampling involves selecting im-
data collection for generating theory whereby the
portant cases regarding the phenomenon of in-
analyst jointly collects, codes, and analyzes his
terest. With this approach, researchers look for
data and decides what data to collect next and
the particularly good story that illuminates crit-
where to find them, in order to develop his theory
ical aspects of the phenomenon.
as it emerges.” The process of theoretical sam-
• Criterion sampling involves studying cases
pling is controlled by the developing grounded
that meet a predetermined criterion of impor-
theory. Theoretical sampling is not envisioned as a
tance. Criterion sampling is sometimes used in
single, unidirectional line. This complex sampling
multimethod studies in which data from the
technique requires researchers to be involved with
quantitative component are used to select cases
multiple lines and directions as they go back and
meeting certain criteria for in-depth study.
forth between data and categories as the theory
Sandelowski (2000) offers a number of helpful
emerges.
suggestions for combining sampling strategies
Glaser stressed that theoretical sampling is not
in mixed-method research.
the same as purposive sampling. Theoretical sam-
• Theory-based sampling involves the selection
pling’s purpose is to discover categories and their
of people or incidents on the basis of their po-
properties and to offer interrelationships that occur
tential representation of important theoretical
in the substantive theory. “The basic question in
constructs. Theory-based sampling is a very fo-
theoretical sampling is: what groups or subgroups
cused approach that is usually based on an a pri-
does one turn to next in data collection?” (Glaser,
ori theory that is being examined qualitatively.
1978, p. 36). These groups are not chosen before
• Sampling confirming and disconfirming cases
the research begins but only as they are needed for
is often used toward the end of data collection in
their theoretical relevance for developing further
qualitative studies. As researchers note trends
emerging categories.
and patterns in the data, emerging conceptual-
izations need to be checked. Confirming cases Example of a theoretical sampling:
are additional cases that fit researchers’ concep- Beck (2002) used theoretical sampling in
tualizations and offer enhanced credibility. her grounded theory study of mothering twins dur-
Disconfirming cases are examples that do not ing the first year of life, in which 16 mothers of
308 ■ PART 3 Designs for Nursing Research
twins were interviewed in their homes. A specific funding for a project. Patton (2002) recommends
example of theoretical sampling concerned what specifying minimum samples that would reasonably
the mothers kept referring to as the “blur period”— be adequate for understanding the phenomenon.
the first few months of caring for the twins. Additional cases can then be added, as necessary, to
Initially, Beck interviewed mothers whose twins achieve saturation.
were around 1 year of age. Her rationale was that
these mothers would be able to reflect back over
Evaluating Qualitative Samples
the entire first year of mothering the multiples.
When these mothers referred to the “blur period,” In a qualitative study, the sampling plan is evaluated
Beck asked them to describe this period more fully. in terms of adequacy and appropriateness (Morse,
The mothers said they could not provide many de- 1991). Adequacy refers to the sufficiency and qual-
tails about this period because it was “such a blur!” ity of the data the sample yields. An adequate sam-
Beck then chose to interview mothers whose twins ple provides researchers with data without any
were 3 months of age or younger, to ensure that “thin” spots. When researchers have truly attained
mothers were still immersed in the “blur period” saturation, informational adequacy has been
and would be able to provide rich detail about what achieved and the resulting description or theory is
this phase of mothering twins was like. richly textured and complete. Appropriateness
concerns the methods used to select the sample.
Sample Size in Qualitative Research An appropriate sample is one resulting from the
identification and use of participants who can best
There are no criteria or rules for sample size in
supply information according to the conceptual
qualitative research. Sample size is largely a func-
requirements of the study. For example, a sam-
tion of the purpose of the inquiry, the quality of the
pling plan that does not include disconfirming
informants, and the type of sampling strategy used.
cases may not meet the information needs of the
For example, a larger sample is likely to be needed
research.
with maximum variation sampling than with typi-
cal case sampling. TIP: No matter what type of qualitative
In qualitative studies, sample size should be de- sampling you use, you should keep a journal
termined based on informational needs. Hence, a or notebook to jot down ideas and reminders re-
guiding principle in sampling is data saturation— garding the sampling process (e.g., who you should
that is, sampling to the point at which no new infor- interview next). Memos to yourself will help you
mation is obtained and redundancy is achieved. remember valuable ideas about your sample.
Morse (2000) has noted that the number of partici-
pants needed to reach saturation depends on a num-
ber of factors. For example, the broader the scope of Sampling in the Three Main
the research question, the more participants will Qualitative Traditions
likely be needed. Data quality can also affect sample
There are similarities among the various qualitative
size. If participants are good informants who are able
traditions with regard to sampling: samples are
to reflect on their experiences and communicate ef-
usually small, probability sampling is almost never
fectively, saturation can be achieved with a relatively
used, and final sampling decisions usually take
small sample. Also, if longitudinal data are collected,
place in the field during data collection. However,
fewer participants may be needed, because each will
there are some differences as well.
provide a greater amount of information.
TIP: Sample size ambiguities sometimes Sampling in Ethnography
create practical dilemmas when you are Ethnographers may begin by initially adopting a
planning a study, or if you are seeking approval or “big net” approach—that is, mingling with and
CHAPTER 13 Sampling Designs ■ 309
nonprobability designs and permit estimates of • Other types of qualitative sampling include ho-
the magnitude of sampling error. mogeneous sampling (deliberately reducing
• Simple random sampling involves the random variation); extreme case sampling (selecting
selection of elements from a sampling frame the most unusual or extreme cases); intensity
that enumerates all the elements; stratified ran- sampling (selecting cases that are intense but
dom sampling divides the population into ho- not extreme); typical case sampling (selecting
mogeneous subgroups from which elements are cases that illustrate what is typical); critical
selected at random. case sampling (selecting cases that are espe-
• Cluster sampling (or multistage sampling) cially important or illustrative); and theory-
involves the successive selection of random based sampling (selecting cases on the basis of
samples from larger to smaller units by either their representation of important constructs).
simple random or stratified random methods. • Samples in qualitative studies are typically
• Systematic sampling is the selection of every small and based on information needs. A guid-
kth case from a list. By dividing the population ing principle is data saturation, which involves
size by the desired sample size, the researcher sampling to the point at which no new informa-
establishes the sampling interval, which is the tion is obtained and redundancy is achieved.
standard distance between the selected ele- • Criteria for evaluating qualitative sampling are
ments. informational adequacy and appropriateness.
• In quantitative studies, researchers should ide- • Ethnographers make numerous sampling deci-
ally use a power analysis to estimate sample sions, including not only whom to sample but
size needs. Large samples are preferable to what to sample (e.g., activities, events, docu-
small ones because larger samples tend to be ments, artifacts); decision making is often aided
more representative, but even a large sample by their key informants who serve as guides
does not guarantee representativeness. and interpreters of the culture.
• Qualitative researchers use the theoretical • Phenomenologists typically work with a small
demands of the study to select articulate and sample of people (10 or fewer) who meet the
reflective informants with certain types of expe- criterion of having lived the experience under
rience in an emergent way, capitalizing on study.
early learning to guide subsequent sampling • Grounded theory researchers typically use theo-
decisions. retical sampling and work with samples of
• Qualitative researchers most often use purposive about 20 to 30 people.
or, in grounded theory studies, theoretical sam-
pling to guide them in selecting data sources
that maximize information richness. STUDY ACTIVITIES
• Various purposive sampling strategies have been
Chapter 13 of the Study Guide to Accompany
used by qualitative researchers. One strategy is
Nursing Research: Principles and Methods, 7th
maximum variation sampling, which entails
edition, offers various exercises and study sugges-
purposely selecting cases with a wide range of
tions for reinforcing concepts presented in this
variation.
chapter. In addition, the following study questions
• Criterion sampling involves studying cases
can be addressed:
that meet a predetermined criterion of impor-
tance. 1. Draw a simple random sample of 15 people from
• Another important strategy is sampling con- the sampling frame of Table 13-3, using the
firming and disconfirming cases, that is, se- table of random numbers that appears in Table
lecting cases that enrich and challenge the re- 8-2 on page 171. Begin your selection by blindly
searchers’ conceptualizations. placing your finger at some point on the table.
CHAPTER 13 Sampling Designs ■ 313
2. Suppose you have decided to use a systematic Diekmann, J. M., & Smith, J. M. (1989). Strategies for
sampling design for a research project. The accessment and recruitment of subjects for nursing
known population size is 5000, and the sample research. Western Journal of Nursing Research, 11,
size desired is 250. What is the sampling inter- 418–430.
Field, P. A. (1991). Doing fieldwork in your own culture.
val? If the first element selected is 23, what
In J. M. Morse (Ed.), Qualitative nursing research:
would be the second, third, and fourth ele-
A contemporary dialogue. Newbury Park, CA:
ments selected? Sage.
3. Suppose you were interested in studying the at- Glaser, B. G. (1978). Theoretical sensitivity. Mill Valley,
titude of clinical specialists toward autonomy CA: Sociology Press.
in work situations. Suggest a possible target Glesne, C., & Peshkin, A. (1992). Becoming qualitative
and accessible population. What strata might researchers: An introduction. White Plains, NY:
be identified if quota sampling were used? Longman.
4. What type of sampling design was used to ob- Kish, L. (1965). Survey sampling. New York: John Wiley
tain the following samples? & Sons.
a. 25 experts in critical care nursing Levey, P. S., & Lemeshow, S. (1980). Sampling for health
b. 60 couples attending a particular prenatal professionals. New York: Lifetime Learning.
MacDougall, C., & Fudge, E. (2001). Planning and re-
class
cruiting the sample for focus groups and in-depth in-
c. 100 nurses from a list of nurses registered terviews. Qualitative Health Research, 11, 117–126.
in the state of Pennsylvania, using a table of Moody, L. E., Wilson, M. E., Smyth, K., Schwartz, R.,
random numbers Tittle, M., & VanCott, M. L. (1988). Analysis of a
d. 20 adult patients randomly selected from a decade of nursing practice research: 1977–1986.
random selection of 10 hospitals located in Nursing Research, 37, 374–379.
one state Morse, J. M. (1991). Strategies for sampling. In J. M.
e. Every fifth article published in Nursing Morse (Ed.), Qualitative nursing research: A contem-
Research during the 1980s, beginning with porary dialogue. Newbury Park, CA: Sage.
the first article. Morse, J. M. (2000). Determining sample size.
5. Suppose you wanted to study the experiences Qualitative Health Research, 10, 3–5.
Patton, M. Q. (2002). Qualitative evaluation and re-
of nursing students during their first clinical
search methods (3rd ed.). Thousand Oaks, CA: Sage.
assignment. Describe what you would need to
Polit, D. F., & Sherman, R. (1990). Statistical power
do to select a sample using maximum variation analysis in nursing research. Nursing Research, 39,
sampling, critical case sampling, typical case 365–369.
sampling, and homogeneous sampling. Porter, E. J. (1999). Defining the eligible, accessible pop-
ulation for a phenomenological study. Western
Journal of Nursing Research, 21, 796–804.
SUGGESTED READINGS Sandelowski, M. (1995). Sample size in qualitative re-
search. Research in Nursing & Health, 18, 179–183.
Methodologic References
Sandelowski, M. (2000). Combining qualitative and
Babbie, E. (1990). Survey research methods (2nd ed.). quantitative sampling, data collection, and analysis
Belmont, CA: Wadsworth. techniques in mixed-method studies. Research in
Brown, J. S., Tanner, C. A., & Padrick, K. P. (1984). Nursing & Health, 23, 246–255.
Nursing’s search for scientific knowledge. Nursing Sudman, S. (1976). Applied sampling. New York:
Research, 33, 26–32. Academic Press.
Cochran, W. G. (1977). Sampling techniques (3rd ed.). Trinkoff, A. M., & Storr, C. L. (1997). Incorporating
New York: John Wiley & Sons. auxiliary variables into probability sampling designs.
Cohen, J. (1977). Statistical power analysis for the be- Nursing Research, 46, 182–185.
havioral sciences (rev. ed.). New York: Academic Williams, B. (1978). A sampler on sampling. New York:
Press. John Wiley & Sons.
314 ■ PART 3 Designs for Nursing Research