NUR/518 – Analysis of Research Reports

University of Phoenix

Burns & Grove: The Practice of Nursing
Research: Conduct, Critique, and
Utilization, 4th ed.
Copyright 2001, W.B. Saunders, an
Elsevier Imprint.
39777 Burns (saunm) INTERACTIVE RIGHT

top of rh
base of rh

top of txt
base of txt

C
. . .H. . A
..P
. .T 11
. . E. .R. . . . . . . .

SELECTING A RESEARCH
DESIGN

T
he purpose of a design is to achieve greater Designs have been developed by researchers to
control and thus improve the validity of the meet unique research needs as they emerged. The
study in examining the research problem. De- first experimental designs were developed in the
termining the appropriate research design for a 1930s by Sir Ronald A. Fisher (1935) and pub-
study requires the integration of many elements. lished in a book entitled The Design of Experi-
The individual critiquing a nursing study is con- ments. However, most work on design has been
fronted with a similar dilemma. Many published conducted since the 1970s (Abdellah & Levine,
studies do not identify the design used in the study. 1979; Anderson & McLean, 1974; Cook & Camp-
Identifying the design may require putting together bell, 1979; Cox, 1958; Trochim, 1986). During this
bits of information from various parts of the re- time, designs have become much more sophisticated
search report. The questions given at the beginning and varied. There is no universal standard for cate-
of Chapter 10 will help you select a design or gorizing designs. Names of designs change as they
identify the design of a study for a critique. are discussed by various authors. Researchers some-
This chapter describes the designs most com- times merge elements of several designs to meet the
monly used in nursing research, using the categories research needs of a particular study. From these
described in Chapter 3 — descriptive, correlational, developments, new designs sometimes emerge.
quasi-experimental, and experimental. Descriptive
Originally, only experimental designs were con-
and correlational designs examine variables in natu-
sidered of value. In addition, many believe that the
ral environments and do not include researcher-
only setting in which an experiment can be con-
designed treatments. Quasi-experimental and ex-
perimental designs examine the effects of an ducted is a laboratory, in which much stricter con-
intervention by comparing differences between trols can be maintained than in a field or natural
groups that have received the intervention and those setting. This approach is appropriate for the natural
that have not received the intervention. As you re- sciences but not for the social sciences. From the
view each design, note the threats to validity con- social sciences have emerged additional quantitative
trolled by the design, keeping in mind that uncon- designs (descriptive, correlational, and quasi-experi-
trolled threats in the design you choose may mental) and qualitative designs.
weaken the validity of your study. Table 11 – 1 lists At present, nurse researchers are using designs
the designs discussed in this chapter. After the de- developed in other disciplines, such as psychology,
scriptions of the designs, a series of decision trees that meet the needs of that discipline. Will these
are provided to assist the researcher in the selection designs be effective in adding to nursing’s knowl-
of the appropriate design or the reader in the identi- edge base? These designs are a useful starting short
fication of the design used in a published study. point, but nurse scientists must go beyond these standard
247 long
39777 Burns (saunm) INTERACTIVE LEFT

top of rh
248 UNIT 2 ■ THE RESEARCH PROCESS base of rh

top of txt
TABLE 11 – 1 intervention research design described in Chapter base of txt
RESEARCH DESIGNS 13. Developing designs to study the outcomes of
Descriptive designs
nursing actions is also important. This emerging
Typical descriptive study design field of research in nursing is described in Chapter
Comparative descriptive design 12. The use of time-series analysis strategies (de-
Time dimensional designs
Longitudinal designs
scribed later in the chapter) holds great promise for
Cross-sectional designs examining important dimensions of nursing. Nurse
Trend designs researchers must see themselves as credible scien-
Event-partitioning designs
Case study design
tists to dare to develop new design strategies that
Correlational designs facilitate examination of little-understood aspects of
Descriptive correlational design nursing. Developing a new design requires careful
Predictive design
Model-testing designs
consideration of possible threats to validity and
Quasi-experimental designs ways to diminish them. It also requires a willing-
Nonequivalent control group designs ness to risk the temporary failures that are always
The one-group posttest– only design
The posttest– only design with nonequivalent groups
inherent in the development of something new.
The one-group pretest– posttest design
The untreated control group design with pretest and posttest
The removed-treatment design with pretest and posttest DESCRIPTIVE STUDY DESIGNS
The reversed-treatment nonequivalent control group design
with pretest and posttest Descriptive studies (see Table 11 – 1) are de-
Interrupted time-series designs signed to gain more information about characteris-
Simple interrupted time series
Interrupted time series with a nonequivalent no-treatment
tics within a particular field of study. Their purpose
control group time series is to provide a picture of situations as they naturally
Interrupted time series with multiple replications happen. In many aspects of nursing, there is a need
Experimental study designs
Pretest– posttest control group design
for a clearer delineation of the phenomenon before
Posttest– only control group design causality can be examined. A descriptive design
Randomized block design may be used for the purpose of developing theory,
Factorial designs
Nested designs
identifying problems with current practice, justify-
Crossover or counterbalanced designs ing current practice, making judgments, or deter-
Randomized clinical trials mining what others in similar situations are doing
(Waltz & Bausell, 1981). No manipulation of varia-
bles is involved. Dependent and independent varia-
bles should not be used within a descriptive design,
designs to develop designs that will more appropri- because the design involves no attempt to establish
ately meet the needs of nursing’s knowledge base. causality.
To go beyond the current designs, nurse scientists Descriptive designs vary in levels of complexity.
must have a working knowledge of available de- Some contain only two variables, whereas others
signs and of the logic on which they are based. may have multiple variables. The relationships
Designs created to meet nursing needs should be among variables are identified to obtain an overall
congruent with nursing philosophy. They must pro- picture of the phenomenon being examined, but ex-
vide a means to examine dimensions of nursing amination of types and degrees of relationships is
within a holistic framework and must allow exami- not the primary purpose of a descriptive study. Pro-
nation of nursing dimensions over time. Designs tection against bias in a descriptive design is
must be developed that can seek answers to impor- achieved through (1) linkages between conceptual
tant nursing questions rather than answering only and operational definitions of variables, (2) sample
questions that can be examined by existing designs. selection and size, (3) the use of valid and reliable
Innovative design strategies are beginning to ap- instruments, and (4) data collection procedures that short
pear within nursing research. One example is the achieve some environmental control. standard
long
39777 Burns (saunm) INTERACTIVE RIGHT

top of rh
CHAPTER 11 ■ Selecting a Research Design 249 base of rh

top of txt
Typical Descriptive Study Design marked on a 100-mm Visual Analogue Scale.” base of txt
(Mimnaugh et al., 1999, p. 78)
The most commonly used descriptive design is
...................................
presented in Figure 11 – 1. The design is used to
“The following research questions were ad-
examine characteristics of a single sample. It in-
dressed:
volves identification of a phenomenon of interest
and of the variables within the phenomenon, devel- 1. What sensations do patients experience when
opment of conceptual and operational definitions of abdominal JP tubes are removed?
the variables, and description of the variables. The 2. What sensations do patients experience when
description of the variables leads to an interpreta- CTs are removed?
tion of the theoretical meaning of the findings and 3. What is the intensity of the sensations patients
provides knowledge of the variables and the study experience when abdominal and chest tubes
population that can be used for future research in are removed?
the area. 4. What factors affect the perception and inten-
Very few studies use a typical descriptive design; sity of sensations that are experienced?
however, many contain descriptive components. An 5. What types of information are patients com-
example of a descriptive design is Mimnaugh and monly given before tube removal?
colleagues’ (1999) study of sensations experienced 6. What types of information do patients indi-
during removal of tubes in acute postoperative pa- cate that they would like to receive before
tients. The following excerpt describes the design of removal of abdominal or chest tubes?” (Mim-
their study. naugh et al., 1999, p. 79) ■
...................................
“The major purpose of this study was to deter-
Comparative Descriptive Design
mine the types and intensity of sensations that pa-
tients experience when chest tubes (CTs) and Jack- The comparative descriptive design (see Figure
son-Pratt (JP) abdominal tubes are removed. A 11 – 2) examines and describes differences in varia-
convenience sample of 62 hospitalized subjects, 31 bles in two or more groups that occur naturally in
with CTs and 31 with JP tubes, participated. Each the setting. Descriptive statistics and inferential sta-
subject was interviewed after tube removal. Sensa- tistical analyses may be used to examine differences
tions were identified, and intensity of sensation was between or among groups. Commonly, the results

CLARIFICATION MEASUREMENT DESCRIPTION INTERPRETATION

description
variable 1
of variable 1

description
variable 2
of variable 2
Phenomenon Interpretation
of Interest of meaning
description
variable 3
of variable 3

description Development
variable 4 of
of variable 4
hypotheses

short
FIGURE 11 – 1 ■ Descriptive study design. standard
long
39777 Burns (saunm) INTERACTIVE LEFT

top of rh
250 UNIT 2 ■ THE RESEARCH PROCESS base of rh

top of txt
base of txt
Group I
(variable[s] Describe
measured)

Comparison
Interpretation
of groups on
of meaning
select variables

Group II Development
(variable[s] Describe of
measured) hypotheses

FIGURE 11 – 2 ■ Comparative descriptive design.

obtained from these analyses are not generalized to into a treatment protocol, point of entry into a new
a population. An example of this design is the study lifestyle, and age at which the subject started smok-
by Williams and associates (1999) of fatigue in ing. An understanding of temporal sequencing is an
mothers of infants discharged home with apnea important prerequisite to examining causality be-
monitors. The following extract describes the study. tween variables. Thus, the results of these designs
lead to the development of hypotheses and are often
................................... forerunners of experimental designs.
“A comparative study was done to determine Epidemiological studies that use time dimen-
differences in caregiver fatigue between two groups sional designs are developed to determine the risk
of mothers of preterm infants at baseline, in the factors or causal factors of illness states. Cause de-
hospital (Time 1), 1 week postdischarge (Time 2), termined in this manner is called inferred causality.
and 1 month postdischarge (Time 3). Group 1 in- The best-known studies in this area are those on
fants were discharged home on apnea monitors smoking and cancer. Because of the strength of
(AM) (n ⫽ 28), and Group 2 infants were not on studies that have undergone multiple repetitions, the
apnea monitors (nonAM) (n ⫽ 46).” (Williams causal link is strong. The strategy is not as power-
et al., 1999, p. 69) ■ ful as experimental designs in supporting causality;
however, in this situation, as in many others, one
can never ethically conduct a true experiment.
Epidemiologists use two strategies to examine a
Time Dimensional Designs situation over time, retrospective studies and pro-
Time dimensional designs were developed within spective studies. The norm in epidemiological stud-
the discipline of epidemiology, in which the occur- ies is to use cohorts to refer to groups of subjects
rence and distribution of disease among populations in prospective studies, but the term is generally not
are studied. These designs examine sequences and used in retrospective studies. In retrospective stud-
patterns of change, growth, or trends over time. The ies, both the proposed cause and the proposed effect
dimension of time, then, becomes an important fac- have already occurred. For example, the subjects
tor. Within the field of epidemiology, the samples could have a specific type of cancer, and the re-
in time dimensional studies are called cohorts. Orig- searcher could be searching for commonalities
inally, cohorts were age categories; however, the among subjects that may have led to the develop-
concept has been expanded to apply to groups dis- ment of that type of cancer. In a prospective study,
tinguished by many other variables. Other means of causes may have occurred, but the proposed effect
classifying populations that have relevance in rela- has not. short
tion to time are time of diagnosis, point of entry The Framingham study is the best-known exam- standard
long
39777 Burns (saunm) INTERACTIVE RIGHT

top of rh
CHAPTER 11 ■ Selecting a Research Design 251 base of rh

top of txt
ple of a prospective study (U.S. Department of given for the points of time selected for measure- base of txt
Health and Human Services, 1968). In this study, ment. There is often a bias in selection of subjects
members of a community were monitored for 20 because of the requirement for a long-term commit-
years by researchers who examined variables such ment. In addition, loss of subjects (mortality) can be
as dietary patterns, exercise, weight, and blood lipid high and can decrease the validity of findings.
levels. As the subjects experienced illnesses, such as Power analysis must be calculated according to
heart disease, hypertension, or lung disease, their the number of subjects expected to complete the
illnesses could be related to previously identified study, not the number recruited initially. The re-
variables. searcher must invest considerable energy in devel-
Prospective studies are considered more powerful oping effective strategies to maintain the sample;
than retrospective studies in inferring causality, be- some strategies for this purpose are discussed in
cause it can be demonstrated that the risk factors Chapter 10. The period during which subjects will
occurred before the illness and are positively related be recruited into the study must be carefully
to the illness. Both designs are important for use in planned, and a time line depicting data collection
nursing studies, because a person’s responses to points for each subject must be developed to enable
health situations are patterns that developed long planning for the numbers and availability of data
before the health situation occurred (Newman, collectors. If this issue is not carefully thought out,
1986). These patterns then influence responses to data collectors may be confronted with the need to
nursing interventions. continue to recruit new subjects while they are at-
Several designs are used to conduct time dimen- tempting to collect data scheduled for subjects re-
sional studies: longitudinal, cross-sectional, trend, cruited earlier. The researcher must also decide
and treatment partitioning. whether all data from a particular subject will be
collected by a single data collector or whether a
LONGITUDINAL DESIGNS different data collector will be used at each point to
ensure that data are collected blindly.
Longitudinal designs examine changes in the Because of the large volumes of data acquired in
same subjects over an extended period. This design a longitudinal study, careful attention must be given
is sometimes called a panel design (see Figure to strategies for managing the data. The repetition
11 – 3). Longitudinal designs are expensive and re- of measures requires that data analysis be carefully
quire a long period of researcher and subject com- thought through. Analyses commonly used are re-
mitment. The area to be studied, the variables, and peated measures analyses of variance, multivariate
their measurement must be clearly identified before analyses of variance (MANOVA), regression ana-
data collection begins. Measurement must be care- lysis, cluster analysis, and time-series analysis (Bar-
fully planned and implemented because the mea- nard et al., 1987).
sures will be used repeatedly over time. If children An example of a longitudinal design is the study
are being studied, the measures must be valid for by Martinson and colleagues (1999), “Comparison
all the ages being studied. The researcher must be of Chinese and Caucasian Families Caregiving to
familiar with how the construct being measured is Children with Cancer at Home.” An abstract of that
patterned over time, and clear rationale must be study follows.

Time 1 Time 2 Time 3 Time 4 Time..n

measure measure measure measure measure
FIGURE 11 – 3 ■ Longitudinal design. variables variables variables variables variables

Sample 1 Sample 1 Sample 1 Sample 1 Sample 1 short
standard
long
39777 Burns (saunm) INTERACTIVE LEFT

top of rh
252 UNIT 2 ■ THE RESEARCH PROCESS base of rh

................................... top of txt
base of txt
“An exploratory, longitudinal design was used to Time 1
compare the caregiving patterns and practices that measure
occur within the family in Chinese and matched variables
Caucasian families who have a child newly diag-
Sample 1
nosed with cancer. Both quantitative and qualita- Immediate Loss
tive methods were used to collect and analyze the
data. . . . A total of 18 families participated in the Time 1
study. Group A consisted of 19 Chinese families measure
having a child newly diagnosed with cancer. Two variables
of the Chinese children could not be matched dur-
Sample 2
ing the span of the study. Group B consisted of Loss 6 mo
eight Caucasian families having a child matched for
diagnosis, age, and sex with a child in group A. Time 1
Five Chinese families from Taiwan and five from measure
Mainland China participated. . . . Data were col- variables
lected at three points: (1) as soon as the Chinese Sample 3
child living in North American was diagnosed and Loss 1 yr
the interview arrangements were able to be made, Time 1
(2) after remission or 4–6 months after diagnosis,
and (3) 1 year after diagnosis.” (Martinson et al., measure
variables
1999, pp. 101–103) ■
Sample 4
Loss 2 yr
CROSS-SECTIONAL DESIGNS
Cross-sectional designs are used to examine Time 1
groups of subjects in various stages of development measure
simultaneously (see Figure 11 – 4). The assumption variables
is that the stages are part of a process that will Sample 5
progress over time. Selecting subjects at various Loss 5 yr
points in the process provides important information
about the totality of the process, even though the FIGURE 11 – 4 ■ Cross-sectional study design.

same subjects are not monitored through the entire
process. The processes of development selected for
the study might be related to age, position in an All these groups could be studied during one pe-
educational system, growth pattern, or stages of riod, but a pattern of grief reactions over a 5-year
maturation or personal growth (if they could be period could be described. The design is not as
clearly enough defined to develop criteria for inclu- strong as the longitudinal design but allows some
sion within differentiated groups). Subjects are then understanding of the phenomenon over time when
categorized by group, and data on the selected vari- time allowed for the study is limited.
ables are collected at a single point in time. Aaronson and Kingry (1988) describe a strategy
For example, one might wish to study grief reac- for mixing cross-sectional data and longitudinal data
tions at various periods after the death of a spouse. to maximize the strengths of both designs. In their
With a cross-sectional design, a group of individu- study of health behavior and beliefs during preg-
als whose spouse had died 1 week ago could be nancy, data were collected cross-sectionally on sub-
tested, another whose loss was 6 months ago, an- jects, subjects were asked to recall previous atti-
other 1 year, another 2 years, and another 5 years. tudes, and a small portion of the sample were short
standard
long
39777 Burns (saunm) INTERACTIVE RIGHT

top of rh
CHAPTER 11 ■ Selecting a Research Design 253 base of rh

top of txt
followed longitudinally to validate the cross-sec- is the study by Wilbur and associates (1990) of the base of txt
tional findings. career trends of master’s-prepared family nurse
Doering and colleagues (1999) conducted a practitioners. They described their design as fol-
cross-sectional study entitled “Evidence of Time- lows.
Dependent Autonomic Reinnervation after Heart ...................................
Transplantation.” The following excerpt describes “The study consisted of FNPs [family nurse prac-
the design of their study. titioners] who had completed graduate studies be-
................................... tween 1974 and 1984. . . . A structured survey
“A sample of 33 orthotopic heart transplant re- questionnaire was used to obtain information on
cipient volunteers from an urban, university-affili- employment, NP practice, professional, and demo-
ated tertiary care center and 16 healthy normal graphic variables. . . . A second questionnaire was
control volunteers matched to the age and sex of used to structure the past employment and demo-
transplant recipients were enrolled in the study. Of graphic variables. . . . The subjects were asked to
the transplant recipients, 16 were less than 5 recall employment information for a maximum of
months posttransplant (early transplant group) and four jobs held since graduation. . . . To identify
17 were more than 1 year posttransplant (late trans- and compare career trends, the decision was made
plant group).” (Doering et al., 1999, p. 310) to group graduating classes into three cohorts repre-
senting the early, intermediate, and recent gradu-
...................................
ates. To identify career trends, the role characteris-
“Handgrip and deep breathing tests, passive 80⌭
tics of both the first job held following graduation
head-up tilt, and heart rate (HR) responsiveness of
and the present job were examined.” (Wilbur et al.,
33 transplant recipients . . . were compared with
1990, pp. 71–72) ■
those of 16 age- and sex-matched control partici-
pants.” (Doering et al, 1999, p. 308) ■
EVENT-PARTITIONING DESIGNS
A merger of the cross-sectional or longitudinal
TREND DESIGNS
and trend designs, the event-partitioning design, is
Trend designs examine changes in the general used in some cases to increase sample size and to
population in relation to a particular phenomenon avoid the effects of history on the validity of find-
(see Figure 11 – 5). Different samples of subjects are ings. Cook and Campbell (1979) refer to these as
selected from the same population at preset inter- cohort designs with treatment partitioning (see Fig-
vals of time, and at each selected time, data are ures 11 – 6 and 11 – 7). The term treatment is used
collected from that particular sample. The re- loosely here to mean a key event that is thought to
searcher must be able to justify generalizing from lead to change. In a descriptive study, the re-
the samples to the population under study. Analysis searcher would not cause or manipulate the key
involves strategies to predict future trends from ex- event but rather would clearly define it so that
amination of past trends. An example of this design when it occurred naturally, it would be recognized.

Time 1 Time 2 Time 3 Time 4 Time..n

measure measure measure measure measure
variables variables variables variables variables PREDICTION

Sample 1 Sample 2 Sample 3 Sample 4 ....n

FIGURE 11 – 5 ■ Trend study design. short
standard
long
39777 Burns (saunm) INTERACTIVE LEFT

top of rh
254 UNIT 2 ■ THE RESEARCH PROCESS base of rh

top of txt
base of txt
Key Time 1
event
1 Sample 1

Key Time 2
event
2 Sample 2

Key Time 3
event
3 Sample 3

Time..n
Key
event Merger of data for analysis
and interpretation of meaning
4 Sample..n

FIGURE 11 – 6 ■ Cross-sectional study with treatment partitioning.

For example, the event-partitioning design could not the absolute length of time. Data would be
be used to study subjects who have completed pro- assumed to be comparable, and a larger sample size
grams to stop smoking. Smoking behaviors and in- would be available for analysis of changes over time.
cidence of smoking-related diseases might be mea- An example of this design is Lepczyk and asso-
sured at intervals of 1 year for a 5-year period. ciates’ (1990) study on the timing of preoperative
However, the number of subjects available at one patient teaching. The following excerpt describes
time might be insufficient for adequate analysis of the study design.
findings. Therefore, subjects from several programs ...................................
offered at different times could be used. Data would “The convenience sample was selected from a
be examined in terms of the relative time since the population of patients who presented themselves
subjects’ completion of the stop-smoking program, for preoperative classes before coronary artery by-

Key Time 1 Time 2 Time 3 Time 4
event
1 Sample 1 Sample 1 Sample 1 Sample 1

Time 1 Time 2 Time 3 Time 4
Key
event
2 Sample 2 Sample 2 Sample 2 Sample 2

Key Time 1 Time 2 Time 3 Time 4
event
3 Sample 3 Sample 3 Sample 3 Sample 3

Merger of data
for analysis and
interpretation of
meaning short
FIGURE 11 – 7 ■ Longitudinal design with treatment partitioning. standard
long
39777 Burns (saunm) INTERACTIVE RIGHT

top of rh
CHAPTER 11 ■ Selecting a Research Design 255 base of rh

top of txt
pass surgery at two large metropolitan hospitals. can lead to the design of large sample studies base of txt
Group 1 (from one hospital, n ⫽ 32) received to examine factors identified through the case
teaching 2–7 days prior to hospital admission. study.
Group 2 (from the other hospital, n ⫽ 42) received The design of a case study depends on the cir-
teaching on the afternoon of hospital admission. cumstances of the case but usually includes an ele-
Seventy-four patients consented to participate. Data ment of time. History and previous behavior pat-
were collected over 8 months.” (Lepczyk et al., terns are usually explored in detail. As the case
1990, p. 302) ■ study proceeds, the researcher may become aware
of components important to the phenomenon being
examined that were not originally built into the
study. A case study is likely to have both quantita-
Case Study Design tive and qualitative elements, and these components
The case study design involves an intensive ex- must be incorporated into the study design. Large
ploration of a single unit of study, such as a person, volumes of data are generally obtained during a
family, group, community, or institution, or a very case study. Organizing the findings of a case study
small number of subjects who are examined inten- into a coherent whole is a difficult but critical
sively. Although the number of subjects tends to be component of the study. Generalization of study
small, the number of variables involved is usually findings in the statistical sense is not appro-
large. In fact, it is important to examine all varia- priate; however, generalizing the findings to
bles that might have an impact on the situation theory is appropriate and important (Barnard et al.,
being studied. 1987; Crombie & Davies, 1996; Gray, 1998; Yin,
Case studies were commonly used in nursing re- 1984).
search in the 1970s but appear in the literature less An example of this design is Lewis’s (1995)
frequently today. Well-designed case studies are study, “One Year in the Life of a Woman with
good sources of descriptive information and can be Premenstrual Syndrome: A Case Study.” An ab-
used as evidence for or against theories. Sterling stract of that study follows.
and McNally (1992) recommend single-subject case ...................................
studies for examining process-based nursing prac- “Over the course of 1 year (13 menstrual cy-
tice. The strategy allows the researcher to investi- cles), data were collected on a daily basis using
gate daily observations and interventions that are a Likert-scale ratings of symptom presence and sever-
common aspect of nursing practice. Case studies are ity, as well as narrative journal entries. The partici-
also commonly used in qualitative studies (Sande- pant was a 37-year-old healthy woman (mean
lowski, 1996). There are even experimental designs cycle length ⫽ 26.7 days, SD ⫾ 1.8) with prospec-
for single case studies (Barlow & Hersen, 1984). A tively screened well-defined premenstrual syndrome
variety of sources of information can be collected (PMS), not on hormones or other drugs, and with-
on each concept of interest through the use of dif- out a psychiatric history. Using the autocorrelation
ferent data collection methods. This approach al- function (ACF), there was evidence for a statistically
lows detailed study of all aspects of a single case. significant predictive cycle-to-cycle symptom pat-
Such a strategy can greatly expand the understand- tern (ACF r ⫽ .49, p ⬍ .05; Bartlett Band range of
ing of the phenomenon under study. significance ⫽ ⫾ .13). Cycle-phase-dependent co-
Case studies are also useful in demonstrating the existence of symptoms was noted, along with par-
effectiveness of specific therapeutic techniques. In ticular narrative themes, most dramatically exempli-
fact, the reporting of a case study can be the vehi- fied by the theme of death. For this subject, the
cle by which the technique is introduced to other findings provided evidence for predictive symptom
practitioners. The case study design also has poten- patterns and an effect of symptom presence on her
tial for revealing important findings that can gener- interpretation of her environment and herself.” short
ate new hypotheses for testing. Thus, the case study (Lewis, 1995, p. 111) ■ standard
long
39777 Burns (saunm) INTERACTIVE LEFT

top of rh
256 UNIT 2 ■ THE RESEARCH PROCESS base of rh

top of txt
Surveys low score of 1 to a high score of 50, truncated base of txt
scores might indicate only scores in the range 10 to
The term survey is used in two ways within
40. More extreme scores would be combined with
scientific thought. It is used in a broad sense to
scores within the designated range. If truncation is
mean any descriptive or correlational study; in this
performed, the researcher may not find a correlation
sense, survey tends to mean nonexperimental. In a
when the variables are actually correlated.
narrower sense, survey is used to describe a tech-
Neophyte researchers tend to make two serious
nique of data collection in which questionnaires
errors with correlational studies. First, they often
(collected by mail or in person) or personal inter-
attempt to establish causality by correlation, reason-
views are used to gather data about an identified
ing that if two variables are related, one must cause
population.
the other. Second, they confuse studies in which
Surveys, in the narrower definition, are used to
differences are examined with studies in which rela-
gather data that can be acquired through self-report.
tionships are examined. Although the existence of a
Because of this limitation in data, some researchers
difference assumes the existence of a relationship,
view surveys as rather shallow and as contributing
the design and statistical analysis of studies examin-
in a limited way to scientific knowledge. This belief
ing differences are different from those examining
has led to a bias in the scientific community against
relationships. In a study examining two or more
survey research. In this context, the term survey is
groups in terms of one or more variables, the re-
used derisively. However, surveys can be an ex-
searcher is exploring differences between groups as
tremely important source of data. In this text, the
reflected in scores on the identified variables. In a
term survey is used to designate a data collection
study examining a single group in terms of two or
technique, not a design. Surveys can be used within
more variables, the researcher is exploring relation-
many designs, including descriptive, correlational,
ships between variables. In a correlational study, the
and quasi-experimental studies.
relationship examined is that between two or more
research variables within an identified situation.
CORRELATIONAL STUDIES
Correlational studies examine relationships
Descriptive Correlational Design
among variables. The examination can occur at sev-
eral levels. The researcher can seek to describe a The purpose of a descriptive correlational design
relationship, predict relationships among variables, is to examine the relationships that exist in a situa-
or test the relationships proposed by a theoretical tion. Using this design facilitates the identification
proposition. In any correlational study, a representa- of many interrelationships in a situation in a short
tive sample must be selected for the study, a sample time. Descriptive correlational studies are also used
reflecting the full range of scores possible on the to develop hypotheses for later studies (see Figure
variables being measured. In correlational designs, a 11 – 8).
large variance in the variable scores is necessary to A descriptive correlational study may examine
determine the existence of a relationship. Therefore, variables in a situation that has already occurred or
correlational designs are unlike experimental de- in a currently occurring situation. No attempt is
signs, in which variance in variable scores is con- made to control or manipulate the situation. As with
trolled (limited). descriptive studies, variables must be clearly identi-
In correlational designs, if the range of scores is fied and defined. An example of a descriptive corre-
truncated, the obtained correlation will be artificially lational design is the study by Johnson and col-
depressed. Truncated means that the lowest scores leagues (1999) that examined the relationships
and the highest scores either are not measured or among attitudes, beliefs, knowledge, and values re-
are condensed and merged with less extreme scores. lated to adolescent sexuality and sexually transmit- short
For example, if an attitude scale were scored from a ted diseases. An abstract of their study follows. standard
long
39777 Burns (saunm) INTERACTIVE RIGHT

top of rh
CHAPTER 11 ■ Selecting a Research Design 257 base of rh

top of txt
Measurement base of txt

research
Description Interpretation
variable
of variable of meaning
1

EXAMINATION
OF RELATIONSHIP

research Description Development
variable of variable of
2
hypotheses

FIGURE 11 – 8 ■ Descriptive correlational design.

................................... dent variables (see Figure 11 – 9). Independent vari-
“This study described rural adolescents’ attitudes,
ables most effective in prediction are highly corre-
beliefs, knowledge, and values with regard to sexu-
lated with the dependent variable but not highly
ality and sexually transmitted diseases (STDs). Rot-
correlated with other independent variables used in
ter’s Social Learning Theory (1954) provided the
the study. Predictive designs require the develop-
theoretical framework for this descriptive, correla-
ment of a theory-based mathematical hypothesis
tional design. The convenience sample consisted of
proposing the variables that are expected to predict
170 students from one rural high school.” (Johnson
the dependent variable effectively. The hypothesis is
et al., 1999, p. 177) ■
then tested with regression analysis.
Jennings-Dozier (1999) conducted a predictive
Predictive Design correlational study to predict intentions to obtain a
Papanicolaou (Pap) smear among African American
Predictive designs are developed to predict the and Latina women. The study was designed as a
value of one variable on the basis of values ob- test of the Theory of Planned Behavior. The follow-
tained from another variable or variables. Prediction ing abstract describes this study.
is one approach to examining causal relationships
between variables. Because causal phenomena are ...................................
being examined, the terms dependent and indepen- “A correlational design was used, and a conven-
dent are used to describe the variables. One variable ience sample of 108 African American and 96
is classified as the dependent variable, and all other Latina adult women were recruited from urban
variables are classified as independent variables. community-based agencies located in a large mid-
The aim of a predictive design is to predict the Atlantic metropolitan area. . . . Direct relation-
level of the dependent variable from the indepen- ships between attitude and perceived behavioral

Predicted
Value of Value of Value of
+ + = Value of
Intercept Independent Independent
Dependent
Variable 1 Variable 2
Variable
short
FIGURE 11 – 9 ■ Predictive design. standard
long
39777 Burns (saunm) INTERACTIVE LEFT

top of rh
258 UNIT 2 ■ THE RESEARCH PROCESS base of rh

top of txt
control and intention to obtain an annual Pap of this model. Endogenous variables are those base of txt
smear were found for African American and Latina whose variation is explained within the theoretical
women. The subjective norm did not significantly model. Exogenous variables influence the variation
predict intention. . . . [T]he study findings did not of endogenous variables. Residual variables indicate
support the empirical adequacy of the Theory of the effect of unmeasured variables not included in
Planned Behavior for either of the ethnic groups.” the model. These variables explain some of the var-
(Jennings-Dozier, 1999, p. 395) ■ iance found in the data but not the variance within
the model (Mason-Hawkes & Holm, 1989).
In Figure 11 – 10, the illustration of a model-
Model-Testing Designs testing design, paths are drawn to demonstrate di-
Some studies are designed specifically to test the rections of cause and effect. The arrows (paths)
accuracy of a hypothesized causal model. The de- from the exogenous variables 1, 2, and 3 lead to the
sign requires that all variables relevant to the model endogenous variable 4, indicating that variable 4 is
be measured. A large, heterogeneous sample is re- theoretically proposed to be caused by variables 1,
quired. All the paths expressing relationships be- 2, and 3. The arrow (path) from endogenous vari-
tween concepts are identified, and a conceptual map able 4 to endogenous variable 5 indicates that vari-
is developed (see Figure 11 – 10). The analysis de- able 4 causes variable 5.
termines whether or not the data are consistent with The researcher measures exogenous and endoge-
the model. In some studies, data from some subjects nous variables by collecting data from the experi-
are set aside and not included in the initial path mental subjects and analyzes the accuracy of the
analysis. These data are used to test the fit of the proposed paths. Initially, these analysis procedures
paths defined by the initial analysis in another data were performed with a series of regression analyses.
set. Now, statistical procedures have been developed
Variables are classified into three categories, ex- specifically for path analysis using the computer
ogenous variables, endogenous variables, and resid- programs LISREL and EQS; these programs are
ual variables. Exogenous variables are within the described in Chapter 20. Path coefficients are calcu-
theoretical model but are caused by factors outside lated that indicate the effect that one variable has

EXOGENOUS ENDOGENOUS
VARIABLES VARIABLES

variable
1 CAUSAL
CORRELATIONS

NONCAUSAL variable variable variable
CORRELATIONS 2 4 5

a b
variable
3

RESIDUAL
VARIABLES short
FIGURE 11 – 10 ■ Model testing design. standard
long
39777 Burns (saunm) INTERACTIVE RIGHT

top of rh
CHAPTER 11 ■ Selecting a Research Design 259 base of rh

top of txt
on another. The amount of variance explained by ternative means of examining causality in situations base of txt
the model as well as the fit between the path coeffi- not conducive to experimental controls. Quasi-
cients and the theoretical model indicates the accu- experimental designs were first described as a group
racy of the theory. Variance that is not accounted by Campbell and Stanley in 1963, when only exper-
for in the statistical analysis is attributed to residual imental designs were considered of any worth.
variables (variables a and b) not included in the Cook and Campbell expanded this description in
analyses (Mason-Hawkes & Holm, 1989). 1979. Quasi-experimental designs facilitate the
An example of this design is Hagerty and Wil- search for knowledge and examination of causality
lams’s (1999) study, “The Effects of Sense of Be- in situations in which complete control is not possi-
longing, Social Support, Conflict, and Loneliness on ble. These designs have been developed to control
Depression.” The following abstract describes the as many threats to validity as possible in a situation
study. in which at least one of the three components of
................................... true experimental design (random sampling, control
“A sample of clients with major depressive disor- groups, and manipulation of the treatment) is lack-
der and students in a midwestern community col- ing.
lege participated in the study by completing ques- There are differences of opinion in nursing about
tionnaires. Path analysis showed significant direct the classification of a particular study as quasi-
paths as postulated, with 64% of the variance of experimental or experimental. The experimental de-
depression explained by the variables in the model. signs emerged from a logical positivist perspective
Social support had only an indirect effect on de- with the purpose of determining cause and effect.
pression, and this finding supported the buffer the- The focus is on determining differences between
ory of social support. Sense of belonging was a groups using statistical analyses on the basis of de-
better predictor of depression.” (Hagerty & Wil- cision theory (see Chapter 18 for an explanation of
liams, 1999, p. 215) ■ decision theory). The true experimental design
(from a logical positivist view) requires the use of
QUASI-EXPERIMENTAL STUDY random sampling to obtain subjects, random assign-
ment to control and experimental groups, rigorous
DESIGNS
control of the treatment, and designs that controlled
The purpose of quasi-experimental and experi- threats to validity.
mental designs is to examine causality. The power A less rigorous type of experimental design, re-
of the design to accomplish this purpose depends ferred to as comparative experimental design, is be-
on the extent to which the actual effects of the ing used by some researchers in both nursing and
experimental treatment (the independent variable) medicine for clinical situations in which the expec-
can be detected by measurement of the dependent tation of random sampling is difficult if not impos-
variable. Obtaining an understanding of the true ef- sible to achieve. These studies use convenience
fects of an experimental treatment requires action to samples with random assignment to groups. For ex-
control threats to the validity of the findings. ample, clinical trials do not use randomly obtained
Threats to validity are controlled through selection samples but tend to be considered experimental in
of subjects, control of the environment, manipula- nature. The rationale for classifying such studies as
tion of the treatment, and reliable and valid meas- experimental is that they have internal validity if
urement of the dependent variables. the two groups are comparable on variables impor-
Experimental designs, with their strict control of tant to the study even though there are biases in the
variance, are the most powerful method of examin- original sample. However, threats to statistical con-
ing causality. For many reasons, both ethical and clusion validity and threats to external validity by
practical, however, experimental designs cannot al- the nonrandom sample are not addressed by these
ways be used in social science research. Quasi- designs. Threats to external validity have not, in the short
experimental designs were developed to provide al- past, been considered a serious concern because standard
long
39777 Burns (saunm) INTERACTIVE LEFT

top of rh
260 UNIT 2 ■ THE RESEARCH PROCESS base of rh

top of txt
they affect not the claim that the treatment caused a Stout and colleagues (1994) suggest a similar base of txt
difference but rather the ability to generalize the strategy they refer to as urn randomization, which
findings. The importance of external validity, al- they describe as follows:
though discounted in the past, is taking on greater
...................................
importance in the current political and health policy
“One would begin the study with two urns, each
climate. See Chapter 12 on outcomes research for a
urn containing a red marble and a blue marble.
discussion of concerns related to the validity of
There is one urn for each level of the stratifying
clinical trials.
variable; that is, in this example there is an urn for
severely ill patients and another urn for the less
Random Assignment to Groups severe patients. When a subject is ready for ran-
domization, we determine whether or not he/she is
Random assignment is a procedure used to as-
severely ill and consult the corresponding urn.
sign subjects to treatment or control groups ran-
From this urn (say, for the severely ill group) we
domly. Random assignment is most commonly used
randomly select one marble and note its color. If
in nursing and medicine to assign subjects obtained
the marble is red we assign the patient to Treat-
through convenience sampling methods to groups
ment A. Then we drop that marble back into the
for purposes of comparison. Random assignment
urn and put a blue marble into the urn as well.
used without random sampling is purported to de-
This leaves the ‘severely ill’ urn with one red and
crease the risk of bias in the selection of groups.
two blue marbles. The next time a severely ill pa-
However, a meta-analysis by Ottenbacher (1992),
tient shows up, the probability that he/she will be
examining the effect of random assignment versus
assigned to Group B will be 2/3 rather than 2, thus
nonrandom assignment on outcomes, failed to re-
biasing the selection process toward balance. A
veal significant differences in these two sampling
similar procedure is followed every time a severely
techniques. He suggests that previous assumptions
ill subject presents for randomization. After each
about design strategies should be empirically tested.
subject is assigned, the marble chosen from the urn
Traditional approaches to random assignment in-
is replaced together with a marble of the opposite
volve using a random numbers table or flipping an
color. The urn for the less severely ill group is not
unbiased coin to determine group assignment. How-
affected. If a low-severity patient presents for the
ever, these procedures can lead to unequal group
study, that patient’s probability of assignment to ei-
sizes and, thus, a decrease in power. Hjelm-Karls-
ther treatment is not affected by the assignment of
son (1991) suggests using what is referred to as a
patients in the other stratum. To some extent, urn
biased coin design to randomly assign subjects to
randomization can be tailored to maximize balanc-
groups. With this technique, selection of the group
ing or to maximize randomization.” (Stout et al.,
to which a particular subject will be assigned is
1994, p. 72). ■
biased in favor of groups that have smaller sample
sizes at the point of the assignment of that subject.
These authors also provide strategies for balancing
This strategy is particularly useful when assignment
several variables simultaneously during random as-
is being made to more than two groups. Calcula-
signment.
tions for the sequencing of assignment to groups
Chang (1999) used random assignment in her
can be completed before initiation of data collec-
study of a cognitive-behavioral intervention for
tion, thus freeing the researcher for other activities
homebound caregivers of people with dementia. She
during this critical period. Hjelm-Karlsson (1991)
describes her sampling procedure as follows.
suggests using cards to make group assignments.
The subject numbers and random group assignments ...................................
are written on cards. As each subject agrees to “Objective: To examine the effects of an 8-week
participate in the study, the next card is drawn from cognitive-behavioral (C-B) intervention tailored to short
the stack, indicating that subject’s number and the specific deficits of persons with dementia standard
group assignment. (PWDs) on selected outcomes for homebound care- long
39777 Burns (saunm) INTERACTIVE RIGHT

top of rh
CHAPTER 11 ■ Selecting a Research Design 261 base of rh

top of txt
givers and the functional status of the PWD.” researcher must carefully assess the potential threats base of txt
(Chang, 1999, p. 173) to validity in interpreting statistical results, because
................................... statistical analysis cannot control for threats to va-
“Method: The design was a two-group random- lidity. The following sections describe examples of
ized trial with measures taken at baseline, 4 weeks, nonequivalent control group design.
8 weeks, and 12 weeks.” (Chang, 1999, p. 173).
................................... THE ONE-GROUP POSTTEST-ONLY
“Procedure: Caregiving dyads were randomly as- DESIGN
signed to C-B or A-O groups.” (Chang, 1999, The one-group posttest-only design is referred to
p. 175) ■ as pre-experimental rather than quasi-experimental
Each of the quasi-experimental designs described because of its weaknesses and the numerous threats
in this section involves threats to validity owing to to validity. It is inadequate for making causal infer-
constraints in controlling variance. Some achieve ences (see Figure 11 – 11). Usually in this design,
greater amounts of control than others. When no attempt is made to control the selection of sub-
choosing designs, the researcher selects the design jects who receive the treatment (the experimental
that offers the greatest amount of control possible group). Generalization of findings beyond those
within the study situation. Even the first designs tested is difficult to justify. The group is not pre-
described in this section, which have very low tested; therefore, there is no direct way to measure
power in terms of establishing causality, can pro- change. The researcher cannot claim that posttest
vide useful information on which to design later scores were a consequence (effect) of the treatment
studies. if scores before the treatment are unknown. Because
there is no control group, one does not know
whether groups not receiving the treatment would
Nonequivalent Control Group have similar scores on the dependent variable. The
Designs one-group posttest-only design is commonly used
by the inexperienced researcher to evaluate a treat-
A nonequivalent control group is one in which
ment program or a nursing intervention.
the control group is not selected by random means.
Cook and Campbell (1979) suggest situations in
Some groups are more nonequivalent than others,
which the one-group posttest-only design can be
and some quasi-experimental designs involve using
appropriate and adequate for inferring causality. For
groups (control and treatment) that have evolved
example, the design could be used to determine that
naturally rather than being developed randomly.
a single factory’s use of vinyl chloride is causing an
These groups cannot be considered equivalent be-
increase in the rate of neighborhood and employee
cause the individuals in the control group may be
cancers. The incidence of cancer in the community
different from individuals in the treatment group.
at large is known. The fact that vinyl chloride
Individuals have selected the group in which they
causes cancer and the types of cancer caused by it
are included rather than being selected by the
are also known. These norms would then take the
researcher. Thus, selection becomes a threat to va-
place of the pretest and the control group. Thus, to
lidity.
use this design intelligently, one must know a great
The approach to statistical analysis is problem-
deal about the causal factors interacting within the
atic in quasi-experimental designs. Although many
situation.
researchers use the same approaches to analysis as
are used for experimental studies, the selection bias
THE POSTTEST-ONLY DESIGN WITH
inherent in nonequivalent control groups makes this
NONEQUIVALENT GROUPS
practice questionable. Reichardt (1979) recommends
using multiple statistical analyses to examine the Although the posttest-only design with nonequi- short
data from various perspectives and to compare lev- valent groups offers an improvement on the previ- standard
els of significance obtained from each analysis. The ous design, because of the addition of a nonequi- long
39777 Burns (saunm) INTERACTIVE LEFT

top of rh
262 UNIT 2 ■ THE RESEARCH PROCESS base of rh

top of txt
base of txt
Manipulation of Measurement of
independent variable dependent variable(s)
TREATMENT POSTTEST

Treatment—often ex post facto
Experimental group—those who receive the treatment and the posttest
Pretest—inferred—norms of measures of dependent variable(s) of population
from which experimental group taken
Control group—implied—norms of measures of dependent variable(s) of population
from which experimental group taken
Approach to Analysis: • comparison of posttest scores with inferred norms
• confident inferences about change

Example: Lamb (1979). Effect of positioning of postoperative fractured-hip patients
as related to comfort.

Uncontrolled threats to validity: • no link between treatment and change
• no control group
• maturation
• undetected confounding variables
• inability to access threats to validity

FIGURE 11 – 11 ■ One-group posttest– only design.

valent control group, it is still referred to as jects in many studies using this design are selected
pre-experimental (see Figure 11 – 12). The addition on the basis of high or low scores on the pretest.
of a nonequivalent control group can lead to a false Thus, there is an additional threat that changes in
confidence in the validity of the findings. Selection the posttest may be due to regression toward the
threats are a problem with both groups. The lack of mean. The addition of a nonequivalent control
a pretest remains a serious impediment to defining group, as described in the next design, can greatly
change. Differences in posttest scores between strengthen the validity of the findings.
groups may be caused by the treatment or by differ-
ential selection processes. THE UNTREATED CONTROL GROUP
DESIGN WITH PRETEST AND
THE ONE-GROUP PRETEST – POSTTEST POSTTEST
DESIGN
The pretest-posttest design with an untreated or
Another pre-experimental design, the one-group placebo group is most commonly used in social
pretest – posttest design, is one of the more com- science research (see Figure 11 – 14). This quasi-
monly used designs but has such serious weak- experimental design is the first design discussed
nesses that findings are often uninterpretable (see here that is generally interpretable. The uncontrolled
Figure 11 – 13). Pretest scores cannot adequately threats to validity are primarily due to the absence
serve the same function as a control group. Events of randomization and, in some studies, the inability
can occur between the pretest and posttest that alter of the researcher to manipulate the treatment. The
responses to the posttest. These events then serve as effects of these threats on interpreting study find-
alternative hypotheses to the proposal that the ings are discussed in detail by Cook and Campbell
change in posttest scores is due to the treatment. (1979).
Posttest scores might be altered by (1) maturation Variations in this design include the use of short
processes, (2) administration of the pretest, and (1) proxy pretest measures (a different pretest that standard
(3) changes in instrumentation. Additionally, sub- correlates with the posttest), (2) separate pretest and long
39777 Burns (saunm) INTERACTIVE RIGHT

top of rh
CHAPTER 11 ■ Selecting a Research Design 263 base of rh

top of txt
base of txt
Manipulation of Measurement of
independent variable dependent variable(s)
Experimental
TREATMENT POSTTEST
group
Nonequivalent
control group POSTTEST

Treatment—often ex post facto
may not be well defined
Experimental group—those who receive the treatment and the posttest
Pretest—inferred—norms of measures of dependent variable(s) of population
from which experimental group taken
Control group—not randomly selected—tend to be those who naturally in the situation
do not receive the treatment
Approach to Analysis: • comparison of posttest scores of experimental and
control groups
• comparison of posttest scores with norms

Example: Arakawa (1995). Use of relaxation to reduce side effects of chemotherapy
in Japanese patients.

Uncontrolled threats to validity: • no link between treatment and change
• no pretest
• selection

FIGURE 11 – 12 ■ Posttest– only design with nonequivalent groups.

Measurement of Manipulation of Measurement of
dependent variable(s) independent variable dependent variable(s)
Experimental
PRETEST TREATMENT POSTTEST Group
group

Treatment—greater researcher control
usually not ex post facto
Experimental group—greater researcher control
expected to serve as control
Control group—pretest scores of treatment group expected to serve as control
Approach to Analysis: • comparison of pretest and post-test scores

Example: Mirolo et al. (1995). Psychological benefits of postmastectomy lymphedema
therapy.
Uncontrolled threats to validity: • history
• statistical regression
• maturation
• testing
• instrumentation
short
standard
FIGURE 11 – 13 ■ One-group pretest– posttest design. long
39777 Burns (saunm) INTERACTIVE LEFT

top of rh
264 UNIT 2 ■ THE RESEARCH PROCESS base of rh

top of txt
Measurement of Manipulation of Measurement of base of txt
dependent variable(s) independent variable dependent variable(s)
Experimental
PRETEST TREATMENT POSTTEST
group
Nonequivalent
control group PRETEST POSTTEST

Treatment—experimental group
control group not treated
Control group—not randomly selected
Approach to Analysis: • comparison of control and experimental pretest
• comparison of pretest and posttest
• comparison of control and experimental posttest

Example: Blair (1995). Combining behavior management and mutual goal setting
to reduce physical dependency in nursing home residents.

Uncontrolled threats to validity: • selection–maturation
• instrumentation
• differential statistical regression
• interaction of selection and history

FIGURE 11 – 14 ■ Untreated control group design with pretest and posttest.

posttest samples, and (3) pretest measures at more be equivalent. In nursing situations, the ethics of
than one time interval. The first two variations removing an effective treatment must be considered.
weaken the design, but the last variation greatly Even if doing so is ethically acceptable, the re-
strengthens it. sponse of subjects to the removal may make inter-
In some studies, the placebo treatment consists preting changes difficult.
of the usual treatment provided in the clinical set-
ting. Because there is commonly wide variation in THE REVERSED-TREATMENT
usual treatment among patients, this strategy re- NONEQUIVALENT CONTROL GROUP
duces the effect size of the experimental treatment, DESIGN WITH PRETEST AND
increases the variance, and decreases the possibility POSTTEST
of obtaining a significant difference between groups.
The reversed-treatment nonequivalent control
group design with pretest and posttest introduces
THE REMOVED-TREATMENT DESIGN
two independent variables — one expected to pro-
WITH PRETEST AND POSTTEST
duce a positive effect and one expected to produce
In some cases, gaining access to even a nonequi- a negative effect (see Figure 11 – 16). There are two
valent control group is not possible. The removed- experimental groups, each exposed to one of the
treatment design with pretest and posttest creates treatments. The design tests differences in response
conditions that approximate the conceptual require- to the two treatments. This design is more useful
ments of a control group receiving no treatment. for theory testing than the no-treatment control
The design is basically a one-group pretest-posttest group design because of its high construct validity
design. However, after a delay, a third measure of of the cause. The theoretical causal variable must be
the dependent variable is taken, followed by an rigorously defined to allow differential predictions
interval in which the treatment is removed, followed of directions of effect. To be maximally interpreta- short
by a fourth measure of the dependent variable (see ble, the following two groups must be added: (1) a standard
Figure 11 – 15). The periods between measures must placebo control group in which the treatment is not long
39777 Burns (saunm) INTERACTIVE RIGHT

top of rh
CHAPTER 11 ■ Selecting a Research Design 265 base of rh

top of txt
base of txt

Sample Experimental Dissipation of Controlled Condition
Condition Treatment Effect (Treatment Withheld)
#1 M(1) M(2) M(3) M(4)

Equivalent Time Intervals

M = Measurement of dependent variable(s)
T = Treatment—Manipulation of the independent variable
Approach to Analysis: • comparison of changes in dependent variable
• scores between measurement points

Example: Hanneman (1994). Multidimensional predictors of success or failure
with early weaning from mechanical ventilation after cardiac surgery.

Uncontrolled threats to validity: • statistical conclusion validity
• construct validity of the cause
• selection

FIGURE 11 – 15 ■ Removed-treatment design with pretest and posttest.

expected to affect the dependent variable and (2) a treatment may be different in important ways from
no-treatment control group to provide a baseline. the posttreatment group. Seasonal variation or other
cyclical influences can be interpreted as treatment
effects. Therefore, identifying cyclical patterns that
Interrupted Time-Series Designs may be occurring and controlling for them are criti-
The interrupted time-series design is similar to cal to the analysis of study findings.
descriptive time designs except that a treatment is McCain and McCleary (1979) suggest the use of
applied at some point in the observations. Time- the ARIMA (autoregressive integrated moving aver-
series analyses have some advantages over other age) statistical model (see Chapter 20) to analyze
quasi-experimental designs. First, repeated pretest time-series data. ARIMA is a relatively new model
observations can assess trends in maturation before that has some distinct advantages over regression
the treatment. Second, the repeated pretest observa- analysis techniques. For adequate statistical analysis,
tions allow measures of trends in scores before the at least 50 measurement points are needed; how-
treatment, decreasing the risk of statistical regres- ever, Cook and Campbell (1979) believe that even
sion, which would lead to misinterpretation of find- small numbers of measurement points can provide
ings. If records are kept of events that could influ- better information than that obtained in cross-sec-
ence subjects in the study, the researcher can tional studies. The numbers of measures shown in
determine whether historical factors that could mod- the designs illustrated here (Figures 11 – 17 through
ify responses to the treatment were in operation 11 – 19) are limited by space and are not meant to
between the last pretest and the first posttest. suggest limiting measures to the numbers shown.
Some threats, however, are particularly problem-
SIMPLE INTERRUPTED TIME SERIES
atic in time-series designs. Record-keeping proce-
dures and definitions of constructs used for data The simple interrupted time series is similar to
collection tend to change over time. Thus, maintain- the descriptive study, with the addition of a treat- short
ing consistency can be a problem. The treatment ment that occurs or is applied at a given point in standard
can result in attrition so that the sample before time (see Figure 11 – 17). The treatment, which in long
39777 Burns (saunm) INTERACTIVE LEFT

top of rh
266 UNIT 2 ■ THE RESEARCH PROCESS base of rh

top of txt
base of txt
Measurement of Manipulation of Measurement of
dependent variable(s) independent variable dependent variable(s)

Experimental proposed positive
PRETEST effective treatment POSTTEST
group 1

Experimental proposed positive
group 2 PRETEST effective treatment POSTTEST

Approach to Analysis: • comparison of group 1 and group 2 pretest
• comparison of group 1 and group 2 posttest
• comparison of changes between pretest and
posttest between groups
Example: Shiao et al. (1995). Nasogastric tube placement: Effects on breathing
and sucking in very low birthweight infants.
Uncontrolled threats to validity: • statistical conclusion validity

FIGURE 11 – 16 ■ The reversed-treatment nonequivalent control group design with pretest and posttest.

Measurement of Manipulation of Measurement of
dependent variable(s) independent variable dependent variable(s)

PRETESTS TREATMENT POSTTESTS

Experimental group M(1) M(2) M(3) M(4) T M(5) M(6) M(7) M(8)

Subjects: may remain the same across the study or different individuals may be
selected at each measurement period (as in the descriptive
time-dimensional designs)
Pretest dependent variable measures: In addition to researcher-initiated tools,
measures may be obtained from archival data sources such as patient
records, scores available in student records, or employee records.
Treatment: may be a key event occurring at one specific time that is expected to modify
subject responses to the dependant variable; that is, curriculum change,
administrative change, change in nursing care of a particuar type of patient.
The treatment tends to continue after inplementation rather than occurring and
and then being withdrawn.
Approach to Analysis: • changes in trends of scores before and after treatment
Example: Whitney, Stotts & Goodson (1995). Effects of dynamic exercise on
subcutaneous oxygen tension and temperature.
Uncontrolled threats to validity: • history
• seasonal trends
• instrumentation
• selection
• mono-operation bias
• cyclical influences interpreted as treatment effects
short
FIGURE 11 – 17 ■ Simple interrupted time series. standard
long
39777 Burns (saunm) INTERACTIVE RIGHT

top of rh
CHAPTER 11 ■ Selecting a Research Design 267 base of rh

top of txt
some cases is not completely under the control of agement practices or patient teaching strategies), the base of txt
the researcher, must be clearly defined. The use of initial response to the change may differ from later
multiple methods to measure the dependent variable responses.
greatly strengthens the design. Threats that are well
controlled by this design are maturation and statisti- INTERRUPTED TIME SERIES WITH
cal regression. MULTIPLE REPLICATIONS
The interrupted time series with multiple replica-
INTERRUPTED TIME SERIES WITH A
tions is a powerful design for inferring causality
NONEQUIVALENT NO-TREATMENT
(see Figure 11 – 19). It requires greater researcher
CONTROL GROUP TIME SERIES
control than is usually possible in social science
The addition of a control group to the interrupted research outside closed institutional settings, such as
time-series design greatly strengthens the validity of laboratories or research units. The studies that led to
the findings. The control group allows examination the adoption of behavior modification techniques
of differences in trends between groups after the used this design. For significant differences to be
treatment and of the persistence of treatment effects interpretable, the pretest and posttest scores must be
over time (see Figure 11 – 18). Although the treat- in different directions. Within this design, treat-
ment may continue (e.g., a change in nursing man- ments can be modified through substitution of one

Measurement of Manipulation of Measurement of
dependent variable(s) independent variable dependent variable(s)

PRETESTS TREATMENT POSTTESTS
Experimental group M(1) M(2) M(3) M(4) T M(5) M(6) M(7) M(8)
Experimental group M(1) M(2) M(3) M(4) M(5) M(6) M(7) M(8)

Subjects: may remain the same across the study or may be part of a group in which
some individuals may change (e.g., class, work group, patient teaching,
or patient care group)
Pretest dependent variable measures: In addition to researcher-initiated tools,
measures may be obtained from archival data sources.
Treatment: may be a key event occurring at one specific time that is expected to modify
experiment subjects' responses to dependant variable measures. Treatment
tends to continue after implementation rather than occurring and then being
withdrawn.
Approach to Analysis: • changes in trends of scores before and after treatment
• comparison of trends in experimental and control groups
• temporal persistence of treatment effects
Example: Quayhagen et al. (1995). A dyadic remediation program for care recipients
with dementia.
Uncontrolled threats to validity: • selection and history interaction
• as experimental and control groups increase in
noncomparability, threats to validity increase
• interaction of populations and treatment
• cyclical influences interpreted as treatment effects

short
FIGURE 11 – 18 ■ Interrupted time series with a nonequivalent no-treatment control group time series. standard
long
39777 Burns (saunm) INTERACTIVE LEFT

top of rh
268 UNIT 2 ■ THE RESEARCH PROCESS base of rh

top of txt
base of txt
PRETEST T POSTTEST PRETEST T POSTTEST etc
Single sample
M(1) M(2) T M(3) M(4) T M(5) M(6) T M(7) M(8) T M(9) M(10)

Treatment: • provided repeatedly
• effects must dissipate rapidly
• should be scheduled randomly
Approach to Analysis: • powerful for inferring causal effects
• differences in pretest and posttest scores must be in
opposite directions to be interpretable
Example: Fridriksdottir & McCarthy (1995). The effect of caloric density of food
on energy intake and body weight in tumor-bearing rats.

FIGURE 11 – 19 ■ Interrupted time series with multiple replications.

treatment for another or combining of two treat- The Classic Experimental Design
ments and examination of interaction effects.
The original or classic experimental design is
still the most commonly used experimental design
(see Figure 11 – 20). There are two randomized
EXPERIMENTAL STUDY DESIGNS groups, one receiving the experimental treatment
and one receiving no treatment, a placebo treatment,
Experimental designs are set up to provide the or the usual treatment. Comparison of pretest scores
greatest amount of control possible to examine cau- allows evaluation of the effectiveness of randomiza-
sality more closely. To examine cause, one must tion in providing equivalent groups. The treatment
eliminate all factors influencing the dependent vari- is under the control of the researcher. The depen-
able other than the cause (independent variable) be- dent variable is measured twice, before and after the
ing studied. Other factors are eliminated by control- manipulation of the independent variable. As with
ling them. The study is designed to prevent any all well-designed studies, the dependent and inde-
other element from intruding into observation of the pendent variables are conceptually linked, conceptu-
specific cause and effect that the researcher wishes ally defined, and operationalized. Instruments used
to examine. to measure the dependent variable clearly reflect the
The three essential elements of experimental re- conceptual meaning of the variable and have good
search are (1) randomization, (2) researcher-con- evidence of reliability and validity. Often, more
trolled manipulation of the independent variable, than one means of measuring the dependent vari-
and (3) researcher control of the experimental situa- able is advisable to avoid mono-operation and mono-
tion, including a control or comparison group. Ex- method biases.
perimental designs exert much effort to control vari- Most other experimental designs are variations of
ance. Sample criteria are explicit, the independent the classic experimental design. Multiple groups
variable is provided in a precisely defined way, the (both experimental and control) can be used to great
dependent variables are carefully operationalized, advantage in the pretest-posttest design and the
and the situation in which the study is conducted is posttest-only design. For example, one control
rigidly controlled to prevent the interference of un- group could receive no treatment and another con-
studied factors from modifying the dynamics of the trol group could receive a placebo treatment. Multi-
process being studied. ple experimental groups could receive varying lev- short
standard
long
39777 Burns (saunm) INTERACTIVE RIGHT

top of rh
CHAPTER 11 ■ Selecting a Research Design 269 base of rh

top of txt
base of txt
Measurement of Manipulation of Measurement of
dependent variable(s) independent variable dependent variable(s)
Randomly selected
experimental group PRETEST TREATMENT POSTTEST
Randomly selected
PRETEST POSTTEST
control group

Treatment: under control of researcher
Approach to Analysis: • comparison of pretest and posttest scores
• comparison of control and experimental groups
• comparison of pretest/posttest differences between samples
Example: Youssef (1990). The impact of group reminiscence counseling on a
depressed elderly population.
Uncontrolled threats to validity: • testing
• instrumentation
• mortality
• restricted generalizability as control increases

FIGURE 11 – 20 ■ Pretest– posttest control group design.

els of the treatments, such as differing frequency, Randomized Block Design
intensity, or duration of nursing care measures.
These additions greatly increase the generalizability The randomized block design uses the two-group
of the study findings. pretest-posttest pattern or the two-group posttest
pattern with one addition, a blocking variable. The
blocking variable, if uncontrolled, is expected to
Posttest-only Control Group Design confound the findings of the study. To prevent con-
In some studies, the dependent variable cannot founding of the findings, the subjects are rank-
be measured before the treatment. For example, be- ordered in relation to the blocking variable.
fore the beginning of treatment, it is not possible to For example, if effectiveness of a nursing inter-
make meaningful measurements of the responses to vention to relieve postchemotherapy nausea were
interventions designed to control nausea from the independent variable in a study, severity of nau-
chemotherapy or postoperative pain. Additionally, sea could confound the findings. Subjects would be
in some cases, subjects’ responses to the posttest ranked according to severity of nausea. The two
can be due, in part, to learning from or subjective subjects with the most severe nausea would be
reaction to the pretest (pretest sensitization). If this identified and randomly assigned, one to the experi-
issue is a concern in a specific study, the pretest mental group and one to the control group. The two
may be eliminated and a posttest-only control group subjects next in rank would then be identified and
design can be used (see Figure 11 – 21). However, randomly assigned. The pattern would be followed
this elimination prevents the use of many powerful until the entire sample was randomly assigned as
statistical analysis techniques within the study. Ad- matched pairs. This procedure ensures that the ex-
ditionally, the effectiveness of randomization in ob- perimental group and the control group are equal in
taining equivalent experimental and control groups relation to the potentially confounding variable.
cannot be evaluated in terms of the study variables. The effect of blocking can also be accomplished
Nevertheless, the groups can be evaluated in terms statistically (through the use of analysis of covari-
of sample characteristics and other relevant variables. ance) without categorizing the confounding variable short
standard
long
39777 Burns (saunm) INTERACTIVE LEFT

top of rh
270 UNIT 2 ■ THE RESEARCH PROCESS base of rh

top of txt
base of txt
Manipulation of Measurement of
independent variable dependent variable

Randomly selected
experimental group TREATMENT POSTTEST

Randomly selected POSTTEST
control group

Treatment—under control of researcher
Approach to Analysis: • comparison of control and experimental groups
Example: Wikblad & Anderson (1995). A comparison of three wound dressings in
patients undergoing heart surgery
Uncontrolled threats to validity: • instrumentation
• mortality
• limited generalizability as control increases

FIGURE 11 – 21 ■ Posttest– only control group design.

into discrete components. However, for this analysis A 2 ⫻ 2 factorial design produces a study with
to be accurate, one must be careful not to violate four cells (A through D). Each cell must contain an
the assumptions of the statistical procedure (Spec- equivalent number of subjects. Cells B and C allow
tor, 1981). An example of this design is McCurren examination of each intervention separately. Cell D
and associates’ (1999) study testing an intervention subjects receive no treatment and serve as a control
strategy for depression among nursing home elders. group. Cell A allows examination of the interaction
The following excerpt describes the study design. between the two independent variables. This design
................................... can be used, as in the randomized block design, to
“A randomized block design was used; i.e., sub- control for confounding variables. The confounding
jects were first categorized in blocks according to variable is included as an independent variable, and
initial Geriatric Depression Scale (GDS) score interactions between it and the other independent
(mildly or severely depressed) and then randomly variable are examined (Spector, 1981).
assigned to either the experimental or the control Extensions of the factorial design to more than
group.” (McCurren et al., 1999, p. 13) ■ two levels of variables are referred to as M ⫻ N
factorial designs. Within this design, independent
variables can have any number of levels within
Factorial Designs
In a factorial design, two or more different char-
acteristics, treatments, or events are independently
varied within a single study. This design is a logical Level of Level of Distraction
approach to examining multiple causality. The sim- Relaxation
Distraction No Distraction
plest arrangement is one in which two treatments or
factors are involved and, within each factor, two Relaxation A B
levels are manipulated (for example, the presence or
absence of the treatment); this is referred to as a
No Relaxation C D
2 ⫻ 2 factorial design. This design is illustrated in
Figure 11 – 22, in the two independent variables are
relaxation and distraction as means of pain relief. FIGURE 11 – 22 ■ Example of factorial design. short
standard
long
39777 Burns (saunm) INTERACTIVE RIGHT

top of rh
CHAPTER 11 ■ Selecting a Research Design 271 base of rh

top of txt
practical limits. Note that a 3 ⫻ 3 design involves 9 much larger in number than those illustrated, be- base of txt
cells and requires a much larger sample size. A 4 ⫻ cause each unit would be considered a subject and
4 design would require 16 cells. A 4 ⫻ 4 design would be randomly assigned to a treatment.
would allow relaxation to be provided at four levels The following excerpt from Harris and Hyman’s
of intensity, such as no relaxation, relaxation for 10 (1984) study, “Clean vs. Sterile Tracheotomy Care
minutes twice a day, relaxation for 15 minutes three and Level of Pulmonary Infection,” is an example
times a day, and relaxation for 20 minutes four of this design.
times a day. Distraction would be provided at simi- ...................................
lar levels. “The purpose of this research was to determine
Factorial designs are not limited to two indepen- if clean tracheotomy care was more effective than
dent variables; however, interpretation of larger sterile as measured by levels of postoperative pul-
numbers becomes more complex and requires monary infection. Ten hospitals with large Head
greater knowledge of statistical analysis. Factorial and Neck/ENT services were selected as data col-
designs do allow the examination of theoretically lection sites. . . . To increase external validity,
proposed interrelationships between multiple inde- possible unique effects associated with hospitals
pendent variables. However, very large samples are were controlled by use of a nested design—hos-
required. pital nested within treatment procedure. . . . At
An example of factorial design is the study by these centers a minimum of 15 tracheostomy pa-
Henker and associates (1995), which evaluated four tient charts were reviewed pre- and postoperatively
methods of warming intravenous fluids (IVF). An for clinical and laboratory data related to infection.
excerpt from that study follows. Patient level of infection was defined using the
................................... Weighted Level of Pulmonary Infection Tool, which
“The methods of warming IVF and the control was constructed for this study.” (Harris & Hyman,
were evaluated with a 5 ⫻ 5 factorial experimental 1984, pp. 80–83) ■
design. Each method of warming and the control
were evaluated at each flow rate and replicated
Crossover or Counterbalanced
two times, yielding three measurements for each
Designs
warming method at each rate.” (Henker et al.,
1995, p. 386) ■ In some studies, more than one treatment is ad-
ministered to each subject. The treatments are pro-
vided sequentially rather than concurrently. Com-
Nested Designs
parisons are then made of the effects of the
In some experimental situations, the researcher different treatments on the same subject. For exam-
wishes to consider the effect of variables that are ple, two different methods known to achieve relax-
found only at some levels of the independent varia- ation might be used as the two treatments. One
bles being studied. Variables found only at certain difficulty encountered in this type of study is that
levels of the independent variable are called nested exposure to one treatment may result in effects
variables. Possible nested variables are gender, race, (called carryover effects) that persist and influence
socioeconomic status, and education. A nested vari- responses of the subject to later treatments. Also,
able may also be the patients who are cared for on subjects can improve as they become more familiar
specific nursing units or at different hospitals; the with the experimental protocol, which is called a
statistical analysis in this case would be conducted practice effect. They may become tired or bored
as though the unit or hospital were the subject with the study, which is called a fatigue effect. The
rather than the individual patient. Figure 11 – 23 il- direct interaction of one treatment with another,
lustrates the nested design. In actual practice, nurs- such as the use of two drugs, can confound differ-
ing units used in this manner would have to be ences in the two treatments. short
standard
long
39777 Burns (saunm) INTERACTIVE LEFT

top of rh
272 UNIT 2 ■ THE RESEARCH PROCESS base of rh

top of txt
base of txt
Pain Control Primary Nursing Care
Management
Primary Care No Primary Care

Unit Unit Unit Unit Unit Unit Unit Unit
A B C D E F G H
Traditional Care
Unit
A
PRN Medication
Unit
B
Unit
C
Unit
D

Unit
E
Unit
New Approach F
"around the clock"
medication Unit
G
Unit
H

FIGURE 11 – 23 ■ Design using nesting.

Crossover, or counterbalancing, is a strategy de- ity that changes may be due to factors such as
signed to guard against possible erroneous conclu- disease progression, the healing process, or the ef-
sions resulting from carryover effects. With counter- fects of treatment of the disease rather than the
balancing, subjects are randomly assigned to a study treatment. The process of counterbalancing
specific sequencing of treatment conditions. This can become complicated when more than two treat-
approach distributes the carryover effects equally ments are involved. Counterbalancing is effective
throughout all the conditions of the study, thus can- only if the carryover effect is essentially the same
celing them out. To prevent an effect related to from treatment A to treatment B as it is from treat-
time, the same amount of time must be allotted to ment B to treatment A. If one treatment is more
each treatment, and the crossover point must be fatiguing than the other or more likely to modify
related to time, not to the condition of the subject. response to the other treatment, counterbalancing
In addition, the design must allow for an ade- will not be effective. The crossover design controls
quate interval between treatments to dissipate the the variance in the study and thus allows the sam-
effects of the first treatment; this is referred to as a ple size to be smaller. The sample size required to
washout period. For example, the design would detect a significant effect is considerably smaller
specify that each treatment would last 6 days and because the subjects serve as their own controls.
that on the eighth day, each subject would cross Because the data collection period is longer, how-
over to the alternative treatment after a 2-day wash- ever, the rate of subject dropout may increase
out period. (Beck, 1989).
The researcher also must be alert to the possibil- An example of this design is Legault and Gou- short
standard
long
39777 Burns (saunm) INTERACTIVE RIGHT

top of rh
CHAPTER 11 ■ Selecting a Research Design 273 base of rh

top of txt
let’s (1995) study comparing kangaroo and tradi- clinical trial categories were developed specifically base of txt
tional methods of removing preterm infants from for testing experimental drug therapy. Phase I, the
incubators. An excerpt of the study follows. initial testing of a new drug, focuses on determining
................................... the best drug dose and identifying safety effects.
“The intervention was use of the kangaroo or Phase II trials seek preliminary evidence of efficacy
traditional method of maintaining body temperature and side effects of the drug dose determined by the
of preterm infants. The dependent variables were Phase I trial. Phase I and Phase II trials do not
physiologic parameters (skin temperature, heart include control groups or randomization and there-
rate, respiratory rate, and oxygen saturation) mea- fore could not be classified as experimental. They
sured five times with each method. Mother’s satis- are more similar to pilot studies (Whitehead, 1992).
faction was measured at the end of each testing Phase III trials are comparative definitive studies
period and mother’s preference at the end of the in which the new drug’s effects are compared with
experiment. The kangaroo method produced less those of the drug considered standard therapy.
variation in oxygen saturation and longer duration Phase III trials are sometimes referred to as “full
of testing, and it was preferred by most of the scale definitive clinical trials,” suggesting that a de-
mothers.” (Legault & Goulet, 1995, p. 501) ■ cision is made on the basis of the findings as to
whether the experimental drug is more effective
than standard treatment. In some Phase III clinical
Randomized Clinical Trials trials, the sample size is not determined before initi-
Randomized clinical trials have been used in ation of data collection. Rather, data are analyzed at
medicine since 1945. Wooding (1994) describes the intervals to test for significant differences between
strategies that were used to introduce new medical groups. If a significant difference is found, data
therapies before that time: collection may be discontinued. Otherwise, the data
................................... collection will continue and retesting is initiated af-
“Until very recently, the genesis and use of new ter accrual of additional subjects (Meinert & Tonas-
treatments came about by means having little to do cia, 1986; Whitehead, 1992). Phase IV trials occur
with the scientific method. For millennia, the ma- after regulatory approval of the drug, are designed
jority of therapies appear to have evolved by one of to follow patients over time to identify uncommon
three methods: accidental discovery of treatments side effects and test marketing strategies, and do not
with unmistakable efficacy; the use of hypotheses include a control group or randomization (Pianta-
alone, without any experimentation; or the utiliza- dosi, 1997; Wooding, 1994).
tion of experimentation without controls, randomi- Piantadosi (1997) recommends redefining these
zation, blinding . . ., or adequate sample sizes. stages to be broader and applicable to more types
Treatments originating by one of the latter two of trials. He recommends using the following termi-
routes frequently persisted for a very long time de- nology: early development, middle development,
spite a lack of unbiased evidence of their efficacy. comparative studies, and late development. The pur-
Bloodletting, purging, and the use of homeopathic pose of early development trials would be to de-
dosages of drugs are examples. Failure of a treat- velop and test the treatment mechanism (thus, they
ment in any particular case was usually attributed could also be called TM trials). Middle development
by its practitioners to its misuse, to poor diagnosis, or studies would focus on clinical outcomes and treat-
to complicating factors.” (Wooding, 1994, p. 26) ■ ment “tolerability.” Tolerability would have three
components: feasibility, safety, and efficacy; thus,
The methodology for a clinical trial uses strate- Piantadosi (1997) refers to middle development
gies for medical research (Meinert & Tonascia, studies as safety and efficacy or SE trials. In this
1986; Piantadosi, 1997; Pocock, 1996; Whitehead, phase, the researcher would estimate the probability
1992; Wooding, 1994). The Phase I, II, III and IV that patients would benefit from the treatment (or short
standard
long
39777 Burns (saunm) INTERACTIVE LEFT

top of rh
274 UNIT 2 ■ THE RESEARCH PROCESS base of rh

top of txt
experience side effects from it). Performance criteria measure may be death, a nonfatal clinical event, or base of txt
such as success rate might be used. a laboratory test. The period of observation may be
Comparative studies or late development trials, short or long depending on the outcome measure.”
according to Piantadosi (1997), would have defined (Meinert & Tonascia, 1986, p. 3) ■
clinical end points and would address comparative Conceptually, the term clinical trial, as it is used
treatment efficacy (so could be called CTE trials). in the nursing literature, seems to be associated with
These studies would include a concurrent control a Phase III trial and has the following expectations:
group that receives the standard treatment and an
1. The study is designed to be a definitive test of
experimental group that receives the experimental
the hypothesis that the intervention causes the
treatment. Late development studies would be de- defined effects.
signed to identify uncommon side effects, interac- 2. Previous studies have provided evidence that
tions with other treatments, or unusual complica- the intervention causes the desired outcome.
tions. They would be developed as expanded safety 3. The intervention is clearly defined and a proto-
or ES trials. col has been established for its clinical applica-
Elwood (1998) suggests that clinical trial meth- tion.
odology could be used for prevention intervention 4. The study is conducted in a clinical setting, not
studies as well as testing treatments. Murray (1998) in a laboratory.
proposes methods of randomizing groups rather 5. The design meets the criteria of an experimen-
than subjects in prevention studies and explores is- tal study.
sues related to community-based trials such as sam- 6. Subjects are drawn from a reference population
ple mortality. through the use of clearly defined criteria.
Until recently, the term clinical trial has not Baseline states are comparable in all groups in-
been used to describe studies conducted in nursing cluded in the study. Selected subjects are then
research. The clinical trial is perceived by many to randomly assigned to treatment and control
be the “Cadillac” of designs. (There are serious groups; thus the term randomized clinical trial.
criticisms of the clinical trial, however; they are 7. Subjects are accrued individually over time as
discussed in Chapter 12.) If the clinical trial is to be they enter the clinical area, are identified as
used in nursing, the methodology should be rede- meeting the study criteria, and agree to partici-
fined to fit the knowledge-building needs of nurs- pate in the study.
ing. Sidani and Braden (1998) have made a start in 8. The study has high internal validity. The design
this direction by proposing such a methodology, is rigorous and involves a high level of control
which is described in Chapter 13. Criteria for defin- of potential sources of bias that will rule out
ing a study as a clinical trial as opposed to referring possible alternative causes of the effect. The
to it as an experimental study have not been clari- design may include blinding or double-blinding
fied in the nursing literature. to accomplish this purpose. Blinding means that
Meinert & Tonascia (1986) defines a clinical trial either the patient or those providing care to the
as follows: patient are unaware of whether the patient is in
................................... the experimental group or the control group.
“[A] planned experiment designed to assess the Double-blinding means that neither the patient
efficacy of a treatment in man by comparing the nor the caregivers are aware of the group as-
outcomes in a group of patients treated with the signment of the patient.
test treatment with those observed in a comparable 9. The treatment is equal and consistently applied
group of patients receiving a control treatment, to all subjects in the experimental group.
where patients in both groups are enrolled, treated, 10. Dependent variables are measured consistently.
and followed over the same time period. The 11. The proposed study has been externally re-
groups may be established through randomization viewed by expert researchers who have ap- short
or some other method of assignment. The outcome proved the design. standard
long
39777 Burns (saunm) INTERACTIVE RIGHT

top of rh
CHAPTER 11 ■ Selecting a Research Design 275 base of rh

top of txt
12. The study has received external funding suffi- designed to examine nursing strategies for improv- base of txt
cient to allow a rigorous design with a sample ing recovery from cardiac surgery. The study was
size adequate to provide a definitive test of the funded by the National Center for Nursing Re-
intervention. search. Baseline data were collected on individual
13. If the clinical trial results indicate a significant health, psychological status, self-efficacy, family
effect of the intervention, the evidence is suffi- functioning, and coping. These variables were also
cient to warrant application of the findings in used as outcome measures. A model of the design
clinical practice. is illustrated in Figure 11 – 24. Gilliss and Kulkin’s
14. The intervention is defined in sufficient detail (1991) description of their study follows.
so that clinical application can be achieved. ...................................
“Improving Recovery from Cardiac Surgery, or
Clinical trials may be carried out simultaneously in the Family Heart Study, was a randomized clinical
multiple geographical locations to increase sample trial of in-hospital and postdischarge nursing inter-
size and resources and to obtain a more representa- ventions designed to improve individual and family
tive sample (Meinert & Tonascia, 1986). In this recovery from cardiac surgery. Eligible patients
case, the primary researcher must coordinate activi- were between 25 and 75 years of age, spoke En-
ties at all the sites. Meinert & Tonascia (1986) glish, were scheduled for CABG [coronary artery
indicates that the costs per patient per year of study bypass graft] or valve repair, were available for
are less for multicenter studies than for single-center telephone contact for 6 months, and had a consent-
trials. The researcher using this technique must con- ing partner who would care for them in early re-
front several problems. Coordination of a project of covery. Subjects who experienced postsurgical
this type requires much time and effort. Keeping up complications of significance were excluded after
with subjects is critical but may be difficult. Com- surgery. All pairs inducted into the study were as-
munication with and cooperation of staff assisting signed to experimental or control status using a
with the study in the various geographical locations block randomization scheme. Patients received
are essential but sometimes difficult. The researcher “standard care” (Intervention 1) at the two study
may encounter attempts to ignore the protocol and hospitals. This involved viewing a slide-tape pre-
provide traditional care (Fetter et al., 1989; Tyzen- sentation (“Move into Action”) on cardiac risk re-
house, 1981). Meinert & Tonascia (1986) recom- duction prepared by the Santa Clara (California)
mends the development of a coordinating center for Heart Association. Experimental patient/caregiver
multisite clinical trials that will be responsible for pairs received additional education and counseling
receiving, editing, processing, analyzing, and storing (Intervention 2) prior to discharge. This included an
data generated in the study. original slide-tape presentation (“Working Together
The use of the clinical trial is growing in nursing for Recovery”) that was patterned after the Ameri-
research. Brooten and colleagues (1986) conducted can Heart Association tapes and educated the pairs
a clinical trial of early hospital discharge and home about common psychological and relationship re-
follow-up of very low birth weight infants. Burgess sponses to recovery from surgery. Following the
and associates (1987) performed a clinical trial of viewing, each pair met with a study nurse to dis-
cardiac rehabilitation. Later studies defined in the cuss the applicability of the slide-tape content to
literature as clinical trials are those by Clarke their particular situation.
(1999), deMoissac and Jensen (1998), Griebel and ...................................
associates (1998), Ippoliti and Neumann (1998), “Finally, experimental patients and caregivers
Rawl and colleagues (1998), and Turner and associ- were contacted at 1, 2, 3, 4, 6, and 8 weeks after
ates (1998). discharge by graduate nurse researcher assistants
An example of a clinical trial in nursing is the for outcall telephone monitoring and recovery
study by Gilliss and Kulkin (1991), who described coaching (Intervention 3). These calls followed a short
the administrative management of a clinical trial semistructured guide and were intended to rein- standard
long
39777 Burns (saunm) INTERACTIVE LEFT

top of rh
276 UNIT 2 ■ THE RESEARCH PROCESS base of rh

top of txt
base of txt
150 pairs
Random Assignments

71 pairs 79 pairs
(Experimental Protocol) (Usual Care/Intervention 1)
Schedule of Intervention and Evaluation
Pre-op/ Prior to discharge Postdischarge 3 months/12 weeks 6 months/24 weeks
Baseline (in hospital) (weeks 1–8 in home)

Consent Telephone call Telephone call Telephone call
F-COPES (P & S)a
POMS (P & S) Self-efficacy + Self-efficacy + Self-efficacy +
FACES III (P & S)
APGAR (P & S) Activity (P) Activity (P) Activity (P)
Self-efficacy (P)
Family profile/ Self-report/ Self-report/ Self-report/
Expected benefits Recovery index (P) Recovery index (P) Recovery index (P)
Medical abstract (weeks 4 and 8)
Self-efficacy +
Activity (P) Mail follow-up Mail follow-up Mail follow-up
POMS (P & S) POMS (P & S) POMS (P & S)
(week 4) APGAR (P & S) APGAR (P & S)
ZCI/SSS (S) F-COPES (P & S) F-COPES (P & S)
(week 4) FACES III (P & S) FACES III (P & S)
SSS (P) Caregiving index (S)
(week 4) Social support (P & S)

Experimental Interventions Interventions 3
group 1 and 2 (calls at weeks
1, 2, 3, 4, 6, and 8)
Control group Intervention 1
a
P = patient; S = spouse/family caregiver

FIGURE 11 – 24 ■ Design model for clinical trial. (From Gilliss, C. L., & Kulkin, I. L. [1991]). Monitoring nursing interventions
and data collection in a randomized clinical trial. Western Journal of Nursing Research, 13(3), 416– 422.

force hospital-based education programs, detect Primary Prevention and Health
early complications and drug effects, answer ques- Promotion Studies
tions, and offer support for recovery; however, an
overriding aim was to offer personalized nursing Studying primary prevention and health promo-
care.” (Gilliss & Kulkin, 1991, pp. 416–417) ■ tion involves applying a treatment of primary pre-
vention (the cause) and then attempting to measure
the effect (an event that does not occur if the treat-
STUDIES THAT DO NOT USE ment was effective). Primary prevention studies,
then, attempt to measure things that do not happen.
TRADITIONAL RESEARCH DESIGNS
One cannot select a sample to study, apply a treat-
In some approaches to research, the research de- ment, and then measure an effect. The sample must
signs described in this chapter cannot be used. be the community. The design involves examining
These studies tend to be in highly specialized areas changes in the community, and the variables are
that require unique design strategies to accomplish called indicators. A change in an identified indica-
their purposes. Designs for primary prevention and tor is inferred to be a consequence of the effective-
health promotion, secondary analysis, meta-analysis, ness of the prevention program (treatment).
and methodological studies are described here. Specific indicators would depend on the focus of short
standard
long
39777 Burns (saunm) INTERACTIVE RIGHT

top of rh
CHAPTER 11 ■ Selecting a Research Design 277 base of rh

top of txt
prevention. For example, the indicators identified visits but had access to all traditional health care base of txt
by the National Institution on Drug Abuse are services.
(1) changes in drug-related perceptions, attitudes,
knowledge, and action; (2) changes in prevalence
Secondary Analysis
and incidence of drug use as well as drug-related
mortality or morbidity, or both; (3) institutional pol- Secondary analysis design involves studying data
icy or programs, or both; (4) involvement of youth previously collected in another study. Data are reex-
or parent, or both, in the community; and (5) acci- amined with the use of different organizations of
dent rates (French & Kaufman, 1981). Because one the data and different statistical analyses from those
indicator alone would be insufficient to infer effect, previously used. The design involves analyzing data
multiple indicators and statistical analyses appropri- to validate the reported findings, examining dimen-
ate for these indicators must be used. sions previously unexamined, or redirecting the fo-
Van Rossum and colleagues (1991) describe the cus of the data to allow comparison with data from
design of a study to determine the effect of preven- other studies (Gleit & Graham, 1989). As data sets
tive home visits to the elderly by public health accumulate from the research programs of groups of
nurses. The goals of the visits were to increase faculty, secondary analyses can be expected to in-
independence of the elderly and prevent or postpone crease. This approach allows the investigators to
institutional care. Outcome measures, or indicators, examine questions related to the data that were not
used in the study were perceived state of health, originally posed. These data sets may provide
well-being, functional state, mental state, use of opportunities for junior faculty members or grad-
health care services, institutionalization, and mortal- uate students to become involved in a research pro-
ity. A parallel group randomized trial was used. gram.
Subjects were limited to people 75 to 84 years of Of concern in secondary analyses of data is the
age. A questionnaire and letter were sent to all tendency of some researchers to write as many pa-
individuals of this age group living in Weert, a pers as possible from the planned analyses of a
town in The Netherlands with a population of study to increase the number of their publications —
60,000. The response rate was 85%, and 92% of a strategy referred to as “salami slicing.” Research-
those responding agreed to participate. Those al- ers performing secondary analyses should always
ready receiving home nursing care were excluded. identify the original source of data and the previous
A sample of 600 subjects was randomly selected publications emerging from the analysis of that data
through the use of a computer from the possible set. Aaronson (1994) points out the problem with
pool of 1285 questionnaire respondents. this practice, as follows:
To control for extraneous variables, the partici- ...................................
pants were stratified by the following four variables “Fundamentally, each paper written from the
before being randomly assigned to groups: (1) gen- same study or the same dataset must make a dis-
der, (2) composition of household, (3) perceived tinct and significant scientific contribution. Presum-
state of health, and (4) social class. Subjects were ably this is not only the major overriding criterion
then randomly assigned to treatment and control used by reviewers, but also the author’s intent
groups within each stratum, again with the use of a when writing the paper. When a particular paper is
computer. Those receiving the treatment were ran- one of several from the same study, project, or
domly assigned to a public health nurse. Subjects dataset, the author’s responsibility to identify the
were monitored for 3 years. Those receiving public source of the data is that much greater. To lead
health nursing visits were visited every 3 months by readers to think a report is from a new study or a
the same public health nurse. The nurse could be different dataset than that used in the authors’ pre-
contacted daily, and additional visits were made vious work is dishonest, particularly if the second
when needed. Control group subjects received no paper purports to substantiate findings of the first short
standard
long
39777 Burns (saunm) INTERACTIVE LEFT

top of rh
278 UNIT 2 ■ THE RESEARCH PROCESS base of rh

top of txt
one. . . . Apart from the overriding concern about clude the means and standard deviations for each base of txt
‘milking the data,’ the most common objection to group in the study. One of the outcomes of a meta-
multiple articles from a single study is concern analysis is the estimation of a population effect size
about the age of the data. . . . Concerns in nurs- for the topic under study. Because studies seldom
ing about the number of papers generated from a have exactly the same focus, conclusions are never
single study may reflect the emerging status of sec- absolute but do give some sense of unity to knowl-
ondary analysis as a legitimate approach to nursing edge within that area (O’Flynn, 1982).
research. . . . One problem consistently encountered in meta-
“All of the reasons offered for using secondary analyses is that the studies being examined are in-
analysis—answering new questions with existing consistent in design quality. However, researchers
data, applying new methods to answer old ques- conducting meta-analyses have not been successful
tions, the real exigencies of cost and feasibility— in identifying a generally acceptable means to mea-
serve equally to justify the continued use of data sure the research quality of studies. Brown (1991)
collected years ago, by the original investigator of a proposed a research quality scoring method to ac-
large project, as well as by others. . . . The issue complish this important task. Another problem en-
remains one of sound science. The question that countered by researchers performing meta-analyses
must be asked is: Does this particular paper make a is that basic research information is often missing
meaningful and distinct contribution to the scien- from research reports. Calculating effect sizes re-
tific literature?” (Aaronson, 1994, pp. 61–62) ■ quires means and standard deviations in each study
for both experimental and control groups. In addi-
An example of secondary analysis is Lauver and tion, the beginning sample size must be reported as
Tak’s (1995) study, “Optimism and Coping with a well as the sample size at the time of the posttest.
Breast Cancer Symptom.” An excerpt from this Often, the treatment has been poorly described,
study follows. making it difficult to determine the most effective
................................... treatments (Brown, 1991).
“This cross-sectional study was a part of a larger An example of a meta-analysis is Brown and
investigation to explain care seeking for breast can- Grimes’s (1995) meta-analysis of nurse practitioners
cer symptoms. Complementary findings have been and nurse midwives in primary care, an abstract of
described elsewhere (Lauver, 1992, 1994). Data which follows.
were collected in a breast surgery clinic of an ur- ...................................
ban hospital that provided care for the medically “This meta-analysis was an evaluation of patient
indigent. Participants were seeking evaluation for outcomes of nurse practitioners (NPs) and nurse
self-identified breast cancer symptoms, such as a midwives (NMs), compared with those of physi-
lump or discharge. Eligible participants were older cians, in primary care. The sample included 38 NP
than 18 years of age, had no personal history of and 15 NM studies. Thirty-three outcomes were an-
cancer, and could communicate in English.” (Lau- alyzed. In studies that employed randomization to
ver & Tak, 1995, p. 203) ■ provider, greater patient compliance with treatment
recommendations was shown with NPs than with
physicians. In studies that controlled for patient risk
Meta-Analysis Designs
in ways other than randomization, patient satisfac-
Meta-analysis design involves merging findings tion and resolution of pathological conditions were
from many studies that have examined the same greater for NP patients. NPs were equivalent to
phenomenon. The design uses specific statistical MDs on most other variables in controlled studies.
analyses to determine the overall findings from a In studies that controlled for patient risk, NMs used
combined examination of reports of statistical find- less technology and analgesia than did physicians
ings of each study. The statistical values used in- in intrapartum care of obstetric patients. NMs short
standard
long
39777 Burns (saunm) INTERACTIVE RIGHT

top of rh
CHAPTER 11 ■ Selecting a Research Design 279 base of rh

top of txt
base of txt
Is there a treatment?

No Yes

Is the primary purpose Is the treatment tightly
examination of relationships? controlled by the researcher?

No Yes No Yes

Descriptive Will the sample be Quasi-experimental Will a randomly
design studied as a study assigned control
single group? group be used?

No Yes
No Yes

Is the original
correlational sample randomly
design selected?

No Yes

? Experimental
study

FIGURE 11 – 25 ■ Type of study.

achieved neonatal outcomes equivalent to those of the point of appropriate use in a study is 5 years.
physicians. Limitations in data from primary studies An example of this design is Ludington-Hoe and
precluded answering questions of why and under Kasper’s (1995) study, “A Physiologic Method for
what conditions these outcomes apply and whether Monitoring Premature Infants.” The abstract from
these services are cost-effective.” (Brown & Grimes, that study follows.
1995, p. 332) ■
...................................
“Instrumentation capable of handling 12 contin-
uous hours of nine-channel real-time physiologic
Methodological Designs
data sampled at 10 Hz was needed to test within
Methodological designs are used to develop the and between subject variability and preterm infant
validity and reliability of instruments to measure responses to skin-to-skin contact with the mother. A
constructs used as variables in research. The process review of basic electrical components, electrical
is lengthy and complex. The average length of re- principles related to physiologic monitoring, and
searcher time required to develop a research tool to electrophysiology concepts generic to physiologic short
standard
long
39777 Burns (saunm) INTERACTIVE LEFT

top of rh
280 UNIT 2 ■ THE RESEARCH PROCESS base of rh

top of txt
base of txt
Examining sequences across time?

No Yes

One group? Following same subjects
across time?

No Yes
No Yes

Comparative Descriptive
descriptive design Data collected Single unit
design across time? of study?

No Yes No Yes

Cross- Study events Longitudinal Case
sectional partitioned study study
design across time?

No Yes

Trend Repeated
analysis measures of
each subject?

Yes No

Cross-sectional design Longitudinal design
with treatment with treatment
partitioning partitioning

FIGURE 11 – 26 ■ Descriptive studies.

monitoring is presented. The development, specifi- DECISION TREES FOR SELECTING
cation and applications of a new instrument to RESEARCH DESIGNS
monitor premature infant cardiorespiratory adapta-
tions are discussed.” (Ludington-Hoe & Kasper, Selecting a research design involves following short
1995, p. 13) ■ paths of logical reasoning. A calculating mind is standard
long
39777 Burns (saunm) INTERACTIVE RIGHT

top of rh
CHAPTER 11 ■ Selecting a Research Design 281 base of rh

top of txt
base of txt
Describe relationships Predict relationships Test theoretically
between/among between/among proposed
variables? variables? relationships?

Descriptive correlational Predictive correlational Model testing
design design design

FIGURE 11 – 27 ■ Correlational studies.

needed to explore all the possible consequences of of variables is involved. Descriptive designs vary in
using a particular design in a study. In some ways, levels of complexity. Some contain only two varia-
selecting a design is like thinking through the bles, whereas others may include multiple variables.
moves in a chess game. One must carefully think The relationships between variables are identified to
through the consequences of each option. The re- obtain an overall picture of the phenomenon being
search design organizes all the components of the examined. Protection against bias is achieved
study in a way that is most likely to lead to valid through conceptual and operational definitions of
answers to the questions that have been posed. variables, sample selection and size, valid and reli-
To help you select the most appropriate design, a able instruments, and data collection procedures that
series of decision trees have been provided here. achieve some environmental control.
The first decision tree, Figure 11 – 25, helps identify Correlational studies examine relationships be-
the type of study being conducted. The next four tween variables. This examination can occur at sev-
decision trees (see Figures 11 – 26 through 11 – 29) eral levels. The researcher can seek to describe a
assist in the selection of specific designs for each of relationship, predict relationships among variables,
the types of studies. The selection of a design is not or test the relationships proposed by a theoretical
a rigid, rule-guided task. The researcher has consid- proposition. In correlational designs, a large vari-
erable flexibility in choosing a design. The path- ance in the variable scores is necessary to determine
ways within the decision trees are not absolute and the existence of a relationship. Researchers tend to
are to be used as guides. make two serious errors with correlational studies.
First, they often attempt to establish causality by
correlation, reasoning that if two variables are re-
■ lated, one must cause the other. Second, they con-
. . . .S. .U. .M. .M
..A
..R
. .Y
...................... fuse studies in which differences are examined with
Designs have been developed by researchers to studies in which relationships are examined.
meet unique research needs as they emerge. At The purpose of quasi-experimental and experi-
present, nursing research is using designs developed mental designs is to examine causality. The power
by other disciplines. These designs are a useful of the design to accomplish this purpose depends
starting point, but nurse scientists must go beyond on the degree to which the actual effects of the
them to develop designs that will more appropri- experimental treatment (the independent variable)
ately meet the needs of nursing’s knowledge base. can be detected by measurement of the dependent
Descriptive studies are designed to gain more variable. Obtaining an understanding of the true ef-
information about variables within a particular field fects of an experimental treatment requires action to
of study. Their purpose is to provide a picture of control threats to the validity of the findings. short
situations as they naturally happen. No manipulation Threats to validity are controlled through selection standard
long
39777 Burns (saunm) INTERACTIVE LEFT

top of rh
282 UNIT 2 ■ THE RESEARCH PROCESS base of rh

top of txt
No Control group? Yes base of txt

Pretest? Pretest?

No Yes No Yes

One-group Repeated Posttest
posttest only measures? only with
design nonequivalent Repeated
group measures?

Comparison No Yes
with
population No Yes
values? Strategy for Compare
comparison? experimental and
control conditions Compare Interrupted
treatments? time series
No Yes with a
nonequivalent
No Yes control group
Suggest
reevaluating No Yes
design Compare Treatment
variables? replications?
No Yes
Suggest Compare
reevaluating treatment
design and
control Untreated Reversed-treatment
conditions No Yes control nonequivalent
group with control group
One-group pretest and with pretest
posttest only posttest and posttest
design Simple Interrupted
interrupted time series
time with
series nonequivalent
dependent No Yes
No Yes variables

Interrupted Interrupted
Nonequivalent Repeated time series time series
dependent treatment with with
variables design removed repeated
design treatment replications

FIGURE 11 – 28 ■ Quasi-experimental studies.

of subjects, manipulation of the treatment, and reli- amount of control possible to examine causality
able measurement of variables. Experimental de- more closely. To examine cause, one must eliminate
signs with their strict control of variance are the all factors influencing the dependent variable other
most powerful method of examining causality. than the cause (independent variable) being studied.
Quasi-experimental designs were developed to pro- Other factors are eliminated by controlling them.
vide alternate means for examining causality in situ- The three essential elements of experimental re-
ations not conducive to experimental controls. Ex- search are (1) randomization, (2) researcher-con- short
perimental designs are set up to provide the greatest trolled manipulation of the independent variable, standard
long
39777 Burns (saunm) INTERACTIVE RIGHT

top of rh
CHAPTER 11 ■ Selecting a Research Design 283 base of rh

top of txt
No Yes base of txt
Control group?

Post-test Repeated
only control measures?
group design

No Yes

Examine effects Repeated
of confounding measures design
variables?

No Yes

Multiple sites? Blocking?

No Yes No Yes

Pretest/posttest Randomized Comparison of Randomized
control group clinical multiple levels block design
design trials of treatment

No Yes

Examination of Nested
complex relationships designs
among variables in
relation to treatment

No Yes

Examination Multivariate
of multiple designs
causality?

No Yes

Solomon Factorial
four-group designs
design

FIGURE 11 – 29 ■ Experimental studies.

and (3) researcher control of the experimental situa- Studying primary prevention and health promotion
tion, including a control or comparison group. involves applying a treatment of primary prevention
Studies in highly specialized areas require unique (the cause) and then attempting to measure the ef- short
design strategies to accomplish their purposes. fect (an event that does not occur if the treatment standard
long
39777 Burns (saunm) INTERACTIVE LEFT

top of rh
284 UNIT 2 ■ THE RESEARCH PROCESS base of rh

top of txt
was effective). The design involves examining Brown, S. A., & Grimes, D. E. (1995). A meta-analysis of nurse base of txt
changes in the community, and the variables are practitioners and nurse midwives in primary care. Nursing
Research, 44(6), 332– 339.
called indicators. Secondary analysis is the study of Burgess, A. W., Lerner, D. J., D’Agostino, R. B., Vokonas,
data previously collected in another study. Data are P. S., Hartman, C. R., & Gaccione, P. (1987). A randomized
re-examined using different organizations of the control trial of cardiac rehabilitation. Social Science and
data and different statistical analyses from those Medicine, 24(4), 359– 370.
previously used. Meta-analysis is the merging of Campbell, D. T., & Stanley, J. C. (1963). Experimental and
quasi-experimental designs for research. Chicago: Rand
findings from many studies that have examined the McNally.
same phenomenon. Specific statistical analyses are Chang, B. L. (1999). Cognitive-behavioral intervention for home-
used to determine the overall findings from joint bound caregivers of persons with dementia. Nursing Re-
examination of reported statistical findings. Method- search, 48(3), 173– 182.
ological studies are designed to develop the validity Clarke, D. A. (1999). Advancing my health care practice in
aromatherapy. Australian Journal of Holistic Nursing, 6(1),
and reliability of instruments to measure constructs 32– 38.
used as variables in research. The average length of Cook, T. D., & Campbell, D. T. (1979). Quasi-experimentation:
researcher time required to develop a research tool Design and analysis issues for field settings. Chicago: Rand
to the point of appropriate use in a study is 5 years. McNally.
...................................... Cox, D. R. (1958). Planning of experiments. New York: Wiley.
Crombie, I. K., & Davies, H. T. O. (1996). Research in health
care: Design, conduct and interpretation of health services
research. New York: Wiley.
deMoissac, D., & Jensen, L. (1998). Changing IV administration
REFERENCES sets: Is 48 versus 24 hours safe for neutropenic patients with
Aaronson, L. S. (1994). Milking data or meeting commitments: cancer? Oncology Nursing Forum, 25(5), 907– 913.
How many papers from one study? Nursing Research, 43(1), Doering, L. V., Dracup, K., Moser, D. K., Czer, L. S. C., &
60– 62. Peter, C. T. (1999). Evidence of time-dependent autonomic
Aaronson, L. S., & Kingry, M. J. (1988). A mixed method reinnervation after heart transplantation. Nursing Research,
approach for using cross-sectional data for longitudinal infer- 48(6), 308– 316.
ences. Nursing Research, 37(3), 187– 198. Elwood, J. M. (1998). Critical appraisal of epidemiological stud-
Abdellah, F. G., & Levine, E. (1979). Better patient care ies and clinical trials. New York: Oxford University Press.
through nursing research. New York: Macmillan. Fetter, M. S., Fettham, S. L., D’Apolito, K., Chaze, B. A., Fink,
Anderson, V. L., & McLean, R. A. (1974). Design of experi- A., Frink, B. B., Hougart, M. K., & Rushton, C. H. (1989).
ments: A realistic approach. New York: Marcel Dekker. Randomized clinical trials: Issues for researchers. Nursing
Arakawa, S. (1995). Use of relaxation to reduce side effects of Research, 38(2), 117-120.
chemotherapy in Japanese patients. Cancer Nursing, 18(1), Fisher, R. A. (1935). The design of experiments. New York:
60– 66. Hafner.
Barlow, D. H., & Hersen, M. (1984). Single case experimental French, J. F., & Kaufman, N. J. (1981). Handbook for preven-
designs: Strategies for studying behavior change. New York: tion evaluation: Prevention evaluation guidelines. Rockville,
Pergamon. MD: National Institute on Drug Abuse.
Barnard, K. E., Magyary, D. L., Booth, C. L., & Eyres, S. J. Fridriksdottir, M., & McCarthy, D. O. (1995). The effect of
(1987). Longitudinal designs: Considerations and applications caloric density of food on energy intake and body weight in
to nursing research. Recent Advances in Nursing, 17, 37– 64. tumor-bearing rats. Research in Nursing & Health, 18(4),
Beck, S. L. (1989). The crossover design in clinical nursing 357– 363.
research. Nursing Research, 38(5), 291– 293. Gilliss, C. L., & Kulkin, I. L. (1991). Monitoring nursing inter-
Blair, C. E. (1995). Combining behavior management and mu- ventions and data collection in a randomized clinical trial.
tual goal setting to reduce physical dependency in nursing Western Journal of Nursing Research, 13(3), 416– 422.
home residents. Nursing Research, 44(3), 160– 165. Gleit, C., & Graham, B. (1989). Secondary data analysis: A
Brooten, D., Kuman, S., Brown, L. P., Butts, P., Finkler, S. A., valuable resource. Nursing Research, 38(6), 380– 381.
Bakewell-Sachs, S., Gibbons, A., & Delivoria-Papadopoulos, Gray, M. (1998). Introducing single case study research design:
M. (1986). A randomized clinical trial of early hospital dis- An overview. Nurse Researcher, 5(4), 15– 24.
charge and home follow-up of very-low-birth-weight infants. Griebel, B, Wewers, M. E., & Baker, C. A. (1998). The effec-
New England Journal of Medicine, 315(15), 934– 939. tiveness of a nurse-managed minimal smoking-cessation in-
Brown, S. A. (1991). Measurement of quality of primary studies tervention among hospitalized patients with cancer. Oncology short
for meta-analysis. Nursing Research, 40(6), 352– 355. Nursing Forum, 25(5), 897– 902. standard
long
39777 Burns (saunm) INTERACTIVE RIGHT

top of rh
CHAPTER 11 ■ Selecting a Research Design 285 base of rh

top of txt
Hagerty, B. M., & Williams, R. A. (1999). The effects of sense Mason-Hawkes, J., & Holm, K. (1989). Causal modeling: A base of txt
of belonging, social support, conflict, and loneliness on de- comparison of path analysis and LISREL. Nursing Research,
pression. Nursing Research, 48(4), 215– 219. 38(5), 312– 314.
Hanneman, S. K. G. (1994). Multidimensional predictors of suc- McCain, L. J., & McCleary, R. (1979). The statistical analysis of
cess or failure with early weaning from mechanical ventila- the simple interrupted time-series quasi-experiment. In T. D.
tion after cardiac surgery. Nursing Research, 43(1), 4– 10. Cook & D. T. Campbell (Eds.), Quasi-experimentation: De-
Harris, R. B., & Hyman, R. B. (1984). Clean vs. sterile tracheot- sign and analysis issues for field settings (pp. 233– 293).
omy care and level of pulmonary infection. Nursing Re- Chicago: Rand McNally.
search, 33(2), 80– 85. McCurren, C., Dowe, D., Ratle, D., & Looney, S. (1999). De-
Henker, R., Bernardo, L. M., O’Connor, K., & Sereika, S. pression among nursing home elders: Testing an intervention.
(1995). Evaluation of four methods of warming intravenous Applied Nursing Research, 12(4), 185– 195.
fluids. Journal of Emergency Nursing, 21(5), 385– 390. Meinert, C. L., & Tonascia, S. (1986). Clinical trials: Design,
Hjelm-Karlsson, K. (1991). Using the biased coin design for conduct, and analysis. New York: Oxford University Press.
randomization in health care research. Western Journal of Mimnaugh, L., Winegar, M., Mabrey, Y., & Davis, J. E. (1999).
Nursing Research, 13(2), 284– 288. Sensations experienced during removal of tubes in acute
Ippoliti, C., & Neumann, J. (1998). Octreotide in the manage- postoperative patients. Applied Nursing Research, 12(2), 78–
ment of diarrhea induced by graft versus host disease. Oncol- 85.
ogy Nursing Forum, 25(5), 873– 878. Mirolo, B. R., Bunce, I. H., Chapman, M., Olsen, T., Eliadis, P.,
Jennings-Dozier, K. (1999). Predicting intentions to obtain a Pap Hennessy, J. M., Ward, L. C., & Jones, L. C. (1995). Psy-
smear among African American and Latina women: Testing chosocial benefits of postmastectomy lymphedema therapy.
the Theory of Planned Behavior. Nursing Research, 48(4), Cancer Nursing, 18(3), 197– 205.
198– 205. Murray, D. M. (1998). Design and analysis of group-randomized
Johnson, L. S., Rozmus, C., & Edmisson, K. (1999). Adolescent trials. New York: Oxford University Press.
sexuality and sexually transmitted diseases: Attitudes, beliefs, Newman, M. A. (1986). Health as expanding consciousness. St.
knowledge, and values. Journal of Pediatric Nursing, 14(3), Louis: Mosby.
177– 185. O’Flynn, A. I. (1982). Meta-analysis. Nursing Research, 31(5),
Lamb, K. (1979). Effect of positioning of postoperative frac- 314– 316.
tured-hip patients as related to comfort. Nursing Research, Ottenbacher, K. (1992). Impact of random assignment on study
28(5), 291– 294. outcome: An empirical examination. Controlled Clinical Tri-
Lauver, D. (1992, October). Promoting prompt care seeking for als, 13(1), 50– 61.
breast cancer symptoms in Caucasian and African-American Piantadosi, S. (1997). Clinical trials: A methodologic perspec-
women. Paper presented at University of Illinois Conference: tive. New York: Wiley.
Reframing Women’s Health: Multidisciplinary Research and Pocock, S. J. (1996). Clinical trials: A practical approach. New
Practice, Chicago, IL. York: Wiley.
Lauver, D. (1994). Care-seeking behavior with breast cancer Quayhagen, M. P., Quayhagen, M., Corbeil, R. R., Roth, P. A.,
symptoms in Caucasian and African-American women. Re- & Rodgers, J. A. (1995). A dyadic remediation program for
search in Nursing & Health, 17(6), 421– 431. care recipients with dementia. Nursing Research, 44(3), 153–
Lauver, D., & Tak, Y. (1995). Optimism and coping with a 159.
breast cancer symptom. Nursing Research, 44(4), 202– 207. Rawl, S. M., Easton, K. L., Kwiatkowski, S., Zemen, D., &
Legault, M., & Goulet, C. (1995). Comparison of kangaroo and Burczyk, B. (1998). Effectiveness of a nurse-managed fol-
traditional methods of removing preterm infants from incuba- low-up program for rehabilitation patients after discharge.
tors. Journal of Gynecologic and Neonatal Nursing, 24(6), Rehabilitation Nursing, 23(4), 204– 209.
501– 506. Reichardt, C. S. (1979). The statistical analysis of data from
Lepczyk, M., Raleigh, E. H., & Rowley, C. (1990). Timing of nonequivalent group designs. In T. D. Cook & D. T. Camp-
preoperative patient teaching. Journal of Advanced Nursing, bell (Eds.), Quasi-experimentation: Design and analysis is-
15(3), 300– 306. sues for field settings (pp. 147– 206). Chicago: Rand Mc-
Lewis, L. L. (1995). One year in the life of a woman with Nally.
premenstrual syndrome: A case study. Nursing Research, Rotter, J. B. (1954). Social learning and clinical psychology.
44(2), 111– 116. New York: Prentice-Hall.
Ludington-Hoe, S., & Kasper, C. E. (1995). A physiologic Sandelowski, M. (1996). One is the liveliest number: The case
method for monitoring premature infants. Journal of Nursing orientation of qualitative research. Research in Nursing &
Measurement, 3(1), 13– 29. Health, 19(6), 525– 529.
Martinson, I. M., Leavitt, M., Liu, C., Armstrong, V., Hornber- Shiao, S. P. K., Youngblut, J. M., Anderson, G. C., DiFiore,
ger, L., Zhang, J., & Han, X. (1999). Comparison of Chinese J. M., & Martin, R. J. (1995). Nasogastric tube placement:
and Caucasian families caregiving to children with cancer at Effects on breathing and sucking in very-low-birth-weight short
home: Part I. Journal of Pediatric Nursing, 14(2), 99– 109. infants. Nursing Research, 44(2), 82– 88. standard
long
39777 Burns (saunm) INTERACTIVE LEFT

top of rh
286 UNIT 2 ■ THE RESEARCH PROCESS base of rh

top of txt
Sidani, S. & Braden, C. J. (1998). Evaluating nursing interven- a Dutch study to test preventive home visits to the elderly. base of txt
tions: A theory-driven approach. Thousand Oaks: Sage. Nursing Research, 40(3), 185– 188.
Spector, P. E. (1981). Research designs. Beverly Hills, CA: Waltz, C. F., & Bausell, R. B. (1981). Nursing research: Design,
Sage. statistics and computer analysis. Philadelphia: F. A. Davis.
Sterling, Y. M., & McNally, J. A. (1992). Single-subject re- Whitehead, J. (1992). The design and analysis of sequential
search for nursing practice. Clinical Nurse Specialist, 6(1), clinical trials. New York: Ellis Horwood.
21– 26. Whitney, J. D., Stotts, N. A., & Goodson, W. H., III. (1995).
Stewart, B. J., & Archbold, P. G. (1992). Nursing intervention Effects of dynamic exercise on subcutaneous oxygen tension
studies require outcome measures that are sensitive to and temperature. Research in Nursing & Health, 18(2), 97–
change, part 1. Research in Nursing & Health, 15(6), 477– 104.
481. Wikblad, K., & Anderson, B. (1995). A comparison of three
Stout, R. L., Wirtz, P. W., Carbonari, J. P., & Del Boca, F. K. wound dressings in patients undergoing heart surgery. Nurs-
(1994). Ensuring balanced distribution of prognostic factors ing Research, 44(5), 312– 316.
in treatment outcome research. Journal of Studies in Alcohol- Wilbur, J., Zoeller, L. H., Talashek, M., & Sullivan, J. A.
ism, 12(Suppl.), 70– 75. (1990). Career trends of master’s-prepared family nurse prac-
Trochim, W. M. K. (1986). Advances in quasi-experimental de- titioners. Journal of the American Academy of Nurse Practi-
sign and analysis. San Francisco: Jossey-Bass. tioners, 2(2), 69– 78.
Turner, J. G., Clark, A. J., Gauthier, D. K., & Williams, M. Williams, P. D., Press, A., Williams, A. R., Piamjariyakul, U.,
(1998). The effect of therapeutic touch on pain and anxiety Keeter, L. M., Schultz, J., & Hunter, K. (1999). Fatigue in
in burns patients. Journal of Advanced Nursing, 28(1), 10– mothers of infants discharged to the home on apnea moni-
20. tors. Applied Nursing Research, 12(2), 69– 77.
Tyzenhouse, P. S. (1981). Technical notes: The nursing clinical Wooding, W. M. (1994). Planning pharmaceutical clinical trials:
trial. Western Journal of Nursing Research, 3(1), 102– 109. Basic statistical principles. New York: Wiley.
U. S. Department of Health and Human Services (1968). The Yin, R. (1984). Case study research: Design and methods. Ap-
Framingham Study: An epidemiological investigation of car- plied Social Research Methods Series, Vol. 5. Beverly Hills,
diovascular disease (USDHHS Publication No. RC667F813). CA: Sage.
Bethesda, MD: Author. Youssef, F. A. (1990). The impact of group reminiscence coun-
Van Rossum, E. V., Frederiks, C., Philipsen, H., Lierop, J. K., seling on a depressed elderly population. Nurse Practitioner,
Mantel, A., Portengen, J., & Knipschild, P. (1991). Design of 15(4), 32– 38.

short
standard
long