Qualitative Data Analysis
Qualitative Data Analysis
Qualitative
Data Analysis
A Methods Sourcebook
Edition
xiii
xvii
xxi
xxiii
17
55
69
107
121
161
193
221
255
275
323
Chapter 13 - Closure
339
345
References 363
Author Index
369
Subject Index
373
Contents
Introduction
Display Format Options
Matrices
Networks
Timing of Display Design
Formatting the Matrix Template
Entering Matrix and Network Data
Making Inferences and Drawing Conclusions From Matrices and Networks
The Methods Profiles
Closure and Transition
Chapter 5
107
Introduction
Lengthy, unreduced text in the form of interview transcripts, field notes, documents,
and so on is cumbersome because it is dispersed over many pages and is not easy to see
as a whole. It is sequential rather than simultaneous, making it difficult to look at two
or three variables at once. Comparing several extended texts carefully is very difficult.
It is usually poorly ordered, can get very bulky, and can make us feel monotonously
overloaded. The same objections apply with even stronger force for readers of our final
reports. They need, if not deserve, a concise delivery of what we analyzed. And in
this highly visual culture, showing rather than telling can make a more effective and
memorable impact on our audiences.
A major purpose of this text is to encourage the creation and dissemination of
matrix and network displays for qualitative data. The central argument of this book is
You know what you display. Credible and trustworthy analysis requires, and is driven
by, displays that are focused enough to permit a viewing of a full data set in the same
location and are arranged systematically to answer the research questions at hand. A
full data set does not, of course, mean the complete corpus of interview transcripts,
field notes, documents, and so on. Rather, the condensed, distilled data presented are
drawn from the full range of persons, events, and processes under study. With extended
text, there can easily be selective stacking of the data. An organized display wards off
this problem.
The idea of display is central to this book. By display we mean a visual format
that presents information systematically so the user can draw conclusions and take
needed action. Although such displays may sometimes be busy, they will never be
monotonous. Most important, the chances of drawing and verifying conclusions are
much greater than for extended text, because the display is arranged coherently to
permit careful comparisons, detection of differences, noting of patterns and themes,
seeing trends, and so on.
Quantitative researchers have software packages that can develop publishable
tables, graphs, and charts. Qualitative researchers have CAQDAS programs for our
unique approaches to data analysis. And even basic Microsoft Office programs such
as Word and Excel are sufficient for most matrix and network displays. But the
qualitative analyst has to handcraft appropriate data display formats because each
project is unique. As yet, there are few familiar, agreed-on data setups among qualitative
researchers, so each analyst has to adapt those of others or invent new ones. The display
ideas we offer in this book are nothing more than thatideas, not prescriptions, for
qualitative data display.
Not everyone loves matrices and network displaysand not everyone thinks
visually. But displaying your condensed data in a systematic way has immense
consequences for your understanding. It requires you to think about your research
questions and what portions of your data are needed to answer them; it requires you
to make full analyses, ignoring no relevant information; and it focuses and organizes
your information coherently. These advantages are repeated when you include
displays in a final report; the reader can re-create your intellectual journey with some
confidence.
108
Matrices
A matrix is essentially the intersection of two lists, set up as rows and columns. Lets
take a look at a sample format, explaining and labeling it as we go.
Display 5.1, a table drawn with Microsoft Word software, is aimed at understanding
the effects of assistance supplied to a school sitethe Masepa Caseby various
sources. This was part of a school improvement study that observed how a new project
innovation was implemented. The matrix format calls for the researcher to address five
related variables, to distinguish two of them according to time, to pool responses, to
align some responses along an evaluative scale, and to explain the response pattern
for each type of assistance source. Here, condensed information from 30 pages of field
notes has been packed into a single page.
Note that the data are abstractions: There are no quotes, and generalizations and
other inferential remarks appear in the last two columns. The Longer-Run Consequences
and Researcher Explanations are not direct condensations of participants remarks or
of researchers observations. Rather, for any given consequence, such as the one in the
top row (Users are helped administratively and substantively, feel obliged to do ECRI
[Exemplary Center for Reading Instruction] with minor adaptations), the researcher
Chapter 5
109
110
1. provides materials
2. demonstrates, models
3. answers requests
4.encourages
5. circulates, controls
1.comparing practices with
others
2. debugging, complaining
3. learning about new parts
4.encouragement
1. sharing materials
2. exchanging tips, solutions
3. comparing, encouraging
1. tips for presentations
2.solution to short-term
problems
3.encourages
4. serves as successful model
++
+
++
++
++
+
+
+
+
+
++
++
Helping Teacher
+
+
1. facilitates practice
2.helps expand beyond core
format
3. maintains effort
4.stimulates
1. increases stock
2.new ideas, practices; problems
solved
3. motivates, stimulates
1. encourages, regulates
1. pressures non-users
2.building administrators have
material, administrative
support
Short-Run Effects
(Users State)
reliable, unthreatening
backup provided in school
increases commitment,
regulates use (decreases
deviance)
program is perceived
as supported, assisted,
protected by central
office
Longer-Run
Consequences
Source: Miles, M. B., & Huberman, A. M. (1994). Qualitative data analysis: An expanded sourcebook (2nd ed.). Thousand Oaks, CA: Sage Publications.
Legend
++ = very
effective
+ = effective
= mixed
effective
= ineffective
Trainers in
Target School,
Other School
Teacher-Users in
Other Schools:
Target Schools
1. promotes ECRI
2.answers building
administration, trainers
requests
+
++
Central Office
Administration
User-Helping
Teacher
Meetings
1. authorizes changes
2. eases schedules
3. controls fidelity
4. consults, offers solutions
Types Provided
++
++
+
Users
Assessment
Building
Administration
Location
Display 5.1
additional source of
assistance, which increases
as number of users grows
personalized in-service
mechanism, with both
training and assistance
allows for mastery and
spread of ECRI in faithful
format
administration, authority,
servicing, availability and
flexibility lead to sustained,
faithful implementation
model
Researcher Explanations
has looked at the data segments in the three preceding columns, checked to see
whether they covary in some patterned way, and drawn a second-order generalization.
In this case (see the first row, first columnBuilding Administration), themes such as
Eases schedules, Consults, offers solutions, Relieves pressure, encourages, Helps
early implementation, and the Users Assessment of generally positive magnitude codes
all suggested the reception of help and a sense of user obligation for reasonably faithful
implementation. A similar process of inductive inference occurs under Researcher
Explanations.
Overall, the matrix is a tabular format that collects and arranges data for easy viewing
in one place, permits detailed analysis, and sets the stage for later cross-case analysis
with other comparable cases or sites. As the chapters progress, youll learn how matrices
can order data by case or time and can arrange and stack their cells in such a way as
to create meta-matrices, which contain multiple rows within a single row and multiple
columns within a single column.
Networks
A network is a collection of nodes or points connected by links or lines that display
streams of participant actions, events, and processes. Networks lend themselves well
to a case-oriented approach that re-creates the plot of events over time, as well as
showing complex interrelationships between variables. They give us the kinds of
narratives that tend to get chopped up analytically in matrices. They are very helpful
when you want to focus on multiple variables at the same time for readily analyzable
information at a glance.
Display 5.2 is a network model from McCammon et al.s (2012) study of how high
school speech classes influenced and affected adolescents and their adulthood after
graduation. The network is an outline of plot points for an accompanying research
narrative.
Display 5.2
A Network Model of Lifelong Impact From High School Speech Participation
Adolescents
Friendships
Acceptance
and Status
Competitions
Winning
Speech
Classes
Good
Coaching
Sense of
Belonging
Positive
Affects
Adolescent
and Adult
Confidence
Source: Saldaa, J. (2013). The coding manual for qualitative researchers (2nd ed.). Thousand Oaks, CA: Sage
Publications.
Chapter 5
111
112
Display 5.3
A QDA Miner 4 3-D Map of a Codes Network
Furthermore, for any given research question or issue, you can develop many
different displays (literally dozens) using the same set of variables. Each display makes
somewhat different assumptions; each has trade-offs among advantages and costs.
Another caution is that display formats nearly always evolve, too. The later ones are
more data sensitive than the earlier ones, as things become clearer.
So our general advice is to generate rough formats early during data collection and
revise them to a firmer state closer to the end of data collection, when they can be
grounded more contextually and empirically. Expect to make several passes or iterations
before the first format of the display is working right. Test a proposed format by entering
data. Unworkable or confusing formats, or those that do not incorporate all of the
relevant data, will show themselves rapidly.
Chapter 5
113
will give you reasonable answers to the questions you are asking or suggest promising
new ways to lay out the data to get answers. At a deeper level, the message of this book
is not Use these matrices but Think display. Adapt and invent formats that will serve
you best.
Given these choices, what can we say informally about the best and easiest ways to
build the templates/outlines/frameworks of matrix displays? We state these briskly, as
friendly advice rather than as harsh prescription:
Look at your research question(s) and key variables, and think of the data that
are or will be available. Sketch the matrix outline roughly using paper and
pencil.
Get a colleague to look at your initial format, to help you detect the assumptions
you are making and to suggest alternative ways to display your data.
Dont try to include more than a dozen or so variables in rows or columns; five or
six is more manageable. If you are drawn to a design with larger numbers of
variables, plan to cluster or partition them as meta-matrices. In effect, regroup
the matrix into streams or adjacent families.
The simplest matrices are organized in two dimensions. You have a choice to
move to more complexity if the data demand it by creating partitions for metamatricesrows within a row and/or columns within a column (to be illustrated
in forthcoming chapters).
If the matrix is an ordered one, expect to transpose rows and columns for a
while until you have a satisfactory version. Most text-based software, database
management, and CAQDAS programs can do this quite easily.
Always stay open to the idea of adding new rows or columns, even late in your
analysis operations.
Keep in mind that any particular research question may require a series of
displays; for example, an initial partially ordered descriptive matrix may
lead to a small summary table and then to a network display. Think ahead
to this possibility, but allow new matrix forms to emerge as the analysis
proceeds.
Creating matrix templates is usually a matter of a few minutes; revising them as the
early items of data are entered is also quick work. Virtually all CAQDAS programs can
enable reformatting as data are entered and coded. The time taken to enter data into a
display actually depends on the following: (a) the number of variables or dimensions in
the display, (b) the number of respondents and cases, and (c) the kind and number of
transformations made.
114
Which types of row and column headings or network bin labels are possible?
The set is almost infinite, ranging from cases to variables. To illustrate, here is just a
sample of types adapted from Gobo (2008), Lofland, Snow, Anderson, and Lofland
(2006), and Bogdan and Biklen (2007). For concreteness, we use examples from
health care settings:
Individuals:
Roles:
Sites as wholes:
Chapter 5
115
Even a dense matrix displays only a very small percentage of the available data.
There is always a great deal of selection and condensation from the mass of field
notes. Be aware of how you make that selection and how you boil the data
down. You are not throwing away your field notesyou can always refer back
to the full material.
More information is better than less: Too thin cell entries keep you away from the
meaning of the data.
Be clear about the forms and types of data you want to enter: direct quotes,
paraphrases, general summary judgments, ratings, and so on.
Use codes and software search functions to locate key material. Entering these
data is much easier with text-based software, database management, or a
CAQDAS program having multiple screens; they permit you to retrieve coded
chunks to one screen or region and to select/edit/condense them on another.
Keep an explicit record of the decision rules you followed in selecting data
chunks for entry (e.g., the extent of agreement among respondents or data
types, the intensity of respondents feelings, the basis for making judgments or
ratings). Otherwise, you may delude yourself retrospectively, forget how you did
it, or shift your decision rules during the process.
When data are missing, ambiguous, or were not asked for from certain respondents,
show this explicitly in the display.
Dont lock up your format until later in the process. Entering data tests the adequacy,
realism, and helpfulness of the display format. Keep revising it as needed.
Get a colleague to review your display, along with your decision rules and
written-up field notes, to check the procedural adequacy of your work. Such
audits are time-consuming, but used selectively, they are an important check on
the confirmability of the procedures you used.
You always have choices about the level and type of data to be entered. For example,
you can include the following:
116
Summaries, paraphrases, or
abstracts:
Researcher explanations:
Patient uncertainty about diagnosis is a function of lifethreatening conditions and use of denial as a buffer.
Ratings or summarized
judgments:
Its always helpful to start with a quick scana squint analysis or eyeballing
down columns and across rows and through network pathways to see what
jumps out. Then, verify, revise, or disconfirm that impression through a more
careful review.
Any given display will always have multiple tactics used on it. The ones we
have used most frequently for drawing first conclusions are noting patterns,
themes; making contrasts, comparisons; clustering; and counting.
Remember that analysis usually has to go beyond descriptive summation and reach
toward explanation. Clarify the conceptual implications of your conclusions
that is, how they tie into your or someone elses theory of social behavior.
Analyses that yield verifiable but poor-meaning conclusions are of little use to
anyone.
Conclusions about data displays normally appear in what we call an analytic text
or narrative. Analytic narrative draws attention to the features of the displayed data
and makes sense of them, knitting them together and permitting the analyst to draw
conclusions and add interpretations. It also encourages a return to the field notes to
consult information not in the display and to add it to the text for further clarity. Since
a display does not speak for itself, analytic text does not stand alone without reference
to the display.
Chapter 5
117
In fact, the act of writing text as you ruminate over the meaning of a display is itself
a focusing and forcing device that propels further analysis. Writing does not come after
analysis; it is analysis, happening as the writer thinks through the meaning of data in the
display. Writing is thinking, not the report of thought.
Stay open on the question of whether the display format you use for analysis should
be the same display used for final reporting (the answer is often yes). Readers can see
for themselves how conclusions were drawn, rather than being handed summarized
study results to be taken on faith. Sometimes, of course, a display will be used for
intermediate purposes and will not necessarily be shown to readers of the final report.
The tradition of presenting basic data is deeply ingrained in reports of quantitative
data analysis, so much so that it would be unthinkable for a researcher to present
conclusions without data tables or, at least, without reference to working documents
containing them. We believe that the same norms should apply to qualitative researchers
working within particular traditional genres of inquiry (e.g., ethnography, grounded
theory, content analysis, evaluation research, etc.). Some data displays should be a
normal part of reporting findings and conclusions. In some instances, of course, a
summary table or a boiled-down version may suffice. Where basic matrices or networks
are not presented, you owe the reader a clear explanation of the analysis methods used
to get to the text. In any case, remember what the reader needs from displays. We can
hardly do better than quote Tufte (1986), speaking primarily of quantitative work but
with relevance for qualitative researchers:
What we are seeking in graphic and tabular displays of information is the clear
portrayal of complexity. Not the complication of the simple; rather the task of
the designer is to give visual access to the subtle and the difficultthat is, the
revelation of the complex. (p. 80)
As the saying goes, Things can be complex without being complicated.
Applications: This section outlines the purposes of the method and its
recommended uses for particular research studies or goals.
Example: A study drawn from the coauthors previous work is used to illustrate
the display, its design, and data entry.
Analysis: The discussion shifts to how the assembled data in the matrix or
network were analyzed and interpreted by the researchers. Recommendations
for reader applications are also provided.
The profiles vary in length, depending on the complexity of the method. They are
grouped into subcategories within each chapter, when appropriate, if they share a
common purpose.
118
We also emphasize here that the names we gave to our methods and displays were
for convenient description, not for standardization. Who cares whether it is a caseordered descriptive meta-matrix or a predictor-outcome-consequences matrix? The
names may sound like jargon, but dont worry about that; the issue is how any given
display works and how it can further your understanding of the data. All of the methods
and displays can be adapted to meet the needs of any particular study. If they seem too
complex, simplify; they can be modified easily for studies where the cases are a few
individuals. Not every single method has to be used for every single study. Take only
what is useful for your research purposes and goals. Focus on the doable while keeping
as much richness as you can.
Chapter 5
119