This chapter:
stresses the need for a systematic analysis of qualitative data;
emphasizes the central role of the person doing the analysis, and warns about some
deficiencies of the human as analyst;
discusses the advantages and disadvantages of using specialist computer software;
explains the Miles and Huberman approach to analysis which concentrates on reducing
the bulk of qualitative data to manageable amounts and on displaying them to help
draw conclusions;
suggests thematic coding analysis as a generally useful technique when dealing with
qualitative data;
reviews the widely used grounded theory approach;
summarizes a range of alternative approaches; and
finally, considers issues involved in integrating qualitative and quantitative data in
multi‐strategy designs.
Introduction
Qualitative data have been described as an ‘attractive nuisance’ (Miles, 1979). Their attractiveness
is undeniable. Words, which are by far the most common form of qualitative data, are a
speciality of humans and their organizations. Narratives, accounts and other collections of
words are variously described as ‘rich’, ‘full’ and ‘real’, and contrasted with the thin abstractions
of number. Their collection is often straightforward. They lend verisimilitude to reports.
The ‘nuisance’ refers to the legal doctrine that if you leave an attractive object, such as an
unlocked car, where children can play with it, you may be liable for any injuries they sustain.
The analysis and interpretation
of qualitative data
CHAPTER 18
REAL WORLD RESEARCH460
Naive researchers may be injured by unforeseen problems with qualitative data. This can occur
at the collection stage, where overload is a constant danger. But the main difficulty is in their
analysis. There is no clear and universally accepted set of conventions for analysis corresponding
to those observed with quantitative data. Indeed, many ‘qualitative’ workers would resist
their development, viewing this enterprise as more of an art than a science. But for those who do
wish to work within the kind of scientific framework advocated in this book, and who wish to
persuade scientific or policy‐making audiences, there are ways in which qualitative data can be
dealt with systematically. This chapter seeks to provide an introduction to that task.
In the typology of research strategies that has been adopted in this text, the various types of
flexible and multiple‐strategy designs are the prime generators of large amounts of complex
qualitative data.
Qualitative data are often useful in supplementing and illustrating the quantitative data
obtained from an experiment or survey. Small amounts of qualitative data used as an adjunct
within a largely quantitative fixed design study will not justify detailed and complex analysis.
Often the need is simply to help the account ‘live’ and communicate to the reader through the
telling quotation or apt example. However, when methods generating qualitative data form the
only, or a substantial, aspect of the study, then serious and detailed attention needs to be given
to the principles of their analysis.
Two assumptions
1. If you have a substantial amount of qualitative data you will use some kind of software package to deal
with it. Standard software, even a simple word‐processing package, can do much to reduce
the sheer tedium of qualitative data analysis (see Hahn, 2008 on the use of standard Microsoft
Office software for a small qualitative project). For anything other than a small amount of
data, the amount of drudgery you can avoid, and the ease with which you can relate to the
data, make the use of a computer near to essential. There are also specialist qualitative data
analysis packages which aid the process even more. See Appendix B for further details.
2. Unless you already have experience yourself, you will be helped or advised by someone who does have
experience in this type of analysis. The dominant model for carrying out qualitative analysis has
in the past been that of apprenticeship. Without accepting all the implications of such a model
(which tends, for example, to include a principled inarticulacy about process) there is
undoubted value in expert advice. The help provided by software is very different from that
in quantitative analysis. There the ’expert’s’ role is largely to point you towards an appropriate
test and to ensure that you understand the outcome. In qualitative data analysis, both the
experienced person and the computer help you through a not very well specified process.
Types of qualitative analysis
Box 18.1 provides a typology of possible approaches. Quasi‐statistical approaches rely largely
on the conversion of qualitative data into a quantitative format and have been covered under the
heading of content analysis in Chapter 15. See also Abeyasekera (2005) who provides a range of
suggestions.
THE ANALYSIS AND INTERPRETATION OF QUALITATIVE DATA 461
Thematic coding analysis is discussed in this chapter as a straightforward general approach
which can be used in a wide variety of settings. The widespread popularity of grounded theory
as a basis for qualitative data analysis demands its coverage in any treatment of the topic. There
are a large number of other approaches, many of which call for an extensive understanding of
their theoretical foundations. A brief introduction is provided in a later section of the chapter.
Whatever approach is taken, the researcher has the responsibility of describing it in detail.
You have to be able to demonstrate the quality of your analysis, including how you got from the
data to your interpretation.
BOX 18.1
Different approaches to qualitative analysis
1. Quasi‐statistical approaches
Uses word or phrase frequencies and inter‐correlations as key methods of determining
the relative importance of terms and concepts.
Typified by content analysis.
2. Thematic coding approach
A generic approach not necessarily linked to a particular (or any) theoretical perspective.
All, or parts of, the data are coded (i.e. identified as representing something of potential
interest) and labelled.
Codes with the same label are grouped together as a theme.
Codes and themes occurring in the data can be determined inductively from reviewing
the data, and/or from relevance to your research questions, previous research, or
theoretical considerations.
The themes then serve as a basis for further data analysis and interpretation.
Makes substantial use of summaries of the themes, supplemented by matrices, network
maps, flow charts and diagrams.
Can be used on a purely descriptive or exploratory basis, or within a variety of theoretical
frameworks.
3. Grounded theory approach
A version of thematic coding where, as a matter of principle, the codes arise from
interaction with the data.
Codes are based on the researcher’s interpretation of the meanings or patterns in the
texts.
Used to develop a theory ‘grounded’ in the data.
Can be used very prescriptively following rules laid down by founders of the
approach, or as a general style of analysis using a specialized terminology for different
types of coding.
Note: There are other specialized approaches, including discourse and conversation
analysis, and the analysis of narratives (i.e. stories in written, spoken or other forms). See
Chapter 15, p. 361.
REAL WORLD RESEARCH462
BOX 18.2
Deficiencies of the human as analyst
1. Data overload. Limitations on the amount of data that can be dealt with (too much to
receive, process and remember).
2. First impressions. Early input makes a large impression so that subsequent revision is
resisted.
3. Information availability. Information which is difficult to get hold of gets less attention
than that which is easier to obtain.
4. Positive instances. There is a tendency to ignore information conflicting with hypotheses
already held and to emphasize information that confirms them.
5. Internal consistency. There is a tendency to discount the novel and unusual.
6. Uneven reliability. The fact that some sources are more reliable than others tends to be
ignored.
7. Missing information. Something for which information is incomplete tends to be
devalued.
8. Revision of hypotheses. There is a tendency either to over‐ or to under‐react to new
information.
9. Fictional base. The tendency to compare with a base or average when no base data is
available.
10. Confidence in judgement. Excessive confidence is rested in one’s judgement when once
it is made.
11. Co‐occurrence. Co‐occurrence tends to be interpreted as strong evidence for correlation.
12. Inconsistency. Repeated evaluations of the same data tend to differ.
(adapted and abridged from Sadler, 1981, pp. 27–30)
The importance of the quality of the analyst
The central requirement in qualitative analysis is clear thinking on the part of the analyst.
Fetterman (1998) considers that the analysis is as much a test of the researcher as it is a test of the
data: ‘First and foremost, analysis is a test of the ability to think – to process information in a
meaningful and useful manner’ (p. 93). As emphasized at the beginning of Part V, qualitative
analysis remains much closer to codified common sense than the complexities of statistical analysis
of quantitative data. However, humans as ‘natural analysts’ have deficiencies and biases
corresponding to the problems that they have as observers (see Chapter 14, p. 320). Some of
these are listed in Box 18.2. Systematic, documented approaches to analysis help minimize the
effects of these human deficiencies. However, there is an inescapable emphasis on interpretation
in dealing with much qualitative data which precludes reducing the task to a defined formula.
Hence, the suggestions made in this chapter are more in the nature of guidelines rather than
tight prescriptions.
THE ANALYSIS AND INTERPRETATION OF QUALITATIVE DATA 463
Common features of qualitative data analysis
While the possible approaches to analysis are very diverse, there are recurring features. Miles,
Huberman and Saldana (2014, p. 10) give a sequential list of what they describe as ‘a fairly classic
set of analytic moves’:
giving labels (‘codes’) to chunks (words, phrases, paragraphs, or whatever), labelling them as
examples of a particular ‘thing’ which may be of interest in the initial set of materials obtained
from observation, interviews, documentary analysis, etc.;
adding comments, reflections, etc. (commonly referred to as ‘memos’);
going through the materials trying to identify similar phrases, patterns, themes, relationships,
sequences, differences between sub-groups, etc.;
using these patterns, themes, etc. to help focus further data collection; gradually elaborating a
small set of generalizations that cover the consistencies you discern in the data; and
linking these generalizations to a formalized body of knowledge in the form of constructs or
theories.
This general approach forms the basis of thematic coding analysis discussed below.
Similarity and contiguity relations
Maxwell and Miller (2008) are concerned that an emphasis on coding and categorizing is in
danger of losing connections within accounts and other qualitative material. They make a distinction
between similarity relations and contiguity relations. When using coding, similarities and
differences are commonly used as the basis for categorization. Relationships based on contiguity
involve seeing connections between things, rather than similarities or differences. We look for
such relationships within a specific interview transcript or observational field notes, seeking
connections between things which are close together in time or space. They can also be sought
between categories and codes once they have been established as a next step in analysis.
Similar distinctions have been proposed previously, including Coffey and Atkinson’s (1996)
‘concepts and coding’ as against ‘narratives and stories’ and Weiss’s (1994) ‘issue‐focused’ and
‘case‐focused’ analysis.
Maxwell and Miller review several advantages of combining categorizing and connecting
strategies for analysing qualitative data. They suggest that it may be useful to think in terms of:
categorizing and connecting ‘moves’ in an analysis, rather than in terms of alternative or
sequential overall strategies. At each point in the analysis, one can take either a categorising
step looking for similarities and differences, or a connecting step, looking for actual
(contiguity based) connections between things. In fact, it is often productive to alternate
between categorizing and connecting moves (p. 470).
They provide an exemplar (pp. 471–2) illustrating one way in which the two strategies can be
integrated.
In their view, the ‘grounded theory’ method discussed later in the chapter actually uses this
strategy although with a different terminology. In particular, Corbin and Strauss’s (2008) ‘axial
REAL WORLD RESEARCH464
coding’ is effectively the same as ‘connecting analysis’. The other main approach covered below,
‘thematic coding’ analysis, while essentially based on categorizing, does not preclude following
Maxwell and Miller’s suggestions.
Using the computer for qualitative data analysis
The single constant factor reported by qualitative researchers is that such studies generate very
large amounts of raw data. A small ethnographic style study will generate many pages of field
notes including observations, records of informal interviews, conversations and discussions.
This is likely to be supplemented by copies of various documents you have had access to, notes
on your own thoughts and feelings, etc., etc. A multi‐method case study will produce a similar
range and amount of material. Even a strictly limited grounded theory study relying solely on
interviews leaves you with 20 or more tapes to be transcribed and subsequently analysed.
Before getting on with any type of analysis, you need to ensure that you know what data you
have available and that they are labelled, stored and, if necessary, edited and generally cleaned
up so that they are both retrievable and understandable when you carry out the analysis. A typical
first analytic task of labelling or coding the materials (e.g. deciding that a particular part or
segment of an interview transcript falls into the category of ‘requesting information’ or ‘expressing
doubt’ or whatever) involves not only assigning that code but also having a way of seeing it
alongside other data you have coded in the same way.
In the pre‐computer era, these tasks were accomplished by means of file folders containing the
various sources of data, markers and highlighters, and copious photocopying. One strategy was to
make as many photocopies of a page as there were different codes on that page, then to file all examples
of a code together. It is clear that much of the drudgery of this task can be eliminated by using
a word processor. Many data sources will either be directly in the form of computer files or can be
converted into them without difficulty. It may be feasible to enter field notes directly into a laptop
computer. An interview tape can be entered into the word processor as it is being transcribed.
Incidentally, if you have to do this yourself there is much to be said for the use of speech recognition
software for this task (listen to each sentence on the tape through headphones, then repeat it out
loud to activate speech recognition). It will need to be checked but modern systems can reach high
standards of accuracy. Similarly, if you have access to a scanner with optical character recognition
software (OCR), it is now straightforward to convert many documents into word processor files.
There are some types of data for which this may not be feasible (e.g. handwritten reports).
Word processors are a boon in storing, organizing and keeping track of your data. Obviously
you need to observe good housekeeping practices and should take advice on how you can
survive possible hard disk crashes, loss, theft, fire, etc. Essentially, this means having multiple
copies of everything, regularly kept up to date in more than one location, and in both paper and
computer file versions. Word processors can also help with the coding task through ‘copy’ and
‘paste’ functions. In this way it is easy to build up files containing all instances of a particular
coding whilst retaining the original file with the original data to which codes have been added.
Word processors can also be used to assist in the ‘connecting’ (as against categorizing) analysis
of qualitative data advocated by Maxwell and Miller (2008), discussed earlier in the chapter.
Marking, extracting and putting together selected data from a longer text can greatly simplify
the task of data reduction needed for producing case studies, narratives, etc.
THE ANALYSIS AND INTERPRETATION OF QUALITATIVE DATA 465
Should you go beyond using standard word processors to one of the many specialist software
packages designed to help with qualitative data analysis?
Using specialist qualitative data analysis (QDA) packages
There are many computer packages specifically designed for researchers to use when analysing
qualitative data (commonly referred to as CAQDAS – computer‐assisted qualitative data analysis).
The most widely used has probably been NUD*IST (Non‐numerical, Unstructured Data
Indexing, Searching and Theorizing), a catchy acronym which encapsulates the central features
of many of the packages – indexing, searching and theorizing. NUD*IST has now been superseded
by NVivo, developed by the same organization, QSR International (http://www.qsrinternational.com).
It can be used profitably in most situations where you have substantial amounts of
qualitative data, and for many different types of study, including grounded theory, conversation
and discourse analysis, ethnographic studies, phenomenological studies, action research, case
studies, and mixed method research. If you have facility in its use it is also a valuable tool when
carrying out literature reviews.
While NVivo is the preferred option for qualitative data analysis in many institutions and
hence is likely to be readily available and to receive support, there are several other packages
worth considering for particular situations or types of data – see Appendix B for details.
When deciding whether or not to use specialist software, the advantages of time‐saving and
efficiency when analysing large amounts of data (once you have gained familiarity with a package),
should be weighed against the time and effort taken to gain that familiarity. Box 18.3 lists
some general advantages and disadvantages in their use. García‐Horta and Guerra‐Ramos
(2009) discuss the use of two different packages with interview data, concluding that ‘CAQDAS
is of great help and can enhance interview data analysis; however, careful and critical assessment
of computer packages is encouraged. Their capabilities must not be overestimated, since
computers are still unable to perform an independent rational process or substitute the analyst’s
capacities’ (p. 151).
Richards (2002), the prime mover in the development of the NUD*IST and NVivo packages,
expresses concerns that the full potential of computer‐based analysis is not being realized. More
seriously, the packages may actually be having negative effects. Because the coding and sorting
tasks can be carried out more effectively and efficiently using a computer package, users tend to
focus excessively on this aspect:
The code‐and‐retrieve techniques most easily supported by computers and most demanded
by users are techniques most researchers had used at some time, for sorting out the mess
of complex data records. But they were not much discussed in the literature before computing,
and not at all clearly associated with the goal of theorizing common to most qualitative
methodologies. So computing became associated with techniques that are generic,
easily learnt and that emphasize data management, and description. Significantly, these
are aspects of practical research ignored or even spurned by theoretical writers (p. 266).
It could be argued that users are simply replicating their previous paper‐based, cut and
paste, highlighter employing, practices on the computer, but it is undoubtedly true that packages
are capable of much more than this; for example, they can include tools for doing more
REAL WORLD RESEARCH466
interpretation once the coding is done. Encouragingly, theory‐building software has in recent
years been developed to such an extent that it is probably the most widely used type. While, as
pointed out by Maxwell and Miller (2008), most of these uses have been based on a prior categorizing
analysis, many of the current programs allow the user to create links among and between
any segments, both within and between contexts, and to display the resulting networks.
Dealing with the quantity of qualitative data
Qualitative data can easily become overwhelming, even in small projects. Hence you need to
find ways of keeping it manageable. This process starts before any data are collected when you
focus the study and make sampling decisions about people to interview, places to visit, etc.
During and after data collection you have to reduce the data mountain through the production
of summaries and abstracts, writing memos, etc. Miles et al. (2014, p. 12) refer to this as date
condensation. They emphasize that this is a part of analysis and not a separate activity. Decisions
about what to select and to summarize, and how this is then to be organized, are analytic choices.
BOX 18.3
Advantages and disadvantages of specialist QDA packages
Advantages
They provide an organized single location storage system for all stored material (also
true of word processing programs).
They give quick and easy access to coded material (e.g. examples of a particular theme)
without using ‘cut and paste’ techniques.
They can handle large amounts of data very quickly.
They force detailed consideration of all text in the database on a line‐by‐line (or similar)
basis.
They help the development of consistent coding schemes.
They can analyse differences, similarities and relationships between coded elements.
Many have a range of ways of displaying results.
Disadvantages
Proficiency in their use takes time and effort.
There may be difficulties in changing, or reluctance to change, categories of information
once they have been established.
Particular programs tend to impose specific approaches to data analysis (depends on the
program – see Appendix B).
Tendency to think that simply because you have used specialist software you have
carried out a worthwhile analysis. A focus on coding and other technical aspects can
give less emphasis to interpretation.
THE ANALYSIS AND INTERPRETATION OF QUALITATIVE DATA 467
Good housekeeping
Even a small project producing qualitative data can easily leave you overwhelmed with lots of
pieces of information of many different types. Possible ways of keeping track include the use of:
Session summary sheets. Shortly after a data collection session (e.g. an interview or observation
session) has taken place and the data have been processed, a single sheet should be prepared
which summarizes what has been obtained. It is helpful if this sheet is in the form of answers
to summarizing and focusing questions. These might include who was involved, what issues
were covered, what is the relevance to your research questions (effectively what was the purpose
of the session), new questions suggested and implications for subsequent data
collection.
Document sheets. A similar sheet prepared for each document collected. This clarifies its context
and significance, as well as summarizing the content of lengthy documents. The session summary
and document sheets assist in data reduction, an important part of the analysis process.
Memoing. A memo can be anything that occurs to you during the project and its analysis.
Memoing is a useful means of capturing ideas, views and intuitions at all stages of the data
analysis process.
The interim summary. This is an attempt to summarize what you have found out so far and highlight
what still needs to be found out. It is recommended that this is done before you are halfway
through the time you have available for data collection. The summary should cover not
only what is known but also the confidence you have in that knowledge, so that gaps and
deficiencies can be spotted and remedied. Flexible designs enable you to do this in a way which
would not be feasible in a fixed design study but to capitalize on this flexibility you must force yourself to
find the time to do this interim summary while you can still take advantage of its findings to direct and
focus the later phases of data collection. The summary can also usefully incorporate a data accounting
sheet which lists the different research questions and shows, for different informants, materials,
settings, etc., whether adequate data concerning each of the questions have been collected.
Thematic coding analysis
Thematic coding analysis is presented here as a generic approach to the analysis of qualitative
data. It can be used as a realist method, which reports experiences, meanings and the reality of
participants or as a constructionist method, which examines the ways in which events, realities,
meanings, and experiences are the effects of a range of discourses operating within society.
Coding has a central role in qualitative analysis. Gibbs (2007), in a very clear and accessible
discussion, introduces it as follows:
Coding is how you define what the data you are analyzing are about. It involves identifying
and recording one or more passages of text or other data items such as the parts of
pictures that, in some sense, exemplify the same theoretical or descriptive idea. Usually,
several passages are identified and they are then linked with a name for that idea – the
code. Thus all the text and so on that is about the same thing or exemplifies the same thing
is coded to the same name (p. 38).