CHAPTER 2 from the manual of AQUAD Five:
Huber, G. L. (1997). Analysis of qualitative data with AQUAD Five for Windows (pp.23-36). Schwangau: Ingeborg Huber Verlag.
In this chapter we try to outline typical phases in the process of a
qualitative text analysis. Particular analytical steps are marked as characteristics
of these phases, although concrete processes of qualitative analysis usually
follow a cyclic path, i.e., during every phase a researcher may be engaged
part of the time in activities characterising other phases of analysis.
In addition, this chapter tries to give an overview on the usage of computer
assistance for qualitative analysis as provided by Aquad. Details are explained
in later chapters. However, the following chapters of this manual cannot
offer more than a general introduction to methodological approaches to
qualitative analyses and possible contributions of computers. We suggest
that users who want to learn more about underlying principles read the
following books:
This list is far from being exhaustive, but names some selected books which seem appropriate for beginners in qualitative analysis and which demonstrated their usefulness in Huber's seminars at the University of Tübingen. All of these books also offer excellent approach to more specialized literature. In the following, this chapter borrows from an introduction by Huber (1992) which unfortunately is not available in English.
1. Which are the steps in qualitative analysis?
As typical phases of qualitative analyses of texts we can distinguish the
reduction of the original data base, the reconstruction of linkages, and
the comparison of findings. Especially in psychological studies researchers
often are interested in inferring inductively from
one
subject's
data regularities in this person's experiences and behaviors. Whether communalities
can be found when taking into account data of several persons is interesting,
too, but only during a later stage of investigation.
The first phase of qualitative analysis is characterized by reducing the overwhelming amount of text data by identifying the content of more or less encompassing text segments. Then, a "code" as abbreviation or name is attached to this text segment. In the following, these codes are used as representants of text segments or "units of meaning" in the text. Fundamentally, this is a process of categorization, where the categories may emerge during text interpretation or may be taken from an already existing category system depending on the researcher's epistemological orientation.
During the second phase researchers try to reconstruct the text producer's subjective meaning system from the units of meaning in their text data. By "text producer" we refer to the researcher's interview partners, to writers of diaries, to observers who took field notes in a setting, etc. In order to reconstruct meaning systems we are looking for regular linkages between units of meaning in the text data, which are characteristic for text producers and/or their situation.
In the third phase finally researchers try to infer invariants or general
communalities by
comparing
individual systems of meaning (see Ragin,
1987).
It is important to keep in mind that these phases neither are strictly
demarcated nor do they follow each other in a linear sequence, but they
overlap and are linked to each other in circular patterns (cf. Shelly &
Sibert, 1992). During data reduction we may start to ponder about the text
producer's implicit theory or we may permanently compare the text at hand
with other texts which we have read earlier. Thus we may perhaps detect
in person C's text an aspect of meaning which we overlooked in person A's
text. As a consequence, we repeat the process of data reduction for person
A. In all of these phases it is necessary to affirm deductively the validity
of our generalizations. That is, we try to infer particularities from our
general findings and then return to our text data and try to find evidence
in form of specific data, i.e., statements in the texts.
2. How to reduce qualitative data
The principles of reduction are obviously simple, but their application
soon proves to demand very much work, to consume very much time, and to
be very prone to errors. Voluminous verbal materials have to be reduced
to the units or categories of meanings they contain. Tesch (1992) describes
these principles as retrieving and marking of text segments, which are
relevant for the question under study, by an abbreviation, that is a code
for the particular category of meaning. Tesch also compared computer assisted
analysis and "traditional" approaches, which use segments of
texts or text clippings in the literal sense or which transfer relevant
text segments to index cards. The work load is tremendous in both cases.
Computer software not only assists in reducing the amount of data, but
also in reducing all the mechanical labor otherwise necessary. Instead
of handling verbose clippings of text distributed over many piles of index
cards, further computer assisted analyses use just the codes of these text
segments; that is, after data reduction you work with category names and
information where to find the categorized text segments in your texts.
If you need to scrutinize the original text segment again during the course
of your work, the computer will retrieve it immediately for you.
The critical question in this phase of analysis is:
How and where
do I find units of meaning in my texts?
Beginners in text analysis
as well as experts who try to get familiar with a new content domain ask
this question again and again. Weber (1985) describes six widely used general
possibilities to define text segments, namely to choose as unit of analysis
single words, meanings of words, sentences, topics, paragraphs, and the
complete text (for instance, if the texts are short as in the case of letters
to the editor or if you want to produce head lines or abstracts). However,
this choice cannot be made mechanically, but it needs itself tentative
qualitative decisions. Not so obvious when single words are used as units
of analysis, but quite obvious when we use more complex alternatives like
word meanings or sentences, we need preceding insights or hypotheses that
the unit we have chosen will contribute to answer our research question.
Additional qualitative decisions are necessary, for instance which words
are used synonymously or which idiomatic expressions have similar meanings
for the writers/speakers of our texts.
The strategies of defining units of meaning describe an important difference
between quantitative and qualitative approaches to text analysis since
the beginnings of a broader reception of text analysis in the social sciences.
In the same year, 1952, two trendsetting articles were published by Berelson
and by Kracauer. While text analysis serves for Berelson to assess systematically,
objectively, and quantitatively the
manifest
contents of communication,
qualitative content analysis according to Kracauer tries to reveal the
categories of meaning hidden or
latent
in the text. Both authors
relate their controversial positions to a debate, which Thomas & Znanecki
(1918) had initiated 35 years earlier with a meanwhile classical analysis
of letters of Polish immigrants to the United States. Obviously the question
of adequate units of analysis -- words vs. meanings -- is confounded with
the particular goal of text analysis.
On the level of processing both approaches do not necessarily exclude
each other. Especially with assistance of computers we can apply various
strategies from the quantitative approach to qualitative text analysis
in order to gain support for our interpretational endeavors.
Principally we should keep in mind to develop the analytic units
during
text interpretation. This is true in analyses which try to
understand
the experiences and actions of people from their own verbal descriptions
as well as in analyses which try to
explain
specific actions by
relating them to the frame of reference of these people's implicit theories.
Only this approach helps to open a door to the subjective world views of
our interviewees or producers of other types of text. Otherwise, we would
be in danger of grasping only some partial aspects of their world views,
maybe isolated from their subjective context, which our personal analytic
grid is able to comb out. The strategy of developing categories "on
the fly" corresponds to the approach of "grounded theory",
an empirically based procedure of generating theories recommended by Glaser
& Strauss (1979).
However, this procedure demands enormous work as soon as more than two
or three texts are to be analyzed. Because we usually want to compare the
results from single texts in advanced stages of analysis, we have to ensure
that we defined and coded the units of meaning in all of the texts coherently.
Usually this process has to be repeated again and again, and in this process
units of meaning and their codes have to be modified. Miles & Huberman
(1984; 1994) suggested a compromise, which structures the process of data
reduction from the beginning: Before you start reading your data texts
you state a very general
frame of orientation
without any references
to particular contents; then you try to find specific units of meaning
within this framework. This structuring does not contradict the demand
of openness for emerging categories while reading your texts, because usually
the reduction of potential data by structuring starts even before you enter
the phase of coding your texts, for instance when planning and deciding
which persons, cases, sites, types of texts, etc. should be included in
a study.
3. How to find units of meaning
When looking for units of meaning in the data texts, you should be open
for emerging structures in the texts. However, recipes enforcing structures
on the data may often be welcome because they help to shorten a phase of
maximal uncertainty particularly during the first steps of data analysis.
On the other hand, the price for certainty and savings of time and effort
may be too high: concrete guidelines would lead only to those units of
meaning which could be foreseen, while surprising, rare, but maybe most
interesting aspects are in jeopardy of being omitted from further analysis.
The following hints should be understood as heuristics, that is, as general
directions which provide assistance in identifying units of meaning and
attaching appropriate codes, but not as algorithmic rules which could be
followed step by step to a predetermined goal. Three heuristics will be
described in the following. Aquad offers support particularly for the first
two of them and their variations, which are also outlined.
3.1 How to find categorical codes
In addition to looking for units of meaning in the text, we have to decide,
whether our categories and the corresponding codes should describe, interpret
or explain the content of a particular text segment (see Miles & Huberman,
1994, p. 56). This is a decision for which we cannot expect computer assistance.
Here, the computer assists "only" in documenting, in sorting,
and in revising -- if necessary -- all the decisions a researcher has made
on his/her own responsibility. Then, three possibilities are widely used
to find categorical codes:
Applying pre-determined category systems
If a qualitative study does
not
aim at constructing theories from
concepts emerging in the texts, there is a very simple means of finding
codes: You can use an available category system and reduce your texts according
to the interpretational schemata contained in the system chosen. Categorical
systems may be "available" from earlier studies on the same topic
or from publications of other authors.
When using a pre-determined category system we have to decide which segments
of a given text correspond to which of the given category definitions.
Within Aquad we have then to note where each of these segments is located
in our text and which codes apply to them. How this is done will be explained
in detail in chapter 5. Without fail we will have trouble assigning specific
text segments to one of the given categories, and we will not always use
the codes consistently. In these cases, the function to
retrieve coded
text segments
is very helpful to control our work. After entering a
critical code (or a whole list of these codes), the program shows us very
quickly all text segments in all texts analyzed, which up to now were marked
with this code.
However, this approach to test the reliability of our coding does lead
us only to those text segments which were assigned erroneously to a particular
category, but not those segments which belong to the definition of this
category but were
not
assigned to this category. In conventional
approaches to text analysis we would detect this type of error when we
notice inconsistencies within the sets of segments assigned to other categories.
With computer assistance it is no problem to control for "missing"
text segments by checking all segments which were assigned to related and
thus "error prone" categories. In addition, we can make use of
the complementary relations of manifest and latent units of meaning in
our text. Three strategies appear to be useful:
We define key words as manifest indicators of a critical meaning and have the computer retrieve all their occurrences in the texts. If we find our key words in a specific text segment, we can decide whether this segment should not better be assigned to the category indicated by this key word. For this strategy most text processors can be used.
We assemble key words in a dictionary for analysis , called a word catalog in Aquad, that is we use a list of key words for the retrieval of critical text segments. Then we find the information needed in a single run of the program.
Aquad offers both of these possibilities, automatically combined with
a third strategy called
key-words-in-context
(Popko 1980). This
function tries to retrieve one or more key words in our texts and prints
it within the context of the line where it was found in a text. The researcher
is responsible then for further decisions, for instance about the range
of a corresponding unit of meaning and about marking it with a particular
code.
Hypothesis-based categorization
The research question often supplies the researcher with hypotheses that
may be used as a sort of guidance when looking for units of meaning in
the texts. Based on these hypotheses the researcher tries to define categories
and rules of coding for her/his data.
For instance, the approach of Marcelo (see p. 18) to code his interviews
with beginning teachers was hypothesis based. A theoretical model of professional
socialization served as a framework to design a preliminary category system.
With increasing familiarity with the texts and deepened insights into the
subjective points of view of the beginning teachers, some of the categories
had to be omitted from the system as inadequate, while some new categories
emerged from the analytic process. Some "narrow" categories could
be combined to more comprehensive units in terms of the teachers' subjective
theories (see Huber & Marcelo, 1992). Of course, new or combined categories
had to be compared to critical text segments in all interviews again.
Properly, the researcher decides about concrete categories already when
designing the collection of data. Thus, for instance, the formulation of
guiding questions for an interview would be deductively determined by the
researcher's basic hypotheses. On the other hand, the researcher will for
sure experience interview situations or find text segments in the interview
transcriptions which cannot be assigned to pre-determined categories. For
these text segments categories have to be developed inductively. The new
categories may relate the text to the original hypotheses -- but they may
also be a stepping stone from which modifications of the original frame
of reference can be realized. We see that the researcher's degrees of freedom
as well as challenges to her/his interpretational aptitudes are increasing
when applying this second strategy. Most important, the decreasing structuredness
of this approach augments the chances to take into account the subjects'
points of view.
When applying this strategy, the researcher has also to meet the demands
of the "method of permanent comparison", which is a trade mark
of the approach of grounded theory (Glaser & Strauss, 1967, 1979; Strauss
& Corbin, 1990). Shelly & Sibert (1992) described this method in
detail and related it to models of the researcher's cognitive activities.
An activity of major importance in "permanent comparison" is
to test every inductive conclusion from particular data to more general
principles -- here the analytical categories -- by means of deductive conclusions,
that is deductions of specific units of meaning in the data base.
Together with the demand to apply interpretation in text analysis the
importance of computer assistance is increasing. When we try to get an
overview on linkages between categories, simple retrieval functions reach
their limits. From this point in text analysis on, Aquad serves you especially
well, because it supplies you with routines for deductive conclusions based
on the principles of logic programming (Tesch 1990; Shelly & Sibert
1992).
The functions for
retrieval of coded text segments
and for retrieval
of
key words
(including retrieval dictionaries and the KWIC-function)
may be used here to check the consistency of coding within and across text
in the same way as in studies applying ready-made category systems (see
above).
From the hypotheses, which mark out a framework for the development of
categories, we get hints to specific
relations of categories
. Aquad
supports hypothesis-based data reduction with functions for testing relations
of categories. Among other relations you can find out about the
super-ordination/sub-ordination
of categories,
sequences
of categories or
clusters
of particular
categories. These three types of relations probably represent the most
frequently tested relational patterns tested in text analyses. The test
activities demand that you focus your attention not only on one category
and the text segments it represents, but on two or more categories and
the defined relations between them -- which may include negations! Of course,
non-events or missing relations have to be registered in this process,
too.
Categories by theory-building
The most exacting mode of reducing qualitative data refrains from any prescriptions
for data reduction as for instance category systems and from structuring
the process of qualitative analysis by hypothetical frameworks. Consequently
the researcher has to keep in mind his/her subjective experiences, opinions
or prejudices as regards the world views or behavior of his/her subjects,
and the researcher has to avoid premature stabilization of emerging principles
of data reduction by permanently comparing the momentary categories and
relevant statements in all available texts. To be sensible of one's own
way of reading a text is particularly important if texts are analyzed which
were written long ago or in a context that differs markedly from the researcher's
way of life (Fischer, 1982). Often specific information about the speaker
or writer of a text is very helpful. If this information is not available
in a text, it may be necessary to look for additional sources. These sources
and their information help to approach better the goal of viewing the world
of the subjects of a study through their own eyes and to understand it
from their own perspectives (or, for more auditive readers: these sources
may support listening to the speakers' own voices in their texts).
In this process we try not only to describe subjective world views, but
to order them by matching concepts and to reconstruct systematic relations
between these concepts, that is, we are occupied with what Glaser &
Strauss (1967; 1979) called the
discovery
of a "grounded theory".
The term "discovery" accentuates an essential difference from
methodological approaches which are applied to confirm given theories.
Grounded theories are developed during text analysis, therefore we do not
start from an available theory looking for verifying or falsifying data
in the texts, but we start from a phenomenon the text refers to, which
we want to understand and to explain (Strauss & Corbin, 1990).
This approach demands a maximum of interpretational efforts. Exactly this
aspect seems to cause problems for beginners of text analysis. Pre-defined
category systems come with explicit interpretation rules or they transport
these rules implicitly in a set of examples for each category. Researchers
using a hypothesis-based approach to categorization can at least rely on
a general orientation what to look for in the texts. In the process of
theory construction, however, researchers have to find out first what a
text is telling them. Beginners often tend to avoid the risk of errors
and to interpret as parsimoniously as possible. In extreme cases this tendency
results in reading a text for the appearances of critical formulations
-- comparable to the application of key words when using ready-made category
systems. If text segments containing such a critical formulation are marked
by a specific code, this code does not necessarily represent a "unit
of meaning" or signify a subjective viewpoint of the writer/speaker,
but more probably the mere fact, that a particular formulation was used
in this text segment. Its meaning still may be unclear.
This tendency is especially pronounced, if a researcher does not know about
the possibilities of tentative interpretation and easy revisions in computer-assisted
qualitative analysis. Supported by Aquad a researcher is not punished for
"playing" with ideas by tremendous labor when it comes to revisions
of codes during a later phase of analysis, for instance, when the researcher
finds text segments not matching a category introduced earlier. On the
contrary, computer assistance encourages creative interpretation, because
we have to introduce codes as markers of text segments from the very beginning
of an analysis, and changes, summaries, differentiations, etc., of categories
can be realized smoothly (Tesch, 1992).
As a
rule of thumb
for theory construction by categorization
we can state: You should look in the text for units of meaning
as large
as possible
in order to find something to interpret at all; at the
same time these units should be
as small as necessary
to avoid representing
incongruent contents by the same code. This approach could be called a
strategy of differentiation
. Aquad supports this strategy, because
units of meaning can be defined without limitations; above all, units of
meaning may overlap, and the same text segment may be assigned to several
categories, that is, it may be marked by several codes.
If you are already familiar both with qualitative analysis and with the
domain addressed in the texts of your study, you may reverse this method
and approach your texts following a
strategy of generalization
These prerequisites given, firstly the probability is low that you fall
a victim of details in the text and miss to reveal its essential meanings.
Secondly you are then also accustomed to software functions for producing
meta-codes, which assemble a number of too detailed categories into one
more comprehensive category. In a generalizing approach it is particularly
important to compare coded text segments permanently in order to find out
inductively about all content dimensions expressed in these segments and
the adequate super-ordinate categories. Aquad supports this strategy, too.
You can retrieve easily those text segments which are similar to a particularly
coded one, which contradict its meaning, which depend on it or are related
in other, specified ways to this text segment. Based on findings about
such relations you can then try to combine individual codes into a more
comprehensive one.
When we reduce texts to categorical codes, we try to distinguish text
segments in such a way that we can assign them to well defined, mutually
exclusive categories. Unequivocal assignment of meanings to categories
does not imply, however, that the text segments involved always have to
be clearly different from each other. Depending on writing styles or communicative
styles of the text producers or depending on the researchers´ ways
of reading a text , segments assigned to differing categories may overlap
or the same text segment may even become assigned to more than one category.
3.2 How to find sequential codes
The starting point for sequential coding is the detection of specific relations
between text segments. If specific linkages of segments emerge from the
text, a researcher may want to mark their appearance in a text and to represent
the type of linkage by a particular code -- in the same way as the researcher
used categorical code. Here, however, the code represents a defined sequence
of meanings found in a text. The unit of meaning is the complete text section,
in which the sequence of meanings was found.
Which strategies are at hand that may help us to go beyond assigning
categories or subcategories to text segments and to detect sequences or
linkages of meaning in a text? In the following we differentiate between
strategies that inquire into simple sequences and complex sequences of
meaning.
Looking for simple sequences
Besides looking for
hierarchical sequences
of super- or subordinate
categories as recommended by Strauss & Corbin (1990), a number of other
simple sequences appear to be interesting from the point of view of grammatical
or linguistic properties of the text. When using these strategies, we should
be aware, however, that we may be about to abandon looking for emerging
categories and begin to enforce categories on the text (see Glaser, 1992).
Whether this switch is permissible and which of the possible strategies
is adequate cannot be answered absolutely, but depends on the research
question. Often researchers are looking for
causal sequences
(using
key words like "because", "based upon", etc.),
temporal
sequences
("while", "then", "before",
etc.),
concessive sequences
(positive concessions like "not
... but ..." or negative concessions like "indeed ... otherwise
..."),
conditional sequences
("if ... then..."),
final sequences
("so that", "lest", "in
order to", etc.),
comparative sequences
,
modal sequences
and
defining sequences
.
Neither this list of types of sequences nor the quoted key words claim
to be complete. They only want to stimulate your own trials to retrieve
sequences of meaning in the texts according to the researcher´s questions.
In this context we should again accentuate the function of computers and
software as useful
tools
for text interpretation, but not as agents
of qualitative analysis. Functions for word retrieval may be especially
helpful, but within very narrow limits. Except in cases where a researcher
analyses carefully formulated texts, the yield of dictionaries of particular
sequences of meaning is usually limited. In freely spoken interview texts
even sentence structures are often hard to define: Where is the principal
clause, where does a subordinate clause begin? Often conjunctions critical
for defining a specific type of sequence may have been in the mind of a
speaker, but they do not appear in the spoken and transcribed text. In
addition, many speakers do not use critical conjunctions or constructions
according to the grammatical rules. Therefore, looking for key words or
applying whole dictionaries of these words never substitutes for empathetic
interpretation of texts -- but these strategies may be useful heuristics.
Looking for complex patterns of sequences
Structural or content characteristics of texts may inspire the search
for complex sequences of meaning. For example, when a researcher tries
to reconstruct theories of action implicit in a text, he or she could look
for sequences of appraisals of situations, reflexions of alternatives for
action, expectations of action effects, and evaluations of potential personal
consequences like satisfaction or disappointment. This means, the researcher
would have to keep in mind a number of categories and to look for them
and their proper sequence simultaneously when interpreting a text. Such
an analysis is both demanding and prone to errors. The variety of structures
of possible linkages of categories is demanding for software, too. It would
not make much sense to implement all sequential combinations of possibly
linked categories in form of abstract deductive algorithms into a program,
so that the researcher would have to enter during run-time only some concrete
codes as variables. For the retrieval of complex patterns a researcher
needs access to the source code of her/his software in order to add those
rules with minimal programming effort, which s/he expects to apply to the
linkage of categories in the texts. Aquad offers exactly this possibility.
3.3 How to find thematic codes
The most radical approach to text reduction is applied, if we try to code
a text just by one central category, that is, its message or its topic.
Since qualitative analysis is a cyclic process, we cannot locate the strategy
of thematic coding at a particular place within the interpretational process,
let's say at the end as a sort of summary of findings. Thematic coding
may play an important role at this stage of text analysis, but this strategy
is also useful at other stages:
When we analyze relatively short and homogenous texts (for instance, "letters to the editor") or paragraphs, thematic coding may be all we need or want to reach our interpretative goals.
In the case of long, heterogeneous, and complicated texts interpretation may start with thematic reduction. Here, this strategy may serve us as an important heuristic. It may help us to find one or some main ideas in the text and not get entangled in a great number of potentially controversial details.
Usually, thematic coding appears in the final stage of text analysis,
often only after several cycles of data reduction, reconstruction and comparison
of meaning structures during which the "Leitmotiv" of different
texts was elaborated more and more clearly (see Shelly & Sibert 1992;
Strauss & Corbin 1990).
Strauss & Corbin (1990) recommend five steps for thematic reduction
of texts, which are compatible with each of the three locations of thematic
reduction:
(1) We start with identifying a main idea or a central category;
(2) then we look for subordinate categories,
(3) which often need to be differentiated or to be linked with each other;
(4) then we examine hypothetical relations among sub-categories as well as between these categories and the main idea by comparing text segments;
(5) discrepancies, for instance, inconsistent categories stimulate further cycles of analysis.
4. How to reconstruct systems of meaning
In theory-constructing qualitative analysis we try to generate the speaker's
or writer's theory of the issue her/his text is talking about, that is,
we try to reconstruct the speaker's or writer's subjective system of meaning.
Solutions can be developed inductively, deductively or by combining inductive
and deductive strategies. Which approach is to be preferred depends on
the research question.
If we analyze transcriptions of minimally structured interviews or texts,
for instance entries in a diary, we usually start by identifying units
of meaning and assigning them to specific categories. Then we may look
for typical sequences of categories. Finally we will try to subsume such
sequences under more abstract categories, that is the message or the topics
of the text producer. This analysis starts with inductive processes, but
alternates between inductive and deductive conclusions at least from the
stage of thematic reduction on.
If we approach a text under a particular theoretical orientation or
from the point of view of particular interests, we will try to retrieve
specified relations from the beginning. That is, we start with hypotheses
of possible relations and try to confirm them deductively in the text.
Inconsistencies or contradictions then will cause processes of inductive
analysis, which in turn may modify our hypothetical orientation.
In inductive approaches we try to generalize categories and their systematic
linkages from concrete text segments. In deductive approaches we try to
find concrete text segments which may confirm general assumptions or hypotheses
about how specific categories are linked. The following overview on strategies
offered in Aquad for computer-assisted reconstruction of systematic linkages
is structured according to these two approaches.
4.1 Reconstruction from the context of particular data
We are looking for systematic occurrences of a particular category (or
several particular categories) in a data text which is characterized by
one or more profile codes, usually socio-demographic codes. For computer-assisted
reconstruction we create code matrices (Miles & Huberman 1984) or code
tables (Shelly & Sibert 1992). The contents (that is, text segments)
in the cells of such a table are determined twice, both by the category
serving as column header (profile code) and by the category serving as
row header. Thus, the interpretation of findings also is much more guided
than in the case of just registering frequent closeness of otherwise "unconditioned"
categories in space (line-numbers) and time (sequence of text production).
On the other hand, the construction of a table for analysis demands much
more conceptual investments, that is greater progress in the process of
text analysis. The module "Tables" (see chapter 10.2) is available
for this type of reconstructions in Aquad.
4.2 Reconstruction by testing particular relations
Here we can also distinguish two approaches, which correspond to the just
described strategies of coding simple or complex sequences of meaning in
the text.
(1) Confirming simple code sequences . Let us assume that an interview with parents talking about their educational practices suggests that a father is trying hard to justify his ways of education. Then we could activate the module "Linkages" (see above and chapter 10.3) and examine our tests, looking for sequences of relevant codes, for instance for final or causal codes, in order to confirm our impression deductively.
(2)
Confirming complex relations of codes
. If a researcher wants
to confirm relations of codes which exceed the complexity of those linkage
structures already built into Aquad, s/he can use an "open" programming
module in Aquad, a so-called "development environment". This
environment contains part of the software package's source code, to which
the researcher can add her or his case-specific linkages of codes. How
you put your assumptions about complex linkages of codes into a form which
can be confirmed or rejected by Aquad is described in detail with the help
of several examples in chapter 11.
5. How to compare relations of meaning
Permanent comparisons of interpretations within a text and across different
texts are at the core of all procedures of qualitative analysis (see Shelly
& Sibert 1992). Even when reducing the data within a text to codes
it appears to be impossible to attain reliable codings without permanently
comparing codes/categories and text segments in the text at hand as well
as in the other texts of the research project. However, in most projects
we want to achieve more than to get access to unique world views expressed
in individual texts by means of a category system valid for all texts.
At some point in the research process we usually want to establish general
assumptions across texts. This task confronts the researcher with the proverbial
danger not to notice the wood because of all the various trees. Tackling
all the different formulations in the texts and elaborating their precise
meanings should not prevent us from looking for common elements. Therefore
we have to notice systematic relations across texts and to compare them,
too. Comparing relations between social phenomena, however, regularly leads
to findings of numerous conditions in manifold, sometimes controversial
combinations. In addition, in many studies one's own findings have to be
compared with findings from other studies, that is to perform a
meta-analysis
of qualitative findings. Even if other researchers have used the same research
questions as we did or at least comparable questions, they will have found
answers that may be only partially compatible with ours. When looking for
configurations of conditions of a particular phenomenon, sciences dealing
with the complexities of natural systems usually find highly varying constellations
across different studies.
Let us take an example that is often discussed hotly in everyday life:
"Is there a relation between cancer of the lungs and smoking?"
Whoever wants to affirm this question is regularly confronted with the
case of an 80 years old grandfather, who smoked all his life long, or with
the opposite case of a victim of cancer that never touched a cigarette.
Obviously a lot of conditions are to be considered. Empirical studies provide
us with many empirical arguments and you can choose the ones you need in
order to immunize your point of view against counter-arguments. Only a
meta-analysis of all the relevant studies or at least a representative
sample of these studies could give clarity about the relevant constellations
of conditions.
By applying Boolean algebra to qualitative data Ragin (1987) has developed an important approach comparing qualitative data. The procedure is based on the Quine-McClusky-algorithm of "logical minimization". According to Ragin (1987, p. 121) this approach fulfils the demands for a qualitative comparative research strategy, because
If we compare this strategy to variable-oriented approaches, which assume that variables can be combined additively, we find that Ragin's case-oriented comparisons (1987, p. 51 f.) appear to be able to
In order to apply the rules of Boolean algebra to qualitative comparisons
we reduce in every single case all codes radically to "truth values".
We are satisfied with the binary statement "condition true" (i.e.,
given) respectively "condition false" (i.e., not given). The
configuration of conditions of a particular case is represented by a row
in the table of truth values. The conditions for a case are combined through
Boolean multiplication (logical
and
). The different configurations,
i.e., the rows in the resulting table are combined additively (logical
or
). In this way, we first represent the single cases as configurations
of characteristics or conditions and then compare the patterns of configurations.
It does not matter if we do not use exclusively qualitative data; scores
on a quantitative dimension can also be reduced to truth values.
Aquad offers the possibilities of logical minimization for qualitative
analysis in its module "Implicants". Whenever comparative operations
are necessary in the process of qualitative analysis, we should consider
these possibilities. Because of the cyclic nature of qualitative analysis,
logical minimization may be helpful during each of its phases.
As a
heuristic
this strategy is already helpful when we start
to generate categories for text interpretation, even if we have only analyzed
a few texts, that is even with a small basis for comparative operations.
During the final phase of analysis, when we want to create a
summary
of results
or when we want to group our findings, to distinguish types
of writers or speakers, to determine key texts in our data base, etc.,
the strategy of logical minimization appears to be indispensable. As Ragin
(1987, p. 51) has stated, "the potential volume of the analysis increases
geometrically with the addition of a single case, and it increases exponentially
with the addition of a single causal condition." Aquad offers well
elaborated procedures for grouping or clustering cases by means of logical
minimalization of critical conditions. Finally, we can apply the strategy
of logical minimization when we intend to
compare several qualitative
studies
. This is a methodological step usually called
meta-analysis
In the area of qualitative research, the strategy of logical minimization
could supply the simplicity, transparency, reliability, and documentation
desired for meta-analytical comparisons.
References
Berelson, B. (1952).
Content analysis in communication research
Glencoe, Ill.: The Free Press.
Fischer, P. M. (1994). Inhaltsanalytische Auswertung von Verbaldaten [Content
analytic analysis of verbal data]. In G. L. Huber & H. Mandl (Hrsg.),
Verbale Daten
[Verbal data]. (2nd edit.) (pp. 179-196). Weinheim:
Beltz.
Glaser, B. G. (1978).
Theoretical sensitivity.
Mill Valley, CA:
Sociology Press.
Glaser, B. G. (1992).
Emergence vs. forcing. Basics of grounded theory
analysis.
Mill Valley, CA: Sociology Press.
Glaser, B., & Strauss, A. L. (1967):
The Discovery of Grounded
Theory: Strategies for Qualitative Research
. Chicago: Aldine.
Glaser, B. G., & Strauss, A. L. (1979). Die Entdeckung gegenstandsbezogener
Theorie: Eine Grundstrategie qualitativer Sozialforschung [The detection
of grounded theory: A basic strategy of qualitative social research]. In
C. Hopf & E. Weingarten (Eds.),
Qualitative Sozialforschung
[Qualitative social research]. (pp. 91-111). Stuttgart: Klett-Cotta.
Huber, G. L. (1992). Qualitative Analyse mit Computerunterstützung
[Computer-assisted qualitative analysis]. In G. L. Huber (Ed.),
Qualitative
Analyse. Computereinsatz in der Sozialforschung
[Qualitative analysis.
Application of computers in social research]. (pp. 115-176). München:
Oldenbourg.
Huber, G. L., & Marcelo García, C. (1992). Voices of beginning
teachers: Computer-assisted listening to their common experiences. In M.
Schratz (Ed.),
Qualitative voices in educational research
. (pp.
139-156). London: Falmer.
Kelle, U. (Ed.) (1995).
Computer-aided qualitative data analysis.
Theory, methods, and practice
. London: Sage.
Kracauer, S. (1952). The challenge of qualitative content analysis. Public Opinion Quarterly, 16 , 631- 642.
Miles, M. B., & Huberman, A. M. (1994).
Qualitative data analysis.
An expanded sourcebook
. (2nd edition). Thousand Oaks, CA: Sage.
Popko, E. S. (1980).
Key-word-in-context bibliographic indexing. Release
4.0 users' manual
. Cambridge, Mass.: Harvard University, Laboratory
for Computer Graphics and Spatial Analysis.
Ragin, C. C. (1987):
The Comparative Method. Moving Beyond Qualitative
and Quantitative Strategies
. Berkeley, CA: University of California
Press.
Schratz, M. (Ed.) (1993).
Qualitative voices in educational research
London: Falmer Press.
Shelly, A. L., & Sibert, E. E. (1992). Qualitative Analyse: Ein computerunterstützter
zyklischer Prozeß [Qualitative analysis: A computer-assisted cyclic
process]. In G. L. Huber (Ed.),
Qualitative Analyse. Computereinsatz
in der Sozialforschung
[Qualitative analysis. Application of computers
in social research]. (pp. 71- 114). München: Oldenbourg.
Strauss, A., & Corbin, J. (1990).
Basics of qualitative research.
Grounded theory procedures and techniques
. Newbury Park, CA: Sage.
Tesch, R. (1990).
Qualitative research. Analysis types and software
tools
. New York: Falmer.
Tesch, R. (1992). Verfahren der computerunterstützten qualitativen
Analyse [Procedures of computer-assisted qualitative analysis]. In G. L.
Huber (Ed.),
Qualitative Analyse. Computereinsatz in der Sozialforschung
[Qualitative analysis. Application of computers in social research]. (pp.
43-70). München: Oldenbourg.
Thomas, W. I., & Znanecki, F. (1918).
The Polish peasant in Europe
and America
. Boston: Badger.
Weber, R. P. (1985).
Basic content analysis.
Sage University Paper
series on Quantitative Applications in the Social Sciences, series no.
07-049. Beverly Hills: Sage.
