Coding

Coding is the fundamental analytic process used by the researcher. In grounded theory research,
there are three basic types of coding: open, axial, and selective.
1. Open Coding. Open coding is the interpretive process by which data are broken down
analytically. Its purpose is to give the analyst new insights by breaking through standard ways of
thinking about or interpreting phenomena reflected in the data. (See Wicker, 1985 on "breaking
out of conceptual ruts.")
A series of techniques have been developed to further this process.
In open coding, events/actions/interactions are compared with others for similarities and
differences. They are also given conceptual labels. In this way, conceptually similar
events/actions/interactions are grouped together to form categories and subcategories. For
example, an analyst might note several incidents, actions, and interactions between nurse and
client which appear to be directed at providing comfort. The analyst labels these as "comfort
work." This category can then be broken down into specific properties and their dimensions.
"Comfort work" has the property of type, which can be broken down into subtypes. Another
property is duration, which can be dimensionalized as ranging from long to short episodes. Still
another property is the manner in which comfort work is carded out, and so forth. Specification
thus develops categories while also furthering the precision of a grounded theory.
Once identified, categories and their properties become the basis for sampling on theoretical
grounds. In making the next observations, the researcher should look closely for instances of
comfort work and take specific note of the different kinds, how long they last, and so forth.
Open coding stimulates generative and comparative questions to guide the researcher upon return
to the field: What is comfort work and how does it manifest itself?. How does it differ from other
types of work that nurses do, safety work, for example? Asking such questions enables the
researcher to be sensitive to new issues and more likely to take notice of their empirical
implications (theoretical sensitivity), while comparisons help to give each category specificity.
Once aware of distinctions among categories, the researcher can spell out specific properties and
dimensions of each. Ambiguities can be resolved through additional field work and specification.
Open coding and the use it makes of questioning and constant comparisons enables investigators
to break through subjectivity and bias. Fracturing the data forces preconceived notions and ideas
to be examined against the data themselves. A researcher may inadvertently place data in a
category where they do not analytically belong, but by means of systematic comparisons, the
errors will eventually be located and the data and concepts arranged in appropriate
classifications.
2o Axial Coding. In axial coding, categories are related to their subcategories, and the
relationships tested against data. Also, further development of categories takes place and one
continues to look for indications of them.
Through the "coding paradigm" of conditions, context, strategies (action/interaction), and
consequences, subcategories are related to a category. This paradigm does not differ from
schemes used in other types of qualitative research, but perhaps is used more concertedly in
grounded theory studies. To continue with our example of "comfort work," as soon as the analyst
notes an indication of this type of work, the data should be scrutinized to determine the
conditions that gave rise to the work, the context in which it was carried out, the
action/interactions through which it occurred, and its consequences. If one does not alternately
collect and analyze data, there will be gaps in the theory, because analysis does direct what one
focuses upon during interviews and observations.
During the analytic process, the analyst can draw upon previous experience to think through the
conditions that might lead a nurse to perform comfort work and what the consequences for the
patient might be. All hypothetical relationships proposed deductively during axial coding must
be considered provisional until verified repeatedly against incoming data. Deductively arrived at
hypotheses that do not hold up when compared with actual data must be revised or discarded.
A single incident is not a sufficient basis to discard or verify a hypothesis.
To be verified (that is, regarded as increasingly plausible) a hypothesis must be indicated by the
data over and over again. An unsupported hypothesis must be critically evaluated to determine if
it is false or if the observed events indicate a variation of the hypothesis (different conditions,
indicating a different form). A major strategy in grounded theory is to seek systematically the full
range of variation in the phenomena under scrutiny. "Researchers too often buy a falsification
mode of thinking, which blinds them to the issue of variation and conditions." (David Maines,
personal communication.)
Let us again turn to an example for clarification. Suppose the analyst conceived the following
hypothesis: when cancer patients complain of pain and request relief, nurses provide comfort not
only by giving them pain medication, but also through touch, soothing talk, and so on. If in
another observation, however, a cancer patient complains of pain, yet the nurse does not respond
in the expected manner, the hypothesis is not necessarily false. The researcher should investigate
why the nurse did not respond as predicted. The incident may suggest a variation of the original
hypothesis, which can then be revised to include various new, provisional, conditional
relationships. Doing so makes the theory conceptually denser, and makes the conceptual linkages
more specific. The analyst can say: "Under these conditions, action takes this form, whereas
under these other conditions, it takes another.
3. Selective Coding. Selective coding is the process by which all categories are unified around a
"core" category, and categories that need further explication are filled-in with descriptive detail.
This type of coding is likely to occur in the later phases of a study.
The core category represents the central phenomenon of the study. It is identified by asking
questions such as: What is the main analytic idea presented in this research? If my findings are to
be conceptualized in a few sentences, what do I say? What does all the action/interaction seem to
be about? How can I explain the variation that I see between and among the categories? The core
category might emerge from among the categories already identified or a more abstract term may
be needed to explain the main phenomenon. The other categories will always stand in
relationship to the core category as conditions, action/interactional strategies, or consequences.
Diagramming can assist in integration of categories.
Poorly developed categories are most likely to be identified during selective coding. A poorly
developed category is one for which few properties have been uncovered in the data or for which
a subcategory contains only a few explanatory concepts. In order for a theory to have
explanatory power, each of its categories and subcategories must have conceptual density. When
this is lacking, the analyst can return to the field, or to fieldnotes to obtain data that will allow
gaps in the theory to be filled.
In some grounded theory studies, researchers have had difficulty making a commitment to a core
category. As one astute reader (Adele Clarke) of a draft of this paper wrote, " . . . at this stage one
can commonly confront several [analytical] schemes that can link it all together. Then one must
choose among them that which best captures the whole shebang." Her question was: how does
one do this? How does one compare "robustness in the face of alternatives"?
The conventional answer has been that sufficient coding will eventually lead to a clear
perception of which category or conceptual label integrates the entire analysis. Sometimes this
does not take place, however, even for experienced researchers. Then, the investigators have to
struggle with the problem of integration, playing one analytic scheme against the other to see
which captures the essence of what the research is all about. There are techniques for advancing
this process (Glaser, 1978; Strauss, 1987; Strauss and Corbin, forthcoming 1990.)
The generalizability of a grounded theory is partly achieved through a process of abstraction that
takes place over the entire course of the research. The more abstract the concepts, especially the
core category, the wider the theory's applicability. At the same time, a grounded theory specifies
the conditions under which a phenomenon has been discovered in this particular data.
A range of the situations to which it applies or has reference is thereby specified. In utilizing
theory, practitioners or others may encounter somewhat different or not-quite-the-same
situations, but still wish to guide their actions by it. They must discover the extent to which the
theory does apply and where it has to be qualified for the new situations.
A grounded theory is reproducible in the limited sense that it is verifiable.
One can take the propositions that are made explicit or left implicit, whatever the case may be,
and test them. However, no theory that deals with social psychological phenomena is actually
reproducible in the sense that new situations can be found whose conditions exactly match those
of the original study, although major conditions may be similar. Unlike physical phenomena, it is
very difficult in the social realm to set up experimental or other designs in which one can
recreate all of the original conditions and control all extraneous variables impinging upon the
phenomenon under investigation. When testing hypotheses derived from a grounded theory, the
investigator should specify the test conditions carefully and adjust the theory to fit them,
assuming they cannot match the conditions originally specified. The more abstract the concepts,
and the more variation uncovered in the original study, the more likely it is that the propositions
apply to a broad range of situations.
Another way of explaining reproducibility is as follows: Given the theoretical perspective of the
original researcher and following the same general rules for data collection and analysis, plus
similar conditions, another investigator should be able to arrive at the same general scheme. The
discrepancies that arise should be resolvable through re-examining the data and identifying
special conditions operating in each case.
A grounded theory is generalizable insofar as it specifies conditions that are linked through
action/interaction with definite consequences. The more systematic and widespread the
theoretical sampling, the more completely the conditions and variations will be discovered,
permitting greater generalizability, precision, and predictive capacity. If the original theory fails
to account for variation uncovered through additional research, the new specifications can be
used to amend the original formulation.
Criteria for Evaluating a Grounded Theory
The success of a research project is judged by its products. Except where results are only
presented orally, the study design and methods, findings, theoretical formulations, and
conclusions are judged through publication. Yet, how are writings to be evaluated and by what
criteria? As noted at the outset of this paper, different modes of research require different
methods and criteria of evaluation. When judging qualitative research, it is not appropriate to
apply criteria ordinarily used to judge quantitative studies. One aim of this paper has been to
show that the grounded theory approach accepts the usual scientific canons, but redefines them
carefully to fit its specific procedures. For any grounded theory study, the specific procedures
and canons described above should provide the major basis of evaluation.
It is important to recognize that in judging a research publication that claims to generate,
elaborate, or "test" a theory, the reader should distinguish four issues. First, judgments should be
made about the validity, reliability, and credibility of the data (Le Compte and Goetz, 1982;
Guba, 1981; Kidder, 1981; Kirk and Miller, 1985; Miles and Huberman, t984; Sandelowski,
1986).
Second, judgments should be made about the plausibility and value of the theory itself or, if the
publication is less ambitious, then of its modest theoretical formulations. Third, judgments
should be made about the adequacy of the research process which generated, elaborated, or
tested the theory. Fourth, judgments should be made about the empirical grounding of the
research findings.
We shall not address the criteria for judging either data or theories. The first have been much
discussed in the literature. Nor shall we offer criteria for judging the plausibility and value of
theories, a matter that belongs better in the province of philosophers of science. To the degree
that a grounded theory publication provides information bearing on the criteria for assessing its
data, research process and empirical grounding, however, its readers can evaluate its plausibility
and value. It is the latter issues that need discussion here--the assessment of the adequacy of the
research proces and the grounding of the research findings. We hope to offer evaluative guidance
to readers of grounded theory publications and suggest more systematic guidelines to authors
themselves. Our description may also stimulate researchers in other qualitative traditions to
specify and publish criteria for judging their own research processes and grounding their
empirical findings.
The Research Process
A grounded theory publication should help the reader to assess some of the components of the
actual research process on which it reports. However, even in a monograph--which, after all,
consists primarily of theoretical formulations and analyzed data--there may be no way that
readers can accurately judge how the researcher carried out the analysis. The readers are not
present during the actual analytic sessions, and the monograph does not necessarily help them to
imagine these sessions or their sequence. To remedy this, it would be useful for readers to be
given information bearing on the criteria given below. The detail need not be great even in a
monograph. But it should provide some reasonably good grounds for judging the adequacy of the
research process.
The kinds of information needed are indicated in these questions:
Criterion #1: How was the original sample selected? On what grounds (selective sampling)?
Criterion #2: What major categories emerged?
Criterion #3. What were some of the events, incidents, actions, and so on that indicated some of
these major categories?
Criterion #4. On the basis of what categories did theoretical sampling proceed? That is, how did
theoretical formulations guide some of the data collection? After the theoretical sample was
carried out, how representative did these categories prove to be?
Criterion #5: What were some of the hypotheses pertaining to relations among categories? On
what grounds were they formulated and tested?
Criterion #6: Were there instances when hypotheses did not hold up against what was actually
seen? How were the discrepancies accounted for? How did they affect the hypotheses?
Criterion #7: How and why was the core category selected? Was the selection sudden or gradual,
difficult or easy? On what grounds were the final analytic decisions made? How did extensive
"explanatory power" in relation to the phenomena under study and "relevance" as discussed
earlier figure in the decisions?
Some of these criteria are unconventional (for instance, emphasizing theoretical rather than
statistical sampling and the injunction to account for discrepancies explicitly) for most
quantitative and many qualitative researchers.
Yet, these standards are essential to evaluating grounded theory studies. If a grounded theory
researcher provides the pertinent information, they enable readers to assess the adequacy of a
complex coding procedure. Detail reported in this way and supplemented with appropriate cues
can, at least in longer publications, highlight thorough tracking of indicators, conscientious and
imaginative theoretical sampling, and so on.
Empirical Grounding of Findings
Criterion #1: Are concepts generated? Since the basic building blocks of any scientific theory is a
set of concepts grounded in the data, the first questions to be asked of any grounded theory
publication are: Does it generate (via coding-categorizing activity) or at least use concepts? And
what are their sources? If concepts are drawn from common usage (such as, "uncertainty") but
are not put to technical use, they are not parts of a grounded theory, for they are not actually
grounded in the data themselves. Any monograph that purports to present theoretical
interpretations of data based on grounded theory analysis should permit a quick, if crude,
assessment of concepts merely by a check of the index to determine whether the listed concepts
seem to be technical or common sense ones, and whether there are many of them. For a more
complete assessment of such points, one must at least scan the book.
Criterion #2: Are the concepts systematically related?
The key to scientific research is systematic conceptualization through explicit conceptual
linkages. So a grounded theory publication must be asked: Have such linkages been made? Do
they seem to be grounded in the data? Are the linkages systematically developed? As in other
qualitative writing, the linkages are unlikely to be presented as a list of hypotheses, set of
propositions, or other formal terms, but be woven throughout the text of the publication.
Criterion #3: Are there many conceptual linkages and are the categories well developed? Do the
categories have conceptual density? If only a few conceptual relationships are specified, even if
they are grounded and identified systematically, there is something to be desired in terms of the
overall grounding of the theory. In final integration, a grounded theory should tightly relate
categories to one another and subcategories in terms of the basic paradigm features--conditions,
context, actions/interactions (including strategies) and consequences. Categories should also be
theoretically dense, having many properties richly dimensionalized. It is tight linkages, in terms
of paradigm features and density of categories, that give a theory explanatory power. Without
them a theory is less than satisfactory.
Criterion #4: Is there much variation built into the theory?
Some qualitative studies report on a single phenomenon, establish only a few conditions under
which it appears, specify only a few actions/interactions that characterize it, and address a
limited number or range of consequences.
By contrast, a grounded theory monograph should be judged in terms of the range of variations
and the specificity with which they are analyzed in relation to the phenomena that are their
source. In a published paper, the range of variations discussed may be more limited, but the
author should at least suggest that a larger study includes their specification.
Criterion #5: Are the broader conditions that affect the phenomenon under study built into its
explanation?
The grounded theory mode of research requires that the conditions noted in explanatory analysis
should not be restricted to ones that bear immediately on the phenomenon under study. The
analysis should not be so "microscopic" as to disregard conditions that derive from more
"macroscopic" sources, such as economic conditions, social movements, cultural values, and so
forth.
Macrosocial conditions must not simply be listed as background material but linked directly to
the phenomena under study through their effect on action/interaction and, through these, to
consequences. Any grounded theory publication that omits the broader conditions or fails to
explicate their specific connections to the phenomena under investigation falls short in empirical
grounding.
Criterion #6: Has "process" been taken into account? Identifying and specifying change or
movement in the form of process is important to grounded theory research. Any change must be
linked to the conditions that gave rise to it. Process may be described in terms of stages or phases
and as fluidity or movement of action/interaction over time in response to prevailing conditions.
Criterion #7: Do the theoretical findings seem significant and to what extent?
It is entirely possible to complete a grounded theory study, or any study, yet not produce findings
that are significant. The question of significance is generally viewed in terms of a theory's
relative importance for stimulating further studies and explaining a range of phenomena. We
have in mind, however, assessing a study's empirical grounding in relation to its actual analysis
insofar as this combination of activities produces useful theoretical findings. If the researcher
simply follows the grounded theory procedures/canons without imagination or insight into what
the data are reflecting--because he or she fails to see what they really indicate except in terms of
trivial or well known phenomena--then the published findings fail on this criterion. Because
there is an interplay between researcher and data, no method, certainly not grounded theory, can
ensure that the interplay will be creative. Creativity depends on the researcher's analytic ability,
theoretical sensitivity, and sensitivity to the subreties of the action/interaction (plus the ability to
convey the findings in writing). A creative interplay also depends on the other pole of the
researcher-data equation, the quality of data collected or utilized. An unimaginative analysis may
in a technical sense be adequately grounded in the data, yet be insufficiently grounded for the
researcher's theoretical purposes. This occurs if the researcher does not draw on the complete
resources of data or fails to push data collection far enough.
This double set of criteria--for the research process and for the empirical grounding of theoretical
findings--bears directly on the issues of how fully verified any given grounded theory study is
and how this is to be ascertained.
When a study is published, if key components of the research process are clearly laid out and if
sufficient cues are provided, then the theory or theoretical formulations can be assessed in terms
of degrees of plausibility. We can judge under what conditions the theory might fit with "reality,"
convey understanding, and prove useful in practical and theoretical terms.
CONCLUSION
Two last comments about evaluative criteria may be useful. First, the criteria should not be
regarded as hard and fast evaluative rules, either for researchers or for readers who are judging
the publications of others. They are intended as guidelines. New areas of investigation may
require that procedures and evaluative criteria be modified to fit the circumstances of the
research.
Imaginative researchers who are wrestling with unusual or creative use of materials will, at
times, depart somewhat from "authoritative" guidelines for procedures. Having said this, we
strongly urge grounded theory investigators to adhere to its major criteria unless there are
exceptional reasons for not doing so. In such unusual cases, researchers should know precisely
how and why they depart from the criteria, say so in their writing, and submit the credibility of
their findings to the reader.
Second, we suggest that researchers using grounded theory procedures should discuss their
procedural operations, even if briefly, especially in longer publications. They should include a
listing of any special procedural steps taken in addition to the ones discussed in this paper.
Readers are then in a better position to judge the overall adequacy of the research. It would also
make readers more aware of how this particular research differs from research employing other
modes of qualitative research. Researchers themselves would become more aware of what their
operations have been and of possible inadequacies of their operations. In other words, they
would be able to identify and convey the limitations of their studies.