kerlind
Gerlese S. A
Centre for Educational Development and Academic Methods, The Australian National
University, Canberra, Australia
This paper focuses on the data analysis stage of phenomenographic research,
elucidating what is involved in terms of both commonality and variation in
accepted practice. The analysis stage of phenomenographic research is often not
well understood. This paper helps to clarify the process, initially by collecting
together in one location the more concrete of the existing descriptions of
phenomenographic analysis. The analytic process is then further elucidated by a
unique analysis of variation in practice, based on the principles underlying that
practice. This work was inspired by the ongoing efforts of John Bowden to
clarify the nature and rigour of phenomenographic research methods,
commencing in particular with Bowden and Walshs volume in 1994.
Introduction
Phenomenography is a relatively new approach to educational research, with the first
publications describing the approach appearing in the early 1980s (Marton, 1981,
1986). However, it has reached a surprising degree of popularity over the subsequent
25 years, particularly in the UK, Australia and Hong Kong, as well as in Sweden, its
point of origin (Bruce and Gerber, 1997). As a research approach, phenomenography
initially emerged from a strongly empirical rather than theoretical or philosophical
basis. Indeed, it is only recently that epistemological and ontological assumptions, a
theoretical basis and specification of methodological requirements underlying the
approach have been more clearly developed (Bowden & Walsh, 1994, 2000; DallAlba
& Hasselgren, 1996; Marton & Booth, 1997; Bowden & Marton, 1998; Bowden &
Green, 2005). Most recently, these developments have led to the proposal of a Variation
Theory of learning and awareness, with associated implications for approaches to learning and teaching (Marton & Tsui, 2004).
While methodological debates and critiques of phenomenography have become
more common over the past decade, these debates typically neglect to address the
issue of accepted variation in phenomenographic practice. This encourages a lack of
awareness of this variation among all but the most active phenomenographic researchers, and can lead to confusion about the nature of the approach. Aggravated by the relative lack of published discussion of phenomenographic methodology, this has led to a
situation in which critiques of the research approach may be founded on misunderstandings of the nature of phenomenography (e.g. Francis, 1996; Webb, 1997), and
Email: gerlese.akerlind@anu.edu.au
116
G.S. A kerlind
117
118
G.S. A kerlind
particular aspects or criteria, but still within a framework of openness to new interpretations, and the ultimate aim of illuminating the whole by focusing on different perspectives at different times. The whole process is a strongly iterative and comparative one,
involving the continual sorting and resorting of data, plus ongoing comparisons
between the data and the developing categories of description, as well as between
the categories themselves.
A primary feature of the constitution of categories of description is the search for key
qualitative similarities within and differences between the categories. In practical terms,
transcripts or selected quotes are grouped and regrouped according to perceived similarities and differences along varying criteria. At times the groupings precede explicit
description of the similarities and differences, at other times the groupings are made
according to tentative descriptions for categories, as a checking and validation procedure:
categories are tested against the data, adjusted, retested, and adjusted again. There is,
however, a decreasing rate of change and eventually the whole system of meanings is
stabilized. (Marton, 1986, p. 42)
119
[We] independently assigned the transcripts to particular draft categories . . . In discussing the categorisation of those transcripts, our focus was on determining the qualitatively different ways in which these students understood learning [the phenomenon
under investigation]. This process occurred at two levels of analysis. First, we
attempted to identify the conception of learning that was evident in each transcript
and second, we sought to clarify the features of each conception by comparing and contrasting it with the other conceptions that were emerging . . . When we had agreed on the
categorisation of many of the transcripts, we attempted to describe the most characteristic features of each conception, with constant reference to the transcripts . . .
At each stage of our discussions about what characterized each conception we read
the transcripts again, each time from a slightly different perspective as our initial understanding of them developed . . . We sought to formulate progressively more complete
and refined descriptions of the six conceptions. As we did this, we continually
sought evidence within the transcripts that either was consistent with our draft categories or conflicted with them. This procedure was carried out within each transcript
so that we always considered the transcript as a whole. In addition, we looked for commonality from one transcript to another within the same category. Through this process
we jointly drafted categories of description based on the evidence in the transcripts. In
refining those categories we engaged in a process of discussion that involved formulating or justifying each aspect of a category, referring back to the relevant transcripts as
we did so (DallAlba, 1994, pp. 7980).
She [the research assistant] was asked to read through the whole set of transcripts . . .
several times until she felt she was reasonably familiar with them. She was then to try to
construct a set of categories which she felt encompassed her perceptions of what the
students were trying to say. She then went back over the transcripts, adjusted the categories, and cycled between the categories and the transcripts until she felt she had a
reasonably stable set of categories. When she had completed this task, we met to
discuss the set . . . My task at this stage was to read through the transcripts, decide
whether I felt they reasonably represented the conceptions reflected in the transcripts,
and to adjust the categories in a way to construct a more logically related set. This
was done by analysing the categories in terms of their structural and referential
aspects. After some detailed discussions we agreed on a set of more logically related
categories. The research assistant took this set, again cycled between the categories
and the transcripts, adjusted the categories and produced a third set. We then cycled
through the whole process, until we felt we had developed a reasonably stable set of
categories . . .
The next stage was to return to the individual transcripts and analyse them in terms
of the categories we constructed. We did this independently . . . we examined the categorisations, and where there seemed to be mismatches, we returned to the transcripts,
and either adjusted our categories, adjusted our categorisations or left the mismatch
remaining, depending on our interpretation of the transcripts. This we repeated
several times over a number of meetings (Prosser, 1994, p. 34).
The matter of focus is all important. As you read the transcript there must be, to my
mind, a focus. In all of the analyses we have done, I have read all the transcripts many
times at least six and sometimes a dozen times. On each occasion, some new perspective is being sought in order to clarify what the student means. On each occasion, the
reading of the transcript is a new experience. To read the transcripts in order to query
the similarities and differences represented in say, version 3 of the categories of description, is a different experience from reading them all again in order to illuminate
120
G.S. A kerlind
version 10. The multiple readings are necessary in order to explore all possible perspectives and because whenever an aspect is being queried it must always, I believe, be
explored with reference to the whole transcript rather than one small section of it
(p. 48).
All the time I am reading a transcript, I have in the back of my mind the question
What does this tell me about the way the student understands terminal velocity [the
phenomenon under study]? In other words, what must terminal velocity mean to the
student if he or she is saying this or that? . . . Students often say similar things but
their underlying meaning is different . . . Students also express similar ideas in quite
different terms. These similarities and differences can only be discovered by holding
all the ideas in mind at one time and trying to draw a picture that explains the underlying
meaning of virtually the whole transcript. If the student understands terminal velocity in
this way, then it may be no surprise that he or she has described an aspect of the motion
in a particular way. But why has the student discussed another aspect of the motion in
another way if the way of seeing terminal velocity is as we thought? And so on
(Bowden, 1994b, pp. 5051).
All of the material that has been collected forms a pool of meaning. It contains all
that the researcher can hope to find, and the researchers task is simply to find it. This is
achieved by applying the principle of focusing on one aspect of the object and seeking
its dimensions of variation while holding other aspects frozen. The pool contains two
sorts of material: that pertaining to individuals and that pertaining to the collective. It is
the same stuff, of course, but it can be viewed from two different perspectives to
provide different contexts for isolated statements and expressions relevant to the
object of research . . . The analysis starts by searching for extracts from the data that
might be pertinent to the perspective, and inspecting them against the two contexts:
now in the context of other extracts drawn from all interviews . . . now in the context
of the individual interview.
One particular aspect of the phenomenon can be selected and inspected across all of
the subjects, and then another aspect, that to be followed, maybe, by the study of whole
interviews to see where these two aspects lie in the pool relative to the other aspects and
the background. In a study that involves a number of problems for solution, for
instance, the analysis might start by considering just one of the problems as tackled
and discussed by all the subjects, and then a selection of whole transcripts that
include particularly interesting ways of handling the problem. This process repeated
will lead to vaguely spied structure through and across the data that our researcher/
learner can develop, sharpen, and return to again and again from first one perspective
and then another until there is clarity (Marton & Booth, 1997, p. 133).
Variation in practice
Examining these descriptions of practice as a set provides a rare insight into what phenomenographic research involves in concrete terms. At the same time, variation in practice amongst researchers is also highlighted.
Variation in the amount of each transcript considered
While all researchers would acknowledge the importance of the context provided by the
larger transcript in interpreting any one segment of a transcript, practice varies from
considering the whole transcript (or large sections of the whole) related to a particular
121
issue (Bowden, 1994a, 1994b; Prosser, 1994; Bowden & Green, 2005) to the selection
of smaller excerpts or quotes seen as representing particular meanings (Svensson &
Theman, 1983; Marton, 1986). In the latter approach, the smaller chunks are separated
from the transcript and combined for analysis in one decontextualized pool of meanings, though these segments are interpreted within the larger interview context
(Marton, 1986). In the former approach, while certain sections of each transcript are
inevitably seen as more pertinent to the research question than others, they continue
to be considered in situ (Bowden, 2005).
The underlying argument in favour of the contextualized within the transcript
approach appears to be that the whole transcript should be seen and treated as a set
of interrelated meanings, which can best be understood in relation to each
other. From this perspective, the decontextualized from the transcript approach
carries the danger of reducing appropriate consideration of the context within which
the selected quotes are made, which might affect the perceived meaning (Bowden,
1994a, 1994b).
Conversely, while proponents of the decontextualized approach also agree on the
importance of considering the larger context when interpreting and selecting
excerpts from the transcripts, working with whole transcripts is seen as having the
danger of encouraging an analytic focus on the individual interviewee, rather than
the group of interviewees as a collective. Another possible argument in favour of
the decontextualized approach is that taking a whole transcript approach to analysis
may reduce the clarity of the key aspects of meaning that researchers search for,
because the meaning a phenomenon holds for an individual may vary during the
course of an interview. Furthermore, from a practical point of view, it is obvious
that some statements within a transcript seem to address the research theme more
directly than others. Selecting excerpts that seem to exemplify meanings present
in the larger interview, while removing perceived irrelevant or redundant components of the interview, should assist in making the data more manageable (Svensson & Theman, 1983).
Variation in emphasis placed on collaboration
Another key variation highlighted by the above descriptions of practice lies in the
emphasis placed on collaboration during the constitution of an outcome space. Most
phenomenographic researchers work individually during their data analysis.
However, some authors argue for the importance of bringing in additional researchers
during the analysis to encourage greater open-mindedness and awareness of alternative
perspectives, as a way of improving the final outcome space (Bowden, 1994b; Walsh,
1994; Trigwell, 2000).
The large number of existing phenomenographic doctoral theses indicates that highquality phenomenographic research can be accomplished as an individual researcher
working on ones own, though this does not preclude the possibility that group research
work may produce a better outcome. It is also relevant to acknowledge here that any
outcome space is inevitably partial, with respect to the hypothetically complete range
of ways of experiencing a phenomenon. So, what we are considering when we talk
about better or worse outcomes is more or less complete outcome spaces, not right
or wrong outcome spaces. Thus, an individual researcher can, at the least, make a substantial contribution to our understanding of a phenomenon, even if group research
might have taken that understanding further.
122
G.S. A kerlind
123
1. the degree to which the logical structure of the outcome space needs to emerge as
directly as possible from the data; and
2. the degree to which it may more explicitly reflect the professional judgement of
the researcher.
This is a question of degree only, as the final outcome inevitably reflects both the data
and researchers judgements in interpreting the data.
Walsh suggests that too strong an emphasis on constituting structural relationships
may lead to potentially ignoring aspects of the data. I would argue, however, that the
search for structural relationships does not necessarily involve ignoring data. For
instance, where data are perceived as indicating variation that does not appear to
form part of a logical relationship between categories, these data may be reported as
representing non-critical variation within one or more ways of experiencing, or as
kerlind & Kayrooz, 2003,
sub-categories of a primary category of description (e.g. A
pp. 341342). Furthermore, the structure of an outcome space need not always take
the form of a linear hierarchy of inclusiveness; branching structures or hierarchies
are also a possibility.
Validity and reliability: credibility and trustworthiness
Qualitative researchers are still traditionally expected to address issues of the validity
and reliability of their research, even though these notions derive from a positivist
approach to research that attempts to study an objective reality, rather than the more
intersubjective reality that most interview-based qualitative research is attempting
kerlind, 2005a). Consequently, these notions
to study (Guba, 1981; Kvale, 1996; A
need to be reframed within the context of the ontological and epistemological assumptions of the research approach being used. Phenomenography has much in common
with the assumptions underlying other qualitative research traditions, and thus draws
on their practices, as well as having differences that necessitate its own set of practices.
Validity
Validity is widely regarded as the extent to which a study is seen as investigating what it
aimed to investigate, or the degree to which the research findings actually reflect the
phenomenon being studied. However, a phenomenographic researcher asks not how
well their research outcomes correspond to the phenomenon as it exists in reality,
but how well they correspond to human experience of the phenomenon (Uljens,
1996). With the widespread understanding that an interpretive process can never be
objective and, in phenomenographic terms, represents the data as experienced by the
researcher (Svensson & Theman, 1983; Bowden, 1996; Sandberg, 1996; Marton &
Booth, 1997), the focus of research quality shifts to ensuring that the research aims
are appropriately reflected in the research methods used (Bowden, 1994b; Francis,
1996; Ashworth & Lucas, 2000).
Two types of validity checks, termed communicative and pragmatic validity by
Kvale (1996), are commonly practised within phenomenographic research, though
the extent to which each is utilized varies.
124
G.S. A kerlind
125
Reliability
From a qualitative research perspective, reliability may be seen as reflecting the use of
appropriate methodological procedures for ensuring quality and consistency in data
interpretations (Guba, 1981; Kvale, 1996). Two primary forms of reliability checks
on the influence of the researchers perspective on the research outcomes are commonly
used with qualitative, interview-based research (Kvale, 1996). Both involve the use of
several researchers for evaluating or offsetting the potential impact of having only one
researchers perspective on the data:
1. Coder reliability check, where two researchers independently code all or a
sample of interview transcripts and compare categorizations; and
2. Dialogic reliability check, where agreement between researchers is reached
through discussion and mutual critique of the data and of each researchers interpretive hypotheses.
Both checks are used within phenomenographic research to varying degrees of popularity; neither is uniformly used.
The principle of dialogic reliability checks has been argued for strongly by
Bowden (1994b, 1996) and illustrated by Prosser (1994), though it is not common
in phenomenographic research at this stage. Similarly, while some phenomenographic
researchers argue for the value of employing a coder reliability check (Prosser, 1994;
Marton, 1996), others regard this as inappropriate for phenomenographic research
(Sandberg, 1994, 1996). (It is also possible for coder reliability checks to be used
to inform revisions to the proposed outcome space, making this check more dialogic
in nature.)
Arguments for not employing a check of coder reliability reflect elements of the
arguments presented above for not checking the categorization of interviews with the
original interviewees. That is, the set of categories of description are based on an analysis of the set of interview transcripts as a group, not an individual transcript basis. This
means that a single transcript may represent more or fewer aspects of the phenomenon
being investigated than does a single category of description, making one-to-one
matching of transcripts and categories of description difficult.
A common alternative to these particular forms of reliability checks is for the
researcher to make their interpretive steps clear to readers by fully detailing the
steps, and presenting examples that illustrate them (Guba, 1981; Sandberg, 1994,
kerlind, 2005b). Indeed, Sandberg argues that a danger of
1996; Kvale, 1996; A
employing coder reliability checks is that it directs attention away from these more fundamental checks of research reliability. These checks involve documenting how
researchers have adopted a critical attitude towards their own interpretations, that is
how they have analysed their own presuppositions and the checks and balances that
they have employed to help counteract the impact of their particular perspectives on
the research outcomes.
Summary
This paper has focused on the data analysis stage of phenomenographic research, elucidating what is involved in terms of variation and commonality in accepted practice,
including:
G.S. A kerlind
126
how much of each transcript is considered at one time during the analysis;
the emphasis placed on analytic collaboration with other researchers;
variation in ways of practically managing the large amount of data involved;
the degree to which the logical structure of the outcome space is seen as needing
to emerge as directly as possible from the data versus more explicitly reflecting
the professional judgement of the researcher;
5. use of communicative and pragmatic validity checks; and
6. use of coder and dialogic reliability checks.
1.
2.
3.
4.
Acknowledgements
I first met John Bowden at a higher education conference in 1982; by chance, we sat next to each
other at the conference dinner. Over the subsequent 23 years, John has been a friend, colleague
and mentor, fulfilling each role to perfection. Thank you, John.
An earlier version of this paper was included in the non-refereed electronic proceedings of
the International Symposium on Current Issues in Phenomenography, held in Canberra, 2002.
References
kerlind, G.S. (2005a). Learning about phenomenography: Interviewing, data analysis and the
A
qualitative research paradigm. In J. Bowden & P. Green (Eds.), Doing developmental phenomenography (pp. 63 73). Melbourne: RMIT University Press.
kerlind, G.S. (2005b). Phenomenographic methods: A case study. In J. Bowden & P. Green
A
(Eds.), Doing developmental phenomenography (pp. 103 127). Melbourne: RMIT
University Press.
kerlind, G.S., Bowden, J., & Green, P. (2005). Learning to do phenomenography: A reflective
A
discussion. In J. Bowden & P. Green (Eds.), Doing developmental phenomenography
(pp. 74 100). Melbourne: RMIT University Press.
kerlind, G.S., & Kayrooz, C. (2003). Understanding academic freedom: The views of social
A
scientists. Higher Education Research & Development, 22, 327 344.
Ashworth, P., & Lucas, U. (2000). Achieving empathy and engagement: A practical approach to
the design, conduct and reporting of phenomenographic research. Studies in Higher
Education, 25, 295 308.
Bowden, J. (1994a). The nature of phenomenographic research. In J. Bowden & E. Walsh
(Eds.), Phenomenographic research: Variations in method (pp. 1 16). Melbourne:
RMIT: EQARD.
Bowden, J. (1994b). Experience of phenomenographic research: A personal account. In J.
Bowden & E. Walsh (Eds.), Phenomenographic research: Variations in method
(pp. 44 55). Melbourne, RMIT: EQARD.
Bowden, J. (1996). Phenomenographic research: Some methodological issues. In G. DallAlba
& B. Hasselgren (Eds.), Reflections on phenomenography: Toward a methodology?
(Gothenburg Studies in Educational Sciences No. 109). Gothenburg: Acta Universitatis
Gothoburgensis.
Bowden, J. (2005). Underpinnings of phenomenography: Theory and practice. In J. Bowden &
P. Green (Eds.), Doing developmental phenomenography. Melbourne: RMIT University
Press.
Bowden, J., & Green, P. (Eds.). (2005). Doing developmental phenomenography. Melbourne:
RMIT University Press.
Bowden, J., & Walsh, E. (Eds.). (1994). Phenomenographic research: Variations in method.
Melbourne: RMIT University Press, (subsequently revised as Bowden & Walsh (2000)).
Bowden, J., & Marton, F. (1998). The university of learning. London: Kogan Page.
Bowden, J., & Walsh, E. (Eds.). (2000). Phenomenography. Melbourne: RMIT University
Press.
Bruce, C., & Gerber, R. (1997). Editorial: Special issue: Phenomenograpy in higher education.
Higher Education Research & Development, 16, 125 126.
127