Generalizing: The descriptive struggle
Barney G. Glaser, Ph.D.; Hon Ph.D.
The literature is not kind to the use of descriptive
generalizations. Authors struggle and struggle to find and
rationalize a way to use them and then fail in spite of trying a
myriad of work-arounds. And then we have Lincoln and Guba’s
famous statement: “The only generalization is: there is no
generalization” in referring to qualitative research. (op cit, p. 110)
They are referring to routine QDA yielding extensive
descriptions, but which tacitly include conceptual generalizations
without any real thought of knowledge about them.
In this chapter I wish to explore this struggle for the purpose
of explaining that the various contra arguments to using
descriptive generalizations DO NOT apply to the ease of using
conceptual generalizations yielded in SGT and especially FGT. I
will not argue for the use of descriptive generalization. I agree
with Lincoln and Guba with respect to QDA, “the only
generalization is: there is no generalization.” It is up to the QDA
methodologists, of whom there are many; to continue the struggle
and I wish them well.
The Descriptive Generalization Struggle
Most, if not all, qualitative research method writers talk of
the near impossibility to generalize as they struggle to make
descriptive generalizations realistic. Most fail.
There are several dimensions to this struggle which help
explain the struggle and then the failure. Their principal
concerns of descriptive generalization are worrisome accuracy of
descriptions which soon become stale dated, transferability,
internal vs. external validity, unit comparisons to determine
similarity and differences (not for concepts), unit comparability
for transferability, volume solutions (the more units the better),
downing abstract leveling of SGT to a local description, and can a
descriptive generalization become a scientific law. The reader
may think of more, but considering these dimensions will give the
idea that the descriptive generalization struggle is never solved
and it does not apply to conceptual generalization. Indeed,
focusing on descriptive generalization in the struggle has two
negative consequences: 1. conceptual generalizations are missed
or passed over and 2. They leave the substantive fields involved
open to speculative theory. I will consider these dimensions in
linear fashion keeping in mind they are highly interrelated.
The writers I refer to are Lincoln and Guba (op cit, chapter
5), Ian Dey (op cit, chapter 11), Janet Ward Schofield, “Increasing
the Generalizability of Qualitative Research” in Miles and
Huberman, The Qualitative Researchers Companion, (op cit,
chapter 8), Margaret Kearny, “New Directions in Grounded
Formal Theory” in Using GT in Nursing,) op cit, chapter 12), and
Glaser, “Conceptual Generalizing” in the GT Perspective I,
chapter 7, and Joy L. Johnson, “Generalizability in Qualitative
Research,” Chapter 10. The many other writers such as Creswell,
Silverman, Walcott, Morse, Schutt, etc, on qualitative
methodology deal with the struggle to generalize but in less than
a chapter focused way. See bibliography for this book.
Missing conceptual generalization: One major source of
the descriptive generalization struggle is the down leveling of
SGT by the remodeling impact of QDA on GT. (See GT
Perspective II: Description Remodeling of GT). What occurs is that
QDA forces a description out of GT and/or GT is taken as
description, not as theory. It becomes local to the area of
research. Descriptive generalization becomes the problem. The
quest is to see if the description applies to another area, if the
area is comparable on enough dimensions. The pressure to
generalize releases a fearful caution of generalizing descriptions
as the research seems particularistic, not general. The fear is
turned into keeping the particular description special, possessed,
savored and uniquely original. Hopefully it is possessed for
colleague respect and career purposes. And by the time the
research gets published a year or so later, the description is stale
dated, the research site changed, thus even less generalizable.
It is no wonder that conceptual generalizing and general
implications of the core category is missed totally as the down
leveling descriptive struggle continues. It is no wonder that the
enduring grab of GT categories over time and place is lost and
their power to explain is lost, since the focus is so strongly on the
struggle to use the description elsewhere, or to block it. Keep in
mind that SGT is not a site description; it is, when done right, a
conceptual theory, which generalizes with ease. Missing the
conceptual nature of GT in total favor of QDA description is clear
in the following critique of GT generalizations by Ian Dey “Thus
the logic of discovery justifies procedures that maximize the
production of new ideas. But the same procedures do not provide
a strong basis for generalizing these ideas to particular
populations.” (op cit, p. 38) Further he says “Glaser and Strauss
legitimate such generalization by linking formal theory closely
with the substantive studies through which it is both generated
and applied. Nevertheless, there seem to be problems with the
use of grounded theory for generalizations. At least three such
problems stand out. One is the preference to sample situation and
processes rather than cases. This may make it difficult to locate
the resulting theory in its local context, and take this into account
when generalizing. A second is the use of theoretical sampling to
select these situations and processes. This tends to leave open the
question of how representative these may be. A third is the
temptation to generalize without reference to the specific spatial
and temporal context with which the generalization may apply
(ibid, p. 246).
Dey is clearly focused on the requirements of descriptive
generalization, and misses that SGT and FGT are conceptual, not
descriptive, and deal with conceptual generalization. Criticism
“one” clearly focuses on the descriptive need to sample for
description for a specific situation or context in order to describe
it well. Dey localizes GT, thus losing its conceptual abstraction
from time, place and people, by turning it into a description of
just what it is abstract of.
Criticism “two” continues this descriptive claim by wanting
theoretical sampling for data to abide by the rules of
representation clearly for describing a unit accurately.
Theoretical sampling is too “anywhere” as concepts drive it to
“wherever” for more data, to generate more conceptualization.
Again he localizes GT by down leveling it to a description of a
unit, which description is not carefully representative by using
theoretical sampling. He misses totally the idea that GT is
involved in conceptual development and using the
interchangeability of indicators and theoretical saturation, to stop
the excessive collection of data, required in QDA description.
Criticism “three” acknowledges the pressure to generalize,
but cautions against it if the original substantive area was not
properly described by representative sampling. Once again a
QDA requirement for accurate description. He misses the
conceptual generation of GT by theory driven sampling and then
when applying the concept, conceptually contextualizing it with
emergent fit. His contextualizing is down leveling GT to
description. Contextualizing comes after the GT generation, not
before as a description. After application then contextualizing a
concept earns relevance with emergent fit. Remember that SGT
and FGT originally come from data, but it gets applied to data
when used.
Focusing on descriptive generalization and missing
conceptual generalization leads Dey to this confused paragraph.
“Thus the problem of producing theory that is complex and
parsimonious is not so much resolved as recast in a new guise …
the distinction between substantive and formal theory allows
complexity in the generation of theory to be condensed into a
more parsimonious formulation at a formal level. Theoretical
reduction allows the elimination of superfluous specificity in the
construction of generalizations. Focusing on a core category
allows the research to set some boundaries to the analysis.” (p.
45) Extending the general implications of a core category by FGT
does not entail the reduction or increase in complexity and
parsimony, or the theory. Description would do this.
Conceptualization, as one goes from substantive to FGT, is
emergent in either direction. There is no preconceived “trade off
between complexity and parsimony in the process of
conceptualization” as Dey says. It is emergent.
Dey concludes this chapter with a descriptive conclusion
about GT: “In grounded theory the emphasis on comparison
across a range of “areas of inquiry” (not cases) may preclude
effective study of any particular phenomenon…. Grounded theory
offers a way of producing generalizations through comparison.”
(p. 229) He faults GT for not producing a fully accurate
description of an area, missing the conceptual concern of the area,
while even mentioning that its focus is on conceptual
generalizations, “by reducing the rigor by relaxing the canons of
comparative inquiry.” So in realizing what GT does, it is critiqued
by descriptive canons. Clearly, the struggle for descriptively
generalizing dismisses conceptual generalization. Faulting GT
with comparative inquiry that conceptualizes instead of
comparative inquiry that describes differences and similarities
again dismisses conceptual work. The power, use and grab of GT,
whether SGT or FGT is subverted without cause.
Generalizing in QDA research is tedious and tenuous. The
struggle can be so intense that others discount descriptive
generalizations as irrelevant. “For some qualitative researchers,
questions of generalizability are seen as irrelevant. ‘Naturalists
eschew generalizations on the grounds that virtually all
social/behavioral phenomena are contexts bound’ (Guba, op cit, p.
81). Generalizability simply is ignored or dismissed as an
oppressive, positivistic concept that hampers creative and
emancipatory qualitative research…. The tendency to reject
generalizability has to be their close alignment with (research)
which focuses on the study of unique cultures.” (Joy Johnson
“Generalizability in Qualitative Research” in Completing a
Qualitative Project, (chapter 10, Sage 1997). Joy closes this
thought about descriptive generalizing with the bear recognition
of conceptual generalization by quoting Guba (1985) “to
generalize one must develop abstract theories, yet abstractions
are not well grounded in what informants experience and think”.
She suggests using GT to obtain these abstractions. This thought
is close but does not go far enough to push for FGT.
It is simply true that QDA research gives new perspectives
on what is going on in a situation. It gives understanding, if not
grounded theory explanation of how the main concern is
continually processed. This descriptive perspective is abetted by
the natural tendency to over generalize particularistic and
speculative views as lessons on supposed patterns, not truly
grounded. Over generalizing broad statements, based on
particularistic limited assumptions and information, emerge to
achieve a credible, but unattainable goal to look like a scientific
law. Sometimes findings and criteria from quantitative studies
are used to broaden the credibility of the generalizations, since
the broad sampling and volume give the appearance of tapping
general descriptions. The unwavering rules of quantitative
validity when applied to QDA do not, in and of themselves,
generate descriptive generalizations that last. These efforts at
descriptive struggle for generalization ignore, waylay or dismiss
the ease and application of conceptual generalization.
The struggle for descriptive generalizations is described by
Johnson as follows: “It is clear that different qualitative
approaches are aimed at developing different kinds of knowledge.
The Grounded Theory Review (2006), vol.6, no.1
6
It is therefore inappropriate to assume that it is possible to
develop a generic set of procedures for enhancing generalizability
across all forms of qualitative research. All approaches and
strategies involve assumptions, judgments and compromises: all
are claimed to have deficiencies. The challenge for researchers is
to be aware of those deficiencies and to refrain from making
claims (generalizations) that extend beyond the purview of the
research study.” (Johnson, op cit, p. 203) “It is for the researcher
to be clear about the aim of the study and how the findings may
apply across settings, person and time.” (p. 205) Johnson again
closes with a slight recognition of conceptual generalization,
when she says. “The theory (referring to GT) if it has been
developed in a rigorous manner, is applicable across numerous
kinds of person and contexts,” but there is no follow through to
fostering doing SGT or FGT.
All the articles referred above show the struggle for
descriptive generalization, but the classic one is the chapter by
Lincoln and Guba stating the only generalization is that there is
no generalization. I said above that I agree and will detail more
reasons why in several subsequent sections in this chapter. Here
I want to underscore that the L&G struggle fails because they
look for unreachable truths and scientific laws as their goal,
whereas other strugglers do not. They discuss many ways that
QDA descriptions are not possible to generalize. This extreme
requirement for descriptive generalizations steers them far from
realizing the power of conceptual generalizations.
Additionally their struggle misses conceptual generalization
because they reduce GT to a description. They say “GT is local
theory as it brings together and systematizes isolated, individual
theory. Local understanding, aggregated leads to partial
understanding.” Lincoln and Guba reverse the escape from time,
place and people of conceptual generalization to wanting to
contextualize the GT and make it a unit description. Thus they
completely miss GT’s conceptual power of generalization by
missing that the constant comparative method yields concepts. It
does not yield similarities and differences leading to and for
description. They see GT as just another description of a locale or
context. What could have saved their struggle for generalization
is steered to assure failure of the struggle.
At least Schofield (in Miles and Huberman, Qualitative
Research Companion, Sage 2002, p.191–193) gives descriptive
generalization a bit of success by saying it is always conditional,
provisional and particularistic. She tries to revive the demise of
descriptive generalization by such conditions. Her struggle is a
stretch to use them, but still a mild failure. Her struggle, as is
typical, steers her clear of realizing conceptual generalizations
exist and are generated by SGT and especially FGT. She says:
“Although qualitative researches have traditionally paid scant
attention to the attaining of generalizability, sometimes even
disdaining such a goal, this situation has changed noticeably in
the past ten to fifteen years. A consensus appears to be emerging
that for qualitative researchers, generalizability is best thought
of as a matter of the ‘fit’ between the situation studied and the
other to which one might be interested in applying the concepts
and conclusions of that study. This conceptualization makes thick
description crucial…” So near and yet so far from GT, since
comparative description is the goal not conceptualization to as
she says “achieve generalizability through the aggregation of
extant independently designed case studies.” She says
“structuring qualitative studies in a way that enhances their
implication for the understanding of another situation,” but it is
all comparative description.
She is so near and yet so far from comparing for generating
conceptual generalizations. The descriptive struggle blinds
researchers to conceptual ease. The descriptive struggle is the
norm and one property of it is that it is perpetual. There are lots
of close calls, but the struggle diverts the best researchers to the
standard arguments of descriptive generalization, which
arguments are then doomed to failure. One among many reasons
is that comparisons are not made for conceptualization. They are
made for showing differences and similarities among units, thus
brought down to pure description. For example, as Schofield says:
“there is another approach to increasing the generalizability of
qualitative case studies that should not be ignored.… finding
ways to aggregate, compare or contrast already existing studies.”
Clearly, she is so close to conceptual comparison, but diverted to
descriptive comparisons.
The diversion to descriptive generalizations also have the
effect of subverting the GT requirement of letting the problem
emerge and not studying the literature of the field before the
research so as not to preconceive and force. The descriptive
generalization steers researchers to the professional problem and
to study the literature on it first before the research in order to
get a descriptive need for more research to fill in gaps. Fine, this
is pure QDA research, which by far outnumbers GT research.
The descriptive quest gets easily framed by the speculative
theory of scholarly great men/women which subsequently
legitimizes it. In spite of the resulting struggle to generalize, GT
is totally subverted by this pattern. And the researcher finds
himself working on the speculative concepts of these great men.
Paradoxically, the researcher is back to conceptualization. But it
is not grounded; rather it is conjecture brought on by the
orientation to and requirement of descriptive generalization, for
example see Dey (op cit, p.265) on the leveling of abstraction. This
leveling poses an interesting problem and paradox, since these
QDA authors know descriptive generalizations do not hold up, yet
they have no grasp of conceptual generalizations.
I am not saying descriptive generalizations are to be avoided;
they are just another focus of research. See Mark Granovetter:
“Finding a Job: 2nd edition” (Univ of Chicago Press, 1995) for a
masterful comparison of his study to others based on differences
and similarities. His writing is laced with conjectural accounting
for the findings leading to strained struggle for descriptive
generalization. Our focus in this book is the conceptual grounded
generation of the general implications of a substantive core
category into a FGT which is clearly quite a different research
goal. Needless to say under the notion that people generalize
naturally, there are those who do it descriptively by nature and
those who do it conceptually by nature. And of course, this is
another source of scant attention to FGT.
In sum, the unit oriented struggle to descriptive
generalization is never satisfactory. It is too absolute, too factual,
too philosophical by conjectural accountings, too inclusive in the
need to account for everything and too argumentative to try to do
it and thusly further driving the researcher to intuitive,
conjectural remarks and the borrowings of grand speculative
theory. No wonder that some QDA researchers thoroughly dislike
descriptive generalizations and seek only pure description.
Descriptive generalization never really works. The reader should
read the full articles referred to above to witness the struggle for
him/herself. In other published sources the struggle is in sections.
I turn now to looking at the elements of descriptive
generalizations which make them so unsatisfactory. These are
problems that conceptual generalization does not have. I will deal
with transferability, external vs. internal validity, worrisome
accuracy and single case studies. The reader should keep in mind
I only touch on these topics enough as regards the generalization
struggle. I do not give the full coverage the literature does, which
the reader can easily search for. I will close this chapter with a
section on conceptual comparison contrasted with descriptive
generalizing.
Transferability
Generalization implies the transferability of a finding from
one context or unit to another as valid. This makes the finding a
descriptive generalization. But is transferability possible? It
depends. “The level of generalizability depends entirely on the
sampling scheme used and on the demographic resemblance
between sample and target population” (Dahlgren, p. 50.) Using
face sheet data to establish sufficient resemblance between
context assumes a relevance that may not exist, hence
demographic resemblance between units may be of no relevance.
Using face data is often an effort to not only transfer the
finding to a similar unit, but to a larger unit. Thus a small study
is generalized to a much larger population based on demographic
data. For example, a study of one nursing school is generalized to
nursing training throughout the nation. This kind of descriptive
generalization is especially jeopardous, since the larger
population brings in so much atypicality that the larger
population is just largely different even though compared to a few
similarities.
How does one make units similar enough to transfer a
finding from one to the other? At the other end from demographic
similarity is judgment and logical reasoning, which Dahlgren
calls “analytic generalization” (op cit, p. 51). There is no
demographic resemblance between study unit and target unit. It
is the fit of the researched problem to the new unit that is
generalized. It is the comparability of the topic or the problem
that is of concern. This borders on conceptual generalization, but
only if it does not drop to the descriptive level and the problem is
fully conceptualized in a GT. In the hands of a QDA researcher it
will indeed drop in level of abstraction.
In between demographic and analytic generalizations is
comparative description of differences and similarities trying to
make more of the similarities than the differences to justify
transferability. Sometimes different cases are just likened with
little comparison. But no matter what technique is used to justify
a transfer of a finding from one unit to another, it is never totally
justified or works. It is always a struggle with a level of
unsatisfactory outcome. The transferred to unit is never known
about fully enough to sharpen the similarities. The researcher
cannot research every unit fully like the initially researched unit.
Thick descriptions (Geetz 1973) are not the answer as it leads to
overload of clarity. Furthermore descriptions of units do not stand
still, they do not endure; they become stale dated at whatever
pace. Thus even a cogent transfer of a finding between units,
making it general, is likely to be short lived and of limited
enduring value. The effort to make the transfer between unit
types based on descriptions is still doomed, since they are
destined to become stale dated.
Conceptual generality has none of these problems. It is just
applied and conceptually modified by contextualization that
varies the categories to fit. Variation in unit resemblance is just
grist for modifying the concepts of the theory. For example (and I
could give many) our theory of awareness of dying was based on
four contexts. Then we went to the premature baby ward and
discovered awareness was not an issue, the patients could not
hear their prognosis. Thus it was a totally open context and the
new concept was the hearability of patients.
Dahlgren et al (p. 52) agree with me, when they say: “In
conceptual qualitative research, such as Grounded Theory, the
results have transcended the empirical data. Here analytic
generalizations make sense.” Dahlgren et al, do research on
international health problems in which generalizations between
very different units are vitally needed irrespective of
demographics or similarities and difference, such as the study of
spousal abuse in “wherever” nations.
The applicability of GT, especially FGT, can be used wisely
by informed laymen, e.g. client, and by other GT researchers
using emergent fit in application. The generalizing is located by
applicability not by descriptive commonalities. The generalizing
is never a factual transfer, as in description, it is just multiple,
integrated conceptual hypotheses modified to fit where applied by
using constant comparison to conceptualize the modified fit,
workability and relevance.
This kind of generalizing is not only easy with some
thoughtful work, but is fun, as it feels good to have a conceptual
grasp on what is going on. For example, when one knows the
subcore categories of credentializing it is fun to see how they vary
in different credentializing contexts. Like, what kinds of reading
and lectures are involved, how long is the schooling (2 weeks to
10 years, so to speak), what kinds of final tests, what are entry
procedures, how much active experience, what kinds of final
legitimating ceremonies and so forth. I have seen the study of
worsening progression in many other areas such as chronic
illness, acute illness, smoking or other drug excesses, super
normalizing denial, excessive exercising and on and on. The
conceptual general implications of core categories know no
bounds in transferability, and the problems of descriptive
generalization for conceptual generalizations are moot.
Internal vs. External Validity
The transferability of descriptive findings to other units
brings on the fundamental problem of internal vs. external
validity. This demand on transferability usually dooms the
descriptive generalization from the start. Internal validity means
generalizing from a small study to all the people in the unit or
context studied. The boundaries remain the same, thus it is
cogent to generalize a finding to all involved. External validity
means transferring the finding to a different unit. External
validity, as we have seen above on transferability, raises the
suspicion that the finding cannot be legitimately transferred
because of grave differences in the ‘transferred to’ new unit.
One author suggests making the new unit typical regarding
the finding being transferred or better do multiple studies of
numerous new units for description generalization of qualitative
data. (RK Schutt, Investigating the Social World, Pine Forge
Press by Sage 2004, p. 154) This struggle is clear, especially the
limited by time and expense approach of doing multiple studies.
It is especially clear by Schutt’s half page mention of the dangers
of qualitative descriptive generalizability compared to six pages
on the generalizability of quantitative surveys and experiments
using random sampling for participants. Yet volunteer
participant experiments are suspect. The argument for external
transfers of descriptions increases in intensity the less it is
doable, therefore suggesting suspicion of or raising doubts of
transferability. Regarding external validity, “The criterion of
transferability is easily met in GT because Grounded Theory
almost automatically transfers findings” (Dahgren, op cit, p. 56).
The automatic transfer comes from the emergent fit of fit,
workability and relevance achieved by the constant comparative
method. The criterion of credibility is to a large degree irrelevant
in a SGT and FGT, once the GT is seen to fit work and be
relevant in another area.
A student wrote me (Judith Holton, 2/23/03,) “I believe that
there is another way in which GT is transferable and in a way
that is closer to the notion of external validity. The sociologist as
theorist provides this theoretical abstraction in many ways when
the theory fits, works and is relevant.” She is quite right. The
layman in the know spots these criteria instantly when the
theory he hears rings true and relevant. ‘“That’s right, that’s the
way it is” are comments we often hear upon presenting SGT to
the knowledgeable. These people then transfer the theory and
apply it. This is a type of generalizing (transferring) which is
done by laymen in the substantive areas which is quite different
than the researcher applying a GT for a client by consultation.
Here the layman applies it himself. “This aspect of transferability
is a very strong advantage for GT.”
Hans Thulesius experiences immediate seeing his GT core
category by others all the time when he gives a lecture to other
MDs on his GT on balancing palliative care. The MDs see it
immediately as “it works” and they want more. It engenders spill
relief — that’s what is going on! It engenders an imagery that
endures forever. It imbues thought and further interventions. It
helps the intelligent layman “go conceptual” without losing site of
the ground. This applicability whether by researcher or layman
occurs by the reversibility of the interchangeability of indicators
in the original GT. That is more indictors of the same latent
pattern are seen as the path to the applied to next unit or
context, so it applies with conceptual modification. Odis Simmons’
theory of “grounded action” leans heavily on this empirical tie to
the next unit.
Qualitative research design planning is sometimes based on
the possible future external validity of the yet to come findings.
The planning questions are many. What is the best population to
interview, how deep should the interviews be, versus more
extensive coverage, how much direct observation is advisable,
what kind of “truths” or goal is desired, what kind of data best
suits the wanted generalizations, what kind of qualitative method
is best used, should a few quantitative measures be used
especially for face sheet data. These are but a few of the questions
bearing on external validity.
Furthermore one author added to these routine research
design issues, how can the “fit be shown with the reality of the
wider world” by “involving a larger number of people” which is a
“tension” for qualitative researchers. (see Social Research: The
Basics: David and Sutton, Sage, 2004, p. 28) These are several of
the research design problems for the external validity goal.
Internal validity is less problematic, as it seems normal to
generalize to a larger number of people in the same population
being studied. David and Sutton further talk of depth validity
when it comes to internal validity. They say: “qualitative research
is associated with the prioritization of depth validity over
generalization.” (p. 34) Whatever the take on it, both forms of
validity are a struggle.
Maxwell highlights internal validity because he sees the lack
of trust in external validity. He says: “Internal generalizability in
this sense is far more important for most qualitative researchers
than external generalizability because qualitative researchers
rarely make explicit claims about the external generalizability of
their accounts. Indeed the value of a qualitative study may
depend on its lack of external generalizability. Thus internal
generalizability is a crucial issue in interpreting interviews.”
(Miles and Huberman, op cit, p. 54) He talks of how a researcher
cannot observe it all and interview everyone “even in one small
setting” thus some sort of generalization to a larger population is
needed. Hence internal validity predominates since it is safer to
generalize to similar people and situations, rather than take on
the challenge of external validity which includes total strangers.
Whatever the take on it, both forms of validity are a struggle.
Denzin and Guba detail Sartre, 1981, in their “soft” view of
internal validity. They say, “that no individual or case is ever just
an individual or case. It must be studied as a single instance of
more universal social experiences and social processes…. Thus to
study the particular is to study general. For this reason, any case
will necessarily bear the traces of the universal. The researcher
assumes that the reader will be able to generalize subjectively or
“naturally” from the case in question to their own experiences.”
(“Strategies of Qualitative Inquiry,” Sage 1998, op cit, p. xiv ,xv).
All I can say is of course people generalize naturally, BUT that is
not science. And the natural generalization is usually conceptual,
but D&G do not recognize this. So close a realization to seeing
conceptual generalization as the answer, but again a miss.
Denzin and Guba in another paper (p. 288), in reference to
“critical trustworthiness” they say, “… critical researchers reject
the traditional notion of external validity. The ability to make
pristine generalizations from one research study to another
accepts a one dimensional cause-effect universe. Many critical
researchers have argued that this traditionalist concept of
external validity is far too simplistic and assert that if
generalizations are to be made — that is, if researchers are to be
able to apply a finding of context A to context B — then we must
be sure that the contexts being compared are similar.” Again
descriptive struggle demands comparative descriptions of
similarities and differences.
As D&G say, researchers learn a lot from these comparisons,
but cannot safely generalize them. Although they tacitly
recognize conceptual generalization by acknowledging “in
everyday situations men and women do not make generalization
in ways implied by external validity…. They reshape cognitive
structures to accommodate unique aspects of what they perceive
in new contexts.” (p. 288). So they essentially state that everyday
persons contextualize the natural need to generalize. But they
miss the power of conceptual generalization, yet are so close.
Descriptive capture wins again. (See GT Perspective I, Glaser,
2001)
Anssi Perakyla struggles heroically to make conversation
analytic research generalizable in terms of external validity.
(“Reliability and Validity in Research Based on Natural Social
Interaction”, in Silverman editor, Qualitative Research
Companion, Sage 2004, p.266–268) First she doubts the
generalizability to other cases and then she discovers its
possibility if one takes exact descriptions and sees them as
comparative possibilities in other units. Thus she comes close to
conceptual generalization, but misses it. She says: “A crucial
dimension of validity of research concerns the generalizability of
research findings. How wide can the results from relatively small
samples be generalized. The comparative approach directly
tackles the question of generalizability by demonstrating the
similarities and differences across a number of settings …
whether the results present in studies are in any way
generalizable. Are they particular to a site or do the results have
some wider relevance. The possibilities of various practices can be
considered generalizable even if the practices are not actualized
in similar ways across different settings….As possibilities, the
practices that are analyzed are very likely to be generalizable, not
as a description of what other counselors or professionals do with
their clients, but they were generalizable as description of what
other professionals can do with his or her clients.” Sure,
possibilities on the conceptual level deal in probabilities, but she
cannot get off the descriptive level and label her practices
conceptually. So there is just struggle … when she is so near.
Dey’s (op cit, p. 219) approach to external validity is that of
bounded generalizations. They are never free of time and place,
but always bounded. He says: “Generalization about society or
social interaction must always be bounded by space and time.
Thus generalizations apply neither to the particular nor to the
eternal, but to events within some implicitly bounded space and
time in which they are assumed to occur.” Thus on the verge as
seeing them conceptually they must be described by a space and
time. Descriptive capture wins and we thusly have bounded
descriptive generalizations. Bounded generalizations supposedly
add to their external validity. Conceptual generalizations get
applied to a space and time, they are not described by them.
Bounded generalizations reverse contextualization in order to
stay with description.
The arguments for or against descriptive validity whether
internal or external are legion. The reader need only check the
index of any of the myriad of QDA methods books to see the pages
on generalization and witness the constant struggle. I have only
given examples here to give the reader the image of this struggle.
Careful grounded conceptual generalizations apply with ease, or
without struggle.
Worrisome Accuracy
In the Grounded Theory Perspective I: Conceptualization
Contrasted with Description, I discussed at length the QDA
methodologist’s concern with the accuracy of data collected in
qualitative research. I called it worrisome accuracy. They are
never sure the descriptions are trustworthy, credible, testable,
stale dated or confirmable. They engage in audit and member
checks to assure accuracy. They see much of their data as
interpreted — the constructivists — and interpretations vary,
which confounds any assurance of accuracy. What truths or
better yet what version of the truths is the researcher collecting
and reporting? What is described as “real” findings can always be
challenged. Given these constant doubts, how could a researcher
possibly generalize a descriptive finding with any confidence
other readers will accept it. What is fact for some is fantasy for
others. Transferability and validity of descriptive generalization
unit to unit are continually suspect and even seen as stale dated.
Dey agrees when he says: “Events are shaped by people with
their own particular perceptions, purposes and projects. However
GT seems to offer a way out of this impasse.” (p. 213.)
As I said in GT Perspective I, p. 50, “this tyranny of the QDA
quest for collecting accurate data is replaced in GT by the
conceptual coding of interchangeable indicators. The concepts
soon become abstract of time, place and people as they emerge.
When applied the concepts are easily modified by what ever
context they are applied to using the constant comparative
method to produce sub concepts. Modification, not verification,
yields credibility. The freedom and power of these concepts is
amazing and they yield conceptual generalizations, begging for
general implications and emergent fit. As I have said many times
in this book, core categories can be seen operating virtually
everywhere, whether by natural views or by further research for
FGT.
One standard strategy to assuage worrisome accuracy is to
test the finding. And then of course can the reader believe the
test procedure. The struggle continues. Donald Cressey in his
book Other People’s Money (1953, p. 156) rambles over this
problem in his conclusions. He says: “the search for negative
cases guided the research in all its phases and often a case which
is clearly exceptional to the theory has not been located. To the
question of whether negative evidence has been neglected or
unwittingly distorted, there is no positive answer. The fact that
our first hypothesis was revised several times before the final
generalization could be formulated implies that the final
generalization also must be revised if negative cases appear. In
other words, the testing of the theory must remain as somewhat
inconclusive in a single case study. The final test will be the
cumulative results of attempts at proof and disproof in research
which follows … crucial negative cases which, if found, will
require revision of the theory toward a more efficient
formulation.”
Thus negative finding test, to assure some accuracy, never
really works. This type of test just shows the perpetually found
inaccuracy of descriptive generalizations. Revision of the theory
by new researches is perpetually doomed by comparative
description, unlike the modification of a SGT by conceptual
comparison. The latter just adds more sub-concepts to the
emerging FGT as it is theoretically sampled for in new
researches.
Miles and Huberman in their classic book Qualitative Data
Analysis (Sage 1994, p. 263) talk of several ways of testing or
confirming findings for generalization. Their prescriptions are
numerous and beyond the resources and skill of most researchers.
They say: “Data quality can be assessed through checking for
representativeness (1) checking for researcher effects on the case,
and vice versa, (2) triangulating across data sources and
methods, (3) weighting the evidence and (4) deciding which kinds
of data are most trustable. Looking at ‘unpatterns’ can tell us a
lot. Checking the meaning of outliers, (5) using extreme cases, (6)
following up surprises, and (7) looking for negative evidence, are
all tactics that test a ‘pattern’ by saying what it is not like. We
can also test our explanations by making if-then tests (9)
replicating a finding, (10) ruling out spurious relations and (11)
checking out rival explanations. Finally, a good explanation
deserves attention from the very people whose behavior it is
about—the informants: getting feedback from them.”
What a struggle to be sure one has an accurate description
that could be generalized. The finding is doubted from the start to
the end. No average researcher could begin to test his findings
with such a long list. Credibility of findings is lost to doubt. Thus
why transfer a finding to another unit, why generalize under this
cloud of suspicion.
Clearly conceptual categories carefully generated by GT
procedures do not have this burden of accuracy, nor testing of it,
since they are abstract of time, place and people. For example are
the concepts of psuedo-friending or worsening progressions
accurate or not. The question is irrelevant, except for the fact
they are grounded concepts nor reified. Reified concepts, that is
concepts with no empirical references, are inapplicable, thus
inaccurate in this manner.
The QDA methodologists, as close as they come, do not have
a clue or thought for FGT generalizing. All they do is struggle
over descriptive generalizing and use all kinds of tactics to
increase the probability of the generalizing for transferability,
validity and accuracy. The criteria of credibility or accuracy
defeats transferability and validity from the start.
Increasing the volume of data by increasing the
representativeness of the data, the size of the unit or the number
of units researched, may increase accuracy, but while this is
routine doable for quantitative research, it is very difficult for
qualitative research. (See David Silverman, Qualitative Research,
second edition, Sage 2004, p. 295–299) on the doubts of small
samples and need to increase the number studies to generalize
descriptively.
Qualitative research deals in small numbers and small units
relative to quantitative research and therefore appear more
particularistic. As a basic goal of science, generalization does not
apply to qualitative research’s descriptive generalization. (See
Lincoln and Guba, op cit, p. 111).
Interestingly enough whether a generalization may be more
of less accurate based on the origin of the data from which the
generalization is made, may be moot for the researcher doing the
study and generalizing. It can be a self fulfilling, self referential
process of screening and evaluating the generalization based on
the framework used in the research, like it was accurate in the
first place. Thus the generalization is based on the generalized
framework applied, before hand, to the research so of course one
can generalize descriptions. For example if a researcher starts his
research on prostitutes with a preconceived framework based on
poor self images based on identity theory then he/she will start
generalizing about poor self images with his descriptions. Then in
fact, who knows if the theory and the findings are accurate and
not just self-fulfilling. Often the theory is speculative. The
descriptive generalizations that follow from a preconceived theory
do not test the theory. They simply continue with support for the
preconceived speculation. The conjecture increases non stop until,
if ever, the claims lead to evidentiary and accuracy problems.
The summary of this section is said well by Anssi Peraklya
(Silverman, op cit, p. 299) “All serious qualitative research
involves assuring the accuracy of recordings and testing of
truthfulness of analytic claims.” Indeed the only way to get away
from this struggle is to turn to conceptual generalization. But
realizing this in the face of the squelching by worrisome accuracy
of descriptive generalizations is difficult for QDA methodologists.
The Single Case
A lot can be learned from a small single case when
generating concepts from it that name latent patterns and the
concepts have general implications. This affirmation somewhat
answers Anssi Perakyla’s question (op cit, p. 295) “How widely
can the results, derived from relatively small samples be
generalized.” The answer is not widely if they are descriptive
generalizations and quite extensively if they are GT categories.
Silverman in Doing Qualitative Research, (Sage 2000, p.
102), states referring to case studies, “make a lot out of a little.”
And this means to make it analytically interesting. However he
says “nagging doubts remain.” This doubt surfaces in a regular
refrain heard from student researchers. “I have so few data, only
just one case. How can I possibly generalize about it?” Silverman
continues by quoting Stake, “Stake refers to the intrinsic case
study where the case is of interest in all its particularity and
ordinariness. No attempt is made to generalize beyond the single
case or even to build theories.” Then Silverman quotes Jennifer
Mason who says, “Qualitative research should therefore produce
explanations which are generalizable in some way or which have
a wider resonance (1996). Silverman then concludes with what
the reader now knows clearly: it’s a struggle. He says: “the
problem of representativeness is a perennial worry of many
qualitative researchers. Can we generalize from cases to
populations without following a purely statistical logic.”
Silverman concludes with four possible solutions to “obtain
generalizability: 1. combining qualitative research with
quantitative measures of populations, 2. purposive sampling
guided by time and resources, 3. theoretical sampling, and 4.
using an analytic model which assumes that generalizability is
present in the existence of any case.” (p. 103) All these solutions
focus on obtaining accuracy to the best of one’s ability and all are
beyond the resources typically of the lone qualitative researcher
in the field, just collecting data.
Using qualitative with quantitative data is hard unless the
quantitative literature — article — backs up a qualitative
finding. We have touched on research guided by preframing
analytic models as being self-fulfilling, unless the model comes
from a GT. Purposive sampling and theoretical sampling will
work, as we will see in Chapter 5, if the researcher does sampling
for conceptualization not description. Silverman is still on the
descriptive level trying to achieve accuracy to justify
transferability and validity. But he is very close to doing FGT, if
he would only turn to constant conceptual comparison and drop
seeking similarities, differences and negative cases to achieve
descriptive generalizations. But alas, not so. He says: “The
comparative approach directly tackles the question of
generalizability by demonstrating similarities and differences
across a number of settings” (Perhyl, 1997, p. 214) “In this sense
the literature review has as much to do with the issue of
generalizability as with displaying academic credentials.” The
reader would enjoy his chapter. Clearly single case studies put an
excessive strain on descriptive generalization.
We are at the point in this chapter where the ideas offered
and the literature covered are fairly conclusive in resulting in the
struggle for making descriptive generalizations. Now I wish to
turn to my section on single case generality in The GT Perspective
I, Sociology Press 2001, pp. 96–98).
Case Generality
In seems like a conflict of intent but case studies are
conceptually generalizable in many ways. A case study is a study
of a specific case, in depth, intensely and descriptively. It is
specific, special, unique, yet relevant. Relevance itself implies
general significance. The latent patterns within the case, as
revealed descriptively, are used as a basis for generalizing
conceptually. For example our case study of a patient dying of
cancer in hospital brought out many of the general problems of
dying in hospital. For example the properties of lingering status
passages. (See Anguish, Strauss and Glaser, Sociology Press,
1970)
From a FT conceptual point of view, the pressure to
generalize aspects of this case was great, WHETHER OR NOT it
was a typical case or an atypical case, if the generalizing is
conceptual. If the generalizing is descriptive, we have the
struggle showing similarity between units or if atypical, the
contrast.
If the generalizing is conceptual, the privacy and confidence
of the case and the people therein is maintained because the
conceptualization is abstract of time, place and people. The source
remains anonymous. If the generalizing is descriptive, confidence
is easily broken, even violated, and most importantly so for well
known cases preferring anonymity. GT conceptualization
detaches itself from the intensely specific descriptions.
Often case studies are done because there is a special issue
involved within the case, such as organizationally produced error,
a violation of a normative social process, or an untoward travel
disaster because of lack of cautionary control, etc. Yet the very
issue and its structural production, has great general implication
for what to do in other similar cases. The issue, descriptively, will
often be distorted descriptively, by multiple realities,
impressions, confusion and by impermeable complexities and
changes over time, etc. Again descriptive generality is poor
because of inaccuracies in data description. If GT is used to
generalize conceptually using constant comparison, the
inaccuracies become the data and are conceptualized as part of
the issue and are easily generalizable. A theory explaining
aspects of concern about the issue is generated. The issue
becomes an area of interest, as we say in GT.
A descriptive case study tells the whole story. This is totally
unnecessary in GT research, given its delimiting procedures
yielding a theory about a concept. But the GT researcher can
start with the details of an existing case and constantly compare,
generate a core category and their properties on the issue and
start a substantive theory about its resolution which has general
implications, hence could be taken to a FGT level. The case study
and perhaps other cases can be theoretically sampled. The other
comparable cases may be only knowledge fragments, but that is
all the data that is necessary for theoretical sampling on a
category. Again the uniqueness of the case study is lost in its
general implications by a GT approach to it.
This is a form of secondary analysis; that is the researcher
uses ongoing research on a case as data, while bringing in other
data from extant case studies for generating a substantive theory.
A lot is learned from the case as a result of the secondary
analysis. The case study is no longer a concentrated bounded
inquiry with a focused description. It starts a conceptualized GT.
It becomes both a unique, intrinsically interesting research
problem and a start of a substantive theory based on the latent
patterns in the case study data, which with theoretically sampled
secondary data bring out the main concern of the participants
and its constant resolving by a core category. The complexity of
the description which may be hard to grasp or relate to other
data, becomes organized conceptually and related easily by the
GT which provides conceptual handles.
The case’s latent patterns become abstracted and
generalized using secondary analysis and theoretical sampling.
The smallest aspect of the case study can lead to a main concern
with great relevance. The unit orientation of the case study with
struggle for descriptive generalization to other units changes to a
conceptual generalization that can be applied wherever with
emergent fit.
For example the study of one adolescent’s transition to
college can begin to yield a substantive GT theory on the
haphazard status passage from high school to college of working
class children. (see Toni A William Sanchez, PhD dissertation,
1998). Or, the study of one non-political social movement such as
female domestics seeking health care benefits, can lead to
generating a GT on non-political social movements.
If a case study is actually one of a class of cases studied, it
makes the secondary analysis easier with more varied data.
Studying a class of cases — many cases — may appear to yield
better empirical generalization because of the survey effect. Yet it
is still a struggle to generalize descriptions to other units. A GT
generated from relevant categories emergent in the case will
always work better.
In summary, the purpose of the case study is not to represent
the world, but to represent the case. However, the utility of case
study research to practitioners and policy makers is in its
extension of experience: to wit its conceptual generalization.
Ultimately, the case study researcher is interested in a
conceptual process for a population of cases, not the individual
case. Thus GT secondary analysis leads effective particularization
of a case to valuing conceptual generalization with confidence.
The value of a case study is thusly achieved in great measure by
a grounded theory approach to it.
Now I would like to summarize as a wrap up of this chapter
by quoting a section in the Grounded Theory Perspective I,
(Sociology Press, 2001, p. 90–94). I beg the reader’s forgiveness
for the redundancy, but the points are so significant for doing
FGT that a little repetition is in order. This section will make a
good quote for those researchers who are trying to put forward a
FGT for career and colleague purpose, as well as for a
contribution. If the reader is total confident in knowing what was
said above, this section can be skipped.
Conceptual Contrasted with Descriptive
Generalizing
In contrast QDA generalizing differs from GT generalizing
substantially, because the former is on a descriptive level and the
latter on a conceptual level. QDA generalizing of description is
often very difficult and based on assumptions that do not hold.
The problem is does a set of findings that hold in one unit, hold in
another unit, whether the unit is at the same level or a larger
unit. Sufficient commonality dimensions must be ascertained
between the units to apply or generalize the findings in one to the
other. This is a stretch that is difficult, and even if done, is short
lived as the contexts are always changing. For example, does a
finding in one nursing school apply to another or in all nursing
schools? Particularism impacts at every point.
Generalizing a finding from one unit to another is often done
by a subsequent researcher using random sampling to achieve
commonalities. It can also be done by replication and testing to a
modest degree. This is expensive in time and money, making GT
conceptual generalizing faster and more economical. GT can
generalize faster and better through conceptual constant
comparisons, thus raising the level of generality of the
descriptions of both units. Given the short life of description and
whatever the qualitative method used, the problem remains of
keeping accuracy accurate long enough to generalize descriptively
for more time than the short run.
As we have seen above, the problem for description
generalization is how PROPERLY to get the descriptions to a
generalizing level, SINCE it is so natural to bust the limits of a
description and see it generally anyway. Using random sampling
in another larger unit or roughly equal units, feels good, since
piling up units feels general even if the required commonalities
are not there. The analysis of commonalities easily becomes
conditionally contorted. Using negative or deviant cases or
similar dimensions of another unit is standard and works to some
degree in the struggle. Selective, bias reanalysis of the case
enters to iron out relevant non-commonalities.
Some will combine a qualitative research with a quantitative
survey of a larger population that touched on the same finding to
indicate how it may be general. The researcher will be lucky to
find this backup. He must search the quantitative literature
extensively to find such support of generality.
The use of an analytic theory model which assumes that
generalizability is present in a particular description is also used.
Of course, this easily can compound the lack of grounding if the
theory model is speculative. For example one can use traditional
self-image identity, role theory, reference group theory to
generalize a finding. We find in GT that this is a deductive,
conceptual elaboration which is usually dangerously irrelevant,
does not fit and only works in the fertile mind of the author. GT
researchers love to bust these erudite, speculative myths.
Statistical descriptions accomplish a probability
generalizing, but of course, qualitative data lacks statistics. A
QDA researcher can by piling up descriptions — use volume —
use probability statements that the description is general, but
where they get the qualitative statistics who knows. Usually the
need comes from a wish to increase the relevancy of “tiny topic”
research by an “ought” deduction.
“Tiny topic” research in QDA comes from a preference to
study people’s manifest problem rather than a conceptual, latent
problem. This can lead to generalizing more with empathy for the
manifest problem, yet with little abstraction. The descriptive
patterns are reduced to a description of one tiny topic, for
example: “Lived Experience in Early Stage Dementia,”
(Qualitative Health Research, April 2004, p. 453). Generalizing to
another unit is a struggle at best, but easily ignored for the
specific problem concern. GT generalizing stays on a conceptual
level of analysis and is applied to similar structural units with
ease.
“Tiny topic” research also focuses on routine but pressing
problems in the medical, nursing, educational and business fields.
It is sometimes very hard to look beyond the problem description
to the more general, conceptual implications, because it is so
important. It is hard to think theoretically, when the data is so
significant. For example it was hard for Hans Thulesius to go
from studying palliative cancer care to the abstraction of
balancing care.
Another QDA researcher says, in the struggle to generalize a
tiny topic, “As the numbers of cases is increased, so does the
scope of the generalizability.” An example of such descriptive
research is that of demonstrating the comforting role of the
trauma nurse when the patient is conscious. This descriptor will
enable comfort talk to be taught and eventually formally
integrated as a part of the responsibility of trauma nursing. The
description becomes conceptual and then becomes general to all
trauma situations of any sort. Thus GT’s ability to generalize
easily gets used by QDA researchers without them really
knowing it, in spite of their cautionary statements such as about
“limited causal implications.” GT concepts have such grab they
break through description capture, the pressure to generalize is
so great.
Janice Morse reflects on this pressure and struggle to
generalize QDA and its not great success. “In explanatory theory,
concepts and linkages are identified and described. These
theoretical ideas are complex and important. However, few have
been developed from qualitative research because of the
limitation inherent in qualitative method in sample size, and the
context bound nature of qualitative inquiry.” She suggests “using
two or more qualitative studies” simultaneously so that
qualitative findings can be broadened and inform quantitative
research. Once again the piling up of researched units is used to
raise the level of abstraction. She says: “It is the level of
abstraction reached, the quality of the interpretation and the use
of concepts and principals of abstraction that make the theory
generalizable.” Her struggle to generalize description gets so close
to conceptual generalization, but never quite reaches it because of
descriptive capture. Her journal “Qualitative Health Research”
more and more has become a journal of tiny topic research as her
dictum for theory generation has been ignored, forgotten or just
plain hopeless for so many researchers.
Others struggle with the pressure to generalize while
unawaredly in conflict with the simple properties of QDA
description. Listen to Howard Becker, a well known QDA
researcher: “Sampling is a major problem for any kind of
research. We cannot study every case of whatever we are
interested in, nor should we want to. Every scientific enterprise
tries to find out something that will apply to everything of a
certain kind by studying a few examples, the results of the study
being, as we say, ‘generalizable’ to all members of that class of
stuff. We need the sample to persuade people that we know
something about the whole class.” (Becker, “Tricks of the Trade”
1998, p. 67) Here we see the same generalizing model referred to
many, many times in this chapter. Many cases and random
representation of one case making it typical of larger cases. Both
are too difficult in QDA, not realistic, and so simple persuasion is
necessary.
Alasuutari suggests another type of struggle: change words!
She says: “Perhaps ‘generalizability’ is the wrong word to describe
what we attempt to achieve in qualitative research.
Generalization is a word that should be reserved for surveys only.
What can be analyzed instead is how the researcher
demonstrates that the analysis relates to things beyond the
material at hand … extrapolation better captures the typical
procedures in qualitative research” (1995, p.156–157). The spin
and struggle continues.
In all these solutions the QDA generalizing problem remains
a struggle which solutions are never quite believable. Pressure to
generalize makes the wish fulfillment “so”, but data and strategy
doubts have an easy time in making it not “so”. It is clear that
generalizing to a population or a unit is very hard, and often very
attackable by those who do not want it applied to them or by
them.
The descriptive capture struggle is moot for GT which easily
generalizes a conceptualization of a range, typology, process,
tolerance limits or any core category. GT procedures can help a
QDA description get generalized by doing some theoretical
sampling and constant conceptual comparison. It raises the level
of description to the abstract general level of conceptualization.
For example a good QDA on stock broker selling of
international securities was easily conceptualized as using “story
selling”. Nonrelevant stories told by the client made him feel
comfortable and trusting, whether or not the story was a sharable
or unique experience. Thus the client was disposed to buying.
Story selling and its compadre story talk is used by all of us to
gain comfort, trust and sharing with others. A description was
raised to a conceptual level with general implications.
Clearly GT conceptual generalizing applied with ease and
emergent fit by constant comparison is the powerful way to go.
Description generalization from unit to unit leaves too much to
struggle and subsequent doubts.
_____________________________
In sum, let the QDA methodologists continue the perennial
struggle to find solutions to descriptively generalize. There is no
real, lasting solution for them. It is their problem. I wish them
the best. In this book we focus on conceptual generalization which
occurs with ease, is seen everywhere and is applicable with
emergent fit, when doing FGT. We now turn to procedures for
generating a FGT.
