METHODS
published: 25 June 2019
doi: 10.3389/feduc.2019.00053
Frontiers in Education | www.frontiersin.org 1June 2019 | Volume 4 | Article 53
Edited by:
Douglas F. Kauffman,
Medical University of the
Americas–Nevis, United States
Reviewed by:
Jana Uher,
University of Greenwich,
United Kingdom
Barbara Hanfstingl,
Alpen-Adria-Universität Klagenfurt,
Austria
*Correspondence:
Nina Baur
Specialty section:
This article was submitted to
Educational Psychology,
a section of the journal
Frontiers in Education
Received: 30 June 2018
Accepted: 22 May 2019
Published: 25 June 2019
Citation:
Baur N (2019) Linearity vs. Circularity?
On Some Common Misconceptions
on the Differences in the Research
Process in Qualitative and Quantitative
Research. Front. Educ. 4:53.
doi: 10.3389/feduc.2019.00053
Linearity vs. Circularity? On Some
Common Misconceptions on the
Differences in the Research Process
in Qualitative and Quantitative
Research
Nina Baur*
Department of Sociology, Technische Universität Berlin, Berlin, Germany
Methodological discussions often oversimplify by distinguishing between “the”
quantitative and “the” qualitative paradigm and by arguing that quantitative research
processes are organized in a linear, deductive way while qualitative research processes
are organized in a circular and inductive way. When comparing two selected quantitative
traditions (survey research and big data research) with three qualitative research traditions
(qualitative content analysis, grounded theory and social-science hermeneutics), a
much more complex picture is revealed: The only differentiation that can be upheld is
how “objectivity” and “intersubjectivity” are defined. In contrast, all research traditions
agree that partiality is endangering intersubjectivity and objectivity. Countermeasures are
self-reflexion and transforming partiality into perspectivity by using social theory. Each
research tradition suggests further countermeasures such as falsification, triangulation,
parallel coding, theoretical sensitivity or interpretation groups. When looking at the overall
organization of the research process, the distinction between qualitative and quantitative
research cannot be upheld. Neither is there a continuum between quantitative research,
content analysis, grounded theory and social-science hermeneutics. Rather, grounded
theory starts inductively and with a general research question at the beginning of analysis
which is focused during selective coding. The later research process is organized in a
circular way, making strong use of theoretical sampling. All other traditions start research
deductively and formulate the research question as precisely as possible at the beginning
of the analysis and then organize the overall research process in a linear way. In contrast,
data analysis is organized in a circular way. One consequence of this paper is that mixing
and combining qualitative and quantitative methods becomes both easier (because the
distinction is not as grand as it seems at first sight) and more difficult (because some tricky
issues of mixing specific to mixing specific types of methods are usually not addressed
in mixed methods discourse).
Keywords: research process, mixed methods, survey research, big data, qualitative content analysis, grounded
theory, social-science hermeneutics, objectivity
Baur Linearity vs. Circularity?
INTRODUCION
Since the 1920s, two distinct traditions of doing social science
research have developed and consolidated (Kelle, 2008, p. 26
ff.; Baur et al., 2017, p. 3; Reichertz, 2019), which are typically
depicted as the “qualitative” and the “quantitative” paradigm
(Bryman, 1988). Both paradigms have a long tradition of
demarcating themselves from each other by ignoring each other
at best or criticizing as well as pejoratively devaluating the
respective “other” tradition at worst (Baur et al., 2017, 2018,
pp. 8–9; Kelle, 2017; Baur and Knoblauch, 2018). Regardless,
few authors make the effort of actually defining the difference
between the paradigms. Instead, most methodological texts
in both research traditions make implicit assumptions about
the properties of “qualitative” and “quantitative” research. If
one sums up both these (a) implicit assumptions and (b) the
few attempts of defining what “qualitative” and “quantitative”
research is, the result is a rather crude and oversimplified picture.
“Qualitative research” is typically depicted as combination of
the following elements (Ametowobla et al., 2017, pp. 737–776;
Baur and Blasius, 2019):
– an “interpretative” epistemological stance (e.g., Knoblauch
et al., 2018) which is associated e.g., with phenomenology or
social constructivism (Knoblauch and Pfadenhauer, 2018) or
some branches of pragmatism (Johnson et al., 2017);
– a research process that is circular or spiral (Strübing, 2014);
– single case studies (Baur and Lamnek, 2017a) or small
theoretically and purposely drawn samples meaning that
relatively few cases are analyzed (Behnke et al., 2010,
pp. 194–210);
– for these cases, a lot of data are collected, e.g., by qualitative
interviews (Helfferich, 2019), ethnography (Knoblauch and
Vollmer, 2019) or so-called “natural” data, i.e., qualitative
process-produced data such as visual data (Rose, 2016) or
digital data such as web videos (Traue and Schünzel, 2019),
websites (Schünzel and Traue, 2019) or blogs (Schmidt, 2019).
In all these cases, this means that a lot of information per case
is analyzed;
– both the data and the data collection process are open-ended
and less structured than in quantitative research;
– data are typically prepared and organized either by hand or
by using qualitative data analysis software (such as NVivo,
MAXqda or Atlas/ti);
– data analysis procedures themselves are suitable for the more
unstructured nature of the data.
In contrast, “quantitative research” is seen as a
combination of (Ametowobla et al., 2017, pp. 752–754;
Baur and Blasius, 2019):
– a “positivist” research stance;
– a linear research process (Baur, 2009a);
– large random samples meaning that many cases are analyzed;
– relatively little information per case, collected via (in
comparison to qualitative research) few variables;
– data are collected in a highly structured format, e.g., using
surveys (Groves et al., 2009; Blasius and Thiessen, 2012;
Baur, 2014) or mass data (Baur, 2009a) which recently have
also been called “big data” (Foster et al., 2017; König et al.,
2018) and which may comprise e.g., webserver logs and log
files (Schmitz and Yanenko, 2019), quantified user-generated
information on the internet such as Twitter communication
(Mayerl and Faas, 2019) as well as public administrational data
(Baur, 2009b; Hartmann and Lengerer, 2019; Salheiser, 2019)
and other social bookkeeping data (Baur, 2009a);
– the whole data collection process is highly structured and as
standardized as possible;
– data are prepared by building a data base and analyzed using
statistical packages (like R, STATA or SPSS) or advanced
programming techniques (e.g., Python);
– data are analyzed using diverse statistical (Baur and Lamnek,
2017b) or text mining techniques (Riebling, 2017).
Once these supposed differences are spelled out, it immediately
becomes obvious how oversimplified they are because in social
science research practice, the distinction between the data types
is much more fluent. For example, “big data” are usually
mixed data, containing both standardized elements (Mayerl
and Faas, 2019) such as log files (Schmitz and Yanenko,
2019) and qualitative elements such as texts (Nam, 2019) or
videos (Traue and Schünzel, 2019). Accordingly, it is unclear,
if text mining is really a “quantitative” method or rather a
“qualitative” method. While the fluidity between “qualitative” and
“quantitative” research becomes immediately obvious in big data
analyses, this issue has also been lingering in “traditional” social
science research for decades. For example, many quantitative
researchers simultaneously analyse several thousand variables.
Survey research has a long tradition of using qualitative methods
for pretesting and evaluating survey questions (Langfeldt and
Goltz, 2017; Uher, 2018). Almost all questionnaires contain open-
ended questions with non-standardized answers which have to
be coded afterwards (Züll and Menold, 2019), and if interviewees
or interviewers do not agree with the questionnaire, they might
add comments on the side—so-called marginalia (Edwards et al.,
2017). During data analysis, results of statistical analyses are often
“qualified” when interpreting results. While Kuckartz (2017)
provides many current examples for qualification of quantitative
data, a well-known older example is Pierre Bourdieu’s analysis
of social space by using correspondence analysis. Likewise,
qualitative research has a long tradition of “quantification” of
research results (Vogl, 2017), and similarly to text mining, it is
unclear, if qualitative content analysis is a “quantitative” method
or rather a “qualitative” method.
Despite these obvious overlaps and fluent borders between
“qualitative” and “quantitative” research, the oversimplified view
of two different “worlds” or “cultures” (Reichertz, 2019) of social
science research practice is upheld in methodological discourse.
Accordingly, methodological discourse has reacted increasingly
by attempting to combine these traditions via mixed methods
research since the early 1980s (Baur et al., 2017). However,
although today many differentiated suggestions exist how to best
organize a mixed methods research process (Schoonenboom and
Johnson, 2017), mixed methods research in a way consolidates
this simple distinction between “qualitative” and “quantitative”
Frontiers in Education | www.frontiersin.org 2June 2019 | Volume 4 | Article 53
Baur Linearity vs. Circularity?
research, as in all attempts of mixing methods, qualitative
and quantitative methods still seem distinct methods—which is
exactly why it is assumed that they need to be “mixed.” Moreover,
many qualitative researchers complain that current suggestions
for mixing methods ignore important principles of qualitative
research and instead enforce the quantitative research logic on
qualitative research processes, thus robbing qualitative research
of its hugest advantages and transforming it into a lacking version
of quantitative research (Baur et al., 2017, for some problems
arising when trying to take qualitative research logics seriously
in mixed methods research, see Akremi, 2017; Baur and Hering,
2017; Hense, 2017).
In this paper, I will address this criticism by focusing on
social science research design and the organization of the research
process. I will show that the distinction between “qualitative” and
“quantitative” research is oversimplified. I will do this by breaking
up the debate about “the” qualitative and “the” quantitative
research process up in two ways:
Firstly, if one looks closely, there is not “one way” of doing
qualitative or quantitative research. Instead, in both research
traditions, there are sub-schools, which are characterized by
the same degree of ignoring themselves or infighting as can be
observed between the qualitative and quantitative tradition.
– More specifically, “quantitative research” can be at least
differentiated into classical survey research (Groves et al., 2009;
Blasius and Thiessen, 2012; Baur, 2014) and big data analysis of
process-generated mass data (“Massendaten”) (Baur, 2009a).
Survey data are a good example for research-elicited data,
meaning that data are produced by researchers solely for
research purposes which is why researchers (at least in
theory) can control every step of the research process and
therefore also the types of errors that occur. In contrast,
process-produced mass data are not produced for research
purposes but are a side product of social processes (Baur,
2009a). A classic example for process-produced mass data
are public administrational data which are produced by
governments, public administrations, companies and other
organizations in order to conduct their everyday business
(Baur, 2009b; Hartmann and Lengerer, 2019; Salheiser, 2019).
For example, governments collect census data for planning
purposes; pension funds collect data on their customers
in order to assess who later has acquired which types of
claims; companies collect data on their customers in order
to send them bills etc. Digital data (Foster et al., 2017;
König et al., 2018), too, are typically side-products of social
processes and therefore count as process-produced data. For
example, each time we access the internet, log files are created
that protocol our internet activities (Schmitz and Yanenko,
2019), and in many social media, users will leave quantified
information—a typical example is Twitter communication
(Mayerl and Faas, 2019). Process-produced data can also be
analyzed by researchers. In contrast to survey data, they have
the advantage of being non-reactive, and for many research
questions (e.g., in economic sociology) they are the only
data type available (Baur, 2011). However, as they are not
produced for research purposes, researchers cannot control
the research process or types of errors that may occur during
data collection—researchers can only assess how the data
are biased before analyzing them (Baur, 2009a). Regardless
of researchers using research-elicited or process-produced
data, many quantitative researchers aim at replicating results
in order to test, if earlier research can uphold scrutiny1.
Therefore, one can distinguish between primary research (the
original study conducted by the first researcher), replication
(when a second researcher tries to produce the same results
with the same or different data) and meta-analysis (where a
researcher compares all results of various studies on a specific
topic in order to summarize findings, see Weiß and Wagner,
2019). In contrast, for secondary analysis, researchers re-use
an existing data set in order to answer a different research
question than the primary researcher asked. As can be seen
from this short overview, there are many diverging research
traditions within quantitative research, and accordingly, there
are many differences and unresolved issues between these
traditions. However, for the purpose of this paper, I will
subsume them under the term “quantitative research”, as I
have shown in Baur (2009a) that at least regarding the overall
organization of research processes, these various schools of
quantitative research largely resemble each other.
– The situation is not as simple for “qualitative research”:
Not only are there more than 50 traditions of qualitative
research (Kuckartz, 2010), but these traditions widely diverge
in their epistemological assumptions and the way they do
research. In order to be able to better discuss these differences
and commonalties, in this paper, I will focus on three
qualitative research traditions, which have been selected
for being as different as possible in the way they organize
the research process, namely “qualitative content analysis”
(Schreier, 2012; Kuckartz, 2014, 2018; Mayring, 2014; see
also Ametowobla et al., 2017, pp. 776–786), “social-science
hermeneutics” (“sozialwissenschaftliche Hermeneutik”),
which is sometimes also called “hermeneutical sociology of
knowledge” (“hermeneutische Wissenssoziologie”) (Reichertz,
2004a; Herbrik, 2018; Kurt and Herbrik, 2019; see also
Ametowobla et al., 2017, pp. 786–790) and “grounded theory”
(Corbin and Strauss, 1990; Strauss and Corbin, 1990; Clarke,
2005; Charmaz, 2006; Strübing, 2014, 2018, 2019). Please note
that within these traditions, some authors try to combine and
integrate these diverse qualitative approaches. However, in
order to be able to explore the commonalities and differences
better, I will focus on the more “pure,” i.e., original forms of
these qualitative paradigms.
Secondly, while it is not possible of speaking of “the” qualitative
and “the” quantitative research, it is neither possible of speaking of
“the” research process in the sense that there is only one question
to be asked when designing social inquiry. Instead, when it comes
to discussing the differences between qualitative and quantitative
research, at least six issues have to be discussed:
1Note that for many social phenomena, replication is not possible due to the nature
of the research object, e.g. for macro-social or fast-changing social phenomena
(Kelle, 2017) – see below for more details.
Frontiers in Education | www.frontiersin.org 3June 2019 | Volume 4 | Article 53
Baur Linearity vs. Circularity?
1. How is researchers’ perspectivity handled during the
research process?
2. How can intersubjectivity be achieved, and what does
“objectivity” mean in this context?
3. When and how is the research question focused?
4. Does the research process start deductively or inductively?
5. Is the order of the diverse research phases (sampling, data
collection, data preparation, data analysis) organized in a
linear or circular way?
6. Is data analysis itself organized in a linear or circular way?
In the following sections, I will discuss for each of these
six issues how the four research traditions (quantitative
research, qualitative content analysis, grounded theory,
social-science hermeneutics) handle them and how they
resemble and differ from each other. I will conclude the
paper by discussing what this means for the distinction
between qualitative and quantitative research as well as mixed
methods research.
HANDLING PERSPECTIVITY BY USING
SOCIAL THEORY
There are many different types of philosophies of sciences and
associated epistemologies, e.g., pragmatism (Johnson et al., 2017),
phenomenology (Meidl, 2009, pp. 51–98), critical rationalism
(Popper, 1935), critical theory (Adorno, 1962/1969/1993;
Habermas, 1981), radical constructivism (von Glasersfeld, 1994),
relationism (Kuhn, 1962), postmodernism (Lyotard, 1979/2009),
anarchism (Feyerabend, 1975), epistemological historism
(Hübner, 2002), fallibism (Lakatos, 1976) or evolutionary
epistemology (Riedl, 1985). Moreover, debates within these
different schools of thought are often rather refined and
organized in sub-schools, as Johnson et al. (2017) illustrate for
pragmatism. Regardless, current social science debates simply
crudely distinguish between “positivism” and “constructivism”.
While this is yet another oversimplification which would be
worth deconstructing, for the context of this paper it suffices to
note that this distinction is rooted in the demarcation between
the natural sciences (“Naturwissenschaften”) and humanities
(“Geisteswissenschaften”) in the nineteenth century. It has been
the occasion of several debates on the nature of (social) science
as well as the methodological and epistemological consequences
to be drawn from this definition of (social) science (e.g., Merton,
1942/1973; Smelser, 1976; for an overview see Baur et al., 2018).
In current social science debates, the “quantitative paradigm”
is often depicted as being “positivist”, while the “qualitative
paradigm” is depicted as being “constructivist” or “interpretative”
(e.g., Bryman, 1988) which has consequences on how we conceive
social science research processes.
One of the issues debated is, whether social reality can
be grasped “per se” as a fact. This so-called “positive
stance” was taken e.g., by eighteenth and nineteenth century
cameralistics and statistics who collected census and other public
administrational data in order to improve governing practices
and competition between nation states and who strongly believed
that their statistical categories were exact images of social reality
(Baur et al., 2018). This “positive stance” was also taken e.g., by
the representatives of the German School of History who claimed
that facts should speak for themselves and focused on a history of
events (“Ereignisgeschichte”) (Baur, 2005 pp. 25–56).
The criticism of these research practices of both the natural
sciences (exemplified by early statistics) and the humanities
(exemplified by historical research) goes back to the nineteenth
century. For example, early German-language sociologists such
as Max Weber criticized both traditions because they argued
that no “facts” exist that speak for themselves, as both the
original data producers of sociological or historical data and
the researchers using these data see them from a specific
perspective and subjectively (re-)interpret them. In other words:
Data are highly constructed. If researchers do not reflect this
construction process, they unconsciously (re-)produce their
own and the data producer’s worldview. As in the nineteenth
century, both statistical data and historical documents were
mostly originally produced by or for the powerful, nineteenth
century statistics and humanities unconsciously analyzed society
from their own perspective and the perspective of the powerful
(Baur, 2008, p. 192). Consequently, early historical science
served to politically legitimate historically evolved orders
(Wehler, 1980, p. 8, 44, 53–54).
These arguments are reflected in current debates, e.g., by
the debates on how social-science methodology in general and
statistics in particular are tools of power (e.g., Desrosières,
2002). They are also reflected in postmodern critiques
that every research takes place from a specific worldview
(“Standortgebundenheit der Forschung”), which is a particular
problem for social science research, as researchers are always
also part of the social realities they analyze, meaning that their
particular subjectivity may distort research. More specifically, as
academia today is dominated by white middle-class men from
the Global North, social science research is systematically in
danger of creating blind spots for other social realities (Connell,
2007; Mignolo, 2011; Shih and Lionnet, 2011)—an issue Merton
(1942/1973) had already pointed out.
At the same time, it does not make sense to dissolve social
science research in extreme “constructivism”, as this will make
it impossible to assess the validity of research and to distinguish
between solid research and “fake news” or “alternative facts”
(Baur and Knoblauch, 2018).
In other words: The distinction between “positivism” and
“constructivism” creates a dilemma between either denying the
existence of different worldviews or abolishing the standards of
good scientific practice. In order to avoid this deadlock, early
German sociologists (e.g., Max Weber) and later generations
of historians reframed this question: The problem is not, if
subjectivity influences perception (it does!), but how it frames
perception (Baur, 2005, 2008; Baur et al., 2018). In other
words, one can distinguish between different types of subjectivity,
which have different effects on the research process. In modern
historical sciences, at least three forms of subjectivity are
distinguished (Koselleck, 1977):
1. Partiality (“Parteilichkeit”): As shown above, subjectivity can
distort research because researchers are so entangled in their
Frontiers in Education | www.frontiersin.org 4June 2019 | Volume 4 | Article 53
Baur Linearity vs. Circularity?
own value system that they systematically misinterpret or even
peculate data. This kind of subjectivity has to be avoided at
all costs.
2. “Verstehen”: Subjectivity is necessary to understand the
meaning of human action (and data in general), so in this
sense, it is an important resource for social science research,
especially in social-science hermeneutics.
3. Perspectivity (“Perspektivität”): Subjectivity is also a
prerequisite for grasping reality. The first important steps
in social science research are framing a research question as
“relevant” and “interesting”, addressing this question from a
certain theoretical stance and selecting data appropriate for
answering that question.
Starting from this distinction, early German sociologists argued
that—as one cannot avoid perspectivity—it is important to reflect
it and make it explicit. And one does this by making strong use of
social theory and methodology when designing and conducting
social science research (Baur, 2008, pp. 192–193). The point
about this is that social science research still creates blind spots
(because reality can never be analyzed as a whole) but as these
blind spots are made explicit, they become debatable and can be
openly addressed in future research.
If one reframes the question, the debate between “positivism”
and “constructivism” implodes, as the comparison of the four
research traditions reviewed in this paper illustrates: Quantitative
research, qualitative content analysis, grounded theory and
social-science hermeneutics all make a strong argument that
social theory is absolutely necessary for guiding the research
process2. In order to establish how social theory and empirical
research should be linked, one first has to define what “social
theory” actually is (Kalthoff, 2008). This is important as theories
differ in their level of abstraction and at least three types of
theories can be distinguished (Lindemann, 2008; Baur, 2009c;
Baur and Ernst, 2011):
1. Social Theories (“Sozialtheorien”), such as analytical sociology,
systems theory, communicative constructivism, actor network
theory or figurational sociology, contain general concepts
about what society is, which concepts are central to analysis
(e.g., actions, interactions, communication), what the nature
of reality is, what assumptions have to be made in order to
grasp this reality and how—on this basis—theory and data can
be linked on a general level.
2. Middle-range theories (“Theorien begrenzter Reichweite”)
concentrate on a specific thematic field, a historical period and
a geographical region. They model social processes just for
this socio-historical context. For example, Esping-Andersen’s
(1990) model of welfare regimes argues that there have been
typical patterns of welfare development in Western European
and Northern American societies since about the 1880s. In
contrast, in their study “Awareness of Dying”, Glaser and
Strauss (1975), address topics of medical sociology and claim
2To clarify a common misconception of qualitative research: When qualitative
researchers demand that research should be ‘open-ended’ (‘Offenheit’), they do not
mean that they are not using theory but that they are using an inductive analytical
stance (see below).
to have identified typical patterns that are valid for the U.S. in
the 1960s and 1970s.
3. Theories of Society (“Gesellschaftstheorien”) try to characterize
complete societies by integrating results from various studies
to a larger theoretical picture, e.g., “Capitalism”, “Functionally
Differentiated Society,” “Modernity,” and “Postmodernity.”
In other words, theories of society build on middle-range
theories and further abstract them. Middle-range theories
and theories of society are closely entwined as an analysis
of social reality demands “a permanent control of empirical
studies by theory and vice versa a permanent review of
these theories via empirical results” (Elias, 1987, p. 63). For
example, in figurational sociology, the objective is to focus
and advance sociological hypotheses and syntheses of isolated
findings for the development of a “theory of the increasing
social differentiation” (Elias, 1997, p. 369), of planned and
unplanned social processes, and of integration and functional
differentiation (Baur and Ernst, 2011).
These types of theories are entwined in a very typical way during
the research process. Namely, all social science methodologies
are constructed in a way that social theory is used to build,
test and advance middle-range theories and theories of society
(Lindemann, 2008; Baur, 2009c). Therefore, social theory is a
prerequisite for social research as it helps researchers decide
which data they need and which analysis procedure is appropriate
for answering their research question (Baur, 2005, 2008). Social
theory also allows researchers to link middle-range theories and
theories of society with both methodology and research practice,
as not all theories can make use of all research methods and
data types (Baur, 2008). For example, rational choice theory
needs data on individuals’ thoughts and behavior, symbolic
interactionism needs data on interactions, i.e., what is going on
between individuals.
Due to the importance given to social theory, it is unsurprising
that all research traditions stress that the theoretical perspective
needs to be disclosed by explicating the study’s social theoretical
frame and defining central terms and terminology at the
beginning of the research process (Weil et al., 2008). The dispute
between the four methodological traditions discussed in this
paper is whether one needs to have a specific middle-range
theory in mind at the beginning of the research process or not.
In quantitative research, specifying one or more middle-range
theories in advance is necessary in order to formulate hypotheses
to be tested. The opposing point of view is that of grounded
theory which explicitly aims at developing new middle-range
theories for new research topics and therefore by nature cannot
have any middle-range theory in mind at the beginning of the
research process. Qualitative content analysis and social-science
hermeneutics are someway in between these extreme positions.
All in all, explicating one’s social theoretical stance is a major
measure against partiality, as assumptions are explicated and thus
can be criticized. All research traditions analyzed for this paper
also agree on a second measure against partiality: self-reflection.
In addition, each research tradition has developed distinct
methodologies in order to handle subjectivity and perspectivity, i.e.,
in order to avoid partiality crawling back in via the backdoor.
Frontiers in Education | www.frontiersin.org 5June 2019 | Volume 4 | Article 53
Loading more pages...