Available via license: CC BY-NC-ND 4.0
Content may be subject to copyright.
Process-Oriented Sampling
JANNIS HERGESELL
Technische Universität Berlin
NINA BAUR
Technische Universität Berlin
LILLI BRAUNISCH
Technische Universität Berlin
Abstract
Using the concepts of “duration” and “temporal patterns,” this paper
discusses how key steps during sampling change, if researchers take
temporality seriously: When defining cases, scholars have to select a
suitable temporal scale and reflect on possible changes of boundaries and
properties of cases. When defining the population or field, researchers
need to set an appropriate time frame and define periods within this
time frame to be analyzed. When selecting the actual cases for analysis,
researchers have to choose an appropriate sampling procedure, decide
upon relevant periods of analysis as well as the number of points in time
to be analyzed.
Résumé
En utilisant les concepts de “durée” et de “modèles temporels,” cet article
examine comment les étapes clés du processus d’échantillonnage
changent, quand la temporalité est prise au sérieux: en déterminant leur
cas d’étude, les chercheurs doivent choisir une échelle temporelle
appropriée et réfléchir aux changements possibles de la démarcation et
des propriétés des cas. En définissant la population ou le domaine, les
chercheurs ont besoin de fixer un cadre temporel et des intervalles de
Jannis Hergesell, Department of Sociology, Technische Universität Berlin, Fraunhoferstr. 33–36 (FH
9-1), 10587 Berlin, Germany. E-mail: jannis.hergesell@tu-berlin.de
C2020 The Authors. Canadian Review of Sociology/Revue canadienne de sociologie published by
Wiley Periodicals LLC on behalf of Canadian Sociological Association/La Société canadienne de sociologie
This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-
NoDerivs License, which permits use and distribution in any medium, provided the original work is prop-
erly cited, the use is non-commercial and no modifications or adaptations are made.
2 CRS/RCS, 00.0 2020
temps en son sein. En sélectionnant les cas, les chercheurs doivent
choisir une procédure d’échantillonnage, déterminer les périodes
d’analyse et le nombre de points dans le temps.
TIME AND SOCIAL RESEARCH
Time is one of the core categories of sociology (Baur 2005:13) since every
kind of sociality is shaped by its temporality. Without taking the historicity
of social processes into account, researchers can neither theoretically con-
ceptualize nor empirically examine other social dimensions, such as power,
knowledge, or space, as well as their entanglement. However, this does not
imply an examination of the past for its own sake (Bühl 2003:32), as is the
case in historical studies. Instead, process-oriented social research stands
in the long tradition of historical sociology that grasps current phenomena
by reconstructing their “being historically so and not otherwise” (Weber
2002:103, translated by authors). Process-oriented social research thus fo-
cuses on the temporal course of analyzed phenomena to understand the
processes “by which we have become what we are” (Schützeichel 2004:9,
translated by authors).
Many classical sociologists systematically reconstructed social pro-
cesses in order to understand the temporal course of phenomena and to un-
cover causal relationships unfolding over time. For example, Weber (2001)
explained the emergence of occidental rationalism through the historically
grown elective affinity between Protestant ethics and capitalist logics. By
reconstructing the sociogenesis of civilizing process, Elias (2000) showed
how external constraints were increasingly internalized and so formed the
contemporary habitus in Western societies. However, from the middle of
the last century, sociology increasingly moved away from its roots of (his-
torical) process-oriented research. Although there are currently some ap-
proaches that still concern themselves with processuality, these have one
thing in common with the classics: The have neglected to develop a sys-
tematic process-oriented methodology, which addresses the specific char-
acteristics of time-sensitive research designs.
The absence of a process-oriented methodology is especially evident
in process-oriented sampling, which is essential for applying a process-
oriented perspective in social research regarding current sociological ques-
tions. Thus, the potential of process-oriented social research for investigat-
ing contemporary society comprehensively and causally by considering the
social dimension of time remains unused. This holds true not only for re-
search explicitly addressing questions of contemporary social change (such
as digitalization, globalization, or biographical research) but also for other
areas of sociology not primarily concerned with the processual nature of
their phenomena (such as social inequality research, cross-cultural com-
parison, or science and technology studies).
Process-Oriented Sampling 3
This paper aims at starting to fill this gap by developing some
guidelines for process-oriented sampling. We will first discuss and crit-
icize existing state-of-the-art approaches and then introduce two key
concepts that have proven to be relevant for methodological issues
concerning process-oriented sampling: durée (duration) and temporal
patterns. We will continue to reflect on how we think the key steps
during sampling change if we take temporality seriously. In our argu-
mentation, we refer to a social-constructivist understanding of sociality
in general and to the methodology of interpretative social research in
particular. However, we limit our discussion neither to purely qualitative
or quantitative research designs nor to individual theoretical schools.
Our goal is to advance the methodological debate on process-oriented
sampling in general, which is connectable to as many sociological concepts
as possible.
SAMPLING IN PROCESS-ORIENTED RESEARCH
Sampling strongly influences if and how results can be generalized, how
data can be linked, and is an essential part of any research process
(Akremi 2019). Accordingly, almost all methodological traditions in the so-
cial sciences have extensively discussed issues of sampling. In the course of
this debate, various sampling procedures have been developed, which are
as varied as random sampling (Baur, Behnke, and Behnke 2010:139–69),
purposeful sampling (Akremi 2019; Creswell and Poth 2018:159; Marshall
1996:523; Miles, Hubermann, and Saldana 2013), theoretical sampling
(Strauss and Corbin 1990:176–94), fuzzy-sets (Ragin 2000), and selection
of single cases for case studies (Baur and Lamnek 2017; Yin 2014).
As diverse as these sampling procedures are, all traditions agree that
researchers have to reflect both on how they define cases and popula-
tions and on how they sample substantially, spatially, and temporally.
Nevertheless, process-oriented sampling has been long neglected both in
methodological debates and in process-oriented sociology (Bidart, Longo,
and Mendez 2013; Lybeck 2017; Baur 2017).
The methodological discussion on sampling typically focuses on dis-
cussing the strengths and weaknesses of various sampling procedures
(Onwuegbuzie and Leech 2007) and on developing new sampling strate-
gies. This includes issues like how many cases researchers should se-
lect and how research findings can be generalized from a small (Abbott
2004:21–23) or large number of cases (Ebbinghaus 2005). There has also
been a long-lasting debate on how different social fields and substantial
research questions influence sampling (Byrne and Ragin 2013; Creswell
and Poth 2018). Comparative and cross-cultural research (Baur 2014), as
well as approaches of historical sociology such as world system analysis
(Demetriou 2012) and comparative historical analysis (Tilly 1984), have
4 CRS/RCS, 00.0 2020
also addressed issues linked to the spatial aspects of the research question.
For instance, comparative historical analysis has extensively discussed
questions such as specifying the level and scale of analysis (Mills, van
de Bunt, and de Bruijn 2006), how the relationship between individual
phenomena’s embeddedness in and interactions with their social context
affect sampling, and how sampling issues are related to generalizability of
results and assessing causality (Goertz and Starr 2003; Kiser and Hechter
1991; Mahoney 2004). World system analysis shows how identifying the
units of analysis is relevant as the research question is a necessary precon-
dition for proper case selection (Moran 2009:116). In summary, literature
has extensively discussed how substantial and spatial aspects of the re-
search question affect sampling but has been long neglecting the temporal
aspects of sampling.
Likewise, historical sociology (Schützeichel 2004) and processual so-
ciology (Abbott 2016) have extensively discussed the nature of social pro-
cesses. However, as these debates focus on theorizing either the nature of
social processes or the theoretical implications of social processes, they
generally lack discussing methodological implications of taking tempo-
rality and processuality of social life seriously (Abbott 2001; Aljets and
Hoebel 2017; Baur 2005; Kaven 2018:2). The rare discussions on process-
oriented methodology either focus on data collection (e.g., Baur 2009a,
2009b) or data analysis (for an overview, see Baur 2005). So all in all, there
is a general lack of guiding principles for time-sensitive sampling, which
both accounts for the characteristics of processes and allows for generaliz-
ability of results (Aljets and Hoebel 2017; Baur 2017).
TEMPORALITY: ANALYSIS OF SOCIAL CHANGE
When systemizing the debate on sampling and the nature of social pro-
cesses, two key concepts have proven especially fruitful for process-
oriented methodology: durée (duration) and temporal patterns (Baur
2005:125–63). Which duration and temporal patterns are to be found and
should be investigated are methodological questions that need to be clar-
ified prior to sampling and data collection. Regardless of the researchers’
theoretical perspective and the methods they apply, a time-sensitive study
of social change always requires investigating the temporal structure of
the phenomenon in order to make further decisions in the research process
possible. If it turns out that the social change of interest extends contin-
uously over several decades, completely different sampling and data col-
lection strategies will be required than in the case of rapid social change
characterized by permanent restructuring. Dealing with the key concepts
of process-oriented research can result in the researcher’s decision to fo-
cus precisely on describing a short section of a process. It may also become
apparent that the phenomenon needs to be examined over a longer period
Process-Oriented Sampling 5
and in a more complex methodological framework with its social context
to answer the research question.
Duration
Social processes differ in the amount of time they need to unfold. They vary
in duration, “time layer” (Zeitschicht) (Koselleck 2018) or “durée” (Braudel
1958). The duration of a social process does not only influence the choice
of suitable and available data (Baur 2005:99–103, 138–42) but is also im-
portant for the way the cases and population have to be defined during
sampling. Analyzing the duration of a process includes at least four com-
ponents: overall duration, the timing of key events, the pace of change,
and the rhythm of change. In addition, research questions can address
phenomena on various time layers.
Heuristically, three types of overall duration can be distinguished
(Baur 2005:99–103, 138–42; Wehler 1972).
First, short-term social processes (temp cours) unfold in moments,
hours, or days. Knoblauch, Wetzels, and Haken (2019) provide the exam-
ple of an analysis of a short-term process by analyzing collective cheer-
ing at sports matches that emerge and end in fractions of a second. To
be able to analyze the temporal patterns of these collective emotions, the
research team uses video data, which provide very dense information for
processes unfolding rapidly. Second, medium-term processes (time of gen-
erations) cover everything that is reflected in the memory of the living
and can thus be accessed, for instance through interviews or surveys—as
a rule of thumb, people’s memories cover a maximum period of 60 to 90
years. For example, using biographical interviews, Rosenthal and Bogner
(2017) show how life stories and life courses of individuals from the Global
South are embedded into and entwined with the dynamic figuration of
larger social groups and we-groups (such as religious or political organiza-
tions or movements). Third, long-term processes (longue durée) (Braudel
1958) go beyond the memory of the living, which means that only process-
generated data can be used as a data source (Baur 2009a,b). Many so-
cial processes of theoretical interest unfold only within one or several cen-
turies or even millennia: among others, modernization, industrialization,
colonialization, state-building, or the transformation of institutions, orga-
nizations, values, and so on. For example, in The Civilizing Process, Elias
(2000) uses etiquette books, documents, and various archival sources to
show that since the Middle Ages, within Europe, social class was expressed
by a distinguished habitus, which over the centuries resulted in a slow
transformation of standards regarding violence, sexual behavior, bodily
functions, table manners, and forms of speech by increasing thresholds of
shame and repugnance (Baur and Ernst 2011). In parallel, competition be-
tween elites resulted in a slow concentration of power and the formation
of modern states.
6 CRS/RCS, 00.0 2020
Figure 1
Trajectory: Continuous Increase of Unemployment in Germany
(1990–97)
Source: Own calculation based on data by DISI and iab.
Temporal Patterns
From a methodological point of view, not only the duration of a process is
relevant but also its pattern over time. Before sampling, scholars should
reflect upon which types of patterns they are aiming at identifying, be-
cause the types of patterns of social change researchers want to grasp
influence the principles of case selection. One can roughly distinguish
three basic patterns of social change (Baur 2005:107–109, 125–37; Kaven
2015):
1. Social change may take systematic patterns or develop path-
dependently along so-called trajectories. An example of such a tem-
poral pattern is the development of unemployment numbers in
Germany in the 1990s. There was a continuous increase in unem-
ployment (see Figure 1). The temporal pattern shows a clear, unin-
terrupted development within the analyzed time frame. This spe-
cific trajectory led to various deductions in German policy debates,
such as persistent economic problems and a decreasing quality of
life, which in turn led to the retrenchment of the German welfare
state symbolized by the so-called “Hartz reforms” in 2004/2005.
Process-Oriented Sampling 7
Figure 2
Turning Point in the Development of Unemployment in Germany
in 1990
Source: Own calculation based on data by DISI and iab.
2. Social processes can also be structured in different phases, and the
transition between these phases can be characterized by abrupt
changes. These rapid and profound transformations of the defin-
ing features of the process are called turning points (Bidart,
Longo, and Mendez 2013:749). For example, the unemployment
rate in West Germany continued to decline until the mid-1990s,
and then rose steadily from 1990 onward (Figure 2). Empirically
such turning points are often caused, for example, by economic
crises, political changes (revolutions, wars), or technical innova-
tions. Possible causes for this specific turning point could include
increasing globalization or the systemic shock due to German
reunification.
3. Social processes can also be characterized by cycles or repetitions:
over time, events recur similarly. To provide an example, following
business cycles, since World War II, unemployment in Germany
has continuously fluctuated between higher and lower unemploy-
ment (see Figure 3). From this point of view, the increase in unem-
ployment in the 1990s was caused by recession, which was followed
by a decrease in unemployment during the next boom.
8 CRS/RCS, 00.0 2020
Figure 3
Cyclical Fluctuations the Development of Unemployment in
Germany (1948–2008)
Source: Own calculation based on data by DISI and iab.
Temporal patterns can be seen as heuristics linked to duration: De-
pending on the time span taken into consideration, the same interaction
pattern may appear as a trajectory, turning point, or cycle—this, in turn,
means that the temporal pattern researchers are interested in also influ-
ence the time frame of analysis. The above examples show that this choice
should not be made arbitrarily because the theoretical conclusions drawn
from results may be completely different depending on the types of pat-
terns found.
CASES AND TEMPORAL SCALE
Regardless of which research tradition researchers follow, the major
challenge when defining cases in process-oriented analysis is that cases
do not remain stable over time. Instead, they have a beginning and an
end, and their main properties can change. When regarding meso- and
macrosocial phenomena such as organizations and states, it becomes
clear that cases can change both in composition and in their boundaries,
resulting in open boundaries and fuzziness (Abbott 2001:145–60, 261–
80). In addition, cases are usually embedded into social contexts, which
include historical circumstances, spatial relations, economic, legal and
Process-Oriented Sampling 9
aesthetic circumstances, other cases, and persons involved in the case
(Stake 1994:238). Cases interact with these social contexts, and both the
relation to and interaction with the context may change over time. This
means that it is often not clear how cases are demarcated from their
contexts (Baur and Lamnek 2017:276).
Somehow, researchers have to handle these changes in the boundaries
of cases. One option is to analyze the changes themselves empirically. If
there is no time for this or this is not a study’s focus, it is of course also an
option to simply define the substantial properties and temporal and spa-
tial boundaries of the cases. However, if researchers choose the latter, they
should at least reflect if and how the definition of cases may influence or
even distort results. In the examples given in Figures 1 to 3, the empirical
case was “Germany.” Since 1945, there have been three changes of territo-
rial borders. While these border changes do not change the basic temporal
patterns discussed above, they do affect the overall level of unemployment
to the extent that policy implications would have been completely differ-
ent, if these changes of case properties had been taken into account (Baur
2014:274–83).
When thinking about the order of events and changes in cases, re-
searchers are basically thinking about “before,” “after,” and “simultane-
ously.” To be able analyze either the history within one case or between
cases, they need to set a temporal scale of comparison that allows for as-
sessing temporal patterns. Researchers usually do this by using calendars
and ordering events or case histories chronologically.
However, what researchers seldom reflect upon is that there is no such
thing as an “objective” time scale—instead, calendars themselves are so-
cially constructed (Baur 2005:121–32): Calendrical systems divide contin-
uous time (Naturzeit) into intervals, and they diverge in how long the
time intervals are, when they begin, and when they end. Besides, time
intervals can often be scaled up and down: A century can be divided into
decades, years, months, weeks, days, hours, seconds, and so on, and the
rules of transforming one scale into another vary between calendrical sys-
tems. For example, in Western societies, most people use the Gregorian
solar calendar. In contrast, the Hellenic calendars, Hebrew calendar, and
Islamic calendar are lunar calendars. In the lunar calendars, months are
shorter than in the solar calendar. Many Mesoamerican calendars used
by the Aztecs and Mayas divided the year into 260 days, while the Gre-
gorian calendar has 365 days. While the Gregorian and Islamic calendars
have 12 months, the Hebrew calendar has 12 or 13 months. The Grego-
rian calendar uses a 7-day week, while Ancient Rome used an 8-day week,
and the ancient Chinese calendar, ancient Egyptian calendar, and French
Republican calendar had a 10-day week.
As can be seen from these examples, there is no simple pattern when
comparing calendrical systems. Nevertheless, choosing an appropriate
10 CRS/RCS, 00.0 2020
Figure 4
Durée and Measurement Scale
Source: Graph created by authors.
calendar and measurement scale is methodologically important for sev-
eral reasons:
Depending on the duration of the social phenomenon of interest, a
measurement scale may be too fine-grained or too coarse (Figure 4; see also
Baur 2005:113–24, 125–37). For example, Knoblauch et al. (2019) could
not have analyzed collective cheering if they had not had data that allow
for the reconstruction of social interactions on the level of milliseconds.
They needed many measurement points over a very short period. Medium-
term phenomena, on the other hand, usually require a calendrical scale of
months or years, while long-term social processes are best measured in
decades or centuries. However, this does not necessarily mean that many
or few measurement points are per se good. Rather, this depends on the
specific research question. For example, Knoblauch et al.’s (2019) measure-
ment scale would not have done at all for Elias (2000) who was interested
in social change over centuries: Even if Elias would have been able to get
such fine-grained data for a time span of over a millennium, he would have
simply drowned in data.
When comparing documents, it can turn out that the same date is ac-
tually not the same date. For example, at first sight, Miguel de Cervantes
and William Shakespeare both seem to have died on April 23,1616. How-
ever, until the fifteenth century, both Spain and England used the Julian
calendar. Spain then introduced the Gregorian calendar in 1582, England
only in 1752. Considering this, Shakespeare actually died 11 days later
Process-Oriented Sampling 11
than Cervantes. Calendrical systems do not only vary temporally but also
between cultures and religions. In interviews or surveys, interviewees
will refer to the calendrical system they have used without explicitly
saying so. For example, in many East Asian countries, the moon calendar
is used for ordering religious festivals (such as the Chinese Ghost Festival
or Buddhist holidays) while the Gregorian calendar is used for ordering
business events, meaning that interviewees might even unconsciously
jump between calendrical systems within the same interview.
In quantitative research, unequal time spans within a given calen-
drical system may affect results. For instance, in the Gregorian calendar,
February usually has 28 days, the other months have 30 or 31 days—which
may affect the overall activities that can be completed in a month.
For many research questions, these issues can be ignored, such as
when the time frame analyzed is short and researchers conduct qualita-
tive research within one country. Should the calendrical system be rele-
vant for a research question, it can still easily be handled. For example,
in quantitative research, one could either define a finer time scale (e.g.,
days instead of months) or calculate rates (e.g., activity per day) instead of
absolute numbers (e.g., overall activity). In interviews, one can ask inter-
viewees what calendrical system they are referring to, and different cal-
endrical systems can be converted into each other using chronology (Baur
2005:121–32).
TIME FRAME AND PERIODIZATION
For process-oriented sampling, it is not only crucial to reflect on general
questions of temporal scale. Researchers should also systematically deter-
mine the concrete time frame of the analysis as well as the subdivision
of the selected time frame into smaller units. Only by doing so, can they
accurately examine how the investigated phenomenon’s processuality or
analyzed social change takes place, and thus exploit the full potential of
process-oriented social research.
Time Frame
As a general rule of thumb, an ideal study starts either when the social
process of interest begins or earlier, and it ends when the social process
ends or even later. In the examples introduced above, Knoblauch et al.
(2019) were interested in typical forms of collective cheering (i.e., a cycli-
cal pattern). This means that the data have to grasp the whole process
of cheering (which in this example is not difficult, as the whole process
unfolds within a couple of seconds). However, the actual time frame of
analysis needs to be longer as data also need to cover several of those
cheering processes because the scholars were interested in typical interac-
tions. What is typical can only be assessed if many cheering processes are
12 CRS/RCS, 00.0 2020
observed. Still, for this type of research question, it is usually enough to
collect ethnographic data over several weeks or months. In contrast, Elias
(2000) needed to cover several centuries to observe how the process of civ-
ilization unfolded. Moreover, as Elias (2000) was looking for a trajectory
and not for a repetitive pattern, it was not as easy to define the starting
and end points of the analysis.
Trajectories often do not have a fixed beginning and end. If there is
a beginning or end, it is often marked by a turning point. However, even
if there is a turning point, it is often difficult to fix it to a specific date.
For example, historical sociologists often begin their analysis with the be-
ginning of modernity. While this seems straightforward at first sight, it is
not clear at all when modernity actually begins because this strongly de-
pends on the social phenomenon under investigation: If one is interested in
military or political processes in Western Europe, modernity begins with
the discovery of America (1451–1506). In the case of religious transforma-
tion, Western European modernity begins with the Reformation, with the
work of Martin Luther (1483–1546). If, on the other hand, the history of
ideas is of interest, the beginning of Western European modernity can be
set in the Renaissance, with its pioneer Frederick II (1194–1250). In the
case of administrative and legal processes, Western European modernity
either begins with the French Revolution (1789) or the Civil Code (Code
Napoleon) (1804) (Baur 2005:82–83).
As can be seen from this example, setting the beginning of the analysis
is often difficult to assess for several reasons:
1. The example of the beginning of legal modernity shows that it is
often not even clear what the critical event is.
2. The critical event is often not a single point in time, but an extended
time period—in the above example, both Luther and Frederick II’s
lifetimes lasted more than 55 years.
3. Which type of critical event one chooses is a highly momentous
decision for data collection and analysis. For the four phenomena
mentioned above, modernity begins somewhere between 1194 and
1804. It will most definitively affect research findings if one starts
analysis 600 years earlier or later.
To decide upon the analysis’ time frame, researchers can use different
ways to identify key events.
Scholars can simply define key events based on analytical specifica-
tions: they can use their prior knowledge of the research field, the state
of literature, or theoretical considerations in order to derive key events.
However, this will most likely produce blind spots as setting key events
has not been empirically verified.
A much more elaborate method for identifying an appropriate time
frame is the so-called backward reading (Rückwärtslesen) (Hergesell
Process-Oriented Sampling 13
2019:96–99). Although more time-consuming, the advantages of backward
reading are that the definition of the time frame is empirically grounded,
customized to the specific substantial research question, and much more
open for the unknown and unexpected. The basic idea of backward read-
ing is that phenomena occurring in the present can only be understood
causally if they are reconstructed in sequence since their inception. To
capture an entire process, it must therefore be analyzed from the begin-
ning. When applying backward reading, researchers empirically search for
this beginning.
Backward reading starts with empirically specifying which character-
istics shape the phenomenon of interest in the present. Depending on the
phenomenon and the research question, this reconstruction of the phe-
nomenon’s current state can be quite extensive and time-consuming be-
cause this might require a first phase of primary research, using any
type of data and analysis procedure suitable for the research question
(Hergesell 2019:59–62). For example, Hergesell (2019) conducted ethno-
graphies to show that in geriatric care in Germany in the 2010s, there is a
conflict between economic and ethical interests. Also, in the 2010s, assis-
tive care technologies were introduced because field actors believed that
these technologies could resolve this conflict.
Next, researchers trace the development of the phenomenon back
in time. Starting with contemporary references and causes of the phe-
nomenon in the recent past, step by step, the development is empirically
traced back in history. For example, Hergesell (2019) tracks the evolve-
ment of the conflict between economic and ethical interests in geriatric
care back in time.
Backward reading is completed when a key event occurs; that is, when
empirical analysis reveals the phenomenon’s first occurrence. This is the
point in time when the social structures have emerged that are causally
related to the present phenomenon, the so-called formative period (forma-
tive Periode) (Berking and Schwenk 2011:256). For example, the forma-
tive period of the conflict Hergesell (2019) is interested in is the institu-
tionalization of geriatric care related to the approval of the German “law
on invalidity and pension insurance” (Gesetz, betreffend die Invaliditäts-
und Altersversicherung) in 1889. Accordingly, Hergesell’s (2019) analysis
of the process—that leads to the introduction of assistive care technologies
in the present—begins at this point in time. In this way, possible causal
relations between the constitution of a process, its temporal patterns, and
its effects in the present can be identified.
Periodization
Once the starting point of analysis (formative period) has been identified,
the question arises if one should really analyze the whole time frame or if
the entire time frame can be divided into shorter (sub)periods (intervals) to
14 CRS/RCS, 00.0 2020
reduce complexity and the amount of data that are needed to be collected.
When reducing complexity, researchers need to take care that they are
still able to answer the research question and identify social change by
comparing subperiods with each other (Baur 2017). The subdivision of a
whole process into individual shorter periods is called periodization.
Periodization is an essential step of process-oriented sampling and
especially important for long-term social processes because the sheer
amount of time needed for data collection and data analysis is often un-
manageable within a single research project if researchers also want to do
detailed analysis. For example, in Hergesell’s (2019) analysis of the devel-
opment of geriatric care, the time frame covers almost 130 years, so the
question he faced during periodization was: Is it really necessary to ana-
lyze all 130 years and collect data for each year, or can the whole process
be subdivided into meaningful periods? This allows analyzing fewer points
in time and instead spending more time on data analysis. In historical so-
ciology, this is especially important because data collection often involves
time-intensive archival work and archives are scattered across different
locations.
In addition, the further one goes back in the time, the more urgent
periodization develops, as it becomes increasingly more difficult to find
suitable data because social processes were either not recorded in the
first place or data were destroyed. If scholars manage to condense the
time frame of analysis to relevant periods, they might be able to substi-
tute nonfindable data. For example, in the development of geriatric care
in Germany, one relevant period was that of marginalization and lack of
resources (1933–1968) (Hergesell 2019:157–98), which covers National So-
cialism and World War II (1933–1945). During that time, many data were
destroyed. However, for Hergesell’s (2019) research question, it was not
necessary to cover the whole research period but just to collect any data
from that period. Instead of having to collect data for the NS time, he could
instead collect data for the time after 1950 for which much more data are
available.
For process-oriented sampling, periodization is an important step,
which has major consequences on sampling decisions and the results of
analysis. Periodization avoids the common mistake of “‘flattening’ the pro-
cess” and reducing “change to the initial objective or the final outcome”
(Bidart, Longo, and Mendez 2013:744). Instead, it allows for a focused and
detailed analysis of main characteristics of the process.
In time-sensitive research designs, there are various approaches to
defining periods. The most common approach in process-oriented analysis
is defining fixed periods of equal length. The problem with this approach
is that it is difficult to assess the adequate duration (Kaven 2011). If
the intervals are too short, one might not be able to effectively reduce
the amount of data collection—which is one of the main goals of peri-
odization. For example, fixing intervals at 10-year lengths will result in
Process-Oriented Sampling 15
20 measuring points in the course of 200 years. In a quantitative study,
this is not a problem (if one finds suitable data). However, if researchers
conduct social science hermeneutics, they will end up with too much data
for analysis. In contrast, if the intervals are too long, the process cannot be
properly grasped in the empirical analysis. For example, fixing intervals
at 50-year length for analyzing a 200-year time frame starting 1800 will
result in five subperiods (1800–1850, 1850–1900, 1900–1950, 1950–2000,
since 2000). Among others, this will result in possibly missing World War
I and II in the data, although both were important turning points for
many macrosociological phenomena.
To avoid this schematic setting of intervals, researchers can identify
periods by empirically analyzing social change (Hergesell 2019:96–103).
In doing so, periods are not selected arbitrarily but specified inductively,
after and through an intensive analysis of the collected data. There are
several methods of empirically identifying periods. Which one of them is
preferable depends on the researcher’s theoretical stance and the research
question.
The first empirical method for periodization is forward reading (Vor-
wärtslesen): Commencing at the formative period, scholars empirically
identify phases of relative stability and structural shifts concerning the
social process of interest (Hergesell 2019:98). For example, Hergesell
(2019:96–103) identifies four periods for geriatric care in Germany: consti-
tution and differentiation of geriatric care (1889–1933), marginalization
and lack of resources of geriatric care (1933–1968), professionalization of
caregivers (1968–2005), and innovation and technology (since 2005). As
can be depicted from this example, the periods empirically specified can be
of different lengths. For example, the fourth period is significantly shorter
than the second period. Furthermore, central phenomena can coincide in
one period. For example, both NS era and postwar era are part of the sec-
ond period because (in contrast to other social processes), for geriatric care,
the end of the NS era empirically caused no profound structural change.
The advantage of this type of periodization is that periods can be analyt-
ically separated and individually reconstructed. The immense complexity
of causal relationships lasting for decades or even centuries is thus re-
duced and empirically accessible. In addition, the periods can be compared
with one another after their reconstruction. This makes it possible to iden-
tify characteristics of a social process that are specific to a period and to
detect crucial transformations of the process during its development.
Periods can also be identified empirically based on discursive dynam-
ics. The concept of discourse is in itself a processual concept that does
not refer to purely linguistic or language aspects but to epistemes, that
is, knowledge and knowledge policies that are constituted, (re)produced,
disseminated, changed, or discarded in time (Keller 2011:17). These
knowledge dynamics are analyzed by looking at statements, actors, and
practices in which knowledge is discursively (re)produced. Against this
16 CRS/RCS, 00.0 2020
background, it is not the change of trajectories of organizations, institu-
tions, or events but the change of semantics that can be used as a basis for
periodization. For example, Braunisch, Hergesell, and Minnetian analyze
the use of the concept of “innovation” in the German parliament between
1949 and 2017 by mixing quantitative content analysis and hermeneutical
deep-reading (Braunisch, Hergesell, and Minnetian 2018; Braunisch and
Minnetian 2018). This results in four periods: In period 1 (1949–1965),
the words “innovation” and “innovative” are merely used as rhetorical
means or adjectives. In period 2 (1965–1980), “innovation” is increasingly
linked to content. In period 3 (1980–2009), the term is used much more
frequently both as fillers and as strategic topoi. In period 4 (2009–2017),
the use of “innovation” becomes reflexive and the semantic demand for
innovation is systematically institutionalized.
If researchers lack time or resources for an empirically grounded peri-
odization customized to the research question (“Gegenstandsangemessen-
heit,” Strübing et al. 2018:85), they can also periodize based on their
knowledge of the state of research.Thisispossibleasmanysocialpro-
cesses share similar typical periods. In the case of Germany, it often makes
sense to set the beginning of the analysis either at the Prussian reforms
(1807) or the March Revolution (1848/49). The next major structural frac-
ture is often 1871, which is related to the founding of the German Reich,
the kickoff of industrialization, and the constitution of the German wel-
fare state. In some cases, it is advisable to define World War I (1914–1918)
and the Weimar Republic (1918–1933) as separate periods. Often, it suf-
fices to treat the end of World War II and the beginning of decolonization
(1945) as the next fracture. The next two turning points are the 1970s
(when various political and economic crises converged, commonly referred
to as the “oil crisis”) and 1990 (collapse of the Eastern Bloc and the onset
of globalization). History will show whether 9/11 (2001), the 2009 financial
crisis, the refugee crisis (2015), and/or the Corona crisis (2020) are turning
points, too.
SAMPLING PROCEDURE
As we have discussed, process-oriented social research involves reflecting
on several methodological aspects that are important prerequisites for ac-
tual sampling. Only this ensures that temporality is systematically and
comprehensively engrained in the research design. These steps are inde-
pendent of the researchers’ theoretical stance or methodological orienta-
tion. In particular, these methodological considerations apply to both qual-
itative and quantitative approaches. The actual sampling procedure can
begin after researchers have considered duration and temporal patterns,
decided on a time frame and concluded periodization.
Even if only a short time frame was recognized as relevant for the
analysis, the number of potential sampling time points (measurement
Process-Oriented Sampling 17
Figure 5
Selecting Measurement Points within Periods of Analysis
Source: Graph created by authors.
points) within this time frame is often still too high for the analysis of
data from all periods. Therefore, researchers might have to select relevant
periods of analysis. In quantitative research, they can draw a random
sample stratified (Baur, Behnke and Behnke 2010:162–64) by periods. In
qualitative approaches, researchers typically purposefully select certain
cases (Akremi 2019; Marshall 1996:523). When doing so, they have to
make several fundamental decisions, namely at what time points they
want to collect how much and which data.
If researchers aim at analyzing periods of relative stability within
periods, they should collect data from the middle of the periods.Ifre-
searchers aim at investigating fractures, they should focus data collection
on the transitions between periods (Figure 5).
When comparing different social processes, researchers need to as-
sess, if the periodization is the same for all social processes. For example,
Baur and Hering (2017) show that the “economic crisis of the 1970s”
unfolded very differently in Frankfurt, Dortmund, Glasgow, and Birm-
ingham in terms of timing and rhythm: In Glasgow, the crisis began in
1945, in Dortmund and Frankfurt in the 1970s, and Birmingham as late
as the 1980s. In Frankfurt, the crisis had already ended in the 1970s, in
18 CRS/RCS, 00.0 2020
Dortmund, it lasted until the mid-1980s, in Glasgow until 1995, while in
Birmingham, it continues to this day.
Finally, the question needs to be addressed, which periods should be
compared? For example, Grunow (2006) compares Danish and German
globalization processes. Sampling revealed the problem that in Denmark
globalization started in the 1920s, in Germany only in the 1990s. This is a
problem for sampling insofar as all other social circumstances in Denmark
in the 1920s and Germany in the 1990s were completely different. As a
result, it is unclear whether it is better to compare Germany to Denmark
in the 1920s or 1990s.
Irrespective of which periods are deemed relevant for a specific study,
researchers also have to reflect how many measurement points they need
in order to be able to capture the properties of the process. This, in turn,
depends on the expected temporal pattern (Baur 2005:191–209): At least
two periods need to be selected to grasp trajectories. Data should be col-
lected for phases of relative stability, that is, rather in the middle of each
period. This enables researchers to compare the results across all periods
during data analysis and thus to observe long-term social change or/and
the specific characteristics of each period. This sampling strategy has the
advantage that it is not necessary to clarify exactly when the transition
phases begin and end. Also, the exact dating of the data collected is not
important as long as they originate from the stable phase.
If the analyzed process is characterized by turning points, at least four
sampling points in time are needed: Two are needed to identify the trajec-
tory before the fracture, and another two to identify the turning point af-
ter the fracture. If the temporal pattern of the process is cyclic, researchers
need to select as many time points of data collection as possible.Particu-
larly, quantitative methods such as time series analysis only work, if there
are many measurement points (Baur and Hergesell 2020).
Finally, it should be noted that these are just general guidelines. De-
pending on the specific research question, researchers might have to de-
viate from these sampling strategies. Say there has been a long phase of
stability, after which a revolution takes place when numerous events occur
in rapid succession. In such a case, it may make sense to select only a few
measurement points for the stable phase and then many measurement
points for the phase of transition.
CONCLUSION
In this paper, we have discussed what additional questions researchers
have to reflect upon when taking temporality seriously during sampling.
Based on this, we have developed a guideline for a time-sensitive sam-
pling: First, researchers have to determine the durée (duration) and tem-
poral patterns of the social process of interest. Next researchers have to
reflect, if the cases’ boundaries and properties change over time and if
Process-Oriented Sampling 19
this might affect sampling. In this course, they also have to decide upon
a calendrical system and how fine-grained or coarse the temporal scale of
analysis should be. Thinking about how cases are embedded into fields or
populations reveals that researchers have to decide upon the time frame
of their analysis, that is, the beginning and end of the analysis. During pe-
riodization, the process is subdivided into subperiods (intervals). Finally,
researchers have to think about how many measurement points they need
and on how to select relevant periods for their analysis.
Our analysis has revealed that these issues are not only relevant for
process-oriented analysis but also any kind of social research. All in all, the
systematic elaboration of the genuinely temporal dimension of social pro-
cesses is highly relevant from a methodological point of view. In this way,
it is possible to utilize the numerous theoretical preliminary works on the
temporality of social processes from various disciplines in empirical stud-
ies of social change. An adequate process-oriented sampling will improve
the validity and significance of studies in the field of process-oriented re-
search.
References
Abbott, A. 2001. Time Matters. On Theory and Method. Chicago, IL: University of Chicago Press.
Abbott, A. 2004. Methods of Discovery. New York: Norton.
Abbott, A. 2016. Processual Sociology. Chicago, IL: The University of Chicago Press.
Akremi, L. 2019. “Stichprobenziehung in der qualitativen Sozialforschung.” Pp. 265–82 in
Handbuch Methoden der empirischen Sozialforschung, edited by N. Baur and J. Blasius.
Wiesbaden: Springer Fachmedien.
Aljets, E. and T. Hoebel. 2017. “Prozessuales Erklären.” Zeitschrift für Soziologie 46(1):4–21.
Baur, N. 2005. Verlaufsmusteranalyse. Wiesbaden: VS Verlag für Sozialwissenschaften.
Baur, N., ed. 2009a. “Linking Theory and Data: Process-Generated and Longitudinal Data for
Analysing Long-Term Social Processes.” Special Issue of Historical Social Research (HSR)
34(1).
Baur, N. 2009b. “Measurement and Selection Bias in Longitudinal Data.” Historical Social
Research (HSR) 34(3):9–50.
Baur, N. 2014. “Comparing Societies and Cultures.” Historical Social Research 39(2):257–91.
Baur, N. 2017. “Process-Oriented Micro-Macro-Analysis.” Historical Social Research 42(4):43–
74.
Baur N., Behnke, J. and N. Behnke. 2010. Empirische Methoden der Politikwissenschaft.Pader-
born: Schöningh.
Baur, N. and S. Ernst. 2011. “Towards a Process-Oriented Methodology.” Sociological Review
59(777):117–39.
Baur, N. and J. Hergesell. 2020. “Time Series.” in Research Methods in the Social Sciences,
edited by J.-F. Morin, C. Olsson and E. Atikcan. Oxford: Oxford University Press.
Baur, N. and L. Hering. 2017. “Learning from the Past.” Human Figurations 6(2). Retrieved
July 2, 2018 (http://hdl.handle.net/2027/spo.11217607.0006.208).
20 CRS/RCS, 00.0 2020
Baur, N. and S. Lamnek. 2017. “Einzelfallanalyse.” Pp. 274–84 in Qualitative Medienforschung,
edited by L. Mikos and C. Wegener. Konstanz: UVK.
Berking, H. and J. Schwenk. 2011. Hafenstädte. Frankfurt am Main: Campus.
Bidart, C., M. Longo and A. Mendez. 2013. “Time and Process.” European Sociological Review
29(4):743–51.
Braudel, F. 1958. “Histoire et Sciences sociales: La longue durée.” Annales 13(4):725–53.
Braunisch, L., J. Hergesell and C. Minnetian. 2018. “Stumme Ökonomisierung.” Zeitschrift für
Diskursforschung 2:183–215.
Braunisch, L. and C. Minnetian. 2018. “‘Soziale Innovationen’ für den Fortschritt von morgen.”
Pp. 145–66 in Innovationsphänomene, edited by J. Hergesell, A. Maibaum, C. Minnetian
and A. Sept. Wiesbaden: Springer VS.
Bühl, W. 2003. Historische Soziologie – Theoreme und Methoden. Münster: LIT Verlag.
Byrne, D. and C. Ragin. 2013. The Sage Handbook of Case-Based Methods. London: Sage.
Creswell, J.W. and C. Poth. 2018. Qualitative Inquiry & Research Design. Thousand Oaks, CA:
Sage.
Demetriou, C. 2012. “Processual Comparative Sociology.” Sociological Theory 30(1):51–65.
Ebbinghaus, B. 2005. “When Less is More: Selection Problems in Large-N and Small-N Cross-
National Comparisons.” International Sociology 20(2):133–52.
Elias, N. 2000. The Civilizing Process. London: Wiley-Blackwell.
Goertz, G. and H. Starr. 2003. Necessary Conditions. New York: Rowman & Littlefield.
Grunow, D. 2006. Convergence, Persistence and Diversity in Male and Female Careers.Opladen:
Barbara Budrich.
Hergesell, J. 2019. Technische Assistenzen in der Altenpflege. Weinheim: Juventa.
Kaven, C. 2011. “Über den begrenzten Nutzen formaler Periodisierungen für die Histographie.”
ZÖSS Discussion Paper, 26. Hamburg: Universität Hamburg.
Kaven, C. 2015. “Langfristige soziale Prozesse.” Pp. 233–45 in Prozesse,editedbyR.
Schützeichel and S. Jordan. Wiesbaden: Springer.
Kaven, C. 2018. “Anmerkungen zu Andrew Abbotts ‘Processual Sociology’ aus Perspek-
tive der Nobert Elias’schen Prozess und Figurationssoziologie.” Gesis, Leibniz-Institut
für Sozialwissenschaften. Retrieved April 10, 2019 (https://nbn-resolving.org/urn:nbn:de:
0168-ssoar-57263- 7).
Keller, R. 2011. Sociological Knowledge Approach to Discourse. Wiesbaden: VS Verlag für
Sozialwissenschaften.
Kiser, E. and M. Hechter. 1991. “The Role of General Theory in Comparative-Historical Sociol-
ogy.” American Journal of Sociology 97(1):1–30.
Knoblauch, H., M. Wetzels and M. Haken. 2019. “Videography of Emotions and Affectivity in
Social Situations.” Pp. 162–79 in Analyzing Affective Societies, edited by A. Kahl. London:
Routledge.
Koselleck, R. 2018. Zeitschichten. Frankfurt a.M.: Suhrkamp.
Lybeck, E. 2017. “Comparative-Historical Sociology as Professional Practice.” Human Figura-
tions 6(2). Retrieved from (http://hdl.handle.net/2027/spo.11217607.0006.207).
Mahoney, J. 2004. “Comparative-Historical Methodology.” Annual Review of Sociology 2004:81–
101.
Marshall, M. 1996. “Sampling for Qualitative Research.” Family Practice 13(6):522–26.
Process-Oriented Sampling 21
Mills, M., G. van de Bunt and J. de Bruijn. 2006. “Comparative Research. Persisting Problems
and Promising Solutions.” International Sociology 21(5):619–31.
Miles, M., Huberman, A. and J. Saldana, eds. 2013. Qualitative Data Analysis. A Methods
Sourcebook.London:Sage.
Moran, T. 2009. “Studying Long-Term Large-Scale Change.” American Sociological Association
XV(1):115–23.
Onwuegbuzie, A. and N. Leech. 2007. “Sampling Designs in Qualitative Research.” Qualitative
Report 12(2):238–54.
Ragin, C. 2000. Fuzzy-Set Social Science. Chicago, IL: University of Chicago Press.
Rosenthal, G. and A. Bogner, eds. 2017. Biographies in the Global South. Frankfurt a.M.:
Campus.
Schützeichel, R. 2004. Historische Soziologie. Bielefeld: transcript Verlag.
Stake, R.E. 1994. “Case Studies.” Pp. 236–47 in Handbook of Qualitative Research, edited by N.
Denzin and Y. Lincoln. Thousand Oaks, CA: Sage.
Strauss, A. and J.M. Corbin. 1990. Basics of Qualitative Research. Thousand Oaks, CA: Sage.
Strübing, J., S. Hirschauer, R. Ayaß, U. Krähnke and T. Scheffer. 2018. “Gütekriterien qualita-
tiver Sozialforschung. Ein Diskussionsanstoß. ” Zeitschrift für Soziologie 47(2):83–100.
Tilly, C. 1984. Big Structures, Large Processes, Huge Comparisons. New York: Russel Sage
Foundation.
Weber, Max. 2001. The Protestant Ethic and the Spirit of Capitalism. Abingdon: Routledge.
Weber, Max. 2002. “Die „Objektivität“ sozialwissenschaftlicher und sozialpolitischer Erkennt-
nis. ” Pp. 77–149 in Weber. Schriften 1894–1922. Ausgewählt von Dirk Kaesler, edited by
Dirk Kaesler. Stuttgart: Kroner Verlag.
Wehler, H.-U. 1972. “Geschichte und Soziologie.” Pp. 42–58; 317–325 reprinted 1980 in His-
torische Sozialwissenschaft und Geschichtsschreibung, edited by H.-U. Wehler. Göttingen:
Vandenhoeck & Ruprecht.
Yin, R.K. 2014. Case Study Research. Design and Methods. Thousand Oaks, CA: Sage.
