Content uploaded by Christian Alexander Mahringer
Author content
All content in this area was uploaded by Christian Alexander Mahringer on Sep 16, 2020
Content may be subject to copyright.
Sequence Analysis in Routine Dynamics
Christian A. Mahringer (University of Stuttgart)
Brian T. Pentland (Michigan State University)
To cite this chapter:
Mahringer, C.A. & Pentland, B.T. (In Press). Sequence Analysis in Routine Dynamics.
In M. S. Feldman, B. T. Pentland, L. D’Adderio, K. Dittrich, C. Rerup and D. Seidl, eds.,
Cambridge Handbook of Routine Dynamics. Cambridge: Cambridge University Press.
2
Sequence analysis in routine dynamics
Abstract
Implicitly or explicitly, sequence analysis is at the heart of research on routine dynamics.
Sequence analysis takes many forms in many different disciplines, because sequence is
central to temporality, process, language, and narrative. In this chapter, we focus on sequence
analysis in routine dynamics research. The goal of this chapter is to help researchers use
sequence analysis in their research on routine dynamics. Hence, the chapter reviews prior
literature that has used sequence analysis, it shows how to carry out sequence analysis and it
provides implications as well as an agenda for future research.
1 Introduction
Sequence analysis can be defined as a family of methods that can be used to identify,
describe, compare and visualize patterns in sequentially ordered data. The disciplinary origins
of these methods include computer science (Sankoff and Kruskal, 1983), bioinformatics
(Durbin et al., 1998), history (Griffin, 2007), life course sociology (Aisenbrey and Fasang,
2010) career research (Abbott and Hrycak, 1990), research on decision making (Levitt and
Nass, 1989) and innovation research (Van de Ven et al., 1999). Sequence is essential to
concepts like progression, temporality, and flow that are central to process organization
studies (Langley and Tsoukas, 2017) and routine dynamics (Feldman et al., 2016).
In this chapter, we focus on sequence analysis as it applies to routine dynamics. The
goal of this chapter is to help scholars use sequence analysis in their research on routine
dynamics. We begin by considering the kinds of questions we can address with sequence
analysis. We review prior routine dynamics research and show how it has used sequence
analysis. Excellent resources are available for the mechanics of particular methods (e.g.,
3
Sankoff and Kruskal, 1983, Cornwell, 2015, Poole et al., 2017). Hence, rather than zooming
in on particular sequence analysis methods (e.g., optimal string matching), we are zooming
out to consider how sequence analysis can help identify, describe, visualize and compare
routines and their dynamics. We show how scholars can get started with sequence analysis
with any kind of sequential data (e.g., from ethnographic observation, interviews or digitized
event logs). Finally, we suggest avenues for future research.
To illustrate our arguments, we draw on the example of Scrum software development
routines in a medium-sized high-tech manufacturing company (Mahringer, 2019, Mahringer,
Dittrich and Renzl, 2019). Scrum is a software development framework which splits the
software development process into phases of two to four weeks (i.e., sprints) (Schwaber and
Beedle, 2002). The study includes ethnographic fieldwork of how the software development
teams enacted the Scrum routines over a period of 12 months. The software development
teams organized their work by using a software tool called Zoe (all names are pseudonyms).
that recorded approximately 4.500 sequences and 90.000 events in a database. Sequence
analysis can be used with any or all of this data.
2 How does sequence analysis help to understand routine dynamics?
Abbott (1990: 375) identifies three kinds of questions where sequence analysis can be
useful: “(1) questions about whether a typical sequence or sequences exist, (2) questions
about why such patterns might exist, and (3) questions about the consequences of such
patterns.” Of these, he argues that the first question is most important. To the extent than an
organizational routine contains recognizable patterns of action, we expect to find typical
sequences in any routine. Sequence analysis can help us identify, describe, visualize and
compare those sequences.
4
Abbott’s (1990) questions are generic to any kind of sequential data, but there are
more specific questions that are relevant to the analysis of organizational routines. By
definition, routines are repetitive, so they generate multiple performances (Feldman and
Pentland, 2003). To the extent that each performance of an organizational routine can be
treated as a sequence of actions, we can ask the following kinds of questions (Salvato, 2009b):
• What are the typical patterns of a routine? Because routines are patterns in variety
(Cohen, 2007) they can potentially generate a large number of different performances.
Some of these performances occur more often and, hence, are more typical while other
performances are less typical. Abbott (1990: 378) refers to these as “typical-
sequence/families-of-sequences” questions; he argues that these are “the central
questions of the sequence area.”
• How varied are the performances of a routine? While the performances of some
routines are more similar to each other, the performances of other routines differ
tremendously. Pentland (2003a) offers metrics for measuring sequential variety.
• How does sequence matter? The sequential relations among actions of a routine
performances are essential (LeBaron et al., 2016). Sequence analysis can be used to
unpack sequential relations between actions.
• How does the pattern of a routine change? Sequence analysis can also be used to
show how the pattern of a routine changes over time. Dittrich, Guérard and Seidl
(2016), for instance, analyze how the routine pattern changed, and identify reflective
talk as a critical mechanism of routine change.
• How do different action patterns influence performance outcomes? Sequence
analysis provides opportunities to better understand how different patterns influence
performance outcomes. For example, first writing an exam and then learning the
relevant content most likely results in a different performance outcome than the other
way round.
5
While Abbott’s (1990) primer provides a useful starting point, it has some important
limitations, especially when applied to routine dynamics. First, Abbott treats events as
objective, which undermines that routine participants might interpret events in different ways.
However, the significance of events for participants is central to the formation and dynamics
of the routine (Sele and Grand, 2016). Therefore, sequence analysis of routines should include
the notion of meaning and interpretation. An event, then, can be defined as an actual
happening that sufficiently coheres to be experienced as similar, but which still incorporates
different points of view (Hernes, 2014).
Second, Abbott (1990) treats patterns as stable or stationary. While Abbott's own
research places history and temporality in the foreground (e.g., in the formation of
professions), the methodologies he discusses in his primer are a-historical. They focus on
sequences of events, but not on how these sequences might change over time. In contrast,
research on routine dynamics is explicitly concerned with change and temporality (Pentland et
al., 2012).
Third, focusing particular sequences tends to obscure the significance of multiplicity
in routines. Routines are generative systems that can display a substantial number of different
sequential patterns. Like other processual phenomena, routines are multiplicities (Pentland et
al., In Press). Hence, when comparing routines or measuring change over time, we may need
to compare whole action patterns, not just particular linear sequences.
3 Sequence analysis in prior routine dynamics research
Sequence methods in routine dynamics research can be sorted into three different
categories: whole sequence methods, pattern-mining methods, and network methods. These
methods can be differentiated according to the length of sequence that is considered in the
analysis (Table 1).
6
Table 1: Three types of sequence analysis in routine dynamics
Whole sequence
methods
Pattern mining
methods
Network methods
Sequence length
Variable, up to length
of longest
performance
Variable, typically
three to five actions
or events
Fixed, pairs of events
Typical
applications
Identifying different
types of patterns
Identifying a typical
pattern of actions of
a routine
Identifying handoffs
between actions
Major
drawbacks
Only considers
differences between
whole sequences, not
within sequences
Size of the lexicon
has a critical effect
on the findings;
limited applicability
in comparing
patterns
Only considers the
immediate context
(i.e., one action
before and one after)
Exemplary
references
Salvato (2009a,
2009b)
Pentland and Rueter
(1994), Hansson,
Pentland and Hærem
(2017)
Pentland, Hærem and
Hillison (2010),
Goh and Pentland
(2019)
3.1 Whole sequence methods
As the name implies, whole sequence methods operate on complete sequences of
action (Salvato, 2009b). These methods build on the rationale that differences between
empirically observable sequences yield meaningful insights. These methods treat whole
performances of a routine as the unit of analysis. They derive from molecular biology and
computer science (Sankoff and Kruskal, 1983). Abbott and Hrycak (1990) pioneered the use
of these methods in career research.
Salvato (2009a), for instance, analyzes new product development processes at Alessi
over a period of 15 years. The author uses dossiers that report details about 90 new product
development projects to identify the sequences of events for each project. He applies optimal
matching (Abbott and Tsay, 2000) to identify clusters of similar sequences. To interpret the
meaning of these clusters the author went back to his raw data or asked participants. A similar
approach is applied by Sabherwal and Robey (1993). These authors use optimal matching and
cluster analysis to develop a taxonomy of implementation processes. Analyzing 53 computer-
based information system implementation processes they identify six archetypes of these
7
processes. Pentland (2003a) also uses whole sequence methods as one way to characterize
variety in routines.
It is important to note that narrative analysis, based on ethnographic fieldwork, can
also be considered a whole sequence method (Pentland, 1999a). When ethnographers describe
the typical performance of a routine, from start to finish, they are engaging in sequence
analysis. Constructing a narrative from fieldnotes requires the same basic steps as a more
formal, mathematical analysis: collecting the data, defining the lexicon, choosing a point of
view, identifying the sequence and creating a representation.
3.2 Pattern mining methods
In contrast to whole sequence methods, pattern mining methods seek to identify
common subsequences within performances of a routine. There is a broad class of algorithms
and techniques for empirically identifying patterns (e.g., Mabroukeh and Ezeife, 2010,
Fournier-Viger et al., 2014).
Hansson et al. (2017) investigate the application of these methods to organizational
routines. They examine the use of regular expressions and inductive pattern mining. Regular
expressions are a pattern matching tool that is available in nearly every modern computing
language. Regular expressions provide a flexible tool for searching a corpus of sequence data
(typically in the form of text) for particular combinations of letters and words. Hansson et al.
(2017) refer to regular expressions as a deductive method because the search pattern must be
defined in advance. In contrast, inductive pattern mining methods are algorithms that search
through a corpus of text to find patterns that do not need to be defined in advance.
Keegan, Lev and Arazy (2016) analyze editorial events in Wikipedia articles. The
authors use pattern mining to identify the most frequent sub-sequences of how articles are
edited. The authors are interested in different contribution styles to Wikipedia articles, such as
8
solo contributing or reactive contributing. They emphasize the opportunities that sequence
analysis offers to better understand routines in online knowledge collaboration.
Pentland and Rueter (1994) apply a simple pattern mining approach to identify
grammatical rules that could be used to describe organizational routines. The authors use a
sample of 335 calls from a software support hotline. They define grammars (i.e., patterns) for
the data set based on observations of the routine. Subsequently, they use the grammatical
patterns to rewrite (i.e., substitute) the events in the actual sequences. This analysis led to the
insight that a large number of performances could be described by a small number of patterns.
3.3 Network methods
In contrast, network methods operate on adjacent pairs of actions or events within a
sequence or set of sequences. Because they do not require whole sequences, network methods
can be used where fragments of performances are observed, as often occurs in fieldwork
(Pentland and Feldman, 2007). Network methods provide a convenient way to describe what
Cohen (2007) referred to as the “pattern-in-variety” of organizational routines. There are
repetitive, recognizable patterns, but there may also be a large number of variations.
Pentland and Feldman (2007) propose that sequences of events can be used to
compute narrative networks, a special class of network in which nodes represent events and
edges represent sequential relations between those events. “A narrative network is an
analytical device for describing, visualizing, and comparing these patterns.” (Pentland and
Feldman, 2007: 782). This method has been used to visualize routine patterns from
observational data (Danner-Schröder and Geiger, 2016, Dittrich et al., 2016), but it also offers
the possibility to compute event networks from digitized digital trace data.
Pentland et al. (2010) analyze 4781 invoice processing sequences in four Norwegian
organizations. They aggregate those performances into an event network that represents the
routines in each organization, and they compare those networks to determine whether the
9
routines in these organizations are different. Note that this method is not comparing specific
performances of the routine. Rather, it is comparing the relative frequency of sequential pairs
of action in all observed performances of the routine. Pentland et al. (2011) extend these
insights by showing that patterns change over time due to endogenous factors.
In summary, the network approach can be used to better understand the variability of
routine patterns (Pentland, 2003b), complexity (Hærem, Pentland and Miller, 2015, Hansson,
Hærem and Jeong, In Press) and multiplicity of routines (Pentland et al., In Press).
4 A guide to sequence analysis in routine dynamics research
In this section we step-by-step show how to analyze sequential data based on the
example of Scrum software development. As with any routine, in practice these steps may not
follow a fixed, linear sequence. Rather, it may be necessary to jump back and forth.
4.1 Collecting the data
As with any empirical work, sequence analysis starts with collecting data. The
predominant empirical approach to understand routine dynamics is ethnographic fieldwork
(Feldman et al., 2016). It should be evident from the prior review that sequence analysis can
be applied to many different kinds of data, including observational data collected during
fieldwork. Indeed, any source of data that includes temporal sequence can be used for
sequence analysis. A major strength of ethnographic fieldwork is that it enables scholars to
capture the mundane everyday actions and the meaning that actors associate with specific
events (Dittrich, In Press). However, a drawback of ethnographic fieldwork is that it is limited
to specific times and places.
Digital trace data is gaining popularity (Berente, Seidel and Safadi, 2019). It offers
possibilities for extending ethnographic data in two important ways. First, digital trace data
can extend the temporal scope, because these data oftentimes extend across several years or
10
decades. This enables seeing patterns of stability and change over longer periods of time. For
example, while the first-author of this chapter spent twelve months observing the Scrum
routines, the digital trace data covers a period of approximately four years. Taking earlier
periods into consideration shows that the actors used different functions in Zoe than they used
during the observation period.
Second, analysis of digital trace data can extend the spatial scope. Ethnographic
fieldwork oftentimes focuses on local settings (Marcus, 1995). Digital trace data, however,
particularly when provided by digital software tools that are used by actors in different
locations, enable scholars to analyze sequences that extend across many locales. In the Scrum
study, for instance, the Product Owner (i.e., the actor who was responsible for the software to
be developed) typically spent some time in his private office space in the early morning to
check the product backlog (i.e., a list of issues to be resolved). During this time he also
clarified issues in Zoe and communicated with customers. These events are tracked in Zoe
while they had not been directly observed.
By extending the spatial and temporal scope of research, digital trace data provides
researchers with methods that can identify otherwise hidden patterns and dynamics.
Computing a network of events from the Scrum sequences showed how events typically
connected to each other. This network showed that the event ‘PrioritizeIssue’ (i.e., an event
that signifies changes in the order of issues in the product backlog) took a central position in
the network and connected with many other events. This made us reflect the relevance of
prioritizing issues.
4.2 Selecting software tools
Data in hand, the next step is to determine if any kind of software is needed to assist in
the analysis. With a small amount of data, it is perfectly possible to identify, describe,
visualize and compare patterns by hand (Barley, 1986, Pentland, 1999b). With larger amounts
11
of data, and for specialized questions, it may be necessary to find a software tool that helps to
analyze sequential data, such as TraMineR or ThreadNet.
TraMineR is a software package for R. After having installed R the TraMineR
package can be downloaded. TraMineR offers many different methods, including whole
sequence analysis, pattern mining and network models. Gabadinho et al. (2011) offer a
detailed user guide that explains the methods available.
ThreadNet is also a software package in R, available on GitHub
(https://github.com/ThreadNet/ThreadNet). As the name implies, it converts threads
(sequential data) into networks based on sequentially adjacent pairs of events. ThreadNet
allows users to define events in a flexible manner, based on any combination of contextual
factors. This allows users to quickly explore action patterns from different points of view
(e.g., the actor, the location, etc.). ThreadNet itself is limited to visualization, but it can export
network structures for analysis in TraMineR and other software packages. In analyzing the
Scrum data, for example, we started with ThreadNet and later extended to TraMineR when
more specific functions were required.
We suggest the use of specialized tools rather than general qualitative analysis tools,
like nVivio or Atlas/ti because sequence analysis poses some unique challenges. We are not
just trying to sift and sort categories; we are looking for patterns of sequential relations
between categories. The number of possible relations grows exponentially (as the square of
the size of the lexicon). As a result, sequential relations can be difficult to keep track of
without some kind of specialized, computerized help.
4.3 Identifying the limitations of your data
All kinds of data have limitations. These limitations shape which kinds of questions
can be answered and where additional inquiry is required. As we have discussed before,
ethnographic data is limited in its temporal and spatial scope. Because ethnographic fieldwork
12
requires the researcher to observe a setting in detail, the data typically covers a period of
several months, and sometimes few years, but rarely extends to a longer time horizon such as
decades. The degree of detail of ethnographic fieldwork also requires researchers to make
choices on what to observe and what not to observe (Van der Waal, 2009). Hereby, it is
necessarily limited to a specific setting.
By contrast, digital trace data also face several limitations. First, trace data are limited
to events that are captured ‘on-line’, as part of the digital environment. Hence, they do not
capture events that happened ‘offline’. In the Scrum case, for instance, the daily standup was
a routine which the developers enacted to synchronize their work. Even though this routine
was an important part of software development, the developers did not use Zoe when
performing it. Hence, the digital trace data did not contain information about the daily standup
routine.
Second, trace data may not capture the differences in meaning associated with events.
For instance, the same event in the digital trace data can have different meanings depending
on the situation in which it is performed. A major feature of Zoe was the product backlog,
which was an extensive list of issues to be resolved. When actors dragged an issue from the
product backlog and dropped it to another position this resulted in an event which we called
‘PrioritizeIssue’. The actors oftentimes coincidentally dragged and dropped and issue to
another position when they were discussing about the product backlog. In other cases, by
contrast, actors intentionally dragged issues from the bottom of the product backlog to its top.
Since issues at the top of the product backlog had the highest priority and were added to the
next sprint this was a significant event. Zoe, however, did not allow us to account for such
differences. Above we noted that this event took a central position in the event network. We
considered the possibility that coincidentally moving an issue could be an explanation for the
central position of the event in the network. However, depending on the data it might also be
13
possible to take account of such differences by considering additional data sources (e.g.,
fieldnotes).
Third, trace data may be limited in capturing the duration of events. The start of an
event may be recorded with a specific time stamp, but the duration might not be recorded.
This might be challenging if researchers try to interpret time lags between events, assuming
that events do not have a specific duration. In the Scrum case, each issue in Zoe contained a
description field. When an actor pushed the ‘safe’ button Zoe created an event which included
a specific time stamp. Whereas sometimes a description change was minor (e.g., correcting a
spelling mistake) in other cases such an event could signify an extensive discussion about a
complex issue. The duration of this discussion, however, was not captured in the event log
data.
In summary, it is essential to identify the limitations of the data set. Ethnographic data
might be limited to particular times and places. Digital trace data is limited to what happens
on-line, might not capture meanings associated with events and is limited in capturing
durations of events. Sometimes it may be possible to gather additional data that helps to
resolve such limitations, but often it is not. This limits the kinds of questions that can be
answered. Most likely, the list of limitations is continuously revised during the course of
analysis, as new limitations are discovered and old limitations are resolved.
4.4 Defining the lexicon of events
Another critical step in sequential analysis is defining the lexicon of events (Berente et
al., 2019). The lexicon is the set of event types that are used to depict the sequences. The key
point is that there does not need to be a one-to-one correspondence between the raw data
collected and the lexicon that is analyzed. For example, some items in the raw data may be
‘filler’. In the Scrum case, for instance, simultaneously adding multiple attachments to an
issue in Zoe produced sequences of similar events in the event log data. Moreover, several
14
different items in the raw data may be used as indicators of the same higher order category
(Gioia, Corley and Hamilton, 2013). In general, the move from raw data to the lexicon of
events is an essential part of making sense of your data (Abbott, 1990, Pentland and Liu,
2018).
The move from raw data to higher order constructs affects the granularity of the data.
Selecting the granularity of events is a major challenge, because granularity can have a
tremendous impact on the findings (Pentland, 2003a). Selecting a finer granularity (i.e., a
larger lexicon) increases differences between sequences and makes it more difficult to
identify patterns. Selecting a coarser granularity (i.e., a smaller lexicon) makes the sequences
more similar, but might lead to the false assumption that there is only minor variation in
routine performances. Hence, it is important to define granularity based on one's
understanding of the setting and the phenomenon of interest. In the Scrum case actors used
different functions in Zoe to indicate interdependencies. This resulted in different events in
the raw data such as ‘component’, ‘link’ and ‘epic link’. Since all of these events were used to
indicate interdependencies, we aggregated them into the event ‘AddInterdependency’.
4.5 Defining sequences
The next step is to define according to which rationale events are sequenced. Similarly
to ethnographers who have to make choices on ‘what to follow’ (Marcus, 1995), digital trace
data may provide several ways of defining sequences, which could yield different insights
into the phenomenon. For example, one could follow the Product Owner on a regular work
day. From that point of view, one would see how the Product Owner interacts with customers,
the developers, and Zoe. Alternatively, one could follow an issue in Zoe (i.e., a bug to be
resolved or a new feature to be developed in the software). From that point of view, one
would see how a customer reports the issue in Zoe, how the Product Owner specifies the
description of the issue and how the developers resolve it. Either point of view presents a
15
partial view of the overall routine, which is why Feldman and Pentland (2003) emphasize that
the ostensive and performative aspects of organizational routines are multiple.
In general, it is worthwhile to think about different ways of sequencing the data and
which insights this could yield. In the Scrum case, we explored different ways of sequencing
the data such as actors, weekdays and issues. Issues were promising because they described
the sequences of events that were performed in order to implement new features in the
software or resolve bugs.
4.6 Identifying, describing, comparing and visualizing patterns
Now that your data are ready, we can apply sequence methods. Three major questions
are important for research on routine dynamics.
Identifying: is there a sequential pattern? Whole sequence methods are useful to
identify different types of sequences, but they cannot be used to identify common patterns of
events across these sequences. Pattern mining is more useful to identify such patterns across
sequences. The major challenge here is that the size of the lexicon influences the findings,
because a larger lexicon makes the sequences more different. Hence, this approach has to
trade-off pattern length and generalizability of patterns. The network approach overcomes this
issue because it does not rely on pattern length, but on handoffs between events.
Describing and visualizing: what is the pattern? The most common method for
describing sequential patterns in routine dynamics research is via narrative (i.e., texts, stories).
Sequences of action have a natural narrative structure, and different characters or roles can
enter and exit the story as needed. However, narrative tends to portray routines as having a
specific, linear structure. It is difficult to capture the pattern-in-variety (Cohen, 2007) in
narratives. Of course, one can describe exceptions and variations, but this quickly becomes
tedious if the routine has a large number of variations.
16
Visualization is another common strategy for describing routines. However, as
Feldman (2016) notes, most published visualizations are abstract simplifications. They may
be easy to grasp, but they convey less information than a linear textual description.
Visualizations based on detailed empirical data are starting to become available through
software tools like TraMineR and ThreadNet.
Comparing: how do patterns differ? Identifying a pattern of events is useful to gain
an understanding of the routine, but it does not yield further insights. Comparing patterns
across contexts can yield further insights into what influences these patterns. We could, for
instance, ask whether the Scrum routines show more or less regularities in more or less
institutionalized contexts. We could also look at whether differences in complexity and
multiplicity (e.g., more or less complex software, more or less actors involved) shape the
patterns of events.
More specifically, patterns can also be compared for different time periods. Because
we are interested in change over time, routine dynamics creates an additional requirement for
conventional sequence analysis. Where Abbott (1990) emphasized synchronic methods,
routine dynamics suggests the need for diachronic analysis (De Saussure, 1916). Diachronic
analysis not only considers a pattern at a specific point in time, but takes its development over
time into consideration (Barley, 1990, Berente et al., 2019).
All three methodological approaches could be used in the context of diachronic
analysis. The whole sequence approach identifies differences between entire sequences.
Comparing sequences across different time windows could help to understand that the
sequences are changing over time. What is missing here is how the pattern changed. The
network approach can be used to compare patterns for different time windows. Pattern mining
methods face similar challenges, but could be suitable to understand whether routines become
less or more patterned over time. The question whether patterns are changing over time
requires iterating between synchronic and diachronic approaches (Berente et al., 2019).
17
4.7 Interpreting
Identifying, describing, comparing and visualizing patterns of action provides us with
either numerical or visual results. However, we need to interpret these visualizations and
numbers (Keegan et al., 2016). Interpreting results shows that we need an in-depth
understanding of the context that we are analyzing. We need to tell a story about the patterns.
Even though this chapter presents interpreting as a discrete step in the analysis, we rather see
it as a process that continuously unfolds during the analysis.
5 Implications and agenda for future research
Clearly, sequence analysis has helped advance our understanding of routine dynamics
and will continue to do so in the future. In the final section of this chapter, we offer some
ideas for future research.
5.1 Mutually contextualizing visualizations and narratives
The most common way to describe routines is through narrative (Feldman et al.,
2016). Well written textual descriptions can be very detailed and compelling. Narrating is
particularly valuable, because it strives to convey the researcher’s experience of local
meanings in the field (Yanow, 2012). However, narratives are limited as a way to describe
processual phenomena (Mesle and Dibben, 2017), because it is difficult to portray variety,
and the linear quality of narrative tends to lead to an understanding of routines as unitary
sequences of action.
As our capability to analyze and visualize sequential data improves, we are beginning
to have visualizations (and metrics) of action patterns, as well. Network methods, for instance,
provide a particularly promising source of visualizations (Moody, McFarland and Bender-
deMoll, 2005). A strength of visualizing lies in depicting multiplicity. Moreover, visualizing
can help to structure and process complex data, which might reveal patterns that we did not
18
see before. However, visualizations are impossible to interpret without some form of narrative
explanation.
As shown in Figure 1, sequence analysis can provide narratives and visualizations (or
metrics) of routines, both of which should contextualize each other. Many kinds of data can
be used to create visualizations (or metrics) and narratives. These outputs mutually
contextualize each other: visualizations add a sense of structure to narratives, and narratives
help interpreting visualizations. Sequence analysis informs both sides of this equation. It is the
foundation for both the visualizations and the narratives.
In a sense, we are specifying Feldman et al’s (2016: 511) statement that
“[e]thnographic fieldwork will always be needed to interpret archival results, but digitized
trace data provide a way to visualize and compare patterns of action that have not previously
been available.” While Abbott (1992: 430) argues that “[t]here is nothing about thinking
processually that requires interpretive attention to complexity of meaning,” we argue that in
routine dynamics, the opposite is usually true. Thus, we encourage future research on routines
that embraces and integrates both approaches, since they are not mutually exclusive, but
mutually contextualizing.
Figure 1: Mutually contextualizing visualizations and narratives generated by
sequence analysis
19
5.2 Extending the spatial and temporal scope
The rise of digital trace data provides new opportunities to further extend the spatial
scope of research on routine dynamics. Dispersed settings such as platform collaboration and
open source software development make it difficult to understand how people coordinate their
work through ethnographic fieldwork (Marcus, 1995). Oftentimes, however, these new ways
of working are supported by software tools that provide rich data. Lindberg et al. (2016) is an
example of a study that has taken advantage of these kinds of data to better understand
routines. Because these digitized contexts become more and more important we call for more
empirical research in these contexts.
Moreover, we encourage routine dynamics scholars to extend the temporal scope of
their analysis. Even though ethnographic fieldwork typically studies a considerable amount of
time, sequence analysis also provides opportunities that show changes and patterns over
several years or even decades (Salvato, 2009a). Hence, it might be fruitful to both zoom into
the details of everyday work, but also zoom out on longer time horizons to better understand
routines.
5.3 Dynamics implies diachronic analysis
Ferdinand De Saussure (1916) introduced the distinction between synchronic and
diachronic analysis in linguistics. Synchronic analysis refers to studies of language structure
or comparative language structure within a specific period of time. In contrast, diachronic
analysis refers to changes in a language over time (De Saussure, 1916). Diachronic analysis
attempts to describe and understand changes over time. Barley (1986, 1990) translated these
concepts for use in organizational research.
Diachronic analysis is an obvious fit for routine dynamics because it provides a way to
conceptualize change in a complex system of sequential relationships over time. Pentland et
al. (2019) offer a methodology for applying diachronic analysis to organizational routines
20
using sequence data. As an example of diachronic analysis based on fieldwork, consider
Barley’s (1986) classic study of the introduction of new technology in the radiology
departments of two hospitals. As a participant observer, Barley recorded the sequential
interactions of radiologists, nurses and technicians, over a one year period, pre- and post-
implementation. Using this data, Barley was able to conduct a diachronic analysis of the roles
and action patterns (see also Barley, 1990).
5.4 Moving from singularity to multiplicity
Sequence methods make it easy to measure similarity between sequences. However,
routines are multiplicities, not singular sequences (Pentland et al., In Press). This ontological
claim has methodological implications because we require approaches that operationalize
multiplicity. Goh and Pentland (2019), for instance, introduce the notion of paths that could
be used as an indicator of multiplicity of routines. More research is required to better
understand multiplicity in and of routines.
5.5 Adopting methodological innovation
Business process management scholars have developed, and continue to develop, tools
for analyzing sequential data (Wurm et al., In Press). Research on routine dynamics research
has just started to recognize the possibilities of adopting these methods. These include
methods for analyzing drift and variants, among other things. Research on machine learning is
also providing a variety of tools for sequence analysis (Witten et al., 2016). There is a great
deal of uncharted terrain that waits to be discovered and we hope that routine dynamic
scholars will continue exploring.
21
References
Abbott, A. (1990). A primer on sequence methods. Organization Science, 1(4), 375-92.
Abbott, A. (1992). From causes to events: Notes on narrative positivism. Sociological
Methods & Research, 20(4), 428-55.
Abbott, A. & Hrycak, A. (1990). Measuring resemblance in sequence data: An optimal
matching analysis of musicians' careers. American Journal of Sociology, 96(1), 144-85.
Abbott, A. & Tsay, A. (2000). Sequence analysis and optimal matching methods in sociology:
Review and prospect. Sociological Methods & Research, 29(1), 3-33.
Aisenbrey, S. & Fasang, A. E. (2010). New life for old ideas: The "second wave" of sequence
analysis bringing the "course" back into the life course. Sociological Methods &
Research, 38(3), 420-62.
Barley, S. R. (1986). Technology as an occasion for structuring: Evidence from observations
of CT scanners and the social order of radiology departments. Administrative Science
Quarterly, 31(1), 78-108.
Barley, S. R. (1990). Images of imaging: Notes on doing longitudinal field work.
Organization Science, 1(3), 220-47.
Berente, N., Seidel, S. & Safadi, H. (2019). Research commentary — data-driven
computationally intensive theory development. Information Systems Research, 30(1),
50-64.
Cohen, M. D. (2007). Reading Dewey: Reflections on the study of routine. Organization
Studies, 28(5), 773-86.
Cornwell, B. (2015). Social Sequence Analysis: Methods and Applications, New York:
Cambridge University Press.
Danner-Schröder, A. & Geiger, D. (2016). Unravelling the motor of patterning work: Toward
an understanding of the microlevel dynamics of standardization and flexibility.
Organization Science, 27(3), 633-58.
De Saussure, F. (1916). Cours de Linguistique Générale, Paris: Payot.
Dittrich, K. (In Press). Ethnography in Routine Dynamics. In M. S. Feldman, B. T. Pentland,
L. D’Adderio, K. Dittrich, C. Rerup and D. Seidl, eds., Cambridge Handbook of
Routine Dynamics. Cambridge: Cambridge University Press.
Dittrich, K., Guérard, S. & Seidl, D. (2016). Talking about routines: The role of reflective talk
in routine change. Organization Science, 27(3), 678-97.
Durbin, R., Eddy, S. R., Krogh, A. & Mitchison, G. (1998). Biological Sequence Analysis:
Probabilistic Models of Proteins and Nucleic Acids, Cambridge: Cambridge University
Press.
Feldman, M. S. (2016). Routines as process. In J. Howard-Grenville, C. Rerup, A. Langley &
H. Tsoukas, eds., Organizational Routines. How They Are Created, Maintained, and
Changed. Oxford: Oxford University Press, pp.23-46.
Feldman, M. S. & Pentland, B. T. (2003). Reconceptualizing organizational routines as a
source of flexibility and change. Administrative Science Quarterly, 48(1), 94-118.
22
Feldman, M. S., Pentland, B. T., D’Adderio, L. & Lazaric, N. (2016). Beyond routines as
things: Introduction to the special issue on routine dynamics. Organization Science,
27(3), 505-13.
Fournier-Viger, P., Gomariz, A., Gueniche, T., Soltani, A., Wu, C.-W. & Tseng, V. S. (2014).
SPMF: A java open-source pattern mining library. The Journal of Machine Learning
Research, 15(1), 3389-93.
Gabadinho, A., Ritschard, G., Studer, M., & Müller, N. S. (2011). Mining Sequence Data in R
with the TraMineR Package: A User’s Guide. Geneva: University of Geneva.
Gioia, D. A., Corley, K. G. & Hamilton, A. L. (2013). Seeking qualitative rigor in inductive
research: Notes on the Gioia methodology. Organizational Research Methods, 16(1),
15-31.
Goh, K. & Pentland, B. T. (2019). From actions to paths to patterning: Toward a dynamic
theory of patterning in routines. Academy of Management Journal, 62(6), 1901-29.
Griffin, L. J. (2007). Historical sociology, narrative and event-structure analysis: Fifteen years
later. Sociologica, 1(3).
Hærem, T., Pentland, B. T. & Miller, K. D. (2015). Task complexity: Extending a core
concept. Academy of Management Review, 40(3), 446-60.
Hansson, M., Hærem, T. & Jeong, J. (In Press). Complexity in Routine Dynamics: A Source
of Creativity and Inertia. In M. S. Feldman, B. T. Pentland, L. D’Adderio, K. Dittrich,
C. Rerup and D. Seidl, eds., Cambridge Handbook of Routine Dynamics. Cambridge:
Cambridge University Press.
Hansson, M., Pentland, B. T. & Hærem, T. (2017). Identifying mid-range patterns of action:
Tools for the analysis of organizational routines. Academy of Management Proceedings.
Hernes, T. (2014). A Process Theory of Organization. Oxford: Oxford University Press.
Keegan, B. C., Lev, S., & Arazy, O. (2016). Analyzing organizational routines in online
knowledge collaborations: A case for sequence analysis in CSCW. Proceedings of the
19th ACM Conference on Computer-Supported Cooperative Work & Social
Computing, 1065-79.
Langley, A. & Tsoukas, H. (2017). The Sage Handbook of Process Organization Studies,
London: Sage.
LeBaron, C., Christianson, M. K., Garrett, L. & Ilan, R. (2016). Coordinating flexible
performance during everyday work: An ethnomethodological study of handoff routines.
Organization Science, 27(3), 514-34.
Levitt, B. & Nass, C. (1989). The lid on the garbage can: Institutional constraints on decision
making in the technical core of college-text publishers. Administrative Science
Quarterly, 34(2), 190-207.
Lindberg, A., Berente, N., Gaskin, J. & Lyytinen, K. (2016). Coordinating interdependencies
in online communities: A study of an open source software project. Information
Systems Research, 27(4), 751-72.
Mabroukeh, N. R. & Ezeife, C. I. (2010). A taxonomy of sequential pattern mining
algorithms. ACM Computing Surveys (CSUR), 43(1).
Mahringer, C. A. (2019). Exploring Routine Ecologies – A Characterization and Integration
of Different Perspectives on Routines, Dissertation, Stuttgart: University of Stuttgart.
23
Mahringer, C. A., Dittrich, K., & Renzl, B. (2019). Interdependent routines and innovation
processes – An ethnographic study of Scrum teams. Academy of Management
Proceedings.
Marcus, G. E. (1995). Ethnography in/of the world system: The emergence of multi-sited
ethnography. Annual Review of Anthropology, 24(1), 95-117.
Mesle, C. R. & Dibben, M. R. (2017). Whitehead’s process relational philosophy. In A.
Langley & H. Tsoukas, eds., The Sage Handbook of Process Organization Studies.
London: Sage, pp.29-42.
Moody, J., McFarland, D. & Bender-deMoll, S. (2005). Dynamic network visualization.
American Journal of Sociology, 110(4), 1206-41.
Pentland, B. T. (1999a). Building process theory with narrative: From description to
explanation. Academy of Management Review, 24(4), 711-24.
Pentland, B. T. (1999b). Organizations as networks of action. In J. Baum & B. McKelvey,
eds., Variations in Organization Science: In Honor of Donald T. Campbell. Thousand
Oaks: Sage, pp.237-53.
Pentland, B. T. (2003a). Conceptualizing and measuring variety in the execution of
organizational work processes. Management Science, 49(7), 857-70.
Pentland, B. T. (2003b). Sequential variety in work processes. Organization Science, 14(5),
528-40.
Pentland, B. T. & Feldman, M. S. (2007). Narrative networks: Patterns of technology and
organization. Organization Science, 18(5), 781-95.
Pentland, B. T., Feldman, M. S., Becker, M. C. & Liu, P. (2012). Dynamics of organizational
routines: A generative model. Journal of Management Studies, 49(8), 1484-508.
Pentland, B. T., Hærem, T. & Hillison, D. (2010). Comparing organizational routines as
recurrent patterns of action. Organization Studies, 31(7), 917-40.
Pentland, B. T., Hærem, T. & Hillison, D. (2011). The (n)ever-changing world: Stability and
change in organizational routines. Organization Science, 22(6), 1369-83.
Pentland, B. T. & Liu, P. (2018). Network models of organizational routines: Tracing
associations between actions. In R. Mir & S. Jain, eds., The Routledge Companion to
Qualitative Research in Organization Studies. New York: Routledge, pp.422-38.
Pentland, B. T., Mahringer, C. A., Dittrich, K., Feldman, M. S. & Ryan Wolf, J. (In Press).
Process multiplicity and process dynamics: Weaving the space of possible paths.
Organization Theory.
Pentland, B. T. & Rueter, H. H. (1994). Organizational routines as grammars of action.
Administrative Science Quarterly, 39(3), 484-510.
Pentland, B. T., Ryan, J. L., Xie, Y., Kim, I., Frank, K. & Pentland, A. P. (2019). Visualizing
clinical routines: What can we see with digital trace data. 11th International Symposium
on Process Organization Studies. Chania, Greece.
Poole, M. S., Lambert, N., Murase, T., Asencio, R. & McDonald, J. (2017). Sequential
analysis of processes. In A. Langley & H. Tsoukas, eds., The Sage Handbook of Process
Organization Studies. London: Sage, pp.254-70.
24
Sabherwal, R. & Robey, D. (1993). An empirical taxonomy of implementation processes
based on sequences of events in information system development. Organization
Science, 4(4), 548-76.
Salvato, C. (2009a). Capabilities unveiled: The role of ordinary activities in the evolution of
product development processes. Organization Science, 20(2), 384-409.
Salvato, C. (2009b). The contribution of event-sequence analysis to the study of
organizational routines. In M. C. Becker & N. Lazaric, eds., Organizational Routines.
Advancing Empirical Research. Northampton: Edward Elgar Publishing, pp.68-102.
Sankoff, D. & Kruskal, J. B. (1983). Time Warps, String Edits, and Macromolecules: The
Theory and Practice of Sequence Comparison, Reading: Addison-Wesley Publishing.
Schwaber, K. & Beedle, M. (2002). Agile Software Development with Scrum, Upper Saddle
River: Prentice Hall.
Sele, K. & Grand, S. (2016). Unpacking the dynamics of ecologies of routines: Mediators and
their generative effects in routine interactions. Organization Science, 27(3), 722-38.
Van de Ven, A. H., Polley, D., Garud, R. & Venkatraman, S. (1999). The Innovation Journey,
New York: Oxford University Press.
Van der Waal, K. (2009). Getting going: Organizing ethnographic fieldwork. In S. Ybema, D.
Yanow, H. Wels & F. Kamsteeg, eds., Organizational Ethnography: Studying the
Complexities of Everday Life. London: Sage, pp.23-39.
Witten, I. H., Frank, E., Hall, M. A. & Pal, C. J. (2016). Data Mining: Practical Machine
Learning Tools and Techniques, Cambridge: Morgan Kaufmann.
Wurm, B., Grisold, T., Mendling, J. & vom Brocke, J. (In Press). Business Process
Management and Routine Dynamics. In M. S. Feldman, B. T. Pentland, L. D’Adderio,
K. Dittrich, C. Rerup and D. Seidl, eds., Cambridge Handbook of Routine Dynamics.
Cambridge: Cambridge University Press.
Yanow, D. (2012). Organizational ethnography between toolbox and world‐making. Journal
of Organizational Ethnography, 1(1), 31-42.