Coding Observed Interaction

Full text

1
CODING OBSERVED INTERACTION
Alan Sillars
Department of Communication Studies
University of Montana
Nickola C. Overall
School of Psychology
University of Auckland
Sillars, A., & Overall, N.C. (2016). Coding observed interaction. In D. Canary & A. VanLear
(Eds.), Researching Interactive Communication Behavior: A Sourcebook of Methods and
Measures (pp. 199-215). Thousand Oaks, CA: Sage.

2
CODING OBSERVED INTERACTION
In this chapter, we discuss practical and conceptual issues when coding observed
communication. At first glance, the process can seem straightforward -- one selects a coding
system, trains coders to use the manual, and checks reliability. However, coding requires more
than mechanically applying categories or ratings to message units. Coding is a form of message
interpretation, analogous to what happens in all communication (Folger, Hewes, & Poole, 1984).
Coders, like participants in communication, apply interpretive rules to discourse and nonverbal
behavior in order to discern meaning -- either conventional meaning or meaning specific to
observer/participant goals. In observational coding, as in everyday communication, standardized
coding rules promote shared meaning (i.e., reliability) but do not remove all ambiguity (Sillars &
Vangelisti, 2006). Coders must improvise when interpreting novel or ambiguous examples,
drawing on their own experience and anticipating how others would view the same message.
Coding is also an exercise in selective perception. Because messages are multi-functional (Sillars
& Vangelisti, 2006) and have different levels of meaning (e.g., content vs. relational), the same
interaction can be coded many ways that do not inherently compete. Coding methods selectively
highlight functions of communication (e.g., persuasion or support), levels of analysis (e.g., molar
vs. molecular), intended meanings (e.g., observer vs. participant), structural properties (e.g., base
rates vs. sequential structure), and so forth. Thus, many alternative ways of coding exist that may
be appropriate (or not), depending on one’s purpose and perspective.
Our experience with observational coding mostly stems from research on couple and
family conflict. We draw on this experience to ground discussion of general issues in coding.
Conflict is one of the most researched aspects of family communication (Sillars & Canary, 2013)
and an area with a long tradition of observational work. Whereas a later chapter provides a

3
review (see Canary, this volume), we cite conflict coding methods selectively to illustrate issues,
options, and tradeoffs when conducting any form of interaction analysis.
Conceptual Foundations of Observational Coding
Observational coding typically involves coders independently categorizing or rating the
verbal and nonverbal content of a recorded interaction according to specified protocols and
coding schemes. Coding yields a systematic record of ongoing communication, albeit a selective
one structured by researcher assumptions and theories. As Krippendorf (2004) stresses,
inference is inherent to content analysis of communication, because the outward (physical)
features of messages have no meaning of their own -- messages only acquire “content” by people
engaging them conceptually. Even automated coding performed by computers rests on theories
of programmers about how humans read and respond to messages (Krippendorf, 2004). Coding
supplies content by filtering, segmenting, and highlighting aspects of communication that have
meaning relative to one’s purpose and conceptual framework. Of necessity, the process
highlights certain features while disregarding many others. Moreover, interaction analysis (i.e.,
content analysis of free-flowing conversation) is especially selective. The verbal, vocal, and
kinetic activities people carry out while speaking and listening are so complex and information
dense per unit of time that formal analysis cannot presume to yield more than partial
understanding (Street & Cappella, 1985, p. 4).
Given the interpretive and selective nature of coding, tradeoffs occur when deciding to
adopt a coding system, adapt one, or invent one’s own. Well-studied aspects of communication,
including most topics in this book, have already spawned multiple systems. It is clearly more
efficient to use an existing system than to begin from the ground up. The proliferation of coding
schemes also complicates synthesis of results; leading some authors to even call for a

4
moratorium on development of new methods (Kerig, 2001). On the other hand, adopting a
coding scheme means buying into particular assumptions about what message features are
important and what they signify. Thus, well-established coding options are not all-purpose.
Bakeman and Gottman (1997, p. 15) comment that borrowing a coding scheme can feel like
“wearing someone else’s underwear;” as coding represents a theoretical act originating within
the confines of a particular research program.
Research on couple conflict illustrates connections between coding methods and
researcher perspectives. Table 1 reports categories from familiar coding schemes for couple
conflict, including the Marital Interaction Coding System (MICS-IV), Kategoriensystem für
Partnerschaftliche Interaktion (KPI), Couples Interaction Scoring System (CISS), and Verbal
Tactics Coding Scheme (VTCS). Table 2 reports similar codes from two rating systems -- the
Conflict Rating System (CRS), and Communication Strategies Coding Scheme (CSCS).
(Categorical codes and ratings are discussed further under Forms of Coding.) Collectively, the
systems share much in common. Systems used to code couple conflict tend to reflect two broad
dimensions: valence and directness (see Overall, Fletcher, Simpson, & Sibley, 2009; Sillars &
Canary, 2013). The valence dimension is explicit in systems that collapse into positive-negative
supra-categories (KPI, CRS, CSCS); however, all of the coding systems have been used to
operationalize positive-negative communication. Directness is reflected in engagement versus
avoidance of conflict (e.g., the demand and withdraw subscales of the CRS), along with direct
and indirect influence attempts (as in the CSCS). The coding systems in Tables 1 and 2 are also
similar in what they omit. That is, they foreground relational aspects of conflict at the expense of
other potentially important processes; for example, bargaining tactics (Putnam & Jones, 1982)
and argument structure (see Seibold and Weger, this volume). Thus, the coding schemes are well

5
suited to research on valence and directness of conflict communication but disregard many other
potentially important features.
Despite broad similarities, the coding schemes in Tables 1 and 2 also reflect important
differences that stem from research goals and observational contexts. Some coding schemes
originating in clinical psychology, such as the MICS, KPI, and CISS, were designed to isolate
communication skill deficits of unhappy couples as a basis for couple therapy. Early studies in
this tradition conceptualized communication according to social learning principles, as
contingent patterns of positive and negative behavioral reinforcement (Birchler, Weiss, &
Vincent, 1975; Gottman, 1982). Thus, codes are organized and aggregated into positive/negative
forms of communication, partly based on how messages are presumed to affect marital
outcomes. Although this division serves a purpose for behaviorally-oriented therapists, others
might find the approach limiting. In their dialectical critique of the satisfaction literature, Erbert
and Duck (1997) chafe at the notion that interaction characteristics discriminating adjusted-
maladjusted relationships can be dichotomized as positive/negative communication. In their
view, the positive/negative duality reinforces an idealized view of relationships as either happy
or conflicted and obscures ways that interactions may be simultaneously positive and negative.
In contrast to clinically-based research, Sillars developed the VCTS with the assumption
that dyadic interaction styles may have variable associations with outcomes, depending on
relationship context (see Sillars & Wilmot, 1994).1 Similarly, Overall, Fletcher, Simpson, and
Sibley, C.G. (2009) developed the CSCS to move past assumptions that “positive” and
“negative” messages inherently benefit or harm relationships by distinguishing between direct
(e.g., coercion) and indirect (manipulation) influence strategies. Research using the CSCS and

6
VCTS provides evidence that seemingly “negative” acts can sometimes help couples directly
tackle relationship problems (McNulty & Russell, 2010; Overall, et al., 2009).
The treatment of avoidance also differs across conflict coding schemes. Early generation
coding systems in psychology (e.g., MICS, CISS; Table 1) primarily featured direct forms of
conflict engagement (although withdrawal was added as a category in the fourth revision of the
MICS). This reflects the main observational method, the problem-solving paradigm, whereby
couples interact in a lab under instruction to discuss and resolve an acknowledged problem
(Gottman, 1994, pp. 18-19). Although the problem-solving paradigm remains a dominant
approach, later generation systems (e.g., CRS2; Table 2) focus more on withdrawal from
interaction, accounting for the fact that individuals sometimes disobey researcher instructions to
engage. Moreover, withdrawal in response to partner demand predicts relationship dissatisfaction
(Eldridge & Christensen, 2002). In contrast to research using the problem-solving paradigm, the
VTCS (Table 1) was developed from research that allowed greater latitude for conflict avoidance
and neutrality; for example, couples were instructed to discuss potential conflicts “until they had
nothing further to say” (e.g., Sillars, Pike, Jones, & Murphy, 1984). Consequently, the VTCS
distinguishes non-engagement tactics more than do other coding schemes.
Despite these contrasts, all coding systems in Tables 1 and 2 rely on structured
observation, at home or in a lab, whereby researchers prompt couples to discuss relationship
issues. No doubt, naturalistic observation of conflict would reveal other forms of avoidance, such
as leaving the scene, retreating to electronic devices (Heyman et al., 2014), or interspersing
confrontation with attention to daily tasks (Sillars & Wilmot, 1994). Observational context also
affects the dimensions of communication readily observed. For example, the coding schemes in
Tables 1 and 2 contain more “negative” codes than “positive” or constructive ones. Heyman

7
(2001, p. 7) notes that, “Whereas it is relatively easy to get unhappy couples to argue on
command, behaviors that promote the various forms of love … are much more challenging to
witness in the laboratory.”
In sum, coding schemes connect to researcher assumptions, goals, and observational
methods. No coding scheme can suffice for all purposes and most require significant adaptation
when there is a shift from the original context in which methods were developed.
Forms of Coding
Coding may take a variety of forms, including categorical codes, checklists, and ratings.
Each approach invokes conceptual and practical tradeoffs.
Discrete Coding Systems
Categorical codes. In the classic sense, coding involves classifying message units into
mutually exclusive and exhaustive categories (Krippendorf, 2004). Categorical coding schemes
are sometimes referred to as micro codes, because they code communication at the level of
individual messages; whereas macro codes (e.g., ratings) describe longer segments of interaction
(Lindahl, 2001). The CISS, KPI, MICS, and VTCS (Table 1) illustrate categorical coding
schemes. These systems first identify a unit of observation (such as the speaking turn or thought
unit3) and then exhaustively code these units into a fixed set of categories. Sub-categories might
be nested under broader categories in order to yield a more detailed description at the level of
sub-categories, while providing sufficient observations for quantitative analyses after collapsing
codes (e.g., blame in the MICS-IV is a combination of criticize, mindread, putdown, turn-off).
The primary advantages of categorical codes are their descriptiveness and flexibility.
Although not nearly as fine-grained as qualitative conversation analysis (Robinson, 2011),
categorical coding yields a more detailed record than do other forms of quantitative interaction

8
analysis.4 Categorical coding is also conducive to statistical analysis of sequential structure,
which examines whether specific codes elicit an immediate response (VanLear and Davis this
volume). In relationship conflict, important sequences include the probability that negative codes
are reciprocated by the partner (negative reciprocity) or demand is followed by withdrawal. The
categorical coding systems in Table 1 were developed in a period marked by influential calls to
focus on the temporal organization of interaction as a way to operationalize systems thinking
about relationships (e.g., Gottman, 1979; Watzlawick, Beavin, & Jackson, 1967). Categorical
codes also offer flexibility in subsequent aggregation, assuming that the initial round of coding
identifies more than a few categories. When detailed codes are aggregated into broad categories,
the research can document how specific codes contribute to summary scores. Unfortunately, this
step is often omitted when researchers report aggregate codes.
The time and expense of categorical coding poses a clear tradeoff. For instance, trained
coders need 1½-2 hours to analyze a 10 minute interaction using the MICS (Heyman, 2004) and
even longer periods using the CISS (Notarius, Markman, & Gottman, 1983). Detailed coding of
interactions requires, at minimum, an audio (and sometimes video) record, and is usually assisted
by written transcripts. In addition to the time and expense of transcription, the interaction record
must be unitized, which requires separate coder training and reliability assessment if the unit of
analysis involves significant coder judgment (as with thought units). Coding itself can require
difficult decisions about how to assign borderline examples to similar categories, which fatigue
coders and contribute to poor reliability. Thus, as Heyman et al. (2014) note, microanalytic
coding carries a poor cost-benefit tradeoff when a large number of initial categories are later
aggregated into just a few (e.g., positive vs. negative communication).

9
One way to make coding more efficient is to apply coding schemes selectively, using
only the categories of greatest relevance. For example, McNulty and Russell (2010) limited their
use of the VTCS (Table 1) to negative (i.e., confrontative) codes, as their purpose was to assess
longitudinal impacts of negative messages on marital satisfaction. Others have developed “rapid”
coding systems, such as the RCISS (Krokoff, Gottman, & Hass, 1989) and RMICS (Heyman,
2004), which mimic the CISS and MICS (Table 1) but dispense with detailed subcategories.
These rapid coding systems make restrictive assumptions about what aspects of interaction are of
interest (again focusing primarily on positive vs. negative communication), which can represent
an advantage or limitation depending on one’s point of view.
Mutually exclusive and exhaustive coding schemes pose conceptual as well as practical
challenges. Mutual exclusivity requires the assignment of a single code per unit, although, in
theory, messages perform multiple functions simultaneously (Jacobs, 2002; Robinson, 2011). For
example, friendly joking during conflict might show affection at the same time that it conveys
tacit criticism. Thus, coders must judge the primary function of a message relative to the purpose
of the coding system. To assist coders, categorical coding sometimes invokes rules of precedence
that assign a coding unit to one particular category when it potentially fits multiple categories.
For example, the MICS-IV and VTCS (Table 1) assign priority to codes seen as more important
or as offering clearer interpretation.
Folger et al. (1984) advise against strict adherence to mutual exclusivity and suggest that
validity concerns can require one to code each unit into multiple categories or along more than
one dimension. However, one can readily see practical limitations to such advice. Allowing
multiple codes increases the complexity of coding and subsequent analysis -- one must determine
when and how to assign multiple codes without compromising reliability, how to collate variable

10
codes per unit, and how to analyze sequential structure if there are multiple antecedent and
consequent acts. Instead of multiple codes, another way to address multi-functionality is to use
more than one coding system. For example, the CISS has separate codes for verbal content and
nonverbal affect. Of course, this approach also multiplies the time and expense of coding.
The conventional requirement of exhaustiveness raises a different conceptual issue. To
ensure exhaustiveness, categorical systems routinely include a default category, such as
uncodable, other, or neutral, which provide designations for units that are not otherwise
classified by the system. Krippendorf (2011) advises against overly broad application of the
default category, as this suggests that the coding system is logically incomplete and yields
unusable information. An overly broad default category also provides coders with an easy way of
avoiding difficult decisions that can be a source of unreliability (Krippendorf, 2011). On the
other hand, coding every unit risks over-interpreting messages that lack clear meaning on the
dimensions coded. An alternative involves sieve coding (Guetzkow, 1950), whereby researchers
designate only certain units for coding based on their research aims (Folger et al., 1984).
McNulty and Russell’s (2010) selective coding of negative messages illustrates this strategy, as
does coding of question sequences in physician-patient interviews (Robinson, 2011).
Checklists. When using checklists, coders identify all categories that apply to the coding
unit in binary fashion (i.e., each code is either present or absent). Checklist coding methods are
especially common in observational studies of parent-child interaction (e.g., Roggman, Cook,
Innocenti, Norman, & Christiansen, K., 2013). The RCISS illustrates use of a checklist system
for coding couple conflict (Krokoff et al., 1989). Checklists might apply to short units, such as
speaking turns (as in the RCISS), longer time-based intervals (e.g., Vivian, Langhinrichsen-
Rohling, & Heyman, 2004), or entire interactions. In contrast to categorical systems, checklist

11
codes are not mutually exclusive and are not necessarily exhaustive. For example, one could
code for verbal confrontation without discerning any relevant forms in a given interaction.
Checklists thereby simplify coding relative to categorical systems because coders do not have to
fit each unit into one and only one category. This makes it practical in some cases to conduct
coding “live” during naturalistic observation or to code recorded interactions without transcripts.
However, the relative efficiency of checklists can partly rest on application of a relaxed
reliability standard, in which reliability is assessed in terms of summary scores (e.g., overall
positivity/negativity) rather than unit-by-unit coder agreement (e.g., Krokoff et al., 1989).
Rating Systems
Rating systems involve coders rating the degree to which people display targeted
communicative acts. As with the rapid versions of the categorical systems described above
(RMICS and RCISS), rating systems typically focus on higher-order categories that categorical
micro-codes are often combined into. Rather than distinguishing a large list of distinct acts,
coders consider a range of relevant acts to determine the presence of broadly defined dimensions,
such as positive, negative, and avoidance (Gill, Christensen & Fincham, 1999; Julien, Markman
& Lindahl, 1989). Researchers using this approach recognize that theoretically relevant
dimensions often represent clusters of interrelated acts. These clusters of interrelated acts might
not all be exhibited or enacted to the same degree by a particular person. Whereas categorical
codes indicate whether a code happens or not, ratings often integrate information on frequency,
intensity, and duration to index the magnitude of the targeted act (Margolin et al., 1998).
A good example of a rating system is the Conflict Ratings Scale (CRS; see Table 2),
which was designed to assess demand-withdraw patterns in couple conflict. Observers watch the
entire interaction and rate the degree to which each partner exhibited each dimension (e.g.,

12
discussion, blames, pressures for change) during the interaction (1 = none, 9 = a lot). Coders are
instructed to consider the frequency, intensity, and duration of the verbal and nonverbal
behaviors relevant to each dimension, and make a judgment of magnitude relative to other
individuals in similar interactions. Christensen, Heavey, and colleagues decided to use global
ratings to focus on interaction patterns that can manifest in a variety of ways and to assess the
intensity rather than frequency of such patterns (Sevier, Simpson & Christensen, 2004). The
resulting ratings distinguish between mild and severe forms of demand-withdraw that may or
may not occur at the same frequency. For example, mild but frequent hesitation to discuss topics
would produce a lower “withdraws” rating (see codes in Table 2) than extreme disengagement
and silence that occurred for a shorter time. Balancing frequency with intensity in ratings of
magnitude is important because instances of extreme disengagement at pivotal moments in the
interaction are likely to have a more pronounced impact on problem resolution and subsequent
relationship outcomes (see Sevier et al., 2004).
A central benefit of rating systems is that they reduce the time and expense required to
obtain analyzable data while producing similar results as categorical codes (Gill et al., 1999;
Julien et al., 1989). Gill et al. (1999) coded couples’ conflict interactions using the VTCS (Table
1), a categorical code system, and the revised CRS to contrast the utility of each system. The
VTCS required more training for coders to reliably distinguish specific codes (about 15 hours)
and additional hours to transcribe, unitize, and code interactions. In contrast, the CRS assumes
that coders are already equipped with a general understanding of coding constructs and thus
require only a short training period to fine-tune this existing knowledge (about 8 hours). Rating
entire interactions (vs. speaking turns) directly from video recordings (vs. transcripts and video
for the VTCS) took less than an hour per couple. After combining VTCS discrete codes into

13
similar dimensions as the CRS, the scores derived from each coding system were associated. The
systems also predicted concurrent and longitudinal satisfaction in similar ways. The one
difference, however, was that global ratings of avoidance in the CRS appeared to capture a
broader array of communicative acts than those assessed by the VTCS, which could enhance
predictive utility but might also reduce understanding of the meaning and impact of specific acts.
Although ratings are an efficient approach to coding, this can be partially offset by the
need for multiple raters per interaction to ensure adequate reliability. For example, Gill et al.
(1989) had eight raters (four per spouse) analyze each interaction, with reliability based on
combined ratings (Spearman-Brown formula). A single coder applied the VTCS, except for 20%
of interactions that were double-coded to check reliability (kappa).
Critically, rating systems allow messages to own multiple functions. As described above,
in most categorical code systems, observers need to assign one code to each unit, which can
involve tough decisions regarding the principal function of the unit. In rating systems,
communication can be indexed as a blend of different acts, with the final ratings capturing the
relative weight of applicable categories. For example, the CSCS (Table 2) organizes ratings into
higher-order categories that reflect the valence and directness of communication strategies.
Partners’ communication across the interaction or within a specific speaking turn can be a blend
of all four types. For example, a person might try to reason with their partner (positive-direct)
while also threatening negative consequences if his/her solution is not adopted (negative-direct).
The resulting ratings represent the relative magnitude of each type, such as high levels of
positive-direct (5 out of 7) and relatively mild negative-direct (3 out of 7) or vice versa. By
assessing the relative presence of different strategies, this approach does not truncate assessment
to the primary strategy only, but still maintains the ability to hone in on which aspects of

14
communication are most predictive of outcomes. For example, accounting for the associations
across direct strategies, Overall et al. (2009) found that both positive-direct and negative-direct
strategies were independently associated with greater problem resolution over time. Rating the
magnitude of all strategies also avoids the difficulty of trying to classify polysemous (i.e.,
multiple-meaning) messages into discrete codes.
Rating systems also contain important drawbacks. Global ratings lack detail regarding the
specific acts present and therefore which acts might have the strongest explanatory power.
Rating systems also lack information about time and sequential contingencies across partners,
such as the likelihood that demand prompts withdraw. Although the CRS ratings of one partner’s
demand and the other partner’s withdraw can be combined to create demand-withdraw
composites, such an index does not reveal whether withdraw was contingent on (i.e., was
influenced by) the partner’s demand (Sevier, et al., 2004).
Alternatively, the presence of specific sequences can be rated, such as the degree to
which a parent demands and child withdraws across an interaction (e.g., Caughlin & Ramey,
2005). This approach does not constrain assessment of sequences to each turn or unit of analysis
(as does sequential analyses). Such lack of constraint proves useful if important interaction
patterns occur across wider time spans and, more importantly, if the time course of dyadic
patterns or the length of interaction varies across the sample. In addition, rather than rating the
entire interaction, the interaction can be divided into shorter time intervals, rating systems
applied to each interval, and then time-series analyses used to test contingency-based predictions.
For example, Overall, Simpson and Struthers (2013) used the CSCS to rate interactions every 30-
seconds to test whether positive-indirect strategies by one partner were associated with
reductions in withdrawal in the next 30-second interval.

15
The most important limitation of rating systems might be that they rely heavily on
coders’ interpretation of the communication exhibited, even more so than typical categorical
systems. By coding more global categories, rating systems focus on what the researcher believes
is theoretically relevant. This helps ensure that the design tests research questions of interest and
is valuable when the wider context of the interaction alters the meaning of the same specific act,
such as whether advice on how to tackle a problem represents reasoning or autocracy (CSCS,
Table 2). However, focusing on broader categories asks coders to make inferences about the
meaning of observed communication and then aggregate these inferences with frequency and
intensity to generate a holistic rating (Margolin et al., 1998). Both the CRS and CSCS (Table 2)
adopt a “cultural informant” approach (Gottman & Levenson, 1986), which assumes that coders
possess a deep understanding of social interactions, make such interpretations in their day-to-day
lives, and thus can reliably decode the meaning of communication. Nonetheless, relying on
coders’ interpretations inevitably provides more room for idiosyncratic views to bias ratings. In
contrast, the descriptiveness of many categorical codes reduces the level of inference required,
which may reduce coder bias. We discuss coder inference and bias in more detail below.
The Role of Inference in Communication Coding
Sources and Levels of Inference
Although inference is inherent to observational coding (Krippendorf, 2004), it is not
always clear what kinds of inferences are carried by communication codes (Folger et al., 1984).
Much of the time, observational codes are simply called “communication behaviors,” suggesting
that codes reference outward features of communication only (i.e., what people “actually” do).
Although actual behavior is the starting point for observational research, coding schemes
typically do not describe behavior so much as produce structured inferences about functional

16
properties of communication (e.g., messages as forms of affection, social support, or conflict
avoidance). Even basic observations, such as the recognition that vocalizations constitute an
utterance or that simultaneous speech constitutes an interruption, interpret observable signals in
terms of their function, including meaning and intention.
As Stone, Tai-Seale, Stults, Luiz, and Frankel (2012) observe, inferences made by coders
can be ambiguous in ways that are not obvious from the usual description of coding procedures.
These authors coded illness-related emotions expressed by patients and empathic responses by
physicians, phenomena that have parallels in the way couples express and respond to
emotionally-laden disclosures during conflict. Although they used a previously validated coding
system, Stone et al. (2012) found that patient verbal expression of emotion was ambiguous in
unanticipated ways. For example, emotion words and other cues were often “fuzzy” and varied
from one patient to another; moreover, discussion of illness appeared emotionally-laden to
coders even in the absence of emotion cues recognized by the coding system.
Coding systems differ in how they resolve such ambiguities. On the one hand, a system
might restrict attention to readily observable emotion cues, as in automated analysis of affect
based on word valence (Baek, Cappella, & Findman, 2011), facial expressions (Cohn & Sayette,
2010), or acoustic features of speech (Black et al., 2013). Alternatively, coders might identify
emotions from context, based on their own implicit cultural knowledge and experience.
The different approaches reflect a distinction between manifest (physical or surface)
versus latent (symbolic) content analysis (e.g., Holsti, 1969). Most obviously, manifest content
includes nonverbal behaviors recorded without assistance by human coders or inference about
sender intent. Whereas inferences about message intent are essential to interpretation of verbal
communication (Jacobs, 20002), Buck and VanLear (2002) argue that many nonverbal behaviors

17
are emitted and apprehended spontaneously (i.e., unintentionally and automatically) based on
biologically programed response patterns. Coding of spontaneous communication still involves
inference, insofar as it rests on theoretical assertions about which manifest cues are important to
observe and what functions they serve. Nonetheless, coding of physical cues (e.g., movement of
facial muscles) does not require inference about conventional or personal meaning, as does
coding of verbal communication or symbolic forms of nonverbal expression.5 In-between strictly
manifest and latent content lie forms of coding that involve low level inferences about speaker
intent that are performed easily by any competent language user (e.g., whether a question is
rhetorical). However, most interaction coding is more inferential – the codes identify abstract
relational events (e.g., confrontation) and associated acts (e.g., criticism). Here again,
considerable variation occurs in the discretion afforded to coders. Some systems constrain coder
inferences through extensive rules and training, whereas others (such as ratings systems noted
above) treat coders as cultural informants and allow them greater latitude to fill-in meaning.
In addition to the inferences conveyed by coders, a second level of inference occurs when
researchers aggregate codes into summary measures. For example, most categorical coding
systems confine coder judgments to moderate inferences (e.g., whether an utterance represents
acceptance or denial of responsibility) but aggregate based on researcher theories connecting
specific codes to summary constructs (e.g., overall positivity/negativity).6 Notably, coding
methods do not always collapse codes in the same way. For example, avoidance and withdrawal
are treated as communicative negativity in some systems (RCISS, RMICS) but not others (CRS,
VTCS) and problem description may be construed as positivity (RCISS) or neutrality (RMICS).
Moreover, researchers often modify constructs ad hoc when collapsing codes. Heyman (2001)

18
notes that researchers have “mixed and matched” codes from the MICS to such an extent that
virtually no studies evaluate identical constructs.
Locus of Meaning
Another general principle of message interpretation is that the same overt signals can
mean something different to participant versus observer (Surra & Ridley, 1991) or to multiple
observers with different frames of reference. Coding methods also assess meaning from varying
perspectives. Poole, Folger, and Hewes (1987) identify four such perspectives (see also the
chapter, Establishing Reliability and Validity, this volume). Generalized observer meanings are
those available to any uninvolved onlooker to an interaction (e.g., a vocalized pause), whereas
restricted observer meanings are derived from application of a specialized interpretive scheme
by outsiders (e.g., conversational coherence). Generalized subject meanings are available to any
member of a cultural or subcultural group (e.g., topic shifts), whereas restricted subject
meanings are accessible only to relationship insiders (e.g., inside jokes or conflict triggers).
In what domain does most communication coding reside? The perspective of the
generalized observer is well-represented in interaction research but limited to features that can be
assessed through manifest content. Restricted subject meaning is not assessable via observer
coding, at least as practiced in quantitative interaction research. Instead, most interaction
research spans the boundary of restricted observer and generalized subject meaning. For
example, all of the coding schemes in Tables 1 utilize specialized interpretive rules applied by
trained observers, which suggests restricted observer meaning. However, the systems also rely
on coders to use their own cultural knowledge to fill in where coding rules are incomplete; for
example, when discriminating friendly versus hostile joking or criticism versus neutral
description based on context.

19
Herein lies the central dilemma of interaction coding. A primary reason for doing
interaction coding is to provide an “objective” (i.e., standardized, outsider) perspective on
communication that avoids the biases of self-report data and provides a contrast to participant
meaning. However, because it is not always possible to codify interaction constructs in terms of
manifest content or clearly identifiable stimulus features, coding methods ultimately rely on
intuitive judgments by observers to interpret meaning. An advantage of human coders over
automated coding is that coders can use their own cultural knowledge to make sense of implicit
features of communication. A limitation is that coders can interject their own knowledge in ways
that threaten reliability and validity.
Coder Bias
To the extent that observational methods rely on coders to fill-in meaning from cultural
knowledge, the methods assume that coders represent cultural or subcultural groups in which
meanings often reside. Coding methods also assume that coders can apply cultural knowledge to
the specific context under investigation. Coders are usually undergraduate or graduate college
students. Students can represent broader cultural meanings when these meanings are widely
shared. This should be the case with low level inferences about speech acts but not necessarily so
with abstract relational events. Moreover, student coders often fail to represent the cultural and
socio-economic mix of the sample, which potentially affects interpretation of the acts coded.
The relative homogeneity, and therefore interpretation, across student coders might also mean
that potentially distinct interpretations are not revealed by reliability checks. Their life and
relationship experiences can also mean that student coders are ill-equipped with contextual
knowledge central to the domain of investigation, such as examining communication during the

20
transition to parenthood, within parent-child dynamics, or in distressed samples, such as people
suffering depression, coping with chronic illness, or facing high levels of violence.
Indeed, as Margolin and colleagues’ (1998) note, life experience, gender, and ethnicity
can all affect coder judgments. Male coders have a greater propensity than females to view adult
behavior as angry and resentful (Davidson, et al., 1996) and to see aggressive behavior in
children’s interactions (Pellegrini, et al., 2011). Gender stereotypes are also likely to affect the
way women and men are coded, including the inferred intent behind similar behaviors (e.g.,
silence as sullen guilt-induction versus withdrawal). Similarly, stereotypes of ethnic and cultural
groups can bias coding (Bente, Senokozlieva, Pennig, Al-Issa, & Fischer, 2008). Cultural
differences can also affect coder inferences because of the way targeted constructs manifest
across cultural groups. For example, cultural differences in the appropriateness of direct conflict
(Sillars & Canary, 2013) could mean that interactions that appear contentious or avoidant to
observers are not experienced in the same way by cultural insiders.
Coders’ own relationship experiences are also likely to affect how coders evaluate and
infer meaning from other people’s communication. The relationship field is replete with
examples of individual and contextual factors that shape how relationship events are construed
and responded to, such as attachment insecurity, relational standards, or levels of relationship
satisfaction. Examining families within diagnostic contexts, such as discussing areas of conflict
or supporting each other, will undoubtedly activate associated expectations, preferences, and
perceptual sets that affect the way interactions are perceived. People are also highly motivated to
maintain positive evaluations of their own relationships, and one way this is managed is by
downplaying the positivity of other relationships (e.g., Rusbult, Van Lange, Wildschut, Yovetich
& Verette, 2000). This bias might produce a tendency to perceive others’ communication as less

21
constructive or loving than is justified (Gagné & Lydon, 2004). Finally, coders might generate
their own understanding of the goals of the research (Harris & Lahey, 1982). By extension,
individual coders possess their own conceptions about what constitutes “good” or “bad”
communication. Coders’ application of these tendencies can potentially undermine the
assessment intended by the researcher.
What can be done to counteract coder bias? Margolin et al. (1998) recommend ensuring
coding teams are diverse in gender, culture, and general background, including replacing or
combining student coders with coders sourced from the wider community. However, achieving
representativeness among coders in relation to the target population may not be practical, and it
can lead to other problems, such as the coding schedule being applied in unintended ways and
increasing training time. Nonetheless, coder bias is a significant issue. The potential for bias does
not render observational coding invalid or useless; however, we do think it necessary to assess
results of coding in light of the limitations of human judgment and the perspectives and
dispositions coders bring to the task. Moreover, researchers should take every step to minimize
coder bias by structuring, limiting, and monitoring coder inference during the coding process.
Managing the Coding Process
Ultimately, coding procedures are designed to coordinate inferences while maintaining
the integrity of coding constructs; which equates to the topics of reliability and validity. Whereas
a subsequent chapter provides a comprehensive discussion of reliability and validity (Poole &
Hewes, this volume), we highlight how reliability and validity are affected by coding procedures
and coder characteristics. Reliability and validity are analogous to the problem of inter-
subjectivity that is the crux of symbolic communication. To coordinate inferences, coders must
apply coding rules consistently and fill-in meaning by adopting the perspective of others who

22
operate within a particular (generalized or restricted) meaning domain. The success of this
enterprise is affected by characteristics of the coding scheme, coding procedures, and coders.
With respect to the coding scheme, more inferential codes are potentially subject to
greater bias, as noted above. More inferential codes also tend to be, but are not inevitably, less
reliable. As Krippendorf (2004, p. 20) notes, coders can sometimes read between the lines with
remarkable consistency. On the other hand, Stone et al. (2012) ultimately limited their coding of
emotional expression to the most explicit examples after attempts to code indirect emotional
expression proved unreliable. Similar compromises are built into most coding schemes.
Researchers often omit subtle and variable features of communication for reliability reasons, no
matter how theoretically heuristic these features might be. The complexity of a coding system
also affects inter-coder reliability. Heyman et al. (2014) advise that coders generally cannot
maintain adequate agreement when there are a large number of subtle codes. However,
exceptions exist (e.g., Cegala, McClure, Marinelli, & Post, 2000; Sillars et al., 1984).
Procedures can reduce the burden on coders when categorizing or rating a large number
of constructs or difficult to judge constructs. For example, in the CSCS, interactions are coded
for one category at a time to ensure coders focus on the particular influence strategy targeted
during that wave. Coding in waves reduces cognitive demand; although coders still need to
distinguish between multiple strategies, they only need to assess the strategy they are rating in
that wave. Applying rating systems to small time intervals, rather than rating multiple
dimensions across entire interactions, has the same benefits and may enable coders to more
effectively rate and distinguish between multiple codes. These procedures might also reduce the
degree to which coders’ subjective evaluations can infiltrate the coding process. Furthermore,
additional coding waves can minimize the degree to which the tone of the interaction influences

23
coding. For example, utilizing a separate team of coders to index broad dimensions, such as
general valence or problem resolution, can provide a way of ensuring that more specific codes
are not “infected” by coders general sense of the interaction.
Although more complex coding systems are not inherently less reliable or subject to bias,
they might require more detailed coding manuals, greater rule specification, and more extensive
training. A coding manual extends the coding scheme by specifying and illustrating coding rules
in detail. A more complete coding manual simplifies coding by anticipating and resolving areas
of confusion. Inexperienced coders may expect the coding manual to remove all ambiguity; that
is, they assume that there is always a “correct” code under the coding rules. Inevitably, however,
examples emerge that the author(s) of the coding manual had not anticipated. Further, even
familiar examples can become ambiguous due to a shift in context. In such cases, some
unreliability is preferable to perfect reliability achieved through arbitrary decision rules that
sacrifice validity. Ideally, observers should code clear examples with a very high degree of
consistency and make ambiguous judgments with reasonable (at least above chance) reliability
while retaining the spirit of coding distinctions.
The coding manual alone cannot always convey subtle distinctions and ambiguities that
must be understood to code reliably. Much of this information is transmitted during the training
phase. Even systems that rely on coders’ existing culturally-relevant knowledge need to organize
that knowledge into the constructs and language of the coding system and ensure coders apply
that knowledge in the same way. Coder training typically occurs in a stacked fashion. Coders
first get familiar with the manual, and then examples of specific codes and difficult distinctions
are used to enhance understanding. For rating systems, examples of levels (e.g., low, medium,
high) should also be presented to anchor coders’ ratings of relative magnitude. Practice sessions

24
are then conducted, which are used to check coder application, isolate areas of confusion, and
build coder confidence. Extensive discussion throughout this process can help identify and
clarify any problematic areas, and to revise coding rules if needed. Low reliability in this phase
provides important information about needed refinements and can assist the researcher in
clarifying distinctions, both procedurally and theoretically (see Poole & Hewes, this volume).
The amount of coder training and practice needed is relative to the demands of the coding
system. Some codes can be applied reliably by observers after only minimal training. Lorber
(2006) had minimally trained raters assess overreactive discipline of mothers after receiving a
10-minute introduction to coding. Compared with “gold standard” raters, who participated in
weekly training and practice sessions over eight weeks, minimally trained raters were less
reliable, but primarily in terms of mean ratings. Rank order was relatively consistent between
coders (r = .61). Further, minimally trained raters had good concurrent validity with raters who
underwent gold standard training (r = .72). These results suggest that minimal training may
suffice for assessing relative (vs. absolute) scores for interaction, which is often all that is needed
to test hypotheses. However, minimal training is most likely to suffice if coding is confined to
surface features of communication (e.g., overreactive discipline was partly defined in terms of
yelling, pushing, pulling) and simple constructs that tap shared meanings and experiences among
coders (e.g., similar experiences of student coders with parental overreaction).
If two or more coders are reliable, this does not necessarily mean that they applied the
coding scheme in the same way any other set of coders would or as the researcher intends. For
example, under pressure to improve reliability, coders may independently or collectively
improvise ad hoc rules that simplify judgments but transform the meaning of codes (Harris &
Lahey, 1982). As much as possible, ad hoc rules should be self-consciously identified and, if

25
appropriate, formalized and incorporated into the coding manual. In that way, one can assess
whether coder improvisations maintain the integrity of conceptual distinctions. A common
temptation is to fashion an ad hoc default category (i.e., “when in doubt, assign code X”) for
ambiguous examples. This tendency makes the code less descriptive and offers a potential source
of spurious observation, especially when coders apply ad hoc rules inconsistently (e.g.,
ambiguous examples are interpreted as verbal aggression when the interaction “feels” tense but
are seen as neutral communication otherwise).
Coder training typically should not stop after coding has begun. Instead, regular meetings
with coding teams provide the opportunity for continual discussion and reflection regarding areas
of uncertainty. Reliability problems and discrepancies in codes should be carefully examined as a
team to reiterate or refine coding categories and rules. In this way, and throughout the coding
process, the researcher explicitly and implicitly clarifies the coding terms. Frequent meetings
with discussion of discrepancies help to counteract against coders drifting from the coding
system. The more interactions that are viewed and coded, the more opportunity coders have to
generate their own rules and for idiosyncratic biases to creep into coders’ understanding and
application of the coding system. Thus, continuous monitoring of reliability and frequent
discrepancy discussions are essential to maintaining reliability.
Further, when coders are aware that their ratings are checked, they are more likely to stay
on task (Harris & Lahey, 1982). Regular checks also provide the chance to consider the presence
of coder biases. Discussing bias openly can help coders recognize the filters they bring to the
coding process and, in turn, may reduce the impact coder bias has on the resulting data.
However, regular meetings and joint coding also has the potential to produce new rules and
definitions, or to create “consensual drift” away from the original meaning of particular

26
categories, as coders’ discussions generate shared implicit rules for evaluating interactions
(Harris & Lahey, 1982). This drift from the original coding manual may result, as described
above, in greater reliability across coders but codes that do not represent the theoretical construct
as originally conceptualized. Guidance by a principal assessor to keep coders true to the coding
system and to record systematic alterations or formal clarifications may be crucial to prevent this
from occurring. However, the assessor must also be reflexive enough to enable coders to query
and challenge in order to prevent coders from simply mimicking the investigator’s view.
Investigators also should ensure they do not label, discuss, or interpret codes in ways that convey
the central hypotheses to coders, thereby compromising coder neutrality (Harris & Lahey, 1982).
Another way to check consensual drift, and reduce the variability that might occur as coders
become more accurate across the sample, is to recode the first 10-20% of interactions.
Along with characteristics of the coding system and coding process, characteristics of
coders affect reliability and validity. The sources of coder bias noted above highlight that coder
demographics can impact results of coding. Moreover, reliability tends to reflect the similarity of
coders in terms of their cultural, educational, and professional background, as well as experience
with texts (Krippendorf, 2004, p. 128). College students are the default choice as coders, both for
convenience and familiarity with coding constructs. Many of the coding schemes used in clinical
psychology and family studies (see Kerig & Lindahl, 2001) require coders with advanced,
specialized education (reflecting a restricted observer perspective). However, researchers using
systems that rely on lay concepts (generalized subject meaning) could prefer coders without
specialized training, because they are less prone to over-interpret interactions. As with decisions
regarding the type of coding system used, coders should also be selected according to the aims of
the research, the coding being conducted, and the nature of the sample assessed.

27
Conclusion: Coordinating Perspectives on Communication
Observational coding of communication represents a form of message interpretation that
parallels everyday communication but with a formal structure for interpretation and self-
reflexive attention to the reliability and validity of inference. As we have noted, most
communication coding represents a standardized observer perspective, which combines elements
of restricted (theory-driven) and generalized (culturally-derived) observer meaning.
Observational coding provides an “objective” perspective in the sense that observations are not
tainted by involvement in the communication episode and are replicable across observers. A key
motivation for doing observational coding is to provide a more objective assessment of
communication than participants’ own self-reports typically provide. Participant accounts of
communication are subject to many known biases, and we often assume that people may not
know, or cannot accurately assess, the acts they and others enact during interactions.
Nonetheless, as we have discussed, coding constitutes an inferential act that often reflects
bias. Whereas participant perspectives are biased by involvement in communication and other
limitations of informal observation, observers are biased by their own goals and experiences.
Observers also lack access to insider context that informs meaning for participants, such as
relationship history and culture. Thus, we caution against treating observational coding as an
unfiltered behavioral description and the only valid/true representation of actual communication.
Kerig (2001, p. 2) sums this point nicely:
People behave in ways that are discrepant from their self-perceptions, and only direct
observation can capture their behavior independently of their appraisals of it….However,
saying that the observer has a unique viewpoint does not mean that it necessarily is the
most valid one. Observational methods are no more purely “objective” than any other

28
tool in the researcher’s toolbox. Underlying every coding category lie choices, and every
choice…is informed by the investigator’s conceptual framework.
In sum, the coding methods we considered in this chapter offer an important way in
which social interaction can be assessed. Nonetheless, the value and utility of the outsider
perspective has to be considered in light of the ways coding methods are applied and, in turn, the
degree to which coding procedures rely on or reduce coder inference and bias. We see
observational methods as a valuable addition to insider perspectives rather than a superior
assessment of communication. Some interaction constructs are best assessed by insider
perspectives. Participants’ subjective emotional experiences, internal dialogue and
communication intentions are difficult (and perhaps impossible) to discern accurately because
insiders’ shared histories and understandings impact the meaning of communicative acts
(restricted subject meaning). Moreover, regardless of the veracity of people’s reports, subjective
experiences and perceptions have a powerful impact on people’s relationship evaluations and
ultimately the course of their relationships. The most complete approach, therefore, is to assess
both insider and outsider perspectives in order to examine how both participants’ subjective
perceptions and the observable patterns that stimulate and result from participant sense-making
shape relationships and the people in them.

29
Notes
1. The original version of the VTCS collapsed into three macro-categories (i.e., integrative,
distributive, avoidance), but was revised to reflect more descriptive macro-categories that
avoid a priori assumptions about which messages serve positive or negative functions.
2. An even more recent system that evolved from the CRS, the Couples Interaction Rating
System (CIRS), has summary scores for demand and withdrawal, but lacks the positive and
negative scales of the CRS (see Sevier, Simpson, & Christensen, 2004).
3. A thought unit is a segment of speech that expresses a single, unified thought.
4. Although conversation analysis (CA) and interaction analysis are separate research traditions
with very different methods and assumptions, Robinson (2011) argues that the two
approaches can form a symbiotic relationship. In an observational study of physician-patient
interviews, CA insights gleaned from close analysis of individual interactions have informed
development of traditional coding schemes, thereby contributing to validity. Traditional
coding methods have helped demonstrate that CA informed distinctions matter by
documenting their statistical association with outcomes (Robinson, 2011).
5. In practice, it can be near impossible to discern the difference between spontaneous
communication versus intentional manipulation of the same signals (i.e., pseudo-spontaneous
communication; Buck & VanLear, 2002). Regardless of their true origins, nonverbal signals
may be interpreted at the level of manifest content or symbolic meaning.
6. The same may be said for coding of relational control (see chapter this volume), which
begins with low-level inferences about the grammatical and pragmatic form of utterances but
aggregates specific codes into patterns of dominance and domineeringness.

30
References
Baek, Y.M., Cappella, J.N., & Bindman, A. (2011). Automating content analysis of open-ended
responses: Wordscores and affective intonation. Communication Methods and Measures,
5, 275-296.
Bakeman, R., & Gottman, J. M. (1997). Observing interaction: An introduction to sequential
analysis. (2nd ed.). New York: Cambridge University Press.
Bente, G., Senokozlieva, M., Pennig, S., Al-Issa, A., & Fischer, O. (2008). Deciphering the
secret code: A new methodology for the cross-cultural analysis of nonverbal behavior.
Behavior Research Methods, 40, 269-277.
Birchler, G.R., Weiss, R.L., & Vincent, J.P. (1975). Multimethod analysis of social
reinforcement exchange between maritally distressed and nondistressed spouse and
stranger dyads. Journal of Personality and Social Psychology, 31, 349-360.
Black, M.P., Katsamanis, A., Baucom, B.R., Lee, C. Lammert, A.C., Christensen, A., Georgiou,
P.G., & Naravanan, S.S. (2013). Toward automating a human behavioral coding system
for married couples’ interactions using speech acoustic features. Speech Communication,
55, 1-21.
Buck, R., & VanLear, C.A. (2002). Verbal and nonverbal communication: Distinguishing
symbolic, spontaneous, and pseudo-spontaneous nonverbal behavior. Journal of
Communication, 52, 522–541.
Caughlin, J.P., & Ramey, M.E. (2005). The demand/withdraw pattern of communication in
parent-adolescent dyads. Personal Relationships, 12, 339-355.

31
Cegala, D.J., McClure, L., Marinelli, T.M., & Post, D.M. (2000). The effects of communication
skills training on patients’ participation during medical interviews. Patient Education and
Counseling, 41, 209–222.
Cohn, J.F., & Sayette, M.A. (2010). Spontaneous facial expression in a small group can be
automatically measured: An initial demonstration. Behavior Research Methods, 42, 1079-
1086.
Davidson, D., MacGregor, M.W., MacLean, D.R., McDermott, N., Farquharson, J., & Chaplin,
W.F. (1996). Coder gender and potential for hostility ratings. Health Psychology, 15,
198-302.
Eldridge, K.A., & Christensen, A. (2002). Demand-withdraw communication during couple
communication: A review and analysis. In P. Noller & J. A. Feeney (Eds.),
Understanding marriage: Developments in the study of couple interaction (pp. 289-322).
New York: Cambridge University Press.
Erbert, L.A., & Duck, S.W. (1997). Rethinking satisfaction in personal relationships from a
dialectical perspective. In R. J. Sternberg & M. Hojjat (Eds.) Satisfaction in close
relationships (pp. 190-217). New York: Guilford Press.
Folger, J.P., Hewes, D.E., & Poole, M.S. (1984). Coding social interaction. In. B. Dervin & M.
Voight (Eds.). Progress in in communication sciences, Vol. 4 (pp. 115-161). Norwood,
NJ: Ablex.
Gagné, F.M., & Lydon, J.E. (2004). Bias and accuracy in close relationships: An integrative
review. Personality and Social Psychology Review, 8, 322-338.
Gill, D.S., Christensen, A., & Fincham, F.D. (1999). Predicting marital satisfaction from
behavior: Do all roads really lead to Rome? Personal Relationships, 6, 369-387.

32
Gottman , J.M. (1979 ). Marital interactions: Experimental investigations. New York, NY:
Academic Press.
Gottman, J.M. (1982). Temporal form: Toward a new language for describing relationships.
Journal of Marriage and the Family, 44, 943-962.
Gottman, J.M. (1994). What predicts divorce? The relationship between marital processes and
marital outcomes. Hillsdale, Erlbaum.
Gottman, J.M., & Levenson, R.W. (1986). Assessing the role of emotion in marriage.
Behavioral Assessment, 8, 31-48.
Guetzkow, H. (1950). Unitizing and categorizing problems in coding qualitative data. Journal of
Clinical Psychology, 6, 47-58.
Harris, F.C., & Lahey, B.B. (1982). Recording system bias in direct observational methodology.
Clinical Psychology Review, 2, 539-556.
Hahlweg, (2004), Kategoriensystem für partnerschaftliche interaktion (KPI): Interactional coding
system (ICS). In P.K. Kerig & D.H. Baucom (Eds.), Couple observational coding
systems (pp. 122-142). Mahwah, NJ: Erlbaum.
Heavey, C. L., Layne, C. & Christensen, A. (1993). Gender and conflict structure in marital
interaction: A replication and extension. Journal of Consulting and Clinical Psychology,
61, 16-27).
Heyman, R.E. (2001). Observation of couple conflicts: Clinical assessment applications,
stubborn truths, and shaky foundations. Psychological Assessment, 13, 5-35.
Heyman, R.E. (2004). Rapid Marital Interaction coding System (RMICS). In P.K. Kerig & D.H.
Baucom (Eds.), Couple observational coding systems (pp. 67-93). Mahwah, NJ: Erlbaum.

33
Heyman, R.E. Lorber M.F., Eddy J.M., & West T.V. (2014). Behavioral observation and coding.
In H. T. Reis, & C. M. Judd (Eds.), Handbook of research methods in social and
personality psychology (2nd Ed.) (pp. 343-370). New York: Cambridge University Press.
Heyman, R.E., Weiss, R. L., & Eddy, J. M. (1995). Marital Interaction Coding System: Revision
and empirical evaluation. Behaviour Research and Therapy, 33, 737-746.
Holsti, O.R. (1969). Content analysis for the social sciences and humanities. Reading, MA:
Addison-Wesley.
Jacobs, S. (2002). Language and interpersonal communication. In M.L. Knapp & J.A. Daly
(Eds.), Handbook of interpersonal communication (3rd Ed.) (p. 213-239). Thousand Oaks,
CA: Sage.
Julien, D., Markman, J.J., & Lindahl, K.M. (1989). A comparison of global and a microanalytic
coding system: Implication for future trends in studying interactions. Behavioral
Assessment, 11, 81-100.
Kerig, P.K. (2001). Introduction and overview: Conceptual issues in family observational
research. In P.K. Kerig & K.M. Lindahl (Eds.), Family observational coding systems:
Resources for systemic research (pp. 1-22). Mahwah, NJ: Lawrence Erlbaum Associates.
Kerig, P.K., & Lindahl, K.M. (Eds.) (2001). Family observational coding systems: Resources for
systemic research. Mahwah, NJ: Erlbaum.
Krippendorf, K. (2004). Content analysis: An Introduction to its methodology (2nd Ed.).
Thousand Oaks, CA: Sage.
Krippendorf, K. (2011). Agreement and information in the reliability of coding. Communication
Methods and Measures, 5, 93–112, 2011.

34
Krokoff, L.J., Gottman, J.M. & Hass, S.D. (1989). Validation of a global rapid couples
interaction scoring system. Behavioral Assessment, 11, 65-79.
Lindahl, K.M. (2001). Methodological issues in family observational research. In P.K. Kerig &
K.M. Lindahl (Eds.), Family observational coding systems: Resources for systemic
research (pp. 23-32). Mahwah, NJ: Lawrence Erlbaum Associates.
Lorber, M. F. (2006). Can minimally trained observers provide valid global ratings? Journal of
Family Psychology, 20, 335-338.
Margolin, G., Oliver, P.H., Gordis, E.B., O’Hearn, H.G., Medina, A., Ghosh, C.M., & Morland,
L. (1998). The nuts and bolts of behavioral observation of marital and family interaction.
Clinical Child and Family Psychology Review, 1, 195-213.
McNulty, J. K., & Russell, V. M. (2010). When “negative” behaviors are positive: A contextual
analysis of the long-term effects of problem-solving behaviors on changes in relationship
satisfaction. Journal of Personality and Social Psychology, 98, 587–604.
Notarius, C.I, Markman, H.J., & Gottman, J.M. (1983). Couples interaction scoring system:
Clinical applications. In E.E. Filsinger (Ed.), Marriage and family assessment (p. 117-
151). Beverly Hills, CA: Sage.
Overall, N.C., Fletcher, G. J. O., Simpson, J. A., & Sibley, C.G. (2009). Regulating partners in
intimate relationships: The costs and benefits of different communication strategies.
Journal of Personality and Social Psychology, 96, 620-639.
Overall, N.C., Simpson, J.A., & Struthers, H. (2013). Buffering attachment-related avoidance:
Softening emotional and behavioral defenses during conflict discussions. Journal of
Personality and Social Psychology, 105, 854-871.

35
Pellegrini, A.D., Bohn-Gettler, C.M., Dupuis, D., Hickey, M., Roseth, C., & Solberg, D. (2011).
An empirical examination of sex differences in scoring preschool children’s aggression.
Journal of Experimental Child Psychology, 109, 232-238.
Poole, M.S., Folger, J.P., & Hewes, D.E. (l987). Methods of interaction analysis. In G. R. Miller
and M. Roloff (Eds.), Interpersonal processes: New directions in communication
research (pp. 220-256). Beverly Hills: Sage.
Putnam, L.L., & Jones, T.S. (1982). Reciprocity in negotiations: An analysis of bargaining
interaction. Communication Monographs, 49, 171-191.
Robinson, J. D. (2011). Conversation analysis and health communication. In T.L. Thompson, A.,
R. Parrott, & J.F. Nussbaum (Eds.), The Routledge handbook of health communication
(2nd Ed.) (pp. 501-518). NY: Routledge.
Roggman, L.A., Cook, G.A., Innocenti, M.S., Norman, V. J., & Christiansen, K. (2013).
Parenting interactions with children: Checklist of observations linked to outcomes
(PICCOLO) in diverse ethnic groups. Infant Mental Health Journal, 34, 290-306.
Rusbult, C.E., Van Lange, P.A.M., Wildschut, T., Yovetich, N.A., & Verette, J. (2000).
Perceived superiority in close relationships: Why it exists and persists. Journal of
Personality and Social Psychology, 79, 521-545.
Sevier, M., Simpson, L.E., & Christensen, A. (2004). Observational coding of demand-withdraw
interactions in couples. In P. K. Kerig & D. H. Baucom (Eds.), Couple observational
coding systems (pp. 159-172). Mahwah, NJ: Erlbaum.
Sillars, A., & Canary, D.J. (2013). Conflict and relational quality in families. In A.L. Vangelisti
(Ed.), Routledge handbook of family communication (2nd ed.) (pp. 338-357). New York:
Routledge.

36
Sillars, A. L. (1986). Procedures for coding interpersonal conflict: The verbal tactics coding
scheme (VTCS). Unpublished coding manual, University of Montana, Missoula, MT.
Sillars, A.L., Pike, G.R., Jones, T.S. & Murphy, M. A. (1984). Communication and
understanding in marriage. Human Communication Research, 3, 317-350.
Sillars, A., & Vangelisti, A.L. (2006). Communication: Basic properties and their relevance to
relationship research. In A.L. Vangelisti & D. Perlman (Eds.), The Cambridge handbook
of personal relationships (pp. 331-351). New York: Cambridge University Press.
Sillars, A.L., & Wilmot, W.W. (1994). Communication strategies in conflict and mediation. In J.
Wiemann & J.A. Daly (Eds.), Strategic interpersonal communication (pp.163-190).
Hillsdale, NJ: Erlbaum.
Street, R.L., & Cappel1a, J.N. (1985). Sequence and pattern in communicative behavior: A
model and commentary. In R.L. Street, Jr., & J.N. Cappella (Eds.), Sequence and pattern
in communicative behavior (pp. 243-276). London: Edward Arnold.
Stone, A.L., Tai-Seale, M., Stults, C.D., Luiz, J.M., & Frankel, R.M. (2012). Three types of
ambiguity in coding empathic interactions in primary care visits: Implications for
research and practice. Patient Education and Counseling, 89, 63-68.
Surra, C.A., & Ridley, C.A. (1991). Multiple perspectives on interaction: Participants, peers, and
observers. In B. Montgomery & S. Duck (Eds.), Studying interpersonal interaction (pp.
35-55). New York: Guilford.
Vivian, D., Langhinrichsen-Rohling, J., & Heymay, R.E. (2004). The thematic coding of dyadic
interactions: Observing the context of couple conflict. In P.K. Kerig & D.H. Baucom
(Eds.), Couple observational coding systems (pp. 273-288). Mahwah, NJ: Erlbaum.

37
Watzlawick, P., Beavin, J., & Jackson, D. D. (1967). Pragmatics of human communication: A
study of interactional patterns, pathologies, and paradoxes. New York: Norton.

38
Table 1
Categorical Coding Systems for Couple Conflict
Marital Interaction Coding System (MICS-IV) (Heyman, Weiss, & Eddy, 1995)
Blame (criticize, mindread negative, putdown, turn-off)
Description (problem description, internal and external)
Dysphoric Affect
Facilitation (assent, disengage, humor, mindread positive, positive touch, paraphrase/reflect, question,
smile/laugh)
Invalidation (disagree, disapprove, deny responsibility, excuse, non-comply)
Irrelevant (unintelligible talk)
Propose Change (compromise, negative and positive solution)
Validation (agree, approve, accept responsibility, comply)
Withdrawal
Kategoriensystem für Partnerschaftliche Interaktion (KPI) (Hahlweg, 2004)
Positive Verbal
Self-disclosure (expression of feelings, wishes, attitudes or behavior)
Positive Solution (constructive proposal, compromise suggestions)
Acceptance of the Other (paraphrase, open question, positive feedback, understanding,
agreement)
Neutral Verbal
Problem Description (neutral description, neutral questions)
Meta Communication (clarifying requests, related to topic)
Rest (inaudible or does not fit other categories)
Listening
Negative Verbal
Criticize (devaluation of partner, specific criticism)
Negative solution (destructive solution, demand for omission)
Justification (excuse own behavior, deny responsibility)
Disagreement (direct disagreement, yes-but, short disagreement, blocking off)
Couples Interaction Scoring System (CISS) (Gottman, 1979)
Content Codes
Problem Information or Feelings about a Problem
Mindreading
Proposing a Solution
Communication Talk
Agreement
Disagreement
Summarizing Other
Summarizing Self
Nonverbal Behavior
Positive (face, voice, and body cues such as smiling, warm voice, touching)
Negative (face, voice, and body cues such as frown, cold voice, inattention)
Neutral (absence of positive or negative nonverbal cues)

39
Verbal Tactics Coding Scheme (VTCS) (Sillars, 1986)
Denial and Equivocation (direct or implicit denial, evasive replies)
Topic Management (topic shifts, topic avoidance)
Noncommittal Remarks (noncommittal statements and questions, abstract or procedural remarks)
Irreverent Remarks (friendly joking)
Analytic Remarks (descriptive, disclosive, or qualifying statements; soliciting disclosure or criticism)
Confrontative Remarks (personal criticism, rejection, hostile imperatives, jokes, or questions, presumptive
attribution, denial of responsibility)
Conciliatory Remarks (supportive remarks, concessions, acceptance of responsibility)

40
Table 2
Rating Systems for Couple Conflict
Conflict Rating System (CRS) (Heavey, Lane, & Christensen, 1993)
Demand Subscale
Discussion (tries to discuss a problem, is engaged and emotionally involved)
Blames (blames, accuses, or criticizes; uses sarcasm or character assassination)
Pressures for change (requests, demands, nags, or otherwise pressures)
Withdraw Subscale
Avoidance (hesitating, changing topics, diverting attention, or delaying discussion)
Withdraws (withdraws, becomes silent, refuses to discuss topic, looks away, disengages)
Positive Subscale
Negotiates (suggests solutions and compromises)
Backchannels (shows listening through positive minimal responses)
Validates Partner (indicates verbal understanding or acceptance of partner’s feelings)
Positive Affect (expresses caring, concern, humor, or appreciation)
Communicates Clearly (expresses self in a way that is easy to understand)
Negative Subscale
Expresses Critical Feelings (verbally expresses hurt, anger or sadness directed at partner)
Interrupts
Dominates Discussion (dominates, tries to take control of the discussion)
Negative Affect (verbal or nonverbal anger, frustration, hostility, hurt or sadness)
Communication Strategies Coding Scheme (CSCS) (Overall, Fletcher, Simpson, Sibley,
2009)
Negative-direct
Coercion (derogate partner, indicate negative consequences for partner, display negative affect,
accuse and blame partner)
Autocracy (insist or demand, talk from a position of authority, invalidate partner’s point of view,
take a domineering and/or non-negotiative stance)
Negative-indirect
Manipulation (attempt to make partner feel guilty, appeal to partner’s love and concern)
Supplication (use emotional expression of hurt, debase self and/or present self as needing help,
emphasize negative consequences for self)
Positive-direct
Rational Reasoning (use and seek accurate information, use logic and rational reasoning, explain
behavior or point of view in a way the partner would find reasonable)
Positive-indirect
Soft Positive (soften persuasion attempts, encourage partner to explain point of view and express
feelings, acknowledge and validate partner’s views, be charming and express positive
affect)