The Specific Affect Coding System (SPAFF)

Abstract

In 1989, Gottman and Krokoff introduced the Specific Af- fect Coding System (SPAFF) for the purpose of systemati- cally observing affective behavior in the context of marital conflict. The original SPAFF conferred a host of advantages over earlier “microanalytic” coding strategies, the primary innovation being the ability to code affect at the construct level instead of at the level of extremely discrete bits of be- havior, such as specific gestures or facial movements (Gott- man, McCoy, Coan, & Collier, 1995).

Full text

16
The Specific Affect Coding System (SPAFF)
James A. Coan
John M. Gottman
267
In 1989, Gottman and Krokoff introduced the Specific Af-
fect Coding System (SPAFF) for the purpose of systemati-
cally observing affective behavior in the context of marital
conflict. The original SPAFF conferred a host of advantages
over earlier “microanalytic” coding strategies, the primary
innovation being the ability to code affect at the construct
level instead of at the level of extremely discrete bits of be-
havior, such as specific gestures or facial movements (Gott-
man, McCoy, Coan, & Collier, 1995).
Since its debut, the SPAFF has, in one version or another,
informed dozens of published scientific findings deriving
from numerous laboratories (e.g., Burman, Margolin, & John,
1993; Carstensen, Gottman, & Levenson, 1995; Coan, Gott-
man, Babcock, & Jacobson, 1997; Cohan & Bradbury, 1994,
1997; Jacobson et al., 1994; Notarius, Benson, Sloane,
Vanzetti, & Hornyak, 1989). For example, the SPAFF has
been used to study affective behavior among newlyweds (Cohan
& Bradbury, 1997; Gottman, Coan, Carrere, & Swanson,
1998), domestically violent couples (Babcock, Waltz, Jacobson,
& Gottman, 1993; Coan et al., 1997), couples in long-term
marriages (Carstensen et al., 1995; Levenson, Carstensen, &
Gottman, 1994), and, most recently, gay and lesbian couples
(Gottman, Levenson, Gross, et al., 2003; Gottman, Levenson,
Swanson, et al., 2003). Although initially developed for the
study of emotional communication among romantic couples,
the SPAFF is now used for coding interactions among chil-
dren, their parents, and their peers (Joanne Wu Shortt, per-
sonal communication, April 9, 2002), and even to therapy
situations (Janine Giese-Davis, personal communication,
October 11, 2003). Indeed, individuals in applied settings
have expressed interest in learning the SPAFF (Coan, 1998).
In a recent review of observational couples research, Heyman
(2001) noted that the SPAFF has “by far the best evidence of
construct and criterion validity for its constructs” of all cur-
rent microanalytic coding systems (Heyman, 2001, p. 25).
History of the SPAFF
Early observational coding systems, such as the Marital In-
teraction Coding System (MICS; Hops, Wills, Weiss, &
Patterson, 1972), and the Facial Affect Scoring Technique
(FAST; Ekman, Friesen, & Tomkins, 1971), sought to iden-
tify extremely discrete bits of behavior that might prove useful
in understanding how such behaviors function in the con-
text of interpersonal relationships. Initially, Gottman fol-
lowed in this tradition with the development of the Couples
Interaction Scoring System (CISS; Gottman, 1979). CISS
coders were instructed to hierarchically scan behaviors for
specific cues, starting with the face, moving to the voice, and
finally coding body movements. Notably, the CISS ignored
verbal content altogether. A short while later, frustrated with
perceived inadequacies in the CISS, Gottman sought a revi-
sion of his system that was heavily influenced by the Facial
Action Coding System (FACS; Ekman & Friesen, 1978).
Nevertheless, Gottman’s frustrations mounted as his lists of

268 Emotion Assessment
discrete codable behaviors grew at an alarming rate. As he
recounted in his first published SPAFF manual:
I did not want my summary codes to read something
like: “Husband shows zygomatic major contracts on
face with contraction of the cheek raiser muscle, with
shift downward in fundamental frequency, decrease in
amplitude and voice in a major key and rapid inhala-
tion and exhalation of breath with hut hut vocaliza-
tions.” Instead, I wanted to say that the husband
laughed. (Gottman et al., 1995, p. 3)
The point was not to ignore the identification of zygo-
matic major contractions or modulations of frequency and
amplitude in vocal communication. Rather, the point was
that modern affect coding, although informed by a thor-
ough knowledge of discrete behaviors such as those de-
scribed by Ekman, Scherer, and others (e.g., Banse &
Scherer, 1996; Ekman & Friesen, 1975; Scherer, 1974),
often missed the forest for the trees. Gottman sought to
devise a coding system that made explicit use of discrete
bits of information in the service of describing constructs
representing generalizable human affective behavior. Thus
Gottman reintroduced verbal content to the specification of
those constructs.
Of course, previous microanalytic coding systems had
been used to cluster codes into “macrocodes” or constructs,
but these attempts had been empirical and, in any event, had
often been implemented after the coding had been com-
pleted. Gottman (Gottman et al., 1995) sought instead to let
his coders code theoretically specified constructs directly. This
new approach assumed that coders had, or could be taught,
the ability to integrate a variety of different affective cues into
broader constructs both rapidly and accurately. It was hoped
that such an approach would increase the speed with which
such coding could be done and also that it would render such
coding more reliable and externally valid.
It was from these efforts that the first major version of
the SPAFF was born. The original version of the SPAFF com-
prised 10 gestalt behavior codes: Neutral, Humor, Affection/
Caring, Interest/Curiosity, Anger, Disgust/Scorn/Contempt,
Whining, Sadness, and Fear. This was later expanded to a
second major version that comprised 16 such codes, adding
Surprise and Validation to the positive set and expanding the
negative set to include Belligerence, Dominance, Stonewall-
ing, and Defensiveness, as well as separating Disgust and
Contempt into distinct codes (Gottman et al., 1995). Since
the publication of the first SPAFF manual, the SPAFF has
been revised yet further. This chapter holds to a description
of the SPAFF in its most current form, with a full listing of
its revised and updated list of codes and their indicators. It
also includes advice for training SPAFF coders, for assessing
coding reliability, and for solving various data analytic issues.
Recent innovations in weighting SPAFF codes for use as a
continuous scale, as well as attempts to utilize SPAFF codes
as separate continuous variables, are also described.
Learning to Code Behavior: The Philosophy
of the SPAFF
Among the core ideas underlying the SPAFF is the uncon-
troversial notion that emotions are expressed in a wide vari-
ety of ways and that this variety should be respected. If there
is a second major idea, it is that SPAFF coding requires the
use of human beings with a personal history of interpersonal,
affective communication. Such a personal history provides
access to subtle cues that even many years of strict training
in the identification of discrete physical features may neglect.
Thus learning to observe emotional behavior means, on the
one hand, learning to identify multiple discrete indicators,
any one of which may or may not be present during a par-
ticular emotional episode, and, on the other hand, drawing
from one’s own personal history of affective communication
in order to spot the complexities of behavior that remain
outside the grasp of highly detailed discrete analysis. SPAFF
coding means learning to integrate voice, physical features,
verbal content, and more—indicators that are sometimes
hard to describe (e.g., “positive energy”) but that are easily
grasped by most coders.
SPAFF Codes Are Latent Psychological Constructs
Figure 16.1 depicts, for the purpose of illustration,
1
a latent
variable model (cf. Bollen, 2002) representing the SPAFF code
Validation. In this model, the core, latent construct Valida-
tion (represented in the oval) is not directly observable.
Rather, it is assumed to exist and to actually cause the ex-
pression of its various observable indicators (represented as
rectangles). One would not be able to “see” Validation with-
out directly observing at least one of its indicators. An indi-
cator is an objective piece of evidence that any observer can
see or hear directly. It is called an indicator because when it
is present, it literally “indicates” the underlying construct we
are interested in—it tells us that our latent construct is hap-
pening. Importantly, we are rarely interested in any one of
the indicators of Validation per se. Rather, we are interested
in the construct that those indicators indicate. Put another
way, it is of little specific consequence to us as SPAFF cod-
ers whether we observe direct statements of agreement or
apology, whether we observe summarizing behaviors, or
whether we observe head-nodding behavior with eye contact.
These bits of observation are merely the media through which
we become aware of the thing we are really interested in, which
is Validation. We cannot “see” Validation without the presence
of one or more or its indicators, but without the construct of
Validation, those indicators are by themselves of little theo-
retical value. This is true even when discrete and easily iden-
tified behaviors wind up predicting important outcomes, such
as happens, for example, in the association between the “eye
roll” and marital dissolution. The importance of a discrete
behavior such as the eye roll lies in its connection to the con-
struct of contempt (cf. Gottman, 1993b).

The Specific Affect Coding System (SPAFF) 269
Physical Features and Cultural Informants
In the language and history of behavior coding, at least two
broad approaches can be identified. These are the physical
features and cultural informant approaches. Physical-features
approaches hold strictly to the detailed description of physi-
cal observables, such as changes in vocal acoustic properties,
facial expressions, specific gestures, and body postures. In
theory, anyone, or nearly anyone, can be taught to be a physi-
cal-features coder; even computers are now becoming ca-
pable of doing so (Xiao, Moriyama, Kanade, & Cohn, 2003).
By contrast, the cultural-informant approach utilizes in-
dividuals who are, for one reason or another, sensitive ob-
servers in a specific cultural setting. They may, for example,
have specific knowledge about a certain culture or group of
cultures and, by virtue of this specific knowledge, be uniquely
capable of decoding the meaning of specific behaviors within
the context of that culture. Anthropologists have employed
cultural informants to study cultures with which they were
not intuitively familiar. Cultural informants aid researchers
in the interpretation of specific observable events.
In SPAFF coding, both physical-features and cultural-
informant approaches are utilized in the service of captur-
ing meaningful affective constructs. In training SPAFF coders,
the assumption is made that most individuals will be sensi-
tive to subtle differences between certain instances of ob-
servable events. Take, for example, an instance in which a
young woman gently rubs her cheek on her shoulder while
making eye contact with her partner. A purely physical-
features approach might note her cheek-shoulder distance
and append that particular unit of distance to a tally of other
such distances for later analysis. Such units of distance
would certainly constitute a kind of information about this
woman. Indeed (as was noted earlier), a computer could
very probably do this kind of coding. However, it is prob-
ably still true that only a live human being with a lifetime
of experience observing people interacting with each other
would be able to distinguish between a cheek rubbed on
the shoulder as part of an emotional display (of, say, coy
affection) versus a cheek rubbed on the shoulder to relieve
an itch.
Such a large and complex number of indicators are
needed to distinguish these two possibilities that computer
algorithms, as sophisticated as they are becoming, are prob-
ably still many years from being able to do so. On the other
hand, a human being—that is, a cultural informant—could
reliably note the difference in an instant, and the difference
is likely to be a meaningful one.
Becoming a Cultural Informant:
Seeing Versus Observing
Perhaps the most recognizable quotation of Sir Arthur Conan
Doyle’s literary creation Sherlock Holmes is “you saw, but
you did not observe.” The line was used to chastise the af-
fable and earnest Dr. Watson, but it applies equally well to
most of us, most of the time. We usually attend to only some
fraction of the sensory information that is available to us.
Most of our time is spent reflexively responding to a kind of
social and affective rhythm, to the pitch and meter of con-
versation, to the style of the clothes people are wearing, to
the moods implied in the ways people carry themselves and
the obvious components of the looks on their faces. All of
these bits of information influence our behaviors, and our
behaviors in turn influence the individuals who are influenc-
ing us. Presumably, our brains “see” all of this information
as it occurs, at least in the sense that the information is caus-
ing us to respond in certain ways, but under ordinary cir-
cumstances, we do not reflect on the information explicitly.
Becoming an effective cultural informant means learning
to be an active observer of the kinds of information just de-
scribed. The first step toward becoming an active observer
simply involves learning to be mindful of the information that
is available at any one moment. For example, when convers-
ing with an office mate, it may be useful to pay close atten-
tion to how he expresses his feelings with his face and to note
what he’s wearing and how he typically carries himself. Such
exercises prompt a number of interesting questions. What,
for example, might this person be intending to communi-
cate other than what he is explicitly saying? Perhaps one
would conclude that he is interested in portraying himself
as a serious and highly skilled worker but also a fun-loving
and adventuresome person in other contexts. Fair enough.
But what is he actually doing to convey this information? In
other words, after we have come to some tentative conclu-
sions about who this person is and what he is trying to tell
us in subtle ways about himself, we can ask the next ques-
tion: How do we know? By doing so, we are already well on
the way to becoming active observers.
Three Rules of People Watching
We have found that active observation can be facilitated with
some simple exercises that we refer to as the rules of people
watching (Gottman et al., 1995). These rules are predicated
on the idea that when people behave in certain characteris-
Validation
Agreement/
apology
Summarizing
Head nodding
/eye contact
Figure 16.1. The SPAFF code Validation represented as a
latent construct.

270 Emotion Assessment
tic ways, or even in ways that seem specific to certain situa-
tions, they are doing so to portray themselves as a kind of char-
acter. Further, they select behaviors for this purpose from a
set of possible choices of how to act at any one moment. That said,
it is important to recognize that we are not asserting that this
is actually what people are doing most of the time, though it
might be. Rather, our intention is to provide the reader with
ways to approach the problem of people watching so that
he or she can become a more sensitive and more accurate
cultural informant. In other words, what follows are not the
rules of behavior but the rules of the active observation of
behavior. They are rules for becoming an active observer.
Rule 1: View a Behavior as Though It Were Chosen
From a Collection of Possible Alternatives
If one imagines two different people with precisely the same
mild disability, a limp for example, it is possible to imagine
further that each individual will behave differently with re-
gard to the limp he or she copes with. One of them may, for
example, work to minimize the extent of the limp through a
variety of movements designed to keep others from seeing
it. Another may exaggerate the limp, forming it into a kind
of a swagger. The obvious point is that either of them could
have chosen either approach to dealing with the limp, and
there may indeed be other approaches as well. There are
multiple options for incorporating the disability into their
day-to-day behaviors. The other point, however, and perhaps
the less obvious one, is that these options could be thought
of as alternative styles of having a mild disability.
One can imagine a multitude of behavioral styles in a
variety of contexts. People watching at parties can be par-
ticularly useful for this. It is informative to note the variety
of dress, the different kinds of laughter people use, the in-
tensity of the smiles one observes, the degree of physical space
that people maintain, and so forth. Any of these dimensions
of behavior could, in theory, be selected by any of the people
in the room. And yet, certain people “select” only certain
behaviors. The question is why certain people select certain
behaviors, and that leads to rule 2.
Rule 2: View Behavior as if It Were Designed to Portray
a Character in a Play or a Film—as if It Were Written
to Follow a Script
As you watch people at our hypothetical party, it is possible
to image that everyone’s role has been scripted and that one
is actually observing actors working to portray certain char-
acters. When actors begin preparing characters, they are fre-
quently given a number of character attributes that they must
then devise ways of communicating to an audience. Thus,
when an individual at a party begins laughing loudly and
becoming very animated, one might ask, What is it that the
actor is trying to portray about that character? He or she may,
for example, be attempting to communicate that he or she is
uninhibited, spontaneous, and warm. When meeting a new
individual at a party, or indeed in any setting, it can be very
useful to ask the question, I wonder how this person is going to
communicate what he or she is like?
Rule 3: Watch a Person as if You Were an Actor
Who Had to Play That Person in a Film
It is instructive to ask oneself what kinds of behaviors would
be necessary to portray any individual being that one is ob-
serving. This exercise can be as simple, at first, as just try-
ing to mirror his or her behavior. Frequently, your body
will know what to do if you just try to mimic someone. In-
deed, research suggests that mirroring behavior can enhance
one’s ability to code it (e.g., Blairy, Herrera, & Hess, 1999).
Mimic the life of the party and contrast that with the feel-
ing of mimicking the person who seemed to spend the
evening in relative seclusion. Each exercise in imitation
reveals a set of feelings, specific movements, props, and even
attitudes that can assist in the development of active ob-
servation. In practice, the three rules of people watching
can begin the process of becoming comfortable with, and
deliberate about, observing rather than merely seeing. But
when it comes to observational coding, being comfortable
with and deliberate about observing people is only the first
step.
Facial Expressions of the SPAFF
As noted in some detail previously, the SPAFF is not a strictly
physical-features-based coding system. Nevertheless, more
than any other such system, the SPAFF has been heavily in-
fluenced by, and indeed incorporates, the Facial Action Cod-
ing System of Paul Ekman and Wallace Friesen (Cohn,
Amdadar & Ekman, chapter 13, this volume; Ekman &
Friesen, 1978). We regard the FACS as the state of the art
in facial coding, and in this section we describe the move-
ments that are regarded as essential to understanding many
SPAFF codes. That said, there are notable departures from
official FACS protocol in the SPAFF, not the least of which
is that the intensity levels of FACS codes are more or less
ignored in favor of coding specific FACS codes as either
present at any level of intensity or not present. Further, for
the sake of brevity and specificity, many FACS codes are not
included in the SPAFF at all.
Specific facial movements are covered in detail elsewhere
in this volume (Cohn, et al., chapter 13, this volume). Thus
we refer the reader to the chapter by Cohn, et al., and, in-
deed, to the FACS manual itself for detailed descriptions of
the action units (AUs) described in this and subsequent sec-
tions. Figure 16.2 depicts a selection of common facial ex-
pressions associated with the SPAFF.
Action Units of the Upper Face
• Action Unit 1 (AU1)—The Inner Brow Raiser. (We
sometimes refer to this as the Woody Allen.)

The Specific Affect Coding System (SPAFF) 271
Figure 16.2. In this figure of the common facial expressions of the SPAFF, arrows highlight the
major action units (AUs) involved in the different expressions. Note that although AUs are
highlighted only once each, several occur in more than one expression.

272 Emotion Assessment
• Action Unit 2 (AU2)—The Outer Brow Raiser. (We
frequently refer to the bilateral manifestation of this
movement as the horns.)
• Action Unit 4 (AU4)—The Brow Lowerer.
• Action Unit 5 (AU5)—The Upper Lid Raiser.
• Action Unit 6 (AU6)—The Cheek Raiser and Lid
Compressor.
• Action Unit 7 (AU7)—The Lid Tightener. (We some-
times refer to this movement as the Clint Eastwood.)
• Action Unit 9 (AU9)—The Nose Wrinkler.
Action Units of the Lower Face
• Action Unit 10 (AU10)—The Upper Lip Raiser.
• Action Unit 12 (AU12)—The Lip Corner Puller.
• Action Unit 14 (AU14)—The Dimpler.
• Action Unit 15 (AU15)—The Lip Corner Depressor.
• Action Unit 17 (AU17)—The Chin Raiser.
• Action Unit 20 (AU20)—The Lip Stretcher.
• Action Unit 25/26 (AU25/26)—Lips Part/Jaw Drop
The facial action units detailed in the preceding lists may
occur singly or in combination. Moreover, some action units,
such as AU14, which is heavily implicated in the SPAFF code
Contempt, frequently manifest on one side of the face only.
Codes of the SPAFF
As described earlier, coding the SPAFF requires that atten-
tion be paid to verbal content, facial behaviors, voice tones,
and other forms of communication. What follows are de-
tailed description of the codes that make up the current
version of the SPAFF (see Table 16.1). Descriptions include
subsections that detail the function of the code in interper-
sonal communication, various indicators of the code, physi-
cal cues for the code, and specific counterindicators regard-
ing the code. Indicators and physical cues provide infor-
mation about behaviors that probably derive from the
presence of the code, whereas counterindicators provide
information about behaviors that probably do not derive from
the presence of the code. Throughout these descriptions,
reference is made to speakers and receivers. Speakers are
those who are observed using the code, and receivers are
those the speakers are speaking to.
Affection
Function
Affection expresses genuine caring and concern and offers
comfort. Often the voice slows and becomes quieter or lower.
Its function is to facilitate closeness and bonding.
Indicators
1. Reminiscing. The speaker shares warm memories of
something she and the receiver enjoyed together.
2. Caring statements. Direct statements of affection or
concern, such as “I love you,” “I care about you,” “I
worry about you,” and so forth.
3. Compliments. Statements that communicate pride in or
admiration of one’s partner (e.g., “you are so smart!” or
“you did such a great job with the . . .”).
4. Empathy. Empathizing individuals mirror the affect of
their partners. Such mirroring need not be verbal, but
however it is expressed, it should be obvious that the
intent of the mirroring is to express an understanding
of the partner’s feelings. Importantly, empathy does
more than simply validate the partner’s thoughts and
feelings—by mirroring the affect of the partner at the
same time, it conveys a level of care that surpasses
validation per se.
5. The common cause. An important indicator of Affection,
similar to empathy, is the common cause, whereby
individuals engage in virtually any affective behavior
together as a form of building trust, closeness, consen-
sus, or bonding. This indicator can sometimes be
confusing. Insults, such as remarking that “Bob is a
jerk,” can be coded Affection if intended to express
obvious agreement. A shared anger, a shared fear, a
shared and vocalized political opinion—all of these
things could be coded Affection.
6. Flirting. When individuals flirt, they are communicat-
ing desire for their partners. The verbal expression
would be “I want you,” but flirting needn’t be verbal.
Flirting can be playful, sweet, warm, intense, or all of
the these.
Physical Cues
There are no particular AUs that indicate affection, but AUs
6 + 12 will commonly be seen.
Table 16.1
Current Codes of the SPAFF
Positive Affects Negative Affects
Affection Anger
Enthusiasm Belligerence
Humor Contempt
Interest Criticism
Validation Defensiveness
Disgust
Domineering
Fear / Tension
Sadness
Stonewalling
Threats
Whining
Neutral

The Specific Affect Coding System (SPAFF) 273
Counterindicators
• Defensive affection. Occasionally, a speaker will insist
that he loves the receiver as a defensive maneuver. The
indicators of defensiveness (discussed later) will
usually give this away. Watch for defensive voice tone,
a defensive context, and a lack of warm, positive
feeling underlying the affectionate message.
Anger
Function
In the SPAFF, anger functions to respond to perceived vio-
lations of the speaker’s rights to autonomy and respect. It
serves as a kind of “affective underlining” of displeasure and
complaint, indicating that an interpersonal boundary has
been transgressed. Some SPAFF coders have called the SPAFF
code of Anger “angry affect without belligerence, contempt,
defensiveness, disgust or attempts to dominate.” This is largely
true.
Indicators
1. Frustration. A relatively low intensity form of Anger,
here facial expressions of anger become apparent
at low levels and the voice may lower in pitch
and tempo. The anger will appear constrained or
out of the obvious awareness of the speaker. Other-
wise, the person may not express anger verbally
at all.
2. Angry “I-statements.” These are verbal statements that
express personal feelings, as in “I am so angry!” or “I
am so frustrated right now!”
3. Angry questions. Questions asked with angry affect and
usually with sharp exhalations, as in “Why?!”
4. Commands. Commands are not attempts to dominate
but rather are strong, affectively intense attempts to
stop a recent or ongoing violation of the speaker’s
autonomy or dignity. Sharp exhalations and strong
angry affect frequently accompany commands.
Examples include “Stop!” or “Don’t speak to me like
I’m a child!”
Physical Cues
AUs 4, 5, 7, 4+5, 4+5+7, 23, 24. The lips will frequently thin,
with the red of the upper lip disappearing or the lips pressed
together; the teeth will clench; and the muscles of the jaw
and neck will tighten. The voice may suddenly increase in
pitch, amplitude, and tempo and may include a kind of “growl”
as when yelling.
Counterindicators
• Blends with other codes. Angry affect is frequently
observed during moments in which indicators of other
negative codes are present. In these instances, Anger is
never coded.
Belligerence
Function
The function of Belligerence is to “get a rise” out of the re-
ceiver through provocation of anger. The belligerent speaker
is, in a sense, looking for a fight.
Indicators
1. Taunting questions. These are questions whose function
is to irritate or confuse the receiver. An example might
include the frequent and irritating use of the question
“Why?” in the context of a serious discussion. Fre-
quently the belligerent speaker is seen struggling to
suppress a smirk while asking taunting questions as
the receiver becomes increasingly enraged.
2. Unreciprocated humor. Sometimes, the belligerent
speaker appears to actually believe he or she is being
funny, even though the receiver is obviously annoyed.
Such moments of unreciprocated humor are neither
playful, fun, and shared (as in humor) nor sarcastic,
mocking, and insulting (as in contempt). Belligerent
speakers do not appear to get the message that the
humor is not universally funny, or the fact that the
jokes are annoying the receiver may increase the level
of humor experienced by the speaker.
3. Interpersonal terrorism. Here, the belligerent speaker is
posing direct challenges to the agreed-on rules or
boundaries of the relationship. Frequently, such
behavior takes the form of a dare, as in “What would
you do if I did?” or “What are you going to do about
it?” It can also be accompanied by a kind of emotional
“strutting,” whereby the belligerent person will make
use of loud commands such as “Don’t interrupt me!”
as a means of demonstrating his or her power. This is
often seen in violent men as a vestigial reminder of
how dangerous they can be.
Physical Cues
AUs 1 or 2. Jaw thrust forward.
Counterindicators
1. Good-natured teasing. Good-natured “jabs” at the
receiver’s foibles are not coded as belligerence,
especially if the humor or the teasing appears to be
shared.
2. Hostile humor. Unreciprocated humor that is obviously
hostile, mocking, belittling, or insulting is coded
Contempt.
Contempt
Function
The function of Contemptuous behavior is to belittle, hurt,
or humiliate. Contempt can be any statement made from a

274 Emotion Assessment
superior position to the partner, such as correcting an angry
person’s grammar. Such behavior deliberately and forth-
rightly communicates an icy lack of respect, often cruelty.
On theoretical and empirical grounds, we regard this behav-
ior as extremely detrimental to interpersonal relationships
(Coan et al., 1997; Gottman, 1993a; Gottman et al., 1998;
Gottman & Levenson, 1992), and so the SPAFF gives it pre-
cedence over most other behaviors.
Indicators
1. Sarcasm. Sarcasm in conversation frequently precedes
derisive laughter at the receiver’s expense or manifests
as a ridiculing comment regarding something the
receiver has said. Frequent examples include the ironic
use of such statements as “sure!” or “I’ll bet you did!”
2. Mockery. When speakers mock, they repeat something
the receiver has said while exaggeratedly imitating the
receiver’s manner of speech or emotional state for the
purpose of making the receiver look ridiculous or
stupid.
3. Insults. Insults are active and straightforward forms of
contempt—they are shows of disrespect for the
receiver through obvious verbal cruelty.
4. Hostile humor. Often, the contemptuous speaker uses a
form of unshared humor that, though an apparent
joke, utilizes sarcasm, mocking, or insults to achieve
the aim of contempt. By delivering such messages as a
“joke,” the speaker may be attempting to leave him- or
herself an “out” (as in, “hey, I was only joking”).
Hostile humor can be momentarily confusing for
coders and receivers alike. The contemptuous speaker
may laugh heartily, and sometimes the receiver will
briefly and reflexively laugh along. Such moments are
not coded as Humor.
Physical Cues
AU 14 (uni- or bilateral). Note: Eye rolls are nearly always coded
as contempt.
Counterindicators
• Good-natured teasing. Good-natured “jabs” at the
receiver’s foibles are not coded as contempt. A good
indication that contempt is not occurring is that the
context of the conversation appears to contradict
contemptuous intentions or that the speaker and
receiver appear to both experience laughter and joy as
a result of the teasing.
Criticism
Function
Criticism functions as an attack on someone’s character or
personality in a way that is not obviously insulting, as in
Contempt. It is a complaint that suggests that the partner’s
personality is defective. It is often accompanied by blame and
is quite distinct from complaining.
2
Complaints refer to spe-
cific instances of behavior, whereas Criticisms are character-
ized by negative global assessments of a person’s abilities or
value as a person. Complaints accompanied by “you always”
or “you never” statements are considered criticisms. Criticism
may or may not make reference to a specific event.
Indicators
1. Blaming. In blaming, one individual assigns fault to
another, along with a personal attack or global
accusation, as in “the reason the engine blew up is that
you never put oil in it.”
2. Character attacks. Often expressed as “you never/you
always” generalizations, character attacks are critical of
a person’s personality or abilities in very general ways.
Examples include statements such as “you don’t care,”
“you always put yourself first,” and so forth.
3. Kitchen sinking. This is essentially a long list of com-
plaints. Even though any particular item on the list
may not fit criteria for Criticism per se, a long list
functions to illustrate the incompetence or personality
defects of the person on the receiving end. For
example, an individual might “kitchen sink” using
complaints and “I” statements, such as, “I don’t feel
listened to by you, and you don’t touch me very often,
and I asked you to do certain chores, but you didn’t,
and we don’t do very many fun things together lately.”
4. Betrayal statements. Similar to blaming, betrayal
statements specifically reference trust and commit-
ment, implying that the person on the receiving end is
either not committed, untrustworthy, or both. “How
could you?” is a question frequently indicative of
Criticism.
5. Negative mind reading. Generally speaking, mind-
reading statements express attributions about
another’s feelings, behaviors, or motives. They indicate
Criticism when negative or accompanied by negative
affect. An example of negative mind reading would be
“you just don’t like Tom because he smokes.”
Physical Cues
There are no particular AUs that indicate Criticism.
Counterindicators
• Insults. Critical statements designed to inflict gratu-
itous emotional pain (e.g., “you’re an idiot”) are coded
contempt.
Defensiveness
Function
Defensiveness functions to deflect responsibility or blame.
It communicates a kind of innocent victimhood or righteous

The Specific Affect Coding System (SPAFF) 275
indignation (e.g., as a counterattack) on the part of the speaker,
implying that whatever bad thing being discussed is not the
speaker’s fault. Defensive speakers can engage in defending
themselves or friends and loved ones who may be under at-
tack by their partners.
Indicators
1. The “yes-but.” SPAFF coders refer to statements that
start off as momentary agreements but very quickly
end in disagreements as “yes-buts.” They are common
indicators of defensiveness.
2. Cross-complaining. This behavior involves meeting one
complaint with an immediate countercomplaint. In
this way, complaints are simply not responded to—
cross-complaints deflect them by leading the conversa-
tion into a suddenly new direction.
3. Minimization. Defensive speakers will frequently try to
minimize a complaint by asserting that the problem
they are potentially responsible for was scarcely a
problem in the first place. A minimizing speaker might
say, for example, “You’re right, I did forget to put the
garbage out, but there was hardly any garbage anyway,
so it really isn’t a problem. It can wait until next week.”
4. Excuses. Excuses are attempts to locate responsibility
or blame in something other than the speaker, as in,
“well, traffic was all backed up, there was nothing I
could do.”
5. Aggressive defenses. Oftentimes a speaker will aggres-
sively assert things, for example, “I did not!” These are
vehement denials of responsibility that come across as
childish, as in “did not/did too” interactions.
Physical Cues
AUs 1, 2, 1 + 2, arms folded across chest. The voice will in-
crease in pitch and amplitude.
Counterindicators
• Invalidations. Statements designed to directly contra-
dict the receiver (e.g., “you are wrong” or “that’s
simply untrue”), spoken in a lower pitched voice tone,
are more properly coded Domineering.
Disgust
Function
Disgust is a relatively involuntary verbal or nonverbal reac-
tion to a stimulus that is perceived to be noxious. Harmful
substances (e.g., feces, rotted food) reliably elicit disgust, but
disgust can also occur for moral or symbolic reasons (Rozin,
Lowery, & Ebert, 1994).
Indicators
1. Involuntary revulsion. Here the object of disgust is some
obvious image of, or reference to, an aversive, noxious
stimulus, as in momentary descriptions of a gruesome
physical injury.
2. Moral objection. Here the object of disgust is an action
or idea that the speaker finds repulsive for moral or
other symbolic reasons, as in responses to undesirable
sexual practices or even political positions.
Physical Cues
The physical cues of Disgust are robust and specific. AUs 9,
10, 4, 15, and 17 can sometimes be seen, either singly or in
any combination. The tongue will sometimes protrude, and
the head will sometimes turn to one side as if avoiding the
noxious stimulus.
Counterindicators
1. Mockery, insults, or belittlement. If the function of a
disgust response, whether verbal or nonverbal, appears
to be to communicate obvious disrespect of the
receiver, it is more properly coded as Contempt. This
includes instances in which the speaker appears to be
disgusted by the behavior of the receiver.
2. Disapproval without Disgust affect. Disapproval, absent
other obvious signs of disgust, can be coded Neutral
(when lacking in obvious affective tone), Domineering
(when spoken in a patronizing tone), or Anger (with
angry affect).
Domineering
Function
The function of Domineering behavior is to exert and dem-
onstrate control over one’s partner or a conversation. Domi-
neering behaviors attempt to impose compliance on the
receiver’s responses or behaviors.
Indicators
1. Invalidation. Invalidation deliberately and forcefully
contradicts the validity of the receiver’s point of view
(e.g., “that’s just wrong”) or expressed feelings (e.g.,
“oh, you are not afraid, quit exaggerating”).
2. Lecturing and patronizing. This indicator identifies
attempts to belittle or disempower a person or a
person’s arguments. Many “subindicators” suggest the
presence of lecturing and patronizing, including
pointing or wagging a finger while talking, citing
authorities (e.g., “well, Dr. Phil says . . .”), speaking in
platitudes and clichés, appealing to an ambiguous
“everyone” (as in “everyone knows”), and so forth. A
distinctly patronizing quality often accompanies these
behaviors. Look for finger pointing used for emphasis.
3. Low balling. Low balling expresses itself in the form of
questions that have predetermined answers. The
questions are not merely rhetorical but also have a
manipulative quality, such as, “You want me to be

276 Emotion Assessment
happy, don’t you?” Low-balling behaviors are similar
to sales ploys that seek to force unwary customers to
answer “yes” to very simple questions (e.g., “Do you
want your children to achieve their potential?”) in
order to manipulate them into purchasing a product.
4. Incessant speech. By using incessant speech, domineer-
ing persons can ensure that the receiver is not allowed
an opportunity to respond. It is a form of forcibly
maintaining the floor in a conversation at all times.
Incessant speech often has a repetitious, steady, almost
rhythmic quality in the voice. When speaking inces-
santly, domineering persons often repeat or summarize
their point of view while paying very little attention to
the verbal content of things said by the people with
whom they are speaking. Look for finger pointing used
for emphasis.
5. Glowering. Glowering is really a kind of steady gaze,
often characterized by the head tilted forward with the
chin down, and the outer portions of the eyebrows
raised—an eyebrow configuration we refer to as “the
horns” because, when configured in this way, the
eyebrows do indeed resemble horns. Thus, when
glowering, the “horns” are emphasized, and the person
may be leaning the head, body, or both forward.
Physical Cues
AU 2 (“the horns”), head forward, body forward, finger point-
ing, head cocked to one side.
Counterindicators
• Contemptuous patronizing. Whenever the content of
patronizing becomes blatantly insulting, it should be
coded Contempt.
Enthusiasm (Formerly Joy)
Function
The function of enthusiasm is to express a passionate inter-
est in a person or activity, as well as a positive valence asso-
ciated with that interest. Enthusiasm is infectious and often
sudden, loud, boisterous, and energetic. Nonverbal behav-
iors prominently accompany verbal expressions of eagerness
and joy.
Indicators
1. Anticipation. Anticipatory behaviors are hopeful,
future-oriented, and often childlike. They may be
accompanied by fidgeting and distraction.
2. Positive surprise. This is an emphatically happy reaction
to some unanticipated event or remark. Prominent
smiles and loud verbalizations characterize this
indicator (e.g., AU 1+2+6+12+24, accompanied by
“Really!?”)
3. Positive excitement. Similar to positive surprise, positive
excitement includes expressions of joy and anticipa-
tion at very high levels of intensity.
4. Joy. Joyful moments reflect high levels of often
suddenly felt happiness, similar to positive surprise
but less intense. Joy will frequently follow receipt of
a compliment and will often be accompanied by
broad, warm smiles and bright, alert, positive facial
expressions.
5. Expansiveness. Expansive individuals feel creative,
motivated, and inspired and convey an effervescent
and elated affect.
Physical Cues
AUs 1+2, 5, 6+12, 23, 24, 25–27 will commonly be seen.
Individuals will sometimes sit up or forward in their chairs,
and their voices will increase in pitch and volume.
Counterindicators
• Interest indicators. Enthusiasm can sometimes look like
Interest and vice versa. Interested questions are
accompanied by positive affect but of a lower intensity
than those coded Enthusiasm.
• Negative Surprise. Surprise reactions are not unequivo-
cally positive, and it is important to be watchful for
surprise reactions that contain either a lack of positive
affect or the presence of negative affect.
Fear/Tension
Function
Fear/Tension communicates, usually involuntarily, fear, worry,
anxiety, nervous anticipation, or dread.
Indicators
1. Speech disturbances. Fearful or tense speakers will often
have a difficult time expressing or even knowing what
they want to say. This will manifest as incomplete or
unfinished statements, stuttering, or frequent and rapid
“uhs” and “ahs.” Watch also for shallow, rapid breath-
ing. (Note that the occasional use of “ah, “er,” or
“umm” can simply reflect attempts to keep the floor or
turn at speech.)
2. Shifts in fundamental frequency. In studies of vocal
quality, chest register refers to a lower pitch character-
ized by vibratory sensations felt in the sternum and
trachea, and head register refers to a higher pitch
characterized by vibratory sensations felt in the head.
Either of these states can characterize a fundamental
frequency, or the lowest frequency, of sound waves
characterizing a person’s speech. In fear/tension, one
can often detect a shift in fundamental frequency that
moves from a chest register to a head register.

The Specific Affect Coding System (SPAFF) 277
3. Fidgeting. Fearful or tense individuals will fidget,
repeatedly shifting their position in their chairs (as if in
the “hot seat”), plucking at clothes or hands, rubbing
their faces (especially the temple, mouth, and chin), or
biting the lips or inside of their mouths.
4. Nervous laughter. Unshared laughter or giggling that
doesn’t appear to fit in the conversation and likely is a
response to nervous tension (e.g., no jokes or humor-
ous moments have occurred). Often, the fearful or
tense individual will seem unable to stop. The smile
will often appear “pasted on” (see “Physical Cues”).
5. Nervous gestures. Certain gestures of the arms and face
can indicate fear/tension, such as arms akimbo (folded
across the chest) and hands frequently touching the
face.
Physical Cues
AUs 1, 2, 4, 12, 20, 1+2+4, 1+2+4+5. Watch for frequent
eye movements, frequent gulping, biting of lips and inside
of mouth, and the “unfelt smile,” a smile without AU6 that
has been associated with neurophysiological patterns sug-
gestive of behavioral withdrawal (Ekman & Davidson, 1993;
Ekman, Davidson, & Friesen, 1990).
Counterindicators
1. Away behaviors. Away behaviors, such as paying
attention to trivial objects in the room, looking at one’s
own hands or nails, and so forth, when unaccompa-
nied by anxious affect and when in the context of high
negative affect, are more properly coded as Stonewall-
ing.
2. Foreign object. Sometimes individuals will become
occupied with picking their teeth or removing some-
thing from their eye in the midst of a conversation.
Such behaviors may be associated with increased
anxiety but are more likely simply Neutral.
3. Shared nervous laughter. Nervous laughter that is
shared among two or more individuals can quickly
escalate into a shared moment of positive affect that is
more properly coded as Humor.
Humor
Function
The function of humor is to share in mutual amusement and
joy following a mutually recognized moment of absurdity or
fun. Humor is relatively unique within the SPAFF in that it
cannot be coded in isolation. The humor code requires a
moment of shared amusement.
Indicators
1. Good-natured teasing. When an individual teases, she
highlights qualities or behaviors in her partner that
both agree are somewhat ridiculous, cute, or otherwise
funny.
2. Wit and silliness. Wit is expressed as an apt or clever
observation that is considered by both individuals to
be humorous. This could manifest as a funny observa-
tion or the straightforward telling of a joke.
3. Private jokes. Private jokes can include moments of
shared laughter and obvious amusement that derive
from coded messages or moments of sudden mutually
recognized humor that are opaque to all but the two
individuals who are communicating.
4. Fun and exaggeration. A very playful form of humor;
here individuals share active, animated, and exagger-
ated play or imitation behavior. High energy and a
deeper form of laughter often accompanies this
indicator.
5. Nervous giggling. Occasionally, individuals will begin to
chuckle with each other for no apparent reason. This
could result from a private joke or may indicate a brief
release of nervous tension given the experimental
context. The affect underlying the giggling should be
obviously positive and shared, unlike a similar form of
giggling associated with the Fear/Tension code.
Physical Cues
AUs include 1, 2, 6, 12, 6 + 12, and 25–27.
Counterindicators
1. Unshared humor. Laughter or amusement that is not
shared is never coded Humor.
2. Tense humor. Humor that is obviously both a nervous
reaction to a high level of tension in the conversation
and either lacking in any positive energy or unshared.
3. Affectionate humor. Sometimes a joke will be coupled
with affectionate messages. Such moments are more
properly coded affection.
4. Belligerent humor. A form of unshared humor, one
individual makes jokes that are intended to “get a rise”
out of the other or make the other angry.
5. Contemptuous humor. Jokes that are intended to be
hurtful or insulting and that are unshared. This is
sometimes confused with teasing. A good rule for
distinguishing contemptuous humor from good-
natured teasing is to attend closely to the degree to
which both individuals are amused.
Interest
Function
The function of this behavior is to communicate genuine in-
terest in one’s partner through active elaboration or clarifica-
tion seeking. As used in the SPAFF, Interest is characterized
as a positively valenced behavior that emphasizes informa-

278 Emotion Assessment
tion gathering about the partner as opposed to minor or
trivial factual information.
Indicators
1. Nonverbal attention with positive affect. Interested
persons will frequently attempt to actively communi-
cate their interest through nonverbal behaviors, such
as leaning forward in their chairs, affecting a warm
tone of voice, and making steady eye contact. The
interested person will communicate focused, respect-
ful, and active engagement with what his or her
partner is saying. If cues associated with Fear/Tension
are not present, the interested person will sometimes
communicate low levels of excitement (not to be
confused with Enthusiasm) that communicates a
desire to hear more.
2. Elaboration and clarification seeking. Interested indi-
viduals will often ask specific questions in order to
gather additional information. Frequently, such
questions will be accompanied by nonverbal behaviors
such as those described in indicator 1. It is important
that questions that serve to elicit more information are
not accompanied by nonverbal negative affect, as such
affect can indicate other affective agendas. Elaboration
and clarification-seeking questions can include
questions about a partner’s opinions and questions
that serve to paraphrase what a partner has been
saying. Paraphrasing questions are easy to confuse
with paraphrasing statements that are coded as
Validation (discussed later).
3. Open-ended questions. Almost any question that does
not require a “yes” or “no” response and that allows
the partner to express him- or herself in greater detail.
Physical Cues
AUs 1+2, 6, 12, 6+12, leaning forward, positive valence.
Counterindicators
1. Lack of eye contact. Eye contact is not absolutely
essential for coding interest, but a lack of eye contact
can indicate that interest is feigned or that questions
are serving some other affective function.
2. No pauses following questions. When questions are
frequent and no opportunity is provided for a partner to
respond to them, it is unlikely that genuine interest is
being observed. Relentless question asking, especially if it
appears to be leading the partner to a very specific series
of answers, can be a sign of Domineering behavior.
3. Low-balling questions. Similar to counterindicator 2,
low-balling questions are those to which there is only
one rational answer. An example would be, “Don’t you
want me to be happy?” Such a question is properly
coded Domineering.
4. Exchange of general factual information. It is important,
though sometimes difficult, to distinguish between
questions that communicate an interest in the partner
and those that communicate an interest in settling
some minor factual issue. An example of a nonin-
terested (per SPAFF) question might be “What time
is it?”
Neutral
Function
The Neutral code represents a sort of “dividing line” between
positive and negative SPAFF codes. It is relatively nonaffective
and is associated with the exchange of unvalenced informa-
tion. The voice will have a relaxed quality, with an even pitch
and volume. It is important to become familiar with an indi-
vidual’s neutral behavior early on in a coding session, as fa-
cial morphology and other characterological mannerisms that
are actually neutral for a given person can often seem affec-
tive to coders unfamiliar with them.
Indicators
1. Information exchanges.
2. Noncodable moments. Sometimes it will be unclear
whether a behavior is affective or what a particular
affective behavior represents. In the SPAFF, such
moments are coded Neutral.
Physical Cues
The neutral face is apparent, though care must be taken to
avoid coding baseline facial morphologies as affective facial
behavior.
Counterindicators
1. Loaded issue. It is possible that a moment of behavior
that seems to be a neutral exchange of information
actually makes reference to an issue that has emotional
relevance to the speaker, the receiver, or both. Such
moments are not properly coded Neutral.
2. Any codable affect.
Sadness
Function
In the SPAFF, the Sadness code refers to behaviors that com-
municate loss, resignation, helplessness, pessimism, hope-
lessness, or a plaintive or poignant quiescence.
Indicators
1. Sighing. Sighs, especially deep sighs, very frequently
occur in the context of Sadness. Thus sighing is nearly
always considered an indication of sad feelings (note,
however, “relief” as a counterindicators).
2. Pouting/Sulking. Sadness physical cues in the context of
being rebuffed, ignored, or not getting one’s way.

The Specific Affect Coding System (SPAFF) 279
Pouting may cause the sad person to appear to
withdraw from the conversation.
3. Resignation. Sad individuals will frequently behave as if
resigned or hopeless. This behavior is communicated
through a pattern of very low energy, slouching, long
pauses between words, and so forth. In the resigned
person, nearly all movement appears to require extra
effort.
4. Crying. Nearly all instances of crying indicate sadness
(but see “happy tears” as a counterindicators.) Some-
times individuals can be observed “choking back
tears,” or trying not to cry. Physical cues and tears
welling up in the eyes will give them away.
5. Hurt feelings. In response to moments of high negativ-
ity, such as belligerence, contempt, or anger, individu-
als will sometimes report or appear to have hurt
feelings. Such moments are coded as Sadness.
Physical Cues
AUs 1, 6, 15, 17, 1+6, 1+15, 1+6+15, 1+6+15+17. Shoul-
ders may droop, and individuals may hang their heads or
look down. The lips and the chin may tremble. The voice
may quaver in terms of pitch and amplitude and may occa-
sionally break.
Counterindicators
1. No back channels. A lack of responding that is attribut-
able to the deliberate attempt to communicate lack of
interest is not a form of pouting and is more properly
coded Stonewalling.
2. Relief. Individuals who display a sudden decrease in
energy as a result of the diffusion of tension or an
escape from responsibility may be showing evidence of
relief, which may be coded as Neutral.
3. Happy tears. Happy tears are here intended to mean
one of two things. First, tears can sometimes result
from intense laughter. Second, tears can sometimes
result from sudden moments of shared intimacy,
compliments, accomplishments, and so forth. These
instances of tears are more properly coded as Humor,
Enthusiasm, or Affection.
Stonewalling
Function
Stonewalling functions to communicate an unwillingness to
listen or respond to the receiver.
Indicators
1. Active away behavior. The speaker focuses on some
trivial object in order to avoid contact with the
receiver. Such away behavior frequently entails the use
of “automanipulation,” a behavior characterized by
playing with hair or hands (e.g., cleaning fingernails or
looking at split ends). This behavior is “active” in
Stonewalling in that it is not a function of idleness but
rather purposefully communicates an unwillingness to
pay attention, especially during conversational
moments characterized by high levels of negative
affect. The “speaker” (i.e., the contemptuous person) is
communicating the message, “I’d rather not be here
right now, and I don’t want to listen to you.”
2. No back channels. The stonewalling person offers no
vocal or nonvocal back channels such as one would
find in Validation. There are no head nods, the neck is
rigid, there are no vocal or verbal assents (as in “umm-
hmmm,” “yeah,” “uh-huh,” etc.), and no other verbal
responses. There is little if any facial movement and
certainly no facial mirroring or eye contact. The “no-
back-channeling” behavior may occur very abruptly, as
if intended to suddenly put up an obvious, though
technically invisible, wall between the speaker and the
receiver.
3. Monitoring gaze. Within the context of “no back
channels,” stonewalling individuals will occasionally
steal glances at their partners, as if to remind their
partners to notice their lack of listening behavior. This
can appear as a intermittent glance in the partner’s
direction, as if the partner is an annoyance that must
be endured, much as one might occasionally glance
over at a noisy person in a library.
Physical Cues
In Stonewalling, the face will typically appear stiff or frozen.
The jaw may be clenched, and the muscles of the neck may
be obviously flexed. Other times, the face will show no ob-
vious signs of emotion at all, deliberately arranged to appear
neutral.
Counterindicators
1. Boredom. Individuals can sometimes become bored or
otherwise run out of things to say to each other.
Sometimes, this will cause them to sit quietly without
interacting for seemingly long periods of time. Away
behavior can characterize these moments, but they
should not be confused with Stonewalling behavior.
Stonewalling does not result from idleness or boredom
but is rather a form of active and aggressive communi-
cation, most frequently observed during heated
moments.
2. Sleepiness. If an individual stops offering back channels
but also appears to be very sleepy (as sometimes
happens), his or her behavior is more properly coded
as Neutral.
3. Resignation. Sometimes individuals will become sad or
defeated during an intense conversation. During such
moments, they can appear to be Stonewalling for want
of back-channeling behavior. It is important to
recognize when this is occurring and to code accord-

280 Emotion Assessment
ingly. Most often, resigned behaviors such as these are
coded as Sadness.
Threats
Function
Threats are a particularly hostile form of domineering behav-
ior in that their function is to control the behavior of the
receiver by setting explicit conditions under which the re-
ceiver will be punished for behaving in ways the speaker finds
undesirable.
Indicators
1. Bans. These are direct “if/then” statements that forbid
certain behaviors and threaten to impose punitive
(sometimes violent) consequences if those behaviors
occur. An example might be “if you ever speak to me
like that again, I’ll. . . .”
2. Ultimatums. Ultimatums reflect demands for change
within some defined context or time period. An
example might include “if you don’t start doing your
share around here by next month, I’m moving out.”
Physical Cues
AU 1, 2 (“the horns”), 1+2, 1+2+5, head forward, body for-
ward, finger pointing, head cocked to one side.
Counterindicators
• Good-natured teasing. Good-natured “jabs” at the
receiver’s foibles and those that include humorous
threats (as in, “ooh, I’m going to get you for that!”) are
coded as Humor.
Validation
Function The function of validation is to communicate sin-
cere understanding and acceptance of one’s partner or of
one’s partner’s views and opinions. In the SPAFF, Validation
is considered to be a positively valenced behavior.
Indicators
1. Back channels. Back channels are behaviors that
indicate attentive and affirmative listening through the
use of paralinguistic and physical cues, such as head
nods and “uh-huhs” or other physical and vocal
assenting behaviors. Usually, back channels are
accompanied by eye contact.
2. Direct expressions of understanding. Direct expressions
of understanding include explicit expressions of
respect or agreement (e.g., “I agree,” or “that’s a very
good point”).
3. Paraphrasing. In this behavior, individuals repeat back
what their partners have told them, usually verbatim,
but sometimes in a slightly altered style.
4. Apologies.
5. Sentence finishing. In this behavior, individuals will
place endings on the sentences their partners have
begun. This behavior lets partners know that both
individuals are “on the same page.” Importantly,
sentence finishing is an indicator of validation only if it
is delivered in a package of positive affect (see “Physi-
cal Cues”).
Physical Cues
AUs 1+2, 6, 12, 6+12. Head nod, eye contact, nonconfronta-
tional voice tone.
Counterindicators
1. Lack of eye contact. A lack of eye contact can mean that
the back channels being offered are insincere, as in
humoring. Back channels without eye contact can also
be associated with sarcastic behavior.
2. Bobbing heads. “Bobbing heads” are head nods that
appear so automatic and repetitive that they essen-
tially become meaningless. Bobbing heads can also be
a sign of exasperation—a kind of nonverbal request
to “shut up.”
3. Affect mirroring. Sometimes, the various indicators of
validation occur in the context of strong mirroring of
affect, as when an individual says, “I understand how
you’re feeling” while expressing facial signs of sadness
in response to their crying partners. The SPAFF
considers such expressions to be signs of empathy,
and such signs are properly coded Affection.
4. Interrupting. Sentence finishing can be an important
indicator of Validation, but if the sentence finishing is
abrupt or is delivered with negative affect, it is likely
nothing more than an interruption related to Domi-
neering, Defensiveness, or other negative affective
behaviors.
Whining
Function
Whining functions to make what might otherwise be an or-
dinary complaint into a plaintive or pleading form of emo-
tional protest. Whining suggests an innocent victim stance,
communicating something like “What are you picking on me
for?” or “What about all the good I do?”
Indicators
• Whiny protest. Whining is really characterized by a
quality of voice paired with a complaint or protest.
This voice quality is high-pitched, nasal, “sing-songy,”
or otherwise annoyingly plaintive. For example, the
question “why” might be expressed in a high-pitched
voice and drawn out with an exaggerated “eeee” sound
at the end, as in “whyyyyeeee?”

The Specific Affect Coding System (SPAFF) 281
Physical Cues
AUs 1, 1 + 2, 1 + 2 + 15.
Counterindicators
• Defensive whining. Sometimes defensive behaviors can
be expressed in a whiny voice style. Such moments are
more properly coded Defensive.
The Nuts and Bolts of SPAFF Coding
Training
In our experience, SPAFF training requires some reading,
some didactic exercises, and a large amount of practice. In-
deed, SPAFF practice is generally ongoing in our labora-
tories, including weekly meetings at which confusions
are clarified and practice tapes are watched collectively.
In the past, we have trained SPAFF coders in steps, as
follows.
3
Step 1. People watching and learning to think about
constructs. Following the structure of this chapter,
students are encouraged to think about and
practice the rules of people watching and are
trained to understand the difference between
constructs and indicators.
Step 2. FACS training. At a minimum, coders should be
trained to reliably recognize those FACS codes that
are commonly observed in the codes of the SPAFF.
It is preferable, though not necessary, for SPAFF
coders to be fully FACS certified through a process
of studying the FACS manual and taking the FACS
exam (see Cohn et al., chapter 13, this volume, for
details).
Step 3: Learning the SPAFF codes. Learning the SPAFF
codes requires careful explanation of the codes, in
addition to frequent viewing of examples of the
codes using videotapes.
4
When viewing videotapes,
we start whenever possible by showing clips
designed to illustrate codes that are the subject of
discussion or recent reading. Eventually, longer
clips are played, and students are encouraged to
discuss what they see in terms of the SPAFF codes
they have learned. Virtually any videotaped
interaction (e.g., tapes of participants, clips from
movies or television) can be useful for this.
Exercises such as these serve at least two purposes.
First, they train people to talk about and discuss
affective behavior using SPAFF as their “language.”
The value of this first purpose is often
underappreciated (more on this later). Second,
they hone their skills in reliably identifying SPAFF
codes.
Data Collection and Reliability Assessment
Video- and Audiotapes
Several pragmatic data collection considerations are impor-
tant for maximizing the effectiveness of the SPAFF. For ex-
ample, it is critical that audio and video are of high quality.
We have found that small microphones with gated compres-
sor mixers attached to people’s shirts or collars provide the
best audio quality. Many video cameras come with their own
built-in microphones, but the audio quality provided by these
is inconsistent. Alternatives include wall-mounted omnidi-
rectional cardoid condenser microphones, generally aimed
at the participants. For video collection, many options now
exist, including sophisticated digital recorders that are quite
compact and relatively unobtrusive. In any case, we recom-
mend that two cameras are used to generate a “split screen,”
at least with dyads (such that both participants can be viewed
simultaneously on the same monitor), and that both cam-
eras are situated to provide full-face shots, including cover-
age from the top of the head to the mid-chest. To create a
split screen, a special-effects generator with “wipe” capabili-
ties will be required. A wide variety of such devices are com-
mercially available. Profile shots should be avoided whenever
possible, as they make it harder to read emotion on the face
(e.g., AU14, a FACS code associated with contempt, is fre-
quently only seen on one side of the face). Moreover, and
despite the risk of participant self-consciousness, cameras
should be low. We have found that cameras placed about 1
foot above the participants’ eye level causes minimal self-
consciousness and that, in any case, participants in emotion-
ally charged situations habituate to the cameras rapidly. In
coding the SPAFF, a time-code generator that is both visible
on the video recording and readable by computer interface
is imperative. Devices that accomplish this, frequently re-
ferred to as vertical interval time code (or “VITC”) generators,
are widely available.
Reliability Assessment
Timing is critical to the utility of the SPAFF as a measure, as
well as to the assessment of reliability. Before coding begins,
a number of timing decisions must be made. For example,
the unit of time per code must be established. Due to tech-
nological limitations, early versions of the SPAFF used very
little or no timing information, using “thought units” based
on transcripts for coding instead. Since then, we have fre-
quently used 1 second as the unit of time for SPAFF coding,
allowing for one code per second per participant. Next, the
time window for assessing reliability must be established. The
time window refers to the amount of time relative to a target
moment around which coders may agree or disagree. For
example, a time window of plus or minus 1 second may be
established around some target second, allowing for a total
window of agreement of 3 seconds within which SPAFF cod-
ers may agree or disagree about how to categorize a particu-
lar behavior. The size of this window is a function of the

282 Emotion Assessment
investigator’s discretion. For most coding, a 3-second win-
dow provides an adequate balance between precision and
agreement.
A variety of options exist for the statistical assessment of
reliability of these kinds of data (Bakeman & Gottman, 1997).
We have assessed reliability by computing confusion matri-
ces, which tally agreements and disagreements between two
coders, allowing for the computation of Cohen’s kappa both
for each interaction and for all interactions within a sample
combined. Cohen’s kappa is useful for assessing the reliabil-
ity of event-coded data to be used for sequential-type analy-
ses in which agreements must be locked in time within a
particular window. We have also computed Cronbach’s al-
pha for this purpose, per Wiggins (1973). However, when
reliability concerns the total frequency of a given code, the
rank-order intraclass correlation, or generalizability coefficient,
provides a sufficient assessment of reliability (see also Shrout
& Fleiss, 1979). Figure 16.3 provides a sample confusion
matrix obtained from a sample of domestically violent couples,
taken from Gottman et al. (1995).
Typically, all tapes in a given study are coded twice, once
each by two independent coders. Checking the reliability of
every interaction in a given study is recommended for opti-
mal data quality. Moreover, reliability should be checked
continuously as a study is ongoing if those reliabilities are to
remain high throughout. It is useful for coders to know that
their reliability will be checked for every tape that they code.
A variety of computer-based coding packages are now avail-
able, including the Long Video Coding System (VCS, avail-
able from the James Long Company; see http://www.jameslong
.net/ for more information). Tape-by-tape reliability checks
are augmented by weekly meetings, with all coders present.
Such meetings are critical to the success of coding any given
study. Because studies differ in sometimes subtle ways, it is
possible that the SPAFF manual will be unclear about a par-
ticular behavior or distinction. Coding meetings cope with
this in at least two ways. First, they can clarify applications
of the SPAFF to particular samples (e.g., is Contempt—a
latent construct—always indicated in the same ways in all
samples?). Second, they contribute to clarifications within
Figure 16.3. A sample confusion matrix (Gottman et al., 1995).
Disagreement Matrix and Overall Kappa
Files processed in this run:
T131rh
High and Low Combined
Column and Row Headings are as follows
(1) Disgust (7) Defensive (13) Validation
(2) Contempt (8) Whining (14) Affection
(3) Belligerence (9) Sadness (15) Humor
(4) Domineering (10) Stonewall (16) Surprise/Joy
(5) Anger (11) Neutral
(6) Fear (12) Interest
OBS#1 OBS#2
12345678910111213141516
1.................000
2.2................006
3.................000
4 . . . 6. . . 3. . . . . . . . . . .007
5.................000
6.................000
7 . 3. . . . . 337. . . . 28. . . . . . .418
8.................000
9.................000
10.................000
11 . . . . . . 30. . . . 458. . 2. 1. . . .551
12.................000
13 . . . . . . . . . . 1. . 6. . . . .009
14 . . . . . . . . . . . . . 2. . . .005
15 . . . . . . . . . . 1. . . 1. 5. . .006
16.................000
.000 .002 .000 ,010 .000 .000 .415 .000 .000 .000 .554 .000 .008 .002 .008 .000
Total frequency 886.000
Pct observed agreement .921
Pct due to chance .479
Kappa .848
Standard Error .031
Z-score 27.652

The Specific Affect Coding System (SPAFF) 283
the manual. In this way, the development of the SPAFF is
analogous to “open source” software, with many groups of
coders, within and between labs, contributing to its devel-
opment. Coding meetings are typically semistructured, with
meeting facilitators bringing clips that are frequently, or re-
cently, the source of disagreement and confusion. Such clips
are often brought to the attention of the meeting facilitator
by the coders themselves, but they need not be. For example,
the VCS program allows for the easy identification of dis-
agreements in the coding record that can be used for discus-
sion. It is important to discuss specific instances, with video,
during these meetings; abstract discussions of the codes are
generally less helpful. In general, the agreement in a single
session of coding should be above 75%, but 60% can be
acceptable with particularly difficult tapes. If the percentage
of agreement is lower, two different coders should recode the
tape.
The Future of the SPAFF
The SPAFF is always in development. It can be used to quan-
tify affective and interactive behaviors, but it can also be
thought of more generally as a language for describing those
behaviors. Indeed, recent applications of the SPAFF include
its transformation into a continuous scale and the use of the
SPAFF terminology to code higher order constructs, and even
sequences of behavior, directly.
The SPAFF as a Scale
The transformation of an earlier version of the SPAFF was
accomplished by using prior research to assign weights to
each code (see Carrere & Gottman, 1999; Gottman, Swan-
son, & Murray, 1999). Thus codes more predictive of nega-
tive outcomes received more negative weights, and codes
more predictive of positive outcomes received more positive
weights. The weighting scheme resulted in a continuous scale
running from –4 to +4 in any given second. These scores were
then summed across 6-second intervals, resulting in a con-
tinuous score of –24 to +24 for any given 6-second epoch.
Weights assigned to a previous version of the SPAFF code
are given in Table 16.2.
The SPAFF as a Language
In new ongoing research, the language of the SPAFF is be-
ing implemented as a means of training coders to observe
specific sequences of behavior across larger periods of time
(e.g., 5 minutes). In this work, very specific sequences of
behavior are of interest, and estimates of the frequency of
such sequences are given by continuous Likert-type rating
scales. For example, of particular interest is a sequence of
behavior that begins with one spouse’s low-level negative
affect (the antecedent) and ends with the other’s high-level
negative affect (the consequent). In previous research, this
sequence has been interpreted as a “rejection of influence”
(Coan et al., 1997; Gottman et al., 1998). Coders under-
stand that low-level negative affect includes such SPAFF
codes as Anger, Fear/Tension, Sadness, and so forth, and
that high-level negative affect includes Stonewalling, Con-
tempt, Criticism, Defensiveness, and so on. They use their
knowledge of the SPAFF to code the sequence directly.
Though currently in the experimental stages, this approach
has dramatically increased the speed with which coding is
done, with little or no apparent cost in terms of reliability.
For example, the intraclass correlation coefficient (ICC)
across three coders and 64 couples for the coding of hus-
band influence rejection was .84, although the same ICC
for the wife was somewhat lower, at .63. This work is still
in development and awaits further testing and analysis.
Nevertheless, it holds the promise of new ways to imple-
ment the SPAFF.
Conclusion
The SPAFF is a flexible, evolving, and reliable language for
describing interactive affective behavior. It has informed
numerous studies and enjoys ample empirical support for
its constructs. The SPAFF has typically been applied to the
study of married couples, but in recent years it has been
applied to parent-child interactions and even to group
therapy sessions. A few new codes have been introduced here,
and recent new directions in the use of the SPAFF have been
covered. Using the SPAFF can be labor intensive and chal-
lenging but also highly rewarding. As use of the SPAFF in-
creases, we enthusiastically anticipate yet more innovation
and development.
Table 16.2
Weighting scheme for the 16-code version
of the SPAFF
Positive Affects Negative Affects
Joy + 4 Contempt – 4
Humor + 4 Disgust – 3
Affection + 4 Defensiveness – 2
Validation + 4 Belligerence – 2
Interest + 2 Stonewalling – 2
Domineering – 1
Anger – 1
Whining – 1
Sadness – 1
Fear / Tension
0</T
B>
Neutral + 0.1

284 Emotion Assessment
Notes
1. Here we are using the language of latent variable models
analogously, not literally. Actual latent variable models imply a
host of mathematical properties that are not necessarily true of
any of our SPAFF codes. Nevertheless, many of the theoretical
properties do apply. It is better, for example, to observe
multiple indicators of a construct if one wishes to infer the
construct’s existence. This is equally true of the SPAFF.
However, in mathematical latent variable modeling, it is
virtually axiomatic that one is required to have at least three
indicators (sometimes referred to as manifest variables) to
properly model the construct. This is not true of SPAFF coding,
in which, in many cases, a single indicator is sufficient.
2. For the sake of clarity, we contrast critical statements
with complaint statements. These examples are taken from John
Gottman’s book Why Marriages Succeed or Fail (Gottman, 1994).
Complaint: We don’t go out as often as I’d like to.
Criticism: You never take me anywhere.
Complaint: It upset me when I came home and there
were dirty dishes in the sink. This morning we agreed you’d
wash them.
Criticism: You left dirty dishes in the sink again. You
promised me you wouldn’t. I just can’t trust you, can I?
Complaint: I expected you to come home right after work.
When you didn’t, it made me feel like you care more about
going out with your friends than spending time with me.
Criticism: I hate that you’re the type of person who never
thinks to call and tell me you’ll be late coming home. You
always leave me hanging. You care more about your friends
than you do about our marriage.
3. One of the authors (JAC) is available for SPAFF work-
shops.
4. Training and test tapes for the SPAFF are not currently
commercially available, although new tapes may be available in
the future, depending on demand. Alternatively, any tapes
involving affective behavior can be used to gain practice in
coding SPAFF. One of the authors (JAC) has even used movie
clips to illustrate examples of SPAFF codes. Virtually any
videotaped social interaction can be useful for SPAFF training,
as long as the faces of the individuals are clearly visible.
References
Babcock, J. C., Waltz, J., Jacobson, N. S., & Gottman, J. M.
(1993). Power and violence: The relation between commu-
nication patterns, power discrepancies, and domestic
violence. Journal of Consulting and Clinical Psychology, 61(1),
40–50.
Bakeman, R., & Gottman, J. M. (1997). Observing interaction: An
introduction to sequential analysis (2nd ed.). New York:
Cambridge University Press.
Banse, R., & Scherer, K. R. (1996). Acoustic profiles in vocal
emotion expression. Journal of Personality and Social
Psychology, 70(3), 614–636.
Blairy, S., Herrera, P., & Hess, U. (1999). Mimicry and the
judgement of emotional facial expressions. Journal of
Nonverbal Behavior, 23(1), 5–41.
Bollen, K. A. (2002). Latent variables in psychology and the
social sciences. Annual Review of Psychology, 53, 605–634.
Burman, B., Margolin, G., & John, R. S. (1993). America’s
angriest home videos: Behavioral contingencies observed in
home reenactments of marital conflict. Journal of Consulting
and Clinical Psychology, 61(1), 28–39.
Carrere, S., & Gottman, J. M. (1999). Predicting divorce among
newlyweds from the first three minutes of a marital conflict
discussion. Family Process, 38, 293–301.
Carstensen, L. L., Gottman, J. M., & Levenson, R. W. (1995).
Emotional behavior in long-term marriage. Psychology and
Aging, 10(1), 140–149.
Coan, J., Gottman, J. M., Babcock, J., & Jacobson, N. (1997).
Battering and the male rejection of influence from women.
Aggressive Behavior, 23(5), 375–388.
Coan, J. A. (1998, March). The systematic observation of affect in
therapy. Paper presented at the Conference on Successful
Relating in Couples, Families, Between Friends, and at
Work, Tucson, AZ.
Cohan, C. L., & Bradbury, T. N. (1994). Assessing responses to
recurring problems in marriage: Evaluation of the Marital
Coping Inventory. Psychological Assessment, 6(3), 191–200.
Cohan, C. L., & Bradbury, T. N. (1997). Negative life events,
marital interaction, and the longitudinal course of newly-
wed marriage. Journal of Personality and Social Psychology,
73(1), 114–128.
Ekman, P., & Davidson, R. J. (1993). Voluntary smiling changes
regional brain activity. Psychological Science, 4(5), 342–345.
Ekman, P., Davidson, R. J., & Friesen, W. V. (1990). The
Duchenne smile: Emotional expression and brain physiology:
II. Journal of Personality and Social Psychology, 58(2), 342–353.
Ekman, P., & Friesen, W. V. (1975). Unmasking the face: A guide
to recognizing emotions from facial clues. Englewood Cliffs,
NJ: Prentice-Hall.
Ekman, P., & Friesen, W. V. (1978). The Facial Action Coding
System (FACS): A technique for the measurement of facial
action. Palo Alto, CA: Consulting Psychologists Press.
Ekman, P., Friesen, W. V., & Tomkins, S. S. (1971). Facial
Action Scoring Technique (FAST): A first validity study.
Semiotica, 3(1), 37–38.
Gottman, J. M. (1979). Marital interaction: Experimental
investigations. New York: Academic Press.
Gottman, J. M. (1993a). The roles of conflict engagement,
escalation, and avoidance in marital interaction: A longitu-
dinal view of five types of couples. Journal of Consulting and
Clinical Psychology, 61(1), 6–15.
Gottman, J. M. (1993b). A theory of marital dissolution and
stability. Journal of Family Psychology, 7(1), 57–75.
Gottman, J. M. (1994). Why marriages succeed or fail. New York:
Simon and Shuster.
Gottman, J. M., Coan, J., Carrere, S., & Swanson, C. (1998).
Predicting marital happiness and stability from newlywed
interactions. Journal of Marriage and the Family, 60(1), 5–22.
Gottman, J. M., & Krokoff, L. J. (1989). Marital interaction and
satisfaction: A longitudinal view. Journal of Consulting and
Clinical Psychology, 57, 47–52.
Gottman, J. M., & Levenson, R. W. (1992). Marital processes
predictive of later dissolution: Behavior, physiology, and
health. Journal of Personality and Social Psychology, 63(2),
221–233.

The Specific Affect Coding System (SPAFF) 285
Gottman, J. M., Levenson, R. W., Gross, J., Frederickson, B. L.,
McCoy, K., Rosenthal, L., et al. (2003). Correlates of gay
and lesbian couples’ relationship satisfaction and relation-
ship dissolution. Journal of Homosexuality, 45(1), 23–43.
Gottman, J. M., Levenson, R. W., Swanson, C., Swanson, K.,
Tyson, R., & Yoshimoto, D. (2003). Observing gay, lesbian
and heterosexual couples’ relationships: Mathematical
modeling of conflict interaction. Journal of Homosexuality,
45(1), 65–91.
Gottman, J. M., McCoy, K., Coan, J., & Collier, H. (1995). The
Specific Affect Coding System (SPAFF) for observing emotional
communication in marital and family interaction. Mahwah, NJ:
Erlbaum.
Gottman, J. M., Swanson, C., & Murray, J. (1999). The
mathematics of marital conflict: Dynamic mathematical
nonlinear modeling of newlywed marital interaction. Journal
of Family Psychology, 13, 3–19.
Heyman, R. E. (2001). Observation of couple conflicts: Clinical
assessment applications, stubborn truths, and shaky
foundations. Psychological Assessment, 13(1), 5–35.
Hops, H., Wills, T. A., Weiss, R. L., & Patterson, G. R. (1972).
Marital Interaction Coding System. Eugene, OR: University of
Oregon and Oregon Research Institute.
Jacobson, N. S., Gottman, J. M., Waltz, J., Rushe, R., Babcock,
J., & Holtzworth-Munroe, A. (1994). Affect, verbal content,
and psychophysiology in the arguments of couples with a
violent husband. Journal of Consulting and Clinical Psychol-
ogy, 62(5), 982–988.
Levenson, R. W., Carstensen, L. L., & Gottman, J. M. (1994).
The influence of age and gender on affect, physiology, and
their interrelations: A study of long-term marriages. Journal
of Personality and Social Psychology, 67(1), 56–68.
Notarius, C. I., Benson, P. B., Sloane, D., Vanzetti, N. A., &
Hornyak, L. M. (1989). Exploring the interface between
perception and behavior: An analysis of marital interaction
in distressed and nondistressed couples. Behavioral
Assessment, 11, 39–64.
Rozin, P., Lowery, L., & Ebert, R. (1994). Varieties of disgust
faces and the structure of disgust. Journal of Personality and
Social Psychology, 66(5), 870–881.
Scherer, K. R. (1974). Acoustic concomitants of emotional
dimensions: Judging affect from synthesized tone se-
quences. In S. Weitz (Ed.), Nonverbal communication (pp.
249–253). New York: Oxford University Press.
Shrout, P. E., & Fleiss, J. L. (1979). Intraclass correlations: Uses in
assessing rater reliability. Psychological Bulletin, 33, 159–174.
Wiggins, J. S. (1973). Personality and prediction: Principles of
personality assessment. Reading, MA: Addison-Wesley.
Xiao, J., Moriyama, T., Kanade, T., & Cohn, J. F. (2003).
Robust full-motion recovery of head by dynamic templates
and re-registration techniques. International Journal of
Imaging Systems and Technology, 13, 85–94