Emotion

Emojis as Social Information in Digital Communication

Thorsten M. Erle, Karoline Schmid, Simon H. Goslar, and Jared D. Martin
Online First Publication, August 5, 2021. [Link]

CITATION
Erle, T. M., Schmid, K., Goslar, S. H., & Martin, J. D. (2021, August 5). Emojis as Social Information in Digital
Communication. Emotion. Advance online publication. [Link]
 Emotion
© 2021 American Psychological Association
ISSN: 1528-3542 [Link]

Emojis as Social Information in Digital Communication

Thorsten M. Erle1, Karoline Schmid2, Simon H. Goslar1, and Jared D. Martin3

1
Department of Social Psychology, Tilburg University
2
Department of Social and General Psychology, Julius-Maximilians-Universität Würzburg
3
OptumLabs, Minnetonka, Minnesota, United States
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Facial expressions of emotion are nonverbal cues that evoke affective, inferential, and social responses
This document is copyrighted by the American Psychological Association or one of its allied publishers.

during face-to-face communication. Given that communication is moving more and more from face-to-
face to digital contexts, the present research tested the functional equivalence of their digital counter-
parts—emojis. Eleven high-powered experiments tested the general effectiveness of emojis to convey
emotionality and to disambiguate discourse during digital communication, as well as predictions about
their social-emotional properties derived from the Emotion as Social Information (EASI) model.
Compared to messages without emojis, those including emojis were perceived as emotionally more
intense and as of more extreme valence. Furthermore, the effects of emojis on perceived valence were
mediated via perceived emotional intensity. This suggests that emojis are effective quasi-nonverbal cues
for digital communication. Furthermore, in line with predictions of the EASI model, emojis produced
patterns similar to what has been observed for facial expressions of emotion in face-to-face communica-
tion, supporting their functional equivalence. Specifically, they instigated affective (emotion contagion)
and inferential (understanding) processes, which subsequently resulted in behavioral intentions (em-
pathic concern). In terms of the predicted mediating processes, we found differences between emojis
and offline facial expressions of emotion. These deviations from our predictions are attributed to inher-
ent differences between digital and face-to-face communication and limitations in the employed meth-
odology. In light of the present findings, we discuss a theoretical synthesis of emojis in digital
communication with the EASI model and propose a research agenda to connect emotion research with
predominant forms of modern communication.

Keywords: emojis, facial expressions, digital communication, nonverbal communication, emotion

Communication is a key aspect of life, involving content that is (Kaye et al., 2017). For instance, whereas we can emphasize the
both verbal and nonverbal (e.g., gestures or facial expressions). A emotional meaning of a sentence during face-to-face communica-
hallmark of the postmillennial society is the rise of digital commu- tion via nonverbal cues such as facial expressions or inflections of
nication (Koch & Frees, 2016) and since the advent of multitouch tone, this is not possible in text-based digital communication. As a
smartphones in 2007; texting has supplanted phone or face-to-face partial remedy for this, emoticons, smileys, and emojis were
conversations as the most popular method of communication developed. Emoticons are ASCII-based sign sequences modeling
(Smith et al., 2015). Although popular, text-based digital commu- facial expressions like “:-)” or “:(,” which are used to digitally
nication is limited in comparison with face-to-face communication refer to prototypical facial expressions. Nowadays, instant messag-
ing apps also offer graphical interfaces for these emoticons, com-
monly referred to as smileys (i.e., graphical interfaces for facial
expressions) or emojis (i.e., graphical interfaces for anything).1
Given that most communication happens digitally via text these
Thorsten M. Erle [Link] days, it is not surprising that these digital aids have taken the world by
Jared D. Martin [Link] storm. In 2015; the emoji U þ 1F602 (a smiling face with tears of joy)
The authors would like to thank Paulina Adler, Benjamin Eichert, Maren
was named the word of the year, recognizing the impact of emojis on
Flottmann, Lea Geraedts, Aileen Halbe, Sofie Hanraths, Paul Heineck, Tina
language and popular culture. In 2017; a feature film dedicated entirely
Hönninger, Leonie Labude, Julia Linden, Paul Lotze-Hermes, Estelle
Knoblauch, Franzisca Maas, Marina Orifici, Julius Rennert, Isabella Russ, to emojis was released, earning over $200 million at the box office.
Anna Schulte, Ilona Vieten, Katharina Wagner, Anna Wehler, Patricia Wilms, Despite their meteoric rise in popular culture, surprisingly little
Margarete Wolf, and Marie-Ann Ziegler for help with the data collection. research exists on potential psychological consequences of emoji usage
All data and materials are available at [Link]
Correspondence concerning this article should be addressed to Thorsten 1
Although the term “emojis” encompasses many images beyond just
M. Erle, Department of Social Psychology, Tilburg University, P.O. faces, we will use it throughout this text rather than “smiley,” because
Box 90153, 5000 LE Tilburg, the Netherlands. Email: [Link]@ “smiley” is potentially confusing given that we did not only investigate
[Link] smiles, but also other facial expressions of emotion.

1
 2 ERLE, SCHMID, GOSLAR, AND MARTIN

in digital communication (for a recent discussion, see Kaye et al., emotional states but rather can serve as social signals that influ-
2017). Since emojis have become pervasive in text-based communica- ence interpersonal relationships (Reis & Collins, 2004). Since
tion, what effects do they have on the meaning of the text they accom- (digital) communication necessarily happens between at least two
pany and the interpersonal relationship between sender and receiver? people, analyzing (virtual) facial expressions of emotion through
Qualitative studies suggest that people use emojis to communi- the lens of social-functional approaches is an intuitive extension,
cate emotional tone, to reduce the ambiguity of discourse, and to which has even been proposed by some of these social-functional
lighten the mood (Derks et al., 2007, 2008; Kaye et al., 2016). approaches (see van Kleef, 2017). Our second research goal was
Thus, emojis are said to serve as “quasi-nonverbal cues” employed thus to provide additional empirical support for the functional
to compensate for the lack of nonverbal channels in digital com- equivalence of emojis and facial expressions of emotion from the
munication (Lo, 2008). However, empirical tests of these commu- perspective of the Emotions as Social Information (EASI) model
nicative functions are surprisingly scarce, especially given the (van Kleef, 2009). In Experiments 2a–4c we tested specific predic-
prominent role emojis have in our daily lives. While a few studies tions based on this model, which we describe in the next section,
have confirmed (Kaye & colleagues, 2016; Kaye & colleagues, and in doing so, we connect research on emojis from computer sci-
2017) qualitative reports, some of them had limited ecological va- ence, consumer behavior, and communication research with basic
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

lidity, presenting chat protocols on paper rather than on an elec- theorizing from emotion research.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

tronic device (e.g., Derks et al., 2007, 2008) or using uncommon

and outdated graphical interfaces for emojis (e.g., Das et al., 2019; The EASI Model
Manganari & Dimara, 2017). Finally, failed attempts at replication
of some studies exist (Walther & D’Addario, 2001). More recent The EASI model posits that facial expressions of emotion influ-
studies have proceeded to investigate the effects of emojis in more ence observers by triggering one (or both) of two pathways, which
specific contexts such as online consumer behavior (see, e.g., Das subsequently affect behavior: inferential processes and affective
et al., 2019; Lohmann et al., 2017; Manganari & Dimara, 2017) or reactions. Inferential processes are conclusions drawn based on fa-
workplace communication (Glikson et al., 2018). Although these cial expressions about the feelings and attitudes of the expresser.
studies overcame some of the previous limitations, they investi- For instance, recent research suggests that smiles serve (among
gated situations in which emojis are not commonly used. In fact, other functions) the purpose of creating and maintaining social
Glikson and colleagues (2018) demonstrated that emojis are per- bonds (Martin et al., 2017). Thus, when a social interaction partner
ceived as inappropriate in these contexts. The present research smiles, a person might infer that the expresser is content with the
moves the field past these limitations in several ways to address way their social relation is going. This inference, in turn, will
two research goals. likely lead to the adoption of future behaviors that keep this rela-
First, we made appropriate changes to our stimulus design to tionship intact. Indeed, research has shown that affiliative displays
reflect current digital communication technologies and we decided of emotion such as smiling increase cooperation in individuals
to locate our research within the context where emojis are actually (Krumhuber et al., 2007) and even among groups (Barsade, 2002).
used most commonly: instant messaging between friends. Similarly, facial displays of anger can signal that past behaviors
Thereby, we aimed to provide an updated test of how well emojis have not satisfied an interaction partner, leading to the inference
fulfill the nonverbal communicative functions that facial expres- that a person should change their behavior. In line with this, stud-
sions of emotion have in face-to-face communication. These func- ies have also shown that negotiators make concessions when nego-
tions, as stated by Kaye and colleagues (2016) are 1) to set an tiation partners display anger, likely because they are afraid that
emotional tone within a conversation, and 2) to reduce the ambigu- otherwise the negotiations might reach an impasse (Sinaceur &
ity of discourse. Specifically, this means that emojis should lead to Tiedens, 2006; van Kleef et al., 2004; van Kleef et al., 2015; van
increased perceptions of emotionality in digital communication, Dijk et al., 2008; Yip & Schweinsberg, 2017). Finally, similar
and this emotionality should serve the purpose to clarify the cen- research (also targeting a wider array of emotional facial expres-
tral content of a message that is sent with an emoji. In the present sions) has been conducted in the areas of leadership, organiza-
studies, this content was always the perceived valence of a mes- tional behavior, and social decision-making (for overviews on
sage, as most emojis were designed to emulate facial expressions these topics, see Koning & van Kleef, 2015; van Kleef, 2014; van
of positive or negative emotions. We tested these communicative Kleef et al., 2010).
functions of emojis in Experiments 1a–1d. The second pathway within the EASI model states that facial
Second, while many of the previous studies contributed to our expressions can influence observers by triggering affective reac-
understanding of individual social-emotional functions of emojis tions. Among other affective processes, van Kleef (2009) proposes
(e.g., emotion contagion; Lohmann et al., 2017); previous research that facial expressions can influence an observer’s emotional state
has not done well to systematically integrate emojis into current via the process of emotion contagion. That is, the same emotional
theorizing in emotion research. Specifically, recent theorizing state implied by the expresser’s facial expression of emotion is
about the social nature of emotional expressions in face-to-face correspondingly elicited in the observer (Hatfield et al., 1993).
communication (Crivelli & Fridlund, 2018; Fridlund, 2014; Martin Extant research supports that vicariously experiencing an emotion
et al., 2017) provides a valuable template for this, with some can influence subsequent judgments and behaviors (Barsade,
approaches making direct reference to emojis as one form of such 2002; Barsade & Gibson, 2007). Again returning to the context of
expressions (van Kleef, 2017). negotiations, prior studies suggest that displays of anger can also
These social-functional approaches to emotions (e.g., Frijda, lead to the exclusion of an angry party from a coalition formation
1986; Keltner & Haidt, 1999; Keltner et al., 2006) posit that facial process (van Beest et al., 2008) and other retaliatory behaviors
expressions of emotion are not simply reflexive by-products of during negotiation (van Kleef & Côté, 2007).
 EMOJIS IN DIGITAL COMMUNICATION 3

Importantly, while the inferential and affective pathways are dis- nonverbal cues in digital communication. Second, after establishing
tinct, they should not be considered independent of one another. The their basic communicative properties, we aimed to test specific pre-
above-mentioned discrepant results on the effects of anger during dictions from the EASI model (van Kleef, 2009) about the social-
negotiations illustrate this: affective reactions (i.e., vicarious anger) emotional functions of emojis. Specifically, we aimed to test whether
can directly lead to retaliatory behaviors (e.g., van Beest et al., 2008). they instigate affective and inferential processing, and thereby affect
However, especially in situations where retaliation competes with social and behavioral outcomes in the same way as facial expressions
other important outcomes, such as negotiations, affect might first and of emotions do in face-to-face communication.
foremost serve as an input for further inferential processing. When To achieve both goals, the general experimental setup was the
considering that angry retaliation likely will stall or end the negotia- same across almost all experiments: participants were asked to
tion process, an individual might adopt different behavioral options read and rate several instant chat messages. Half of these messages
while still feeling angry (e.g., van Dijk et al., 2008). conveyed positive valence while the remaining half conveyed neg-
The EASI model proposes two moderators that describe which ative valence. Orthogonally to this, half of all messages included
of the two pathways will have more sway: information processing an emotion-congruent facial expression emoji while the remaining
and social-relational factors (van Kleef, 2009). First, information messages did not.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Concerning our first research goal, in Experiments 1a partici-

This document is copyrighted by the American Psychological Association or one of its allied publishers.

processing refers to the idea that inferential processes only happen

when the observer of a facial expression has time and motivation pants rated how emotionally intense they perceived these mes-
to process the expression further. This is especially likely in situa- sages, and they indicated how positive or negative they found the
tions of high importance—such as negotiations—and when there messages (perceived message valence). Since emojis reincorporate
are no barriers to systematic processing such as time pressure or at least one nonverbal channel to express emotions into digital
cognitive load. Second, social-relational factors such as norms can communication, we predicted that they would increase a mes-
determine which affective reaction is elicited by a facial expres- sage’s emotional tone or intensity. Furthermore, emotional inten-
sion and which inferences are drawn based on the expression of sity was predicted to be positively correlated with the perceived
emotion. Such factors (among others) include cultural norms about valence of a message. Therefore, we predicted that (a) emotion-
the appropriateness of emotional facial displays. By detailing two congruent emojis increase the emotional intensity, and (b) accentu-
psychological processes (inferential processes and affective reac- ate the perceived valence of a text message, and c) their effect on
tions) and two moderators that constrain activity of those proc- perceived valence should be mediated via increased emotional in-
esses, the EASI model provides a clear theoretical framework tensity. These hypotheses correspond to a qualitative analysis of
from which to derive testable hypotheses. user reports about using emojis in digital communication (Kaye et
Finally, the EASI model not only makes predictions about emo- al., 2016). Experiments 1b–1d were conducted to rule out alterna-
tional facial expressions in face-to-face communication. Rather, tive explanations for these findings and to test the role of demand
van Kleef (2017) states that: within our general experimental setup.
Next, we moved on to test the equivalence of emojis in terms of
“EASI theory posits that expressions of the same emotion that are the affective, inferential, and social consequences that the EASI
emitted via different expressive modalities (i.e., in the face, through model proposes. While the general experimental setup was the
the voice, by means of bodily postures, with words, or via symbols same as in Experiment 1a, the dependent variables in these studies
such as emoticons) have comparable effects, provided that the emo- were modified.
tional expressions can be perceived by others” (p. 213). Concerning the affective pathway of the EASI model, we pre-
dicted that the presence of a (digital) facial expression of emotion
While this functional equivalence tenet has been tested exten- in text messages induces emotion contagion in the recipient of the
sively for voices, bodily postures, and words (for a review, see van message. To assess this, in Experiments 2a–2b we asked partici-
Kleef, 2017); support for the effect of emoticons or emojis through pants to indicate how they themselves felt after reading text mes-
the lens of the EASI model so far has been relatively scarce (see sages with and without emotion-congruent emojis. We expected
our literature review above). Such investigations seem especially that positive messages including an emoji would cause a stronger
warranted given that there are reports suggesting that the last part (i.e., more positive) affective reaction. Conversely, negative mes-
of van Kleef’s quote (“provided that the emotional expression can sages including an emoji were predicted to cause more negative
be perceived by others”) might not always hold true for emojis affective reactions compared to plain negative text messages. Fur-
(Miller et al., 2016). Thus, the second goal of the present studies thermore, we predicted that the strength of emotion contagion
was to test whether emojis fulfill the same social-emotional func- would be tied to the effect of emojis on perceived message va-
tions that the EASI model proposes and thereby their functional lence: the more extreme the perceived valence of a message, the
equivalence to emotional facial expressions of emotions in face- stronger the potential emotion contagion. Thus, we predicted that
to-face communication. the effects of emojis on emotion contagion would be mediated via
perceived message valence.
Overview of the Present Research Similarly, we tested whether (digital) facial expressions of emo-
tion lead to additional inferential processes in Experiments 3a–3b.
As mentioned, the goal of the present research was twofold. First, As a testbed for inferential processing, we assessed the degree to
we aimed to clarify whether or not emojis are able to set an emo- which participants believed they understood the emotional state of
tional tone within digital communication and whether they are able the sender of an instant text-message. We predicted that emojis
to disambiguate the perceived valence of a message. These studies would enhance the level of understanding compared to plain text
aimed to provide an updated assessment of their basic feasibility as messages. Concerning the role of perceived message valence, we
 4 ERLE, SCHMID, GOSLAR, AND MARTIN

Figure 1
Overview of the Present Research
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Note. The top of the figure shows the intended communicative and social-emotional functions that are associated with including an emoji in a message.
The flowchart below shows the variables we used to operationalize these functions, the proposed relations between those variables, and which experi-
ments tested those variables or relations.

predicted that understanding would also be positively related to their functional equivalence to facial expressions of emotion,
the effect of emojis on perceived message valence. The stronger which has been proposed by social-functional theories of emotions
the effect of an emoji on perceived message valence, the more suc- (van Kleef, 2009, 2017). Figure 1 summarizes the central con-
cessful the emoji was in setting an emotional tone. Consequently, structs and their operationalization of all present studies.
inferential processes have a stronger source signal to build upon.
Therefore, we again predicted that the effects of emotion-congru- Power Analyses, Ethics, and Open Practices
ent emojis on understanding (as an indicator of inferential proc-
esses) would be mediated via perceived message valence. All experiments aimed to be sufficiently sensitive to medium-sized
Finally, Experiments 4a–4c took a first step toward testing the effects (h2p = .06) with a power of (1b) = .80. The sample size
needed for this in a 2 3 2 repeated-measures ANOVA is N = 60 (cal-
social functions that emojis serve in digital communication.
culated in G*power; Faul et al., 2007). Experiment 1c was designed
Directly derived from the EASI model, we predicted that both
to mimic the number of datapoints in a between-subjects design and
inferential processes and affective reactions lead to interpersonal
since all participants only rated K = 1 instead of K = 100 messages,
consequences. Specifically, we assessed participants’ empathic
the target was N = 600 participants. For Experiment 1d, we decided
concern (“other-oriented feelings of sympathy and concern”;
to double the sample size (target N = 120) because we anticipated
Davis, 1983, p. 114) for the sender of a message. We decided to
smaller differences due to the design changes made (see below).
measure empathic concern rather than behavior for three reasons:
Based on the effect sizes in Experiments 1a–1b (all hp2s . .30), we
First, behavioral options are usually severely limited in digital
reduced the sample size to N = 40 for Experiment 2a. However,
communication which questions the feasibility of assessing proso-
given the results of Experiment 2a we returned to N = 60 for the re-
cial behavior or intentions in this situation. Second, empathic con- mainder of the studies. The studies received ethical approval at the
cern is reliably correlated with prosociality both in experimental respective institutions, if deemed necessary. We report all measures
(Batson, 2002) as well as survey research (Einolf, 2008), and and manipulations. There were no exclusions in any experiment. All
therefore still provides an adequate test of the proposed conceptual data and materials are available at [Link]
relation to behavior. Third, given that the messages we selected
for the current research were quite mundane in their content, par-
Experiments 1a–1d: Intended Communicative
ticipants might have rightly concluded that no behavior was neces-
Functions of Smileys
sary in response to these messages. Thus, we favored external
validity over a one-to-one correspondence with the EASI model The first four experiments tested whether facial expression emojis
for tests of the social effects of emojis in digital communication. fulfill the communicative functions that users most commonly report
In sum, the present research provides an updated assessment of intending to use them for: setting an emotional tone and reducing am-
how well emojis fulfill their intended role as nonverbal cues in biguity (Kaye et al., 2016). In the present experiments we used emo-
digital communication and for the first time systematically tests tional intensity as a proxy for the emotional tone of a message, and
 EMOJIS IN DIGITAL COMMUNICATION 5

Figure 2
Overview of All Emojis Used in the Present Research

Note. See the online article for the color version of this figure.

perceived message valence as an indicator for the central content that 1F60A; valence: M = 6.29; arousal: M = 4.68). The two negative
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

emojis were the “slightly frowning face” (Unicode: U þ 1F641;

This document is copyrighted by the American Psychological Association or one of its allied publishers.

was to be clarified by emojis. While clearly related, it is important to

point out the differences between emotional intensity and perceived valence: M = 2.35; arousal: M = 4.55) and the “disappointed face”
valence, and how they are likely related. (Unicode: U þ 1F61E; valence: M = 2.13; arousal: M = 4.30).2 All
he emotional tone of a message refers to its delivery, that is, how stimuli of all experiments are available at [Link] and
intensely a verbal communication act is conveyed. For example, in Figure 2 shows all emojis used in the present research.
face-to-face communication the same verbal content can be conveyed Participants were asked to rate the emotional intensity of every
in an indifferent inflection of tone or with a level of enthusiasm. As a message on a scale from 0 (“not emotional at all”) to 10 (“very
common example in the English language, consider the statement emotional”), and its valence on a scale from –5 (“very negative”)
“I’m fine.” Although the valence of this message is the same irre- to 5 (“very positive”). These two ratings were analyzed using a 2
spective of how it is presented (the speaker conveys that they are (Message Valence: Negative vs. Positive) 3 2 (Emoji present:
well, and thus something positive), the delivery affects its perceived Yes vs. No) repeated-measures ANOVA.
valence. When spoken with indifference, it is usually perceived as
Experiment 1b
the standard answer to the communicative ritual of starting a conver-
sation by asking how someone else is doing and it is unlikely that the Since we adopted a within-subjects design, it was possible that
receiver will further process this statement. When spoken enthusiasti- participants gained insight into the crucial manipulation of the
cally, however, the receiver likely will expect that something positive study and guessed that they were expected to alter their responses
has happened to the sender of the message and thus perceive the va- whenever an emoji was used. To investigate such potential effects,
lence of the message as more positive. Thus, higher emotional inten- in Experiment 1b half of participants (n = 68) were told that the
sity correlates with stronger perceived valence and based on this we emojis were randomly added to the messages by the computer pro-
predicted that the effects of emojis on perceived valence would be gram, while the remaining participants (n = 69) were not given
mediated by increased emotional intensity—if emojis indeed are any information. Thus, in the randomly generated emoji condition
equivalent to nonverbal cues in face-to-face communication. participants had no reason to believe that the experimenter strate-
Experiment 1b–1c also tested the role of experimental demand gically placed the emojis in certain messages to evoke specific
for these effects. Experiment 1d tested the alternative explanation responses, and therefore comparing the results of this condition
that smileys affect valence ratings directly. with the uninformed control condition allowed us to gauge the
impact of experimenter expectations on the observed results. Oth-
Method erwise, Experiment 1b was identical to Experiment 1a.
Data of Experiment 1b were analyzed using a 2 (Message Va-
In all four experiments, participants were presented with What- lence: Negative vs. Positive) 3 2 (Emoji present: Yes vs. No) 3 2
sApp messages and were instructed to imagine a friend/acquaint- (Emoji Generation: Randomly Generated vs. Control [no informa-
ance had sent the messages. In Experiments 1a–1c half of these tion]; between) mixed ANOVA.
messages were positive (e.g., “I am happy”) and half were negative
(e.g., “I am sad”). In Experiment 1d there were also neutral mes- Experiment 1c
sages (e.g., “That’s possible”) and the three valences (positive, neg- To most conclusively rule out demand effects, in this experi-
ative, and neutral) were evenly split across trials. The following ment every participant only rated one text message that was either
sections elaborate the crucial differences between these studies. positive or negative and sent with or without a smiley (the smileys
Experiment 1a used here were the “smiling face with smiling eyes” and the “dis-
appointed face” emojis). Only one trial per participant was used
Participants saw a total of K = 100 messages. Randomly because presenting participants with multiple messages that all
selected, half of the positive and half of the negative messages were sent with or without an emoji, could have obscured their
were presented with or without one of two emotion-congruent informational value. This would be akin to a person always using
emojis (iOS sourced). The two positive emojis were the “slightly
smiling face” (Unicode: U þ 1F642; valence: M = 5.80; arousal: 2
The normative ratings for the emojis were taken from Rodrigues and
M = 3.66) and the “smiling face with smiling eyes” (Unicode: U þ colleagues (2018).
 6 ERLE, SCHMID, GOSLAR, AND MARTIN

the same inflection of tone during face-to-face communication, neutral, and negative messages. If the effects of smileys were due
which erases any inherent meaning this tone might have had ini- to their mere perception, on the other hand, the emoji alone would
tially. Thus, Experiment 1c had the same 2 (Message Valence: determine their perceived valence. Granted, it seems implausible
Negative vs. Positive) 3 2 (Emoji present: Yes vs. No) design as to assume that a negative message with a positive emoji would be
Experiment 1a, but instead of a within-subjects design, here a perceived as positive, but the comparison between neutral and pos-
between-subjects design was implemented. itive/negative messages with an emotion-congruent emoji provides
Experiment 1d a fair test of this alternative mechanism. Based on the idea that
smileys directly affect valence judgments by virtue of their mere
Experiment 1d served the purpose of ruling out an alternative ex- perception, (at least) neutral and positive/negative messages that
planation of the predicted results. While we so far assumed that emo- include the same positive/negative emoji should be rated as
jis affect the perceived valence of a message via increased emotional equally positive/negative. To test this, we compared neutral and
intensity, alternatively smileys also could directly cause the previ- positive as well as neutral and negative text messages featuring
ously observed outcomes. That is, smileys could be emblematic for positive/negative emojis with each other. In all experiments, par-
valence and directly translate into valence judgments. This mecha-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

ticipants provided demographic data at the end of the session.

This document is copyrighted by the American Psychological Association or one of its allied publishers.

nism holds that their effects are not related to social information dur-
ing communication at all, but rather perceptual in nature. Samples
To address thus, Experiment 1d had a 3 (Message Valence:
In Experiment 1a, of N = 59 students (n = 44 female; age: M =
Negative vs. Neutral vs. Positive) x 3 (Emoji: Negative vs. Neutral
27.59, SD = 10.24) from the University of Würzburg participated
vs. Positive) within-subjects design. Participants rated K = 18 posi-
in a 60-minute laboratory study involving another irrelevant task
tive, negative, or neutral messages. The messages’ structure was
(Krishna & Eder, 2018). Experiment 1a took 15 minutes to com-
kept as parallel, yet as distinct in valence as possible (e.g., positive
plete. Participants received a compensation of 8 e.
version: “That’s clever,” neutral: “That’s possible,” negative:
In Experiment 1b, N = 137 students (n = 103 female; n = 31
“That’s stupid”). In this experiment, participants only rated the va-
lence of the presented messages on a scale from 1 (“very nega- male; n = 3 other; age: M = 22.98, SD = 3.65) from the University
tive”) to 7 (“very positive”). of Cologne participated in a 30-minute laboratory study involving
N = 46 participants pilot tested the valence of the text messages other irrelevant tasks (Erle & Topolinski, 2018). Experiment 1b took
for the main study on a scale from 1 (very negative) to 7 (very pos- 15 minutes to complete. Participants received compensation of 4 e.
itive). Message triplets with the following average valences were Experiments 1c and 1d were conducted online on Prolific Aca-
selected for the main study: positive (M = 5.79, SD = .88), neutral demic. In Experiment 1c, N = 607 individuals (n = 238 female, n =
(M = 4.17, SD = .96), and negative (M = 2.45, SD = .97). 366 male, n = 3 diverse; age: M = 27.43, SD = 9.79) participated
Orthogonally to the messages’ content, all messages were pre- for a financial compensation of .25£.
sented with a negative, neutral, or positive emoji. As the negative In Experiment 1d, N = 135 individuals (n = 56 female, n = 77
emoji, we chose the “Pensive Face” (M = 2.4, SD = 1.26, U þ male, n = 2 diverse; age: M = 29.01, SD = 9.67) participated for a
1F614). For the neutral emoji, we presented the “Hushed Face” financial compensation of .5£.
(M = 4.00, SD = .84, U þ 1F62F). For the positive emoji, we used
the “Smiling Face with Open Mouth” (M = 5.78, SD = 1.26, U þ Results
1F60A; for ratings, see footnote 2). These smileys were selected to
match the valence ratings of the message texts. The smileys were All descriptive statistics for Experiments 1a–1d can be found in
WhatsApp Version 2.19.352 sourced. Tables 1–2.
Based on the idea that smileys are nonverbal communication Experiment 1a
cues, we expected main effects of verbal (the message) and non-
verbal (emojis) cues on perceived valence. Specifically, any mes- For emotionality, there was only a significant main effect of
sage that is accompanied by a positive/negative emoji should be Emoji, F(1, 58) = 28.50, p , .001, h2p = .33. Messages including
perceived as more valent in line with the emoji’s valence. Impor- an emoji (M = 4.66; SD = 1.73) were rated as more emotionally
tantly, if participants process the verbal content of the message as intense than messages without emojis (M = 3.68; SD = 1.85). The
well, there should also be inherent differences between positive, effect size was medium to large, dz = .70. The main effect of
Table 1
Descriptive Statistics in Experiments 1a–1c
Positive messages Negative messages
Exp. Dependent variable No emoji With emoji No emoji With emoji
1a Valence 2.16 (0.86) 2.43 (0.91) 2.38 (0.80) 2.62 (0.74)
Emotional intensity 3.68 (1.97) 4.76 (1.84) 3.69 (1.85) 4.57 (1.72)
1b Valence 2.17 (1.07) 2.49 (1.08) 2.51 (0.97) 2.68 (0.97)
Emotional intensity 3.22 (1.91) 4.29 (2.10) 3.32 (1.81) 4.46 (1.97)
1c Valence 3.38 (1.93) 3.97 (1.28) 2.47 (1.57) 3.08 (1.52)
Emotional intensity 4.81 (2.96) 5.60 (2.41) 5.52 (2.55) 5.88 (2.56)
Note. Table displays means (and standard deviations). Valence was measured on a scale from 5 to 5. Emotionality was measured on a scale from 0 to 10.
 EMOJIS IN DIGITAL COMMUNICATION 7

Table 2 not reduce the direct effect to nonsignificance, B = .80, 95% CI [.35,
Descriptive Statistics in Experiment 1d 1.25], SE = .23. Thus, the mediation analyses favor the theoretically
predicted model.
Message type M SD
Positive message, Negative smiley 4.09 1.28 Experiment 1b
Positive message, Neutral smiley 5.51 0.94
Positive message, Positive smiley 6.24 0.77
For emotional intensity, the ANOVA yielded a main effect of
Neutral message, Negative smiley 3.07 0.89 Message Valence, F(1, 135) = 4.43, p = .037, h2p = .03, indicating
Neutral message, Neutral smiley 4.33 0.62 that negative messages were rated as more emotionally intense
Neutral message, Positive smiley 5.44 0.90 than positive messages. More importantly, there was a main effect
Negative message, Negative smiley 2.00 0.78 of Emoji, F(1, 135) = 100.32, p , .001, h2p = .43. Messages with
Negative message, Neutral smiley 2.62 0.81
Negative message, Positive smiley 3.53 1.31 emojis (M = 4.37; SD = 1.98) were more emotionally intense than
messages with no emoji (M = 3.27; SD = 1.80). The effect size
Notes. Valence was assessed on a scale from 1 to 7. was large, dz = .86. All other effects were not significant, all Fs ,
2.79, all ps $ .097.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Message Valence and the interaction effect were nonsignificant,

This document is copyrighted by the American Psychological Association or one of its allied publishers.

For valence, there were main effects of Emoji, F(1, 135) = 6.45,
both Fs , 1.93, both ps $ .170. p = .012, h2p = .05, indicating that ratings were more positive after
For valence, there was a significant main effect of Message Va- messages including emojis, and Message valence, F(1, 135) =
lence, F(1, 58) = 622.27, p , .001, h2p = .92, indicating that posi- 900.08, p , .001, h2p = .87, indicating that positive messages were
tive messages were rated as more positive than negative messages. perceived as more positive than negative messages. More impor-
More importantly, there was a significant Emoji 3 Message Va- tantly and qualifying these main effects, the predicted Emoji 3
lence interaction. Positive messages with an emoji were perceived Message Valence interaction was significant, F(1, 135) = 60.76,
as more positive than positive messages without an emoji, t(58) = p , .001, h2p = .31. Messages including and emoji were rated as
4.02, p , .001, dz = .52, and negative messages with an emoji more positive/negative than messages without an emoji. This was
were perceived as more negative than negative messages without true for both the Control (no information) condition, t(68) = 5.94,
an emoji, t(58) = 5.13, p , .001, dz = .67. The main effect of p , .001, dz = .72 for positive messages, t(68) = 2.78, p = .007,
Emoji was not significant, F(1, 58) = .31, p = .579. dz = .33 for negative messages, and the Randomly Generated
Following the analyses on the mean level, a mediation analysis Emoji condition, t(67) = 3.66, p , .001, dz = .44 for positive mes-
with Emoji (0 = No vs. 1 = Yes) as predictor, Emotional Intensity sages, t(67) = 4.16, p , .001, dz = .50 for negative messages.
as mediator, and Perceived Valence as criterion was conducted Importantly, for both dependent variables there was no significant
using the PROCESS Macro for SPSS (model 4; Hayes, 2013). effect involving Emoji Generation, all Fs , 2.79, all ps $ .097. For
Since the same predictions was made for positive and negative both Emotional Intensity and Valence, we computed a 2 (Emoji Gen-
messages, Message Valence was transformed to absolute values to eration: Randomly Generated vs. Control [no information]; between)
map all valence ratings onto a scale from 0 (low valence) to 5 3 2 (Message Valence: Negative vs. Positive) x 2 (Emoji present:
(high valence). Regression weights and 95% CIs were estimated Yes vs. No) mixed Bayesian ANOVA to quantify the evidence for
using 5,000 bootstrap samples. The correlation between variables the absence of an effect of Emoji Generation. Bayes Factors for the
used for mediation analyses in all studies can be seen in Table 3. inclusion of the main effect of Emoji Generation and interactions
The model explained a significant proportion of variance, F(2, including this factor were derived, which quantify the extent to which
233) = 16.48, p , .001, R2 = .35. There was a significant and positive the data support their inclusion. These Bayes Factors all spoke
indirect effect of Emojis on Message Valence via Emotional inten- against the inclusion of the Emoji Generation variable, all BF01 .
sity, B = .14, 95% CI [.07, .25], SE = .05, which reduced the direct 11.23, range: 11.24–250.00. Bayes factors of this magnitude are con-
effect of Emojis on Message Valence to nonsignificance, B = .11, ventionally described as “strong” to “very strong” evidence against
95% CI [.10, .32], SE = .11. Thus, the effect of emojis on valence the inclusion of these effects (Jeffreys, 1961; p. 432).
was fully statistically mediated by message emotional intensity. We
Experiment 1c
also tested the reverse mediation model with Message Valence as
mediator and Emotional Intensity as criterion, but in this model there For Emotionality, the ANOVA yielded main effects of Message
was an indirect effect, B = .18, 95% CI [.03, .40], SE = .09, that did Valence, F(1, 599) = 5.42, p = .020, h2p = .01 (indicating that negative

Table 3
Correlations Between Variables for All Reported Mediation Analyses
Experiment
Correlation tested Exp. 1a Exp. 1b Exp. 2b Exp. 3b Exp. 4b Exp. 4c
Emotional intensity—Perceived valence .35 .50 — — — —
Perceived valence—Emotion contagion — — .68 — — —
Perceived valence—Understanding — — — .37 — —
Emotion contagion—Empathic concern — — — — .34 —
Understanding—Empathic concern — — — — — .51
Note. All correlations are significant at the p , .001 level.
 8 ERLE, SCHMID, GOSLAR, AND MARTIN

messages were perceived as slightly more emotionally intense than directly by perceptual means and demonstrated that participants
positive messages), and Emoji, F(1, 599) = 7.21, p = .007, h2p = .01. pay attention both to the verbal and nonverbal contents of the mes-
Messages including an emoji (M = 5.73; SD = 2.78) were rated as sages in our experiments. However, as noted by an anonymous
more emotionally intense than messages without emojis (M = 5.17; reviewer, while this study provides evidence that participants
SD = 2.48). The interaction between the two independent variables clearly process both the verbal and nonverbal valence cues in our
was not significant, F(1, 599) = 1.03, p = .311. messages, it does not provide direct support for the idea that emo-
For Valence, the ANOVA yielded main effects of Message Va- jis accentuate the meaning of a text and leaves open the possibility
lence, F(1, 599) = 2482.65, p , .001, h2p = .81 (indicating that negative that other mechanisms produced the observed pattern of results.
messages were perceived as more negative than positive messages), Such evidence could be provided only by studies comparing va-
and a significant Message Valence x Emoji interaction, F(1, 599) = lence ratings of emojis to valence ratings of emojis in text mes-
21.77, p , .001, h2p = .04. Positive messages with an emoji were per- sages. However, such comparisons are problematic themselves as
ceived as more positive than positive messages without an emoji, an emoji alone does not represent a communicative act, and the
t(303) = 3.17, p = .002, d = .36, and negative messages with an emoji mechanisms involved in evaluating communication differ from
were perceived as more negative than negative messages without an those employed to evaluate a picture alone.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Finally, an unexpected finding in most of these studies was that

This document is copyrighted by the American Psychological Association or one of its allied publishers.

emoji, t(296) = 3.44, p = .001, d = .40. The main effect of Emoji was
not significant, F(1, 599) = .01, p = .929. The main effect of Emoji the impact of emojis on our dependent measures was stronger
was not significant, F(1, 599) = 1.03, p = .311. effects for positive than for negative text messages. Since this pat-
The effect sizes in this experiment, however, were only small to tern persisted also in the later studies of the article, we will return
medium (emotionality: d = .21, valence positive messages: d = to it in the general discussion. Therefore, irrespective of the exact
.36, valence negative messages: d = .40), indicating that experi- mechanism assumed and this asymmetry in valence, the first con-
mental demand might have affected the previous results, although clusion that is still supported by our data is that emojis are effec-
it cannot fully account for the observed differences. tive at what they were designed to do: they reintroduce emotional
intensity into digital communication: if one seeks to emphasize the
Experiment 1d emotional content of one’s messages, emojis are a viable tool—
both for positive and negative emotions.
Speaking against the alternative explanation of our previous
data, the ANOVA yielded significant main effects of Message Va-
lence, F(2, 133) = 317.00, p , .001, h2p = .83, Emoji, F(2, 133) = Experiments 2a-2b: Affective Reactions
215.02, p , .001, h2p = .76, and a significant interaction between Experiments 2a–2b focused on the first pathway of the EASI
these factors, F(4, 131) = 16.82, p , .001, h2p = .34. model and specifically tested the effects of emojis on emotion con-
Planned comparisons revealed that both verbal (text) and nonverbal tagion (Hatfield et al., 1993). We predicted that since emojis signal
(emojis) cues affected valence ratings. Speaking against the idea that more emotional intensity and thereby accentuate the valence of a
smileys directly translate into valence judgments irrespective of the message, they increase the chance of evoking a congruent emo-
message content, positive messages with positive smileys (M = 6.24, tional response in the recipient. Thus, emojis were predicted to
SD = .77) were perceived as more positive than neutral messages increase emotion contagion (i.e., they lead to more negative feel-
with positive smileys (M = 5.51, SD = .94), t(134) = 11.42, p , ings in the participant after negative messages and to more posi-
.001, dz = .87. Similarly, negative messages with negative smileys tive feelings after positive messages). This effect was tested in
(M = 2.00, SD = .78) were perceived as more positive than neutral both Experiments 2a and 2b. Furthermore, we predicted this effect
messages with negative smileys (M = 3.07, SD = .89), t(134) = to be mediated by increased perceived message valence, which we
12.46, p , .001, dz = 1.08. tested in Experiment 2b.

Discussion of Experiments 1a–1d: Intended Method

Communicative Functions of Smileys
Experiments 2a–2b were identical to Experiment 1a, but instead of
The results of the first experiments largely suggest that emojis rating emotional intensity and valence, in Experiment 2a participants
produce effects that are functionally equivalent to effects observed rated how they felt themselves after reading the message on a scale
for facial expressions of emotion in face-to-face communication. from 5 (“very negative”) to 5 (“very positive”) as a measure of emo-
Experiments 1a–1c demonstrated that they increase the perceived tion contagion. In addition to participants’ ratings of their own emo-
emotional intensity of a message and all four experiments showed tional states, message valence was assessed in Experiment 2b as in
that they also accentuate the perceived valence of a message and Experiments 1a-1c. M ratings on these variables were analyzed using
thus its central content. In Experiment 1a, we also found evidence a 2 (Valence Rating: Negative vs. Positive) 3 2 (Emoji presence:
that their effects on valence are mediated via emotional intensity, Yes vs. No) repeated-measures ANOVA. In both experiments, par-
and results spoke against the inverse mediation path (their effects ticipants provided demographic data at the end of the session.
on emotional intensity are mediated via perceived valence).
Samples
Experiments 1b–1c ruled out that these findings are due to expect-
ations about the experimenter’s research goal (Exp. 1b) or other Students from the University of Würzburg (Experiment 2a, N =
demand effects (Exp. 1c), although effect sizes were reduced 40, 23 female; age: M = 22.15, SD = 3.24; Experiment 2b, N = 61,
when using a between-subjects design (Exp. 1c). Finally, Experi- 29 female; age: M = 24.25, SD = 3.42) participated in a 10-minute
ment 1d suggests that smileys do not affect valence judgments experiment in exchange for a candy bar.
 EMOJIS IN DIGITAL COMMUNICATION 9

Results Contagion as criterion was conducted as in Experiment 1a (model

4; Hayes, 2013).
All descriptive statistics of Experiments 2a-4c are displayed in The model explained a significant proportion of variance, F(2,
Table 4. 241) = 104.87, p , .001, R2 = .47. There was an indirect effect of
Emotion Contagion Emojis on Emotion contagion via Message Valence, B = .28, 95%
CI [.15, .43], SE = .07, which reduced the direct effect of Emoji on
In both Experiments, there were main effects of Message Valence, Emotion contagion to nonsignificance, B = .09, 95% CI [.06,
both Fs . 294.15, both ps , .001, both hp2s . .87, indicating that par- .24], SE = .08. Thus, the effect of emojis on emotion contagion
ticipants felt more positive after positive messages compared to nega- was fully mediated by increased perceived valence. Within the
tive messages. There was also a main effect of Emoji in both reverse mediation model there was again an indirect effect, B =
experiments, both Fs . 15.98, both ps , .001, both hp2s . .23, indi- .24, 95% CI [.12, .39], SE = .07, that did not reduce the direct
cating that valence ratings were more positive after messages with effect to nonsignificance, B = .18, 95% CI [.04, .33], SE = .07,
emojis than without. More importantly and qualifying these main thus favoring the theoretically predicted model.
effects, both in Experiment 2a, F(1, 39) = 15.82, p , .001, h2p = .29,
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

and Experiment 2b, F(1, 60) = 27.33, p , .001, h2p = .31, the predicted
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Experiments 3a-3b: Inferential Processing

Emoji x Message Valence interaction was significant. In Experiment
2a, positive messages with an emoji elicited more emotion contagion Experiments 3a-3b went on to test the second pathway pro-
than messages without an emoji, t(39) = 5.11, p , .001, dz = .81, and posed in the EASI model (van Kleef, 2009). In analogy to facial
unexpectedly negative messages with an emoji did not, t(39) = 1.05, expressions of emotion in face-to-face communication, we pre-
p = .298, dz = .17. In Experiment 2b both comparisons were signifi- dicted that because emojis represent a nonverbal cue that disam-
cant, positive messages: t(60) = 5.74, p , .001, dz = .74, negative mes- biguates the emotional state of a person, they can instigate
sages: t(60) = 3.10, p = .003, dz = .40. inferential processes about the emotional state of that person.
Given that we used only clearly valent messages, emotion-con-
Valence gruent emojis, and that we standardized the relational schema
between participants and the senders, we predicted that these
Experiment 2b replicated the findings of Experiments 1a-1b
inferences would be straightforward. Thus, we predicted that
with significant main effects of Message Valence, F(1, 60) =
emojis would enhance the degree to which participants think
985.31, p , .001, h2p = .94, and Emoji, F(1, 60) = 16.31, p , .001,
they understand the sender’s emotional state. In Experiment 3b
h2p = .21 (see Experiments 1a-1c). More importantly and qualifying
we additionally assessed perceived message valence and we pre-
these main effects, the predicted Emoji x Message Valence inter-
dicted that the effect of emojis on understanding would be medi-
action was significant, F(1, 60) = 32.15, p , .001, h2p = .35. Posi-
ated by enhanced message valence, because the stronger an
tive messages with an emoji were perceived as more positive than
emotional expression, the more likely it is that it will be per-
positive messages without an emoji, t(60) = 6.20, p , .001, dz =
ceived by another person and the higher the likelihood that this
.79, and negative messages with an emoji were perceived as more
person then will engage in further inferential processing.
negative than negative messages without an emoji, t(60) = 3.32,
p = .001, dz = .43.
Method
Mediation
The general setup for Experiments 3a-3b was again similar as in
For Experiment 2b, a mediation analysis with Emoji (0 = No vs. the previous studies. However, in Experiment 3a participants rated
1 = Yes) as predictor, Message Valence as mediator, and Emotion how well they understood sender’s emotional state when writing the

Table 4
Descriptive Statistics in Experiments 2a4c
Positive messages Negative messages
Exp. Dependent variable No emoji With emoji No emoji With emoji
2a Emotional contagion 1.81 (0.77) 1.89 (0.82) 1.61 (0.74) 2.10 (0.76)
2b Valence 2.23 (0.80) 2.83 (0.82) 2.32 (0.10) 2.57 (0.65)
Emotional contagion 1.33 (0.75) 1.87 (0.88) 1.36 (0.65) 1.57 (0.71)
3a Understanding 1.09 (1.82) 3.03 (0.85) 0.05 (1.98) 1.06 (2.71)
3b Valence 2.33 (0.91) 3.02 (0.91) 2.61 (0.68) 2.81 (0.76)
Understanding 5.92 (1.54) 7.19 (0.91) 4.94 (1.89) 5.57 (2.13)
4a Empathic concern 1.09 (1.12) 1.90 (1.18) 0.18 (1.34) 0.33 (1.81)
4b Emotional contagion 1.10 (0.85) 1.59 (0.94) 1.11 (0.99) 1.30 (1.31)
Empathic concern 1.56 (1.75) 2.06 (2.01) 1.18 (1.32) 1.66 (1.76)
4c Understanding 6.66 (1.38) 7.58 (1.12) 4.93 (2.16) 5.60 (2.64)
Empathic concern 1.04 (0.94) 1.70 (0.98) 0.91 (1.17) 1.39 (1.50)
Note. Table displays means (and standard deviations). Emotional Contagion measured on a scale from 5 to 5. Understanding was measured on a scale
from 0 to 10, except in Experiment 3a (scale from 5 to 5).
 10 ERLE, SCHMID, GOSLAR, AND MARTIN

message on a scale from 5 (“very badly”) to 5 (“very well”) as a The model explained a significant proportion of variance, F(2,
(subjective) measure of inferential processing (the scale ranged from 237) = 23.22, p , .001, R2 = .16. There was an indirect effect of
0 to 10 as in the other studies using this variable). In Experiment 3b, Emoji on Understanding via Message Valence, B = .31, 95% CI
they additionally rated the messages’ valence as in Experiments 1a- [.15, .52], SE = .09. This indirect effect did not reduce the direct
1c and 2b. M ratings on these dependent variables were analyzed effect of Emojis on Understanding to nonsignificance, B = .64,
using a 2 (Message Valence: Negative vs. Positive) x 2 (Emoji: Yes 95% CI [.19, 1.09], SE = .23. Thus, the effect of emojis on under-
vs. No) repeated-measures ANOVA. After participants rated all mes- standing was partially mediated by perceived message valence.
sages, demographic data were assessed. The reverse mediation model this time suggested an equally strong
partial mediation with a significant indirect effect, B = .14, 95%
Samples CI [.07, .24], SE = .04, that did not reduce the direct effect to non-
significance, B = .30, 95% CI [.10, .51], SE = .10. Therefore, the
In Experiment 3a, N = 59 students (n = 39 female, n = 19 male,
mediation analyses favored neither the theoretically predicted nor
n = 1 missing; age: M = 22.19, SD = 3.96) from the University of
the reverse mediation model.
Würzburg participated in a 10-minute experiment in exchange for
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

a candy bar.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

In Experiment 3b, N = 60 students (n = 33 female; age: M = Experiments 4a–4c: Behavioral Intentions

22.27, SD = 3.00) from the University of Würzburg participated in Experiments 4a-4c investigated the effects of emojis on em-
a 15-minute experiment in exchange for a candy bar. pathic concern, a social affective reaction that reliably correlates
with behavior and behavioral intentions following emotional expe-
Results riences (see Batson, 2002; Einolf, 2008). We predicted that com-
pared to messages without an emoji, emojis would lead to higher
All descriptive statistics of Experiments 2a-4c are displayed in
empathic concern toward the sender of a message. Furthermore, in
Table 4.
line with the EASI model (van Kleef, 2009) this effect was
Understanding expected to be mediated by affective reactions (emotion conta-
gion) and inferential processes (understanding). In Experiment 4b,
In both Experiments, there were unexpected main effects of Mes- we thus concurrently assessed emotion contagion and predicted
sage Valence, both Fs . 21.86, both ps , .001, both hp2s $ .27, indi- that emojis would lead to greater empathic concern via emotion
cating that participants understood the sender’s emotional state better contagion: As participants experience the sender’s emotions more
for positive compared to negative messages. As predicted, in both strongly, they develop more empathic concern (i.e., vicarious joy
experiments there was also a main effect of Emoji, both Fs . 36.47, for positive messages and compassion for negative messages). In
both ps , .001, both hp2s . .38, indicating that emojis increased Experiment 4c, we predicted the same mediation for understand-
understanding. Unexpectedly and qualifying these main effects, both ing: As participants better understand the sender’s emotional state,
in Experiment 3a, F(1, 58) = 14.89, p , .001, h2p = .20, and Experi- more empathic concern develops.
ment 3b, F(1, 59) = 16.28, p , .001, h2p = .22, there was an Emoji x
Message Valence interaction. Although emojis increased understand- Method
ing for both positive and negative of messages, Experiment 3a: Posi-
tive messages: t(58) = 8.75, p , .001, dz = 1.14, negative messages: In Experiment 4a participants were asked what their feelings to-
t(58) = 4.37, p , .001, dz = .57, Experiment 3b:, Positive messages: ward the sender of the message were on a scale from 5 (“very
t(59) = 6.41, p , .001, dz = .83, negative messages: t(59) = 4.22, p , cold feelings”) to 5 (“very warm feelings”) as a measure of em-
pathic concern (see, e.g., Batson et al., 1997). In Experiment 4b,
.001, dz = .54, this effect was larger for positive messages.
also emotion contagion was assessed as in Experiments 2a–2b,
Valence and in Experiment 4c, also their understanding was assessed as in
Experiments 3a-3b. M ratings on these variables were analyzed
Experiment 3b replicated the findings of Experiments 1a-1b and using a 2 (Message Valence: Negative vs. Positive) 3 2 (Emoji:
2b with significant irrelevant main effects of Message Valence, F Yes vs. No) repeated-measures ANOVA. After participants rated
(1, 59) = 923.69, p , .001, h2p = .94, and Emoji, F(1, 59) = 22.90, all messages, demographic data were assessed.
p , .001, h2p = .28. More importantly and qualifying these main
effects, the predicted Emoji 3 Message Valence interaction was Samples
significant, F(1, 59) = 31.65, p , .001, h2p = .35. Positive messages
In Experiment 4a, N = 61 students (n = 39 female; age: M =
with an emoji were perceived as more positive than those without 21.43, SD = 2.99) from the University of Würzburg participated in
an emoji, t(59) = 6.21, p , .001, dz = .80, and negative messages a 10-minute experiment for a candy bar.
with an emoji were perceived as more negative than those without In Experiment 4b, N = 54 students (n = 41 female, n = 12 male,
an emoji, t(59) = 2.75, p = .008, dz = .36. n = 1 other; age: M = 23.04, SD = 6.60) from the University of Co-
Mediation logne participated in a 15-minute experiment in exchange for a
candy bar.
For Experiment 3b, a mediation analysis with Emoji (0 = No vs. In Experiment 4c, N = 63 students (n = 46 female, n = 13 male,
1 = Yes) as predictor, Message Valence as mediator, and Under- n = 1 other, n = 3 missing; age: M = 22.28, SD = 2.79) from the
standing as criterion was conducted as in previous experiments University of Cologne participated in a 15-minute experiment in
(model 4; Hayes, 2013). exchange for a candy bar.
 EMOJIS IN DIGITAL COMMUNICATION 11

Results emotion contagion. The reverse mediation model again indicated a

partial mediation pattern with an indirect effect, B = .08, 95% CI
All descriptive statistics of Experiments 2a–4c are displayed in [.01, .16], SE = .04, that did not reduce the direct effect to non-
Table 4. significance, B = .30, 95% CI [.09, .51], SE = .11. Thus, the media-
Empathic Concern tion analysis favored the theoretically predicted model.
The model for Experiment 4c also explained a significant pro-
In Experiments 4a-4b, both Fs . 4.85, both ps # .032, both portion of variance, F(2, 249) = 50.71, p , .001, R2 = .29. There
hp2s . .08, but not Experiment 4c, F(1, 62) = 2.52, p = .117, there was an indirect effect of Emoji on Empathic Concern via Under-
was a main effect of Message Valence. Participants reported more standing, B = .17, 95% CI [.06, .31], SE = .06. This indirect effect
of empathic concern after positive messages. Finally, in Experi- did not reduce the direct effect of Emoji on Empathic Concern to
ment 4a there was a significant Message Valence 3 Emoji interac- nonsignificance, B = .36, 95% CI [.15, .57], SE = .11. Thus, the
tion, F(1, 60) = 11.11, p = .001, h2p = .16 (Experiments 4b-4c: effect of emojis on empathic concern was partially mediated by
Both Fs , 3.17, both ps $ .080), indicating that the increase of understanding. The reverse mediation test with Empathic Concern
empathic concern for messages sent with an emoji was larger for as mediator and Understanding as criterion surprisingly indicated
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

positive messages. a full mediation with an indirect effect, B = .58, 95% CI [.33, .89],
This document is copyrighted by the American Psychological Association or one of its allied publishers.

More importantly and as predicted, in all experiments there was SE = .14, that reduced the direct effect to nonsignificance, B = .22,
a significant main effect of Emoji, all Fs . 25.39, all ps , .001, 95% CI [.26, .70], SE = .25. Therefore, in this case the mediation
all hp2s . .29. In Experiments 4a–4c, positive messages including analyses favored the reverse mediation model over the theoreti-
an emoji evoked more empathic concern in the recipient than those cally predicted one.
without an emoji, all ts . 4.91, all ps , .001, all dzs . .66. For
negative messages, emojis significantly enhanced empathic con- Discussion of Experiments 2a–4c: Social-Emotional
cern only in Experiments 4b-4c, both ts . 4.64, both ps , .001, Functions of Smileys
both dzs . .59, whereas this difference was in the same direction,
but nonsignificant in Experiment 4a, t(60) = 1.13, p = .261. Our second goal was to test whether emoji affect the two path-
ways of the EASI model in similar ways as facial expressions of
Emotion Contagion emotion in face-to-face communication or put differently: their
functional equivalence to facial expressions (van Kleef, 2017).
In Experiment 4b, there were again main effects of Message Va-
On the mean level, all our predictions were confirmed: across
lence, F(1, 53) = 122.99, p , .001, h2p = .70, and Emoji, F(1, 53) =
seven experiments, emojis enhanced emotion contagion as an in-
4.58, p = .037, h2p = .08 (see Experiments 2a-2b). More importantly
dicator of the affective pathway, they enhanced understanding as
and qualifying these main effects, the predicted Emoji x Message
an indicator of the inferential pathway, and they also increased
Valence interaction was significant, F(1, 53) = 35.46, p , .001, h2p =
empathic concern as an indicator of behavioral intentions result-
.40. Both positive, t(53) = 5.05, p , .001, dz = .69, and negative mes-
ing from affective and inferential processes. Thus, on the mean
sages including an emoji, t(53) = 2.42, p = .019, dz = .33, elicited
level emojis were functionally equivalent to real facial expres-
more emotion contagion than messages without an emoji.
sions of emotion.
Understanding In terms of procedural characteristics, however, we could only
partially confirm the proposed archetype of the EASI model. In
In Experiment 4c, there was a main effect of Message Valence, terms of the affective pathway, our results perfectly mirror the
F(1, 62) = 38.51, p , .001, h2p = .39, and a Message Valence 3 EASI model: emojis increase emotion contagion (Exps. 2a-2b &
Emoji interaction, F(1, 62) = 4.13, p = .046, h2p = .06 (see Experi- 4b), and mediation analyses suggested that this fully explains their
ments 3a–3b). Finally and most importantly, as predicted there effect on empathic concern (Exp. 4b). One slight caveat about
was a main effect of Emoji, F(1, 62) = 32.27, p , .001 h2p = .34, these relations is that we did not assess actual behavior, but rather
indicating that understanding of the sender was higher after mes- a variable that is closely related to behavior (Batson, 2002; Einolf,
sages including emojis. This was true for positive, t(62) = 6.12, p 2008). Another limitation that relates to the findings for the affec-
, .001, dz = .77, and negative messages, t(62) = 4.31, p , .001, dz tive pathway of the EASI model is the hypothetical nature of our
= .54. studies. Since we only asked participants to imagine receiving
Mediations these messages from friends, it is possible that instead of actual
emotion contagion, participants merely reported their expected
For Experiments 4b-4c, mediation analyses with Emoji (0 = No emotion contagion in this situation.
vs. 1 = Yes) as predictor, Emotion contagion/Understanding as In terms of the inferential pathway, our results were mixed and
mediators, and Empathic Concern as criterion were done as in pre- mostly did not mirror what we predicted based on the EASI
vious experiments (model 4; Hayes, 2013). model. First, although emojis increased understanding (Exp.
The model for Experiment 4b explained a significant proportion 3a–3b & 4c), this effect was not mediated via their effects on per-
of variance, F(2, 213) = 11.23, p = .001, R2 = .22. There was an ceived valence (Exp. 3b). Mediation analyses rather suggested a
indirect effect of Emoji on Empathic Concern via Emotion conta- bidirectional link between understanding and perceived valence.
gion, B = .25, 95% CI [.10, .46], SE = .09. This indirect effect Emojis directly affected both variables, and indirectly affected
reduced the direct effect of Emoji on Empathic Concern to non- them, mediated via the respective other. Rather than mediators and
significance, B = .24, 95% CI [.20, .68], SE = .22. Thus, the criteria, these two variables should therefore be considered corre-
effect of emojis on empathic concern was fully mediated by lates of each other. Finally, we predicted that the effects of emojis
 12 ERLE, SCHMID, GOSLAR, AND MARTIN

on empathic concern would be mediated via understanding. Our The EASI Model in Digital Versus Face-to-Face
results, however, more strongly supported the opposite conclusion, Communication
that is, emojis affect understanding via empathic concern (Exp.
4c). The present experiments were designed to test the most basic
As it relates to the bidirectional link between perceived valence premises of the EASI model (van Kleef, 2009). Thus, we made
and understanding, it is important to note that the EASI model some concessions in terms of the complexity of our experimental
designs, which might have affected the observed results.
focusses on the social–cognitive consequences of discrete expres-
First, we decided to only investigate very mundane text mes-
sions of emotions on observers’ affective, inferential, and behav-
sages which were clearly valenced and we implemented only sim-
ioral reactions. The present research, however, more globally
ple emotion-congruent emojis (except in Experiment 1d). In
investigated the effects of positive and negative perceived valence
addition to the measurement issues discussed above, this might
of emojis. Thus, while our findings did not confirm our prediction
help explain why we observed a bidirectional relation between
that the perceived valence of an emoji translates into inferential perceived valence and understanding, rather than an indirect effect
processing, future research is needed to test this prediction for of emojis on understanding, mediated via perceived valence. Fur-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

emojis that convey specific discrete emotions. However, it should thermore, this might also explain why the effect of emojis on
This document is copyrighted by the American Psychological Association or one of its allied publishers.

be noted that as of now there is only a clear understanding of how understanding was fully mediated via empathic concern. What
positively or negatively different emojis are perceived (Rodrigues these results suggest to us is that we created a scenario where the
et al., 2018); but it is currently unknown how well people classify affective pathway within the EASI model clearly has more sway.
emojis as depictions of discrete emotions, and some research sug- In defense of this decision, however, it needs to be stated that the
gests that they are notoriously hard to understand (Miller et al., mode of communication we chose (i.e., instant messaging between
2016). friends) is the most common form of digital communication and
Upon reflection, also the uncertainties about the relation between we believe that this is the first crucial difference between com-
understanding and empathic concern could be rooted in our methodol- puter-mediated and face-to-face communication: digital communi-
ogy. While the EASI model would predict that inferential processing cation is comparatively more uniform, and we believe that our
affects behavioral intentions and reactions, our mediation analyses results are reflective of one of the most prototypic instances of dig-
rather point to the opposite. This discrepant finding might be due to ital communication.
the measure we used to assess inferential processing. First, compared However, this does not imply that the EASI model is not appli-
to emotion contagion, which can be solely based on introspection, to cable to other digital communication settings. On the contrary,
which participants have full access, self-reports about the intentions of what these findings suggest is that clearly more research is needed
others and emotional understanding are comparatively unreliable as to fully test the EASI model within the context of digital commu-
they involve inferences, which can be biased, for example, by egocen- nication. Future research should aim to create situations that allow
tric projections (see Epley et al., 2004). Second, while understanding is for a fair test of the inferential pathway of the EASI model (van
Kleef, 2009). Conveniently, the EASI model explicitly proposes at
the first inference people have to draw about the emotional state of
least two moderators that could be used to design such research:
another person, it usually is qualified by subsequent inferences and
information processing capacities and social-relational factors. It
appraisals about the meaning of this state for the relation between its
should be noted that digital communication compares quite favor-
expresser and the perceiver. Thus, future research should include more
ably to face-to-face communication in terms of the ability to draw
fine-grained and potentially more reliable measures of subsequent
inferences about the message of another person because texting
inferential processing that overcome these limitations, potentially lead- can happen time-lagged (except for some messaging apps like
ing to patterns of results that can also be aligned with the predictions Snapchat, Telegram, Wickr etc.), whereas it would be quite un-
of the EASI model about the relation between inferential processing usual to respond to a face-to-face request only after a longer delay.
and behavioral intentions and responses. However, as the present results suggest that inferential proc-
esses are not needed to decode simple text messages, a more fruit-
ful approach would be to increase the content complexity of the
General Discussion messages, and more importantly for the EASI model, the complex-
ity of how emotions are expressed nonverbally in digital commu-
As described above, our research supports the feasibility of
nication. More complex emojis have become increasingly popular
emojis as nonverbal cues for digital communication, although it
on Internet streaming platforms like Twitch. Some Twitch emotes
cannot conclusively answer which mechanisms underlie this find-
convey more ambiguous and/or complex emotions such as sar-
ing. Furthermore, they also support their functional equivalence to casm/smugness (the “Kappa” emote), awe/surprise (the “Pog-
facial expressions in most aspects, but not fully. While we Champ” emote), or pain/outrage (the “SwiftRage” emote) and are
acknowledge that some of the few discrepancies we observed can interfaced in a more anthropomorphic manner. Most instant mes-
be attributed to methodological aspects of our work, we also saging apps now also offer animated gifs or “animojis” that mimic
believe that it is equally important to also pay heed to differences expressions dynamically. Previous research has found that more
between digital and face-to-face communication. In the next sec- complex emojis are often hard to understand (Miller et al., 2016);
tion we sketch out what these differences and the limitations of for instance, due to culturally different usage (Wolf, 2000) or tech-
our research are, and how this might explain the observed discrep- nical communication aspects (Miller et al., 2016).
ant results, and how the EASI model could be more thoroughly In terms of how well the EASI model applies to digital commu-
tested in future research in digital communication settings. nication settings, it is therefore possible that inferential processes
 EMOJIS IN DIGITAL COMMUNICATION 13

are the primary determinant of subsequent behavioral effects of less certain what a sad message that at the same time includes a
emotional expressions only if the meaning of an emoji is relatively yellow, nonanthropomorphic comedic relief device implies. For
ambiguous or unclear. In addition to the limitations of assessing positive messages, on the other hand, the functional role of smi-
only understanding as an indicator of inferential processing, future leys is aligned with the message content and therefore it is possible
research therefore should also investigate digital communication that the results were stronger in this context.
using more complex, emotion-incongruent, or dynamic emojis to Second, it is well-documented that positive information has a
provide a stronger test of the inferential pathway proposed by the higher “density” than negative information, that is, positive con-
EASI model (van Kleef, 2009). cepts are more similar to each other than negative concepts
Concerning social-relational schemata, we want to emphasize (Unkelbach et al., 2008). Thus, another explanation for this finding
again that the presently tested situation could be considered the might be that positive emojis are a priori applicable to a wider
default of digital communication. That is not to say that computer- array of contexts than negative emoji and this translated into the
mediated communication does not happen in different contexts, too. differences observed presently. To explore this possibility, future
Previous research even suggests that emojis in different contexts research should conceptually replicate the present studies with
evoke inferential processing. For example, they are considered inap- more discrete emotions, paralleling what has been done for face-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

propriate for digital communication at work and even lead to attribu-

This document is copyrighted by the American Psychological Association or one of its allied publishers.

to-face communication before, although as we mentioned above it

tions of incompetence (Glikson et al., 2018). These results are is unclear how well emojis do represent discrete emotions as of
another clear indication that emojis can affect inferential processing. now (see, e.g., Miller et al., 2016).
It would be prudent for psychological and communication research- Nonetheless, the effects of emojis on all outcomes were also
ers to experimentally determine the exact impact of these moderating significant for negative messages except for a few individual com-
variables on the affective and inferential consequences of emoji use, parisons. Even though our power analysis indicated that the actual
especially given what is at stake in formal or professional contexts. observed statistical power of the present studies was considerably
Given how quickly digital communication (norms) change, this is an higher than the conventional level of (1-b) = .80 (the observed me-
important task for now, as well as prospectively. dian power was asymptotically 1), we computed many individual
The second major aspect in which the present research deviated comparisons for negative messages and observing some nonsigni-
from the EASI model was that instead of observing actual behavior, ficant results is expected based on chance.
we recorded a variable that is strongly associated with prosocial inten-
tions (Batson, 2002; Einolf, 2008); but we did not measure any behav- Conclusion
ioral outcomes. Again, in terms of ecological validity we believe that
this decision is justifiable given that compared to face-to-face commu- In conclusion, the present findings systematically investigated
nication, behavioral options are usually severely limited in digital com- how well emojis can fulfill functions that have been demonstrated
munication. In terms of internal validity, on the one hand, it could be for facial expressions of emotion from the perspective of the EASI
argued that empathic concern is just another indicator of the affective model. They complement many more specific, yet valuable inves-
pathway of the EASI model. However, empathic concern clearly dif- tigations, that undertook this important task, too. These previous
fers from measures of emotion contagion in that it is other-focused and our findings emphasize that emojis are powerful devices in
(Davis, 1983) whereas emotion contagion is very much self-focused. digital communication, potentially explaining the high popularity
Thus, also in terms of internal validity we believe that the present oper- that they have garnered in the past. At the same time, they show
ationalization makes sense. that clearly more research is needed to keep up with the rapid
However, future research should also aim to incorporate tests of advances modern technology affords and the speed with which the
actual behavioral outcomes in digital communication. Given that postmillennial society accepts these changes. This report grounds
the only behavioral option in digital communication usually is these challenges in a social–cognitive model of emotions that can
texting back, future research could focus on this behavior. A help to move emotion research—at least in terms of technology—
potential solution would be to ask participants to write responses into the 21st century.
to text messages with and without emojis and to analyze these
responses in terms of their emotional intensity. Such validation
could be done by different participants or by computer programs References
(e.g., Pennebaker et al., 2015) that analyze texts according to pre- Barsade, S. G. (2002). The ripple effect: Emotion contagion and its influ-
defined features, such as emotional intensity. Such criteria could ence on group behavior. Administrative Science Quarterly, 47(4),
be borrowed from the EASI model to guide the evaluation process 644–675. [Link]
for response messages. Finally, such validation could also include Barsade, S. G., & Gibson, D. E. (2007). Why does affect matter in organi-
criteria like reciprocal emoji use. zations? The Academy of Management Perspectives, 21(1), 36–59.
Lastly, the observed asymmetry between positive and negative [Link]
text-messages warrants attention. Throughout most of our studies, Batson, C. D. (2002). Addressing the altruism question experimentally. In
S. G. Post, L. G. Underwood, J. P. Schloss, & W. B. Hurlbut (Eds.),
emotion-congruent emojis had stronger effects when they were
Altruism and altruistic love: Science, philosophy, and religion in dia-
coupled with positive compared to negative messages. Several
logue (pp. 89–105). Oxford University Press. [Link]
explanations for this unexpected (yet rather consistent) pattern of acprof:oso/9780195143584.003.0010
results exist. First, given that emojis are a generalization of smi- Batson, C. D., Early, S., & Salvarani, G. (1997). Perspective taking: Imag-
leys, by virtue of this name it should be evident that they were ining how another feels versus imagining how you would feel. Personal-
originally devised for funny and informal contexts. Thus, in terms ity and Social Psychology Bulletin, 23(7), 751–758. [Link]
of their affective, inferential, and social meaning, people might be .1177/0146167297237008
 14 ERLE, SCHMID, GOSLAR, AND MARTIN

Crivelli, C., & Fridlund, A. J. (2018). Facial displays are tools for social Krishna, A., & Eder, A. B. (2018). No effects of explicit approach-avoid-
influence. Trends in Cognitive Sciences, 22(5), 388–399. [Link] ance training on immediate consumption of soft drinks. Appetite, 130,
10.1016/[Link].2018.02.006 209–218. [Link]
Das, G., Wiener, H. J., & Kareklas, I. (2019). To emoji or not to emoji? Krumhuber, E., Manstead, A. S., Cosker, D., Marshall, D., Rosin, P. L., &
Examining the influence of emoji on consumer reactions to advertising. Kappas, A. (2007). Facial dynamics as indicators of trustworthiness and
Journal of Business Research, 96, 147–156. [Link] cooperative behavior. Emotion, 7(4), 730–735. [Link]
.jbusres.2018.11.007 1528-3542.7.4.730
Davis, M. H. (1983). Measuring individual differences in empathy: Evi- Lo, S.-K. (2008). The nonverbal communication functions of emoticons in
dence for a multidimensional approach. Journal of Personality and Social computer-mediated communication. Cyberpsychology & Behavior,
Psychology, 44(1), 113–126. [Link] 11(5), 595–597. [Link]
Derks, D., Bos, A. E., & Von Grumbkow, J. (2007). Emoticons and social Lohmann, K., Pyka, S. S., & Zanger, C. (2017). The effects of smileys on
interaction on the internet: The importance of social context. Computers receivers’ emotions. Journal of Consumer Marketing, 34(6), 489–495.
in Human Behavior, 23(1), 842–849. [Link] [Link]
.11.013 Manganari, E. E., & Dimara, E. (2017). Enhancing the impact of online hotel
Derks, D., Bos, A. E., & von Grumbkow, J. (2008). Emoticons in com- reviews through the use of emoticons. Behaviour & Information Technology,
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

puter-mediated communication: Social motives and social context. 369(7), 674–686. [Link]
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Cyberpsychology & Behavior, 11(1), 99–101. [Link] Martin, J., Rychlowska, M., Wood, A., & Niedenthal, P. (2017). Smiles as
.2007.9926 multipurpose social signals. Trends in Cognitive Sciences, 21(11),
Einolf, C. J. (2008). Empathic concern and prosocial behaviors: A test of 864–877. [Link]
experimental results using survey data. Social Science Research, 37(4), Miller, H., Thebault-Spieker, J., Chang, S., Johnson, I., Terveen, L., &
1267–1279. [Link] Hecht, B. (2016). “Blissfully happy” or “ready to fight”: Varying Inter-
Epley, N., Keysar, B., Van Boven, L., & Gilovich, T. (2004). Perspective pretations of emoji. Proceedings of the International AAAI Conference
taking as egocentric anchoring and adjustment. Journal of Personality on Web and Social Media, 10(1). The AAAI Press. [Link]
and Social Psychology, 87( 3), 327–339. [Link] [Link]/ICWSM/article/view/14757.
-3514.87.3.32 Pennebaker, J. W., Francis, M. E., & Booth, R. J. (2015). Linguistic inquiry and
Erle, T. M., & Topolinski, S. (2018). Disillusionment. Experimental Psy- word count: LIWC 2015 [Computer software]. [Link]
chology, 65(6), 332–344. [Link] Reis, H. T., & Collins, W. A. (2004). Relationships, human behavior, and
Faul, F., Erdfelder, E., Lang, A.-G., & Buchner, A. (2007). G*Power 3: A psychological science. Current Directions in Psychological Science,
flexible statistical power analysis program for the social, behavioral, and 13(6), 233–237. [Link]
biomedical sciences. Behavior Research Methods, 39(2), 175–191. Rodrigues, D., Prada, M., Gaspar, R., Garrido, M., & Lopes, D. (2018).
[Link] Emojicon: Norms for emoji and emoticons in seven evaluative dimen-
Fridlund, A. J. (2014). Human facial expression: An evolutionary view. sions. Behavior Research Methods, 50(1), 392–405. [Link]
Academic Press. .3758/s13428-017-0878-6
Frijda, N. H. (1986). The emotions. Cambridge University Press. Sinaceur, M., & Tiedens, L. Z. (2006). Get mad and get more than even:
Glikson, E., Cheshin, A., & van Kleef, G. A. (2018). The dark side of a When and why anger expression is effective in negotiations. Journal of
smiley: Effects of smiling emoticons on virtual first impressions. Social Experimental Social Psychology, 42(3), 314–322. [Link]
Psychological & Personality Science, 9(5), 614–625. [Link] .1016/[Link].2005.05.002
.1177/1948550617720269 Smith, A., Rainie, L., McGeeney, K., Keeter, S., & Duggan, M. (2015). U.S.
Hatfield, E., Cacioppo, J. T., & Rapson, R. L. (1993). Emotion contagion. smartphone use in 2015. Pew Research Center. [Link]
Current Directions in Psychological Science, 2(3), 96–100. [Link] .org/2015/04/01/us-smartphone-use-in-2015/
.org/10.1111/1467-8721.ep10770953 Unkelbach, C., Fiedler, K., Bayer, M., Stegmüller, M., & Danner, D.
Hayes, A. F. (2013). Introduction to mediation, moderation, and condi- (2008). Why positive information is processed faster: The density hy-
tional process analysis: A regression-based approach. Guilford Press. pothesis. Journal of Personality and Social Psychology, 95(1), 36–49.
Jeffreys, H. (1961). Theory of probability (3rd ed.). Oxford University [Link]
Press, Clarendon Press. Van Beest, I., Van Kleef, G. A., & Van Dijk, E. (2008). Get angry, get out:
Kaye, L. K., Malone, S. A., & Wall, H. J. (2017). Emojis: Insights, affor- The interpersonal effects of anger communication in multiparty negotia-
dances, and possibilities for psychological science. Trends in Cognitive tion. Journal of Experimental Social Psychology, 44(4), 993–1002.
Sciences, 21(2), 66–68. [Link] [Link]
Kaye, L. K., Wall, H. J., & Malone, S. A. (2016). “Turn that frown upside- van Dijk, E., van Kleef, G. A., Steinel, W., & van Beest, I. (2008). A social
down”: A contextual account of emoticon usage on different virtual plat- functional approach to emotions in bargaining: When communicating
forms. Computers in Human Behavior, 60, 463–467. [Link] anger pays and when it backfires. Journal of Personality and Social Psy-
.1016/[Link].2016.02.088 chology, 94(4) 600–614. [Link]
Keltner, D., & Haidt, J. (1999). Social functions of emotions at four levels Van Kleef, G. A. (2009). How emotions regulate social life: The emotions as
of analysis. Cognition and Emotion, 13(5), 505–521. [Link] social information (EASI) model. Current Directions in Psychological Sci-
.1080/026999399379168 ence, 18(3), 184–188. [Link]
Keltner, D., Haidt, J., & Shiota, M. N. (2006). Social functionalism and the Van Kleef, G. A. (2014). Understanding the positive and negative effects
evolution of emotions. In M. Schaller, J. A. Simpson, & D. T. Kenrick of emotional expressions in organizations: EASI does it. Human Rela-
(Eds.), Evolution and social psychology (pp. 115–142). Psychosocial Press. tions, 67 (9), 1145–1164. [Link]
Koch, W., & Frees, B. (2016). Dynamische Entwicklung bei mobiler Internet- Van Kleef, G. A. (2017). The social effects of emotions are functionally
nutzung sowie Audios und Videos [Dynamic development in mobile Inter- equivalent across expressive modalities. Psychological Inquiry, 28(2–3),
net usage as well as audios and videos]. Media Perspektiven, 9(1), 418–437. 211–216. [Link]
Koning, L. F., & Van Kleef, G. A. (2015). How leaders’ emotional displays Van Kleef, G. A., & Côté, S. (2007). Expressing anger in conflict: When it
shape followers’ organizational citizenship behavior. The Leadership Quar- helps and when it hurts. Journal of Applied Psychology, 92(6),
terly, 26(4), 489–501. [Link] 1557–1569. [Link]
 EMOJIS IN DIGITAL COMMUNICATION 15

van Kleef, G. A., De Dreu, C. K., & Manstead, A. S. (2004). The interpersonal Science Computer Review, 19(3), 324–346. [Link]
effects of anger and happiness in negotiations. Journal of Personality and 3930101900307
Social Psychology, 86(1), 57–76. [Link] Wolf, A. (2000). Emotional expression online: Gender differences in emo-
Van Kleef, G. A., De Dreu, C. K., & Manstead, A. S. (2010). An interper- ticon use. Cyberpsychology & Behavior, 3(5), 827–833. [Link]
sonal approach to emotion in social decision making: The emotions as 10.1089/10949310050191809
social information model. In M. P. Zanna (Ed.), Advances in experimen- Yip, J. A., & Schweinsberg, M. (2017). Infuriating impasses: Angry expres-
tal social psychology (Vol. 42, pp. 45–96). Academic Press. sions increase exiting behavior in negotiations. Social Psychological & Per-
Van Kleef, G. A., van den Berg, H., & Heerdink, M. W. (2015). The persua- sonality Science, 8(6), 706–714. [Link]
sive power of emotions: Effects of emotional expressions on attitude for-
mation and change. Journal of Applied Psychology, 100(4), 1124–1142.
[Link] Received May 7, 2019
Walther, J. B., & D’Addario, K. P. (2001). The impacts of emoticons on Revision received March 5, 2021
message interpretation in computer-mediated communication. Social Accepted March 9, 2021 n
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.