Académique Documents
Professionnel Documents
Culture Documents
ORIGINAL ARTICLE
Gesture, voice, expression, and context add richness to communications that increase
chances of accurate interpretation. E-mails lack much of this richness, leading readers
to impose their own richness. Three experiments tested the effect of communicator rela-
tionship and availability of context on e-mail writer and reader confidence levels and
accuracy. Effects of nonverbal and verbal behavior were also investigated. Results showed
that confidence levels for both writers and readers were high, yet somewhat unwarranted
based on accuracy rates. Further, writers had more confidence that friends would correctly
interpret e-mails than strangers, although friends showed no more accuracy. Findings
suggest that reliance upon friendship and context, as well as verbal and nonverbal cues, to
interpret emotion in e-mail is ineffective, sometimes detrimental.
doi:10.1111/hcre.12093
rather than affect cues (Vandergriff, 2014), such as by following a criticism with a
smiley face to appear less threatening.
The knowledge that an emoticon carries affective information for a reader has lim-
ited usefulness in understanding CMC unless we know whether the emoticon carries
the correct affective information. In other words, we must examine both the intended
emotion of the writer and the perceived emotion of the reader to determine if a com-
munication is successful.
One study in particular attempted to answer this question: Hancock, Landrigan,
and Silver (2007) had two strangers engage in an instant messaging conversation. One
participant was told to express happiness or sadness toward the naïve other without
explicitly mentioning the emotion. Overall, the intended affect communication was
successful: The naïve partners rated interlocutors as happier if they were in the happy
condition than in the sad condition. An examination of the resulting instant messag-
ing conversation showed that participants assigned to happiness wrote more words,
wrote fewer affect words (negative affect words in particular), disagreed less, and used
more exclamation points than the sadness group (interestingly, emoticon use did not
differ. Interlocutors may have used both smiley and frowny faces).
While Hancock et al.’s (2007) study shows that affect communication can be
successful in CMC, it rather oversimplifies the process. In studying the communica-
tive success of CMC between two people, context becomes important. For example,
Walther and Tidwell (1995) composed a set of e-mails that were either social or
task-related in nature. Participants read and rated the e-mails for dominance and
affection. The authors found that the e-mail context affected these ratings. Although
this study again uses researcher-generated messages (and thus is subject to the same
issues as above studies), it did indicate that the context of the e-mail matters in
interpretation of a message.
A study by Riordan, Kreuz, and Olney (2014) also showed the effect of context.
In a within-subjects experiment, friends engaged in instant messaging conversations
that were either freely written or based in a debate. The dynamics of the resulting
conversations varied distinctly between the two conversation types—in the length
of utterances, the lag time of response, the use of affect words, and semantic align-
ment. In other words, the context of the conversation altered both paralinguistic and
linguistic patterns, even within the same pair of interlocutors.
The effect of context may be rooted in grounding theory. Interlocutors design their
communications based on what they believe a listener knows or does not know, in
order to promote communication success; that is, they “ground” their communication
in a particular context based on mutual knowledge (Clark & Brennan, 1990). As such,
contextual knowledge results in more efficient and better understood communication.
Another factor that has an effect on communication is the relationship between the
writer and reader. Channel expansion theory (Carlson & Zmud, 1994) posited that the
effectiveness of a communication depends upon the level of experience with the per-
son with whom one is communicating—the more experience, the more effective. The
hyperpersonal model (Walther, 1996) further suggested that when interlocutors are
strangers, readers pull from a personal repertoire regarding how people are expected
to feel or act in situations in order to interpret a message. In other words, they inter-
pret messages from an egocentric or stereotypical viewpoint. Friends, by definition,
share a history of personal communication and experience upon which they can rely
to help them interpret each other’s messages, whereas strangers must necessarily draw
upon only general experiences and knowledge of the world. Such relational differ-
ences lead to alterations in interpretation and conversational patterns. For example,
Riordan et al. (2014) found very different conversational dynamics depending upon
the relationship between interlocutors: Instant message exchanges with a friend were
more highly aligned paralinguistically than message exchanges with a stranger—and
friends showed significant effects of context when strangers did not.
Because much research about affect understanding in messages has been
one-sided (i.e., based on what a reader interprets, or what a writer writes, but rarely
both), the impact of context and relational dynamics has not been fully assessed.
Although these studies answered valuable questions about encoding and decoding
messages in CMC and the factors affecting same, conclusions as to whether affect
is communicated accurately in CMC are impossible. These studies failed to accom-
modate both the intentions of the writer and the interpretation of the reader, while
considering contextual and relational dynamics that played a role in communicating
information.
Kruger, Epley, Parker, and Ng (2005), in a series of five experiments, took a step
toward elucidating the problem of determining how successful the communication
of affect between interlocutors using CMC actually is. The authors found that e-mail
writers were more confident that readers would correctly interpret sarcasm, serious-
ness, anger, humor, and sadness in e-mail than was warranted by the actual accuracy
rates of the readers. Though confidence levels were high, accuracy was less than desir-
able. The authors argued that the writer imposed richness upon the message, “hearing”
himself in his e-mail, which readers were not able to imitate.
Instead, readers were forced to impose their own richness upon the e-mail in order
to devise an interpretation of affect. This imposed richness came from an egocentric
viewpoint, which led to an interpretation that was often less accurate than the writer
assumed it would be. Interestingly, in an assessment of relational dynamics, one of the
experiments found that friends were no more accurate than strangers, suggesting that
familiarity with the writer did not overcome this egocentrism during interpretation.
Kruger et al.’s (2005) theory that we impose richness upon a message by “hearing”
ourselves when we write and “hearing” others when we read a message underlines
the importance of nonverbal communication in determining affect. Indeed, Burgoon,
Buller, and Woodall (1996) reported that nonverbal behaviors account for more of a
person’s perception of another’s affect than verbal content does. For example, imagine
noticing a friend with pursed lips and crossed arms. Interpreting her body language,
you ask her what’s wrong. She replies, “I’m fine,” in a biting voice. Burgoon et al. (1996)
suggest it is more likely that you will interpret that she is upset about something based
on her body language and tone than you are to trust her word.
Although historically kinesics and vocal tone were presumed to be absent in CMC,
compelling readers to provide it from an egocentric viewpoint, research suggests that
these nonverbal cues were never actually lost, but instead transformed. Social infor-
mation processing theory (Walther, 1992) suggested that one way users work with and
overcome the limitations of CMC relative to face-to-face encounters is by translating
nonverbal cues into verbal cues. For example, interlocutors asked more questions,
used statements of disclosure (Tidwell & Walther, 2002; Walther, Loh, & Granka,
2005), varied the formality of the language (Byron & Baldridge, 2005), and used ver-
bal strategies such as disagreement (Hancock et al., 2007). Interlocutors also adapted
symbols to serve the same purpose as nonverbal cues. For example, they wrote longer
messages (Byron & Baldridge, 2005; Hancock et al., 2007), used emoticons (Derks,
Bos, & von Grumbkow, 2008; Lo, 2008; Utz, 2000; Walther & D’Addario, 2001) or
symbols such as asterisks and capitalization (Riordan & Kreuz, 2010), varied their
punctuation (Hancock et al., 2007), and even included or excluded standard features
such as capitalization of proper nouns or intentional misspellings (Harris & Paradice,
2007; Riordan & Kreuz, 2010). Such cues may have been imposed before a message
was even crafted, such as the finding that lag time before a response indicates a respon-
dent’s affect (Riordan et al., 2014; Riordan, Markman, & Stewart, 2013; Walther &
Tidwell, 1995). More recently, the use of memes, gifs, and other stimuli that add visual
cues such as kinesics have become a new format for communicating information in a
nonverbal way.
The presence of explicit verbal and nonverbal strategies using affective language
or visual symbols could lead to more confidence and success in communication. In
the current study, participants generate e-mails and rate the emotional content of the
e-mails as well as their confidence that the e-mail would be correctly interpreted. The
e-mails are then read by other participants who make the same ratings. The effects of
situational context and relationship are investigated. We also examine several different
verbal and nonverbal strategies in e-mails that are projected to lead to greater success
in communicating affect.
In the following studies, four hypotheses and three research questions were tested.
First, based on Kruger et al.’s (2005) theory of egocentrism in combination with both
channel expansion theory and the hyperpersonal model to analyze relationship
effects, we hypothesized that:
H1: Writers will be more confident in the interpretation abilities of friends than strangers.
H5: Readers will be more confident in their interpretations when context is present than
absent.
H6: Readers will be more accurate in their interpretations when context is present than
absent.
Experiment 1
Participants
To collect the e-mails, a short task was given to 50 participants (25 males;
M age = 37.22, SD = 9.92) recruited from Mechanical Turk (MT) crowdsourcing
system. Participants were required to have completed a minimum of 5000 tasks
with 98% success and to be located in North America (to avoid cultural effects). All
participants were paid US$1.
The second task was given to 50 participants (27 males; M age = 32.58, SD = 11.23)
recruited using the same requirements. Participants were paid 15 U.S. cents each.
Procedure
In the first task, participants were directed to an online survey page. Participants were
asked to indicate their age and sex and then to write an e-mail. Instructions for this
context-absent e-mail were: “In the box below, write an e-mail to a friend. This e-mail
is to convey an emotion; for example, that you are disappointed after trying a new
restaurant, happy that someone asked you out on a date, or angry with a family mem-
ber.” Participants were given a text box with a standard text editor menu to write
the e-mail. The text editor included an emoticon menu, image uploader, and font
alteration tools. After writing the e-mail, participants then filled out a questionnaire
asking them to rate, on a scale of 1 (not at all) to 7 (very much), how much of each
of eight emotions were present in the e-mail. These emotions were those identified
by Plutchik (2003) as the most basic human emotions, and included joy, trust, fear,
surprise, sadness, disgust, anger, and anticipation.1 Then participants were asked to
rate how confident they were about a friend’s and a stranger’s ability to correctly inter-
pret the affect in the e-mail. Immediately following this series of scales was another
question using the same scale that indicated participants should respond with “2”
to continue the survey. Those who responded with the number 2 were redirected by
the survey flow to continue the survey. Those who responded with any other num-
ber were redirected to an end statement saying that the participant was ineligible to
continue the survey. The intent of this question was to automatically remove possible
acquiescent responding from the data set.
The participant then wrote a second e-mail. Instructions for this context-given
e-mail were: “Pretend that one of the following situations happened to you. In the
box below, indicate which situation you chose. Then write an e-mail to a friend telling
them how you feel about the situation.” Five scenarios depicting common situations
followed (see Appendix). The same ratings were then completed as for the first e-mail.
In the second task, which was completed by different participants than task 1,
participants were directed to an online survey page. The page asked them to indicate
their age and sex and then read a context-absent e-mail that had been generated in the
first task. They rated the e-mail on the eight emotions and indicated their confidence
in their interpretation. The process was repeated for the context-given e-mail from
the same writer, first presenting readers with the situation chosen by the writer (and
thus “given” the context). We counterbalanced which e-mail appeared first. The same
question for detecting acquiescent responding in the first task was also present in this
second task.
H4. Unexpectedly, the writer of the e-mail was no more or less confident in
the accuracy of the reader no matter if the e-mail was context-absent (M = 5.91,
SD = 1.27) or context-given (M = 6.01, SD = 1.10; one-tailed t(198) = .59, ns). This
effect was true of both friends alone (context-absent M = 6.36, SD = .75; context-given
M = 6.32, SD = .84; one-tailed t(98) = .25, ns) and strangers alone (context-absent
M = 5.46, SD = 1.51; context-given M = 5.70, SD = 1.25; one-tailed t(98) = .86, ns).
The writer’s confidence in the accuracy of friends was positively related to their
confidence in strangers (r = .49, p < .001), no matter if the e-mail was context-absent
(r = .48, p < .001) or context-given (r = .52, p < .001). In other words, if the writer
thought that friends would have an easy time interpreting the e-mail accurately, they
thought strangers would, too, although they did tend to have slightly less confidence
in the strangers. Overall, high confidence was predominant: In only 11% of the confi-
dence ratings did writers believe readers would perform at less than a 5 on the 1 (not
at all confident) to 7 (very confident) scale.
H5. Readers in the second task were confident that they were highly accurate,
too: Only 8% of ratings were less than 5. This confidence did not waver between
context-absent (M = 5.90, SD = .93) and context-given e-mails (M = 5.90, SD = .99;
one-tailed t(98) = .00, ns).
To determine this, negative numbers were made positive and averages were
recomputed. Readers tended to be fairly accurate in their assessments of emotion;
on average readers were only 1.41 points (range = 0 to 6) off in either direction,
although this varied quite a bit depending upon the emotion being considered
(Anger M = 1.06; Disgust M = 1.08; Sadness M = 1.09; Fear M = 1.21; Joy M = 1.37;
Surprise M = 1.67; Anticipation M =1.86; Trust M =1.96). The variability suggests
that some emotions may be easier to determine in an e-mail than others; readers
were somewhat better at assessing the four negative emotions than the four positive
ones, though this effect did not approach significance.
Neither the reader’s confidence level nor the reader’s overall accuracy significantly
differed depending on the number of cues used in the e-mail (intercept = .64, confi-
dence = −.17, accuracy = −.11). This same result was true of context-absent e-mails
(intercept = 2.47, confidence = −.44, accuracy = −.81) and context-given e-mails
(intercept = −.56, confidence = .04, accuracy = .09) independently.
Apparently, nonverbal cues are not particularly helpful for interpreting affect, and
their presence does not inspire confidence in one’s interpretations.
RQ3. To determine the number of verbal cues in an e-mail, Linguistic Inquiry and
Word Count (LIWC; Pennebaker, Booth, & Francis, 2007) was employed. LIWC com-
pares a text to an internal dictionary of 4,500 words and word stems that encompass
over 80 categories and returns the percentage of words in the e-mail that fit into a
given category. Two of these 80 categories were used in the current study: positive
and negative emotion words. Because LIWC can only recognize words with proper
spelling, accuracy could only be assured if any misspellings in the e-mails were cor-
rected before LIWC was used. These spell-checked versions of the e-mails were used
for LIWC purposes only, and the e-mails existed in their unaltered state for all other
analyses.
Across all data, no relationship was found between reader confidence level and
the total number of words in the e-mail (r = .02, ns), or the percentage of negative
words (r = −.02, ns) specifically. However, a relationship was found between reader
confidence level and the percentage of positive words (r = .29, p < .01) in an e-mail.
This pattern of significant findings applied specifically to context-absent e-mails (total
words: r = .04; negative words: r = −.001; positive words: r = .39, p < .005) and not
context-given e-mails (total words: r = .01; negative words: r = −.04; positive words:
r = .20).
Overall accuracy across all e-mails was not related to the total word count
(r = .05, ns), percentage of negative words (r = .15, ns), or percentage of positive
words (r = −.16, ns). This pattern of nonsignificant findings was true for both
context-absent (total: r = .06; negative: r = .23; positive: r = −.15) and context-given
e-mails (total: r = .02; negative: r = .05; positive: r = −.19).
To summarize, readers were more confident in their interpretations if a greater
percentage of positive words existed in the e-mail, but this effect was only true for
context-absent e-mails. Verbal cues had no effect on accuracy.
Discussion
As hypothesized, writers had greater confidence in the interpretation abilities of
friends than strangers, suggesting that writers lean on common ground when they
write an e-mail. Such findings are in line with Kruger et al.’s (2005) egocentrism
theory, channel expansion theory, and the hyperpersonal model. Unlike what would
be expected based on grounding theory, confidence level did not vary depending
upon whether context was given or absent, suggesting that situational knowledge is
not part of the common ground that is factored into writer confidence levels. This
lack of effect of contextual knowledge appears to be warranted: A stranger was no
more accurate in his or her interpretations when context was given than when it was
absent. Interestingly, readers themselves recognized this lack of effect: They were no
more or less confident in the accuracy of their interpretation whether context was
given or not. In fact, readers were poor judges of their accuracy.
Neither the assessed nonverbal nor verbal cues appeared to be used as markers for
either writer or reader confidence levels—and rightfully so, as they did not contribute
to accuracy. This is a surprising result, as social information processing theory would
have suggested that if people use cues, they are used for the purpose of improving the
communication success.
As these results were surprising, in Experiment 2 the second task was repeated
with another set of participants. This served as a direct replication of the first exper-
iment as well as generated data to determine, in an analysis of Experiments 1 and 2
together, whether consistency in interpretation was present, if not accuracy.
Experiment 2
Participants
Fifty participants (24 males; M age = 32.54, SD = 8.73) were recruited from MT using
the same requirements as in Experiment 1. All participants were paid fifteen U.S. cents
as in Experiment 1.
Procedure
The procedure was the same as in the second task of Experiment 1.
Possibility of consistency
Having repeated the experimental procedure, the data now included two different
sets of interpretation scores for each e-mail, allowing an analysis of whether there is
consistency in interpretations. In other words, if a reader misinterpreted the writer,
did another reader misinterpret the same way? It is possible that even if there was
a lack of relationship between the writer’s intended interpretation and the reader’s
actual interpretation, there was still a relationship between the interpretations made
by the two readers, which would indicate consistency in interpreting affect. To
determine this, the emotion ratings given by the first group of readers (Experiment 1)
were correlated with the same emotion ratings of the same e-mail from the second
group of readers (Experiment 2). Results showed a significant relationship (r = .47,
p < .001) in emotion ratings between the two readers. Interestingly, however, the cor-
relation between the two readers was slightly weaker than the correlations between
each reader and the writer (Experiment 1: r = .54, p < .001; Experiment 2: r = .51,
p < .001).
An LME model was created in which the ratings of the two readers were compared
while controlling for multiple ratings corresponding to the same writer and the same
e-mail. The model shows no significant difference in the emotion ratings between the
two readers (intercept = 3.22, reader = −.03, ns). This result suggests that no matter
who was reading the e-mail, the same interpretation was likely, regardless of whether
the relationship in affect scores between the writer and reader was strong or weak.
Discussion
The results in Experiment 1 were largely directly replicated in Experiment 2. Situa-
tional knowledge, nonverbal cues, and verbal cues are not factored into confidence
levels nor do they affect accuracy.
Experiment 1 found that writers have higher confidence levels in friends than
strangers. Presumably, this confidence is rooted in a history of common ground,
which should lead to greater accuracy than strangers in interpreting e-mails. How-
ever, Kruger et al. (2005) found this not to be the case. If accuracy is not better
for friends than for strangers, why would writers have more confidence in friends?
Confidence in others is generally built with experience—that is, high accuracy in
past e-mail exchanges begets high confidence in accuracy in future e-mail exchanges.
Kruger et al. (2005) measured accuracy between friends and strangers based
on whether the friend or stranger could explicitly report the correct emotion in
the e-mail. In the two current experiments, using a different data format, the same
Experiment 3
Participants
Participants were recruited from psychology courses at a large University in the Mid-
western United States and were given class credit for participation. Each participant
brought a friend to the study, who was given a choice between class credit or $10 for
participating. Thirty-five dyads participated (N = 70, 22 men; M age = 19.70 years,
SD = 3.58; M length of friendship = 42.88 months, SD = 54.34, range = 2.75–252
months).
Procedure
The two friends reported to a laboratory. These participants were seated in separate
computer rooms and completed a demographics questionnaire asking for age, sex,
and the length of their friendship with the other participant. We then replicated the
first task of Experiment 1 using Google e-mail accounts provided for the purpose,
which included all standard text-editing tools including an emoticon menu and image
uploading capability. We counterbalanced whether writers wrote the context-absent
or context-given e-mail first.
Writers sent the e-mails to their friends at the e-mail address created for the pur-
pose. The friends then replicated the second task of Experiment 1 (and 2). Lastly, in a
deviation from Experiments 1 and 2, each participant wrote a response e-mail to the
writer.
As a third task, each participant was presented with an e-mail written by the pre-
vious dyad (thus reading as a stranger). Participants completed the same affective and
confidence ratings as they did for their friends’ e-mails, also crafting a reply e-mail.
We counterbalanced the presentation of the context-absent and context-given e-mail.
In this manner, each e-mail had three sets of eight affective ratings and ratings of
confidence: those of the writer, those of the writer’s friend, and those of a stranger
from the dyad that followed. Each e-mail had two reply e-mails: one from a friend,
and one from a stranger.
SD = 1.14; one-tailed t(134) = −.36, ns). Also as in the prior experiments, confidence
levels were high for both writers and readers, with the overwhelming majority being
five or above.
On accuracy. We calculated difference scores in the same manner as in prior
experiments. Each e-mail thus yielded two sets of eight difference scores: The dif-
ference scores between the writer and her friend, and the difference scores between
the writer and the stranger. We then entered these scores into an LME model, which
allowed control for multiple difference scores coming from the same e-mail nested
within the same writer-reader pair.
H3 and H6. The model intercept (i.e., difference score) of −.23 did not significantly
differ if the writer and reader were friends rather than strangers (.11, ns), if the context
was given rather than absent (.02, p = .96, as in Experiments 1 and 2), or if the two
variables interacted (−.02, ns).
Although friends were more confident than strangers in their interpretations
when reading an e-mail, this confidence turned out to be unwarranted, since they
were both equally (in)accurate overall in their interpretations. However, when specif-
ically looking only at friends, a longer friendship did lead to better overall accuracy
(r = .23, p < .01).
Another model assessed whether accuracy was different among the positive and
negative emotions; as in prior experiments, no significant difference was found (inter-
cept = −.12, emotion type = .01, ns).
Also as in prior experiments, readers tended to overestimate the presence of
emotion, except in this case for joy and anticipation, for which they tended toward
underestimation: Joy M = .15; Anticipation M = .14; Fear M = −.01; Anger M = −.09;
Sadness M = −.11; Disgust M = −.19; Trust M = −.21; Surprise M = −.35. Also as
in prior experiments, readers tended to be fairly accurate in their assessments of
emotion, averaging only 1.48 points (range = 0 to 6) off in either direction, and
again this varied quite a bit depending upon the emotion being considered (Joy
M = 0.75; Anger M = 1.05; Sadness M = 1.12; Fear M = 1.23; Disgust M = 1.35;
Surprise M = 1.71; Anticipation M =2.15; Trust M =2.48). As in prior experiments,
anticipation and trust were the most difficult emotions to determine accurately, and
the negative emotions were somewhat easier to determine than the positive ones
(with the exception of joy here, same as Experiment 2), though this effect did not
approach significance.
actual overall accuracy—for both friends (r = −.03, ns) and strangers (r = −.07, ns,
as in prior experiments).
other mannerisms, a process called alignment (Ferreira & Bock, 2006, provide
an excellent review). Clark and Brennan (1990) suggested that during grounding,
interlocutors use this alignment as evidence of mutual knowledge. For example: Jack
tells Jill about his day, and during the story Jill smiles when Jack smiles and frowns
when Jack frowns. This alignment tells Jack that Jill understands his story. However, if
Jack mentions something about which Jill lacks knowledge, such as “Hartwood Acres
Park,” Jill may deviate in her alignment, frowning at Jack while he is smiling. Jack will
pick up on this deviation and further explain what and where Hartwood Acres Park
is, in order to again establish mutual knowledge, before proceeding with his story.
The more two people come to understand each other, the more aligned they
become in these aspects. Thus, those who understand each other better (usually
based in a history of communication) have greater levels of alignment than those
who do not understand each other as well. Given that this increased understanding
often comes from a history of communication, friends should have greater levels of
alignment than strangers. We further posit that times in which interlocutors have
more information from which to develop such understanding (i.e., context-given)
should result in greater alignment levels than times in which interlocutors have less
information (i.e., context-absent).
To measure such implicit understanding in our current experiment, each reader
was instructed to write a reply to the e-mail he read. The original e-mail and the reply
e-mail were compared in the length of the e-mail (number of words), the percentage
of negative emotion words, and the percentage of positive emotion words. Difference
scores between the original and reply e-mail on each of the three variables were com-
puted to determine how aligned the two e-mails were. For example, if a writer created
an e-mail that was 121 words long, and the reader’s reply was 95 words, the difference
score was 26. All scores were positive numbers; scores of zero indicated perfect align-
ment. Attempts were made to create this difference score for nonverbal cues as well,
but the low rate of use of these cues artificially inflated alignment scores (i.e., very few
alignment scores were above zero).
LME models were created to control for the fact that each participant wrote two
e-mails, each of which was replied to by two people. As such, four difference scores
correspond to the same e-mail writer and within those four scores, two are from the
same e-mail replier, although each difference score corresponds to only one set of
original and reply e-mails. All difference scores were log-transformed to fit model
assumptions.
The first model indicated greater alignment in the total number of words for
context-absent e-mails than context-given e-mails, but no other significant effects
(intercept = 2.51, context = .50, p < .05, relationship = .32, interaction = −.21).
The second model indicated that greater alignment in positive words occurred
for context-absent e-mails than context-given e-mails, but again, no other significant
effects (intercept = .43, context = .52, p < .05, relationship = .31, interaction = −.20).
The third model indicated no significant effects of alignment in negative words
(intercept = .33, context = .19, relationship = .31, interaction = −.20).
Results suggest that effects of context were present when measured implicitly, a
finding that was lacking when measured explicitly. Readers showed greater implicit
understanding for context-absent than context-given e-mails.
Relational effects did not appear to play a role in implicit understanding. This find-
ing complements the lack of effect in accuracy scores found with measures of explicit
understanding.
Discussion
Three studies were designed to test the manner in which emotion is interpreted in
an e-mail. The studies tested whether writers’ and readers’ confidence levels were
greater when the interlocutors were friends or strangers, whether these levels varied
by contextual knowledge, and whether these levels were justified by accuracy rates.
The studies also tested whether accuracy rate on the readers’ part was related to the
relationship between the interlocutors or the presence of context. Lastly, the studies
sought to determine whether the use of a number of verbal and nonverbal cues led to
increased confidence or accuracy rates.
The first hypothesis, which predicted that writers would have more confidence in
friends than strangers, was supported. Kruger et al. (2005) did not find a difference
in this aspect, although both channel expansion theory (Carlson & Zmud, 1994) and
the hyperpersonal model (Walther, 1996) would have predicted such. The difference
in confidence, however, was quite small; writers had quite high confidence in read-
ers across the board and, furthermore, this did not differ between context-given and
context-absent e-mails as grounding theory suggests would occur.
Tests of the second hypothesis found that friends were significantly more confi-
dent about their interpretations than strangers. This effect is in line with both channel
expansion theory and the hyperpersonal model.
Tests of the third hypothesis probing the effect of relationship on accuracy found
little difference in the accuracy rates of friends and strangers when categorical
variables were used (replicating Kruger et al. (2005)). However, when friendship was
quantified by months, longer friendships led to greater communicative success.
The fourth, fifth, and sixth hypotheses tested the effect of contextual information
on the confidence of both the writer and reader and the accuracy of the reader.
Unlike what would be expected based upon grounding theory, no difference was
found in confidence or accuracy between context-given and context-absent e-mails,
for either the writer or reader. In Experiment 3, we actually found that writers had
less confidence in the reader’s ability to correctly interpret context-given e-mails
than context-absent e-mails. The five situations given may have been artificial to
the participants, such that the writer was not sure what he or she would feel about
the hypothetical situation and was not confident in his or her ability to convey
affect about it. Future studies should attempt different manipulations of contextual
information, both for ecological validity and to determine what elements of context
are truly key to interpretation.
assumptions affected their interpretations remain unknown. Further studies are nec-
essary to examine the effect these factors might have on interpretations, and whether
readers make assumptions about these factors when evidence is lacking.
Further studies might also consider that the effect of friendship may not be a
robust one, thus requiring a larger sample size to assess. Although Experiment 2 repli-
cates all findings of Experiment 1, it does so using participants from the same pop-
ulation given the same procedure. Experiment 3 is based in a different population
(college students), and alters the procedure. The strangers in Experiment 3 largely
followed the results pattern seen in Experiments 1 and 2, which would be expected
given that all three samples consist of strangers.
However, the friends in Experiment 3 varied from this pattern. For example,
Experiments 1 and 2 showed no relationship between confidence and accuracy.
However, in Experiment 3, specifically for friends, higher writer confidence led to
lower accuracy. This deviation suggests that friendship might lead writers to take
certain things for granted that are not taken for granted with strangers. This possi-
bility is supported by the fact that writers were less confident in their friends when
context was given than when it was absent, suggesting that the context supplied in
the experiment may not have allowed them to express themselves in a way that they
could assume the friend would understand.
The small sample size of friends (35 dyads, one e-mail per dyad from both con-
text conditions) in the experiment restricts the ability to determine specific differ-
ences between the context-given and context-present e-mails that could suggest why
context-given e-mails would result in lower confidence. Indeed, other results also
suggested that a larger sample size would be necessary to see effects. For example,
although friends were no more accurate than strangers in Experiment 3, for friends
specifically, a longer friendship did lead to a little more accuracy. Given a larger sam-
ple size, which might supply a wider range of friendship lengths, hypothesis three
might have been supported. However, if effects can only be found with a large sample
and using specific measures or analyses, they are likely of little practical importance
to interlocutors.
Conclusion
George Bernard Shaw once proposed that “the single biggest problem in communi-
cation is the illusion that it has taken place.” Throughout this work, Shaw’s suggestion
appears to be the dominant theme: Evidence supports overconfidence at the keyboard,
and it is clear that reliance upon friendship and situational knowledge when interpret-
ing emotion is ineffective at best, detrimental at worst.
Appendix
Scenarios presented for context-given e-mails
Situation 1. You went to a basketball game last night. Your team was leading by 28 at
the end of the first half and it looked like it was going to be a blow-out as the second
half of the game began. The opponent began to catch up, however, and 18 minutes
into the half, the score was tied. The crowd was roaring, and when the clock ticked
down to the last minute your team is behind by 4. With 15 seconds left on the clock,
your team hit a 3-pointer. The opponent got the ball and tossed it inbounds intending
to run out the clock. With 4 seconds left, the star forward stole the ball and went in
for a layup. As the buzzer sounded, signaling the end of the game, your team sank the
shot and won. It is now the next day.
Situation 2. You are taking a class to learn a new language. A big test is coming
up. You studied for it, doing all the study guide questions and reading the books. At
nearly midnight the night before the test, you realized that you never finished reading
the last chapter of your language book. You lied down to read it, but you fell asleep a
third of the way into it. The next morning, you took the test and you did not know the
answer to half of the questions. It is now two hours after the test.
Situation 3. You were hanging out at the tables outside your workplace. A dog
came up to your table, sniffing at your food. Not seeing the dog’s owner, you walked
around asking if anyone knows where the dog came from. Nobody had any answers.
You called the number on the dog’s tag, but nobody answered. Eventually, you gave
up and left the dog, returning to work.
Situation 4. You went to the gym and ran on the treadmill. You had a great workout.
Afterward, you hit the button to stop the machine but did not realize you did not
actually stop the machine. When you attempted to step off, you lost your footing, fell
on the treadmill belt, and slid off the machine onto the floor. Half the gym saw this
happen, and several tried not to laugh while asking if you’re okay. It is now four hours
later, and you are icing your knee.
Situation 5. Last week you saw a booth in the mall that gave a free piece of pizza
and a raffle ticket for a restaurant gift card if you completed a survey. You completed
the survey, put in your raffle ticket, and got your pizza. It is now the week after. You
get a phone call telling you that you won the raffle.
Note
1 There is considerable debate in the literature about what constitutes the basic human
emotions. While we respect this debate, it is peripheral to our current objective to create a
brief measurement tool.
References
Burgoon, J. K., Buller, D. B., & Woodall, W. G. (1996). Nonverbal communication: The
unspoken dialogue (Vol. ,2nd ed.). New York: McGraw-Hill.
Byron, K. (2008). Carrying too heavy a load? The communication and miscommunication of
emotion by email. Academy of Management Review, 33, 309–327. doi:10.5465/AMR.
2008.31193163
Byron, K., & Baldridge, D. (2005). Toward a model of nonverbal cues and emotion in email.
Academy of Management Proceedings, B1–B6. doi:10.5465/AMBPP.2005.18781269