Download as pdf or txt
Download as pdf or txt
You are on page 1of 13

Accounting, Organizations and Society xxx (xxxx) xxx

Contents lists available at ScienceDirect

Accounting, Organizations and Society


journal homepage: www.elsevier.com/locate/aos

Feedback with feeling? How emotional language in feedback affects


individual performance*
Devon Erickson a, D. Kip Holderness Jr. b, Kari Joseph Olsen c, Todd A. Thornock d, *
a
Utah State University, USA
b
West Virginia University, USA
c
Utah Valley University, USA
d
University of Nebraska-Lincoln, USA

a r t i c l e i n f o a b s t r a c t

Article history: Our study examines the influence of manager emotion on the effectiveness of feedback in motivating
Received 6 March 2020 performance. Feedback Intervention Theory (Kluger & DeNisi,1996) suggests that feedback cues can shift
Received in revised form feedback recipients’ attention away from the implications of the feedback for the task and toward the
14 June 2021
implications for the self. We propose that negative emotional language from a manager acts as a feedback
Accepted 23 November 2021
Available online xxx
cue directing employees’ attention more toward the self, and therefore, reduces employee effort in
response to negative feedback. Consistent with our prediction, we find that negative emotional language
decreases employee performance in response to negative feedback. In contrast, positive emotional
Keywords:
Performance feedback
language has no effect on employee performance in response to positive feedback. In two additional
Emotional language experiments, we provide triangulating evidence e via moderation-of-process and direct measurement e
Feedback valence supporting our proposed mechanism, i.e., that the effect of negative emotional language on performance
Feedback intervention theory occurs by shifting individuals’ attention away from the task and toward the self. In doing so, we also
Self-attention show that managers can attenuate the effect of negative emotional language on subsequent performance
by emphasizing that feedback relates to the task and not the self. Our study identifies a novel feedback
intervention cue and provides corroborating evidence of Feedback Intervention Theory. Overall, our
findings suggest that managers should consider how expressions of emotion accompanying feedback
affect employee attention and subsequent performance.
© 2021 Elsevier Ltd. All rights reserved.

1. Introduction communicate more than explicit employee performance. For


example, when informing employees that their performance
Performance feedback is a central component of management exceeded or missed expectations, anecdotal evidence suggests
control systems and plays a critical role in motivating individual managers may also express how they feel about employees' per-
effort (Hannan, Krishnan, & Newman, 2008; Luckett & Eggleton, formance, i.e., their emotional reaction to employee performance
1991; Tafkov, 2013; Thornock, 2016). Performance feedback typi- (Jensen, 2020; LeaderSharp Group, 2015; Richardson, 2013). We
cally informs employees about how they are performing relative to argue that managers' use of emotional language when providing
benchmarks such as goals, targets, and peer performance. However, performance feedback has implications for employees' subsequent
when providing performance feedback, managers may performance. Specifically, we examine whether managers’ use of

*
*We are thankful for the guidance of the editor and two anonymous reviewers. We gratefully acknowledge funding from an Institute of Management Accountants
Research Foundation grant and from the University of Nebraska-Lincoln, Utah State University, Utah Valley University, and West Virginia University. We are also grateful to
Alexander Brüggen, Jace Garrett, Adam Presslee, Jordan Samet, Tyler Thomas, and workshop participants at the University of North Texas, the University of Waterloo, the BYU
Accounting Research Symposium, the American Accounting Association annual meeting, the American Accounting Association Management Accounting Section meeting, and
the Global Management Accounting Research Symposium for helpful feedback. All errors are our own.
* Corresponding author. University of Nebraska-Lincoln, HLH 445C, PO Box 880488, Lincoln, NE, 68588-0488, USA.
E-mail addresses: devon.erickson@usu.edu (D. Erickson), kip.holderness@mail.wvu.edu (D.K. Holderness), kari.olsen@uvu.edu (K.J. Olsen), tthornock2@unl.edu
(T.A. Thornock).

https://doi.org/10.1016/j.aos.2021.101329
0361-3682/© 2021 Elsevier Ltd. All rights reserved.

Please cite this article as: D. Erickson, D.K. Holderness Jr., K.J. Olsen et al., Feedback with feeling? How emotional language in feedback affects
individual performance, Accounting, Organizations and Society, https://doi.org/10.1016/j.aos.2021.101329
D. Erickson, D.K. Holderness Jr., K.J. Olsen et al. Accounting, Organizations and Society xxx (xxxx) xxx

emotional language moderates the effect of relative performance language used in Experiment 1. Additionally, we manipulate the
feedback on employee effort responses. presence of a statement designed to shift participants’ attention
Feedback Intervention Theory (FIT) provides a comprehensive away from the self. Specifically, for half of participants, we include a
framework to understand feedback responses (Kluger & DeNisi, statement from the manager at the end of the feedback report
1996). Part of FIT suggests that employees' reactions to perfor- stating that the feedback is not a reflection of the employee as a
mance feedback depend on their locus of attention. When em- person and speaks only to their performance in that round of the
ployees are more task-focused, they are more likely to interpret task. To the extent that attention to the self drives the effect of
feedback as a signal of whether more or less effort is required. In emotional language on performance, we hypothesize that this
contrast, when employees are more self-focused, they are more intervention reduces the harmful effect of emotional language. We
likely to fixate on what the feedback implies about them as a per- find that our results from Experiment 1 for the negative feedback
son. The extent to which employees allocate attention to the task or valence condition hold for AMT participants. Participants who
the self depends in part on feedback cues. We expect that a man- received emotional language accompanying their negative feed-
ager's use of emotional language when providing performance back exhibited lower subsequent performance relative to partici-
feedback is a feedback cue that influences employees' locus of pants for whom the emotional language was absent. Further, for
attention. Specifically, we predict that a manager's use of negative participants who received our intervention, the effect of emotional
emotional language increases attention to the self and results in language was attenuated, consistent with the notion that increased
reduced employee effort in response to negative feedback relative attention to the self drives the effect of emotional language on
to feedback that does not contain emotional language. Receiving performance.
positive emotional language may also divert attention from the task In Experiment 3, we again manipulate the presence of
to the self. However, prior research suggests that positive feedback emotional language accompanying negative performance feedback.
is less likely to challenge or change individuals' self-image than However, immediately after receiving feedback, participants are
negative feedback (Blaine & Crocker, 1993, pp. 55e85), which may asked to complete a separate task labeled the “Linguistic Implica-
mute any distracting effect of positive emotional language. Further, tions Form,” which is designed to directly and subtly measure in-
although positive feedback is more likely than negative feedback to dividuals' attention to the self (Wegner & Giuliano, 1980, 1983). In
be accepted as accurate, it often does not inspire increases in effort this task, participants read a series of generic statements, each with
because it can be seen as a signal of sufficiency (Smith, Haynes, a missing pronoun, and are asked to choose among three gram-
Lazarus, & Pope, 1993; Taylor, Fisher, & Ilgen, 1984). As such, it is matically correct pronouns to fill in the blank. One of these pro-
unclear ex ante whether positive emotional language affects em- nouns is self-referent (e.g., “I” or “me”), whereas the other
ployees' effort. pronouns refer to groups or others (e.g., “we,” “he,” or “they”). The
To examine our research questions, we conduct several experi- extent to which participants select self-referential pronouns serves
ments in which participants assume the role of workers in a firm as a proxy for their attention to the self. Consistent with our theory,
and receive relative performance feedback on a simple effort-based we find that participants whose feedback included emotional lan-
task. After receiving feedback, participants complete an additional guage from the manager exhibit higher attention to the self than
round of the task. Our primary dependent variable is participants’ participants whose feedback did not include emotional language. In
performance in the subsequent round, controlling for previous- combination, Experiments 2 and 3 provide triangulating evidence
round performance. with Experiment 1 that the presence of emotional language in
In our first experiment (Experiment 1), we utilize student par- negative feedback increases individuals’ attention to the self,
ticipants in a laboratory setting. We manipulate feedback valence resulting in diminished future performance relative to when
between subjects by comparing each participant's performance feedback is provided without emotional language.
with that of a higher- or lower-performing participant, resulting in Our paper makes several contributions to practice and
valid negative or positive feedback. We also manipulate the pres- academia. We provide new insights on how managers can improve
ence of valence-consistent manager emotional language between the effectiveness of negative performance feedback e an important
subjects. Half of the participants receive a statement communi- tool for managers (Finkelstein & Fishbach, 2011; Grant, 2013) e by
cating management's feelings about their performance (i.e., that better understanding the effects of negative emotional language.
the manager is disappointed or pleased), whereas the remaining The results of our paper suggest that by avoiding negative
participants do not receive such a statement. emotional language when providing feedback, managers can
Consistent with our predictions, Experiment 1 provides evi- enhance employees' subsequent effort. Many companies now uti-
dence that among participants receiving negative feedback from lize software that continuously monitors employee performance
the manager, those who also receive emotional language in the and facilitates feedback provision without the need for interaction
feedback (i.e., an expression of manager disappointment) perform with a manager (Gellman & Baer, 2016; Goetz, 2011; Silverman,
worse in the subsequent round of the task relative to those for 2011, 2016). The gamification of workplace performance, for
whom the emotional language is absent. In contrast, among those example, allows employees to compare their performance with
who receive positive feedback from the manager, those receiving that of coworkers in real time (Manjoo, 2014). From a practical
positive emotional language in the feedback (i.e., an expression perspective, our results suggest that such workplace information
indicating that the manager is pleased) perform no differently than systems, which allow companies explicit control over whether
participants for whom the emotional language is absent. manager emotional language is included in performance feedback,
To provide evidence that our proposed mechanism (attention to can improve the efficacy of feedback by removing language sug-
the self) drives the effect of emotional language on performance in gestive of manager emotion when providing negative feedback. To
the negative feedback valence condition, we perform two addi- the extent emotional language is communicated (e.g., uninten-
tional experiments e one with an intervention designed to atten- tionally), our results also suggest that managers can alleviate the
uate attention to the self, and another with a direct measure of negative effects of emotion by specifying that the feedback reflects
attention to the self. We administer Experiment 2 online using employees’ performance but does not reflect on who they are as a
Amazon Mechanical Turk (AMT) workers as participants. In this person.
experiment, we manipulate the presence of emotional language Our study of manager emotional language also contributes to
accompanying negative performance feedback, consistent with the academic literature on feedback. To our knowledge, this study is
2
D. Erickson, D.K. Holderness Jr., K.J. Olsen et al. Accounting, Organizations and Society xxx (xxxx) xxx

the first to provide evidence that emotional expressions within According to FIT, feedback cues can direct individuals’ locus of
feedback serve as feedback cues that cause individuals to shift attention to the task or to the self.4 When feedback cues direct
attention from the task to the self. Our study provides corrobo- attention more to the task, individuals are more likely to consider
rating evidence of specific aspects of FIT through process evidence the implications of the performance feedback for their task moti-
(Asay, Guggenmos, Kadous, Koonce, & Libby, 2019) from both a vation and effort. Consequently, individuals interpret negative
moderation-of-process design and a direct measurement approach performance feedback as a signal that effort should be increased
using the Linguistic Implications Form to measure attention to the and positive performance feedback as a signal that effort should be
self. maintained or possibly reduced (Kluger & DeNisi, 1996, p. 263). In
Prior research emphasizes differing elements of leader emotion contrast, when feedback cues direct attention more to the self, in-
in distinct contexts. For example, Gaddis, Connelly, and Mumford dividuals are more likely to think about meta-task processes (i.e.,
(2004) provide evidence that expressions of support and/or anger processes that link the focal task to higher order goals such as self-
from group leaders who relay feedback from a superior influence esteem, control, and impression management). Thus, attention to
other group members' attitudes regarding the leader's effective- the self may divert cognitive resources from task performance and
ness and the quality of the group's creative output. Also, Van Kleef is likely to result in affective responses that contribute to this
et al. (2009) provide evidence that manager anger interacts with diversion (Belschak & Den Hartog, 2009). For example, individuals
group members' motivation to influence group performance rela- who receive negative performance feedback and are more attuned
tive to manager happiness. In contrast, in our study, we utilize an to the self due to the inclusion of emotional language may disen-
individual task that is effort-driven. Thus, our results speak specif- gage from the task as they devote cognitive resources to manage
ically to the influence of a manager's emotional language on indi- their feelings by diminishing the perceived fairness of the feedback
vidual employee effort, whereas the mechanism by which manager or importance of the task (Kluger & DeNisi, 1996, p. 266).
emotion influences group performance in prior work is unclear.1 In
addition, while this prior work examines the effect of manager 2.1.2. Feedback and manager emotional language
anger on performance, our study examines the influence of man- We build upon this theoretical reasoning to make predictions
ager disappointment e arguably a more appropriate work-place about how managers’ emotional language in performance feedback
expression of emotion than anger (Shellenbarger, 2012). Since influences subsequent employee effort. We define emotional lan-
anger and disappointment lead to different social behaviors guage as language that expresses an emotional reaction of the
(Johnson & Connelly, 2014), it is unclear ex-ante whether manager feedback provider. When delivering performance feedback, man-
disappointment will lead to a distinct effort response.2 agers may include emotional language that reflects their feelings
about employee performance either intentionally (e.g., as an
attempt to further motivate subordinates e see Clark, Pataki, &
2. Emotional language in feedback and employee effort Carver, 1996; Lindebaum & Fielden, 2011) or unintentionally.
(Experiment 1) With the adoption of modern information systems capable of
providing frequent feedback without face-to-face contact with
2.1. Development of hypotheses managers (Gellman & Baer, 2016; Goetz, 2011; Silverman, 2011),
companies can avoid any non-intentional inclusion of emotional
2.1.1. Feedback Intervention Theory e attention to the self and the language with the provision of feedback. In other words, modern
task feedback systems provide companies with enhanced control over
Kluger and DeNisi (1996) provide a theoretical framework for whether emotional language is provided.
evaluating the effects of feedback interventions, known as Feed- As shown in Fig. 1, FIT suggests that the extent to which in-
back Intervention Theory (FIT). FIT suggests that negative and dividuals direct attention to the task or the self depends in part on
positive feedback (i.e., whether an individual performs lower or feedback intervention cues, i.e., characteristics of the task and of
higher than a benchmark) create discrepancies that individuals the feedback recipient/situation. We first consider the case of
attempt to resolve. However, the methods employed to resolve valence-consistent emotional language accompanying negative
these discrepancies (e.g., expending more or less effort) depend on feedback. We posit that a manager's use of negative emotional
various factors, including the extent to which feedback recipients language represents an important feedback intervention cue that
are focused on the task or on meta-task processes (i.e., the self). prompts employees to direct their attention away from the task and
Fig. 1, adapted from Kluger and DeNisi (1996) provides an overview toward the self. When managers express negative emotion in re-
of FIT as well as the specific elements of the theory addressed by our action to employee performance, they increase the likelihood that
study.3 employees also feel negative emotion. This idea is consistent with
the literature regarding “emotional contagion,” which suggests that
emotions experienced by one individual are mimicked by others,
1
Group performance on creative tasks may result from various factors (e.g., resulting in a convergence of emotions (Hatfield, Cacioppo, &
attitude, cooperation, creativity, etc.).
2 Rapson, 1993). Importantly, the psychology literature provides ev-
Johnson and Connelly (2014) provide evidence that displays of anger accom-
panying feedback lead to fewer positive social behaviors than displays of disap- idence that feelings of negative emotion increase individuals'
pointment. The authors also examine the effect of these emotions on performance attention to the self (Salovey, 1992; Sedikides, 1992; Wood,
for a creative task but find no evidence that either anger or disappointment affects Saltzberg, & Goldsamt, 1990).5 Thus, we expect that when man-
performance relative to no emotion. Our paper extends this literature by examining agers use negative emotional language, emotional contagion causes
the influence of disappointment on effort and provides contrasting results.
3
Because we select relative performance information as our form of feedback,
social comparison theory could also be applied in this setting, which suggests that
4
individuals will exert greater effort to maintain their self-concept or status (Hannan Feedback recipients may also direct attention to task details when receiving
et al., 2008; Hannan, McPhee, Newman, & Tafkov, 2013; Kerr et al., 2007; Tafkov, feedback, particularly when the feedback provides specific guidance for learning
2013). However, social comparison theory provides less guidance on the expected about and improving on the task (Kluger & DeNisi, 1996). However, in our study we
effects of emotional language and applies to a narrower set of feedback (relative use a basic, effort-based task that limits the potential for learning.
5
performance information). In contrast, FIT applies to various types of feedback and This idea is also consistent with suggestions in FIT that when feedback
incorporates discussion of feedback cues, which is relevant to our research threatens self-esteem or discourages the recipient, attention is directed more to-
questions. ward the self and less toward the task.

3
D. Erickson, D.K. Holderness Jr., K.J. Olsen et al. Accounting, Organizations and Society xxx (xxxx) xxx

Fig. 1. Theoretical framework e adapted from Feedback Intervention Theory in Kluger and DeNisi (1996).

employees to feel negative emotion, and consequently, increases positive emotional language from the manager will not affect
employees' attention to the self. As a result, our previous discussion employee effort responses relative to when positive feedback does
of FIT suggests that increased attention to the self adversely in- not contain positive emotional language.
fluences subsequent employee performance responses to negative
feedback. In other words, although negative feedback creates a
2.2. Experimental design
discrepancy that employees may seek to resolve via increased
effort, a shift in attention from the task to the self diverts cognitive
2.2.1. Experimental procedures
resources away from task effort and toward management of self-
To test H1 and H2, we first conducted a laboratory experiment
perception. Accordingly, we predict that emotional language
(Experiment 1). We employed a between-subjects design manip-
accompanying negative feedback diminishes the effect of the
ulating feedback valence (i.e., positive versus negative) and the
feedback on employee effort.
presence of valence-consistent emotional language from the
H1. . When employees receive negative performance feedback, manager (i.e., present versus absent). We manipulated the presence
negative emotional language from the manager will decrease of valence-consistent emotional language, rather than fully crossing
employee effort responses relative to when negative feedback does the presence of positive and negative emotional language, to avoid
not contain negative emotional language. potential issues that can arise when participants receive conflicting
signals from the feedback message and the manager's emotional
The influence of emotional language accompanying positive
language (Keltner & Haidt, 1999). For example, participants who
feedback on performance responses to feedback is less clear. Posi-
receive negative feedback that contains language suggesting that
tive feedback often does not motivate increases in effort, as positive
the manager is pleased with their performance may logically infer
feedback can be seen as a sign that past efforts were sufficient
that the feedback message itself is inconsequential.
(Podsakoff & Farh, 1989; Smith et al., 1993; Taylor et al., 1984). To
We recruited 178 students from a large Midwestern university
the extent positive feedback does not affect effort, emotional lan-
to participate in the experiment. Participants were asked to
guage may not moderate effort responses. In addition, although we
perform an effort-based task: a modified version of a letter-
expect that emotional language from the manager creates positive
decoding task (Chow, 1983; Holderness, Olsen, & Thornock, 2017;
emotions for feedback recipients due to emotional contagion
Kelly, Webb, & Vance, 2014). This task was selected because it
(Hatfield et al., 1993), research is mixed on whether positive
provides a clean measure of real effort exerted by individuals while
emotions lead to increased attention to the self (Salovey, 1992;
reducing the likelihood that other factors, such as learning or
Sedikides, 1992; Wood et al., 1990). In fact, research suggests that
creativity, drive any observed effects. Participants first completed a
positive emotions can decrease attention to the self (Green,
pre-experimental questionnaire when they signed up for the study.
Sedikides, Saltzberg, Wood, & Forzano, 2003). Finally, even if pos-
In addition to indicating their availability for the primary experi-
itive emotional language does increase attention to the self, prior
ment, participants completed demographic information.
literature suggests that positive feedback is less likely than negative
Approximately two weeks later, participants completed the lab
feedback to threaten individuals’ self-image (Blaine & Crocker, 1993,
experiment. Upon entering the lab, participants were situated at a
pp. 55e85), and thus, cognitive resources may not be needed to
computer desk. The administrator announced that the participants
manage self-perception. Because it is unclear whether positive
were being asked to assume the role of production workers in a
emotional language will have an effect on employee effort in
fictitious company and that the administrator was assuming the
response to positive feedback, we state the following hypothesis in
role of the “quality control manager.” Participants then read the
null form:
instructions about the production task e a letter-decoding task
H2. . When employees receive positive performance feedback, designed to mirror the productive task of testing circuit boards
4
D. Erickson, D.K. Holderness Jr., K.J. Olsen et al. Accounting, Organizations and Society xxx (xxxx) xxx

(Chow, 1983). After completing a 5-min practice round to famil- performance was immediately adjacent to theirs based on a sorted
iarize themselves with the task, participants were asked to answer list of Round 1 production. This design ensures a balance of high
a brief questionnaire assessing their interest in and opinion of the (low) performers receiving negative (positive) feedback. For
task. Prior to beginning Round 1 production, we provided the example, based on the sorted list, the top performer in the session
following information to all participants: and the second highest performer were paired together, and thus,
TenTech evaluates employee performance by comparing the the top performer received positive feedback, whereas the second
number of circuit boards an employee successfully decodes to the highest performer received negative feedback. This design also
number successfully decoded by another employee. TenTech will maintains the integrity of the lab results if participants were to
provide a performance report with this comparison at the end of discuss their experience with participants yet to take part in the
each round. experiment. Perhaps most importantly, this design helps overcome
This statement eliminates any surprise that relative perfor- the challenge of providing positive or negative feedback that is both
mance information is provided and also specifies that the company valid (i.e., related to actual task performance) and disentangled
evaluates employees based on relative performance rather than on from participant knowledge, ability, and performance. The majority
some other performance metric. Round 1 production, which lasted of prior research provides feedback that either controls for partic-
10 min, then began. ipants’ performance (Podsakoff & Farh, 1989; Vancouver &
Upon completion of Round 1 production, the quality control Tischner, 2004) or relates to actual task performance (Hannan
manager provided participants with a paper-based performance et al., 2008; Loftus & Tanlu, 2018), but not both. The “zipper-
feedback report. We provided paper-based reports to increase both matched” design provides valid feedback based on actual task
the salience of the feedback message and the likelihood that par- performance while reducing the correlation between ability and
ticipants attended to the feedback, consistent with prior research feedback valence.6
(Holderness et al., 2017). The report contained a summary of par- Our second manipulation on the performance feedback report
ticipants’ performance and indicated how their performance adds a sentence containing emotional language from the quality
compared with that of another participant in the same session, control manager that is consistent with the feedback valence. Par-
which we manipulated between subjects as described below. We ticipants who receive emotional language read the following: “I am
also manipulated whether the report included an additional sen- very pleased (very disappointed) that you completed more (fewer)
tence containing emotional language from the quality control circuit boards than the other employee.” We utilize these specific
manager that corresponded to the valence of the feedback received. emotions because they represent professionally appropriate and
After reviewing the performance feedback report, participants anecdotally common manager responses to employee perfor-
completed Round 2 production for an additional 10 min. Partici- mance.7 This phrase attributes the manager's emotion specifically
pants then viewed another feedback report similar to the report to the relative performance of the employee to prevent participants
viewed after Round 1. We ensured that the valence was consistent from interpreting the emotion as deriving from some other source
from the feedback received after Round 1 by comparing partici- (e.g., from raw performance). For the conditions without emotional
pants who previously received negative (positive) feedback with language, this sentence is omitted. See Fig. 2 for an example of the
the highest (lowest) performer in Round 2. We discuss the reasons feedback report.
for including a third round and related results in section 4.2.
Finally, participants were asked to complete a post- 2.2.3. Dependent variable
experimental questionnaire. To avoid any experimenter demand Our primary dependent variable of interest is the number of
effects or ‘end-game’ strategies, we did not tell participants the correct letter decodes in Round 2 of the experiment (hereafter
number of rounds of the task they would be performing Round 2 Production). In order to control for any variation in Round 1
(Murnighan & Roth, 1983). We paid participants $20 for their performance by condition, we control for Round 1 Production in all
participation in the study, which was comprised of approximately analyses.8 Round 2 Production should differ based on how partici-
10 min on the pre-experimental questionnaire and 50 min in the pants respond to the feedback report received between rounds.
behavioral lab.
2.3. Results
2.2.2. Experimental manipulations
We manipulate two variables in this study. Both manipulations Panel A of Table 1 provides descriptive statistics. Of the 178
take place as part of the performance feedback report that partic- participants, 58 percent were female. Participants were 20 years
ipants received between Round 1 production and Round 2 pro-
duction. Our two manipulated variables are (1) the valence (i.e.,
sign) of the feedback, and (2) the presence of valence-consistent 6
We note that by using actual task performance, we reduce, but cannot elimi-
emotional language from the manager. In our analyses, we define nate, the correlation between ability and feedback valence. Average Round 1 per-
Feedback Valence as an indicator variable equal to one if the feed- formance was 33.61 for participants receiving positive feedback and 31.41 for
participants receiving negative feedback. We also note that if we drop the top four
back was positive and zero if the feedback was negative. We define
and bottom four participants in the positive and negative valence conditions,
Emotional Language as an indicator variable equal to one if valence- respectively, we no longer find a statistical correlation between performance and
consistent manager emotional language was presented as part of valence, and our reported results continue to hold without change of inference.
7
the performance feedback and zero otherwise. Other potential negative emotional responses may include anger or sadness.
Our first manipulation on the performance feedback report However, anger may fall outside the realm of what is deemed appropriate manager
behavior (e.g., Shellenbarger, 2012), and sadness seems less targeted to the
provides relative performance information (RPI) that tells partici-
employee performance and more reflective of manager coping ability. In addition,
pants if they completed “more” or “fewer” circuit boards than given our use of the manager being ‘pleased’ in the positive valence condition,
another employee. We utilize RPI as our feedback setting because it manager ‘disappointment’ seems a reasonable counterpart. We note that we do not
allows us to provide valid, non-deceptive feedback without compare the magnitude of the influence of emotional language for positive vs.
negative feedback; rather, we focus on the directional effect of emotional language
conflating participant ability or performance with feedback
within each feedback valence condition.
valence, and because it provides a salient benchmark of perfor- 8
We note that results are inferentially identical using a change variable from
mance. We follow the “zipper-matched” design of Holderness et al. Round 1 to Round 2 performance. We include these analyses alongside those of our
(2017) by matching each participant to the participant whose main dependent variable in all tables.

5
D. Erickson, D.K. Holderness Jr., K.J. Olsen et al. Accounting, Organizations and Society xxx (xxxx) xxx

Fig. 2. Example of feedback report.

Table 1
Descriptive statistics for Experiment 1 (laboratory).

Panel A: Demographics and Psychological Entitlement Scale

Gender Age

Male (n) 75 Mean 19.97


Female (n) 103

Panel B: Performance Change by Sample Condition

Round 1 Round 2 Performance Round 3 Performance


Production Production Change Production Change
R1 to R2 R2 to R3

Condition n Mean (Std. Dev.) Mean (Std. Dev.) Mean (Std. Dev.) n* Mean (Std. Dev.) Mean (Std. Dev.)

Negative Valence/No Emotional Language 44 32.98 (6.53) 35.66 (6.33) 2.68 (2.67) 42 35.69 (6.48) 0.52 (3.37)
Negative Valence/Emotional Language 47 31.11 (5.07) 32.96 (5.79) 1.85 (3.15) 46 35.00 (4.94) 2.37 (2.96)
Positive Valence/No Emotional Language 43 33.14 (6.54) 35.53 (6.35) 2.40 (2.80) 41 36.78 (6.29) 0.71 (2.81)
Positive Valence/Emotional Language 44 32.47 (6.44) 34.75 (5.99) 2.27 (2.18) 41 36.46 (5.92) 1.29 (2.49)

Notes: Raw means are presented in Panel B. All variables are defined in the Appendix. *We excluded eight participants from analyses involving Round 3 Production as the
valence of their feedback was inconsistent between feedback instances. As such, the means for performance change from R2 to R3 will not cross foot.

6
D. Erickson, D.K. Holderness Jr., K.J. Olsen et al. Accounting, Organizations and Society xxx (xxxx) xxx

guage affects Round 2 Production (p ¼ 0.72, two-tailed), consistent


with the null hypothesis in H2.10
Overall, the results of our laboratory experiment support our
hypotheses. We find evidence that the presence of emotional lan-
guage accompanying negative feedback decreases employees'
effort in response to the feedback relative to when such language is
absent, consistent with H1. In contrast, the presence of emotional
language accompanying positive feedback has no effect on em-
ployee's effort in response to the feedback relative to when such
language is absent, consistent with the null hypothesis in H2.

3. Emotional language in feedback, attention to the self, and


employee effort (Experiment 2)

3.1. Development of hypothesis

Experiment 1 was designed to test the effects of emotional


language on employee effort in response to feedback. Next, we
Fig. 3. Experiment 1 (laboratory): Effects of Emotional Language and Feedback Valence consider an additional feedback intervention cue designed to
on Round 2 Production. counteract the effect of emotional language on attention to the self.
Specifically, we examine whether a statement from the manager
indicating that the feedback received relates to performance on the
old on average. Panel B of Table 1 provides descriptive statistics for task, not to who the employee is as a person, mitigates the influence
Round 2 Production broken out by experimental condition.9 of emotional language on employee effort. This investigation serves
Fig. 3 graphically presents the results. Table 2 provides results dual purposes. First, it represents a test of the theoretical mecha-
related to H1 and H2. Panel A of Table 2 reports results for our main nism we propose in Experiment 1 for the effects of emotional
dependent variable e Round 2 Production e while controlling for language on employee effort. This approach follows suggestions by
Round 1 Production. We formally test H1 and H2 by examining the Bonner (2008) and Asay et al. (2019) to provide process evidence
simple effects of Emotional Language within the Negative and Pos- through moderation. If, as we theorize, increased attention to the
itive Valence conditions. In Panel A of Table 2, we report results for self drives the effect of emotional language on employee effort,
the Negative Valence subsample. In support of H1, we find that then an intervention designed to reduce attention to the self and to
Emotional Language has a statistically significant negative effect on increase attention to the task should moderate (attenuate) the ef-
Round 2 Production (p ¼ 0.04, one-tailed). In the Positive Valence fect of emotional language on effort. Second, such an investigation
condition, we do not find evidence that positive Emotional Lan- provides managers with a potential “fix” to counteract the effects of

Table 2
Experiment 1 (laboratory): Effects of Emotional Language and Feedback Valence on performance.

Panel A: GLM on Round 2 Production

Factor df MS F p-value

Round 1 Production 1 5275.45 746.05 <0.01


Emotional Language 1 16.372 2.32 0.13
Feedback Valence 1 0.975 0.14 0.71
Emotional Language  Feedback Valence 1 7.74 1.09 0.15
Error 173 7.07
Simple Effects:
Emotional Language when Valence ¼ Negative 1 21.73 3.07 0.04
Emotional Language when Valence ¼ Positive 1 0.90 0.13 0.72

Panel B: GLM on Performance Change


Factor df MS F p-value
Emotional Language 1 10.10 1.36 0.25
Feedback Valence 1 0.20 0.03 0.87
Emotional Language  Feedback Valence 1 5.57 0.75 0.19
Error 174 7.45
Simple Effects:
Emotional Language when Valence ¼ Negative 1 15.68 2.10 0.07
Emotional Language when Valence ¼ Positive 1 0.33 0.04 0.83

Notes: All variables are defined in the Appendix. Reported p-values are one-tailed as indicated in bold type if predicted; two-tailed otherwise.

emotional language included in feedback, unintentionally or


otherwise. Because theory and the results from Experiment 1
9
suggest that the effect of emotional language only occurs when
Approval for all experiments was granted by the institutional review board
accompanying negative feedback, we limit our discussion to the
(IRB) of the institution at which the experiment took place.
10
In Panel B of Table 2, we report results of identical tests using Performance negative feedback condition.
Change as the dependent variable. We find consistent results with those reported FIT suggests that various feedback intervention cues determine
above.

7
D. Erickson, D.K. Holderness Jr., K.J. Olsen et al. Accounting, Organizations and Society xxx (xxxx) xxx

the attention feedback recipients direct at the self and the task, as incorrectly identifying the valence for one or both of the feedback
outlined in Fig. 1. H1 predicts that emotional language serves as a instances. We also excluded one participant who failed an
feedback cue, shifting feedback recipients’ attention away from the attention-check question regarding what real-world task the
task and toward the self. In contrast, a subsequent statement from decoding task represents. Finally, because feedback should only
the manager indicating that the feedback pertains to performance matter to participants who actually exert effort on the experimental
on the task and does not speak to who the employee is as a person task, we excluded 25 participants who exhibited distracted effort
should prompt employees to shift their attention back to the task by completing fewer than 10 letter decodes in any period.12,13 We
and away from the self. As a result, we propose that when note that at the beginning of the Qualtrics survey, all participants
emotional language is present, the inclusion of this statement from were instructed that they should read all instructions carefully and
the manager will attenuate the influence of emotional language on work on the tasks diligently, and that failure to do so would result
employee effort. in being removed from the study. Our exclusion criteria resulted in
179 participants for our primary analyses.14
H3. . A statement from the manager indicating that feedback
received pertains to performance on the task, not to who the
employee is as a person, attenuates the adverse influence of 3.3. Results
negative emotional language on employee effort responses.
Panels A and B of Table 3 provide descriptive statistics for
3.2. Experimental design Experiment 2. We graphically present related results in Fig. 4.
Panels A and B of Table 4 reports our primary analyses for
We administer Experiment 2 to Amazon Mechanical Turk (AMT) Experiment 2, which are conducted similarly to our analyses of
workers as participants via Qualtrics. We manipulated two factors Experiment 1, with Round 2 Production as our primary dependent
in this experiment. First, in order to replicate our main result of H1 variable, and a control for Round 1 Production included in the
(i.e., that emotional language with negative feedback harms per- model. We find a statistically significant interaction of Emotional
formance) for our online participants, we manipulate the presence Language * Prompt on Round 2 Production, such that performance is
of emotional language accompanying negative feedback, identical lowest when Emotional Language is present and Prompt is absent
to the approach in Experiment 1. Second, we manipulate the (p ¼ 0.07, one-tailed), suggesting support for H3.15 Examining the
presence of an additional statement from the manager designed to pattern of our results, we note that the simple effect of Emotional
prompt participants to focus on the task rather than the self. Spe- Language within the No Prompt condition is significant (p ¼ 0.03,
cifically, for half of the participants we included the following one-tailed), consistent with H1 and replicating our laboratory
statement after the emotional statement: “Please note that this study's finding that emotional language reduces employee effort in
feedback speaks only to your performance on this round of the task response to negative feedback. Providing further support for H3, we
and is not a reflection of who you are as a person.” The other half of find that, in the presence of the Prompt, the simple effect of
participants did not receive this additional message in their feed- Emotional Language is no longer significant (p ¼ 0.83).
back reports. For our analyses, we define Prompt as an indicator As noted by previous research, ANOVA often lacks power to
variable equal to one if the feedback included the prompt and zero identify significant ordinal interactions (Bobko, 1986; Buckless &
if it did not. Ravenscroft, 1990; Guggenmos, Piercey, & Agoglia, 2018; Rosnow
AMT workers completed the same task as in our laboratory & Rosenthal, 1995). Thus, we use a contrast test modeled on our
study with only minor changes. First, the feedback report was theoretical prediction to further explore the implied ordinal inter-
delivered electronically, with the format of the report being action. We test for a planned contrast following the guidelines
designed to mirror that of the paper-based report delivered to in- suggested by Guggenmos et al. (2018). Ex ante, we expected that
person participants. Second, we did not use zipper-matching to our intervention (Prompt) would mitigate the effect of emotional
generate feedback valence. Instead, we compared participants’ language on Emotional Language on performance, consistent with
scores to that of the highest performing participant from a prior attention to the self driving the effect of emotional language pre-
session, ensuring that nearly all participants would receive valid viously documented. However, it was not clear ex ante whether
negative feedback.11 To ensure the validity of using AMT workers, such mitigation would be full or partial. Consequently, we
we followed protocols suggested by extant literature (Bentley,
2018), restricting our sample to AMT workers with 95% or greater
12
approval levels and more than 500 approved tasks completed. We determined the 10-letter-decode cutoff by referencing the performance of
Further, we required U.S. based participants and unique I.P. ad- our laboratory participants (Experiment 1) who completed the study in a controlled
environment absent easily accessible distractions. No lab participants completed
dresses. Participants completed the task in approximately 45 min fewer than 20 letter decodes in any production round. However, our online in-
and were paid $6. strument may have been more subject to page-loading delays. Thus, we selected
We recruited 215 AMT participants to complete the study. One half the minimum lab performance (10) to avoid an overly restrictive cutoff.
13
potential concern with using online participants (outside of a lab To examine whether our experimental manipulations influenced the likelihood
that a participant was excluded from our analyses, we compare the number of
environment) is that they may not pay sufficient attention to the
participants excluded across experimental conditions. We find no evidence that the
case materials, engage in the task, and exert sufficient effort to number of participants excluded differs across assigned experimental conditions
cause them to internalize the feedback provided and our Prompt (Pearson c2 ¼ 0.05, p-value ¼ 0.82).
intervention. To alleviate this concern, we incorporated several 14
As would be expected, including participants who did not attend to the task
exclusion criteria. First, we excluded 10 participants who failed a weakens our reported effects such that the effect of emotional language without
the prompt and the interactive effect of emotional language and the prompt on
comprehension check regarding the valence of their feedback by
Round 2 Production are no longer significant (p-values > 0.10, one-tailed). In prac-
tice, this may suggest that the effects of feedback valence, emotional language, and
our intervention only occur for individuals giving some minimal level of attention
11
We selected the highest performer in the production periods used for reference and effort in their work task. We note that, when including all participants, these
from Experiment 1, who completed 56 and 56 decodes in the first and second effects on performance become significant if we combine performance from both
production periods, respectively. The highest performer from Experiment 2 production rounds after feedback (p-values < 0.10, one-tailed).
15
completed 52 and 52 decodes for these production periods, and thus, all partici- While the F-distribution is not two-tailed, this one-tailed p-value is equivalent
pants in Experiment 2 received valid negative feedback. to a one-tailed t-test.

8
D. Erickson, D.K. Holderness Jr., K.J. Olsen et al. Accounting, Organizations and Society xxx (xxxx) xxx

Table 3
Descriptive statistics for Experiment 2 (online) e negative valence condition.

Panel A: Demographics

Gender Age

Male (n) 92 Mean 39.32


Female (n) 87

Panel B: Performance Change by Sample Condition

Round 1 Round 2 Performance Round 3 Performance


Production Production Change Production Change
R1 to R2 R2 to R3

Condition n Mean (Std. Dev.) Mean (Std. Dev.) Mean (Std. Dev.) Mean (Std. Dev.) Mean (Std. Dev.)

No Emotional Language/No Prompt 44 30.95 (7.40) 32.25 (6.82) 1.30 (4.38) 33.23 (7.95) 0.98 (3.85)
Emotional Language/No Prompt 44 31.75 (7.26) 31.20 (8.13) 0.55 (4.61) 31.25 (9.42) 0.05 (4.80)
No Emotional Language/Prompt 41 30.34 (8.33) 31.24 (7.44) 0.90 (4.38) 32.20 (7.79) 0.95 (3.85)
Emotional Language/Prompt 50 30.12 (8.02) 31.26 (8.08) 1.03 (4.33) 31.26 (8.15) 0.00 (3.12)

Notes: Raw means are presented in Panel B. All variables are defined in the Appendix.

attenuation of the negative effect of emotional language found in


the Prompt subsample suggests that the mechanism by which
emotional language reduces effort in response to negative feedback
is through diverting individuals’ attention from the task to the
self.18
In Panel A of Table 4, we also provide the simple effects by
Emotional Language condition. That is, when emotional language is
absent, we find that the Prompt does not have a statistically sig-
nificant effect on performance (p ¼ 0.56). On the other hand, when
emotional language is present, we find that the Prompt has a pos-
itive and statistically significant effect on performance (p ¼ 0.04,
one-tailed). We note that the main effect of the Prompt is insig-
nificant in this model (p ¼ 0.48). Thus, our evidence is consistent
with the interventional message attenuating the effects of
emotional language by redirecting attention away from the self and
toward the task, rather than having a separate main effect on per-
formance. Overall, the results of Experiment 2 are consistent with
the framework of FIT, suggesting that emotional language acts as a
cue directing feedback recipients’ attention away from the task and
Fig. 4. Experiment 2 (online): Effects of Emotional Language and Prompt on Round 2
Production e negative valence condition.
toward the self, resulting in lower effort relative to when emotional
language is absent.

considered two potential sets of contrast weights. The first set


(þ1, 3, þ1, þ1) represents full mitigation and is consistent with a 4. Supplemental analyses
negative effect of negative emotional language when the prompt is
absent (þ1 versus 3) and no effect of negative emotional language 4.1. Supplemental experiment (Experiment 3): emotional language
when the prompt is present (þ1 versus þ1). The second set and a direct measure of self-attention
(þ2, 3, þ2, 1) represents partial mitigation. We find that the
pattern of means appears to visually fit the contrast weightings Experiment 2 provides evidence through moderation of the role
representing full mitigation (see Fig. 4). Using this set, the contrast that locus of attention plays in responding to negative feedback.
test is significant (F ¼ 3.80, p ¼ 0.03), the residual between-cells Although the intervention in Experiment 2 attenuated the negative
variance test is not significant (F ¼ 0.18, p ¼ 0.83), and the q2 sta- effect of emotional language on the negative feedback condition,
tistic (0.08) suggests that our contrast explains a large majority of the results do not provide a direct measure of attention to self.
the total variance that could have been explained. For the second Thus, as additional process evidence to supplement Experiments 1
set of contrast weights, the visual fit is not strong, but the contrast and 2 (Asay et al., 2019), and to provide triangulating evidence of
test is marginally significant (F ¼ 2.63, p ¼ 0.10). We also find that the theoretical mechanism through which we propose that
the residual between-cells variance test is not significant (F ¼ 0.77, emotional language influences effort, we perform a supplemental
p ¼ 0.47). Further, the q2 statistic (0.35) suggests that this contrast experiment (Experiment 3).
explains about two-thirds of the variance.,16,17 Overall, the Experiment 3 utilized procedures and materials essentially
identical to those from Experiment 2. AMT participants completed
a round of the experimental (letter-decoding) task and received a
16
feedback report. Similar to Experiment 2, all participants were
The results for the custom contrast test using change in performance as the
dependent variable are inferentially similar (untabulated).
17
We acknowledge that the application of multiple sets of contrast weights could
18
potentially lead to a Type I error. However, finding consistent results with both sets In Panel B of Table 4, we again report results of identical tests using Performance
of contrast weights (full and partial mitigation) provides some comfort that our Change as the dependent variable. We find inferentially consistent and statistically
conclusion is not the result of a Type 1 error. stronger results for this dependent variable.

9
D. Erickson, D.K. Holderness Jr., K.J. Olsen et al. Accounting, Organizations and Society xxx (xxxx) xxx

Table 4
Experiment 2 (online): Effects of Emotional Language and Prompt on performance e negative valence condition.

Panel A: GLM on Round 2 Production

Factor df MS F p-value

Round 1 Production 1 7152.65 401.48 <0.01


Emotional Language 1 25.11 1.41 0.24
Prompt 1 8.91 0.50 0.48
Emotional Language  Prompt 1 40.10 2.25 0.07
Error 174 17.82

Simple Effects:
Emotional Language when Prompt ¼ Absent 1 64.17 3.60 0.03
Emotional Language when Prompt ¼ Present 1 0.87 0.05 0.83
Prompt when Emotional Language ¼ Absent 1 6.25 0.35 0.56
Prompt when Emotional Language ¼ Present 1 52.28 2.93 0.04

Panel B: GLM on Performance Change


Factor df MS F p-value
Emotional Language 1 28.61 1.46 0.23
Prompt 1 18.59 0.95 0.33
Emotional Language  Prompt 1 48.08 2.46 0.06
Error 175 19.59

Simple Effects:
Emotional Language when Prompt ¼ Absent 1 74.56 3.81 0.03
Emotional Language when Prompt ¼ Present 1 1.27 0.06 0.80
Prompt when Emotional Language ¼ Absent 1 3.28 0.17 0.68
Prompt when Emotional Language ¼ Present 1 66.49 3.39 0.03

Notes: All variables are defined in the Appendix. Reported p-values are one-tailed as indicated in bold type if predicted; two-tailed otherwise.

compared with the top performer from a previous round so that difference is statistically significant (p ¼ 0.06, one-tailed).21
participants would receive negative feedback.19 We again manip- The results of Experiment 3 provide evidence using direct
ulated between subjects the presence of emotional language measurement that the presence of emotional language impacts
accompanying the feedback. However, immediately following the feedback recipients’ attention to the self. Combined with the results
receipt of feedback, participants were asked to complete a pre- of our previous experiments, we provide triangulating evidence
sumably unrelated task before continuing with the primary work that the adverse effect of emotional language on effort in response
task. This task was labeled “Linguistics Implication Form” (LIF) and to negative feedback occurs due to a diversion of attention away
was adapted from prior work in psychology regarding attention to from the task and toward the self, consistent with FIT.22
the self (Wegner & Giuliano, 1980, 1983). Participants were pro-
vided with a series of 32 sentences from which a pronoun had been
removed, with 16 after the first feedback instance and 16 after the 4.2. Supplemental analyses: repetition of emotional language
second feedback instance. They were asked to select among three
grammatically correct pronouns to fill in the blank for each missing As noted previously, in each of our experiments participants
pronoun. In every case, one of the pronouns referred to the self (e.g., completed a total of three production rounds, receiving feedback
I, me, and my), whereas the other available pronouns referred to a after Round 1 and again after Round 2. We ensured that most
group or a third party (e.g., they, we, he, and she). The number of participants received the same type of feedback after Round 2 as
self-referent pronouns selected by the participant represents the received after Round 1, which allows us to consider the effects of
attention-to-self score they received for the task. repetition of positive or negative emotional language. We first
AMT workers completed the experiment via Qualtrics. We uti- examine the effects of repetition in our laboratory experiment
lized the same exclusion criteria as in Experiment 2, resulting in (Experiment 1).
101 participants for our primary analyses.20 Participants were paid In Experiment 1, we varied (if necessary) the referent peer
$8 to complete the study, which took approximately 50 min on participant to ensure that Round 2 feedback was equivalent to
average. In untabulated analyses, we find that participants in the Round 1 feedback. Thus, participants who received negative feed-
Emotional Language condition received an average attention-to-self back after the first production round had their second production
score of 19.43 (i.e., selected an average of 19.43 self-referent pro-
nouns) on the LIF, whereas those in No Emotional Language condi-
tion received an average attention-to-self score of 17.15. This 21
Breaking down the two LIF measurements (i.e., after the first and second receipt
of feedback) results in significant effects of emotional language on the selection of
self-referent pronouns (untabulated; p ¼ 0.07 and p ¼ 0.08, respectively, both one-
tailed).
19 22
We again used the highest performer from Experiment 1, who completed 56 The purpose of this experiment is to measure attention to the self, not to
and 56 decodes in the first and second production periods, respectively. The highest replicate our results related to emotional language and effort/performance. We
performer from Experiment 3 completed 55 and 51 decodes for these production expected that because participants completed the LIF immediately following the
periods, again resulting in all participants receiving valid negative feedback. receipt of feedback and before continuing with the letter-decoding task, this
20
We recruited 118 participants and excluded 17 for failing attention/compre- interruption could diminish the salience of the feedback previously received during
hension checks or exhibiting distracted effort e 10 for missing one or both of the Round 2 of the work task, therefore diminishing any effects of emotional language.
feedback valence comprehension checks, 1 for missing the attention check about Consistent with this expectation, we do not find a difference in Round 2 Production
what the task represents, and 6 for not completing 10 or more decodes in any one controlling for Round 1 Production in the second supplemental study (p ¼ 0.26, two-
period. Including all participants does not change the statistical inferences of our tailed). We leave the question of whether distractions following the provision of
findings. feedback diminish the effects of emotional language to future research.

10
D. Erickson, D.K. Holderness Jr., K.J. Olsen et al. Accounting, Organizations and Society xxx (xxxx) xxx

Table 5
Experiment 1 (laboratory): Effects of Emotional Language and Time on performance e negative valence condition.

Panel A: Descriptive Statistics

Round 1 Round 2 Performance Round 3 Performance


Production Production Change Production Change
R1 to R2 R2 to R3

Condition n Mean (Std. Dev.) Mean (Std. Dev.) Mean (Std. Dev.) Mean (Std. Dev.) Mean (Std. Dev.)

Negative Valence/No Emotional Language 42 32.60 (6.43) 35.17 (6.04) 2.57 (2.65) 35.69 (6.48) 0.52 (3.37)
Negative Valence/Emotional Language 46 30.78 (4.61) 32.63 (5.39) 1.85 (3.18) 35.00 (4.94) 2.37 (2.96)

Panel B: Repeated Measures GLM on Performance by Period

Factor df MS F p-valueb

Between Subjects:
Round 1 Production 1 3976.71 278.16 <0.01
Emotional Language 1 0.08 0.01 0.94
Error 85 14.30

Within Subjects:
Time 1 10.58 2.11 0.15
Time  Period 1 Production 1 2.69 0.54 0.47
Time  Emotional Language 1 33.26 6.62 0.01
Error 85 5.02

Notes: Raw means are presented in Panel A. All variables are defined in the Appendix. Reported p-values are two-tailed.

round performance compared with that of a participant who evidence regarding whether feedback recipients become accus-
completed more circuit boards in the second production round, tomed to the effects of negative emotional language from man-
whereas participants who received positive feedback after the first agers, reducing its effects on effort over time.25 We leave this issue
production round had their second-round performance compared to future research.
with that of a participant who completed fewer circuit boards in the
second round.23 Given the lack of effect of Emotional Language
within the Positive Valence condition, our analyses emphasize the 5. Discussion and conclusion
influence of repetition for participants in the Negative Valence
condition.24 Performance feedback plays a critical role in the design and
In Panel A of Table 5, we report descriptive statistics for the effectiveness of management control systems. Prior literature
participants who remained in the sample for our analyses of suggests that various features of feedback influence its effective-
repetition within the Negative Valence condition. Panel B of Table 5 ness, and thus, management's ability to influence employee
provides the results of a repeated measures GLM, which accounts behavior and to achieve organizational objectives (Buchheit,
for variation within subject. The variable Time, which is measured Dalton, Downen, & Pippin, 2012; Hannan et al., 2008; Luckett &
as the ordered instance of the observed variable, represents the Eggleton, 1991; Thornock, 2016). Although managers can provide
effect of repetition. Of particular interest is the Time  Emotional feedback in a variety of ways (e.g., in-person, in writing, via com-
Language effect, which represents the change in the effect of pany information systems, etc.), any provision of feedback repre-
Emotional Language from the first chance to perform after receiving sents a form of communication between a manager and an
feedback (Round 2 Production) to the second chance to perform employee. Thus, as with all forms of communication, feedback
after a second instance of feedback (Round 3 Production). We note provisions often convey manager emotions with respect to the
that the negative effect of Emotional Language not only diminishes content of the feedback message (Jensen, 2020; LeaderSharp
with repetition (p ¼ 0.01, two-tailed), but even appears to flip such Group, 2015; Richardson, 2013). In this study, we investigate
that negative emotional language benefits performance. This result whether managers' use of emotional language affects employee
is consistent with the notion that participants becoming condi- responses to performance feedback.
tioned to communications of emotion from the manager, thus Our study contributes to the Feedback Intervention Theory (FIT)
mitigating its effects on performance. literature and the broader academic literature on performance
We also examine the effects of repetition within our online feedback. Our results suggest that emotional language accompa-
experiment (Experiment 2). In untabulated results, we find that the nying negative feedback serves as a relevant feedback cue that in-
interactive effect of Time  Emotional Language is insignificant fluences the relative attention individuals direct to the task and to
(p ¼ 0.29, two-tailed) in the No Prompt condition (i.e., our replica- the self, which in turn influences effort responses to negative
tion of the negative emotion condition of Experiment 1). In fact, the feedback. This evidence serves as validation of FIT (Kluger & DeNisi,
pattern of means seems to suggest a stable negative effect of 1996) and provides new insights on a prevalent cue that influences
emotional language over time. As such, we have conflicting how individuals process feedback. We also find that when man-
agers specify that feedback relates to task performance and not to
who the employee is as a person, the influence of emotional lan-
guage diminishes.
23
In rare cases, participants could not be provided valid and consistent feedback Our findings also have several practical implications for
after Round 2 because they were the highest (lowest) performer, and thus, could
not receive negative (positive) feedback. Such participants were removed from our
analyses regarding repetition of emotional language.
24 25
Within the Positive Valence condition, we find no evidence that repetition Differences in the effects of repetition between our two studies may be driven
(Time) interacts with Emotional Language to influence participants' performance by the setting in which the experiments took place (i.e., in the laboratory vs.
(untabulated). online).

11
D. Erickson, D.K. Holderness Jr., K.J. Olsen et al. Accounting, Organizations and Society xxx (xxxx) xxx

managers with respect to the emotions they convey when contributions and insights that would result from studying the
providing feedback. First, because emotional language reduces effects of manager emotional language in performance feedback
employee effort in response to negative feedback, companies and settings within the workplace. We also acknowledge that various
managers could benefit from actions that help managers learn to factors in real-world settings may moderate the effects we identify.
control their emotions. Training regarding the importance of For example, although real-world feedback and emotional language
emphasizing task-relevant feedback and limiting the expression of is likely much more salient to workers given the implications of
negative emotion (i.e., decoupling negative emotional language such on their compensation and continuing employment, distrac-
from the delivery of negative performance feedback) may help tions in the work environment may diminish the salience of feed-
managers prevent subsequent employee performance declines. Our back and emotional language as work tasks are completed. We
results also imply that to the extent companies can remove the believe future research can address the extent to which our results
opportunity for expressions of emotion (e.g., by formalizing feed- generalize to various real-world settings.
back messages), the effectiveness of negative feedback may be The repeated nature of workplace interactions and development
enhanced. For some companies, formalizing feedback messages of professional associations and familiarities could influence the
may not feasible, and therefore complete control over manager behavioral responses to performance feedback and emotional
expressions of emotion may not be possible. However, the results of language in manager-employee relationships. For example, while
our second experiment suggest that by specifying that feedback managers may express a particular emotion, the personal knowl-
relates to the work performed rather than to who the employee is edge employees have about their manager might suggest alterna-
as a person, managers can counteract the adverse effects of tive perceptions or interpretations of the manager's true emotional
emotional language. state. Our study presents emotions in a stark laboratory or online
Our results are particularly relevant given the increasing prev- environment where performance can only be increased by greater
alence of feedback delivery via company information systems. For effort, whereas workplace relationships contain additional richness
example, some companies now provide employees with real-time and complexities. Although we examine repeated instances of
access to their performance information via workplace gamifica- emotional language, our findings are mixed, and we acknowledge
tion or ranking systems (Manjoo, 2014). To the extent these new that such repetition lacks the familiarity between managers and
feedback methods reduce or eliminate direct manager-employee employees more common in the workplace. We leave the issues of
interactions during the provision of feedback, they also provide the effects of repetition of emotional language and familiarity with
companies with the opportunity to more fully control whether the manager to future research. We further acknowledge that our
emotional language is included when providing performance focus on two specific manager emotions (disappointment and
feedback. Furthermore, while prior work emphasizes verbally pleasure) limits our ability to generalize our findings to other
communicated emotion, our study emphasizes written emotional emotions that may arise in workplace feedback settings such as
language, arguably an increasingly important medium for feedback anger, disgust, resentment, joy, and excitement. Future work could
provisions (Baker, 2010). Another consequence of real-time per- delve further into the array of manager emotions and the effects
formance feedback is a shift from a traditional annual review cycle they may have in performance feedback settings.
to the deliverance of frequent, informal feedback (Cappelli & Travis, We also acknowledge that the results of our study are specific to
2016; Silverman, 2016). Because this study cannot speak to the the setting we employ e written delivery of expected relative
extent to which negative emotional language affects performance performance information with valence-consistent emotional lan-
in an informal setting (i.e., where employees are unaware of the guage. Feedback medium (e.g., in-person vs. impersonal), feedback
timing and format of feedback), we defer this issue to future type (e.g., relative performance information vs. absolute perfor-
research. mance), feedback expectation (expected vs. unexpected), the con-
We acknowledge that although our experimental design and sistency of emotion with feedback valence (e.g., valence-consistent
use of an experimental laboratory provides critical benefits related vs. valence-inconsistent), and the frequency of feedback provisions
to the internal validity and causal inferences derived from the may all influence employee responses to manager emotional lan-
study, our ability to generalize results to real-world workplace guage. We believe future research on these issues is warranted.
settings is limited. We replicate our primary findings using an on-
line worker sample, which provides some added external validity
for our results holding across different samples. We argue that the
theoretical reasoning behind our predictions would extend to other APPENDIX. Variable Definitions
samples and settings, but acknowledge the potential future

Variable Definition

Emotional An indicator variable equal to 1 if valence-consistent emotional language was presented as part of the performance feedback and 0 otherwise
Language
Feedback Valence An indicator variable equal to 1 if the feedback was positive and 0 if the feedback was negative
Performance The change in the number of correct letter decodes between production rounds of the experiment
Change
Prompt An indicator variable equal to 1 if the feedback included a prompt that feedback “speaks only to your performance on this round of the task and is not a
reflection of who you are as a person,” and 0 otherwise.
Round 1 The number of correct letter decodes from Round 1 of the experiment
Production
Round 2 The number of correct letter decodes from Round 2 of the experiment, which is immediately following the first instance of performance feedback
Production
Round 3 The number of correct letter decodes from Round 3 of the experiment, which is immediately following the second instance of performance feedback
Production
Time An indicator variable of whether Performance Change represents change from Round 1 to Round 2 or from Round 2 to Round 3

12
D. Erickson, D.K. Holderness Jr., K.J. Olsen et al. Accounting, Organizations and Society xxx (xxxx) xxx

References Cognition & Emotion, 13(5), 505e521.


Kerr, N. L., Messe , L. A., Seok, D.-H., Sambolec, E. J., Lount, R. B., Jr., & Park, E. S.
(2007). Psychological mechanisms underlying the Ko €hler motivation gain.
Asay, H. S., Guggenmos, R., Kadous, K., Koonce, L., & Libby, R. (2019). Theory testing
Personality and Social Psychology Bulletin, 33(6), 828e841.
and process evidence in accounting experiments. Working Paper. Indiana Uni-
Kluger, A. N., & DeNisi, A. (1996). The effects of feedback interventions on perfor-
versity, Cornell University, Emory University, University of Texas at Austin, and
mance: A historical review, a meta-analysis, and a preliminary feedback
Cornell University.
intervention theory. Psychological Bulletin, 119(2), 254e284. https://doi.org/
Baker, N. (2010). Employee feedback technologies in the human performance sys-
10.1037/0033-2909.119.2.254
tem. Human Resource Development International, 13(4), 477e485.
LeaderSharp Group. (2015, November 15, 2015). Feedback conversations: Dealing
Belschak, F. D., & Den Hartog, D. N. (2009). Consequences of positive and negative
with emotional reactions. Retrieved from https://www.leadersharp.com/
feedback: The impact on emotions and extra-role behaviors. Applied Psychology,
publications/feedback-conversations-dealing-with-emotional-reactions/.
58(2), 274e303. https://doi.org/10.1111/j.1464-0597.2008.00336.x
Lindebaum, D., & Fielden, S. (2011). ‘It's good to be angry’: Enacting anger in con-
Bentley, J. W. (2018). Challenges with Amazon mechanical Turk research in accounting.
struction project management to achieve perceived leader effectiveness. Hu-
Working paper. Amherst: University of Massachusetts.
man Relations, 64(3), 437e458.
Blaine, B., & Crocker, J. (1993). Self-esteem and self-serving biases in reactions to
Loftus, S., & Tanlu, L. J. (2018). Because of “because”: Examining the use of causal
positive and negative events: An integrative review Self-esteem (pp. 55e85).
language in relative performance feedback. The Accounting Review, 93(2),
Springer.
277e297.
Bobko, P. (1986). A solution to some dilemmas when testing hypotheses about
Luckett, P. F., & Eggleton, I. R. (1991). Feedback and management accounting: A
ordinal interactions. Journal of Applied Psychology, 71(2), 323e326.
review of research into behavioural consequences. Accounting, Organizations
Bonner, S. E. (2008). Judgment and decision making in accounting. Upper Saddle
and Society, 16(4), 371e394.
River, NJ: Pearson/Prentice Hall.
Manjoo, F. (2014, January 12, 2014). High definition: The 'gamification' of the office
Buchheit, S., Dalton, D., Downen, T., & Pippin, S. (2012). Outcome feedback, in-
approaches. The Wall Street Journal. Retrieved from https://www.wsj.com/
centives, and performance: Evidence from a relatively complex forecasting task.
articles/high-definition-the-8216gamification8217-of-the-office-approaches-
Behavioral Research in Accounting, 24(2), 1e20.
1389558998.
Buckless, F. A., & Ravenscroft, S. P. (1990). Contrast coding: A refinement of ANOVA
Murnighan, J. K., & Roth, A. E. (1983). Expecting continued play in prisoner's
in behavioral analysis. The Accounting Review, 933e945.
dilemma games: A test of several models. Journal of Conflict Resolution, 27(2),
Cappelli, P., & Travis, A. (2016). The performance management revolution. Harvard
279e300.
Business Review, 58e67. October 2016.
Podsakoff, P. M., & Farh, J.-L. (1989). Effects of feedback sign and credibility on goal
Chow, C. W. (1983). The effects of job standard tightness and compensation scheme
setting and task performance. Organizational Behavior and Human Decision
on performance: An exploration of linkages. The Accounting Review, 58(4),
Processes, 44(1), 45e67.
667e685.
Richardson, L. (2013). Feedback with feeling. Retrieved from https://www.
Clark, M. S., Pataki, S. P., & Carver, V. H. (1996). Some thoughts and findings on self-
richardson.com/blog/feedback-with-feeling/.
presentation of emotions in relationships. In G. J. O. Fletcher, & J. Fitness (Eds.),
Rosnow, R. L., & Rosenthal, R. (1995). Some things you learn aren't so: Cohen's
Knowledge structures in close relationships: A social psychological approach. New
paradox, Asch's paradigm, and the interpretation of interaction. Psychological
York, NY: Psychology Press.
Science, 6(1), 3e9.
Finkelstein, S. R., & Fishbach, A. (2011). Tell me what I did wrong: Experts seek and
Salovey, P. (1992). Mood-induced self-focused attention. Journal of Personality and
respond to negative feedback. Journal of Consumer Research, 39(1), 22e38.
Social Psychology, 62(4), 699e707.
Gaddis, B., Connelly, S., & Mumford, M. D. (2004). Failure feedback as an affective
Sedikides, C. (1992). Mood as a determinant of attentional focus. Cognition &
event: Influences of leader affect on subordinate attitudes and performance. The
Emotion, 6(2), 129e148. https://doi.org/10.1080/02699939208411063
Leadership Quarterly, 15(5), 663e686.
Shellenbarger, S. (2012, August 15, 2012). When the boss is a screamer. The Wall
Gellman, L., & Baer, J. (2016, May 26, 2016). Goldman Sachs to stop rating employees
Street Journal. Retrieved from https://www.wsj.com/articles/
with numbers. The Wall Street Journal. Retrieved from https://www.wsj.com/
SB10000872396390444772404577589302193682244.
articles/goldman-sachs-dumps-employee-ranking-system-1464272443.
Silverman, R. E. (2011, September 6, 2011). Yearly reviews? Try weekly. The Wall
Goetz, T. (2011, June 20, 2011). How facebook uses feedback loops: Meet rypple.
Street Journal. Retrieved from https://www.wsj.com/articles/
Wired.
SB10001424053111903895904576542962030419874.
Grant, H. (2013, January 28, 2013). Sometimes negative feedback is best. Harvard
Silverman, R. E. (2016, June 8, 2016). GE Re-engineers performance reviews, pay
Business Review.
practices. The Wall Street Journal. Retrieved from https://www.wsj.com/articles/
Green, J. D., Sedikides, C., Saltzberg, J. A., Wood, J. V., & Forzano, L. A. B. (2003).
ge-re-engineers-performance-reviews-pay-practices-1465358463.
Happy mood decreases self-focused attention. British Journal of Social Psychol-
Smith, C. A., Haynes, K. N., Lazarus, R. S., & Pope, L. K. (1993). In search of the "hot"
ogy, 42(1), 147e157.
cognitions: Attributions, appraisals, and their relation to emotion. Journal of
Guggenmos, R. D., Piercey, M. D., & Agoglia, C. P. (2018). Custom contrast testing:
Personality and Social Psychology, 65(5), 916.
Current trends and a new approach. The Accounting Review, 93(5), 223e244.
Tafkov, I. D. (2013). Private and public relative performance information under
Hannan, R. L., Krishnan, R., & Newman, A. H. (2008). The effects of disseminating
different compensation contracts. The Accounting Review, 88(1), 327e350.
relative performance feedback in tournament and individual performance
Taylor, M. S., Fisher, C. D., & Ilgen, D. R. (1984). Individuals' reactions to performance
compensation plans. The Accounting Review, 83(4), 893e913.
feedback in organizations: A control theory perspective. Research in Personnel
Hannan, R. L., McPhee, G. P., Newman, A. H., & Tafkov, I. D. (2013). The effect of
and Human Resources Management, 2(8), 1e124.
relative performance information on performance and effort allocation in a
Thornock, T. A. (2016). How the timing of performance feedback impacts individual
multi-task environment. The Accounting Review, 88(2), 553e575.
performance. Accounting, Organizations and Society, 55, 1e11.
Hatfield, E., Cacioppo, J. T., & Rapson, R. L. (1993). Emotional contagion. Current
Van Kleef, G. A., Homan, A. C., Beersma, B., Van Knippenberg, D., Van
Directions in Psychological Science, 2(3), 96e100.
Knippenberg, B., & Damen, F. (2009). Searing sentiment or cold calculation? The
Holderness, D. K., Olsen, K. J., & Thornock, T. A. (2017). Who are you to tell me that?!
effects of leader emotional displays on team performance depend on follower
the moderating effect of performance feedback source and psychological
epistemic motivation. Academy of Management Journal, 52(3), 562e580.
entitlement on individual performance. Journal of Management Accounting
Vancouver, J. B., & Tischner, E. C. (2004). The effect of feedback sign on task per-
Research, 29(2), 33e46.
formance depends on self-concept discrepancies. Journal of Applied Psychology,
Jensen, J. (2020). How to manage emotions when giving and receiving feedback.
89(6), 1092e1098.
Retrieved December 2, 2020, from https://www.think2perform.com/about-us/
Wegner, D. M., & Giuliano, T. (1980). Arousal-induced attention to self. Journal of
blog/how-manage-emotions-when-giving-and-receiving-feedback.
Personality and Social Psychology, 38(5), 719e726.
Johnson, G., & Connelly, S. (2014). Negative emotions in informal feedback: The
Wegner, D. M., & Giuliano, T. (1983). On sending artifact in search of artifact: Reply
benefits of disappointment and drawbacks of anger. Human Relations, 67(10),
to McDonald, Harris, and Maher. Journal of Personality and Social Psychology,
1265e1290. https://doi.org/10.1177/0018726714532856
44(2), 290e293.
Kelly, K. O., Webb, R. A., & Vance, T. (2014). The interactive effects of ex post goal
Wood, J. V., Saltzberg, J. A., & Goldsamt, L. A. (1990). Does affect induce self-focused
adjustment and goal difficulty on performance. Journal of Management Ac-
attention? Journal of Personality and Social Psychology, 58(5), 899.
counting Research, 27(1), 1e25.
Keltner, D., & Haidt, J. (1999). Social functions of emotions at four levels of analysis.

13

You might also like