Vous êtes sur la page 1sur 22

THE ACCOUNTING REVIEW American Accounting Association

Vol. 93, No. 2 DOI: 10.2308/accr-51830


March 2018
pp. 277–297

Because of ‘‘Because’’: Examining the Use of Causal


Language in Relative Performance Feedback
Serena Loftus
Tulane University

Lloyd J. Tanlu
Northeastern University
ABSTRACT: This study examines how the use of causal language in conveying relative performance feedback
impacts subsequent task performance. Research in linguistics has shown that causal language, defined as language
reflecting the search for reasons (commonly expressed through words such as ‘‘because’’ and ‘‘thus’’), impacts how
recipients process received information. We use a laboratory experiment to show that causal language has a
differential effect when used in negative versus positive feedback. In the case where initial relative performance is
low, the high use of causal language in the resulting negative performance feedback leads to a greater improvement
in subsequent performance, compared to low use of causal language. Conversely, when initial relative performance
is high, greater use of causal language in delivering positive feedback results in a smaller improvement in
performance. Our results indicate that employees’ cognitive processes and reactions to performance feedback are
influenced by the language used in explanations.
Keywords: performance evaluation; performance feedback; causal language.

I. INTRODUCTION

T
his study examines how the use of causal language in conveying relative performance feedback impacts subsequent
task performance. Performance feedback has been defined in the literature as information given to a person regarding
the quantity and quality of his or her past performance (Prue and Fairbank 1981). The current literature in accounting
on performance feedback has primarily focused on developing quantitative measures of performance and studying how
employee behavior is influenced by those measures (Hemmer 1996; Ittner, Larcker, and Rajan 1997; Banker, Potter, and
Schroeder 1993). However, little research has focused on the role of qualitative information in performance feedback, although
the provision of qualitative performance feedback is a key component of an organization’s management control system
(Luckett and Eggleton 1991), and managers often devote resources to this issue in practice (Sprinkle 2003; Baiman and Rajan
1995). We argue that the way that managers communicate with employees in performance evaluations has a direct impact on
employee performance. Specifically, we argue that employees react to the language used in explanations, even when the
explanation itself is held constant.
In this study, we investigate how the use of causal language, or language reflecting the search for reasons (Pennebaker and
Francis 1996), influences employees’ reactions to the relative performance information contained in performance evaluations.
Studies in linguistics suggest that the language through which the feedback is communicated can also impact how the
feedback is received. We examine whether the differential use of causal language in performance feedback impacts

We appreciate the helpful comments of Eddy Cardinaels (editor) and two anonymous referees. We thank the following individuals for their thoughtful
comments and suggestions: Jasmijn Bol, Bob Bowen, Asher Curtis, Kristina Demek, Sukari Farrington, Robert Grasser, Katlijn Haesebrouck, Lynn
Hannan, Frank Hodge, Kris Hoang, Patrick Hurley, Kim Ikuta, Jane Jollineau, Mikaela Kursell, Dawn Matsumoto, Jim McGinley, Sarah McVay, Jennifer
Nichol, Emily Rosenzweig, Stephen Rowe, D. Shores, Bill Tayler, Kristy Towry, Sara Toynbee, workshop participants at Northeastern University, Tulane
University, University of Washington, and Wake Forest University, and participants at the 2014 Global Management Accounting Research Symposium,
the 2014 AAA Western Region Meeting, and the 2014 AAA Annual Meeting.
Earlier versions of this paper were circulated under the title ‘‘Because of ‘Because’: The Role of Causal Language in Performance Feedback and Its Impact
on Affect and Subsequent Task Performance.’’
Editor’s note: Accepted by Eddy Cardinaels, under the Senior Editorship of Mark L. DeFond.
Submitted: August 2014
Accepted: June 2017
Published Online: June 2017
277
278 Loftus and Tanlu

performance. Causal language is defined as language reflecting the search for reasons. Common causal words include
connectors such as ‘‘because,’’ ‘‘thus,’’ and ‘‘since,’’ as well as verbs such as ‘‘result’’ and ‘‘infer.’’ To identify causal words that
relate to the search for reasons, we rely on the Linguistic Inquiry and Word Count (LIWC) program, a computational linguistic
tool developed by linguistic scholars James W. Pennebaker, Roger J. Booth, and Martha E. Francis.
We examine managers’ use of causal language for the following reasons: (1) causal language has a strong relationship with
causal reasoning, and causal reasoning is relevant in a performance evaluation process where employees are trying to
understand and address feedback; and (2) managers can vary their use of causal language without changing the explanation for
the performance feedback. Overall, causal language is important to study not only because of its potential ability to impact the
affective and cognitive responses of employees, but because managers can adapt their use of causal language to improve
feedback efficacy.
Based on the extant research that examines causal language, empirical evidence supports the notion that the use of causal
words facilitates understanding and links cause and effect.1 We, thus, predict that greater use of causal language facilitates the
cognitive processing of performance feedback. We also posit that the impact of causal language on subsequent task
performance depends on whether the feedback was positive or negative, as literature on performance feedback suggests that
positive feedback has a different effect than negative feedback on overall cognitive processing and behavioral responses to
feedback (Belschak and Den Hartog 2009; Fishbach, Eyal, and Finkelstein 2010).
Our study uses a 2 3 2 between-subjects experimental design and a relative performance setting, in which we vary the
extent of causal language used in the performance feedback (high versus low causal language) and the valence of the
performance feedback (positive versus negative performance feedback). Our experimental methodology allows us to hold the
explanation contained in the performance feedback constant between the experimental treatment groups; thus, we can attribute
any changes in subsequent task performance to the use of causal language in the performance feedback.2
We hypothesize that when initial performance is poor and the resulting performance feedback is negative, high use of
causal language improves future performance more than low use of causal language. This effect occurs because employees who
receive negative performance reviews that contain a higher amount of causal language are more likely to address the feedback,
as they are better able to understand the rationale behind it. Therefore, those employees use the feedback to guide their future
behavior, which results in improved performance (Ilgen and Davis 2000). This prediction is consistent with the archival studies
on causal language that associate causal language with improved cognitive processing and better outcomes (Pennebaker and
Francis 1996; Pennebaker 1997a, 1997b; Pennebaker and Graybeal 2001).
The literature on causal language offers little insight regarding the effect when feedback is positive (Anderson 1965;
Skowronski and Carlston 1989). Given that a high level of causal language in negative feedback enhances future performance,
it may be tempting to expect this result for positive feedback. However, an alternative conjecture is that improved cognitive
processing of positive feedback leads to less of an improvement in future performance. One potential reason is that higher self-
attention, which prior research has shown to reduce the automaticity of skilled execution (Baumeister, Hutton, and Cairns
1990), potentially results in lower future performance than would occur absent improved processing. In other words, increasing
self-awareness by more effectively pointing out the reasons why the employee is doing well may cause the employee to disrupt
automatic processes and behaviors contributing to his or her success. Given the lack of definitive theory, we make no
directional prediction regarding the effect of the use of causal language in positive feedback on subsequent performance.
We test our theory in a relative performance setting for two reasons. First, the use of relative performance feedback is a
common and significant component of performance evaluation systems within organizations. Tice (2015) finds that relative
performance evaluation use in the public firms in her sample increased from 22 percent in 2006 to 37 percent in 2012, and
recent research has suggested that the use of relative performance feedback leads to increased effort and improved subsequent
performance (Tafkov 2013; Hannan, McPhee, Newman, and Tafkov 2013). Second, compared to other performance evaluation
systems, the use of relative performance feedback generates more uncertainty in performance outcomes for employees, as they
do not have complete information about the distribution of performance outcomes and, therefore, cannot readily engage in self-
comparison. Since our theory develops from the assumption that performance feedback is uncertain, we choose to test our
predictions in a setting where feedback uncertainty is high.
Our results show that participants’ scores improved across all conditions after relative performance feedback was provided.
However, for participants that received negative performance feedback, we find that the improvement in subsequent

1
For example, a student who is told by a professor ‘‘You received a C on the essay because you did not sufficiently vary your sentence structure’’ is less
likely to reject the professor’s explanation for the grade than if the student was told ‘‘You received a C on the essay. You did not sufficiently vary your
sentence structure.’’ In the first example, the use of the word ‘‘because’’ allows the student to clearly connect the feedback on the essay with the
explanation given by the professor, whereas in the second, the student must connect the explanation with the feedback on his or her own.
2
Note that we also hold monetary compensation constant between treatment groups to ensure that any results are due to differences in feedback and not
due to differences in pay.

The Accounting Review


Volume 93, Number 2, 2018
Because of ‘‘Because’’: Examining the Use of Causal Language in Relative Performance Feedback 279

performance is significantly greater for participants receiving feedback with more causal language, which is consistent with our
prediction. Conversely, for participants who received positive feedback, the improvement in subsequent performance is
significantly lower for those who received more causal language. This surprising result is consistent with the view that self-
attention stemming from improved processing of positive valence feedback adversely affects performance.
Our findings contribute to the academic literature in several ways. First, we extend the academic literature on performance
evaluation and feedback efficacy. In their review of performance feedback research in the management accounting literature,
Luckett and Eggleton (1991) call for more research into the behavioral consequences of feedback. Particularly, they note that
behavioral responses differ based on the sign of the feedback (i.e., positive versus negative), and they indicate that there is a
need for more accounting studies that ‘‘investigate questions such as the behavioral impact of different modes of reporting and
communicating negative performance results’’ (Luckett and Eggleton 1991, 391). More recently, Choi, Hecht, Tafkov, and
Towry (2016) find that employees do behave differently when they observe peer evaluations that are negative versus those that
are positive. Our paper directly addresses this issue and shows that causal language improves future performance when
performance feedback is negative, a finding that is particularly notable given how often prior research has shown that negative
feedback does not have the desired effect of improving performance (Ilgen and Davis 2000; Kluger and DeNisi 1996).
Second, prior research in accounting has focused on determinants of performance evaluation and managers’ biases in
performance evaluation and feedback (Bol 2011; Bol and Smith 2011; Lipe and Salterio 2000; Libby, Salterio, and Webb 2004;
Cardinaels and Van Veen-Dirks 2010). While previous studies have investigated managers’ cognitive responses during the
performance evaluation process (Feldman 1981; Knowlton and Mitchell 1980; Brown 1985), our study offers insights into
employees’ cognitive and affective responses to performance feedback—particularly to language in performance feedback. By
documenting the impact of language on employees’ cognitive responses to feedback, we demonstrate how managers can
improve feedback efficacy.
Third, our results extend the literature on causal linkages in strategic performance measurement systems (e.g., Banker,
Chang, and Pizzini 2004; Webb 2004; Luft 2004) by suggesting that the language used to communicate causal linkages may
impact employees’ reactions to the explanation provided. Prior studies (Farrell, Kadous, and Towry 2012) show that
communicating qualitative causal linkages can improve employee effort and firm performance, and we extend this finding by
documenting that the language used in qualitative explanations impacts employee behavior.
Finally, our study also has implications for practitioners. We show that managers can use high causal language when
providing negative feedback to poorly performing employees to reduce the negative affect experienced by the employee and
improve subsequent performance. On the other hand, the use of causal language may work less well when relaying positive
feedback to strong performers. Causal language is a particularly attractive tool for managers to use in performance reviews
because it offers a relatively costless alternative to the method currently purported by the literature—changing the explanation
for the performance feedback (for example, Ray 2007; Chen and Chiu 2013; Hansen 2010). While current research proposes
that managers can modify the explanation contained in performance feedback to facilitate cognitive processing, managers are
often reluctant to do so because they believe that the explanation can be used to fulfill other, often conflicting, objectives
(Hansen 2013; Longenecker, Sims, and Gioia 1987). By documenting that causal language is an alternative tool, we provide
managers with a method of managing employees’ reactions to feedback without varying the explanation for the feedback.
The remainder of this paper is organized as follows. Section II reviews related literature and develops our hypotheses.
Sections III and IV present the design and results of our experiment. Section V concludes.

II. RELATED LITERATURE AND HYPOTHESIS DEVELOPMENT

Performance Feedback
A large body of literature in both management accounting and organizational economics focuses on the use of performance
feedback as a way of motivating effort from employees (Kunz and Pfaff 2002; Ittner and Larcker 1998; Otley and Fakiolas
2000; Vagneur and Peiperl 2000; Hopwood 1972). Since the early 1980s, there has been a continual drive to find and refine
performance measurement systems (Kaplan 1983, 1984; Eccles 1991). Most recently, the body of research has focused on
designing the optimal contract—i.e., the identification and selection of the appropriate performance measures used to evaluate
and incentivize employee performance. In this paper, we focus on how the information that emerges from these performance
measurement and incentive systems is delivered to employees. Specifically, we focus on the role of language in performance
feedback.

Employees’ Cognitive Responses to Performance Feedback


Performance feedback conveyed to employees is uncertain and usually contains new information. While employees may
be able to anticipate some aspects of the performance feedback, they are unlikely to be able to perfectly predict all of the

The Accounting Review


Volume 93, Number 2, 2018
280 Loftus and Tanlu

information in the feedback—especially when relative performance feedback is used. Employees evaluated on their relative (as
opposed to absolute) performance are unlikely to have complete information about the distribution of performance outcomes,
especially as companies are likely to keep relative performance information private (Hannan et al. 2013; Tafkov 2013).
In response to the uncertainty of the information contained in the performance feedback, employees engage in automatic
and unconscious cognitive and emotional processing that helps them understand the feedback (Wilson, Centerbar, Kermer, and
Gilbert 2005). Individuals can use a variety of tools to make sense of unexpected events: (1) they may search for a reason for
the event; (2) they may alter knowledge structures to accommodate the event; and/or (3) they may assimilate the event into their
existing knowledge structures. Once employees have successfully made sense of their performance feedback, they cognitively
adapt to the event, which then allows them to respond to the performance feedback. Although people engage in sense-making
of simple data normally and automatically, research has shown that autonomous information processing is often insufficient to
transform more complex data into simple representations (Chater and Loewenstein 2016).

Causal Language and Facilitated Sense-Making


Because employees must engage in sense-making to adapt to new information, organizational behavior scholars have
focused on tools that managers can use to facilitate the performance feedback process, including providing explanations in the
feedback. A large literature in organizational behavior suggests that the information within performance feedback facilitates (or,
in some cases, obfuscates) employees’ sense-making and reduces employees’ emotional reactions to the performance
evaluations (e.g., Smith and Ellsworth 1985; Liden and Mitchell 1985; Kluger and DeNisi 1996; Gaddis, Connelly, and
Mumford 2004). However, while managers may wish to vary their explanations to impact employee processing, they may also
have other (potentially conflicting) objectives that affect their willingness to do so.3 Accordingly, managers can often do little
to alter the content of the explanations given in the feedback. For instance, if a sales employee did not meet his or her quota,
then the feedback will include this information, as well as reasons why the employee performed poorly—the manager
providing the feedback can do little to vary the explanations for the poor performance.
We argue that in addition to content of the feedback, the language used to convey the information can also affect sense-
making. Specifically, we propose that high use of causal language in performance evaluations facilitates sense-making by
clearly linking the feedback to the explanation through causal terms such as ‘‘since,’’ ‘‘resulting,’’ or ‘‘because.’’ In our
experimental setting, we are able to hold the content or explanation contained in the performance feedback constant while
varying the degree of causal language employed in the performance evaluation.
Because causal language facilitates sense-making, its use makes employees more likely to accept performance feedback.
Acceptance is employees’ belief that the feedback is an accurate portrayal of their performance.4 When employees receive
performance feedback, acceptance of the feedback determines whether they are likely to incorporate the feedback into their
future behavior (Ilgen, Fisher, and Taylor 1979). A wealth of other research suggests that when employees understand the cause
of the feedback, they are more likely to accept performance feedback (Ilgen et al. 1979; Ilgen and Davis 2000), less likely to
deny the information contained in the performance evaluation (see Kluger and DeNisi [1996] for a review), and more likely to
address the performance feedback (Ilgen et al. 1979; Ilgen and Davis 2000).
Research has suggested that there are differential effects of addressing negative and positive feedback on future
performance (Ilies, de Pater, and Judge 1986). Therefore, the impact of addressing performance feedback on future
performance likely depends on the employee’s initial performance and, consequently, on the valence of performance feedback
given. When performance feedback is negative, addressing performance feedback is likely to improve subsequent performance,
provided that the feedback contains information that is pertinent and relevant to future performance (Ilgen and Davis 2000).
This effect occurs because if the employee is not performing well, then the employee’s current approach to the task is unlikely
to be effective. Assuming that the feedback is accurate and constructive, behavioral changes based on the explanation given in
the performance feedback are likely to improve future performance for those employees. If the greater use of causal language in
performance feedback facilitates sense-making and consequently results in employees changing their behavioral strategies, then
the change is likely to result in greater improved future performance. Conversely, the improvement in future performance is not
as great in the absence of causal language in feedback, holding the explanation constant. Therefore, we make the following
prediction:

3
They may, for example, wish to provide all employees with the same explanations in performance feedback, despite any processing consequences, to
avoid potential future wrongful termination lawsuits.
4
Positive performance feedback is more readily accepted than negative performance feedback (M. Jacobs, A. Jacobs, Feldman, and Cavior 1973),
although the association of the feedback with self-image may be a more important determinant of feedback acceptance than valence (Kennedy and
Willcutt 1964; Korman 1970, 1976). Sense-making may also impact employees’ self-efficacy or beliefs that they hold about their ability to perform
particular tasks (Bandura 1982; Bandura and Jourden 1991).

The Accounting Review


Volume 93, Number 2, 2018
Because of ‘‘Because’’: Examining the Use of Causal Language in Relative Performance Feedback 281

H1: When initial performance is poor and feedback is negative, high use of causal language in conveying negative
feedback leads to a greater improvement in subsequent performance relative to low use of causal language.
While our theory predicts that causal language improves sense-making, many of the studies examining the link between
sense-making and future performance have been conducted in negative performance feedback contexts (Kluger and DeNisi
1996). Given the lack of empirical evidence, it is less clear how improved sense-making of positive performance feedback
impacts future performance. One possibility is that addressing positive valence feedback improves future performance, through
the same cognitive processing route that enhanced processing of negative valence feedback improves future performance.
However, whereas employees that initially performed poorly are more able to address the feedback in the presence of more
causal language and ‘‘fix what is broken’’ to improve future performance, it is less clear that employees that initially performed
well will have much to ‘‘fix.’’ Furthermore, positive feedback may be considered less surprising given that most individuals
exhibit a cognitive bias where they overestimate their own abilities relative to others (Hoorens 1993). Thus, sense-making
facilitated by the high use of causal language is likely to have a smaller (or no) effect on future performance.
Further, some empirical evidence suggests that improved cognitive processing of positive feedback leads to higher self-
attention. Self-attention tends to reduce the automaticity of skilled execution (Baumeister et al. 1990; Baumeister 1984),
resulting in lower future performance than would occur absent improved processing. For example, Baumeister et al. (1990) find
that participants that received praise after playing several rounds of a video game showed a deterioration in performance
compared to a control group. The authors suggest that any positive feedback generates greater self-consciousness, which
interferes with skilled performance. Given the lack of definitive theory, we are unable to make a directional prediction
regarding the effect of causal language on subsequent performance when performance feedback is positive. We, therefore, state
the following exploratory hypothesis in null form:
H2: When initial performance is good and feedback is positive, high use of causal language in conveying positive
feedback leads to no differential improvement in subsequent performance relative to low use of causal language.
Note that in order for the performance feedback to be relevant and pertinent to the recipient, our predictions assume that the
feedback valence is linked to actual performance—i.e., negative feedback is provided when performance is poor, and positive
feedback is provided when performance is high. We do not consider the scenario in which positive feedback is provided when
performance is poor, and vice versa.5 Thus, we make predictions regarding the effect of causal feedback only within feedback
valence conditions.
We graphically depict and summarize our research question in Appendix A.

III. RESEARCH DESIGN

Participants, Materials, and Design


To test our predictions, we use a 2 3 2 between-subjects design that varies the following independent variables in the
feedback manipulation: use of causal language (high versus low) and feedback valence (positive versus negative). Feedback
valance is assigned to participants based on their performance on the experimental task prior to receiving the causal language
manipulation.
Participants were 108 undergraduate business students from a large, public state university who performed the
experimental task in a laboratory setting.6,7 The laboratory setting ensured that the task was completed continuously and that
participants were not exposed to any other language during the experimental task. Participants were paid a flat fee of $15 for
their participation after completing the study.8 Participants were recruited from introductory financial and managerial
accounting courses. They were 19.6 years old, on average, and had 1.9 years of work experience. Because our experimental
task does not require prior accounting knowledge or any specialized skills, we believe that these participants are an appropriate
subject pool for testing predictions about the impact of language in performance evaluations.

5
In practice, the use of multiple performance measures means performance is not as easily categorized in a binary fashion—employees may score highly
on some performance measures and poorly on others. While we expect our theory to generalize in a mixed-valence feedback setting, our current design
allows us to observe the relation between feedback valence and performance much more distinctly.
6
The experiment in this study was approved by the Institutional Review Board (IRB) for Human Participants at the university where the study was
administered and completed.
7
At the end of the study, as a manipulation check, we asked participants whether their first-round score was above or below the median (based on the
feedback received). Out of an initial sample of 111 participants, three participants provided the wrong response. Thus, we exclude these three
participants to arrive at our final sample of 108 participants.
8
Holding pay constant allows us to test the effects of causal language absent any incentive effects. Furthermore, prior literature has documented that
relative performance feedback is often provided by companies even when it is not linked to compensation (Hannan et al. 2013).

The Accounting Review


Volume 93, Number 2, 2018
282 Loftus and Tanlu

We ask participants to make as many words as they can consisting of at least three letters from a set of 12 letters within 60
seconds. All participants in all experimental conditions receive the same letters in each round to ensure that the difficulty of
each round’s task is held constant across conditions. We first provide participants with an opportunity to practice the task. Then,
in the first round, we ask participants to complete three turns of this task, where each turn includes a different set of 12 letters
(see Appendix B for an experimental timeline). We score each participant’s performance on the first round of the experiment
using an instantaneous algorithm. The scoring algorithm first performs a validity check to ensure that each word created is at
least three letters long and appears in a dictionary. Next, we assign points based on each letter used in each valid word created.
To assign points, we use the Scrabble system, in which letters that appear frequently in the English language hold lower point
values than letters that appear infrequently.
We disclose the following to participants: (1) their total score for the first round (consisting of the three turns), (2)
information on whether their score lies above or below the median score for the first round, and (3) four determinants of their
total score. In the second round of the experiment, we ask participants to complete another three turns, each again consisting of
different sets of 12 letters. Ultimately, we compare each participant’s first-round score (i.e., prior to receiving the feedback) to
their second-round score.
Each participant’s total score is only based on the four determinants disclosed to participants (the number of words derived,
the length of the words derived, the letters that appear in each word, and the complexity of the letter combinations appearing in
each word). Because participants are informed about how their score will be determined, and their scores and performance
valence reflect their actual performance, our experiment does not involve deception.

Causal Language Manipulation


Our goal is to manipulate the causal language contained in participants’ performance feedback while holding the
explanation of the performance feedback constant. We include causal words in the High_Causal condition that are absent from
Low_Causal condition, substitute non-causal language in the Low_Causal condition with causal language in the High_Causal
condition, and repeat causal language extraneously in the High_Causal condition. Because causal language extends across
multiple categories of words, our causal words encompasses multiple types of language, including nouns (‘‘result,’’ ‘‘reason’’),
an adverb (‘‘why’’), a verb (‘‘made’’), and a conjunction (‘‘because’’), which ensures that our findings result from causal
language and not the type of word used (e.g., verb, conjunction, noun, etc.). We are able to use synonyms for several words,
and those identified in the LIWC dictionary are expected to have a more causal interpretation than the corresponding word that
is not categorized by the LIWC dictionary as causal. This approach allows us to use a broad range of causal language for our
manipulation.9 The disadvantages of this approach are that (1) we cannot isolate which type of causal language impacts
performance (e.g., nouns, verbs, conjunctions, etc.), and (2) we are unable to disentangle the various connotations associated
with different words that are identified as causal by the LIWC.10 Appendix C offers a comparison of the High_Causal and
Low_Causal conditions.

Initial Performance and Valence of Relative Performance Feedback


The valence of performance feedback depends on the participant’s relative performance on the first round of the
experiment. Participants who score above (below) an established median for the first round of the experiment are deemed high
(low) performers and, thus, receive positive (negative) feedback.11 We use relative performance feedback to ensure that the
feedback contains some information that is novel to participants. While participants may have prior beliefs about their ability on
the experimental task, they cannot perfectly predict how their own performance will compare with the performance of other
participants because they are unaware of the distribution of scores on the task.

9
Performance feedback shown to participants in the high (low) Causal Language condition contains 26.25 percent (4.25 percent) causal language.
Furthermore, we are aware that the additional words in the High_Causal condition very slightly reduce the readability of the feedback. However, if
participants were deterred from reading the performance feedback based on readability scores, then we would not observe a differential reaction to the
High_Causal manipulation from participants who scored both above and below the median score in Round 1.
10
For instance, the word ‘‘result’’ is classified as causal by the LIWC, while ‘‘score’’ is not. We use both words in the same context in providing feedback,
and assume that the word ‘‘result’’ helps facilitate sense-making more than the word ‘‘score’’ does. However, we acknowledge the limitation that we are
not able to rule out whether these words have other connotations not related to sense-making, nor if participants who performed well (and received
positive feedback) interpret the causal words differently from those who did not.
11
To initially establish the median, we used the scores from a pilot test sample consisting of 20 participants. We then adjusted the median (which was then
used for all subsequent participants) following the first group of participants. The median for the pilot participants was 156 and the median score for the
first group of participants was 152, with scores ranging from 21–654. Note that only two participants fell within a score of 152 and 156 after we had
adjusted the median, and removing these participants from analysis does not impact results.

The Accounting Review


Volume 93, Number 2, 2018
Because of ‘‘Because’’: Examining the Use of Causal Language in Relative Performance Feedback 283

Capturing and providing feedback based on participants’ actual performance (as opposed to randomly assigning
participants to positive or negative feedback conditions) helps ensure that participants cognitively process feedback naturally
(e.g., believe, internalize, and potentially act upon their performance feedback as they would in the ‘‘real world’’).12 We do not
want to trigger any cognitive processes that are not associated with our manipulation. For instance, cognitive processes that
result from questioning the accuracy of the performance feedback could confound our conclusions about causal sense-making.
A large stream of research suggests that individuals form judgments about the accuracy of their feedback and that these
perceptions influence their acceptance of feedback (Ammons 1956; Nease, Mudgett, and Quinones 1999; Ilgen et al. 1979).
Participants who perceive feedback to be inaccurate may be less likely to accept feedback, and may even reject the performance
feedback (Ilgen et al. 1979; Ammons 1956). Participants’ acceptance or rejection of feedback could impact their exposure to
the causal language manipulation that is our primary research focus, because participants either read and disregard the feedback
(which contains the causal language manipulation), or do not read the feedback at all.13 Moreover, since our goal was to
determine how participants would react to feedback, we provided feedback that could be attributed to their past performance
(rather than chance, which is beyond participants’ control), which they can work on and potentially improve or change.
However, one trade-off of this design choice is that we risk introducing a potential confound into our experiment. Our
conjectures on the effects of feedback may be confounded by other determinants of performance (such as ability) to the extent
that these are correlated with the internalization and interpretation of feedback (i.e., high performers may perceive feedback
differently from low performers). This concern is mitigated by the fact that we examine the impact of causal language within
feedback valence conditions. For instance, all participants who receive positive feedback valence have the same ability, so
ability cannot explain the observed impact of causal language within the positive valence feedback condition.
Nevertheless, we empirically control for participant-specific factors to address the aforementioned concerns. We discuss
our tests and findings in Section IV.

Dependent Measures
We measure participants’ change in scores after they receive the experimental treatment between the two rounds.14 We
choose to use a single-period design to control for differences in expectations that may form over time in a multi-period model.
In addition, individuals may engage in considerable task learning in a multi-period design. A single-period measure of
subsequent task performance will attenuate the confounding impact of learning relative to a multi-period measure of
performance in this setting.

IV. RESULTS
Table 1 presents descriptive statistics of our results. Our experimental task resulted in median scores of 153.5 on Round 1
and 210.5 on Round 2. More than 80 percent of participants improved their scores on the second round of the experiment,
resulting in an overall mean score improvement of 51.6 (Table 1, Panel A). The average scores by condition are presented in
Table 1, Panel B.
Our first hypothesis predicts that when initial relative performance is poor and consequent feedback valence is negative,
high causal language leads to higher subsequent performance than low causal language. As indicated in Table 2, Panel A, for
participants who performed below the median score in the first round and consequently received negative feedback, the average
Change in Score is higher for participants in the High_Causal condition (47.35) than for those in the Low_Causal condition
(23.13). The difference between these means is consistent with our directional prediction at the 10 percent one-tailed
significance level (t-statistic ¼ 1.64, p ¼ 0.053, one-tailed). These results support our first hypothesis by revealing that high
causal language leads to more improved subsequent performance than low causal language when feedback is negative.
When initial performance is high, however, we make no directional prediction on how causal language in the resulting
positive feedback affects subsequent performance. Our results provide evidence to reject our null hypothesis for H2. As
previously mentioned, the Change in Score for participants who performed strongly in the first round are both positive;
however, participants who received feedback with high causal language show a lower average score improvement (45.69; see

12
This design choice mimics many real-world settings where high performers receive positive feedback, while subpar performers receive negative
feedback. Because our research question concerns the cognitive processing of feedback, we want to ensure that participants’ cognitive responses to
feedback in the lab authentically reflect employees’ cognitive responses to feedback in non-laboratory settings.
13
We ask participants of our study whether they believed the feedback they received was accurate, fair, and objective, and on a seven-point scale. All of
the response means are statistically greater than the median response of 4 at the 1 percent significance level (two-tailed, df ¼ 106, untabulated). This
suggests that participants were accepting of the feedback. Furthermore, the scores for all questions did not vary across experimental conditions.
14
Additionally, we also examine the percentage difference in scores (rather than differences in raw scores) as the dependent variable in all our tests. The
untabulated results are qualitatively similar and our conclusions are identical.

The Accounting Review


Volume 93, Number 2, 2018
284 Loftus and Tanlu

TABLE 1
Descriptive Statistics

Panel A: Aggregate Performance Statistics by Round


First Third
n Mean Std. Dev. Quartile Median Quartile
Round 1 Score 108 180.9 107.2 109 153.5 236
Round 2 Score 108 232.5 134.2 126 210.5 316
Change in Score 108 51.6 61.8 7 42.5 90

Panel B: Descriptive Performance Statistics—By Experimental Condition


Average Scores Round 1 Average Scores Round 2
First Round Performance First Round Performance
Below Above Row Below Above Row
Feedback Median Median Means Median Median Means
Low Causal Language 103 254 173 127 349 230
(36.2) (94.7) (102.6) (68.2) (126.8) (149.2)
n ¼ 30 n ¼ 26 n ¼ 56 n ¼ 30 n ¼ 26 n ¼ 56
High Causal Language 107 271 189 154 317 236
(36.3) (102.6) (112.5) (68.7) (98.0) (117.2)
n ¼ 26 n ¼ 26 n ¼ 52 n ¼ 26 n ¼ 26 n ¼ 52
Column Means 105 263 139 333
(35.9) (98.1) (69.3) (113.4)
n ¼ 56 n ¼ 52 n ¼ 56 n ¼ 52
Means, standard deviations (in parentheses), and number of participants are provided for each dependent variable indicated in Panel B.

Table 2, Panel A) than for those who received low causal language (94.58). The difference is statistically significant (t-statistic
¼ 3.01, p , 0.01, two-tailed). In other words, the lower use of causal language in conveying positive feedback leads to higher
subsequent performance than the high use of causal language. These results are consistent with the explanation that facilitated
cognitive processing increases self-attention and reduces automaticity (Baumeister et al. 1990; Baumeister 1984). When self-
attention is heightened, participants tend to become more self-conscious. This disrupts the automatic processes that allowed
them to perform well in the initial round, consequently reducing the improvement in subsequent performance in the second
round. We cannot, however, conclusively rule out alternative explanations for this unanticipated finding, and leave this to
future research.
To complement our within-conditions findings, Table 2, Panel B presents the results of an ANOVA. We find evidence of a
significant interaction of feedback valence (operationalized by the First Round Performance [FRP] indicator variable, which
takes on a value of 1 if the initial performance is below the median and performance feedback is negative, and 0 otherwise) and
causal language (operationalized by the Causal Language [CL] indicator variable, which takes on a value of 1 for feedback
with high causal language, and 0 otherwise) on Change in Score (Table 2, Panel B, F-statistic ¼ 11.14, p , 0.01, two-tailed).
These findings are consistent with our univariate test results in Table 2, Panel A, and confirm that the effect of causal language
varies based on feedback valence.15
In the previous analysis, we use the median score in the first round to distinguish low performers from high performers.
However, the median split is a crude measure to measure initial performance. Participants scoring just below the median are not

15
Social comparison theory would suggest decreased subsequent performance for participants who receive positive valence feedback and increased
subsequent performance for participants who receive negative valence feedback (Tafkov 2013; Suls and Wheeler 2001). Under social comparison
theory, individuals wish to perform above the average, and individuals with below-average performance levels are subject to upward social comparison,
which serves as a motivating factor, despite invoking negative self-feelings. However, for individuals who are performing above average, downward
comparisons have a demotivating effect on future performance, although they invoke more positive self-feelings. While some of our results are
consistent with social comparison theory (particularly for participants whose initial performance was poor), social comparison theory would not explain
the interactive impact of causal language and feedback valence on subsequent performance.

The Accounting Review


Volume 93, Number 2, 2018
Because of ‘‘Because’’: Examining the Use of Causal Language in Relative Performance Feedback 285

TABLE 2
Tests of Hypotheses
Changes in Performance
Low versus High Initial Performance
Panel A: Descriptive Statistics
Dependent Variable: Change in Score
(Round 2 Score  Round 1 Score)
First Round Performance
Feedback Below Median Above Median Row Means
Low Causal Language 23.13 94.58 56.30
(50.57) (61.52) (66.04)
n ¼ 30 n ¼ 26 n ¼ 56
High Causal Language 47.35 45.69 46.52
(59.82) (55.52) (57.14)
n ¼ 26 n ¼ 26 n ¼ 52
Column Means 34.38 70.13
(55.89) (63.05)
n ¼ 56 n ¼ 52
High Causal  Low Causal 24.21þ 48.88***
t-statistic 1.64 3.01
df 54 50

Panel B: ANOVA
Dependent Variable: Change in Score
(Round 2 Score  Round 1 Score)
Source df MS F-Stat p-value
Causal Language (CL) 1 4,093.0 1.27 0.26
First Round Performance (FRP) 1 32,750.6 10.16 0.00***
CL 3 FRP 1 35,928.7 11.14 0.00***
Residual 104 3,223.9
þ þþ þþþ
, , Denote one-tailed statistical significance at 10 percent, 5 percent, and 1 percent, respectively.
*, **, *** Denote two-tailed statistical significance at 10 percent, 5 percent, and 1 percent, respectively.
Means, standard deviations (in parentheses), and number of participants are provided for each dependent variable indicated in Panels A and B.

necessarily poor performers, nor are those who score just above the median superior performers. Thus, the assignment of
positive or negative feedback valence to participants whose scores are around the median introduces noise in our previous
analysis. To address this, we more finely distinguish low and high performers by splitting the sample into terciles, using the
scores from the first round of the study. Participants who scored in the lower (upper) tercile are considered low (high)
performers, thus receiving negative (positive) feedback prior to the second round of the experiment. The middle tercile includes
participants who scored close to the median and, thus, may have been misidentified by the original analysis where the median
score was used to distinguish low and high performers.
We present our results in Table 3. Panel A presents the univariate analysis of means of the differences in scores for low,
medium, and high performers in each of the low and high causal language conditions. Consistent with H1, for low performers
(i.e., participants whose scores in the first round are in the lowest tercile) who received negative feedback, the improvement in
performance is higher for those in the High_Causal condition (difference between groups ¼ 53.63, t-value ¼ 3.29, one-tailed p
, 0.01). Similar to our results from the previous analysis, the improvement in scores for high performers (i.e., participants
whose scores in the first round are in the upper tercile) is higher for participants in the Low_Causal condition (difference
between groups ¼54.63, t-value ¼ 3.18, two-tailed p , 0.01). The difference in scores between the Low_Causal and High_
Causal conditions are not statistically significant for participants whose scores in the first round are closest to the median score

The Accounting Review


Volume 93, Number 2, 2018
286 Loftus and Tanlu

TABLE 3
Tests of Hypotheses
Changes in Performance
Low versus Middle versus High Initial Performance
Panel A: Descriptive Statistics
Dependent Variable: Change in Score
(Round 2 Score  Round 1 Score)
First Round Performance
Feedback Bottom Middle Top Row Means
Low Causal Language 15.84 59.61 93.63 56.30
(37.73) (78.76) (53.23) (66.04)
n ¼ 19 n ¼ 18 n ¼ 19 n ¼ 56
High Causal Language 69.47 31.53 39.00 46.52
(58.91) (57.40) (51.17) (57.14)
n ¼ 17 n ¼ 17 n ¼ 18 n ¼ 52
Column Means 41.17 45.97 67.05
(55.28) (69.68) (58.48)
n ¼ 36 n ¼ 35 n ¼ 37
High Causal  Low Causal 53.63þþþ 28.08 54.63***
t-statistic 3.29 1.20 3.18
df 34 33 35

Panel B: ANOVA
Dependent Variable: Change in Score
(Round 2 Score  Round 1 Score)
Source df MS F-Stat p-value
Causal Language (CL) 1 2,532.5 0.77 0.38
First Round Performance (FRP) 2 6,073.7 1.85 16.24þ
CL 3 FRP 2 28,797.8 8.77 0.00***
Residual 102 3,282.8
þ þþ þþþ
, , Denote one-tailed statistical significance at 10 percent, 5 percent, and 1 percent, respectively.
*, **, *** Denote two-tailed statistical significance at 10 percent, 5 percent, and 1 percent, respectively.
Means, standard deviations (in parentheses), and number of participants are provided for each dependent variable indicated in Panels A and B.

for the round (see also Figures 1 and 2).16 The ANOVA presented in Table 3, Panel B corroborates the findings from Panel A
and Table 2. The interaction between CL and FRP is statistically significant (F-statistic ¼ 8.77, two-tailed p , 0.01). These
results lend further support that the use of causal language has a differential, asymmetric effect when feedback is negative
versus when it is positive.

Controlling for Participant-Specific Factors


As discussed earlier, our research design may be susceptible to the potential confound of participant-specific factors (e.g.,
ability), such that high performers may perceive and incorporate feedback differently from low performers. In order to empirically
address this concern, we employ a multiple linear regression model to validate our main results. We regress participant scores in
Round 2 on their scores in Round 1, along with our variables of interest, Causal Language and Below the Median.17 By including

16
Note that in this method, half of the participants in the middle tercile received positive feedback and the other half received negative feedback. We are
aware that in this group, we are unable to determine the effect of feedback on performance; however, we are much more concerned with determining
the effect of feedback on the lower and upper terciles. We replicate the analysis using quartiles, and results are similar to our tercile analysis—we find
significant results in the same directions for the lowest and highest quartiles, but no statistically significant result for the two middle quartiles.
17
We also use the difference in scores as an alternative dependent variable, and our inferences are identical to the analysis presented here.

The Accounting Review


Volume 93, Number 2, 2018
Because of ‘‘Because’’: Examining the Use of Causal Language in Relative Performance Feedback 287

FIGURE 1
Graphs Depicting Difference in Scores
Low versus High Performers

participant scores in Round 1 as a regressor, we hope to control for any variable unique to the participants (e.g., ability), as
individual and idiosyncratic determinants of past performance are likely to impact future performance, as well. Table 4 presents
the multiple regression analysis results. In all four specifications, the coefficient on Round 1 Score is statistically significant and
close to 1.00, denoting that subsequent scores are largely dependent on participants’ initial scores. In Column (1), the coefficient
on the interaction between Causal Language and Below the Median is positive and statistically significant (coefficient ¼ 73.12,
one-tailed p , 0.01), indicating that the high use of causal language improves performance when initial performance is poor and

FIGURE 2
Graphs Depicting Difference in Scores
Low versus Middle versus High Performers

The Accounting Review


Volume 93, Number 2, 2018
288 Loftus and Tanlu

TABLE 4
Tests of Hypotheses
Multiple Regression Analysis
Dependent Variable: Round 2 Score
(1) (2) (3) (4)
Causal Language (CL) 48.91*** 65.89*** 46.94*** 61.38***
(3.08) (3.13) (2.99) (2.92)
Below the Median (BM) 71.24*** 34.34** 69.51*** 33.82**
(3.72) (2.17) (3.69) (2.15)
CL 3 BM 73.12þþþ 71.90þþþ
(3.32) (3.32)
Round 1 Score (R1) 1.00*** 1.23*** 0.93*** 1.15***
(13.11) (13.22) (11.65) (11.74)
CL 3 R1 0.43*** 0.39***
(4.24) (3.92)
Ability0 7.39 6.50
(1.53) (1.37)
Frequency0 6.72 5.60
(1.49) (1.26)
Enjoyment0 5.14 3.79
(0.79) (0.60)
Change in Effort1 10.32** 9.23**
(2.21) (2.01)
Intercept 94.24*** 35.52 110.69*** 56.33
(4.21) (1.51) (3.03) (1.55)
Observations 108 108 108 108
Adj. R2 0.82 0.83 0.83 0.83
þ þþ þþþ
, , Denote one-tailed statistical significance at 10 percent, 5 percent, and 1 percent, respectively.
*, **, *** Denote two-tailed statistical significance at 10 percent, 5 percent, and 1 percent, respectively.
Standard errors are shown in parentheses.

Variable Definitions:
Causal Language (CL) ¼ 1 if a participant was in the high causal language condition, and 0 otherwise;
Below the Median (BM) ¼ 1 if a participant was in the bottom half of scores in the first round and, thus, received negative feedback, and 0 otherwise;
Ability0 ¼ the answer to the question ‘‘Rate your ability at word unscrambling games,’’ measured at the beginning of the experiment using a seven-point
Likert scale, with higher values denoting better higher self-reported ability;
Frequency0 ¼ the answer to the question ‘‘How often do you play word unscrambling games,’’ with higher values denoting higher self-reported frequency
of playing;
Enjoyment0 ¼ the answer to the question ‘‘How much do you like playing word unscrambling games,’’ measured at the beginning of the experiment using a
seven-point Likert scale, with higher values denoting higher self-reported enjoyment; and
Change in Effort1 ¼ the answer to the question ‘‘Compared to the effort you expended on Part 1 of the study, how much effort do you intend to expend on
Part 2 of the study?’’ measured after feedback is provided after Round 1, using a seven-point Likert scale with higher values denoting greater self-
reported effort.

consequent feedback is negative. Conversely, the coefficient on Causal Language is negative and statistically significant
(coefficient ¼48.91, two-tailed p , 0.01), indicating that for participants that performed well (Below the Median ¼ 0), the high
use of causal language has a negative effect on subsequent performance, ceteris paribus. Column (2) presents the multiple
regression results that include the interaction of Causal Language and Round 1 Score as an independent variable. The coefficient
on this interaction is negative and significant (coefficient ¼ 0.43, two-tailed p , 0.01), suggesting that Causal Language has a
more negative effect as participants’ Round 1 Scores increase.
To further control for participant-specific factors, we include as independent variables participants’ self-reported scores on
measures of perceived ability at word-scramble tasks, frequency in which they engage in similar tasks, perceived enjoyment in
performing these tasks (all three of which are measured at the start of the experiment),18 and the change in effort that

18
Using the self-reported ability and enjoyment measures taken at the other points in the experiment does not change our results or inferences.

The Accounting Review


Volume 93, Number 2, 2018
Because of ‘‘Because’’: Examining the Use of Causal Language in Relative Performance Feedback 289

participants were expecting to exert for the second round of the experiment immediately after receiving the feedback.19 Results
presented in Columns (3) and (4) of Table 4 are similar to those in Columns (1) and (2). Note that among all the participant-
specific self-reported variables, only the coefficients on change in expected effort are statistically significant (two-tailed p ,
0.05 in both specifications indicated in Columns (3) and (4) of Table 4). The coefficient on expected change in effort is
negative, indicating that the subsequent Round 2 scores become worse as expected effort increases. This surprising finding is
once more consistent with self-attention (particularly for high performers), such that an increase in effort may decrease the
automaticity of already good performance.20
Our instrument also allows us to address several additional alternative explanations for our findings. First, we monitor
changes in participants’ beliefs about self-efficacy to ensure that those are not driving our key findings. We ask participants to
assess their ability at similar tasks at three points in the experiment: before the experiment begins, after receiving the
experimental manipulation, and at the end of the experiment. We present the raw scores in Table 5, Panel A. We analyze
changes scores over the course of the experiment, and find that feedback valence impacts changes in perceptions of ability
between the beginning of the experiment and the end of the first round, as well as between the beginning of the experiment and
end of experiment. Participants had lower beliefs about self-efficacy after receiving negative feedback (t ¼ 6.37, df ¼ 55, two-
tailed p , 0.05, untabulated), while those who received positive feedback had higher self-efficacy scores (t ¼ 1.85, df ¼ 51,
two-tailed p , 0.05, untabulated). The impact of feedback valence on beliefs about self-efficacy is consistent with the prior
literature (Stajkovic and Luthans 1998; Baron 1988; Brown and Dutton 1995), providing validation of our measure. However,
results also reassuringly indicate that there are no statistically significant differences between High_Causal and Low_Causal
participants that received the same feedback valence at any given point in the experiment. Thus, our results are unlikely driven
by participants’ beliefs about self-efficacy.
Second, we address the explanation that reactions to both the causal language and feedback valence are driven by
participant engagement in the task by asking participants to assess their enjoyment of similar tasks at the same points that we
take the ability measure above. We report the self-reported enjoyment scores in Table 5, Panel B. We again find statistically
significant differences in enjoyment (t ¼ 2.56, df ¼ 106, two-tailed p , 0.01, untabulated) between participants that initially
performed poorly (and received negative feedback) and those that did well (and received positive feedback). However, the
causal language manipulation within each feedback valence condition does not impact any of the changes in task enjoyment
throughout the experiment (t ¼ 0.33, df ¼ 54, two-tailed p ¼ 0.75, untabulated, for negative feedback condition; t ¼ 0.00, df ¼
50, two-tailed p ¼ 0.50, untabulated, for positive feedback condition).
Finally, we measure whether participants perceive performance feedback containing causal language as objective rather
than subjective and that results are driven by reactions to perceptions of objectivity rather than the causal language. We ask
participants to assess their perceptions of the objectivity (and subjectivity) of the feedback after receiving the language
manipulation and find no significant differences across conditions.
All in all, our conclusions from the results presented in Tables 2 and 3 remain unchanged, even after controlling for
participant-specific factors.

Supplemental Analyses
Process Measures of Sense-Making: Emotional Adaptation
Emotional adaptation is one form through which sense-making and cognitive processing can take place. The AREA
(attend, respond, explain, adapt) model of emotional processing suggests that individuals experience intense emotional
responses when processing unanticipated information.21 Individuals attend or orient to novel, relevant events; they emotionally
respond to the events; they explain or engage in sense-making to understand the event; they emotionally adapt to the event so
that they think about the emotional event less and experience less intense emotions when they do (Wilson et al. 2005).
Research has shown that causal language facilitates emotional adaptation, consequently reducing the emotional intensity of an
event. Individuals are better able to address the emotional event once transformed from a surprising, unexpected event into one
that is easier to understand and rationalize (Wilson, Wheatley, Kurtz, Dunn, and Gilbert 2004; Wilson and Gilbert 2005).

19
There is a statistically significant difference in ex ante perceived ability (t ¼ 3.68, df ¼ 106, two-tailed p , 0.01, untabulated) between participants that
scored above (5.65 out of a score of 7) and below (4.60 out of 7) the median, as reported in Table 5, Panel A. This difference persists throughout the study.
However, there is no difference in perceived ability between participants in the High_Causal and Low_Causal conditions measured at various points
throughout the study. For the other participant-specific self-reported variables, there are no statistically significant differences across all conditions.
20
Note, however, that we do not find a difference in reported change in effort between high and low performers.
21
For instance, Mellers, Schwartz, and Ritov (1999) find that people have more extreme emotional reactions to gambling wins and losses when the
outcome had a low probability of occurring. Additionally, a second stream of research shows that neural reward pathways, such as dopaminergic
neurons, are activated by the predictability of a stimulus (Berns, McClure, Pagnoni, and Montague 2001). Unanticipated outcomes also induce
physiological arousal (Le Poire and Burgoon 1996), which is likely to intensify emotional responses to events (Schachter and Singer 1962).

The Accounting Review


Volume 93, Number 2, 2018
290 Loftus and Tanlu

TABLE 5
Self-Reported Ability and Task Enjoyment Scores

Panel A: Ability
Pre-Round 1 After Experimental Treatment After Round 2
First Round Performance First Round Performance First Round Performance
Below Above Row Below Above Row Below Above Row
Feedback Median Median Means Median Median Means Median Median Means
Low Causal Language 4.40 5.61 4.96 3.27 5.85 4.46 3.43 6.00 4.63
(1.43) (1.50) (1.57) (1.39) (1.41) (1.90) (1.59) (1.36) (1.96)
n ¼ 30 n ¼ 26 n ¼ 56 n ¼ 30 n ¼ 26 n ¼ 56 n ¼ 30 n ¼ 26 n ¼ 56
High Causal Language 4.85 5.69 5.27 3.19 5.96 4.58 3.81 5.88 4.85
(1.64) (1.35) (1.55) (1.52) (1.37) (2.00) (1.70) (1.42) (1.87)
n ¼ 26 n ¼ 26 n ¼ 52 n ¼ 26 n ¼ 26 n ¼ 52 n ¼ 26 n ¼ 26 n ¼ 52
Column Means 4.60 5.65 3.23 5.90 3.61 5.94
(1.53) (1.41) (1.44) (1.38) (1.64) (1.38)
n ¼ 56 n ¼ 52 n ¼ 56 n ¼ 52 n ¼ 56 n ¼ 52

Panel B: Enjoyment
Pre-Round 1 After Experimental Treatment After Round 2
First Round Performance First Round Performance First Round Performance
Below Above Row Below Above Row Below Above Row
Feedback Median Median Means Median Median Means Median Median Means
Low Causal Language 3.87 4.77 4.28 3.33 4.62 3.92 3.30 4.73 3.96
(1.17) (0.95) (1.16) (1.35) (0.98) (1.35) (1.42) (1.00) (1.43)
n ¼ 30 n ¼ 26 n ¼ 56 n ¼ 30 n ¼ 26 n ¼ 56 n ¼ 30 n ¼ 26 n ¼ 56
High Causal Language 4.08 5.15 4.62 3.50 5.15 4.32 3.62 5.12 4.37
(1.02) (1.32) (1.29) (1.17) (1.46) (1.56) (1.13) (1.37) (1.46)
n ¼ 26 n ¼ 26 n ¼ 52 n ¼ 26 n ¼ 26 n ¼ 52 n ¼ 26 n ¼ 26 n ¼ 52
Column Means 3.96 4.96 3.41 4.88 3.45 4.92
(1.09) (1.15) (1.26) (1.26) (1.29) (1.20)
n ¼ 56 n ¼ 52 n ¼ 56 n ¼ 52 n ¼ 56 n ¼ 52
Ability is the answer to the question ‘‘Rate your ability at word unscrambling games,’’ measured using a seven-point Likert scale, with higher values
denoting better higher self-reported ability. Task Enjoyment is the answer to the question ‘‘How much do you like playing word unscrambling games,’’
measured using a seven-point Likert scale, with higher values denoting higher self-reported enjoyment. Both measures are taken before Round 1, at the
end of Round 1 (after the experimental manipulation), and at the end of Round 2. Means, standard deviations (in parentheses), and number of participants
are provided for each variable indicated in Panels A and B.

We use the Positive and Negative Affect Schedule (PANAS) (Watson, Clark, and Tellegen 1988) to measure participants’
emotional states at various points in the study.22 While measuring affect is an intrinsically difficult process, the PANAS scale is
an appropriate measure of affect during performance reviews because emotions included on the scale are likely to be
experienced by employees during the performance review.23 We operationalize higher scores to signify more positive affect.
We measure participants’ affect before the task starts, again after the first round (immediately after participants receive
feedback), and finally after the second round (i.e., at the end of the experiment), and report the results graphically in Figure 3.24

22
The PANAS scale asks individuals to rate the extent to which they feel 20 different emotions (such as anger, hostility, anxiety, and pride) using a nine-
point Likert scale with endpoints ranging from ‘‘Very Slightly or Not at All’’ to ‘‘Extremely.’’ The PANAS scale then categorizes ten emotions with
positive affect and ten emotions with negative affect and derives aggregate measures of both positive and negative affect.
23
In untabulated analysis, we examine whether changes in affect are driven primarily by any one out of the 20 measures of affect, and find that this is not
the case.
24
The literature is ambiguous regarding the timing of the effect of causal feedback on emotional adaptation. Hence, we measure affect scores at various
times in the study. If causal language in feedback results in immediate emotional adaptation, then the measurement taken after Round 1 will already
reflect emotional adaptation. If, on the other hand, it takes longer, then perhaps the effects of emotional adaptation will be more evident in the
measurement taken at the end of the experiment.

The Accounting Review


Volume 93, Number 2, 2018
Because of ‘‘Because’’: Examining the Use of Causal Language in Relative Performance Feedback 291

FIGURE 3
Graph Depicting Affect by Condition throughout the Experiment

The affect scores of participants do not differ by condition at the start of the experiment, but differ immediately after
participants receive feedback. Net affect was significantly lower for participants who received negative feedback than positive
feedback (means of 8.28 versus 19.87 for negative versus positive feedback, t ¼ 4.76, df ¼ 106, two-tailed p , 0.01,
untabulated).
For participants who received negative feedback, those in the High_Causal condition reported significantly higher net
affect than those in the Low_Causal condition (means of 11.22 versus 5.53, t ¼ 1.87, df ¼ 54, two-tailed p ¼ 0.07,
untabulated). Further, net affect remained higher for participants who received negative feedback in the High_Causal
condition after completing the experiment (mean of 15.50 versus 7.60 for Low_Causal, t ¼ 2.50, df ¼ 54, two-tailed p ¼ 0.02,
untabulated). We do not find any evidence that causal language impacts net affect for participants that received positive
feedback.
We then examine whether affect mediates the relation between causal language and future performance. Using a two-group
structural equations model, we find that net affect partially mediates the relation between causal language and future
performance for negative feedback, but does not appear to mediate the relation for positive feedback (Figure 4). For this test,
we omit participants in the middle tercile of initial performance, as the link between performance and feedback is noisiest for
these participants. Untabulated results show that inferences do not change if these participants are included, although statistical
significance is weaker.
Taken together, our results suggest that for negative feedback valence, emotional adaptation is a process that links, but
does not fully explain, the relationship between causal language and future performance. We find no evidence of emotional
adaptation for recipients of positive feedback.
One reason we may find an effect of causal language on net affect for negative, but not positive, performance feedback
valence is due to the well-documented positive-negative asymmetry effect (Anderson 1965; Skowronski and Carlston 1989),
which suggests that negative feedback elicits stronger affective responses than positive feedback. The positive-negative
asymmetry arises because it is generally more costly for individuals to ignore negative events than positive events (e.g.,

The Accounting Review


Volume 93, Number 2, 2018
292 Loftus and Tanlu

FIGURE 4
Mediating Role of Affect

Panel A: Mediation Analysis for Low Performers Receiving Negative Feedback

Panel B: Mediation Analysis for High Performers Receiving Positive Feedback

þ þþ þþþ
, , Denote one-tailed statistical significance at 10 percent, 5 percent, and 1 percent, respectively.
*, **, *** Denote two-tailed statistical significance at 10 percent, 5 percent, and 1 percent, respectively.
This figure summarizes the structural equation modeling results for the mediating role of affect, conditional on prior performance. Affect is
measured as the difference in PANAS scores taken after the manipulation and those taken at the beginning of the study. Low (high)
performers include participants whose scores in the first round appear in the bottom (upper) tercile. Unstandardized estimates and standard
errors (in parentheses) are provided for each effect. Solid lines denote statistically significant effect. Dotted lines denote non-statistically
significant effects. Model fit statistics for the exploratory model are as follows: CFI ¼ 1.00; SRMR , 0.01; v2/df ¼ 22.39/6 (p ¼ 0.001),
indicating the model is good fit.

negative events at work may be more predictive of continued employment than positive events). Thus, negative events require
more cognitive processing than positive events (which results in participants spending more time thinking about the negative
feedback than the positive feedback), and negative events make stronger impressions on individuals than positive events
(Baumeister, Bratslavsky, Finkenauer, and Vohs 2001).

Robustness: Analyzing the Explanations Received by Participants


As previously discussed, the feedback provided to each participant provides four reasons why his or her performance is
above (or below) the median: his or her (1) number of words, (2) average word length, (3) number of letters used, and (4)
complexity of words are higher (or lower) than those of other participants. Our experimental manipulation provides all
participants with the same four determinants of their score, but it is not necessarily the case that all four determinants apply to
each participant. Since this design choice is constant across all four experimental conditions, it cannot explain our findings, but
it could add noise to our results if it hampers participants’ sense-making processes. To address the fact that some participants
may have scored above (below) the median on one of the four determinants, but were given negative (positive) Feedback
Valence, we exclude certain participants who score above (below) the median on any of the four determinants, but earn
negative (positive) feedback overall. We then replicate our tests in Table 2, and our results and inferences are consistent with
our earlier analysis.

The Accounting Review


Volume 93, Number 2, 2018
Because of ‘‘Because’’: Examining the Use of Causal Language in Relative Performance Feedback 293

Robustness: Learning
To provide evidence that our central finding is not due to learning, we run our tests on only participants’ performance on
the set of letters immediately preceding and following our manipulation. Recall that each round of our experiment included
three sets of 12 letters that participants were asked to use to form words. If our results were not driven by the information
contained in the experimental manipulation, then participants’ performance on the last set of letters in the first round should be
roughly equivalent to participants’ performance on the first set of letters in the second round. Our inferences do not change
when using these data.

V. CONCLUSION
In this study, we present theory and experimental evidence consistent with the idea that the quantity of causal language
contained in relative performance feedback impacts both future performance and employees’ emotional reactions to the
performance evaluation. Theory suggests that causal language facilitates the sense-making process that employees use when
they attempt to interpret the information contained in performance evaluations. We predict and find that when initial
performance is poor and feedback is negative, the presence of causal language in performance feedback leads to a greater
improvement in performance. Further, when performance feedback is positive, we find that the high use of causal language
(relative to low use of causal language) results in a lower improvement in subsequent task performance. In supplemental
analyses, we find that the use of causal language improves affect when performance feedback is negative, but has no significant
effect when performance feedback is positive.
Our study makes several contributions. First, our study contributes to the literature on the determinants of performance
evaluation efficacy by documenting that the causal language used in relative performance feedback (holding compensation and
all other content in the feedback constant) can impact subsequent task performance. Second, our study contributes to a growing
literature in financial accounting that documents that managers’ language choices in qualitative disclosures impact the
judgments and decisions of the users of those disclosures (Tafkov 2013). We examine a previously unexamined language
choice—causal language—and document its role in facilitating sense-making, as well as demonstrate that the effect of causal
language on performance improvement is dependent on whether initial performance is good or bad. Finally, our results provide
managers with the potential costs and benefits of their use of causal language in performance feedback. Specifically, managers
should use more causal language in communicating negative feedback when initial performance is poor; however, managers
should avoid the use of causal language in communicating positive feedback when initial performance is strong.
Our study is not without limitations. First, in order to make the feedback believable, true, and objective, our participants are
not randomly sorted into the feedback valence conditions; rather, the feedback valence they receive is based on their initial
performance on the experimental task. This may introduce the potential confound that poorer performers may perceive and
interpret feedback differently from stronger performers. Although we try to control for this potential confound, we are unable to
completely rule out this alternative explanation. Second, the task in this setting is one that requires skill in terms of word
unscrambling. We are not certain that the results generalize to tasks that are simple and menial that do not require skill, or to
tasks requiring high degrees of creativity, as we do not have definitive theory to inform our predictions in these cases. Third, we
attempt to hold feedback content constant while varying the quantity of causal language contained in the performance feedback.
However, it is possible that the placement of the causal language or the various connotations elicited by some of the causal
words from participants, rather than the ratio of causal words to non-causal words in the text, may also inadvertently influence
our results. Finally, we examine the effect of causal language on single-period performance, which may differ in a multi-period
setting. We do not account for the effects of learning and experience, which may influence how employees interpret and react to
feedback (Berger, Klassen, Libby, and Webb 2013). We leave the task of exploring these additional questions to future
research. Despite these limitations, our theory and results are likely to be useful to managers who are seeking to improve the
efficacy of their performance evaluation practices.

REFERENCES
Ammons, R. B. 1956. Effects of knowledge of performance: A survey and tentative theoretical formulation. Journal of General
Psychology 54 (2): 279–299. doi:10.1080/00221309.1956.9920284
Anderson, N. H. 1965. Averaging versus adding as a stimulus-combination rule in impression formation. Journal of Experimental
Psychology 70 (4): 394–400. doi:10.1037/h0022280
Baiman, S., and M. V. Rajan. 1995. The informational advantages of discretionary bonus schemes. The Accounting Review 70: 557–579.
Bandura, A. 1982. Self-efficacy mechanism in human agency. American Psychologist 37 (2): 122–147. doi:10.1037/0003-066X.37.2.122

The Accounting Review


Volume 93, Number 2, 2018
294 Loftus and Tanlu

Bandura, A., and F. J. Jourden. 1991. Self-regulatory mechanisms governing the impact of social comparison on complex decision
making. Journal of Personality and Social Psychology 60 (6): 941–951. doi:10.1037/0022-3514.60.6.941
Banker, R. D., G. Potter, and R. G. Schroeder. 1993. Reporting manufacturing performance measures to workers: An empirical study.
Journal of Management Accounting Research 5: 33–55.
Banker, R. D., H. Chang, and M. J. Pizzini. 2004. The balanced scorecard: Judgmental effects of performance measures linked to strategy.
The Accounting Review 79 (1): 1–23. doi:10.2308/accr.2004.79.1.1
Baron, R. A. 1988. Negative effects of destructive criticism: Impact on conflict, self-efficacy, and task performance. Journal of Applied
Psychology 73 (2): 199–207. doi:10.1037/0021-9010.73.2.199
Baumeister, R. F. 1984. Choking under pressure: Self-consciousness and the paradoxical effects of incentives on skillful performance.
Journal of Personality and Social Psychology 46 (3): 610–620. doi:10.1037/0022-3514.46.3.610
Baumeister, R. F., D. G. Hutton, and K. J. Cairns. 1990. Negative effects of praise on skilled performance. Basic and Applied Social
Psychology 11 (2): 131–148. doi:10.1207/s15324834basp1102_2
Baumeister, R. F., E. Bratslavsky, C. Finkenauer, and K. D. Vohs. 2001. Bad is stronger than good. Review of General Psychology 5 (4):
323–370. doi:10.1037/1089-2680.5.4.323
Belschak, F. D., and D. N. Den Hartog. 2009. Consequences of positive and negative feedback: The impact on emotions and extra-role
behaviors. Applied Psychology 58 (2): 274–303. doi:10.1111/j.1464-0597.2008.00336.x
Berger, L., K. J. Klassen, T. Libby, and A. Webb. 2013. Complacency and giving up across repeated tournaments: Evidence from the
field. Journal of Management Accounting Research 25 (1): 143–167. doi:10.2308/jmar-50435
Berns, G. S., S. M. McClure, G. Pagnoni, and P. R. Montague. 2001. Predictability modulates human brain response to reward. Journal of
Neuroscience 21: 2793–2798.
Bol, J. C. 2011. The determinants and performance effects of managers’ performance evaluation biases. The Accounting Review 86 (5):
1549–1575. doi:10.2308/accr-10099
Bol, J. C., and S. D. Smith. 2011. Spillover effects in subjective performance evaluation: Bias and the asymmetric influence of
controllability. The Accounting Review 86 (4): 1213–1230. doi:10.2308/accr-10038
Brown, C. 1985. Causal reasoning in performance assessment: The effects of cause and effect temporal order and covariation. Accounting,
Organizations and Society 10 (3): 255–266. doi:10.1016/0361-3682(85)90019-4
Brown, J. D., and K. A. Dutton. 1995. The thrill of victory, the complexity of defeat: Self-esteem and people’s emotional reactions to
success and failure. Journal of Personality and Social Psychology 68 (4): 712–722. doi:10.1037/0022-3514.68.4.712
Cardinaels, E., and P. Van Veen-Dirks. 2010. Financial versus non-financial information: The impact of information organization and
presentation in a balanced scorecard. Accounting, Organizations and Society 35 (6): 565–578. doi:10.1016/j.aos.2010.05.003
Chater, N., and G. Loewenstein. 2016. The under-appreciated drive for sense-making. Journal of Economic Behavior and Organization
126 (Part B): 137–154.
Chen, B. R., and Y. S. Chiu. 2013. Interim performance evaluation in contract design. Economic Journal 123 (569): 665–698. doi:10.
1111/ecoj.12018
Choi, J. W., G. Hecht, I. D. Tafkov, and K. L. Towry. 2016. Vicarious learning under implicit contracts. The Accounting Review 91 (4):
1087–1108. doi:10.2308/accr-51293
Eccles, R. G. 1991. The performance measurement manifesto. Harvard Business Review 69 (1): 131–137.
Farrell, A. M., K. Kadous, and K. L. Towry. 2012. Does the communication of causal linkages improve employee effort allocations and
firm performance? An experimental investigation. Journal of Management Accounting Research 24 (1): 77–102. doi:10.2308/jmar-
50149
Feldman, J. M. 1981. Beyond attribution theory: Cognitive processes in performance appraisal. Journal of Applied Psychology 66 (2):
127–148. doi:10.1037/0021-9010.66.2.127
Fishbach, A., T. Eyal, and S. R. Finkelstein. 2010. How positive and negative feedback motivate goal pursuit. Social and Personality
Psychology Compass 4 (8): 517–530. doi:10.1111/j.1751-9004.2010.00285.x
Gaddis, B., S. Connelly, and M. D. Mumford. 2004. Failure feedback as an affective event: Influences of leader affect on subordinate
attitudes and performance. Leadership Quarterly 15 (5): 663–686. doi:10.1016/j.leaqua.2004.05.011
Hannan, R., G. McPhee, A. Newman, and I. D. Tafkov. 2013. The effect of relative performance information on performance and effort
allocation in a multi-task environment. The Accounting Review 88 (2): 553–575. doi:10.2308/accr-50312
Hansen, S. E. 2010. The Benefits of Limited Feedback in Organizations. Working paper, Universitat Pompeu Fabra.
Hansen, S. E. 2013. Performance feedback with career concerns. Journal of Law, Economics, and Organization 29 (6): 1279–1316.
doi:10.1093/jleo/ews032
Hemmer, T. 1996. On the design and choice of ‘‘modern’’ management accounting measures. Journal of Management Accounting
Research 8: 87–116.
Hoorens, V. 1993. Self-enhancement and superiority biases in social comparison. European Review of Social Psychology 4 (1): 113–139.
doi:10.1080/14792779343000040
Hopwood, A. G. 1972. An empirical study of the role of accounting data in performance evaluation. Journal of Accounting Research 10:
156–182. doi:10.2307/2489870

The Accounting Review


Volume 93, Number 2, 2018
Because of ‘‘Because’’: Examining the Use of Causal Language in Relative Performance Feedback 295

Ilgen, D., and C. Davis. 2000. Bearing bad news: Reactions to negative performance feedback. Applied Psychology 49 (3): 550–565.
doi:10.1111/1464-0597.00031
Ilgen, D. R., C. D. Fisher, and M. S. Taylor. 1979. Consequences of individual feedback on behavior in organizations. Journal of Applied
Psychology 64 (4): 349–371. doi:10.1037/0021-9010.64.4.349
Ilies, R., I. E. de Pater, and T. A. Judge. 2007. Differential affective reactions to negative and positive feedback, and the role of self-
esteem. Journal of Managerial Psychology 22 (6): 590–609. doi:10.1108/02683940710778459
Ittner, C. D., and D. F. Larcker. 1998. Innovations in performance measurement: Trends and research implications. Journal of
Management Accounting 10: 205–238.
Ittner, C. D., D. F. Larcker, and M. V. Rajan. 1997. The choice of performance measures in annual bonus contracts. The Accounting
Review 72: 231–255.
Jacobs, M., A. Jacobs, G. Feldman, and N. Cavior. 1973. The ‘‘credibility gap’’: Delivery of positive and negative and emotional and
behavioral feedback in groups. Journal of Consulting and Clinical Psychology 4: 214–223.
Kaplan, R. S. 1983. Measuring manufacturing performance: A new challenge for managerial accounting research. The Accounting Review
58: 686–705.
Kaplan, R. S. 1984. The evolution of management accounting. The Accounting Review 59: 390–418.
Kennedy, W. A., and H. C. Willcutt. 1964. Praise and blame as incentives. Psychological Bulletin 62 (5): 323–332. doi:10.1037/
h0042917
Kluger, A. N., and A. DeNisi. 1996. Effects of feedback intervention on performance: A historical review, a meta-analysis, and a
preliminary feedback intervention theory. Psychological Bulletin 119 (2): 254–284. doi:10.1037/0033-2909.119.2.254
Knowlton, W. A., and T. R. Mitchell. 1980. Effects of causal attributions on a supervisor’s evaluation of subordinate performance.
Journal of Applied Psychology 65 (4): 459–466. doi:10.1037/0021-9010.65.4.459
Korman, A. K. 1970. Toward an hypothesis of work behavior. Journal of Applied Psychology 54 (1, Pt. 1): 31–41. doi:10.1037/h0028656
Korman, A. K. 1976. Hypothesis of work behavior revisited and an extension. Academy of Management Review 1 (1): 50–63. doi:10.
5465/AMR.1976.4408762
Kunz, A. H., and D. Pfaff. 2002. Agency theory, performance evaluation, and the hypothetical construct of intrinsic motivation.
Accounting, Organizations and Society 27 (3): 275–295. doi:10.1016/S0361-3682(01)00031-9
Le Poire, B. A., and J. K. Burgoon. 1996. Usefulness of differentiating arousal responses within communication theories: Orienting
response or defensive arousal within nonverbal theories of expectancy violation. Communication Monographs 63 (3): 208–230.
doi:10.1080/03637759609376390
Libby, T., S. E. Salterio, and A. Webb. 2004. The balanced scorecard: The effects of assurance and process accountability on managerial
judgment. The Accounting Review 79 (4): 1075–1094. doi:10.2308/accr.2004.79.4.1075
Liden, R. C., and T. R. Mitchell. 1985. Reactions to feedback: The role of attributions. Academy of Management Journal 28 (2): 291–308.
doi:10.2307/256202
Lipe, M. G., and S. E. Salterio. 2000. The balanced scorecard: Judgmental effects of common and unique performance measures. The
Accounting Review 75 (3): 283–298. doi:10.2308/accr.2000.75.3.283
Longenecker, C. O., H. P. Sims, and D. A. Gioia. 1987. Behind the mask: The politics of performance appraisal. Academy of Management
Executive 1 (3): 183–193. doi:10.5465/AME.1987.4275731
Luckett, P. F., and I. R. C. Eggleton. 1991. Feedback and management accounting: A review of research into behavioral consequences.
Accounting, Organizations and Society 16 (4): 371–394. doi:10.1016/0361-3682(91)90028-D
Luft, J. L. 2004. Discussion of ‘‘Managers’ commitment to the goals contained in a strategic performance measurement system.’’
Contemporary Accounting Research 21 (4): 959–964. doi:10.1111/j.1911-3846.2004.tb00076.x
Mellers, B., A. Schwartz, and I. Ritov. 1999. Emotion-based choice. Journal of Experimental Psychology. General 128 (3): 332–345.
doi:10.1037/0096-3445.128.3.332
Nease, A. A., B. O. Mudgett, and M. A. Quinones. 1999. Relationships among feedback sign, self-efficacy, and acceptance of
performance feedback. Journal of Applied Psychology 84 (5): 806–814. doi:10.1037/0021-9010.84.5.806
Otley, D., and A. Fakiolas. 2000. Reliance on accounting performance measures: Dead end or new beginning. Accounting, Organizations
and Society 25 (4/5): 497–510. doi:10.1016/S0361-3682(98)00007-5
Pennebaker, J. W. 1997a. Opening Up: The Healing Power of Expressing Emotions. New York, NY: Guilford Press.
Pennebaker, J. W. 1997b. Writing about emotional experiences as a therapeutic process. Psychological Science 8 (3): 162–166. doi:10.
1111/j.1467-9280.1997.tb00403.x
Pennebaker, J. W., and M. E. Francis. 1996. Cognitive, emotional, and language processes in disclosure. Cognition and Emotion 10 (6):
601–626. doi:10.1080/026999396380079
Pennebaker, J. W., and A. Graybeal. 2001. Patterns of natural language use: Disclosure, personality, and social integration. Current
Directions in Psychological Science 10 (3): 90–93. doi:10.1111/1467-8721.00123
Prue, D. M., and J. A. Fairbank. 1981. Performance feedback in organizational behavior management: A review. Journal of
Organizational Behavior Management 3 (1): 1–16. doi:10.1300/J075v03n01_01

The Accounting Review


Volume 93, Number 2, 2018
296 Loftus and Tanlu

Ray, K. 2007. The retention effect of withholding performance information. The Accounting Review 82 (2): 389–425. doi:10.2308/accr.
2007.82.2.389
Schachter, S., and J. Singer. 1962. Cognitive, social, and physiological determinants of emotional state. Psychological Review 69 (5):
379–399. doi:10.1037/h0046234
Skowronski, J. J., and D. E. Carlston. 1989. Negativity and extremity biases in impression formation: A review of explanations.
Psychological Bulletin 105 (1): 131–142. doi:10.1037/0033-2909.105.1.131
Smith, C. A., and P. C. Ellsworth. 1985. Patterns of cognitive appraisal in emotion. Journal of Personality and Social Psychology 48 (4):
813–838. doi:10.1037/0022-3514.48.4.813
Sprinkle, G. B. 2003. Perspectives on experimental research in managerial accounting. Accounting, Organizations and Society 28 (2/3):
287–318. doi:10.1016/S0361-3682(01)00058-7
Stajkovic, A. D., and F. Luthans. 1998. Self-efficacy and work-related performance: A meta-analysis. Physiological Bulletin 124: 240–
261.
Suls, J., and L. Wheeler. 2001. Handbook of Social Comparison Theory and Research. New York, NY: Kluwer Academic/Plenum
Publishers.
Tafkov, I. D. 2013. Private and public relative performance information under different compensation contracts. The Accounting Review
88 (1): 327–350. doi:10.2308/accr-50292
Tice, F. 2015. The Effect of Relative Performance Evaluation on Investment Efficiency and Firm Performance. Working paper, University
of Colorado Boulder.
Vagneur, K., and M. Peiperl. 2000. Reconsidering performance evaluation style. Accounting, Organizations and Society 25 (4/5): 511–
525. doi:10.1016/S0361-3682(98)00002-6
Watson, D., L. A. Clark, and A. Tellegen. 1988. Development and validation of brief measures of positive and negative affect: The
PANAS scales. Journal of Personality and Social Psychology 54 (6): 1063–1070. doi:10.1037/0022-3514.54.6.1063
Webb, R. A. 2004. Managers’ commitment to the goals contained in a strategic performance measurement system. Contemporary
Accounting Research 21 (4): 925–958. doi:10.1506/KHW0-G7PY-AQEA-718J
Wilson, T. D., and D. T. Gilbert. 2005. Affective forecasting: Knowing what to want. Current Directions in Psychological Science 14 (3):
131–134. doi:10.1111/j.0963-7214.2005.00355.x
Wilson, T. D., T. P. Wheatley, J. L. Kurtz, E. W. Dunn, and D. T. Gilbert. 2004. When to fire: Anticipatory versus postevent reconstrual
of uncontrollable events. Personality and Social Psychology Bulletin 30 (3): 340–351. doi:10.1177/0146167203256974
Wilson, T. D., D. B. Centerbar, D. A. Kermer, and D. T. Gilbert. 2005. The pleasures of uncertainty: Prolonging positive moods in ways
people do not anticipate. Journal of Personality and Social Psychology 88 (1): 5–21. doi:10.1037/0022-3514.88.1.5

APPENDIX A
Summary of Research Question

Our research question focuses on the impact of causal language on responses to feedback (Link E), and in the experiment,
we manipulate observable variables in solid boxes, although we are unable to observe variables in dotted boxes. We maintain
the assumption that the feedback contains information that is pertinent and relevant to future performance, and operationalize
this assumption by (1) basing past and future performance on the same task, and (2) basing feedback valence on past
performance.
Our first hypothesis predicts that when performance feedback is addressed by poor-performing individuals, it is likely to
improve performance (Link D). However, it is less clear in the literature whether performance feedback can improve future
performance for high-performing individuals. Several streams of literature have suggested that performance feedback can have
detrimental effects for high-performing individuals (Link D). We have no directional prediction of the effect of causal language
on subsequent performance when initial performance is good and resulting feedback is positive and, thus, state our second
hypothesis in null form.

The Accounting Review


Volume 93, Number 2, 2018
Because of ‘‘Because’’: Examining the Use of Causal Language in Relative Performance Feedback 297

APPENDIX B
Experimental Design and Order of Tasks

APPENDIX C
Performance Feedback Given to Participants during Experiment Following First Round
Note: Words classified as causal by LIWC 2007 shown in italics.

a. Condition: high causal feedback (below-median performance shown in parentheses)


Your result on Part 1: XX
Your result is higher (lower) than the median result of other participants.
The reason why your result is higher (lower) than the median result is:
 Because you generated more (less) words than other participants.
 Because the words you generated were longer (shorter) than the average word generated by other participants.
 Because the words you made used more (less) letters than words made by other participants.
 Because the words you made used more (less) complex letter combinations than other participants used.

b. Condition: low causal feedback (below-median performance shown in parentheses)


Your score on Part 1: XX
Your score is higher (lower) than the median score of other participants.
Your score is higher (lower) than the median:
 You generated more (fewer) words than other participants.
 Your words were longer (shorter) than the average word of other participants.
 Your words used more (fewer) letters than other participants’ words.
 You used more (less) complex letter combinations than other participants.

The Accounting Review


Volume 93, Number 2, 2018
Copyright of Accounting Review is the property of American Accounting Association and its
content may not be copied or emailed to multiple sites or posted to a listserv without the
copyright holder's express written permission. However, users may print, download, or email
articles for individual use.

Vous aimerez peut-être aussi