Description
This paper provides experimental evidence about how the interaction between a company's earnings and
its information system influences the degree of honest reporting by managers in a capital budgeting task.
Specifically, the results show that participants overstate cost less when the manager's cost report determines
whether the firm earns a gain or loss than when their report does not affect whether the firm
earns a profit or loss (i.e., the firm always earns either a profit or loss regardless of the cost report). Further,
the results suggest that the impact of the earnings situation on the degree of honesty depends on whether
the firm uses an information system that improves its ability to detect misreporting. Specifically, the
earnings situation has less effect on the degree of honesty when the firm uses an information system. This
is because the information system decreases honesty when the manager's report determines whether the
firm earns a profit or loss but increases it otherwise. This study provides important insights into the
conditions under which information systems can crowd out prosocial behavior.
Earnings benchmarks, information systems, and their impact on the
degree of honesty in managerial reporting
*
Eddy Cardinaels
a, b, *
a
KU Leuven, Department of Accountancy, Finance and Insurance, Naamsestraat 69, B-3000 Leuven, Belgium
b
Tilburg University, Department of Accountancy, P.O. Box 90153, 5000 LE Tilburg, The Netherlands
a r t i c l e i n f o
Article history:
Received 16 May 2012
Received in revised form
9 September 2015
Accepted 10 September 2015
Available online xxx
Keywords:
Honesty
Capital budgeting
Earnings
Information system
Prosocial behavior
a b s t r a c t
This paper provides experimental evidence about howthe interaction between a company's earnings and
its information system in?uences the degree of honest reporting by managers in a capital budgeting task.
Speci?cally, the results show that participants overstate cost less when the manager's cost report de-
termines whether the ?rm earns a gain or loss than when their report does not affect whether the ?rm
earns a pro?t or loss (i.e., the ?rmalways earns either a pro?t or loss regardless of the cost report). Further,
the results suggest that the impact of the earnings situation on the degree of honesty depends on whether
the ?rm uses an information system that improves its ability to detect misreporting. Speci?cally, the
earnings situation has less effect on the degree of honesty when the ?rmuses an information system. This
is because the information systemdecreases honesty when the manager's report determines whether the
?rm earns a pro?t or loss but increases it otherwise. This study provides important insights into the
conditions under which information systems can crowd out prosocial behavior.
© 2015 Elsevier Ltd. All rights reserved.
1. Introduction
Research has demonstrated that in settings where managers are
able to misrepresent cost reports, many managers still produce
honest reports because their dishonest reporting may negatively
affect the wealth of others (Evans, Hannan, Krishnan, & Moser,
2001). More evidence is however, needed on when managers
more strongly pursue this motive to act honestly. This study shows
that the company's earnings situation can serve as an important
contextual feature. Speci?cally, I presume that the degree of
honesty is higher when the manager's cost report determines
whether the ?rm earns a gain or a loss than when the manager's
report does not affect the ?rm's earnings situation. Studying the
effect of managerial in?uence on company's earnings is important.
While many studies on earnings management have explored the
impact of important earnings benchmarks on external reporting
(Burgstahler & Dichev, 1997), the effects of these benchmarks on
internal decisions like budgeting have received scant attention.
The lack of attention in prior studies is partially based on their
focus on companies that earn pro?ts from a manager's production
(e.g., Rankin, Schwartz, & Young, 2003, 2008). I argue that consid-
ering the ?rm's pro?t situationcanenrichour understanding of why
information systems to detect misreporting are sometimes not
effective (Christ, Emett, Summers, & Wood, 2012; Salterio & Webb,
2006). I predict that information systems and the earnings of the
?rminteract such that the bene?cial effects on honesty of the ?rm's
earnings situation are mitigated once an information system is
present. I use a capital budgeting task to test this prediction. The
?rm's earnings situationis manipulatedas the ?rst between-subject
factor. In the gain/loss condition, the participant's cost report can
determine whether the company earns a gain or a loss. In the two
other conditions, labeled as the positive earnings condition and the
negative earnings condition, the participant's cost report cannot
affect the ?rm's pro?t situation; that is, the ?rm always earns a
pro?t or a loss regardless of the manager's cost report. The second
*
The author would like to thank the editor Mike Shields and two anonymous
reviewers for their helpful comments. I would further like to thank Nathalie Beckers,
Jan Bouwens, Bart Dierynck, Stefan Hollander, Laurence van Lent, Donald Moser,
Bernhard Reichert, Jeroen Suijs, Kristy Towry, Michael Williamson, Andrew Yim,
Huaxiang Yin, and Rick Young for their helpful suggestions. I further acknowledge
the useful comments of participants at the European Accounting Association Con-
ference 2010 (Istanbul) and the Management Accounting Section conference 2011
(Atlanta) and seminar participants at Tilburg University, VU University of Amster-
dam, KU Leuven, and the University of Bern. This paper was formally titled “Refer-
ence Points and Budget Requests: Can Controls Destroy Honesty in Managerial
Reporting?” Experimental materials available on authors request.
* KU Leuven, Department of Accountancy, Finance and Insurance, Naamsestraat
69, B-3000 Leuven, Belgium.
E-mail address: [email protected].
Contents lists available at ScienceDirect
Accounting, Organizations and Society
j ournal homepage: www. el sevi er. com/ l ocat e/ aoshttp://dx.doi.org/10.1016/j.aos.2015.09.002
0361-3682/© 2015 Elsevier Ltd. All rights reserved.
Accounting, Organizations and Society xxx (2015) 1e13
Please cite this article in press as: Cardinaels, E., Earnings benchmarks, information systems, and their impact on the degree of honesty in
managerial reporting, Accounting, Organizations and Society (2015),http://dx.doi.org/10.1016/j.aos.2015.09.002
between-subject factor manipulates the absence or presence of
information systems that improve the ?rm's ability to detect mis-
reporting. Across all conditions, the pecuniary bene?ts of dishon-
esty for participants (and its costs for the ?rm) are held constant.
In the absence of an information system, the results show
that the degree of honesty is greater when the participant's cost
report can determine whether the ?rm earns a loss or pro?t (the
gain/loss condition) than in the positive or negative earnings con-
dition. In the gain/loss condition, a larger fraction of participants
remain honest or even underreport their costs and thus sacri?ce
money to avoid losses and keep the company pro?table. The
reporting feature of deliberately understating costs has not
received much attention but can be economically relevant (Erat &
Gneezy, 2012). Results further show that reporting behavior does
not differ between the positive and negative earnings conditions.
Similar to earlier work (Evans et al., 2001; Hannan, Rankin, &
Towry, 2006), many managers produce partially honest reports in
these conditions. The results also show that the effect of the ?rm's
earning situation interacts with the use of an information system
that improves the ?rm's ability to detect suspicious reporting.
Speci?cally, the information system decreases the degree of
honesty when the manager's report determines whether the ?rm
earns a pro?t or loss but increases it otherwise (i.e., the positive or
negative earnings condition).
This study provides important insights into the conditions under
which controls can reduce prosocial behavior. Prior work has shown
that reliance on control systems to reduce misreporting may crowd
out some of the preferences for honesty but nonetheless has shown
that the ?rm's pro?t is still higher with a control system than
without one. Rankin et al. (2008) showed that opportunities for
principals to reject budget requests reduce the level of misreporting
and thus are bene?cial to the ?rm's pro?t. Hannan et al. (2006) also
showed that when information systems are used to improve the
?rm's ability to detect misreporting, honesty is increased compared
to when they are not used. This study, however, shows that the
otherwise positive effects of information systems may not always
materialize. It also offers a rationale for this crowding-out effect
using the self-concept maintenance theory of Mazar, Amir, and
Ariely (2008). In the gain/loss condition participants start to pro-
duce small dishonest reports, which the ?rm cannot detect as
misreporting, once an information system is present instead of
reporting behavior that could be bene?cial to ?rm pro?t. If their
reporting behavior has less in?uence on the ?rm's earnings situa-
tion, such as in the negative or positive earnings conditions, infor-
mation systems to detect suspicious reporting still tend to reduce
misreporting by reducing the range of dishonest reports.
The ?ndings concerning these earnings benchmarks also offer
many practical insights. Strong variations in earnings are often
caused by changes in the economic condition, such as temporary
?uctuations in prices or pro?t margins, or by the type of product
that the business unit produces. For example, business units pro-
ducing new products or products with spillover effects on other
products are often close to breakeven.
1
If companies feel that the
manager can make a difference between experiencing losses or
pro?ts, it may be tempting for them to install systems that help the
company to detect misreporting. Such detective forms of controls
are often part of the ?rm's internal control procedures (Christ et al.,
2012), and resorting to such systems maybe a natural response in
case pro?ts start to erode. The results show, however, that infor-
mation systems to detect misreporting can be less effective, in
particular when pro?ts are under pressure.
Besides earnings levels, many other situations in a company
may alter participants' views of the repercussions of their mis-
reporting for the organization. Business units need to achieve
certain targets before bonus pools are paid out to employees, ?rms
need to meet or beat analyst expectations, and certain actions can
hurt only a few but also many other business units. Prior work by
Church, Hannan, and Kuang (2012) showed for example that people
report more dishonest and thus care less about the ?rm when
bene?ts of misreporting are shared with other managers in the
company. This paper shows that considering these repercussions of
managerial dishonesty on the ?rm is important, as doing so may
help organizations to utilize their controls more effectively.
2. Theory and hypothesis development
Evans et al. (2001) showed that many agents in capital budg-
eting produce partially honest reports even when ?nancial in-
centives for misreporting are fully present. Based on the ?nding
that individuals value honesty, follow-up studies have focused on
incentive mechanisms, monitoring systems, or other types of con-
trol systems that can help companies to improve honest reporting.
For example, prior studies have investigated competition among
agents or whistle-blowing by fellow agents in relation to honesty
(Brüggen & Luft, 2011; Zhang, 2008) or examined the effects of
social norms or peer behavior (Tayler & Bloom?eld, 2011;
Cardinaels & Jia, 2015). Other studies examine changes in eco-
nomic incentives, opportunities for principals to reject budget
proposals, or systems to reduce information asymmetry between
the business owner and the agent in relation to truthful reporting
(Evans et al., 2001; Hannan et al., 2006; Rankin et al., 2008).
However, fewer studies have focused on the organizational
settings in which managers more strongly pursue honest reporting
without touching upon costly incentive devices or control systems.
An exception is, for example, Church et al. (2012), who documented
that people are more honest when they fully bear the consequences
of their dishonesty than when the bene?ts of their dishonesty are
shared with other organizational members. This paper examines if
the ?rm's earnings situation can also serve as an important
contextual factor which may affect reporting behavior by man-
agers. Considering this variation may offer additional insights into
the crowding-out effects of information systems to detect mis-
reporting. The ?rst section will argue that participants overstate
costs less when their reports can make a difference between gains
or losses for the ?rm than in two other conditions where the
company always realizes either positive or negative earnings
regardless of the cost report. Next, I will discuss how the ?rm's
pro?t situation interacts with information systems that companies
use to detect misreporting.
2.1. The ?rm's earnings situation and the effect on honesty
Given the information asymmetry that exists between the agent
and the principal in a capital budgeting context, agency theory
would predict that the company's earning situation would not
matter because agents will always try to pro?t from dishonesty.
This study predicts that the degree of honesty - measured by the
level of cost overstatements - is higher when participants' cost
1
New products like, for example, new generations of smart phones are often not
pro?table. Nevertheless, once demand increases and learning takes place, pro?ts
start to accrue. Products with spillover effects are, for example, ink-jet printers.
Typically these printers are sold for a small loss, but business units producing the
cartridges that are used with them make a pro?t. Because ?rms often commit to a
customer base, they sometimes need to accept small losses when prices are under
pressure. On average, such ?rms expect to be pro?table by serving their customer
base, but temporary price ?uctuations can lead to losses, pro?ts, or pro?ts that are
close to zero. From a decision-control perspective (Zimmerman, 2009), various
types of cost allocations and transfer-pricing policies may further lead to differ-
ences in the division's contribution to organizational pro?ts.
E. Cardinaels / Accounting, Organizations and Society xxx (2015) 1e13 2
Please cite this article in press as: Cardinaels, E., Earnings benchmarks, information systems, and their impact on the degree of honesty in
managerial reporting, Accounting, Organizations and Society (2015),http://dx.doi.org/10.1016/j.aos.2015.09.002
reports can determine whether the company incurs a gain or a loss
than when participants' reporting behavior does not change the
company's earnings situation.
2
The arguments are two-fold.
The ?rst argument rests on the theory of self-concept mainte-
nance by Mazar et al. (2008). This theory argues that individuals try
to derive some bene?ts of cheating while at the same time trying to
maintain their positive self-concept of being an honest person. In
other words, there is a potential range of dishonesty in which
people will act dishonestly but their behaviors do not bear nega-
tively on their self-concept. Mazar et al., however, hinted at the fact
that this range may decrease and people may thus be shifted to-
ward more honest behavior if the context makes it dif?cult for
them to interpret a dishonest act as still being consistent with this
positive self-concept. I predict that the range of dishonest reports
that can be made without altering the individual's self-concept is
smaller in the gain/loss condition than in the two other conditions.
In the negative earnings condition or the positive earnings
condition, a dishonest report will negatively affect the other party's
wealth, but it however, does not really change the ?rm's pro?t
situationdthat is, the ?rm will still either earn a pro?t or incur a
loss regardless of the manager's report. This may offer room for
participants to produce a larger range of partially dishonest reports,
which would still be seen as consistent with the self-concept of
being an honest person. In the situation where participants'
reporting behavior can produce gains or losses for the ?rm, the
agent, however, may observe that a dishonest act, sometimes even
a small one, can turn pro?ts into losses for the ?rm. When the ?rm
would earn a pro?t for an actual cost draw, agents may avoid the
dishonest choice of overstating costs because it may turn this pro?t
into a loss. Conversely, if the company incurs a loss for an actual
cost draw, agents can turn this loss into a pro?t by sacri?cing part of
their resources through submitting a cost that is less than the actual
cost. Turning pro?ts into losses through lying is dif?cult for agents
to reconcile with a positive self-concept, while making small sac-
ri?ces to make the ?rm pro?table can reinforce participant's self-
concept of an honest person. Hence, the potential range of
dishonest reports that can be reconciled with a positive self-
concept is smaller in the gain/loss condition, and may shift par-
ticipants toward prosocial behavior. Choices to underreport also
occurred more frequently in the experiment of Evans et al. (2001),
where participants received a hurdle contract.
3
This paper argues
that such behavior can become relevant (Erat & Gneezy, 2012)
when participants perceive their reporting behavior to have
stronger repercussions on the company's pro?t.
A second argument, leading to the same prediction, is that
people may be intrinsically motivated to avoid losses for other ac-
tors. Indirect evidence from strategic games on cooperation
(Cachon &Camerer, 1996; Drake &Haka, 2008; Feltovich, Iwasaki &
Oda, 2012) suggests that people behave differently when either one
or more parties in the game can experience a loss than when all
parties would only be able to realize positive pay-offs.
4
Many
agents then utilize socially rational strategies to avoid losses and
expect other people to also avoid such points (Cachon & Camerer,
1996). In the gain/loss condition, participants observe that report-
ing behavior can make a difference between losses of pro?ts for the
other party. Individuals’ own reactions to losses are often severe
and most individuals would try to avoid them if possible. While in
the setting, individuals do not experience the loss themselves, they
nevertheless may still adapt their behavior vicariously (Welten,
Zeelenberg, & Breugelmans, 2012; Wood & Bandura, 1989). That
is, because they observe the consequences of dishonesty on the
other party, people may refrain from dishonest reports that push
the other party into a loss. This is labeled as vicarious loss avoidance
behavior, which can only take place in the gain/loss condition,
where participants can avoid losses for the ?rm.
2.2. Different earnings situation, information systems and their
impact on honesty
Consistent with Hannan et al. (2006) and Schulze and Frank
(2003), the information system provides the company with a nar-
rower range of potential costs and thus enables the company to
detect suspicious reporting if participants report a higher cost than
the upper limit of the interval.
5
Consistent with Hannan et al.
(2006, p. 896), the information system also does not introduce a
?nancial punishment for misreporting in order to ensure that in-
centives to misrepresent costs are equivalent across all conditions.
Prior work on capital budgeting has shown that people's pref-
erences for honesty may diminish as a result of control systems.
Rankin et al. (2008) showed that the participant's preferences for
honesty are lower when principals have the authority to reject their
budget requests. However, organizations still bene?t from this au-
thority because participants report more truthfully when rejection
opportunities are present compared to when they are not. Similarly,
Hannan et al. (2006) showed that the use of an information system
that increases the ?rm's ability to detect misreporting is not
necessarily bad news. When ?rms install information systems to
reduce the information asymmetry, people can still give the
impression of an honest appearance. In order to do so, they reduce
their misreporting to a level that will not be detected by the infor-
mation system. Overall, an information system still reduces
2
Prior work has indirectly shown that earnings may in?uence prosocial behavior.
In the study of Hannan (2005), agents act reciprocally toward higher wage offers of
the principal by voluntarily delivering more than an optimal effort when earnings
for the ?rm decreased (as, in such a case, higher wages are seen as a generous act)
compared to when earnings increased. Although the ?rm is always pro?table in this
study, agents care about random shocks in the principal's pro?t. An unanticipated
?nding in Evans et al. (2001, p. 547, fn. 11) also hinted at the fact that agents behave
differently depending on the context. Within the hurdle contract game, a setting in
which production would not take place if costs are too high, a sizeable number of
participants voluntarily sacri?ced money by submitting reports for costs less than
the actual costs. Participants may have had a social motive to understate the costs
to ensure production, so that principals would earn some pro?t.
3
In Evans et al. (2001, p. 547, fn. 11), 34 out of 138 agents (24.6 percent) delib-
erately understate costs at the expense of their ?xed wage when actual costs were
above the production hurdle. In this way, production took place and the company
would still make some money. Evans et al. (2001, p. 547, fn.11), also argue that
subjects knowledgeably sacri?ce part of their wealth to increase ?rm pro?t, to
ensure that production would take place.
4
These strategic games use simple transformations to the players' pay-off matrix
(which are irrelevant from a game-theoretic perspective) and show that behavior is
different when such losses come into play. While the capital budgeting game in-
volves little strategic interactions as the agent always decide about the report, the
manipulation of the ?rm's earnings situation resembles the manipulations
observed in these more strategic games. By manipulating the product's contribu-
tion margin, I focus on simple transformations, which would not change the eco-
nomic self-interested outcome. Yet, the manipulations allow to differentiate
between settings in which participant's cost report can make a difference between
losses or pro?ts for the company and where losses can thus be avoided and settings
in which participant's report has less in?uence on company pro?ts (i.e. positive or
negative earnings conditions).
5
The information system is a form of detective control offering a warning for
suspicious reporting (Christ et al. 2012). Section 3 describes the manipulation in
detail. In short, when an information system is absent, participants made ?ve
reporting decisions under full information asymmetry. When an information sys-
tem is present, the company can examine two reporting decisions for which the
?rm receives a narrower interval of potential costs. If reported cost are higher than
the maximum cost of the interval, participants get a warning and they need to
submit a new cost report to the ?rm on the basis of a new cost draw. Regardless of
whether an information system is present or absent, all participants thus play ?ve
full rounds with information asymmetry, for which company earnings and their
pay-offs are calculated. In each condition, the maximum pay-off is achieved when
people lie to the full extent.
E. Cardinaels / Accounting, Organizations and Society xxx (2015) 1e13 3
Please cite this article in press as: Cardinaels, E., Earnings benchmarks, information systems, and their impact on the degree of honesty in
managerial reporting, Accounting, Organizations and Society (2015),http://dx.doi.org/10.1016/j.aos.2015.09.002
misrepresentation compared to not having one at all. Participants
avoid large cost overstatements under both a coarse and precise
information system, although a precise information system crowds
out some of the motives to appear honest (Hannan et al., 2006).
In contrast to these prior studies, I presume that the impact of an
information system, as a detective form of control, may depend on
the earnings situation. That is, compared to the negative earnings or
the positive earnings conditions, where information systems may
still reduce misreporting, the same systemmay have aweaker impact
or potentially increase misreporting if agents’ reporting behavior can
make a difference between losses or pro?ts for the company.
Experimental economists argue that detective forms of control
can have a ?attening effect on dishonesty (Schulze & Frank, 2003).
Even though they may crowd out some of the motives to act pro-
social, sel?sh behavior is also reduced. Prior work by Hannan et al.
(2006) showed that people are willing to give up some bene?ts of
misreporting and thus report more honestly under the presence of
systems that allow the company to make inferences regarding the
manager's honesty level. Using self-concept maintenance theory
(Mazar et al., 2008), I argue that such bene?ts are more likely to be
realized in the negative or positive earnings situation. As previously
argued, participants in these earnings situations may have a larger
range of dishonest reports for which they can maintain their self-
concept as being an honest person. Information systems that
allowthe ?rmto make some inferences about the honesty level of a
participant may reduce this range. That is, to preserve their positive
self-concept as an honest person, participants may reduce their
misreporting and thus give up some bene?ts of misreporting by
submitting cost reports that the information system cannot detect
as instances of misreporting (Hannan et al., 2006; Schulze & Frank,
2003). By reducing their misreporting, participants can still appear
to be honest to the company, which would be positively rewarded
in their personal value system (Mazar et al., 2008).
The same information system may work differently when a
participant's cost report is able to determine whether the company
earns a pro?t or loss (gain/loss condition). As mentioned, self-
concept maintenance theory would argue that the fact that par-
ticipant's reporting behavior can make the difference between
gains or losses for the company may already reduce misreporting
and may potentially shift a participant's reporting towards more
honest reporting behavior.
When an information systemis present, participants, receive, as
a result of the information system, an opportunity to reconcile
dishonest reports that can lead to a loss for the company or that
may prevent the company from earning pro?ts, as still consistent
with their positive self-concept. In particular, the detrimental effect
of dishonest reports that would put the ?rm into a loss on the
positive self-concept may be less signi?cant when a company
would use an information system (Matsushima, 2008, p. 354;
Charness & Dufwenberg, 2006). The use of an information system
can signal to the participant that the company detects and does not
accept large cost overstatements. Nevertheless, as long as a person
misreports by small amounts, which the information system
cannot detect, people may reconcile their partially dishonest report
as ?ne. The system allows participants to maintain a positive self-
concept by appearing honest, even if such reports may not be
bene?cial to ?rm pro?t. As a result, in settings where the reporting
behavior can make a difference to ?rm pro?ts more partially
dishonest reports may occur at the expense of honest reporting or
choices to underreport cost to make the ?rm pro?table. While in-
formation systems, as argued before, may reduce high levels of
misreporting, such information systems may also justify small
dishonest reports. In contexts, where earnings can vary between
losses or pro?ts for the ?rm, this can weaken the effect of an in-
formation system on misreporting.
Based on arguments of Tenbrunsel and Messick (1999), weak
sanction systems, such as a detective information system, may also
change the decision frame froma setting where people would worry
about ethical concerns to a setting where people would worry more
about ways to avoidpenalty. Without suchsystems inplace, stronger
ethical concerns for the ?rm may arise in particular when people's
behavior can make a difference between pro?ts and losses for a
company (e.g. people avoid losses vicariously), compared to when
people cannot change the earnings situation. While an information
system may reduce the range of misreporting because persons do
want to be detected as a dishonest person, people in the gain/loss
condition can also refrain from reports that may otherwise be
bene?cial for the company. People may submit partially dishonest
reports that the ?rm cannot detect, and may worry less about con-
sequences of their reports for the company's earnings. The concerns
of a person to avoidlosses for the company by reporting more honest
can play a weaker role once an information system is present.
Both arguments would predict an interaction whereby the
earnings situation has less in?uence on misreporting once an in-
formation system is present. Speci?cally, while an information
systemmay reduce large cost misrepresentations in the negative or
positive earnings conditions, the same information system in the
gain/loss condition may reduce the reporting of choices which can
be bene?cial to ?rm pro?t. Hence, the net effect is that an infor-
mation system has a less positive (or even a negative) impact on
honesty in the gain/loss condition relative to the two other
conditions.
Hypothesis. The effect of an information system on honest
reporting is less positive (or potentially negative) when the man-
ager’s cost report determines whether the ?rmearns a gain or a loss
than when the ?rm always realizes positive or negative earnings.
3. Experiment
3.1. Participants
A total of 186 participants from an accounting course in a
business studies program at a large West-European university,
participated in the study as part of the course requirement across
two different administrations of the experiment. Most of the stu-
dents were undergraduates in the last year of their bachelor degree
(96.8%), although a minority were masters-level students (3.2%).
Participants’ average age was 21.23 years and 59.14% were male.
They had on average taken 2.28 accounting courses and 74.7% of the
students reported to have work experience (Their average part-
time work experience was 19.4 months).
3.2. Experimental task
I used a capital budgeting task, similar to that of Evans et al.
(2001). Participants assumed the role of a production manager
(i.e., the agent) and worked for one of the divisions of Acazia
headquarters (i.e., the principal).
6
They received a ?xed wage of
6
For reasons of task realism, I used a hypothetical company where consequences
of dishonesty come at the cost of the experimenter (See also Evans et al. 2001,
Evans, Moser, Newman, & Stikeleather, 2015). Experimental instructions
mentioned that participants would work for one of the divisions of corporate
headquarters. Depending on the earnings manipulation, headquarters would either
always receive a loss, always receive a pro?t, or receive either a gain or a loss
depending on the participant's cost report. This earnings manipulation is likely to
have greater external validity when a common company is involved. Second, such a
design gives dishonesty a strong change to survive. Any deviation against this
agency prediction can then be attributed to a social motive to be honest.
E. Cardinaels / Accounting, Organizations and Society xxx (2015) 1e13 4
Please cite this article in press as: Cardinaels, E., Earnings benchmarks, information systems, and their impact on the degree of honesty in
managerial reporting, Accounting, Organizations and Society (2015),http://dx.doi.org/10.1016/j.aos.2015.09.002
250 lira (50 lira ¼ 1 euro). Participants then submitted budget
requests to headquarters to receive funding for the production of a
lot of 500 units in each period. At the start of each period, both
headquarters and the participants knew that the per-unit pro-
duction cost was randomly drawn from a uniformly distributed
cost interval between 4.00 and 6.00 lira, with increments of 0.05
(e.g., 4.00, 4.05, …, 5.95, 6.00). Before submitting a cost report to
headquarters, participants, however, received feedback from a
private forecasting system, which offered them an accurate fore-
cast of the actual production cost for the production period.
Headquarters, however, would never learn the actual costs and
thus can never assess if participants report honestly. Participants
then submitted a cost report to headquarters. If participants re-
ported a cost higher than the actual cost, they could keep the
excess resources in addition to their ?xed wage. In sum, a par-
ticipant's earnings are equal to:
(1) Participant's earnings ¼ 250 lira þ 500 lira * (reported
cost e actual cost)
Participants were told that the company earns a contribution on
the output produced in each production period (i.e., the contribu-
tion margin will be manipulated between subjects; see next sec-
tion). In sum, headquarters’ earnings were calculated as follows:
(2) Headquarters' earnings ¼ 500 lira * (contribution per
unit e reported cost) e 250 lira
Instructions also emphasized to participants that earnings are
very important for headquarters and that higher (lower) cost
reports would lead to lower (higher) earnings for the ?rm. If
participants report less honest, their own payoffs would increase,
but earnings for headquarters would decrease. To make this
trade-off clear, the instructions given to participants provided
examples of how budget requests affect their own earnings as
well as their contribution to the ?rm's pro?t. In addition, par-
ticipants received a payoff table based on an actual cost of 4.50
lira, showing all possible outcomes of headquarters' earnings and
their own payoffs for all the potential cost reports they could
submit, including all the combinations that were for costs less
than actual costs. Participants then played ?ve production pe-
riods. Before submitting their cost report, participants again
received a full payoff table containing the pay-off and earnings
consequences for the full range of possible cost reports for the
respective actual cost draw, including also all combinations for
less than actual costs.
3.3. Between-subjects manipulations
The study employs a 3 Â2 between-subjects design. As a ?rst
factor, I manipulated three different earnings situations. One-third
of the participants were told that headquarters realizes a 4.00 lira
contribution margin on each production unit. This scenario is
labeled as the negative earnings (NE) condition because head-
quarters always realizes a loss. The loss can vary between e1250
lira and À250 lira depending on the submitted cost report. Another
third of the participants were told that headquarters receives a
contribution margin of 7.00 lira. This scenario is labeled as the
positive earnings (PE) condition. In this scenario, headquarters will
always realize a pro?t that can vary between 250 lira and 1250 lira.
The rest of the participants were given the gain/loss (GL) condition
with a contribution margin of 5.50 lira. Compared to the conditions
where the ?rmwill either earn a loss (NE) or a pro?t (PE) regardless
of the manager's report, the GL condition is the only situation
where the manager's report can determine whether the company
earns a loss or a pro?t. Earnings can vary fromÀ500 lira to 500 lira
and become negative when the reported cost is above 5.00 lira and
positive when the reported cost is below 5.00 lira. In all cells,
participants are also told that headquarters do not face any risk of
failure.
7
From a self-interested perspective, the ?rm's earnings
situation would not matter. According to formula (1), participants
will maximize their payoff when costs are fully overstated (6.00
lira). Consistent with the theory, however, I presume that when the
report can determine whether the company earns a gain or a loss,
dishonest reporting may be more dif?cult to reconcile with the
participant's self-concept.
As a second between-subject manipulation, the study manipu-
lated the presence or absence of an information system. When the
information system was present, the company had some ability to
make an inference about suspicious reporting of the participant. In
this condition, participants were told that headquarters would
check reports in two out of ?ve production rounds (no more, no
less). Consistent with Hannan et al. (2006), headquarters would
merely receive a narrower range of potential costs if the report was
examined. Participants were told production would be ?nanced
(suspended) if the cost falls within (outside) the range. Participants,
however, received private information. They knew that the com-
pany's information systemhad the ability to reduce the information
asymmetry with regard to potential cost to a range of 0.5 lira
around the actual cost draw. For example, if a participant received
an actual cost of 4.50 lira, headquarters' information systemwould
display that the cost should be in the range of 4.25e4.75 lira rather
than in the range of 4.00e6.00 lira. In the production runs, pro-
duction was ?nanced on the basis of reported costs for all costs
below the maximum cost of this interval. If a participant submitted
a report higher than the maximumcost, headquarters would send a
warning message to the participant indicating that the report is
suspicious and that production will be suspended. Participants
would then receive a newcost draw for that production period and
they again would need to submit a cost report on which basis
production would be ?nanced.
Similar to the information absent condition, participants in the
information system present condition would thus always play ?ve
full production runs for which their own payoffs and their contri-
bution to the ?rm's earnings are calculated.
8
This was done to
ensure economic comparability across all six cells. For the same
reason, I do not attach a ?nancial consequence to suspicious
reporting. This approach is consistent with Hannan et al. (2006)
and allows the researcher to attribute deviations from full
dishonesty to the intrinsic motive of participants to report honestly.
These information systems are labeled as detective forms of control
(Christ et al., 2012). They are often part of a company's internal
control procedures in which reports by managers are (randomly)
checked and not accepted if the manager's report is perceived as
suspicious. Such systems often also provide a warning if the report
is considered to be suspicious (Christ et al., 2012).
3.4. Experimental procedures
Participants were randomly assigned to the experimental cells
7
To ensure that the NE or the GL condition are not perceived differently than the
PE condition in terms of risk, the case states the ?rm does not experience any
failure risk and that it has enough resources available from past activities.
8
In the experiment, headquarters checks the initial budget request of the second
and fourth production run. Only when reported costs are higher than the upper
limit, headquarters will suspend production. Participants are not aware of this.
Similar to in the information absent condition, the agency prediction in the infor-
mation system present condition is again to lie to the maximum amount possible,
in which case earnings for the participant are maximized.
E. Cardinaels / Accounting, Organizations and Society xxx (2015) 1e13 5
Please cite this article in press as: Cardinaels, E., Earnings benchmarks, information systems, and their impact on the degree of honesty in
managerial reporting, Accounting, Organizations and Society (2015),http://dx.doi.org/10.1016/j.aos.2015.09.002
when entering the computer room.
9
Sessions lasted an hour.
Following Evans et al. (2001), participation was anonymous. To
ensure full anonymity, students drew a ticket number, on which
basis they could claim their payoff. Before starting the task, they
provided background information on age, gender, work experience,
the number of accounting courses they followed, etc. Then, they
completed a nine-item instrument, which measured their social
value orientation(see VanLange, Bekkers, Schuyt, &vanVugt, 2007).
This instrument classi?es people into “competitive,” “individual,”
and “prosocial” types of players and can pick up some of the innate
preferences of people to act honestly. Prosocial people may care
more about another person's wealth.
10
Next, participants played a
distracter task and continued into the main task. Before making
reporting decisions, they carefully read the case descriptions. Par-
ticipants performed a true-false quiz about the case descriptions. To
increase task understanding, participants always received feedback
on the right answer for each question, before moving on to the next
question. Participants then played ?ve production runs, in which
actual cost was randomly drawn by the computer from the uniform
distribution of costs (4.00, 4.05, 4.10, …, 5.95, 6.00 lira).
Consistent with Evans et al. (2001) and Hannan et al. (2006),
participants knew that the earnings of one production period
would be randomly selected and converted into real cash at the end
of the experiment (50 lira ¼ 1 euro). After participants ?nished the
task, they ?lled out an exit questionnaire containing items on task
understanding, motivation, etc. and some manipulation checks. The
cash reward was collected two weeks after the last session. On
average, participants earned 7.69 euro (the minimum was 3 euro,
and the maximum, 25 euro).
11
3.5. Manipulation checks
Participants' task understanding and task motivation did not
vary across the manipulations.
12
Most participants (95.7%) under-
stood that their participation was fully anonymous. In total 84.4% of
the participants further agreed with the item “There is a clear
trade-off. The more I misrepresent costs, the lower the earnings for
corporate headquarters.” Importantly, I did not detect differences of
the manipulations on the anonymity question nor on the trade-off
question (all p's > 0.15, F-model ns). To check if results depend on
attendance to these manipulation checks, Section 4.1 also reports
the results for the hypothesis test using a reduced sample of par-
ticipants who agreed with both these items (n ¼ 150, or 80.6
percent of the sample).
13
Participants' perceptions of the ?rm's contribution margin were
directionally consistent with the earnings manipulation. One
seven-point Likert scale item (1 ¼ completely disagree,
7 ¼ completely agree) asked participants whether they felt that the
company earned a too low contribution. Only the main factor
earnings was signi?cant (F
(2, 180)
¼ 29.12, p < 0.01). Participants in
the NE condition achieved a higher score on this question
(M ¼ 5.74, SD ¼ 1.38) than participants in the GL condition
(M ¼ 4.60, SD ¼ 1.37) and participants in the PE condition
(M¼ 3.94, SD ¼ 1.22). The contrast differences between the NE and
GL conditions (F
(1, 180)
¼ 22.86, p < 0.01); the NE and PE conditions
(F
(1, 180)
¼ 56.88, p < 0.01); and the GL and PE conditions (F
(1,
180)
¼7.62, p < 0.01) were all signi?cant. The post questionnaire did
not have a manipulation check for the between-subject factor in-
formation system, as it was obviously either present or absent.
4. Results
4.1. Hypothesis test
Consistent with Brüggen and Luft (2011), I use the degree of
misrepresentation of the private cost signal as the dependent var-
iable, which is de?ned as the reported cost minus the actual cost.
This measure perfectly correlates with the participant's earnings in
the experiment. I use it to avoid losing the information value of
choices where participants submitted reports for costs less than the
actual costs. The frequency of such reports varies strongly across
cells (see Section 4.2). Although Evans et al. (2001) view the choice
to underreport a cost as inconsistent and set it as equal to honest
reporting in their measure, such a choice can be driven by social
9
I checked if randomization was successful (n ¼ 186). I do not ?nd signi?cant
effects of our manipulations for the variables age, the percentage of pro-socials, the
percentage of people who report to have work experience, and the no. of ac-
counting courses. Only when considering gender as dependent variable, I ?nd a
signi?cant interaction of earnings * information system (F
(2, 180)
¼ 2.48, p ¼ 0.086).
The negative and positive earnings conditions contained fewer males when infor-
mation system was absent (48, 4% for NE and 54.8% for PE) compared to when it
was present (67.7% of males for NE and 61.3% for PE). For the GL condition data
shows an opposite pattern with 71.0% males in the IS absent condition and 51.6% in
the IS present condition. The study also used two administrations of the experiment
to increase sample size; 144 students (randomized across six cells) participated in
Nov. 2007 and an additional 42 students (randomized across six cell) participated in
Nov. 2009. Timing of administration did not have an impact on the level of mis-
reporting (M ¼ 0.214; SD: 0.373 (n ¼ 144) vs. M ¼ 0.305, SD: 0.354 (n ¼ 42),
t
184
¼ À1.40, p > 0.16, two-tailed) nor on the aforementioned demographic vari-
ables (All p's > 0.12 two-tailed). Except for gender results show that the ?rst
administration contained more males (62.50%) compared to the second (47.62%; t
184
¼ 1.73, p ¼ 0.085 two-tailed). Results are similar when Gender is added as
covariate in Table 1; the covariate Gender is not signi?cant.
10
The instrument is fully described in Van Lange et al. (2007). When applying this
instrument, a small number of people could not be classi?ed because they make
inconsistent choices (Van Lange et al., 2007). In this study, 37.1% of the participants
were classi?ed as prosocial, 39.2% as individualistic, and 10.8% as competitive. The
remainder of the participants (12.9%) could not be classi?ed. Importantly, the dis-
tribution of people across categories (prosocial, individualistic, competitive, and
unclassi?ed) did not differ across the six experimental cells (c
2
(5, N ¼ 186): 2.973,
p ¼ 0.704), suggesting that randomization was successful. Consistent with the
theory in Van Lange et al. (2007), correlations with the dependent variable show
that people who are classi?ed as prosocial (prosocial is 1; and zero otherwise)
misrepresent costs less (r ¼ À0.156, p ¼ 0.034). Results do not qualitative differ
when I use an ANCOVA, controlling for the dummy prosocial as covariate. The
covariate prosocial is signi?cant (F (1, 179) ¼ 3.92, p ¼ 0.05). The results and sig-
ni?cance levels in Tables 1 and 2 have similar magnitudes and signi?cance levels.
11
The maximum of 25 euro is achieved at a cost of 4 lira and a submitted report of
6 lira. Honest reporting results in a reward of 5 euro (¼250 lira of ?xed wage). By
reporting a cost below the actual cost, some participants sacri?ced part of their
?xed wage or more than their ?xed wage leading to a negative pay-off. If a round
was drawn for pay-out in which payoffs were lower than 3 euro, participants still
received 3 euros for participation. This was, however, not announced to them
beforehand.
12
Task understanding uses two seven-point Likert scale items in the post ques-
tionnaire (Cronbach's a ¼ 0.74). Results showed no signi?cant effects of the ma-
nipulations (all p's > 0.11, F-model ns). Task motivation also uses two seven-point
Likert-scale items (Cronbach's a ¼ 0.66). Again, the main effects and interactions
are not signi?cant (all p's > 0.26, F-model ns). Participants also indicate they had
suf?cient time to carry out the task (all p's > 0.36, F-model ns).
13
Note that the 36 people (who fail either one or both of these two items)
misreport cost less compared to the 150 participants, who agreed with both items
(mean level of misreporting M¼ 0.064, SD: 0.326 (n ¼ 36) vs. M¼ 0.275, SD: 0.368
(n ¼ 150), t
184
¼ 3.15, p < 0.01). Results are driven by the answer on the trade-off
question (correlation trade-off item (agree ¼ 1; 0 otherwise) with level of mis-
reporting r ¼ 0.297, p < 0.01). More honest participants are less likely to agree with
this item, because for them such a trade-off is less present. Demographic charac-
teristics are relatively similar between the two groups with respect to age, gender,
the percentage of pro-socials, and the no. of accounting courses (All p's > 0.16 two-
tailed) except for percentage of participants with work experience (M ¼ 86.1%
(n ¼ 36) vs. M ¼ 72.0% (n ¼ 150), t
184
¼ À1.75; p ¼ 0.081); Task understanding as
measured with two PEQ items also does not signi?cantly differ (M ¼ 5.32, SD: 1.35
(n ¼ 36) vs. M ¼ 5.54, SD: 1.21 (n ¼ 150), t
184
¼ 0.94, p > 0.34). The subsample
however, does exclude one participant (outlier) in the NE/IS present condition who
never realized a positive pay-off (i.e. he or she scari?ed the full ?xed wage or more
than the ?xed wage in all ?ve trials).
E. Cardinaels / Accounting, Organizations and Society xxx (2015) 1e13 6
Please cite this article in press as: Cardinaels, E., Earnings benchmarks, information systems, and their impact on the degree of honesty in
managerial reporting, Accounting, Organizations and Society (2015),http://dx.doi.org/10.1016/j.aos.2015.09.002
concerns. Participants may decide to underreport the cost to
improve the earnings of the company.
14
This measure thus also
better captures the consequences of participant's behavior to ?rm
pro?t.
Because there is less roomto misrepresent costs at high levels of
the actual cost, I also control for the actual cost in a second model
(see also Brüggen & Luft, 2011). Table 1 presents the results of the
full-factor ANOVA (Model 1) and a mixed model procedure (Model
2) with the actual cost as a time-varying covariate (i.e., 41 possible
values between 4.00, 4.05, 4.10, …, 5.95, and 6.00). In both models,
the reporting period is treated as a within-subject effect. Results
concerning the hypothesis test (i.e., Table 1) are based on two-
tailed tests. Comparisons of the simple effects of the interaction
using least-square means (i.e., Table 2) are based on one-tailed
directional tests.
Panel A of Table 1 shows the cell means and the conditional
means controlling for the actual cost. In all cells, participants on
average overstate costs (i.e., misrepresentation > 0) and thus
extract rents from the organization. Similar to Evans et al. (2001),
participants strongly deviate from the full self-interested choice of
reporting the maximum cost of 6 lira, in which case the degree of
misrepresentation would be close to 1 (because the expected value
of the actual cost is 5 lira). Importantly, the degree of misrepre-
sentation differs strongly across cells.
The hypothesis predicts an interaction between earnings and
information systems (E ÂIS), indicating that the effect of an infor-
mation system on the degree of honesty is less positive (and
potentially negative) in the settings where the cost report can
determine whether the ?rm earns a gain or a loss (GL condition)
compared to the two other conditions (NE or PE conditions). Panel B
of Table 1 shows that the interaction ExIS is signi?cant in both
models at the 10% level (p ¼ 0.054 in Model 1 and p ¼ 0.064 in
Model 2). The means in panel A of Table 1 indicate that both in the
negative earnings (NE) and positive earnings (PE) condition, infor-
mation systems reduce the degree of misrepresentation, whereas
information systems increase misrepresentation in the gain/loss
(GL) condition. The contrast analyses in panel B of Table 1 offer more
detail on this interactive pattern. The interaction is signi?cant when
comparing the GL condition against the NE condition (p ¼ 0.023 in
Model 1 and p ¼ 0.039 in Model 2), when comparing the GL con-
dition against the PE condition (p ¼ 0.063 in Model 1 and p ¼ 0.047
in Model 2); and when comparing the GL condition with the pooled
means of the PE and NE conditions (p ¼ 0.017 in Model 1 and
p ¼ 0.019 in Model 2). Information systems have a less positive ef-
fect (i.e. a negative effect) on honesty in the GL condition compared
to the two other conditions. The interaction is not signi?cant when
comparing the PE against the NE condition, because the information
system reduces misreporting in both cases.
15
Panel B of Table 2 explores the means (i.e. conditional means) by
comparing the simple effects of the interaction using the least
square means procedure. When the information system is absent,
the degree of misrepresentation is higher when comparing both
the negative earnings and positive earnings conditions against the
gain/loss condition (the NE vs. GL condition þ0.170, p ¼ 0.034 one
tailed, or þ0.154, p ¼ 0.047 one-tailed when considering condi-
tional means; and the PE vs. GL condition þ206, p ¼ 0.014 one
tailed, or þ0.234, p ¼ 0.006 one-tailed when considering condi-
tional means). This is consistent with the self-concept maintenance
theory: individuals ?nd it more dif?cult to misreport costs in the GL
condition, where the cost report can determine whether the
company earns a gain or a loss. The presence of an information
system reduces the motive to report more honestly in the GL
condition. When the information system is present, results show
that the GL condition has higher degrees of misrepresentation and
the level is comparable to or even higher than the negative or
positive earnings conditions. The simple effects of information
system in Panel B of Table 2 show that information systems
marginally reduce misrepresentation in both the NE (À0.177,
Table 1
Hypothesis test.
Panel A: Summary statistics for misrepresentation (Y ¼ reported cost e actual
cost)
Negative earnings
(NE)
Positive earnings
(PE)
Gain or losses
(GL)
Info system absent 0.309 0.345 0.139
[0.296] [0.375] [0.142]
(0.340) (0.327) (0.451)
n ¼ 31 n ¼ 31 n ¼ 31
Info system present 0.132 0.222 0.262
[0.143] [0.233] [0.259]
(0.398) (0.233) (0.404)
n ¼ 31 n ¼ 31 n ¼ 31
Panel B: Full factor analysis with period as a within-subject factor
Model 1 Model 2
DF MS F Stat Sign. Num.
DF
Den
DF
F Stat Sign.
Between subjects
Earnings (E) 2 0.58 0.86 0.424 2 180 1.46 0.234
Info. System (IS) 1 0.81 1.22 0.271 1 180 1.27 0.262
E*IS (H1) 2 1.98 2.97 0.054* 2 180 2.79 0.064*
Error 180 0.67 e e e
Within subjects
Actual cost e 40 680 11.89 0.17) when the information system is present. In the GL con-
dition, the information system marginally increases misreporting
(one tailed p ¼ 0.05 for the mean; one-tailed p ¼ 0.07 for the
conditional mean), whereas it marginally reduces misreporting in
the PE condition (one tailed p ¼ 0.08 for the mean; one-tailed
p ¼ 0.06 for the conditional mean). In contrast to Table 2, the
decrease in misreporting as a result of information system for the
NE condition is no longer signi?cant (one-tailed p > 0.24 for the
mean; one-tailed p > 0.28 for the conditional mean).
4.2. Supplementary analyses of choice behavior
Table 3 analyzes the frequency of reporting behavior that can
be considered as bene?cial to the company. I count the number of
observations per participant over the ?ve reporting periods
where reported costs are equal to or belowactual costs, labeled as
total social choices in Panel A of Table 3 (is equal to the sum of
honest reports (i.e., reports equal to actual cost) and altruistic
reports (i.e., reports below actual costs)). The remainder are
dishonest reports for which reported cost are higher than actual
cost (total dishonest in Panel A of Table 3). The theory suggested
that individuals may be more honest or underreport costs to
improve the ?rm's pro?ts and thus misreport less if their reports
can determine whether the company earns a pro?t or loss (the GL
condition) compared to negative and positive earnings condi-
tions. Such behavior in the GL condition may occur less frequently
Table 2
Pairwise comparison of the interaction.
Panel A: Display of the means [conditional means controlling for covariate actual cost]
Panel B: Comparison simple effects per earnings condition and per information system condition
Comparison Means Conditional means
Info system absent Info system present Info system absent Info system present
Effect (sign.) Effect (sign.) Effect (sign.) Effect (sign.)
NE vs. GL þ0.170 (0.034)** À0.130 (0.081)* þ0.154 (0.047)** À0.116 (0.104)
PE vs. GL þ0.206 (0.014)** À0.040 (0.332) þ0.234 (0.006)*** À0.026 (0.390)
NE vs. PE À0.036 (0.351) À0.090 (0.167) À0.079 (0.193) À0.090 (0.163)
Effect info system Effect info system
Effect (sign.) Effect (sign.)
Neg. earnings (D-A) À0.177 (0.029)** À0.153 (0.048)**
Pos. earnings (E-B) À0.123 (0.094)* À0.142 (0.061)*
Gain or loss (C-F) þ0.123 (0.094)* þ0.117 (0.102)
***, **, * indicate signi?cance levels of 1%, 5%, and 10% levels of signi?cance using one-tailed directional tests. Panel A shows a graphical plot of the cell means of Table 1
(respectively the conditional means using actual cost as time varying covariate). Panel B performs a mean by mean comparison of the interaction earnings  information
system of Table 1 across information system condition or across earnings condition. Differences between the means are displayed ?rst; the one-tailed p-values are shown
between brackets and are based on directional test using the least squares means procedure within an ANOVA or mixed model procedure (conditional means).
E. Cardinaels / Accounting, Organizations and Society xxx (2015) 1e13 8
Please cite this article in press as: Cardinaels, E., Earnings benchmarks, information systems, and their impact on the degree of honesty in
managerial reporting, Accounting, Organizations and Society (2015),http://dx.doi.org/10.1016/j.aos.2015.09.002
once an information system is present. I thus anticipate an
interaction effect, suggesting that the effect of information sys-
tem on this type of reporting behavior depend on earnings con-
dition. A type 3 analysis, using frequency of social choices on
participant level of as the dependent (multinomial response),
shows, however, that the interaction of ExIS is not signi?cant (c
2
(2, N¼186)
¼ 3.54, p ¼ 0.17). Also the main effects are not signi?cant
(information system c
2
(1, N¼186)
¼ 1.07, p ¼ 0.30; earnings c
2
(2,
N¼186)
¼ 4.29, p ¼ 0.12). Contrasts show that the interactive
pattern is signi?cant at the 10% level when comparing the GL
condition against the NE condition (c
2
(1, N¼186)
¼ 3.13; p ¼ 0.08)
and when comparing the GL condition against the combined cells
PE and NE (c
2
(1, N¼186)
¼ 3.44, p ¼ 0.06). The differential impact is
not signi?cant when comparing the GL condition against the PE
condition (c
2
(1, N¼186)
¼ 2.08, p ¼ 0.15).
Panel B of Table 3 compares the effect of an information
system on the frequency of social choices in each of the three
earnings condition, separately. The effect of information system
in the GL condition reduces the frequency of these reporting
choices from 54.84% to 35.48% (i.e. reduction of minus 19.35%
percent, p ¼ 0.04). As such, the use of information system in the
GL condition shifts participants towards dishonest reporting. In
the NE or PE conditions there is no signi?cant effect of infor-
mation system on the frequency of social choices (i.e.
respectively, þ4.52%; p ¼ 0.73 and À5.81%; p ¼ 0.95). Further
analyses show that in the information system absent condition,
the frequency of social behavior of 54.84% in the GL condition
differs from the respective frequencies of 38.06% and 32.90%
observed in the NE and PE conditions (GL vs. NE, c
2
(1,
N¼62)
¼ 2.71, p ¼ 0.10; GL vs. PE, c
2
(1, N¼62)
¼ 4.62, p ¼ 0.03). In
the condition where the information system is present, the dif-
ference between the GL vs. the PE condition and the difference
between GL vs. NE condition are no longer signi?cant
(p's > 0.45). Panel A of Table 3 further shows that a large part of
the social behavior in the GL condition, when information sys-
tems are absent, can be explained by altruistic choices in which
reported costs are below actual costs. Speci?cally, 30.32% of the
choices can be labeled as altruistic. This frequency differs from
the percentages of 9.03% and 4.52% respectively observed in the
NE and PE conditions (p's < 0.01). Analyses reported in footnote
indicate that participants in the GL condition produce these
altruistic reports to avoid losses and thus to keep the company
pro?table.
16
Table 4 uses the meanof three items froma post-questionnaire to
measure the participants' social concerns for the ?rm. Panel A of
Table 4 shows that in each of the three earnings conditions, infor-
mation systems reduce the social concerns for the ?rm. This is also
con?rmed by an ANOVA where only the main effect of information
system is signi?cant (F
(1, 180)
¼ 5.77; p ¼ 0.017). Results thus do not
provide support for a differential impact of information systems
across earnings conditions on the variable social concerns. Untabu-
lated results show that neither the interaction of Ex IS, nor contrast
tests of the interaction are signi?cant (p's > 0.32). Panel B of Table 4
shows, however, that the reductionof social concerns as a result of an
information systemis only signi?cant in the GL condition (p ¼ 0.03)
and not in the NE condition (p ¼0.47) or the PE condition (p ¼0.18).
This reduction in the GL condition may presumably arise because
participants worry less about reports which are bene?cial to ?rm
pro?t, once headquarters uses an information system.
17
Note that while these analyses provide some insights as to why
information systems produce a negative effect on the degree of
misrepresentation in the gain/loss condition, they do not explain
why information systems still produce positive effects in the pos-
itive earnings or the negative earnings conditions. Consistent with
Gneezy (2005) and Hannan et al. (2006), the next section presents
histograms of participants’ reporting choices to provide more in-
sights on this issue.
Table 3
Supplementary analyses of reporting choices.
Panel A: Summary statistics of choice behavior
Negative earnings (NE) Positive earnings (PE) Gain or losses (GL)
Info system absent Altruistic (a) 9.03% 4.52% 30.32%
Honest (b) 29.03% 28.39% 24.52%
Total social (a þ b) 38.06% 32.90% 54.84%
Total dishonest 61.94% 67.10% 45.16%
n ¼ 31 n ¼ 31 n ¼ 31
Info system present Altruistic (a) 18.06% 9.68% 20.65%
Honest (b) 24.52% 17.42% 14.84%
Total social (a þ b) 42.58% 27.10% 35.48%
Total Dishonest 57.42% 72.90% 64.52%
n ¼ 31 n ¼ 31 n ¼ 31
Panel B: Effect of information system per earnings condition (Y ¼ total social choices)
Effect info system þ4.52% ¡5.81% ¡19.35%
Chi-square (1, N ¼ 62) 0.120 0.003 4.388**
(p-value) (p ¼ 0.73) (p ¼ 0.95) (p ¼ 0.04)
***, **, * indicate signi?cance levels of respectively 1%, 5% and 10% level of signi?cance (two-tailed). I dummy coded each choice on participant level as either social (i.e. 1 if cost
equal to or below actual costs; 0 otherwise) and subsequently calculate for each participant the frequency of how many of the ?ve reporting choices can be labeled as social
(total social). The range of these variables on participant level is either 0%, 20%, 40%, 60%, 80% or 100%. The remainder (1Àtotal social) are the frequency of dishonest choices.
The table also constructed frequency variables on participant level based on choices classi?ed as altruistic choices (i.e. equal to one if reported cost < actual cost; 0 otherwise)
and honest choices (i.e. equal to one if reported cost equals actual cost; 0 otherwise), Panel A displays frequencies. Given that variables are based on frequencies, panel B shows
for each earnings condition, the results of a KruskaleWallis test for differences in distribution in social behavior across the factor information system.
16
I also calculate the conditional pro?t that headquarters would have realized if
participants were to report honestly and compare this to the pro?t headquarters
receives under the altruistic choice. In the scenario where the information system is
absent, the loss for the ?rm would be equal to À162.23 lira if the participant would
have reported honestly, but the ?rm realizes a gain of 40.43 lira because of the
altruistic reporting. Participants still earn positive earnings of 47.34 lira. In the in-
formation system present condition, altruistic reporting also ensures that the
company earned a pro?t of 2.34, rather than a loss of À116.41 if the participant
would have been honest. Yet, fewer of these choices occur and also less is trans-
ferred to the ?rm, because agents still keep on average 131.25 lira.
17
The variables total social choices of Table 3 and the level of social concerns of
Table 4 are positively correlated (r ¼ 0.48, p < 0.01).
E. Cardinaels / Accounting, Organizations and Society xxx (2015) 1e13 9
Please cite this article in press as: Cardinaels, E., Earnings benchmarks, information systems, and their impact on the degree of honesty in
managerial reporting, Accounting, Organizations and Society (2015),http://dx.doi.org/10.1016/j.aos.2015.09.002
4.3. Reporting behavior and impact of an information system
Fig. 1 uses the 155 reports per cell (i.e. 31 participants each
producing ?ve reports) to further explore howinformation systems
change the distribution of reports. Panel A of Fig. 1 shows the
number of cost reports that are either (1) honest or altruistic (i.e.,
misreporting that was equal or below zero); (2) dishonest but for
which the information system cannot detect misreporting (i.e.
0 < misrepresentation 0.25); or (3) dishonest to a level above the
detection limit of the information system (i.e., the degree of
misrepresentation> 0.25).
Consistent with the theory, panel A of Fig. 1 shows that in the
negative and positive earnings conditions, presence of an infor-
mation system (compared to absence of it) reduce the participant's
range of dishonesty. The ?gures show a shift from the interval
involving large dishonest reporting to the interval of dishonest
reporting which cannot be detected by the information system, for
which participants may still be able to maintain a positive self-
concept. The test for differences in distribution of reports across
the factor information system is signi?cant both for the negative
earnings condition (c
2
(1, N¼310)
¼ 8.46, p < 0.01) as well as the
positive earnings condition (c
2
(1, N¼310)
¼ 4.79, p < 0.05).
Conversely, in the gain/loss condition, participants reduce honest
and altruistic reporting (as is evidenced in Table 3), and shift to
more dishonest reports which are below the detection limit of the
information system. The test for differences in distribution of re-
ports across the three intervals for the factor information system is
signi?cant (c
2
(1, N¼310)
¼ 3.402, p
This paper provides experimental evidence about how the interaction between a company's earnings and
its information system influences the degree of honest reporting by managers in a capital budgeting task.
Specifically, the results show that participants overstate cost less when the manager's cost report determines
whether the firm earns a gain or loss than when their report does not affect whether the firm
earns a profit or loss (i.e., the firm always earns either a profit or loss regardless of the cost report). Further,
the results suggest that the impact of the earnings situation on the degree of honesty depends on whether
the firm uses an information system that improves its ability to detect misreporting. Specifically, the
earnings situation has less effect on the degree of honesty when the firm uses an information system. This
is because the information system decreases honesty when the manager's report determines whether the
firm earns a profit or loss but increases it otherwise. This study provides important insights into the
conditions under which information systems can crowd out prosocial behavior.
Earnings benchmarks, information systems, and their impact on the
degree of honesty in managerial reporting
*
Eddy Cardinaels
a, b, *
a
KU Leuven, Department of Accountancy, Finance and Insurance, Naamsestraat 69, B-3000 Leuven, Belgium
b
Tilburg University, Department of Accountancy, P.O. Box 90153, 5000 LE Tilburg, The Netherlands
a r t i c l e i n f o
Article history:
Received 16 May 2012
Received in revised form
9 September 2015
Accepted 10 September 2015
Available online xxx
Keywords:
Honesty
Capital budgeting
Earnings
Information system
Prosocial behavior
a b s t r a c t
This paper provides experimental evidence about howthe interaction between a company's earnings and
its information system in?uences the degree of honest reporting by managers in a capital budgeting task.
Speci?cally, the results show that participants overstate cost less when the manager's cost report de-
termines whether the ?rm earns a gain or loss than when their report does not affect whether the ?rm
earns a pro?t or loss (i.e., the ?rmalways earns either a pro?t or loss regardless of the cost report). Further,
the results suggest that the impact of the earnings situation on the degree of honesty depends on whether
the ?rm uses an information system that improves its ability to detect misreporting. Speci?cally, the
earnings situation has less effect on the degree of honesty when the ?rmuses an information system. This
is because the information systemdecreases honesty when the manager's report determines whether the
?rm earns a pro?t or loss but increases it otherwise. This study provides important insights into the
conditions under which information systems can crowd out prosocial behavior.
© 2015 Elsevier Ltd. All rights reserved.
1. Introduction
Research has demonstrated that in settings where managers are
able to misrepresent cost reports, many managers still produce
honest reports because their dishonest reporting may negatively
affect the wealth of others (Evans, Hannan, Krishnan, & Moser,
2001). More evidence is however, needed on when managers
more strongly pursue this motive to act honestly. This study shows
that the company's earnings situation can serve as an important
contextual feature. Speci?cally, I presume that the degree of
honesty is higher when the manager's cost report determines
whether the ?rm earns a gain or a loss than when the manager's
report does not affect the ?rm's earnings situation. Studying the
effect of managerial in?uence on company's earnings is important.
While many studies on earnings management have explored the
impact of important earnings benchmarks on external reporting
(Burgstahler & Dichev, 1997), the effects of these benchmarks on
internal decisions like budgeting have received scant attention.
The lack of attention in prior studies is partially based on their
focus on companies that earn pro?ts from a manager's production
(e.g., Rankin, Schwartz, & Young, 2003, 2008). I argue that consid-
ering the ?rm's pro?t situationcanenrichour understanding of why
information systems to detect misreporting are sometimes not
effective (Christ, Emett, Summers, & Wood, 2012; Salterio & Webb,
2006). I predict that information systems and the earnings of the
?rminteract such that the bene?cial effects on honesty of the ?rm's
earnings situation are mitigated once an information system is
present. I use a capital budgeting task to test this prediction. The
?rm's earnings situationis manipulatedas the ?rst between-subject
factor. In the gain/loss condition, the participant's cost report can
determine whether the company earns a gain or a loss. In the two
other conditions, labeled as the positive earnings condition and the
negative earnings condition, the participant's cost report cannot
affect the ?rm's pro?t situation; that is, the ?rm always earns a
pro?t or a loss regardless of the manager's cost report. The second
*
The author would like to thank the editor Mike Shields and two anonymous
reviewers for their helpful comments. I would further like to thank Nathalie Beckers,
Jan Bouwens, Bart Dierynck, Stefan Hollander, Laurence van Lent, Donald Moser,
Bernhard Reichert, Jeroen Suijs, Kristy Towry, Michael Williamson, Andrew Yim,
Huaxiang Yin, and Rick Young for their helpful suggestions. I further acknowledge
the useful comments of participants at the European Accounting Association Con-
ference 2010 (Istanbul) and the Management Accounting Section conference 2011
(Atlanta) and seminar participants at Tilburg University, VU University of Amster-
dam, KU Leuven, and the University of Bern. This paper was formally titled “Refer-
ence Points and Budget Requests: Can Controls Destroy Honesty in Managerial
Reporting?” Experimental materials available on authors request.
* KU Leuven, Department of Accountancy, Finance and Insurance, Naamsestraat
69, B-3000 Leuven, Belgium.
E-mail address: [email protected].
Contents lists available at ScienceDirect
Accounting, Organizations and Society
j ournal homepage: www. el sevi er. com/ l ocat e/ aoshttp://dx.doi.org/10.1016/j.aos.2015.09.002
0361-3682/© 2015 Elsevier Ltd. All rights reserved.
Accounting, Organizations and Society xxx (2015) 1e13
Please cite this article in press as: Cardinaels, E., Earnings benchmarks, information systems, and their impact on the degree of honesty in
managerial reporting, Accounting, Organizations and Society (2015),http://dx.doi.org/10.1016/j.aos.2015.09.002
between-subject factor manipulates the absence or presence of
information systems that improve the ?rm's ability to detect mis-
reporting. Across all conditions, the pecuniary bene?ts of dishon-
esty for participants (and its costs for the ?rm) are held constant.
In the absence of an information system, the results show
that the degree of honesty is greater when the participant's cost
report can determine whether the ?rm earns a loss or pro?t (the
gain/loss condition) than in the positive or negative earnings con-
dition. In the gain/loss condition, a larger fraction of participants
remain honest or even underreport their costs and thus sacri?ce
money to avoid losses and keep the company pro?table. The
reporting feature of deliberately understating costs has not
received much attention but can be economically relevant (Erat &
Gneezy, 2012). Results further show that reporting behavior does
not differ between the positive and negative earnings conditions.
Similar to earlier work (Evans et al., 2001; Hannan, Rankin, &
Towry, 2006), many managers produce partially honest reports in
these conditions. The results also show that the effect of the ?rm's
earning situation interacts with the use of an information system
that improves the ?rm's ability to detect suspicious reporting.
Speci?cally, the information system decreases the degree of
honesty when the manager's report determines whether the ?rm
earns a pro?t or loss but increases it otherwise (i.e., the positive or
negative earnings condition).
This study provides important insights into the conditions under
which controls can reduce prosocial behavior. Prior work has shown
that reliance on control systems to reduce misreporting may crowd
out some of the preferences for honesty but nonetheless has shown
that the ?rm's pro?t is still higher with a control system than
without one. Rankin et al. (2008) showed that opportunities for
principals to reject budget requests reduce the level of misreporting
and thus are bene?cial to the ?rm's pro?t. Hannan et al. (2006) also
showed that when information systems are used to improve the
?rm's ability to detect misreporting, honesty is increased compared
to when they are not used. This study, however, shows that the
otherwise positive effects of information systems may not always
materialize. It also offers a rationale for this crowding-out effect
using the self-concept maintenance theory of Mazar, Amir, and
Ariely (2008). In the gain/loss condition participants start to pro-
duce small dishonest reports, which the ?rm cannot detect as
misreporting, once an information system is present instead of
reporting behavior that could be bene?cial to ?rm pro?t. If their
reporting behavior has less in?uence on the ?rm's earnings situa-
tion, such as in the negative or positive earnings conditions, infor-
mation systems to detect suspicious reporting still tend to reduce
misreporting by reducing the range of dishonest reports.
The ?ndings concerning these earnings benchmarks also offer
many practical insights. Strong variations in earnings are often
caused by changes in the economic condition, such as temporary
?uctuations in prices or pro?t margins, or by the type of product
that the business unit produces. For example, business units pro-
ducing new products or products with spillover effects on other
products are often close to breakeven.
1
If companies feel that the
manager can make a difference between experiencing losses or
pro?ts, it may be tempting for them to install systems that help the
company to detect misreporting. Such detective forms of controls
are often part of the ?rm's internal control procedures (Christ et al.,
2012), and resorting to such systems maybe a natural response in
case pro?ts start to erode. The results show, however, that infor-
mation systems to detect misreporting can be less effective, in
particular when pro?ts are under pressure.
Besides earnings levels, many other situations in a company
may alter participants' views of the repercussions of their mis-
reporting for the organization. Business units need to achieve
certain targets before bonus pools are paid out to employees, ?rms
need to meet or beat analyst expectations, and certain actions can
hurt only a few but also many other business units. Prior work by
Church, Hannan, and Kuang (2012) showed for example that people
report more dishonest and thus care less about the ?rm when
bene?ts of misreporting are shared with other managers in the
company. This paper shows that considering these repercussions of
managerial dishonesty on the ?rm is important, as doing so may
help organizations to utilize their controls more effectively.
2. Theory and hypothesis development
Evans et al. (2001) showed that many agents in capital budg-
eting produce partially honest reports even when ?nancial in-
centives for misreporting are fully present. Based on the ?nding
that individuals value honesty, follow-up studies have focused on
incentive mechanisms, monitoring systems, or other types of con-
trol systems that can help companies to improve honest reporting.
For example, prior studies have investigated competition among
agents or whistle-blowing by fellow agents in relation to honesty
(Brüggen & Luft, 2011; Zhang, 2008) or examined the effects of
social norms or peer behavior (Tayler & Bloom?eld, 2011;
Cardinaels & Jia, 2015). Other studies examine changes in eco-
nomic incentives, opportunities for principals to reject budget
proposals, or systems to reduce information asymmetry between
the business owner and the agent in relation to truthful reporting
(Evans et al., 2001; Hannan et al., 2006; Rankin et al., 2008).
However, fewer studies have focused on the organizational
settings in which managers more strongly pursue honest reporting
without touching upon costly incentive devices or control systems.
An exception is, for example, Church et al. (2012), who documented
that people are more honest when they fully bear the consequences
of their dishonesty than when the bene?ts of their dishonesty are
shared with other organizational members. This paper examines if
the ?rm's earnings situation can also serve as an important
contextual factor which may affect reporting behavior by man-
agers. Considering this variation may offer additional insights into
the crowding-out effects of information systems to detect mis-
reporting. The ?rst section will argue that participants overstate
costs less when their reports can make a difference between gains
or losses for the ?rm than in two other conditions where the
company always realizes either positive or negative earnings
regardless of the cost report. Next, I will discuss how the ?rm's
pro?t situation interacts with information systems that companies
use to detect misreporting.
2.1. The ?rm's earnings situation and the effect on honesty
Given the information asymmetry that exists between the agent
and the principal in a capital budgeting context, agency theory
would predict that the company's earning situation would not
matter because agents will always try to pro?t from dishonesty.
This study predicts that the degree of honesty - measured by the
level of cost overstatements - is higher when participants' cost
1
New products like, for example, new generations of smart phones are often not
pro?table. Nevertheless, once demand increases and learning takes place, pro?ts
start to accrue. Products with spillover effects are, for example, ink-jet printers.
Typically these printers are sold for a small loss, but business units producing the
cartridges that are used with them make a pro?t. Because ?rms often commit to a
customer base, they sometimes need to accept small losses when prices are under
pressure. On average, such ?rms expect to be pro?table by serving their customer
base, but temporary price ?uctuations can lead to losses, pro?ts, or pro?ts that are
close to zero. From a decision-control perspective (Zimmerman, 2009), various
types of cost allocations and transfer-pricing policies may further lead to differ-
ences in the division's contribution to organizational pro?ts.
E. Cardinaels / Accounting, Organizations and Society xxx (2015) 1e13 2
Please cite this article in press as: Cardinaels, E., Earnings benchmarks, information systems, and their impact on the degree of honesty in
managerial reporting, Accounting, Organizations and Society (2015),http://dx.doi.org/10.1016/j.aos.2015.09.002
reports can determine whether the company incurs a gain or a loss
than when participants' reporting behavior does not change the
company's earnings situation.
2
The arguments are two-fold.
The ?rst argument rests on the theory of self-concept mainte-
nance by Mazar et al. (2008). This theory argues that individuals try
to derive some bene?ts of cheating while at the same time trying to
maintain their positive self-concept of being an honest person. In
other words, there is a potential range of dishonesty in which
people will act dishonestly but their behaviors do not bear nega-
tively on their self-concept. Mazar et al., however, hinted at the fact
that this range may decrease and people may thus be shifted to-
ward more honest behavior if the context makes it dif?cult for
them to interpret a dishonest act as still being consistent with this
positive self-concept. I predict that the range of dishonest reports
that can be made without altering the individual's self-concept is
smaller in the gain/loss condition than in the two other conditions.
In the negative earnings condition or the positive earnings
condition, a dishonest report will negatively affect the other party's
wealth, but it however, does not really change the ?rm's pro?t
situationdthat is, the ?rm will still either earn a pro?t or incur a
loss regardless of the manager's report. This may offer room for
participants to produce a larger range of partially dishonest reports,
which would still be seen as consistent with the self-concept of
being an honest person. In the situation where participants'
reporting behavior can produce gains or losses for the ?rm, the
agent, however, may observe that a dishonest act, sometimes even
a small one, can turn pro?ts into losses for the ?rm. When the ?rm
would earn a pro?t for an actual cost draw, agents may avoid the
dishonest choice of overstating costs because it may turn this pro?t
into a loss. Conversely, if the company incurs a loss for an actual
cost draw, agents can turn this loss into a pro?t by sacri?cing part of
their resources through submitting a cost that is less than the actual
cost. Turning pro?ts into losses through lying is dif?cult for agents
to reconcile with a positive self-concept, while making small sac-
ri?ces to make the ?rm pro?table can reinforce participant's self-
concept of an honest person. Hence, the potential range of
dishonest reports that can be reconciled with a positive self-
concept is smaller in the gain/loss condition, and may shift par-
ticipants toward prosocial behavior. Choices to underreport also
occurred more frequently in the experiment of Evans et al. (2001),
where participants received a hurdle contract.
3
This paper argues
that such behavior can become relevant (Erat & Gneezy, 2012)
when participants perceive their reporting behavior to have
stronger repercussions on the company's pro?t.
A second argument, leading to the same prediction, is that
people may be intrinsically motivated to avoid losses for other ac-
tors. Indirect evidence from strategic games on cooperation
(Cachon &Camerer, 1996; Drake &Haka, 2008; Feltovich, Iwasaki &
Oda, 2012) suggests that people behave differently when either one
or more parties in the game can experience a loss than when all
parties would only be able to realize positive pay-offs.
4
Many
agents then utilize socially rational strategies to avoid losses and
expect other people to also avoid such points (Cachon & Camerer,
1996). In the gain/loss condition, participants observe that report-
ing behavior can make a difference between losses of pro?ts for the
other party. Individuals’ own reactions to losses are often severe
and most individuals would try to avoid them if possible. While in
the setting, individuals do not experience the loss themselves, they
nevertheless may still adapt their behavior vicariously (Welten,
Zeelenberg, & Breugelmans, 2012; Wood & Bandura, 1989). That
is, because they observe the consequences of dishonesty on the
other party, people may refrain from dishonest reports that push
the other party into a loss. This is labeled as vicarious loss avoidance
behavior, which can only take place in the gain/loss condition,
where participants can avoid losses for the ?rm.
2.2. Different earnings situation, information systems and their
impact on honesty
Consistent with Hannan et al. (2006) and Schulze and Frank
(2003), the information system provides the company with a nar-
rower range of potential costs and thus enables the company to
detect suspicious reporting if participants report a higher cost than
the upper limit of the interval.
5
Consistent with Hannan et al.
(2006, p. 896), the information system also does not introduce a
?nancial punishment for misreporting in order to ensure that in-
centives to misrepresent costs are equivalent across all conditions.
Prior work on capital budgeting has shown that people's pref-
erences for honesty may diminish as a result of control systems.
Rankin et al. (2008) showed that the participant's preferences for
honesty are lower when principals have the authority to reject their
budget requests. However, organizations still bene?t from this au-
thority because participants report more truthfully when rejection
opportunities are present compared to when they are not. Similarly,
Hannan et al. (2006) showed that the use of an information system
that increases the ?rm's ability to detect misreporting is not
necessarily bad news. When ?rms install information systems to
reduce the information asymmetry, people can still give the
impression of an honest appearance. In order to do so, they reduce
their misreporting to a level that will not be detected by the infor-
mation system. Overall, an information system still reduces
2
Prior work has indirectly shown that earnings may in?uence prosocial behavior.
In the study of Hannan (2005), agents act reciprocally toward higher wage offers of
the principal by voluntarily delivering more than an optimal effort when earnings
for the ?rm decreased (as, in such a case, higher wages are seen as a generous act)
compared to when earnings increased. Although the ?rm is always pro?table in this
study, agents care about random shocks in the principal's pro?t. An unanticipated
?nding in Evans et al. (2001, p. 547, fn. 11) also hinted at the fact that agents behave
differently depending on the context. Within the hurdle contract game, a setting in
which production would not take place if costs are too high, a sizeable number of
participants voluntarily sacri?ced money by submitting reports for costs less than
the actual costs. Participants may have had a social motive to understate the costs
to ensure production, so that principals would earn some pro?t.
3
In Evans et al. (2001, p. 547, fn. 11), 34 out of 138 agents (24.6 percent) delib-
erately understate costs at the expense of their ?xed wage when actual costs were
above the production hurdle. In this way, production took place and the company
would still make some money. Evans et al. (2001, p. 547, fn.11), also argue that
subjects knowledgeably sacri?ce part of their wealth to increase ?rm pro?t, to
ensure that production would take place.
4
These strategic games use simple transformations to the players' pay-off matrix
(which are irrelevant from a game-theoretic perspective) and show that behavior is
different when such losses come into play. While the capital budgeting game in-
volves little strategic interactions as the agent always decide about the report, the
manipulation of the ?rm's earnings situation resembles the manipulations
observed in these more strategic games. By manipulating the product's contribu-
tion margin, I focus on simple transformations, which would not change the eco-
nomic self-interested outcome. Yet, the manipulations allow to differentiate
between settings in which participant's cost report can make a difference between
losses or pro?ts for the company and where losses can thus be avoided and settings
in which participant's report has less in?uence on company pro?ts (i.e. positive or
negative earnings conditions).
5
The information system is a form of detective control offering a warning for
suspicious reporting (Christ et al. 2012). Section 3 describes the manipulation in
detail. In short, when an information system is absent, participants made ?ve
reporting decisions under full information asymmetry. When an information sys-
tem is present, the company can examine two reporting decisions for which the
?rm receives a narrower interval of potential costs. If reported cost are higher than
the maximum cost of the interval, participants get a warning and they need to
submit a new cost report to the ?rm on the basis of a new cost draw. Regardless of
whether an information system is present or absent, all participants thus play ?ve
full rounds with information asymmetry, for which company earnings and their
pay-offs are calculated. In each condition, the maximum pay-off is achieved when
people lie to the full extent.
E. Cardinaels / Accounting, Organizations and Society xxx (2015) 1e13 3
Please cite this article in press as: Cardinaels, E., Earnings benchmarks, information systems, and their impact on the degree of honesty in
managerial reporting, Accounting, Organizations and Society (2015),http://dx.doi.org/10.1016/j.aos.2015.09.002
misrepresentation compared to not having one at all. Participants
avoid large cost overstatements under both a coarse and precise
information system, although a precise information system crowds
out some of the motives to appear honest (Hannan et al., 2006).
In contrast to these prior studies, I presume that the impact of an
information system, as a detective form of control, may depend on
the earnings situation. That is, compared to the negative earnings or
the positive earnings conditions, where information systems may
still reduce misreporting, the same systemmay have aweaker impact
or potentially increase misreporting if agents’ reporting behavior can
make a difference between losses or pro?ts for the company.
Experimental economists argue that detective forms of control
can have a ?attening effect on dishonesty (Schulze & Frank, 2003).
Even though they may crowd out some of the motives to act pro-
social, sel?sh behavior is also reduced. Prior work by Hannan et al.
(2006) showed that people are willing to give up some bene?ts of
misreporting and thus report more honestly under the presence of
systems that allow the company to make inferences regarding the
manager's honesty level. Using self-concept maintenance theory
(Mazar et al., 2008), I argue that such bene?ts are more likely to be
realized in the negative or positive earnings situation. As previously
argued, participants in these earnings situations may have a larger
range of dishonest reports for which they can maintain their self-
concept as being an honest person. Information systems that
allowthe ?rmto make some inferences about the honesty level of a
participant may reduce this range. That is, to preserve their positive
self-concept as an honest person, participants may reduce their
misreporting and thus give up some bene?ts of misreporting by
submitting cost reports that the information system cannot detect
as instances of misreporting (Hannan et al., 2006; Schulze & Frank,
2003). By reducing their misreporting, participants can still appear
to be honest to the company, which would be positively rewarded
in their personal value system (Mazar et al., 2008).
The same information system may work differently when a
participant's cost report is able to determine whether the company
earns a pro?t or loss (gain/loss condition). As mentioned, self-
concept maintenance theory would argue that the fact that par-
ticipant's reporting behavior can make the difference between
gains or losses for the company may already reduce misreporting
and may potentially shift a participant's reporting towards more
honest reporting behavior.
When an information systemis present, participants, receive, as
a result of the information system, an opportunity to reconcile
dishonest reports that can lead to a loss for the company or that
may prevent the company from earning pro?ts, as still consistent
with their positive self-concept. In particular, the detrimental effect
of dishonest reports that would put the ?rm into a loss on the
positive self-concept may be less signi?cant when a company
would use an information system (Matsushima, 2008, p. 354;
Charness & Dufwenberg, 2006). The use of an information system
can signal to the participant that the company detects and does not
accept large cost overstatements. Nevertheless, as long as a person
misreports by small amounts, which the information system
cannot detect, people may reconcile their partially dishonest report
as ?ne. The system allows participants to maintain a positive self-
concept by appearing honest, even if such reports may not be
bene?cial to ?rm pro?t. As a result, in settings where the reporting
behavior can make a difference to ?rm pro?ts more partially
dishonest reports may occur at the expense of honest reporting or
choices to underreport cost to make the ?rm pro?table. While in-
formation systems, as argued before, may reduce high levels of
misreporting, such information systems may also justify small
dishonest reports. In contexts, where earnings can vary between
losses or pro?ts for the ?rm, this can weaken the effect of an in-
formation system on misreporting.
Based on arguments of Tenbrunsel and Messick (1999), weak
sanction systems, such as a detective information system, may also
change the decision frame froma setting where people would worry
about ethical concerns to a setting where people would worry more
about ways to avoidpenalty. Without suchsystems inplace, stronger
ethical concerns for the ?rm may arise in particular when people's
behavior can make a difference between pro?ts and losses for a
company (e.g. people avoid losses vicariously), compared to when
people cannot change the earnings situation. While an information
system may reduce the range of misreporting because persons do
want to be detected as a dishonest person, people in the gain/loss
condition can also refrain from reports that may otherwise be
bene?cial for the company. People may submit partially dishonest
reports that the ?rm cannot detect, and may worry less about con-
sequences of their reports for the company's earnings. The concerns
of a person to avoidlosses for the company by reporting more honest
can play a weaker role once an information system is present.
Both arguments would predict an interaction whereby the
earnings situation has less in?uence on misreporting once an in-
formation system is present. Speci?cally, while an information
systemmay reduce large cost misrepresentations in the negative or
positive earnings conditions, the same information system in the
gain/loss condition may reduce the reporting of choices which can
be bene?cial to ?rm pro?t. Hence, the net effect is that an infor-
mation system has a less positive (or even a negative) impact on
honesty in the gain/loss condition relative to the two other
conditions.
Hypothesis. The effect of an information system on honest
reporting is less positive (or potentially negative) when the man-
ager’s cost report determines whether the ?rmearns a gain or a loss
than when the ?rm always realizes positive or negative earnings.
3. Experiment
3.1. Participants
A total of 186 participants from an accounting course in a
business studies program at a large West-European university,
participated in the study as part of the course requirement across
two different administrations of the experiment. Most of the stu-
dents were undergraduates in the last year of their bachelor degree
(96.8%), although a minority were masters-level students (3.2%).
Participants’ average age was 21.23 years and 59.14% were male.
They had on average taken 2.28 accounting courses and 74.7% of the
students reported to have work experience (Their average part-
time work experience was 19.4 months).
3.2. Experimental task
I used a capital budgeting task, similar to that of Evans et al.
(2001). Participants assumed the role of a production manager
(i.e., the agent) and worked for one of the divisions of Acazia
headquarters (i.e., the principal).
6
They received a ?xed wage of
6
For reasons of task realism, I used a hypothetical company where consequences
of dishonesty come at the cost of the experimenter (See also Evans et al. 2001,
Evans, Moser, Newman, & Stikeleather, 2015). Experimental instructions
mentioned that participants would work for one of the divisions of corporate
headquarters. Depending on the earnings manipulation, headquarters would either
always receive a loss, always receive a pro?t, or receive either a gain or a loss
depending on the participant's cost report. This earnings manipulation is likely to
have greater external validity when a common company is involved. Second, such a
design gives dishonesty a strong change to survive. Any deviation against this
agency prediction can then be attributed to a social motive to be honest.
E. Cardinaels / Accounting, Organizations and Society xxx (2015) 1e13 4
Please cite this article in press as: Cardinaels, E., Earnings benchmarks, information systems, and their impact on the degree of honesty in
managerial reporting, Accounting, Organizations and Society (2015),http://dx.doi.org/10.1016/j.aos.2015.09.002
250 lira (50 lira ¼ 1 euro). Participants then submitted budget
requests to headquarters to receive funding for the production of a
lot of 500 units in each period. At the start of each period, both
headquarters and the participants knew that the per-unit pro-
duction cost was randomly drawn from a uniformly distributed
cost interval between 4.00 and 6.00 lira, with increments of 0.05
(e.g., 4.00, 4.05, …, 5.95, 6.00). Before submitting a cost report to
headquarters, participants, however, received feedback from a
private forecasting system, which offered them an accurate fore-
cast of the actual production cost for the production period.
Headquarters, however, would never learn the actual costs and
thus can never assess if participants report honestly. Participants
then submitted a cost report to headquarters. If participants re-
ported a cost higher than the actual cost, they could keep the
excess resources in addition to their ?xed wage. In sum, a par-
ticipant's earnings are equal to:
(1) Participant's earnings ¼ 250 lira þ 500 lira * (reported
cost e actual cost)
Participants were told that the company earns a contribution on
the output produced in each production period (i.e., the contribu-
tion margin will be manipulated between subjects; see next sec-
tion). In sum, headquarters’ earnings were calculated as follows:
(2) Headquarters' earnings ¼ 500 lira * (contribution per
unit e reported cost) e 250 lira
Instructions also emphasized to participants that earnings are
very important for headquarters and that higher (lower) cost
reports would lead to lower (higher) earnings for the ?rm. If
participants report less honest, their own payoffs would increase,
but earnings for headquarters would decrease. To make this
trade-off clear, the instructions given to participants provided
examples of how budget requests affect their own earnings as
well as their contribution to the ?rm's pro?t. In addition, par-
ticipants received a payoff table based on an actual cost of 4.50
lira, showing all possible outcomes of headquarters' earnings and
their own payoffs for all the potential cost reports they could
submit, including all the combinations that were for costs less
than actual costs. Participants then played ?ve production pe-
riods. Before submitting their cost report, participants again
received a full payoff table containing the pay-off and earnings
consequences for the full range of possible cost reports for the
respective actual cost draw, including also all combinations for
less than actual costs.
3.3. Between-subjects manipulations
The study employs a 3 Â2 between-subjects design. As a ?rst
factor, I manipulated three different earnings situations. One-third
of the participants were told that headquarters realizes a 4.00 lira
contribution margin on each production unit. This scenario is
labeled as the negative earnings (NE) condition because head-
quarters always realizes a loss. The loss can vary between e1250
lira and À250 lira depending on the submitted cost report. Another
third of the participants were told that headquarters receives a
contribution margin of 7.00 lira. This scenario is labeled as the
positive earnings (PE) condition. In this scenario, headquarters will
always realize a pro?t that can vary between 250 lira and 1250 lira.
The rest of the participants were given the gain/loss (GL) condition
with a contribution margin of 5.50 lira. Compared to the conditions
where the ?rmwill either earn a loss (NE) or a pro?t (PE) regardless
of the manager's report, the GL condition is the only situation
where the manager's report can determine whether the company
earns a loss or a pro?t. Earnings can vary fromÀ500 lira to 500 lira
and become negative when the reported cost is above 5.00 lira and
positive when the reported cost is below 5.00 lira. In all cells,
participants are also told that headquarters do not face any risk of
failure.
7
From a self-interested perspective, the ?rm's earnings
situation would not matter. According to formula (1), participants
will maximize their payoff when costs are fully overstated (6.00
lira). Consistent with the theory, however, I presume that when the
report can determine whether the company earns a gain or a loss,
dishonest reporting may be more dif?cult to reconcile with the
participant's self-concept.
As a second between-subject manipulation, the study manipu-
lated the presence or absence of an information system. When the
information system was present, the company had some ability to
make an inference about suspicious reporting of the participant. In
this condition, participants were told that headquarters would
check reports in two out of ?ve production rounds (no more, no
less). Consistent with Hannan et al. (2006), headquarters would
merely receive a narrower range of potential costs if the report was
examined. Participants were told production would be ?nanced
(suspended) if the cost falls within (outside) the range. Participants,
however, received private information. They knew that the com-
pany's information systemhad the ability to reduce the information
asymmetry with regard to potential cost to a range of 0.5 lira
around the actual cost draw. For example, if a participant received
an actual cost of 4.50 lira, headquarters' information systemwould
display that the cost should be in the range of 4.25e4.75 lira rather
than in the range of 4.00e6.00 lira. In the production runs, pro-
duction was ?nanced on the basis of reported costs for all costs
below the maximum cost of this interval. If a participant submitted
a report higher than the maximumcost, headquarters would send a
warning message to the participant indicating that the report is
suspicious and that production will be suspended. Participants
would then receive a newcost draw for that production period and
they again would need to submit a cost report on which basis
production would be ?nanced.
Similar to the information absent condition, participants in the
information system present condition would thus always play ?ve
full production runs for which their own payoffs and their contri-
bution to the ?rm's earnings are calculated.
8
This was done to
ensure economic comparability across all six cells. For the same
reason, I do not attach a ?nancial consequence to suspicious
reporting. This approach is consistent with Hannan et al. (2006)
and allows the researcher to attribute deviations from full
dishonesty to the intrinsic motive of participants to report honestly.
These information systems are labeled as detective forms of control
(Christ et al., 2012). They are often part of a company's internal
control procedures in which reports by managers are (randomly)
checked and not accepted if the manager's report is perceived as
suspicious. Such systems often also provide a warning if the report
is considered to be suspicious (Christ et al., 2012).
3.4. Experimental procedures
Participants were randomly assigned to the experimental cells
7
To ensure that the NE or the GL condition are not perceived differently than the
PE condition in terms of risk, the case states the ?rm does not experience any
failure risk and that it has enough resources available from past activities.
8
In the experiment, headquarters checks the initial budget request of the second
and fourth production run. Only when reported costs are higher than the upper
limit, headquarters will suspend production. Participants are not aware of this.
Similar to in the information absent condition, the agency prediction in the infor-
mation system present condition is again to lie to the maximum amount possible,
in which case earnings for the participant are maximized.
E. Cardinaels / Accounting, Organizations and Society xxx (2015) 1e13 5
Please cite this article in press as: Cardinaels, E., Earnings benchmarks, information systems, and their impact on the degree of honesty in
managerial reporting, Accounting, Organizations and Society (2015),http://dx.doi.org/10.1016/j.aos.2015.09.002
when entering the computer room.
9
Sessions lasted an hour.
Following Evans et al. (2001), participation was anonymous. To
ensure full anonymity, students drew a ticket number, on which
basis they could claim their payoff. Before starting the task, they
provided background information on age, gender, work experience,
the number of accounting courses they followed, etc. Then, they
completed a nine-item instrument, which measured their social
value orientation(see VanLange, Bekkers, Schuyt, &vanVugt, 2007).
This instrument classi?es people into “competitive,” “individual,”
and “prosocial” types of players and can pick up some of the innate
preferences of people to act honestly. Prosocial people may care
more about another person's wealth.
10
Next, participants played a
distracter task and continued into the main task. Before making
reporting decisions, they carefully read the case descriptions. Par-
ticipants performed a true-false quiz about the case descriptions. To
increase task understanding, participants always received feedback
on the right answer for each question, before moving on to the next
question. Participants then played ?ve production runs, in which
actual cost was randomly drawn by the computer from the uniform
distribution of costs (4.00, 4.05, 4.10, …, 5.95, 6.00 lira).
Consistent with Evans et al. (2001) and Hannan et al. (2006),
participants knew that the earnings of one production period
would be randomly selected and converted into real cash at the end
of the experiment (50 lira ¼ 1 euro). After participants ?nished the
task, they ?lled out an exit questionnaire containing items on task
understanding, motivation, etc. and some manipulation checks. The
cash reward was collected two weeks after the last session. On
average, participants earned 7.69 euro (the minimum was 3 euro,
and the maximum, 25 euro).
11
3.5. Manipulation checks
Participants' task understanding and task motivation did not
vary across the manipulations.
12
Most participants (95.7%) under-
stood that their participation was fully anonymous. In total 84.4% of
the participants further agreed with the item “There is a clear
trade-off. The more I misrepresent costs, the lower the earnings for
corporate headquarters.” Importantly, I did not detect differences of
the manipulations on the anonymity question nor on the trade-off
question (all p's > 0.15, F-model ns). To check if results depend on
attendance to these manipulation checks, Section 4.1 also reports
the results for the hypothesis test using a reduced sample of par-
ticipants who agreed with both these items (n ¼ 150, or 80.6
percent of the sample).
13
Participants' perceptions of the ?rm's contribution margin were
directionally consistent with the earnings manipulation. One
seven-point Likert scale item (1 ¼ completely disagree,
7 ¼ completely agree) asked participants whether they felt that the
company earned a too low contribution. Only the main factor
earnings was signi?cant (F
(2, 180)
¼ 29.12, p < 0.01). Participants in
the NE condition achieved a higher score on this question
(M ¼ 5.74, SD ¼ 1.38) than participants in the GL condition
(M ¼ 4.60, SD ¼ 1.37) and participants in the PE condition
(M¼ 3.94, SD ¼ 1.22). The contrast differences between the NE and
GL conditions (F
(1, 180)
¼ 22.86, p < 0.01); the NE and PE conditions
(F
(1, 180)
¼ 56.88, p < 0.01); and the GL and PE conditions (F
(1,
180)
¼7.62, p < 0.01) were all signi?cant. The post questionnaire did
not have a manipulation check for the between-subject factor in-
formation system, as it was obviously either present or absent.
4. Results
4.1. Hypothesis test
Consistent with Brüggen and Luft (2011), I use the degree of
misrepresentation of the private cost signal as the dependent var-
iable, which is de?ned as the reported cost minus the actual cost.
This measure perfectly correlates with the participant's earnings in
the experiment. I use it to avoid losing the information value of
choices where participants submitted reports for costs less than the
actual costs. The frequency of such reports varies strongly across
cells (see Section 4.2). Although Evans et al. (2001) view the choice
to underreport a cost as inconsistent and set it as equal to honest
reporting in their measure, such a choice can be driven by social
9
I checked if randomization was successful (n ¼ 186). I do not ?nd signi?cant
effects of our manipulations for the variables age, the percentage of pro-socials, the
percentage of people who report to have work experience, and the no. of ac-
counting courses. Only when considering gender as dependent variable, I ?nd a
signi?cant interaction of earnings * information system (F
(2, 180)
¼ 2.48, p ¼ 0.086).
The negative and positive earnings conditions contained fewer males when infor-
mation system was absent (48, 4% for NE and 54.8% for PE) compared to when it
was present (67.7% of males for NE and 61.3% for PE). For the GL condition data
shows an opposite pattern with 71.0% males in the IS absent condition and 51.6% in
the IS present condition. The study also used two administrations of the experiment
to increase sample size; 144 students (randomized across six cells) participated in
Nov. 2007 and an additional 42 students (randomized across six cell) participated in
Nov. 2009. Timing of administration did not have an impact on the level of mis-
reporting (M ¼ 0.214; SD: 0.373 (n ¼ 144) vs. M ¼ 0.305, SD: 0.354 (n ¼ 42),
t
184
¼ À1.40, p > 0.16, two-tailed) nor on the aforementioned demographic vari-
ables (All p's > 0.12 two-tailed). Except for gender results show that the ?rst
administration contained more males (62.50%) compared to the second (47.62%; t
184
¼ 1.73, p ¼ 0.085 two-tailed). Results are similar when Gender is added as
covariate in Table 1; the covariate Gender is not signi?cant.
10
The instrument is fully described in Van Lange et al. (2007). When applying this
instrument, a small number of people could not be classi?ed because they make
inconsistent choices (Van Lange et al., 2007). In this study, 37.1% of the participants
were classi?ed as prosocial, 39.2% as individualistic, and 10.8% as competitive. The
remainder of the participants (12.9%) could not be classi?ed. Importantly, the dis-
tribution of people across categories (prosocial, individualistic, competitive, and
unclassi?ed) did not differ across the six experimental cells (c
2
(5, N ¼ 186): 2.973,
p ¼ 0.704), suggesting that randomization was successful. Consistent with the
theory in Van Lange et al. (2007), correlations with the dependent variable show
that people who are classi?ed as prosocial (prosocial is 1; and zero otherwise)
misrepresent costs less (r ¼ À0.156, p ¼ 0.034). Results do not qualitative differ
when I use an ANCOVA, controlling for the dummy prosocial as covariate. The
covariate prosocial is signi?cant (F (1, 179) ¼ 3.92, p ¼ 0.05). The results and sig-
ni?cance levels in Tables 1 and 2 have similar magnitudes and signi?cance levels.
11
The maximum of 25 euro is achieved at a cost of 4 lira and a submitted report of
6 lira. Honest reporting results in a reward of 5 euro (¼250 lira of ?xed wage). By
reporting a cost below the actual cost, some participants sacri?ced part of their
?xed wage or more than their ?xed wage leading to a negative pay-off. If a round
was drawn for pay-out in which payoffs were lower than 3 euro, participants still
received 3 euros for participation. This was, however, not announced to them
beforehand.
12
Task understanding uses two seven-point Likert scale items in the post ques-
tionnaire (Cronbach's a ¼ 0.74). Results showed no signi?cant effects of the ma-
nipulations (all p's > 0.11, F-model ns). Task motivation also uses two seven-point
Likert-scale items (Cronbach's a ¼ 0.66). Again, the main effects and interactions
are not signi?cant (all p's > 0.26, F-model ns). Participants also indicate they had
suf?cient time to carry out the task (all p's > 0.36, F-model ns).
13
Note that the 36 people (who fail either one or both of these two items)
misreport cost less compared to the 150 participants, who agreed with both items
(mean level of misreporting M¼ 0.064, SD: 0.326 (n ¼ 36) vs. M¼ 0.275, SD: 0.368
(n ¼ 150), t
184
¼ 3.15, p < 0.01). Results are driven by the answer on the trade-off
question (correlation trade-off item (agree ¼ 1; 0 otherwise) with level of mis-
reporting r ¼ 0.297, p < 0.01). More honest participants are less likely to agree with
this item, because for them such a trade-off is less present. Demographic charac-
teristics are relatively similar between the two groups with respect to age, gender,
the percentage of pro-socials, and the no. of accounting courses (All p's > 0.16 two-
tailed) except for percentage of participants with work experience (M ¼ 86.1%
(n ¼ 36) vs. M ¼ 72.0% (n ¼ 150), t
184
¼ À1.75; p ¼ 0.081); Task understanding as
measured with two PEQ items also does not signi?cantly differ (M ¼ 5.32, SD: 1.35
(n ¼ 36) vs. M ¼ 5.54, SD: 1.21 (n ¼ 150), t
184
¼ 0.94, p > 0.34). The subsample
however, does exclude one participant (outlier) in the NE/IS present condition who
never realized a positive pay-off (i.e. he or she scari?ed the full ?xed wage or more
than the ?xed wage in all ?ve trials).
E. Cardinaels / Accounting, Organizations and Society xxx (2015) 1e13 6
Please cite this article in press as: Cardinaels, E., Earnings benchmarks, information systems, and their impact on the degree of honesty in
managerial reporting, Accounting, Organizations and Society (2015),http://dx.doi.org/10.1016/j.aos.2015.09.002
concerns. Participants may decide to underreport the cost to
improve the earnings of the company.
14
This measure thus also
better captures the consequences of participant's behavior to ?rm
pro?t.
Because there is less roomto misrepresent costs at high levels of
the actual cost, I also control for the actual cost in a second model
(see also Brüggen & Luft, 2011). Table 1 presents the results of the
full-factor ANOVA (Model 1) and a mixed model procedure (Model
2) with the actual cost as a time-varying covariate (i.e., 41 possible
values between 4.00, 4.05, 4.10, …, 5.95, and 6.00). In both models,
the reporting period is treated as a within-subject effect. Results
concerning the hypothesis test (i.e., Table 1) are based on two-
tailed tests. Comparisons of the simple effects of the interaction
using least-square means (i.e., Table 2) are based on one-tailed
directional tests.
Panel A of Table 1 shows the cell means and the conditional
means controlling for the actual cost. In all cells, participants on
average overstate costs (i.e., misrepresentation > 0) and thus
extract rents from the organization. Similar to Evans et al. (2001),
participants strongly deviate from the full self-interested choice of
reporting the maximum cost of 6 lira, in which case the degree of
misrepresentation would be close to 1 (because the expected value
of the actual cost is 5 lira). Importantly, the degree of misrepre-
sentation differs strongly across cells.
The hypothesis predicts an interaction between earnings and
information systems (E ÂIS), indicating that the effect of an infor-
mation system on the degree of honesty is less positive (and
potentially negative) in the settings where the cost report can
determine whether the ?rm earns a gain or a loss (GL condition)
compared to the two other conditions (NE or PE conditions). Panel B
of Table 1 shows that the interaction ExIS is signi?cant in both
models at the 10% level (p ¼ 0.054 in Model 1 and p ¼ 0.064 in
Model 2). The means in panel A of Table 1 indicate that both in the
negative earnings (NE) and positive earnings (PE) condition, infor-
mation systems reduce the degree of misrepresentation, whereas
information systems increase misrepresentation in the gain/loss
(GL) condition. The contrast analyses in panel B of Table 1 offer more
detail on this interactive pattern. The interaction is signi?cant when
comparing the GL condition against the NE condition (p ¼ 0.023 in
Model 1 and p ¼ 0.039 in Model 2), when comparing the GL con-
dition against the PE condition (p ¼ 0.063 in Model 1 and p ¼ 0.047
in Model 2); and when comparing the GL condition with the pooled
means of the PE and NE conditions (p ¼ 0.017 in Model 1 and
p ¼ 0.019 in Model 2). Information systems have a less positive ef-
fect (i.e. a negative effect) on honesty in the GL condition compared
to the two other conditions. The interaction is not signi?cant when
comparing the PE against the NE condition, because the information
system reduces misreporting in both cases.
15
Panel B of Table 2 explores the means (i.e. conditional means) by
comparing the simple effects of the interaction using the least
square means procedure. When the information system is absent,
the degree of misrepresentation is higher when comparing both
the negative earnings and positive earnings conditions against the
gain/loss condition (the NE vs. GL condition þ0.170, p ¼ 0.034 one
tailed, or þ0.154, p ¼ 0.047 one-tailed when considering condi-
tional means; and the PE vs. GL condition þ206, p ¼ 0.014 one
tailed, or þ0.234, p ¼ 0.006 one-tailed when considering condi-
tional means). This is consistent with the self-concept maintenance
theory: individuals ?nd it more dif?cult to misreport costs in the GL
condition, where the cost report can determine whether the
company earns a gain or a loss. The presence of an information
system reduces the motive to report more honestly in the GL
condition. When the information system is present, results show
that the GL condition has higher degrees of misrepresentation and
the level is comparable to or even higher than the negative or
positive earnings conditions. The simple effects of information
system in Panel B of Table 2 show that information systems
marginally reduce misrepresentation in both the NE (À0.177,
Table 1
Hypothesis test.
Panel A: Summary statistics for misrepresentation (Y ¼ reported cost e actual
cost)
Negative earnings
(NE)
Positive earnings
(PE)
Gain or losses
(GL)
Info system absent 0.309 0.345 0.139
[0.296] [0.375] [0.142]
(0.340) (0.327) (0.451)
n ¼ 31 n ¼ 31 n ¼ 31
Info system present 0.132 0.222 0.262
[0.143] [0.233] [0.259]
(0.398) (0.233) (0.404)
n ¼ 31 n ¼ 31 n ¼ 31
Panel B: Full factor analysis with period as a within-subject factor
Model 1 Model 2
DF MS F Stat Sign. Num.
DF
Den
DF
F Stat Sign.
Between subjects
Earnings (E) 2 0.58 0.86 0.424 2 180 1.46 0.234
Info. System (IS) 1 0.81 1.22 0.271 1 180 1.27 0.262
E*IS (H1) 2 1.98 2.97 0.054* 2 180 2.79 0.064*
Error 180 0.67 e e e
Within subjects
Actual cost e 40 680 11.89 0.17) when the information system is present. In the GL con-
dition, the information system marginally increases misreporting
(one tailed p ¼ 0.05 for the mean; one-tailed p ¼ 0.07 for the
conditional mean), whereas it marginally reduces misreporting in
the PE condition (one tailed p ¼ 0.08 for the mean; one-tailed
p ¼ 0.06 for the conditional mean). In contrast to Table 2, the
decrease in misreporting as a result of information system for the
NE condition is no longer signi?cant (one-tailed p > 0.24 for the
mean; one-tailed p > 0.28 for the conditional mean).
4.2. Supplementary analyses of choice behavior
Table 3 analyzes the frequency of reporting behavior that can
be considered as bene?cial to the company. I count the number of
observations per participant over the ?ve reporting periods
where reported costs are equal to or belowactual costs, labeled as
total social choices in Panel A of Table 3 (is equal to the sum of
honest reports (i.e., reports equal to actual cost) and altruistic
reports (i.e., reports below actual costs)). The remainder are
dishonest reports for which reported cost are higher than actual
cost (total dishonest in Panel A of Table 3). The theory suggested
that individuals may be more honest or underreport costs to
improve the ?rm's pro?ts and thus misreport less if their reports
can determine whether the company earns a pro?t or loss (the GL
condition) compared to negative and positive earnings condi-
tions. Such behavior in the GL condition may occur less frequently
Table 2
Pairwise comparison of the interaction.
Panel A: Display of the means [conditional means controlling for covariate actual cost]
Panel B: Comparison simple effects per earnings condition and per information system condition
Comparison Means Conditional means
Info system absent Info system present Info system absent Info system present
Effect (sign.) Effect (sign.) Effect (sign.) Effect (sign.)
NE vs. GL þ0.170 (0.034)** À0.130 (0.081)* þ0.154 (0.047)** À0.116 (0.104)
PE vs. GL þ0.206 (0.014)** À0.040 (0.332) þ0.234 (0.006)*** À0.026 (0.390)
NE vs. PE À0.036 (0.351) À0.090 (0.167) À0.079 (0.193) À0.090 (0.163)
Effect info system Effect info system
Effect (sign.) Effect (sign.)
Neg. earnings (D-A) À0.177 (0.029)** À0.153 (0.048)**
Pos. earnings (E-B) À0.123 (0.094)* À0.142 (0.061)*
Gain or loss (C-F) þ0.123 (0.094)* þ0.117 (0.102)
***, **, * indicate signi?cance levels of 1%, 5%, and 10% levels of signi?cance using one-tailed directional tests. Panel A shows a graphical plot of the cell means of Table 1
(respectively the conditional means using actual cost as time varying covariate). Panel B performs a mean by mean comparison of the interaction earnings  information
system of Table 1 across information system condition or across earnings condition. Differences between the means are displayed ?rst; the one-tailed p-values are shown
between brackets and are based on directional test using the least squares means procedure within an ANOVA or mixed model procedure (conditional means).
E. Cardinaels / Accounting, Organizations and Society xxx (2015) 1e13 8
Please cite this article in press as: Cardinaels, E., Earnings benchmarks, information systems, and their impact on the degree of honesty in
managerial reporting, Accounting, Organizations and Society (2015),http://dx.doi.org/10.1016/j.aos.2015.09.002
once an information system is present. I thus anticipate an
interaction effect, suggesting that the effect of information sys-
tem on this type of reporting behavior depend on earnings con-
dition. A type 3 analysis, using frequency of social choices on
participant level of as the dependent (multinomial response),
shows, however, that the interaction of ExIS is not signi?cant (c
2
(2, N¼186)
¼ 3.54, p ¼ 0.17). Also the main effects are not signi?cant
(information system c
2
(1, N¼186)
¼ 1.07, p ¼ 0.30; earnings c
2
(2,
N¼186)
¼ 4.29, p ¼ 0.12). Contrasts show that the interactive
pattern is signi?cant at the 10% level when comparing the GL
condition against the NE condition (c
2
(1, N¼186)
¼ 3.13; p ¼ 0.08)
and when comparing the GL condition against the combined cells
PE and NE (c
2
(1, N¼186)
¼ 3.44, p ¼ 0.06). The differential impact is
not signi?cant when comparing the GL condition against the PE
condition (c
2
(1, N¼186)
¼ 2.08, p ¼ 0.15).
Panel B of Table 3 compares the effect of an information
system on the frequency of social choices in each of the three
earnings condition, separately. The effect of information system
in the GL condition reduces the frequency of these reporting
choices from 54.84% to 35.48% (i.e. reduction of minus 19.35%
percent, p ¼ 0.04). As such, the use of information system in the
GL condition shifts participants towards dishonest reporting. In
the NE or PE conditions there is no signi?cant effect of infor-
mation system on the frequency of social choices (i.e.
respectively, þ4.52%; p ¼ 0.73 and À5.81%; p ¼ 0.95). Further
analyses show that in the information system absent condition,
the frequency of social behavior of 54.84% in the GL condition
differs from the respective frequencies of 38.06% and 32.90%
observed in the NE and PE conditions (GL vs. NE, c
2
(1,
N¼62)
¼ 2.71, p ¼ 0.10; GL vs. PE, c
2
(1, N¼62)
¼ 4.62, p ¼ 0.03). In
the condition where the information system is present, the dif-
ference between the GL vs. the PE condition and the difference
between GL vs. NE condition are no longer signi?cant
(p's > 0.45). Panel A of Table 3 further shows that a large part of
the social behavior in the GL condition, when information sys-
tems are absent, can be explained by altruistic choices in which
reported costs are below actual costs. Speci?cally, 30.32% of the
choices can be labeled as altruistic. This frequency differs from
the percentages of 9.03% and 4.52% respectively observed in the
NE and PE conditions (p's < 0.01). Analyses reported in footnote
indicate that participants in the GL condition produce these
altruistic reports to avoid losses and thus to keep the company
pro?table.
16
Table 4 uses the meanof three items froma post-questionnaire to
measure the participants' social concerns for the ?rm. Panel A of
Table 4 shows that in each of the three earnings conditions, infor-
mation systems reduce the social concerns for the ?rm. This is also
con?rmed by an ANOVA where only the main effect of information
system is signi?cant (F
(1, 180)
¼ 5.77; p ¼ 0.017). Results thus do not
provide support for a differential impact of information systems
across earnings conditions on the variable social concerns. Untabu-
lated results show that neither the interaction of Ex IS, nor contrast
tests of the interaction are signi?cant (p's > 0.32). Panel B of Table 4
shows, however, that the reductionof social concerns as a result of an
information systemis only signi?cant in the GL condition (p ¼ 0.03)
and not in the NE condition (p ¼0.47) or the PE condition (p ¼0.18).
This reduction in the GL condition may presumably arise because
participants worry less about reports which are bene?cial to ?rm
pro?t, once headquarters uses an information system.
17
Note that while these analyses provide some insights as to why
information systems produce a negative effect on the degree of
misrepresentation in the gain/loss condition, they do not explain
why information systems still produce positive effects in the pos-
itive earnings or the negative earnings conditions. Consistent with
Gneezy (2005) and Hannan et al. (2006), the next section presents
histograms of participants’ reporting choices to provide more in-
sights on this issue.
Table 3
Supplementary analyses of reporting choices.
Panel A: Summary statistics of choice behavior
Negative earnings (NE) Positive earnings (PE) Gain or losses (GL)
Info system absent Altruistic (a) 9.03% 4.52% 30.32%
Honest (b) 29.03% 28.39% 24.52%
Total social (a þ b) 38.06% 32.90% 54.84%
Total dishonest 61.94% 67.10% 45.16%
n ¼ 31 n ¼ 31 n ¼ 31
Info system present Altruistic (a) 18.06% 9.68% 20.65%
Honest (b) 24.52% 17.42% 14.84%
Total social (a þ b) 42.58% 27.10% 35.48%
Total Dishonest 57.42% 72.90% 64.52%
n ¼ 31 n ¼ 31 n ¼ 31
Panel B: Effect of information system per earnings condition (Y ¼ total social choices)
Effect info system þ4.52% ¡5.81% ¡19.35%
Chi-square (1, N ¼ 62) 0.120 0.003 4.388**
(p-value) (p ¼ 0.73) (p ¼ 0.95) (p ¼ 0.04)
***, **, * indicate signi?cance levels of respectively 1%, 5% and 10% level of signi?cance (two-tailed). I dummy coded each choice on participant level as either social (i.e. 1 if cost
equal to or below actual costs; 0 otherwise) and subsequently calculate for each participant the frequency of how many of the ?ve reporting choices can be labeled as social
(total social). The range of these variables on participant level is either 0%, 20%, 40%, 60%, 80% or 100%. The remainder (1Àtotal social) are the frequency of dishonest choices.
The table also constructed frequency variables on participant level based on choices classi?ed as altruistic choices (i.e. equal to one if reported cost < actual cost; 0 otherwise)
and honest choices (i.e. equal to one if reported cost equals actual cost; 0 otherwise), Panel A displays frequencies. Given that variables are based on frequencies, panel B shows
for each earnings condition, the results of a KruskaleWallis test for differences in distribution in social behavior across the factor information system.
16
I also calculate the conditional pro?t that headquarters would have realized if
participants were to report honestly and compare this to the pro?t headquarters
receives under the altruistic choice. In the scenario where the information system is
absent, the loss for the ?rm would be equal to À162.23 lira if the participant would
have reported honestly, but the ?rm realizes a gain of 40.43 lira because of the
altruistic reporting. Participants still earn positive earnings of 47.34 lira. In the in-
formation system present condition, altruistic reporting also ensures that the
company earned a pro?t of 2.34, rather than a loss of À116.41 if the participant
would have been honest. Yet, fewer of these choices occur and also less is trans-
ferred to the ?rm, because agents still keep on average 131.25 lira.
17
The variables total social choices of Table 3 and the level of social concerns of
Table 4 are positively correlated (r ¼ 0.48, p < 0.01).
E. Cardinaels / Accounting, Organizations and Society xxx (2015) 1e13 9
Please cite this article in press as: Cardinaels, E., Earnings benchmarks, information systems, and their impact on the degree of honesty in
managerial reporting, Accounting, Organizations and Society (2015),http://dx.doi.org/10.1016/j.aos.2015.09.002
4.3. Reporting behavior and impact of an information system
Fig. 1 uses the 155 reports per cell (i.e. 31 participants each
producing ?ve reports) to further explore howinformation systems
change the distribution of reports. Panel A of Fig. 1 shows the
number of cost reports that are either (1) honest or altruistic (i.e.,
misreporting that was equal or below zero); (2) dishonest but for
which the information system cannot detect misreporting (i.e.
0 < misrepresentation 0.25); or (3) dishonest to a level above the
detection limit of the information system (i.e., the degree of
misrepresentation> 0.25).
Consistent with the theory, panel A of Fig. 1 shows that in the
negative and positive earnings conditions, presence of an infor-
mation system (compared to absence of it) reduce the participant's
range of dishonesty. The ?gures show a shift from the interval
involving large dishonest reporting to the interval of dishonest
reporting which cannot be detected by the information system, for
which participants may still be able to maintain a positive self-
concept. The test for differences in distribution of reports across
the factor information system is signi?cant both for the negative
earnings condition (c
2
(1, N¼310)
¼ 8.46, p < 0.01) as well as the
positive earnings condition (c
2
(1, N¼310)
¼ 4.79, p < 0.05).
Conversely, in the gain/loss condition, participants reduce honest
and altruistic reporting (as is evidenced in Table 3), and shift to
more dishonest reports which are below the detection limit of the
information system. The test for differences in distribution of re-
ports across the three intervals for the factor information system is
signi?cant (c
2
(1, N¼310)
¼ 3.402, p