Download as pdf or txt
Download as pdf or txt
You are on page 1of 32

Traditional ABC and Time-Driven ABC:

An Experimental Investigation
Karl Schuhmacher∗† Michael Burkert∗
University of Lausanne

ABSTRACT: We conduct an experiment to investigate the accuracy of different time estimation


approaches in activity-based costing (ABC). Employing a 2×2 between-subjects design, we vary
time-driven costing approach and number of tasks per activity. Specifically, we compare traditional
ABC – estimating time in a top-down approach based on total work time (typically in relative units)
– and time-driven ABC (TD-ABC) – estimating time in a bottom-up approach directly focusing
on the single activities (typically in absolute units). Most importantly, the two systems differ in
their reference points for time estimation (total work vs. activities). Moreover, we vary number of
tasks per activity manipulating heterogeneity between activities.
Our design builds on a study by Cardinaels and Labro (2008). We ask 78 student participants to
provide time estimates for activity durations after having performed a set of 20 tasks. As percent-
age estimates in traditional ABC have to add up to 100 %, we adapt time estimates in TD-ABC
correcting for consistent under-/overestimation bias. Our results indicate that TD-ABC results
in more accurate estimates when the number of tasks per activity varies. However, when activity
heterogeneity is low, we do not find either approach to be more accurate. Moreover, when we do
not correct estimates for consistent bias in TD-ABC, our results reveal that participants heavily
underestimate time in TD-ABC. We conduct analyses indicating that this underestimation bias re-
sults in significantly less accurate estimates for TD-ABC compared to traditional ABC questioning
implications on practical capacity in TD-ABC. Finally, we provide evidence that individuals access
different reference points for time estimation in TD-ABC and traditional ABC.
We contribute to accounting literature directly comparing the accuracy of different time-driven
costing approaches. Moreover, our results suggest that organizations should rely on TD-ABC for
time estimation. However, implications about practical capacity have to be treated with caution
as TD-ABC is prone to consistent under-/over-estimation.

The authors acknowledge support from the Swiss National Science Foundation under project number
100014 140612/1. This paper greatly benefited from the comments of participants at the EAA’s Doctoral
Colloquium 2013 in Paris.

Corresponding author: University of Lausanne; Quartier UNIL-Dorigny; Bâtiment Internef, #501; 1015
Lausanne (Switzerland); karl.schuhmacher@unil.ch; phone: +41 21 692 3383.
1 Introduction
Management accounting systems (MAS) and, specifically, cost systems are a central pillar of orga-
nizations representing an important source of information (e.g., Merchant and Van der Stede 2007;
Horngren et al. 2008). The accuracy of information within these systems is essential as they lay
the foundations for decision-making. However, cost systems contain errors and often do not feature
an adequate level of accuracy (Noreen 1991; Datar and Gupta 1994; Labro and Vanhoucke 2007,
2008). It is essential for designers of such systems to know about potential errors in their systems.
Most cost accounting research has focused on errors arising from rather technical issues con-
cerning practicability and feasibility, e.g., merging incongruent resources into cost pools with only
one cost driver. There is an extensive body of research focusing on such issues in the domain of
analytical modeling (e.g., Datar and Gupta 1994; Christensen and Demski 1995), which has been
extended by large simulation studies (Labro and Vanhoucke 2007, 2008; Balakrishnan et al. 2011).
Different from these streams of literature, we investigate errors in cost systems that result from
cognitive limitations.
Cognitive biases are highly relevant for cost systems relying on employees’ time estimates as time
estimation is subject to uncertainty (Ittner 1999; Cardinaels and Labro 2008). This study focuses on
cognitive biases that arise when employees intend to estimate time as accurate as possible. Thus, we
do not reflect on strategic considerations that may distort time estimates. Rather, we focus on “best
estimates” from which employees may deviate to incorporate strategic considerations. Cardinaels
and Labro (2008) provide first evidence on cognitive limitations investigating the accuracy of time
estimates in time-driven costing. In a laboratory experiment, they ask participants to perform a
set of 20 tasks (belonging to different activities) and, subsequently, to estimate the time spent on
these activities. True time is measured by the computer allowing them to nicely benchmark time
estimates against true time.
Based on this study, we use a similar experimental design in order to directly compare different
time-driven costing approaches. In management accounting, time-driven costing has received much
attention since Kaplan and Anderson (2004, 2007) have introduced what they have branded time-
driven activity-based costing (TD-ABC). TD-ABC differs from traditional ABC mainly in its time
estimation technique. While traditional ABC focuses on total work as reference point estimating
time in a top-down approach (typically in relative units, i.e., percentages); TD-ABC directly refers
to the single activities estimating time in a bottom-up approach (typically in absolute units, i.e.,
minutes and seconds). Thus, in traditional ABC, employees estimate time answering the question
“what percentage of total work time did you spend on activity”; whereas employees estimate time
in TD-ABC answering the question “how much time [in absolute units] did you spend on average
on activity X to perform it once”. Importantly, estimates in traditional ABC refer to overall time
spent on each activity. However, estimates in TD-ABC solely refer to the average amount of time
spent to perform an activity once. Thus, time estimates in traditional ABC (different from TD-
ABC) have to account for the number of times a specific activity has been performed. Moreover,

1
estimates in traditional ABC (different from TD-ABC) have to add up to 100 % not allowing for
consistent under-/over-estimation bias. Therefore, Kaplan and Anderson (2004, 2007) claim that
TD-ABC (1) is more accurate, (2) allows to account for unused capacity, and (3) is cheaper to
adapt and implement, arguments highly relevant to practitioners.
In order to investigate such aspects more in detail, we build on the study by Cardinaels and
Labro (2008) and extend their research question asking how different time-driven costing ap-
proaches, i.e., traditional ABC and TD-ABC, influence measurement error in cost systems. In
particular, we apply a 2×2 between-subjects design varying time-driven costing approach and
number of tasks performed per activity manipulating heterogeneity between activities. All partici-
pants in our experiment perform a set of 20 tasks, each belonging to one of four different activities.
While one group performs 5 tasks per activity, the other group performs 6/4/3/7 tasks for the re-
spective activities. Subsequently, participants provide time estimates for the four activities which
are benchmarked against true time (as measured by the computer). Participants estimate time
according to either traditional ABC or to TD-ABC.
We intend to add to literature by investigating cognitive limitations that arise within traditional
ABC and TD-ABC. Specifically, we compare the top-down approach in traditional ABC with total
work as reference point and the bottom-up approach in TD-ABC with single activities as reference
points. We argue that time estimation in TD-ABC is cognitively less demanding and, thus, more
accurate testing different arguments for TD-ABC. First, estimates in TD-ABC refer directly to
the single activities without having to refer to total work time making it easier to come up with
time estimates. While traditional ABC asks for time estimates as partition of total time, i.e., “how
much of total work time did you spend on activity X”, TD-ABC directly asks “how much time
did you spend ... on activity X ...”. Moreover, as time estimates in traditional ABC are typically
demanded in relative units, i.e., percentages (“what percentage of total work time did you spend
on activity X”), we argue that it is cognitively less demanding to provide estimates in absolute
units, i.e., minutes and seconds (“how much time did you spend ... on activity X ...”). Similar to
probabilities, percentages are cognitively challenging constructs (e.g., Kahneman et al. 1982).
Second, we argue that estimates in TD-ABC are more accurate as it is only required to estimate
the average duration to perform an activity once (“how much time did you spend on average on
activity X to perform it once”). However, traditional ABC requires employees to account for the
number of times a specific activity has been performed relying on an internal count of tasks per
activity (‘ ‘what percentage of total work time did you spend on activity X”). Thus, we do not
expect that the accuracy of time estimates in TD-ABC differs depending on the number of times
different activities have been performed. However, estimates in traditional ABC are likely to be
less accurate the more activities differ in the amount of times they have been performed.1
Especially, when we correct for consistent under-/over-estimation bias in TD-ABC, we expect
TD-ABC to be more accurate than traditional ABC. However, when we do not correct for consistent
1
Costing systems typically keep count of the number of times a certain activity has been performed.

2
under-/over-estimation in TD-ABC, these advantages of TD-ABC are likely to be destroyed. As
estimates in traditional ABC have to add up to 100 %, total work time respectively, consistent
under-/overestimation is automatically eliminated. Due to the detachment from total work time in
TD-ABC, proponents of traditional ABC argue that TD-ABC is prone to consistent under-/over-
estimation bias. Such consistent estimation biases endanger implications on practical capacity.
Importantly, our time-driven costing manipulation differs from the differentiation between ab-
solute and relative units in Cardinaels and Labro (2008) in two aspects: First, we make a difference
between the top-down (percentages) and bottom-up estimation approach (minutes and seconds),
while Cardinaels and Labro (2008) refer for both absolute and relative time units to the top-down
approach. Second, we vary estimation technique between subjects, while Cardinaels and Labro
(2008) vary estimation technique within subjects. However, we rely on the approach by Cardinaels
and Labro (2008) in a second time estimation to verify the robustness of our results.
In total, we recruited 78 students from a participants pool for experiments at a large West-
ern European university to participate in our experiment. In order to motivate participants to
estimate activity durations as precise as possible, we compensated them based on the accuracy of
their time estimates. We find that the accuracy of estimates in TD-ABC does not depend on the
number of times different activities are performed. Contrary, the accuracy of estimates in tradi-
tional ABC deteriorates significantly with the amount of times activities are performed. Moreover,
when we correct for consistent under-/over-estimation bias in TD-ABC, we do not find support for
our hypotheses that estimates in TD-ABC are more accurate than estimates in traditional ABC.2
However, when we allow for consistent under-/over-estimation bias in TD-ABC, we find that par-
ticipants have a strong tendency to underestimate time in TD-ABC. We provide analyses indicating
that this bias turns estimates in traditional ABC to be significantly more accurate than estimates
in TD-ABC. This is due to the fact that estimates in TD-ABC have to add up to 100 %, total work
time respectively, and do not allow for consistent under- or over-estimation.
Moreover, our results remain consistent in a second estimation (robustness check). In this
estimation, participants in the traditional ABC condition have to estimate total time in absolute
units and, subsequently, allocate it to each activity, whereas participants in the TD-ABC condition
directly estimate overall time spent on an activity without estimating total work time before.3 This
reflects the idea that traditional ABC represents a top-down approach, whereas TD-ABC rather
represents a bottom-up approach. Interestingly, the results of our robustness check indicate that
participants in the TD-ABC condition continue to underestimate time, while participants in the
traditional ABC condition heavily overestimate time.
We contribute to accounting literature as we directly compare the accuracy of time estimates
2
This evidence is somewhat contrary to a similar study that we conduct on the accuracy of different time
estimation approaches in costing.
3
In order to retain the advantage of TD-ABC that participants do not have to keep an internal count of
times a specific activity has been performed, we provide participants in TD-ABC in the second estimation
with information on the amount of times they have performed a specific activity.

3
in traditional ABC and TD-ABC. Our results suggest that estimates in TD-ABC – corrected for
consistent under-/over-estimation bias – are more accurate than estimates in traditional ABC when
number of tasks per activity varies. However, not corrected for consistent bias, estimates in TD-
ABC are less accurate than estimates in traditional ABC. As estimates in absolute units are a lot
less expensive, we consider our results directly relevant for designers of cost systems. However,
conclusions about practical capacity from estimates in TD-ABC need to be treated with caution.
In the next section, we develop our hypotheses. Section 3 describes the experimental design,
participant recruitment and payment, as well as the manipulations. Section 4 presents our results
and, in section 5, we conclude with final remarks pointing out directions for future research.

2 Literature Review and Hypotheses


In this section, we review literature on cost systems and estimation techniques leading to our
hypotheses on the accuracy of cost systems.

2.1 Literature Review


Cost systems are an important source of information for decision-making. Therefore, the accuracy
of cost information and, more so, the allocation of costs to cost objects has been a major interest
of research in management accounting literature. When costs cannot be directly assigned to cost
objects, an allocation mechanism is needed and uncertainty enters in cost systems. The central
aspect of different allocation mechanisms is that the true consumption of resources is unknown
introducing error to cost systems (Noreen 1991; Datar and Gupta 1994; Labro and Vanhoucke
2007, 2008).
A major concern in any cost system are personnel costs that cannot be directly assigned to cost
objects. In activity-based costing (ABC), these costs are allocated based on the time employees
spent on different activities and the amount of activities that are consumed by a specific cost
object. When time cannot be measured directly by the system, time estimations are necessary.
These estimations are a potential source of error in costing systems.
In traditional ABC, time estimations are based on the total work time of an employee. When
employees estimate how much time they spend on a specific activity, they are asked to provide
this estimate as a proportion of total work time (‘ ‘what percentage of total work time did you
spend on activity X”). Typically, they provide time estimates in relative units, i.e., percentage
estimates of total time. The advantage of such an estimation method is that percentage estimates
are a zero-sum game similar to bonus pools in compensation contracting. Thus, employees cannot
consistently under- or over-estimate time for each activity as percentages have to add up to 100
%.4 Importantly, this approach to estimate time is based on total work time. That is, estimates
4
This is similar to the notion that leniency bias cannot be observed in bonus pools (Bol 2008).

4
for each activity are provided as a proportion of total work time making the estimation procedure
a top-down approach portioning total time into time consumed by each activity. Thus, employees
refer to total work time when estimating time according to traditional ABC.
A large drawback of traditional ABC arises when the amount of tasks per activity varies across
time periods. This problem occurs because estimates in traditional ABC are based on overall time
spent on one activity. Specifically, estimates have to account for the amount of times different
activities are performed. Consequently, practitioners have to frequently update their systems and
time estimates accounting for the variation in the number of times different activities are performed.
Therefore, traditional ABC is considered quite time- and cost-consuming occupying a large amount
of human resources with the elicitation of time estimates. Moreover, when a new activity enters
or a standardized activity is adapted in the system, all other estimates also have to be updated
making ABC itself a large pool of costs.
Due to these drawbacks of traditional ABC, Kaplan and Anderson (2004, 2007) have introduced
a rivaling approach to time-driven costing, namely Time-Driven Activity-Based Costing (TD-ABC).
In particular, they propose to estimate time with a bottom-up approach focusing directly on the
activities without referring to total time. Typically, estimates in TD-ABC are provided in absolute
units, i.e., hours, minutes, and seconds. Importantly, estimates are not provided for overall time
spent on a specific activity in a given time period. Rather, employees have to estimate the average
time spent to perform the specific activity once (“how much time did you spend on average on
activity X to perform it once”). For example, assuming that it takes 30 minutes to perform a
specific activity once and this activity is carried out 10 times in a given time period, TD-ABC
(different from traditional ABC) is not interested in the figure displaying the overall time spent on
the activity, i.e., 5 hours (30 minutes × 10 times), but rather in the figure displaying the time to
perform this activity once, i.e., 30 minutes.
As cost systems usually keep track of the amount of times different activities are performed,
it is sufficient to estimate the average duration spent to perform an activity once. Thus, practical
capacity, can be calculated within cost systems multiplying the amount of times an activity has
been performed and average time to perform the activity once making this method a bottom-
up approach to time-driven costing. Kaplan and Anderson (2004, 2007) argue that TD-ABC (1)
provides more detailed and precise time estimates, (2) allows to account for unused capacity, and
(3) is cheaper to implement and adapt in comparison to traditional ABC. The presumption of their
approach is that information is available concerning the amount of times different activities are
performed in a specific time period. However, a large disadvantage of this method is that it allows
for consistent under-/over-estimation potentially working against the proposed advantages.
Literature provides considerable case-based evidence on the use and implementation of TD-
ABC in practice (Pernot et al. 2007; Everaert et al. 2008; Demeere et al. 2009; Dalci et al. 2010;
Hoozée and Bruggeman 2010; Öker and Adigüzel 2010; Everaert et al. 2012). Moreover, Hoozée
et al. (2012) investigate the refinement of time equations used in TD-ABC analytically. In general,

5
there is an extensive body of research focusing on the accuracy of cost systems in the domain of
analytical modeling (e.g., Datar and Gupta 1994; Christensen and Demski 1995), which has been
extended by large simulation studies (Labro and Vanhoucke 2007, 2008; Balakrishnan et al. 2011).
Different from these streams of literature, we investigate errors in cost systems that result from
cognitive limitations.
Cardinaels and Labro (2008) provide first evidence for errors resulting from cognitive limitations
in time-driven costing. They conduct an experimental study investigating the accuracy of time
estimates in costing. Specifically, they ask participants to perform 20 tasks (belonging to three
main activities, respectively, six sub-activities) and, subsequently, to estimate the time spent on the
different activities. They find that estimates are more accurate when activities are estimated on an
aggregated level and when participants are prospectively notified about estimation. Moreover, they
ask participants to estimate time, first, in relative units (percentages) and, then, in absolute units
(minutes). Precisely, they ask participants in the second estimation (absolute units) to estimate
total time and, subsequently, to allocate estimated total time to the different activities. They find
that participants tend to over-estimate time when they estimate in absolute units and argue that
“[t]his overestimation bias is likely to undo the intended effect of the minutes-based response on
allocating only practical capacity rather than full capacity” (Cardinaels and Labro 2008, p. 737).
Importantly, Cardinaels and Labro (2008) ask participants to estimate time using the top-down
approach known from traditional ABC for both response modes, i.e., relative units and absolute
units. Moreover, they vary response mode within-subjects to provide a sensitivity check for their
results.
In this study, we want to extend the framework provided by Cardinaels and Labro (2008)
varying the approach to costing (traditional ABC versus TD-ABC) and the number of tasks per
activity. The different approaches to time-driven costing can be nicely compared from a reference
point perspective. While the reference point for estimation in traditional ABC is total work time,
TD-ABC directly focuses on the single activities. Therefore, we consider TD-ABC a cost system
with a bottom-up approach estimating single activities durations and, eventually, ending up with
practical capacity; whereas we consider traditional ABC a cost system with a top-down approach
starting out at full (or practical) capacity and establishing activity durations as final output of the
costing process.

2.2 Hypotheses
Comparing the two approaches to time-driven costing, i.e., traditional ABC and TD-ABC, we
argue that providing estimates in traditional ABC is cognitively more demanding than providing
estimates in TD-ABC when consistent under-/over-estimation is accounted for. In particular, we
elaborate on two different arguments to support our hypotheses. However, when consistent under-
/over-estimation in TD-ABC is not corrected for, we suppose that potential benefits from TD-ABC
are likely to be eliminated.

6
The main difference between the two approaches to time-driven costing is their reference point
in time estimation. While traditional ABC asks for estimates as a proportion of total time, i.e.,
“how much of total work time did you spend on activity X”, TD-ABC directly asks for time spent on
a specific activity, i.e., “how much time did you spend ... on activity X ...”. Therefore, employees
can directly focus on the activity itself (as in TD-ABC) without having to make the detour via total
work time (as in traditional ABC). We argue that it is cognitively more challenging to process the
top-down approach in traditional ABC (referring to total work time) compared to the estimation
process in TD-ABC, where employees simply have to picture how they perform the specific activity
once.
Another difference between traditional ABC and time-driven ABC is that time estimates are
typically provided in different response modes in the two approaches. While traditional ABC usually
relies on estimates in relative units, i.e., percentages (“what percentage of total work time did you
spend on activity X”), TD-ABC relies on estimates in absolute units, i.e., minutes (“how much time
did you spend ... on activity X ...”). We argue that it is cognitively more challenging to come up
with time estimates in relative units as compared to time estimates in absolute units. Similar to
probabilities, percentages are perceived as complex constructs that individuals have difficulties to
capture (e.g., Kahneman et al. 1982). However, absolute time units are concepts that individuals
use on a daily basis and are, therefore, better used to estimate time in absolute units.
Finally, when time estimates are provided as a proportion of total work time (as in traditional
ABC), it is likely that individuals have to adapt their initial estimates in order for estimates to add
up to 100 %, total work time respectively. It is not clear, how this adaptation process influences the
accuracy of time estimates. Importantly, time estimates in absolute units are not subject to such
restrictions and can be freely varied. However, they are potentially subject to consistent under- or
over-estimation bias. We correct estimates in TD-ABC for such bias in order to solely focus on the
arguments presented before.5 Based on these arguments, we derive the following hypothesis:

Hypothesis 1 Estimates in traditional ABC are less accurate than estimates in TD-ABC – cor-
rected for potential under-/over-estimation bias – as estimating time in traditional ABC is cogni-
tively more demanding (when the number of tasks per activity does not vary).

However, time estimates in TD-ABC are rarely corrected for consistent under-/over-estimation
bias in practice. Nevertheless, such biases are likely to influence the accuracy of time estimates in
TD-ABC. On the other hand, time estimates in traditional ABC, by definition, do not allow for
consistent under-/over-estimation bias as they have to add up to 100 %, total work time respectively.
Cardinaels and Labro (2008, p. 737) find that participants tend to over-estimate time when they
estimate in absolute units and argue that “[t]his overestimation bias is likely to undo the intended
effect of the minutes-based response on allocating only practical capacity rather than full capacity”.
5
We correct for consistent under-/over-estimation bias rescaling time estimates in TD-ABC assuming
that implicitly estimated total work time corresponds to true total work time as measured by the computer.

7
Moreover, research in psychology provides evidence that individuals tend to consistently under-
/over-estimate time when their estimates are not restricted as in traditional ABC (e.g., Fortin and
Rousseau 1998). Therefore, we argue that benefits from TD-ABC as described above are likely to
be eliminated due to consistent estimation bias when we do not correct time estimates in TD-ABC
for such bias, which leads us to propose the following hypothesis:

Hypothesis 2 Estimates in traditional ABC are more accurate than estimates in TD-ABC – not
corrected for potential under-/over-estimation bias – as estimates in traditional ABC do not allow
for consistent under-/over-estimation bias.

Finally, traditional ABC and TD-ABC differ in one further major aspect. Whereas estimates
in traditional ABC have to account for overall time spent on a specific activity (“what percentage
of total work time did you spend on activity X”), estimates in TD-ABC only have to account for
the average time spent to perform an activity once (“how much time did you spend on average
on activity X to perform it once”). Consequently, employees have to include the number of tasks
performed per activity when they estimate time in traditional ABC, whereas employees solely have
to estimate the time spent on the average task per activity in TD-ABC. Therefore, they have
to additionally rely on an internal count of tasks in traditional ABC making time estimation in
traditional ABC cognitively more demanding than time estimation in TD-ABC.
We argue that it is more challenging to picture, process, and estimate overall time of an activity
including all tasks that have been performed (as in traditional ABC) as compared to the average
duration to perform a specific activity once (as in TD-ABC). However, we do not expect this
difference between the two costing approaches to result in significantly different results when each
activity is performed the same amount of times (symmetry between amount tasks). In this case,
the count of tasks can be ignored and the average tasks can be evaluated as in TD-ABC. However,
when different activities are performed more or less often (asymmetry between amounts of tasks),
task count becomes relevant. Therefore, we argue that it is cognitively more challenging to come
up with time estimates in traditional ABC when the amount of tasks per activity varies. However,
the amount of tasks per activity changes little when estimating time in TD-ABC. This leads to the
following hypotheses:

Hypothesis 3 Compared to estimates in TD-ABC, estimates in traditional ABC are less accurate
when the amount of tasks per activity varies.

3 Experimental Setting
In order to test our hypotheses, we conduct a computerized experiment with 78 student participants
at a large Western European university. We build our experimental set-up on Cardinaels and
Labro (2008). Employing a 2×2 between-subjects design, we vary time-driven costing approach

8
and number of tasks per activity. Particularly, we ask participants to provide time estimates for
activity durations after they have performed a set of 20 tasks which belong to one out of four
different activities (5 tasks per activity versus 6/4/3/7 tasks per activity). Specifically, we compare
two approaches to time estimation and, more specifically, to time-driven costing (traditional ABC
versus TD-ABC). In order to evaluate the accuracy of time estimates we benchmark them against
true time spent on these activities as measured by the computer.

3.1 Experimental Tasks


Similar to the study by Cardinaels and Labro (2008), participants in our experiment assume the
role of a worker in a company that produces different types of wooden blocks. These wooden
blocks can be customized in terms of colors and shape (L shapes and S shapes). Workers in this
company perform 4 simple pre-defined activities. Different from Cardinaels and Labro (2008), we
ask participants to perform 4 instead of 3 different activities as we do not differentiate between
levels of aggregation.6 Similar to Cardinaels and Labro (2008), we ask participants to perform a
set of 20 tasks and to estimate subsequently the time spent on these tasks. Each of the 20 tasks
belongs to one of our 4 main activities (Incoming Customer Order [A], Machine Programming [B],
Material Requests [C], and Billing [D]) resulting in 5 tasks per activity for one experimental group
and 6/4/3/7 tasks per activity for the other experimental group.7
Participants receive an exhaustive training on each of the four activities before they start with
the actual set of 20 tasks in order to assure that there is no learning effect. While participants
perform the set of 20 tasks, the computer counts the time they spend on these tasks. After
participants have finished the set of 20 tasks, they are asked to provide estimates about the time
spent on the corresponding activities. Finally, these estimates are compared to the true time
measured by the computer. Different from Cardinaels and Labro (2008), we include little pauses
in-between consecutive tasks to reduce levels of stress as stress has been shown to have an effect on
time estimates. Nevertheless, pauses are not included in the estimations later on. In what follows,
we briefly sketch the 4 different activities.
In activity A (Incoming Customer Order ), workers have to read an email with details of an
incoming order. Subsequently, they have to enter this information into their order system. In
especial, they enter first name and family name of the client into the ordering system. In addition,
they have to enter the product (L-shape or S-shape), amount of product ordered, and amount of
colors per product into the system. For each incoming order, they have to enter two products in
their ordering systems.
Activity B (Machine Programming) requires workers to program a code that tells a machine
6
We only ask participants to estimate activities on the aggregated level because Cardinaels and Labro
(2008) find that these estimations are significantly more accurate.
7
Cardinaels and Labro (2008) do not include the activity Incoming Customer Order.

9
what kind of product to produce.8 Moreover, information about the batch size needs to be pro-
grammed. Based on a color draft in a 3 × 3 matrix, participants have to enter a program code
transforming colors into numbers. They are given a color table indicating the numbers that cor-
respond to the respective color (10 different colors in total - 0 to 9). Each color has a unique
structure that allows participants to clearly assign numbers to colors when they have problems to
differentiate between colors. Workers either have to program an “L-shaped” code, or an “S-shaped”
code.
Activity C (Material Requests) demands workers to treat material requests from a work station
requiring different colors. Firstly, workers have to indicate which workstation is in need of new
materials (A or B). Moreover, based on a graphic indicating the levels of paint in stock (10 different
colors as in Machine Programming), workers have to initiate replenishments of those colors that
fall below a certain threshold (i.e., 50). Specifically, workers have to demand the difference between
the actual levels of stocks and the maximum possible level (i.e., 100). Color replenishments are
only necessary for those colors that fall below the threshold.
Finally, participants have to bill customers in activity D (Billing [D]). Based on an email from
the Shipping Department, workers have to enter for each product ordered the amount indicated,
the colors included, as well as cuts needed to manufacture the wooden block. Moreover, they
have to indicate whether each article needs a product surcharge for S-shaped articles and a volume
discount. Volume discounts apply when the amount ordered per article exceeds ten. Workers have
to enter two products per bill into the billing system.
Appendix B provides screen shots of the different activities.

3.2 Experimental Parameters


All participants in our experiment perform the 20 tasks in a coherent order. That is, they perform
first all tasks of activity A, then all tasks of activity B, C, and, finally, of activity D. While one
experimental group performs five tasks per activity (A-A-A-A-A–B-B-B-B-B–C-C-C-C-C–D-D-D-
D-D), the other experimental group performs 6/4/3/7 tasks per activity (A-A-A-A-A-A–B-B-B-B–
C-C-C–D-D-D-D-D-D-D). We put tasks in a coherent order because Cardinaels and Labro (2008)
find task coherence to interact with ‘notification’ and ‘aggregation’. As we do not want our results
to be confounded by task coherence, especially in the condition when the amount of tasks per
activity varies, we decide for a coherent task order. After participants have performed their set
of 20 tasks, they provide estimates for the time spent on these activities. We introduce our main
experimental manipulation in this estimation part asking participants to provide estimates either
in traditional ABC or TD-ABC.
We apply a 2 × 2 between-subjects design varying cost system (traditional ABC versus TD-
ABC) and number of tasks per activity (equal number versus varying number). While we ask
8
Activity B, C, and D are based on the activities described in Cardinaels and Labro (2008).

10
participants in traditional ABC to provide time estimates according to the top-down approach on
overall time spent per activity in relative units, i.e., percentages; participants in TD-ABC estimate
time according to the bottom-up approach on the average duration of a task per activity in absolute
units, i.e., seconds. Thus, we ask participants in traditional ABC “What percentage of total work
time did you spend on activity X?”, whereas we as ask participants in TD-ABC “How much time
did you spend on average on activity X to perform it once?” reflecting the two different approaches.
Importantly estimates in traditional ABC have to add up to 100 %. In our second manipulation,
we vary number of tasks per activity (equal number -5- versus unequal number -6/4/3/7).
Moreover, we ask each participant to provide a second estimation that we use as a robustness
check for our analyses. Similar to Cardinaels and Labro (2008), we ask participants in traditional
ABC – for the second estimation – to provide an estimate of total work time (in minutes and
seconds) spent on the set of 20 tasks (“How much time did you spend in total on all 20 tasks?”).
Subsequently, they have to allocate this total estimatedtime to the 4 different activities guaranteeing
that total time is fully allocated (“Distribute estimated total time to the four activities without
residual.”). This estimation procedure resembles the top-down approach of traditional ABC in
which the reference point is total work time. On the other hand, we ask participants in TD-ABC –
for the second estimation – to provide an estimate (in minutes and seconds) on the total time spent
on each activity (“How much time did you spend in total on those tasks that belong to activity X?”).
That is the time spent on all tasks per activity.9 This approach reflects the bottom-up approach
with the single activities as direct reference point in TD-ABC.
Table 1 summarizes the questions asked in the first and second time estimation reflecting the
top-down approach in traditional ABC and the bottom-up approach in TD-ABC.

3.3 Measurement of Accuracy


Our dependent variable is measurement error. We calculate measurement error as Cardinaels and
Labro (2008) do. That is, “[t]otal measurement error is defined as the sum of the absolute devia-
tions of the estimated time spent on an activity versus the actual time for that activity” (Cardinaels
and Labro 2008, p. 745):

Total Measurement Error =


X
% estimated timei − % true timei

activity i

Summing up absolute errors per activity assures that over-estimation in one activity cannot com-
pensate for under-estimation in another activity. % true timei is defined as true overall time spent
on activity i (sum of all tasks belonging to activity i ) divided by true total time spent on all 20
9
In this second part, we indicate to participants in TD-ABC the amount of tasks spent on each activity
to have similar conditions as in the first estimation.

11
Table 1: Estimation Approaches

Traditional ABC Time-Driven ABC

1) What percentage of total work time did 1) How much time did you spend on aver-
you spend on activity X? age on activity X to perform it once?
......... % ......... seconds

2) How much time did you spend in total 2) How much time did you spend in total
on all 20 tasks? on those tasks that belong to activity X?
...... min ......... seconds ...... min ......... sec
Distribute estimated total time to the
four activities without residual!
...... min ......... seconds

tasks as measured by the computer:

% true timei =

100 ∗ true overall timei


true total time

While % estimated timei is simply taken as estimated for traditional ABC, we need to transform
estimates for TD-ABC in order to compare the two approaches to time-driven costing. To do so,
we follow two different procedures resulting in an error term that corrects for consistent estimation
bias and another error term that does not correct for bias. First, correcting for consistent under-
/over-estimation bias, we multiply estimates from TD-ABC with the number of tasks per activity
ni and divide this term by total estimated time. Dividing by total estimated time assures that
transformed estimates add up to 100 %:

% estimated timei in TD-ABC (corrected for consistent under-/over-estimation bias) =

100 ∗ ni ∗ estimated average timei


estimated total time

Second, allowing for consistent under-/over-estimation bias, we multiple again estimates from TD-
ABC with the number of tasks per activity ni , but divide this term by true total time spent on all
20 tasks:

12
% estimated timei in TD-ABC (not corrected for consistent under-/over-estimation bias) =

100 ∗ ni ∗ estimated average timei


true total time

Thus, consistent over-estimation (under-estimation) bias will result in estimates that are larger
(smaller) than those when correcting for estimation bias. When consistent estimation bias is present
in TD-ABC, transformed estimates do not add up to 100 %.
We derive error terms similarly for the second estimation (our robustness check). % true timei
is exactly calculated as before. For traditional ABC, we calculate proportions dividing estimated
time per activity by estimated total time. For TD-ABC, we differentiate again as indicated above
between an error term corrected for consistent estimation bias, and another error term not corrected
for consistent bias.

3.4 Participants and Procedures


We recruited 78 student participants from a participant pool for experiments at a large Western
European university. On average, we have 65.4 % male subjects and our participants have an
average age of 20.5 years. Moreover, they have an average work experience of 11.6 months.10
Similar to Cardinaels and Labro (2008), we recruit student workers as time estimations are tasks
that have to be done by many workers in companies throughout different work domains.
We invite participants to come to an experimental laboratory in which 20 participants may be
seated at the same time. Students receive a short briefing before the start of the experiment as
well as a participation number and are then led into the experimental laboratory. Participation
numbers are randomly mixed and are given to participants upon entry in the laboratory. In the
laboratory, subjects put all their belongings to the side and choose a desk with computer that shows
the experimental instructions. Participants are instructed to take nothing to the desk besides their
identification number. Desks are separated by dividing walls and participants cannot see what
other participants do. Participants begin the experiment by entering their participation number
on the first screen assigning them to the respective experimental condition. All instructions are
provided on computer screens and our sessions take approximately 1.25 hours.
In the first part of the experiment, we ask participants questions concerning their demographic
background. Moreover, we ask them to provide information on different personal characteristics
such as self-esteem and risk aversion. The second part introduces the case material. Participants
are told that they are supposed to perform 20 tasks in a quick and consistent manner. Time
is our measure of quality. They are informed that time is measured by the computer and that
their compensation is based on the time they need to perform the 20 tasks. Moreover, they
receive an extensive training on all four activities allowing them to become acquainted with the
10
Testing for randomization success, we do not find significant difference between experimental groups for
these variables.

13
different activities. In addition, we tell them that their compensation is tournament based. That
is, for each per experimental condition, those participants that perform in the first quartile get the
highest compensation. Compensation decreases with descending quartiles. Specifically, participants
performing in the first quartile receive CHF 24, CHF 22 in the second quartile, CHF 20 in the third
quartile, and, finally, CHF 18 in the last quartile. Importantly, we tell participants prospectively
that they have to estimate time. We do so, as Cardinaels and Labro (2008) find that time estimates
are more accurate when participants are prospectively notified about time estimation.
After the training session, participants perform the set of 20 tasks with pauses of 6 seconds
between subsequent tasks. Each task is shown on one screen and participants can only move on to
the next screen when all information is correctly provided. Thus, time continues to be measured
until all information is provided correctly reinforcing time as quality measure. While participants
perform the tasks, the computer measures the time that they spend on each task. After participants
have successfully finished all tasks they receive a questionnaire on the set of 20 tasks. When they
have finished this questionnaire they begin with the estimation task. Before subjects provide their
estimates, they receive detailed information on the estimation process. After participants have
finished the first estimation, they are asked to provide a second set of estimates. For both estima-
tions, participants are compensated based on the accuracy of their estimates. This compensation
is again done in tournament form. For each experimental condition, the first quartile with the
most accurate estimates receives CHF 14, the second quartile receives CHF 10, the third quartile
receives CHF 6, and, finally, the fourth quartile receives CHF 2. They receive this compensation
for each of the two estimations they perform. Finally, participants have to fill in a final question-
naire with manipulation and comprehension checks before they are told about their performance.
Before participants end the experiment, they also receive a short debriefing about the purpose of
the experiment.

3.5 Manipulation Checks


In case of number of tasks per activity, we notify participants three times about the sequence of
tasks before they actually perform the 20 tasks. Moreover, they perform the tasks in the respective
sequence. In addition, we ask participants before the first estimation tasks to indicate whether they
have realized each activity the same amount of times. Participants performing the same amount
of tasks per activity indicated that they had done so compared to participants not performing the
same amount of tasks (Ftasks (1,76) = 466.55; p < 0.001; mean is 6.55 for participants performing
the same amount of tasks per activity and 1.55 for participants not performing the same amount
of tasks on a seven-point Likert scale).
Concerning estimation approach, participants receive clear instructions how to estimate time
for each estimation approach. The perception of time estimation is also consistent with the way we
manipulate estimation approach. In particular, participants who are estimating time in traditional
ABC indicate to base their estimates more on total time (mean is 5.46) than participants estimating

14
time in TD-ABC (mean is 3.74; Festi (1,76) = 14.82; p < 0.001).

4 Results
First, we report descriptive results indicating absolute levels of measurement error. Consequently,
we test our hypotheses applying OLS regressions.11 Finally, we provide some additional analyses
investigating under-/over-estimation for the different time estimation approaches.

4.1 Descriptive Results


Table 2 provides descriptive results on average measurement error in the first estimation (per
experimental cell).12 ‘Symmetry’ indicates that the number of tasks across activities remains
equal, while ‘Asymmetry’ indicates an unequal number of tasks per activity. The top left of each
cell reveals average total measurement error. Average total measurement error results from the
sum of average measurement errors across activities (indicated at the bottom of each cell). In the
top right of each cell, we report average of log of total measurement error that we use subsequently
to test our hypotheses. For TD-ABC, figures in brackets refer to average measurement error when
not correcting for consistent under-/over-estimation bias. Figures normally displayed, represent
average measurement error corrected for consistent estimation bias.
Whereas average (log of) total measurement error in traditional ABC sums up to 18.44 %
(2.75), average total measurement error in TD-ABC amounts to 15.29 % (2.66) when estimates are
corrected for consistent estimation bias. However, when estimates are not corrected for consistent
estimation bias average total measurement error in TD-ABC amounts to 39.17 % (3.56) indicating
that consistent estimation bias largely impacts time estimates in TD-ABC. Moreover, average total
measurement errors only indicate that TD-ABC is more accurate when the amount of tasks per
activity varies (25.30 % versus 16.75 %). This difference between measurement errors is more
pronounced for activities C (7.0 % versus 3.8 %; difference is 3.2 %) and D (7.8 % and 5.3 %,
difference is 2.5 %) than for activities A (5.7 % versus 3.9 %, difference is 1.8 %) and B (4.8 %
and 3.7 %, difference is 1.1 %) indicating that the effect is stronger for activities varying more
in the amount of tasks per activity (6/4/3/7). Moreover, the difference between ‘Symmetry’ and
‘Asymmetry’ is more pronounced for traditional ABC (11.92 % versus 25.30 %, difference is 13.38
%) than for TD-ABC (13.90 % versus 16.75 %, difference is 2.85 %) when estimates are corrected
for consistent bias. This statement also holds when estimates are not corrected for consistent bias
in TD-ABC (38.08 % versus 40.31 %, difference is 2.23 %).
Table A.1 in Appendix A reports descriptive results for the second estimation. Descriptive
results are very similar in both estimations indicating that results remain robust.
11
OLS regression provides same results as full-factorial ANOVA when independent variables are measured
as dummy variables.
12
Descriptive results for the second estimation are attached in Appendix A.

15
Table 2: Descriptive Results - Error First Estimation

ERROR Symmetry Asymmetry

traditional 11.92 n=20 2.46 25.30 n=19 3.05 18.44


ABC A B C D A B C D
3.1 3.5 3.3 2.1 5.7 4.8 7.0 7.8 2.75

13.90 n=20 2.58 16.75 n=19 2.75 15.29


TD-ABC (38.08) (3.53) (40.31) (3.58) (39.17)
A B C D A B C D
3.4 2.9 3.4 4.2 3.9 3.7 3.8 5.3 2.66
(8.9) (8.4) (3.7) (11.2) (11.3) (8.3) (6.7) (14.0) (3.56)

12.91 2.52 21.02 2.90


(25.00) (3.00) (32.81) (3.32)
Note: This table indicates, for each of the four experimental cells, average total measurement error (top
left in each cell - in bold), number of subjects (top middle in each cell), average log of total measurement
error (top right in each cell - in bold), and average absolute measurement error per activity (bottom line in
each cell). Notice that the sum of average absolute measurement errors across activities adds up to average
total measurement error. Moreover, we indicate errors in TD-ABC both corrected for consistent estimation
bias and not corrected for consistent bias (in brackets).

4.2 OLS Regression Results


We test our hypotheses conducting OLS regression analyses. To this end, we code ‘TD-ABC’ to
equal zero (one) when estimates are provided in traditional ABC (TD-ABC) and ‘Asymmetry’ to
equal zero (one) when number of tasks per activity does (not) vary. Based on these variables, we
code the interaction variable ‘Interaction’ to equal one when both estimates are provided in TD-
ABC and number of tasks per activity varies, and zero otherwise. Moreover, we use the natural
logarithm of measurement errors (instead of measurement error itself) as our dependent variable
to account for outliers. Table 3 indicates the results of different OLS regressions for the first
estimation. Estimates in TD-ABC in Table 3 are corrected for consistent under-/over-estimation
bias. While we show results for log of total measurement error in model (1); model (2), (3), (4),
and (5) reveal results for log of measurement error concerning the four different activities.
We find significant support for our third hypotheses. That is, compared to estimates in TD-
ABC, estimates in traditional ABC are less accurate when the amount of tasks per activity varies.
This is indicated by the negative significant interaction effect (b = - 0.422, t = - 1.582, p < 0.1 – one-
tailed) in model (1). Although only marginally significant, we have confidence in this interaction
effect for two reasons. First, the effect is more pronounced for activities for which the amounts of
tasks vary to a larger extent, i.e., activity C and D, as indicated in model (3) and (4). Second,

16
Table 3: OLS Regression Results – Measurement Error Type 1 – First Estimation
(1) (2) (3) (4) (5)
VARIABLES Total Activity A Activity B Activity C Activity D

TD-ABC 0.117 0.0278 -0.125 0.0258 0.543***


(0.626) (0.130) (-0.575) (0.115) (2.560)
ASYMMETRY 0.593*** 0.459** 0.126 0.482** 0.936***
(3.145) (2.121) (0.571) (2.115) (4.357)
INTERACTION -0.422* -0.242 -0.0125 -0.428* -0.851***
(-1.582) (-0.790) (-0.039) (-1.329) (-2.799)

Constant 2.459*** 1.213*** 1.330*** 1.269*** 0.971***


(18.67) (8.026) (8.628) (7.984) (6.475)

Observations 78 78 78 78 78
R-squared 0.131 0.073 0.017 0.073 0.212
Note: ***,**,* denote significance at the 1%, 5%, and 10% level (one-tailed) and parentheses
show the corresponding t-statistics. The table indicates results for OLS regressions of the
log of measurement error on ‘TD-ABC’, ‘ASYMMETRY’ and their ‘INTERACTION’ term.
‘TD-ABC’ takes values of 1 for TD-ABC and values of 0 for traditional ABC. ‘ASYMME-
TRY’ takes values of 1 for unequal amounts of tasks per activity and 0 for equal amounts
of tasks per activity. ‘INTERACTION’ takes values of 1 for TD-ABC and Asymmetry
equalling 1 and values of 0 otherwise.

17
evidence from our robustness check (second estimation) indicates similar results.13
Providing more conclusive evidence for hypothesis 3, we see from the coefficient for ’Asymme-
try’ in Table 3 (b = 0.593, t = 3.145, p < 0.01 – two tailed) that estimates in traditional TD-ABC
are significantly less accurate when number of tasks per activity varies. On the other hand, addi-
tional analysis, i.e., testing the joint effect of ‘Asymmetry’ and the ‘Interaction’ term, reveals that
estimates in TD-ABC are not significantly different when the number of tasks per activity varies
(b = 0.171, F(1, 74) = 0.39, p = 0.3666 – two-tailed). These results provide evidence for our third
hypotheses following our argumentation that the accuracy of estimates in TD-ABC is not affected
by the amount of times an activity is performed. Contrary, the accuracy of estimates in traditional
ABC is largely affected by the amount of times an activity is performed.
Moreover, the coefficient of ‘TD-ABC’ in Table 3 – model (1) – (b = 0.117, t = 0.626, p =
0.533 – two-tailed) indicates that estimates in TD-ABC are not (significantly) more accurate than
estimates in traditional ABC when the amount of tasks remains equal. This result provides contrary
evidence to our first hypothesis that percentage estimates are cognitively more challenging even
when the amount of tasks remains equal across activities. Thus, we reject hypothesis 1.14
In order to test our second hypothesis, we look at Table 4 displaying OLS regression results for
errors – not corrected for consistent under-/over-estimation bias in TD-ABC. Again the interaction
effect is negatively significant and here more pronounced (b = - 0.542, t = - 2.023, p < 0.05 – two
tailed). To analyze whether hypothesis 2 holds, we, therefore, look at two effects in more detail:
(1) The effect of ‘TD-ABC’ to account for differences between traditional ABC and TD-ABC when
the amount of tasks per activity remains equal (b = 1.074, t = 5.746, p < 0.05 – two tailed)
and (2) the joint effect of ‘TD-ABC’ and the ‘Interaction’ term to account for differences between
traditional ABC and TD-ABC when the amount of tasks per activity varies (b = 0.532, F(1,74)=
7.70, p < 0.01 – two tailed). Both effects indicate that estimates in TD-ABC are significantly less
accurate than estimates in traditional ABC providing support for our second hypothesis.
Furthermore, we find additional support for our third hypothesis. Whereas, estimates in TD-
ABC do not become significantly less accurate when number of tasks per activity varies (b = 0.051,
F(1,74) = 0.07, p = 0.7876 – two-tailed), accuracy of estimates in traditional ABC significantly
deteriorates with task asymmetry (b = 0.593, t = 3.132, p < 0.01 – two tailed).

4.3 Additional Analyses


Moreover, we find strong support that participants in TD-ABC heavily underestimate activity du-
rations. This result is indicated in Table 5 (first estimation) and Table 6 (second estimation). Both
tables show the average measurement error for the different activities per experimental cell (not in
13
Results for total measurement error (without using the log) are highly significant (b = - 10.538, t = -
2.273, p = 0.026 – two-tailed).
14
This result is surprising, especially as we find contrary evidence in a similar study investigating the
accuracy of TD-ABC and traditional ABC.

18
Table 4: OLS Regression Results – Measurement Error Type 2 – First Estimation
(1) (2) (3) (4) (5)
VARIABLES Total Activity A Activity B Activity C Activity D

TD-ABC 1.074*** 0.926*** 0.667*** 1.052*** 1.166***


(5.746) (4.224) (2.798) (4.882) (5.198)
ASYMMETRY 0.593*** 0.459** 0.126 0.482** 0.936***
(3.132) (2.070) (0.522) (2.206) (4.120)
INTERACTION -0.542** -0.360 -0.156 -0.864*** -0.537**
(-2.023) (-1.146) (-0.455) (-2.800) (-1.671)

Constant 2.459*** 1.213*** 1.330*** 1.269*** 0.971***


(18.59) (7.832) (7.893) (8.328) (6.122)

Observations 78 78 78 78 78
R-squared 0.386 0.270 0.142 0.250 0.412
Note: ***,**,* denote significance at the 1%, 5%, and 10% level (one-tailed) and parentheses
show the corresponding t-statistics. The table indicates results for OLS regressions of the
log of measurement error on ‘TD-ABC’, ‘ASYMMETRY’ and their ‘INTERACTION’ term.
‘TD-ABC’ takes values of 1 for TD-ABC and values of 0 for traditional ABC. ‘ASYMME-
TRY’ takes values of 1 for unequal amounts of tasks per activity and 0 for equal amounts
of tasks per activity. ‘INTERACTION’ takes values of 1 for TD-ABC and Asymmetry
equalling 1 and values of 0 otherwise.

19
Table 5: Under-/Over-Estimation – First Estimation

ERROR Symmetry Asymmetry

A B C D A B C D
ABC -1.30 2.87 -1.72 0.15 -1.99 2.63 6.17 -6.81
(70%) (30%) (55%) (50%) (68%) (32%) (21%) (79%)

A B C D A B C D
TD-ABC -6.62 -5.84 -8.67 -7.83 -5.05 -4.52 -4.12 -7.40
(85%) (90%) (90%) (80%) (79%) (79%) (74%) (74%)

Note: This table indicates, for each of the four experimental cells, the average measurement er-
ror per activity (in bold) and the percentage of participants underestimating time per activity
(in brackets). A negative average measurement error reflects underestimation, a positive aver-
age measurement error reflects overestimation. Notice that the sum of average measurement
errors across activities adds up to zero in traditional ABC.

absolute units and not corrected for consistent estimation bias). In Table 5 average measurement
errors per activity add up to zero in traditional ABC as percentage estimates reflect a zero-sum
game. However, average measurement errors in TD-ABC are all largely negative indicating that
participants suffer from consistent under-estimation bias. Figures in brackets show the percent-
ages of participants per cell that actually under-estimate time. Again, figures provided for TD-ABC
indicate that the vast majority consistently under-estimates time (more than 70 % for each activity.
Table 6 reflects the same measures for the second estimation investigating consistent estimation
bias with the sole difference that we base estimated time in traditional ABC on total true time
allowing also estimates in traditional ABC to suffer from consistent under-/over-estimation bias.
Consistent with Cardinaels and Labro (2008), we find that participants heavily overestimate time
in the second estimation in traditional ABC. Interestingly, participants in TD-ABC carry their
underestimation bias over to the second estimation when they estimate the total duration for all
tasks per activity. This result is somewhat puzzling given the fact that psychology literature argues
that short durations below 2 to 5 minutes tend to be over-estimated compared to longer durations
(above this threshold) which tend to be under-estimated.

5 Discussion
In sum, our results suggest that time should only be estimated in TD-ABC when corrected for
consistent under-/over-estimation. However, when estimates in TD-ABC are not corrected for con-
sistent estimation bias, the benefits of TD-ABC are likely to be eliminated by this bias. Moreover,
we provide evidence showing that estimates in TD-ABC are not affected by the amount of times

20
Table 6: Under-/Over-Estimation – Second Estimation

ERROR Symmetry Asymmetry

A B C D A B C D
ABC 9.92 18.17 15.05 13.41 13.37 14.51 17.75 12.44
(35%) (15%) (25%) (15%) (32%) (21%) (21%) (42%)

A B C D A B C D
TD-ABC -6.27 -5.23 -7.75 -6.45 -2.83 -2.13 -1.34 -6.81
(90%) (90%) (85%) (80%) (68%) (58%) (58%) (63%)

Note: This table indicates, for each of the four experimental cells, the average measurement
error per activity (in bold) and the percentage of participants underestimating time per activ-
ity (in brackets). A negative average measurement error reflects underestimation, a positive
average measurement error reflects overestimation.

an activity is performed. However, estimates in traditional ABC are largely affected in their ac-
curacy when the amount of tasks per activity varies. This supports the idea that TD-ABC is a
bottom-up approach allowing employees to directly focus on the single activities when estimating
time. Finally, we provide evidence for different estimation biases in the two estimation approaches,
i.e., under-estimation in TD-ABC and over-estimation in traditional ABC. However, generalizing
on the directions of consistent estimation bias has to be done with care. Nevertheless, we deem
further studies in this direction of interest for research in psychology.
We believe that we contribute to accounting literature in different aspects. To our best knowl-
edge, we are the first to directly compare the different approaches to time-driven costing in an
empirical study. More importantly, we find that task heterogeneity is not an issue in TD-ABC,
whereas it is highly problematic in traditional ABC. In addition, our results suggest that designers
of cost systems should rely on time estimation as suggested in TD-ABC only if corrected for con-
sistent under-/over-estimation bias. Therefore, conclusions about practical capacity in TD-ABC
need to be drawn with caution.
Moreover, further avenues for experimental research in time-driven costing seem worthwhile to
pursue. We restrict our analyses to activities on a rather aggregated level. Moreover, an interaction
of task coherence with number of tasks performed per activity seems to be worthwhile to investigate.
Our results must be treated with caution concerning two aspects: (1) Our sample size is rather
small and more data is needed to underpin our results. (2) Researchers should not conclude
from our results whether activity durations are generally under- or over-estimated when employees
provide best estimates. Under- and over-estimation biases seem to depend very much on the general
duration of a task.
Finally, we deem our results to be highly interesting for practitioners applying time-driven

21
costing.

22
References
Balakrishnan, R., S. Hansen, and E. Labro. 2011. Evaluating heuristics used when designing
product costing systems. Management Science 57 (3): 520 – 541.

Bol, J. C. 2008. Subjectivity in compensation contracting. Journal of Accounting Literature 27: 1


– 27.

Cardinaels, E. and E. Labro. 2008. On the determinants of measurement error in time-driven


costing. The Accounting Review 83 (3): 735 – 756.

Christensen, J. and J. S. Demski. 1995. The classical foundations of ’modern’ costing. Management
Accounting Research 6 (1): 13 – 32.

Dalci, I., V. Tanis, and L. Kosan. 2010. Customer profitability analysis with time-driven activity-
based costing: A case study in a hotel. International Journal of Contemporary Hospitality
Management 22 (5): 609 – 637.

Datar, S. and M. Gupta. 1994. Aggregation, specification and measurement errors in product
costing. The Accounting Review 69 (4): 567 – 591.

Demeere, N., K. Stouthuysen, and F. Roodhooft. 2009. Time-driven activity-based costing in an


outpatient clinic environment: Development, relevance and managerial impact. Health Policy
92 (2): 296 – 304.

Everaert, P., W. Bruggeman, G. Sarens, S. R. Anderson, and Y. Levant. 2008. Cost modeling
in logistics using time-driven ABC: Experiences from a wholesaler. International Journal of
Physical Distribution & Logistics Management 38 (3): 172 – 191.

Everaert, P., G. Cleuren, and S. Hoozée. 2012. Using time-driven ABC to identify operational
improvements: A case study in a university restaurant. Journal of Cost Management 26 (2): 41
– 48.

Fortin, C. and R. Rousseau. 1998. Interference from short-term memory processing on encoding
and reproducing brief durations. Psychological Research 61 (4): 269 – 276.

Hoozée, S. and W. Bruggeman. 2010. Identifying operational improvements during the design
process of a time-driven ABC system: The role of collective worker participation and leadership
style. Management Accounting Research 21 (3): 185 – 198.

Hoozée, S., L. Vermeire, and W. Bruggeman. 2012. The impact of refinement on the accuracy of
time-driven ABC. Abacus 48 (4): 439 – 472.

23
Horngren, C. T., S. M. Datar, G. Foster, M. Rajan, and C. Ittner. 2008. Management and Cost
Accounting: A Managerial Emphasis. Upper Saddle River, NJ: Prentice Hall, 13th edition.

Ittner, C. D. 1999. Activity-based costing concepts for quality improvement. European Management
Journal 17 (5): 492 – 500.

Kahneman, D., P. Slovic, and A. Tversky. 1982. Judgment under Uncertainty: Heuristics and
Biases. Cambridge, UK: Cambridge University Press.

Kaplan, R. S. and S. R. Anderson. 2004. Time-driven activity-based costing. Harvard Business


Review 82 (11): 131 – 140.

Kaplan, R. S. and S. R. Anderson. 2007. Time-Driven Activity-Based Costing: A Simpler and


More Powerful Path to Higher Profits. Boston, MA: Harvard Business Press.

Labro, E. and M. Vanhoucke. 2007. A simulation analysis of interactions among errors in costing
systems. The Accounting Review 82 (4): 939 – 962.

Labro, E. and M. Vanhoucke. 2008. Diversity in resource consumption patterns and robustness of
costing systems to errors. Management Science 54 (10): 1715 – 1730.

Merchant, K. A. and W. A. Van der Stede. 2007. Management Control Systems: Performance
Measurement, Evaluation and Incentives. Essex, England: Prentice Hall, 2nd edition.

Noreen, E. 1991. Conditions under which activity-based cost systems provide relevant costs. Journal
of Management Accounting Research 3 (4): 159– 168.

Öker, F. and H. Adigüzel. 2010. Time-driven activity-based costing: An implementation in a


manufacturing company. Journal of Corporate Accounting & Finance 22 (1): 75 – 92.

Pernot, E., F. Roodhooft, and A. Van den Abbeele. 2007. Time-driven activity-based costing
for inter-library services: A case study in a university. The Journal of Academic Librarianship
33 (5): 551 – 560.

24
Table A.1: Descriptive Results - Error Second Estimation

ERROR Symmetry Asymmetry

13.18 n=20 2.52 21.72 n=19 3.00 17.34


ABC A B C D A B C D
3.2 4.2 3.7 2.1 5.7 4.0 5.4 6.6 2.75

14.50 n=20 2.62 16.57 n=19 2.76 15.51


(39.29) (3.55) (38.24) (3.52) (38.78)
TD-ABC A B C D A B C D
3.5 2.6 4.1 4.3 2.9 3.7 4.0 6.0 2.69
(9.1) (9.5) (12.0) (8.7) (10.4) (8.0) (6.0) (13.8) (3.53)

13.84 2.57 19.14 2.88


(26.23) (3.03) (29.98) (3.26)
Note: This table indicates, for each of the four experimental cells, average total measurement error (top
left in each cell - in bold), number of subjects (top middle in each cell), average log of total measurement
error (top right in each cell - in bold), and average absolute measurement error per activity (bottom line in
each cell). Notice that the sum of average absolute measurement errors across activities adds up to average
total measurement error. Moreover, we indicate errors in TD-ABC both corrected for consistent estimation
bias and not corrected for consistent bias (in brackets).

Appendix A: Additional Tables

Appendix B: Screen Shots of Activities

25
Table A.2: OLS Regression Results – Measurement Error Type 1 – Second Estimation
(1) (2) (3) (4) (5)
VARIABLES Total Activity A Activity B Activity C Activity D

TD-ABC 0.0978 -0.00945 -0.402** -0.00516 0.578***


(0.571) (-0.0463) (-1.952) (-0.0227) (2.718)
ASYMMETRY 0.477*** 0.456** -0.0101 0.105 0.754***
(2.749) (2.202) (-0.0485) (0.456) (3.504)
INTERACTION -0.340* -0.461* 0.273 -0.0653 -0.530**
(-1.387) (-1.574) (0.926) (-0.201) (-1.740)

Constant 2.522*** 1.257*** 1.460*** 1.396*** 0.971***


(20.84) (8.701) (10.02) (8.708) (6.462)

Observations 78 78 78 78 78
R-squared 0.103 0.091 0.062 0.004 0.194
Note: ***,**,* denote significance at the 1%, 5%, and 10% level (one-tailed) and parentheses
show the corresponding t-statistics. The table indicates results for OLS regressions of the
log of measurement error on ‘TD-ABC’, ‘ASYMMETRY’ and their ‘INTERACTION’ term.
‘TD-ABC’ takes values of 1 for TD-ABC and values of 0 for traditional ABC. ‘ASYMME-
TRY’ takes values of 1 for unequal amounts of tasks per activity and 0 for equal amounts
of tasks per activity. ‘INTERACTION’ takes values of 1 for TD-ABC and Asymmetry
equalling 1 and values of 0 otherwise.

26
Table A.3: OLS Regression Results – Measurement Error Type 2 – Second Estimation
(1) (2) (3) (4) (5)
VARIABLES Total Activity A Activity B Activity C Activity D

TD-ABC 1.026*** 0.866*** 0.605*** 0.929*** 1.088***


(5.739) (3.875) (2.618) (4.180) (4.470)
ASYMMETRY 0.477*** 0.456** -0.0101 0.105 0.754***
(2.633) (2.012) (-0.0432) (0.465) (3.061)
INTERACTION -0.503** -0.451* -0.152 -0.561** -0.343
(-1.964) (-1.408) (-0.457) (-1.762) (-0.984)

Constant 2.522*** 1.257*** 1.460*** 1.396*** 0.971***


(19.96) (7.952) (8.930) (8.878) (5.646)

Observations 78 78 78 78 78
R-squared 0.374 0.216 0.127 0.223 0.351
Note: ***,**,* denote significance at the 1%, 5%, and 10% level (one-tailed) and parentheses
show the corresponding t-statistics. The table indicates results for OLS regressions of the
log of measurement error on ‘TD-ABC’, ‘ASYMMETRY’ and their ‘INTERACTION’ term.
‘TD-ABC’ takes values of 1 for TD-ABC and values of 0 for traditional ABC. ‘ASYMME-
TRY’ takes values of 1 for unequal amounts of tasks per activity and 0 for equal amounts
of tasks per activity. ‘INTERACTION’ takes values of 1 for TD-ABC and Asymmetry
equalling 1 and values of 0 otherwise.

27
Table B.1: Screen Shot - Activity A

28
Table B.2: Screen Shot - Activity B

29
Table B.3: Screen Shot - Activity C

30
Table B.4: Screen Shot - Activity D

31

You might also like