Professional Documents
Culture Documents
Performance on
Compensated
Task Dimension
+
+
Performance on
Controlled Task
Dimension
+
Dimension
Figure 1 Conceptual model
Notes:
The conceptual model presents our expectation for how controls can inuence compensated goals,
controlled goals, and intrinsic motivation. Each type of control has differential characteristics
in terms of the timeliness of feedback or autonomy of the control. Timeliness of feedback and
autonomy of the control are both expected to positively inuence the salience of controlled
goals which in turn positively inuences accuracy and negatively inuences speed. The restric-
tion of autonomy also negatively affects an employees intrinsic motivation.
Effects of Preventive and Detective Controls 439
CAR Vol. 29 No. 2 (Summer 2012)
Student participants are appropriate in this experiment for two reasons: (1) the same
type of task is performed by students (or people of similar background) in the real
world and (2) previous psychology research does not suggest age experience should matter
in this task. With regards to the rst point, our task, data entry, is often performed by stu-
dents as part-time or full-time employment. Further, typing is a standard skill required of
and practiced by college students. With regard to the second point, we are unaware of any
research on goals, frames, and salience that would suggest our results would be systemati-
cally biased by using student participants for our task. The phenomena we are examining
are general in nature and are unlikely to differ based on age experience.
Experimental task and manipulations
We employ a 1 4 experimental design. The four conditions we examine vary the type of
formal control (preventive and detective), which implicitly varies the extent to which the
control limits the participants autonomy and the timeliness of the feedback it provides.
We describe each condition after introducing the experimental task.
The experimental task consisted of seven parts: the introduction, demographic ques-
tions, a pretest task, instructions (and the introduction of the experimental manipulation
for that participant), a test of understanding, the experimental (data entry) task, and addi-
tional questions. Participants arrived at a computer lab that was equipped with standard-
ized computers and keyboards. Upon arrival, the participants logged on to a secure
Internet website and read a brief introduction to the experimental task. The introduction
stated (emphasis as in the experiment):
For todays study you will perform a data entry task (i.e., a typing tutor type task). We
want you to type as quickly and as accurately as possible. You will be compensated based
on how fast you enter the data on the screenthe faster you type, the more money you
will make.
We informed participants that they would be compensated based on how fast they
entered data relative to a xed, predetermined benchmark (based on pilot testing) and not
relative to the other participants, and that payoffs would range from $5.00 to $15.00 with
an expected average payoff of $10.00.
6,7
Because participants were paid based on the speed
with which they entered data, we use data entry speed as a proxy for their performance on
the compensated dimension of the task. Similarly, because participants were controlled on
accuracy, we use data entry accuracy as a proxy for performance toward the controlled
dimension of the task. Next, we gave participants specic instructions about how to com-
plete the task, which included an example.
6. The scale, based on extensive pretesting, ranged from payment of $6 for participants who entered at least
40 words per minute to $15 for those whose data entry speed exceeded 77.4 words per minute (with a $10
payoff for 60 words per minute). The scale was designed to create a uniform distribution of payouts over
the advertised payout range. Consistent with several studies in experimental economics, we did not inform
participants of the exact conversion rate between their actions and their payouts, but did provide the
range and expected average payoff that would be earned (see, e.g., Bowlin, Hales, and Kachelmeier 2008).
7. We did not disclose the specic performance benchmark to participants because we wanted to hold con-
stant the economic incentive to perform well on both typing speed and accuracy across conditions. Given
our expectation that speed and accuracy would differ across conditions, revealing specic benchmarks
could cause perceptions of expected earnings to differ by condition, while the intent of this study was to
understand the effect of controls irrespective of perception of expected earnings. In order to isolate the
effect of controls, we wanted to hold constant perceptions of pay and therefore chose to use a do-your-
best goal rather than provide specic benchmarks. Future research may wish to investigate how specic
benchmarks interact with controls.
440 Contemporary Accounting Research
CAR Vol. 29 No. 2 (Summer 2012)
After the introduction, participants answered demographic questions and began the
pretest task. In the pretest, all participants entered the same 1,011 characters (12 lines of
data on three screens). No controls were implemented on any characters. This task was
meant to establish a baseline of performance for each participant. Participants were paid
based on their performance during the pretest task in the same manner as they were com-
pensated for the main task.
After nishing the pretest task, we provided additional instructions to participants.
First, we reminded participants of the importance of both accuracy and speed and the gen-
eral rules for performing the data entry task. Second, we introduced one of the four exper-
imental manipulations (varied among participants).
1. No control condition: no controls were implemented and the task was identical to
the pretest task.
2. Preventive control condition: controls were implemented on some characters such that
if participants did not enter the correct character: (1) they could not advance to the
next character until they had entered the correct character and (2) the incorrectly
entered character would turn red. As part of the instructions, participants were given
an example and told to try entering the wrong characters to observe what happened.
3. Detective control, immediate feedback condition: controls were implemented on some
characters such that, if participants entered the wrong character, the missed charac-
ter would turn red but the cursor would advance (i.e., participants could not correct
the mistake, but would immediately see that they had made the mistake). As part
of the instructions, participants were instructed to try entering the wrong characters
on an example line of text.
4. Detective control, delayed feedback condition: controls were implemented on some
characters such that, if participants entered the wrong character, the missed character
would turn red after the participant nished entering all of the characters on the
screen. Specically, the participant would nish entering all text on a screen, hit con-
tinue, and a second screen would show what characters (with controls) the partici-
pant had missed. Similar to the detective, immediate feedback condition, participants
in this condition could not correct their mistakes. As part of the instructions, partici-
pants were instructed to try entering the wrong characters on an example line of text.
Once participants nished reading the second set of instructions, they began the data
entry task (i.e., the main experimental task). Participants completed 3,209 characters (38 lines,
eight screens) of data entry.
8
Formal controls were assigned to a random set of approximately
20 percent of the characters. The formal controls were imposed on the same characters in all
conditions. Next, participants answered debrieng questions. Finally, participants were paid
based on their performance on the compensated dimension of the task (i.e., data entry speed).
Measures
To test our hypotheses, we develop measures of data entry accuracy (performance on the
controlled dimension), data entry speed (performance on the compensated dimension), and
intrinsic motivation. To measure accuracy, we test whether participants enter the correct
character for each character that does not have a formal control.
9
We examine only those
8. The data entry material for both pretesting and testing came from the company history and a news release
of Millenniata, Inc., a start-up company that produces archival quality DVDs (and DVD writers) that
will preserve data for generations (Millenniata 2011). The material described the history of Barry Lunt,
the originator and founder of the company, how he developed the technology, and the benets of the
technology.
9. To compare the same characters in all conditions, for the no control condition we only analyze the equiva-
lent characters for which there was no control in place in the conditions with controls.
Effects of Preventive and Detective Controls 441
CAR Vol. 29 No. 2 (Summer 2012)
characters without formal controls because, by denition, participants must enter the cor-
rect character in the preventive control condition, which ensures the correct character will
eventually be entered. Further, examining characters for which controls are not in place
allows us to examine how formal controls inuence behavior in a task in which controls
are only activated at key times (i.e., it provides insights into how the existence of a formal
control inuences the control environment). To observe differences in accuracy, we prevent
the alteration of any uncontrolled characters in all conditions and in both pretask and task
sections of the experiment.
10
Participants have only a single opportunity to enter the correct character for uncon-
trolled characters. For each participant, we calculate the ratio of errors to total uncon-
trolled characters in the experimental task and subtract it from the participants ratio of
errors to total uncontrolled characters in the pretest task.
11
A positive value for this metric
suggests the participants performance improved relative to the pretest task and a negative
value suggests they performed worse. To restate using mathematical notation, the accuracy
measure we use for testing is computed as follows:
Accuracy
Total uncontrolled errors made in pretest task
Total uncontrolled characters pretest