1995 Joint Effects of Budgetary Slack and Task Uncertainty 63
If so, then the extent to which task uncertainty influences departmental performance may
be influenced by the level of slack
The frequent conceptualization, however, of task uncertainty as comprising the dimensions
of task difficulty and task variability? (Daft, 1983; Daft and Macintosh, 1981; Macintosh,
1981; Perrow, 1967, 1970; Van de Ven and Delbecq, 1974] presents impediments not only
in terms of theoretically predicting task uncertainty’s relation with subunit performance in
settings of varying degrees of budgetary slack, but also in the empirical assessment of that
relation? Because work units perform tasks that can vary in task difficulty but not in task
variability, and vice versa (Van de Ven and Delbecq, 1974; Brownell and Dunk, 1994], the
theoretical roles of each of the dimensions require evaluation, together with an empirical
assessment of their effects in work units.*
Task difficulty refers to the degree to which procedures lay down the series of steps that
must be performed for successful work completion [Van de Ven and Delbecq, 1974]. When
task difficulty is low, work can be reduced to objective, mechanical steps involving the use
of standard operating procedures, instructions, manuals, or programs [Daft and Macintosh,
1978; Macintosh, 1981]. If tasks are well understood prior to their execution, as is the case
when task difficulty is low, then the steps comprising those tasks can be effectively
planned, mapped and encoded by budgetary control systems [Ashby, 1964; Buckley, 1969;
Cunningham, 1978; Galbraith, 1973; Merchant, 1984].
Merchant [1984] noted that budgeting becomes more reliable when budgets focus on
clearly demonstrable and programmable input/output relations. In such circumstances,
Bruns and Waterhouse [1975] argued that budgets can be designed and implemented that
reflect more accurately the better known and predictable nature of the work of subunits.
Abemethy and Stoelwinder [1991] suggested that budgets may be used as a form of
programmed contro! through the specification of desired product input/output relations.
They indicated, however, that the extent to which budgets can reflect such relations and
reflect performance is dependent upon tasks being routine in nature, which they described
as low in task difficulty and variety. Tasks low in difficulty are well understood and can be
structured to facilitate efficient task performance through the identification of clear criteria
for task performance [Gresov er al., 1989]. Given such a level of task knowledge, precise
quantities of resource inputs can be ascertained to allow for the required outputs.
Consequently, there is little room for slack in such settings. This expectation is consistent
with Merchant (1985b], who pointed out that the ability to set tight and accurate standardsand to measure performance precisely provides the opportunity to prevent the introduction
of slack.
By contrast, if task difficulty is high, there is a shortage of task-specific information
[Gresov er a/., 1989). Work is not analyzable and solutions are hard to identify, because
there are no objective, computational procedures or techniques available to specify
precisely what must be done to perform those tasks (Daft, 1983: Daft and Macintosh, 1981].
Successful problem solving and decision making requires greater expertise, intuition and
judgment, instead of standard operating procedures [Daft, 1983; Daft and Macintosh, 1981;
Perrow, 1967; Van de Ven and Delbecq, 1974]. Such tasks are hard to decompose, and
therefore difficult to evaluate in terms of their costs, benefits, and outcomes [Daft and
Macintosh, 1981]. Lukka [1988] argued that when uncertainty is high, the quantum of
resources required to perform tasks is not clear ex ante, thereby giving rise to the utility of
budgetary slack as a means of redressing the prospect of running short of critical resources.
Since the resource inputs required 10 perform tasks high in difficulty cannot be ascertained
precisely a priori, budgets cannot be constructed to map and encode those tasks as
effectively as when task difficulty is low. Attempts to specify rules and procedures hinder
rather than aid task performance because tasks in such circumstances cannot be codified
precisely or parsimoniously [Gresov er al., 1989]. If insufficient resources are programmed
for highly difficult tasks, then they may not be completed, properly, or on time. In other
words, the downside of low-balling budget estimates is that subunit performance may be
compromised. The provision of resource flexibility through the availability of slack may
enable difficult tasks to be more effectively performed than if those resources had not been
scheduled, Consequently, as suggested by Cyert and March (1963), slack may play an
adaptive and stabilizing role through’the provision of a pool of emergency resources that
can facilitate performance when task difficulty is high. If so, then the detrimental effect of
task difficulty on subunit performance is likely to be less pronounced when slack is high in
contrast to when it is low. This hypothesis is stated in its null form as follows.
Hl: Budgetary slack and task difficulty do not interact to affect subunit performance.
Task Variability
ask variability refers to the extent to which novel and unexpected events are encountered
in the planning and conversion processes that necessitate the use of different procedures or
methods to carry out the work (Daft and Macintosh, 1981; Perrow, 1967; Van de Ven and
Delbecq, 1974]. When variability is low, the array of tasks can be predicted with
‘considerable certainty and few novel or exceptional tasks are undertaken. By contrast,
when task variability is high, unexpected events and problems are often encountered (Daft,
1983; Daft and Macintosh, 1981; Macintosh, 1981; Withey eto, 1983).
‘slack influencing the relation between task variability and performance is
». Brownell and Dunk {1991] argued that in a low task difficulty but high
even though the array of specific tasks ahead cannot be foreseen (high
it arises, is analyzable since it is straightforward to specify theTelevant inputs necessary 10 achieve a desired outcome (low difficulty). They proposed that
flexible manufacturing systems emphasizing CAD/CAM provide an example of the
independence of task variability and task difficulty. Although each unit of output can
constitute a lot size of one. the customization of the production process is dependent on it
being computerized. To enable that software to be written, each of the manufacturing steps
must be clearly identifiable and programmable. Similarly, Goldas [1986] pointed out that a
computer controlled machine can teadily make 12 of the one type of product in a row, or
randomly produce one each of 12 different designs, provided the software for each of those
12 designs is in place
This example illustrates that if task difficulty is low, the level of task variability does not
influence the capacity of the machine to perform its required array of tasks. Similarly, if
task difficulty is high, the levcl of task variability would not affect the extent to which the
machine performs effectively. Whether task variability is high or low makes no statement
of the level of task difficulty, given the independence of the two constructs. Hence, the
ability to specify the relevant set of inputs in the budget to achieve a desired outcome (a
question of task difficulty) and the argued role for budgetary slack is independent of the
level of task variability. Since slack is the incorporation of amounts in the budget to make
it easier to attain [Merchant, 1985a; Lukka, 1988; Young, 1985], then a necessary condition
for slack is that a budgeted array of tasks have been planned. Referring back to the machine
example, when task difficulty is high, machine performance is likely to suffer in the
absence of slack. The application of slack tailored to a given set of budgeted tasks
(whatever the level of variability) may assist in countering an inability to specify a given
set of input/output relations. In contrast, the degree of variability faced by the machine has
no necessary additional call on resources other than for the budgeted array. Consequently,
task variability is less likely than task difficulty to have performance implications that can
be influenced by the provision of slack. To facilitate the empirical assessment of whether
task variability and slack interact to attect performance, the following null hypothesis is
proposed for testing.
Hg: Budgetary slack and task variability do not interact to affect subunit
performance
Method
‘A random sample of 61 companies was selected from the population of manufacturing
ions listed in the Kompass Australia business directory employing more than 100
people in the Sydney, Australia, metropolitan area. Firms with less than this number of
were not expected to have clearly defined areas of responsibility to which
_ managers would be appointed. A total of 118 managers were contacted by telephone, and
were requested personally to take part in the research, No restriction was placed on the
‘responsibility from which managers could be drawn for sampling purposes, since
no argument for restricting sampling to any one functional area. A questionnaire
ter, together with a reply-paid self-addressed envelope was mailed to each
a higher response rate and to provide evidence that the targeted
ed 10 the questionnaire themselves, a telephone follow-up was made.TABLE | : Descriptive Statistics of the Variables in the Study
Theoretical Range — Actual Range
n Mean StdDevn Min Max — Min Max
udgetary Slack 79° 13.71 4.85 4 28 4 26
Task Difficulty 79 25.73 5.83 7 49 10 48
Task Variability Bei 33.72 6.98 7 49 14 44
Subunit Performance 79 4.20 0.71 1 5 2 2
Responses were received from 79 managers; a response rate of 67 percent. In all, 46
companies were represented in the sample. A total of 25 respondents were drawn from
individual companies, another 17 organizations provided two respondents each, three
respondents came from the one organization, in two instances four responses were received
from the one firm, and in one other case nine respondents came from the one company. The
managers represented marketing, manufacturing, research and development, engineering
and administration units. The mean age of the respondents was 42 years, they had held their
current positions on average for 4.5 years, their mean length of experience in the areas they
managed was 13 years, and the average number of employees in their areas of
responsibility was 102
Variable Measurement
Budgetary Slack
Budgetary slack was measured using the Dunk [1993] four item seven-point Likert scaled
instrument, The results of a factor analysis revealed that the four items in the measure load
single factor with an eigenvalue of 1.3878, explaining 34.7 percent of the variance in
ying variable. The Cronbach alpha (Cronbach, 1951] for the measure was 0.68.
ive statistics for the instrument are contained in table J.
instrument «leveloped by Van de Ven and Delbecq [1974] was used to
based on a seven-point Likert scale. The Cronbach alpha (Cronbach,
was 0.57, similar to that calculated by Withey er al. [1983]. Even
istency of the instrument is not high, it is still within acceptable
Guilfo "1; Nunnally, 1967}. Importantly, Van de Ven and
‘between scale scores and the ratings of
:’s construct validity. Descriptive statistics995 Joint Effects of Budgetary Slack and Task Uncertainty 67
1951] for the scale was 0.76, which suggests that it has considerable internal consistency.
Evidence of the validity o! the measure is based on Withey et al.’s (1983] findings that it
correlated highly with the validated Sims er al. [1976] Job Characteristics Inventory of task
variability together with the substantial agreement between independent external judges’
ratings of work unit task veriability and scale scores. Due to these psychometric properties,
Withey et al. [1983] recommended its use in research. Descriptive statistics for the measure
are reported in table 1.
Subunit Performance
A five-point single item Likert-scaled measure consistent with that used by Merchant
{1981, 1984] and Brownell and Merchant [1990] was used to assess subunit performance.
The instrument requested managers to rate the performance of their areas of responsibility
on a scale anchored by (1) not at all effective and (5) very effective. Table 1 presents
descriptive statistics for the instrument.
No attempt was made to obtain corporate records of performance because it is essentially
impossible to obtain properly matched objective data relating to subunit activities across
multiple organizations [Brownell and Merchant, 1990; Govindarajan, 1984; Merchant,
1981, 1984; Williams ef al., 1990}. Even though internal management accounting reports,
cash flows, operating profit and ROI have been classified as objective data (Govindarajan
and Gupta, 1985; Venkatraman and Ramanujam, 1987], there is no evidence to suggest that
these data are more objective than self-ratings [Brownell and Dunk, 1991]. Moreover,
Venkatraman and Ramanujam [1986] argued that organizations may not make available for
reasons of confidentiality and sensitivity data on financial performance.
tantly, Abernethy and Stoclwinder {1991} found no evidence that subunit managers
Jenient in any consistent fashion in rating the performance of their subunits. From a
ical perspective, rating leniency is a form of measurement error, which Nunnally
gorized as either nonrandom (systematic bias) or random. Nonrandom error
d to the addition (or subtraction) of a constant to the otherwise “true” score
ent variable [Carmines and Zellar, 1982; Tsui and Barry, 1986]. That
er, has no effect on the nature of the relation between independent and
in multiple regression models (Nunnally, 1981]. By comparison, the
n error in the model error term reduces the likelihood of the null
[Neter er al., 1985}.
cin
ion matrix of the variables in the study. The table shows that
bility correlate at 0.25, which is far from supportive of task
onal construct. Whether the significant and negativeAccounting and Finance November
TABLE 2 : Correlation Matrix of the Variables in the Study
. Subunit Performance Budgetary Slack Task Difficulty
Budgetary Slack -0.043(n8) Sully 6 ye
Task Difficulty -0.380(p < 0.001) 0.193 (n.s.)
Task Variability —1).087 (n.s.) 0.312 (p < 0.01) 0.251 (p < 0.05)
association between task difficulty and performance is influenced by the level of slack is
subject to hypothesis test. Given the use of performance self-ratings and the potential for
managers to rate the performance of their departments more favorably when slack is
available, the table provides evidence that there is no zero-order positive relation between
these two variables. Consequently, this finding provides some assurance that managers rate
the performance of their areas of responsibility independent of the level of slack that may
be in their budgets. Although budgetary stack and task variability are correlated (r = -0.312,
p<0.01), multicollinearity is unlikely to influence the significance or sign of the coefficient
ofthe interaction term in multiple regression models [Allison, 1977; Hirst and Lowy, 1990]
Results
The following two-way interaction model was employed to test the hypotheses. Separate
regression analyses for task difficulty and task variability were conducted in turn.
Y= by + BX, + OX, + bX,X, +e a)
where
_ ¥ is subunit performance
is budgetary slack
task difficulty; task variability
is that budgetary slack and task difficulty interact to affect subunit
was tested by examining the significance of b;, the coefficient of the
. The results presented in table 3 suggest that b, is significant at p = 0.025,
rejection of the null hypothesis. The two-way interaction model is also
p= 0.001, explaining 21.3 percent of the variance in performance.° Despite
ificance of b, and b,, Southwood [1978] demonstrated that main effectTABLE 3 : Results of Regressing Subunit Performance
on Budgetary Slack, 1 Interaction
Variable Coefficient Value t Pee
aes eee 73267 10.8317 881 0,001
Budgetary Slack b, 0.1342 0.0534 -2.51 0.014
‘Task Difficulty b, 9.1125 00307 3.670.001
Budgetary Slack x Task
SS 0.0046 0.0020 2.28 0.025
————_ TABLE 4: Results of Regressing Subunit Performance on Task Difficulty
bn Variable Coefficient Valu td Error t P
Panel A: Low Budgetary Slack
Constant bo 6.3619 0.5821 10.93 0.001
Task Difficulty b, 0.0799 0.0206 -3.82_—_—0.001
R? = 0.288 F, 4, = 14.57, p = 0.001
Panel B: High Budgetary Slack,
Constant by 4.7183 0.3878 12.16 0.001
Task Difficulty by 0.0204 0.0155 o13) 0.197
_R? = 0.042 F, 49 = 1.72, p = 0.197
TABLE 5: Results of Regressing Subunit Performance
‘on Budgetary Slack, Task Variability and their Interaction
Coefficient Value Std Error t Pp
by 5.6150, 1.1010 5.10 0.001
b, 0.0733 0.0743, -1.04 0.301
0.0376 0.0312 -1.20 0.232
0.92
was dichotomized at the
d as zcro, and scores above the median
ucted, first, regressing performance on
1 is high. The results of these analyses,
low if difficulty negatively affects70
Accounting and Finance November
fi
serlopsaty (panel A). By contrast, when slack is high, task difficulty does not influence
Be ; ince (panel B). These findings support the proposition that the effect of task
ieully On performance is less pronounced when slack is high than when it is low
The Tesults of testing hypothesis two, however, are different. Table 5 presents the findings
of the regression of performance on budgetary slack, task variability, and their interaction
Because 6; is not significant at an acceptable level of probability, there is no evidence to
suggest that budgetary slack and task variability interact to affect performance.”
Conclusions
The results of the study sugyest that care may need to be taken to consider explicitly both
theoretically and empirically the implications associated with each of the dimensions in
studies in which task uncertainty is implicated. The findings of the study suggest that the
‘association between task difficulty and performance is significantly less negative when
budgetary slack is high, as compared to when it is low. Consequently, the presence of slack
in budgets may be effective in terms of managing the relation between task uncertainty and
subunit performance, at least for the task difficulty dimension. There was no evidence that
slack performed a similar function with regard to the association between task variability
and performance. This finding is consistent with Brownell and Dunk [1991], who found
that task variability did not influence the interaction of budgetary participation and budget
emphasis affecting managerial performance. Taken together, these findings suggest that the
role of task variability may be less critical than that of task difficulty in influencing
performance.
The results of the study also provide support for claims made by both Onsi [1973] and
Galbraith [1973] that there may be circumstances when slack is not necessarily
dysfunctional. The findings of this study suggest that attempts to drive slack out of budgets
when task difficulty is high may detrimentally affect performance. Consequently, attempts
to optimize planned efficiency through the elimination of slack may only be of strategic
advantage in terms of performance enhancement in settings of low task difficulty
Despite careful attention to research design, the results of the study are subject to a number
of potential limitations. First, data were drawn only from companies located in the Sydney,
Australia, metropolitan area As such, the results may be generalizable only to that
population. Following Hofstede {1980}, however, this possible limitation may be unduly
festrictive. Second, no statement of causation can be made since the research is
contemporaneous and cross-sectionally based