Description
A judgment modeling experiment which used bank loan officers as participants was performed in a field
setting to examine the impact of an increase in information level on the judgments of individuals and
groups. The results indicate that, as predicted, groups reached judgments of higher quality than
individuals at both information levels. Also as predicted, groups responded more positively
Pergamon Accountfng, Organizations and Socfety, Vol. 20, No. 7/8, pp. 685-700, 1995
Copyright 0 1995 FJsevier Science Ltd
Printed in Great Britain. Au rights reserved.
0361-3682/95 $9.50+0.00
THE IMPACT OF AN INCREASE IN ACCOUNTING INFORMATION LEVEL ON
THE JUDGMENT QUALITY OF INDMDUALS AND GROUPS
MORRIS H. STOCKS
University of Mississippi
and
ADRIAN HARRELL
University of South Carolina
Abstract
A judgment modeling experiment which used bank loan officers as participants was performed in a field
setting to examine the impact of an increase in information level on the judgments of individuals and
groups. The results indicate that, as predicted, groups reached judgments of higher quality than
individuals at both information levels. Also as predicted, groups responded more positively (or less
negatively) than individuals to increases in information level, resulting in an increase in the judgment
quality gap between groups and individuals. lndividuals experienced greater information processing
difficulties than groups as information level increased.
Individuals and groups often must integrate
many distinct pieces of information into their
judgments. Creditors, for example, typically
must incorporate numerous financial ratios,
industry standards and other financial informa-
tion into their lending decisions. Based on
Schroder et al., (1967), it has been proposed
that, as the number of relevant pieces of avail-
able information increases, the quality of a per-
son’s judgments will initially increase. At some
point, however, further increases in the num-
ber of relevant pieces of information will result
in a decline in the quality of these judgments.
Previous accounting research has provided
empirical support for the information utiliza-
tion pattern predicted by the Schroder et al.
(1967) model (e.g. Casey, 1980; Iselin, 1988;
Chewning & Harrell, 1990).
Prior accounting information level studies
have, to date, considered only individual deci-
sion makers. Despite this emphasis upon indi-
viduals, many important judgments using
accounting information are frequently reached
in a group setting. For example, a group pro-
cess is often involved in bankruptcy, credit,
and loan default judgments (Haberlin, 1976;
McNeil, 1986). Although no prior research
has examined the impact of increases in infor-
mation level on the quality of group judgments,
accounting researchers have recognized the
need to study both individual and group judg-
ments. In general, prior research indicates the
judgments reached by groups appear to be of
higher quality than the judgments reached by
individuals (Schultz & Reekers, 1981; Solomon,
1982; Trotman et al., 1983; Chalos, 1985).
Noting that prior studies have focused on
individuals and that many important judg-
ments are reached by groups. Snowball
(1979) called for accounting information level
research which compares individuals and
groups. The purpose of this study is, there-
685
686 M. H. STOCKS and A. HARRELL
fore, to compare the effect of two different
levels of accounting information, six distinct
information cues and nine distinct information
cues, on the quality of the judgments reached
by individuals and groups. It is expected that
the judgment quality of groups will be higher
than that of individuals at both information
levels. In addition, groups are expected to
respond more positively than individuals to
an increase in accounting information level.
Accordingly, it is proposed that, as the informa-
tion level increases, the expected gap between
the judgment quality of groups and individuals
will increase, so that the quality of group judg-
ments will diverge from that of individuals.
RELEVANT LITERATURE AND HYPOTHESIS
DEVELOPMENT
The remainder of this paper is organized into
four sections. First, a brief review of the rele-
vant prior research is provided, along with the
hypothesis which guided this study. Second,
the research methodology is presented. Third,
the data analysis procedures are described and
the results are shown. Finally, the last section
presents a discussion of the implications of the
findings.
Schroder et al . (1967) developed a concep-
tual model of human information processing.
The Schroder, Driver and Streufert (SDS)
model postulates an inverted-U curve when
the complexity of the information environ-
ment is plotted against the level of information
processing exhibited by an individual or group
(Fig. 1). This inverted-U curve suggests that, as
the information environment becomes more
complex, the level of information processing
by the processing unit increases up to some
maximum processing level. After this point,
additional environmental complexity results in
a lower level of information processing.
The SDS model has served as the conceptual
basis for most of the accounting research
which examines the effect of environmental
complexity on judgment quality. Accounting
researchers have generally operationalized
environmental complexity in terms of task
complexity. The widely recognized concep-
tual model of task complexity originated by
F
0
E
_
TASK COMPLEXITY
Fig. 1. Inverted-U curve.
ACCOUNTING INFORMATION LEVEL 687
Wood (1986) can be used to describe the com-
plexity of various tasks. Wood discusses three
aspects of task complexity: component com-
plexity, coordinative complexity, and dynamic
complexity. The component complexity of a
task is determined by the number of distinct
information cues that must be processed and
the number of distinct acts that must be exe-
cuted in performing the task. Coordinative
complexity is determined by the form and
strength of the relationships between informa-
tion cues and acts, including the timing, fre-
quency, intensity, and location requirements
for performances of required acts. Dynamic
complexity refers to the need to adapt to
changes which occur in the cause-effect chain
or means-ends task relationships during the
performance of the task. The experimental
tasks employed in the accounting information
level studies appear to be adequately described
by Wood’s component complexity construct.
Two distinct streams of accounting compo-
nent complexity research can be identified.
The first research stream is relatively well
established. It consists of those studies with
experimental tasks which require their partici-
pants to incorporate a number of distinct
(independent) information cues into a judg-
ment or decision (e.g. Trotman et al ., 1983;
Chewning & Harrell, 1990). In these studies,
the component complexity of the experimen-
tal task may be determined by counting the
number of distinct (independent) cues
involved in the task. Component complexity
may be manipulated directly by increasing or
decreasing the number of distinct information
cues provided to the participant. A second
research stream consists of studies with experi-
mental tasks which involve both elements of
component complexity. In these studies, com-
ponent complexity may be manipulated by
increasing or decreasing the number of dis-
tinct information cues, the number of distinct
acts that must be executed to perform the task,
or both. Only a few studies have been per-
formed which use tasks that appear to involve
both elements of component complexity (Ise-
lin, 1988; Simnett, 1993; Stocks et al ., 1995).
Soon after the introduction of the SDS
model, a series of conceptual papers appeared
in the accounting literature which suggested
that the availability of large volumes of account-
ing information might lead to sub-optimal finan-
cial judgments (Fertakis, 1969; Revsine, 1970;
Miller, 1972; Ashton, 1974; Driver & Mock,
1975; Miller & Gordon, 1975). The first related
empirical study to appear in the accounting
literature was conducted by Casey (1980). In
his study, bank loan officers were assigned to
one of three information levels and were asked
to predict bankruptcy for ten actual firms. The
accuracy of these predictions was used to indi-
cate the effect of information level on judg-
ment quality. Although the participants
perceived an increase in the information
level, the information level manipulation did
not affect the quality of the participants’ judg-
ments. Since the experimental task was not
described in terms of a conceptual model of
task complexity, it’s not clear whether the
manipulation actually increased task complex-
ity or whether task complexity was simply per-
ceived to have increased. Libby and Lewis
(1982, p. 271) suggest Casey’s (1980) manipu-
lation may have been ineffective because the
“high” information level may have lacked addi-
tional information content.
Shields (1983) also used judgment accuracy
as a measure of the impact of information level
on individuals’ information processing. Unlike
the between-persons design used by Casey
(1980) Shields used a within-persons manipu-
lation of information level. The results of this
study provide support for the SDS model. More
recently, Schick et al . (1990) proposed that
information load level (task complexity) can
be considered a function of time available,
rather than a function of cognitive processing
ability.
A study by Chewning and Harrell(l990) may
represent the first effort to relate the number of
distinct information cues used to the number
available, thus determining the effect of infor-
mation levels on information utilization. Their
subjects were asked to make repeated financial
distress predictions at three information levels
688 M. H. STOCKS and A. HARRELL
(four, six and eight distinct information cues).
Accordingly, the actual information level was
manipulated by increasing the number of dis-
tinct information cues to increase complexity.
By estimating a regression model for each sub-
ject at each cue level, it was possible to mea-
sure cue usage at each information level. The
results showed that when cue usage was
plotted against cue availability, approximately
one-third of the subjects exhibited the informa-
tion usage implied by the SDS model. Because
the cases employed in the study represented
hypothetical firms, it was impossible to mea-
sure judgment accuracy. Therefore, cue
usage, consistency and consensus served as
indicants of judgment quality. The results indi-
cate that the subjects who exhibited the
inverted-U pattern scored significantly lower
on these indicants of judgment quality at the
eight-cue level.
Iselin (1988) may have been the first to
employ an experimental task which involves
both facets of component complexity: distinct
information cues and distinct acts to be exe-
cuted in the performance of the task. His parti-
cipants completed an experimental task which
required them to perform relatively routine
computations to arrive at a mathematically
determined response. Thus Iselin’s study
might be described as an examination of the
impact of variations in the two elements of
component complexity on individuals’ perfor-
mance of a relatively routine, highly structured
task.
Two recent studies have attempted to
extend Iselin’s research to contexts which
involve professional judgment. Simnett (1993)
asked auditors to make going concern predic-
tions for a series of firms using either a low or a
high information level. Some subjects in the
high information level received additional
cues which were considered to be distinct
from those used in the low information level,
while others received additional cues which
were considered to be redundant. This manip-
ulation was designed to “. . . separately iden-
tify the effect of both information load and
additional information” (Simnett, 1993, p. 7).
Stocks et al . (1995) manipulated the number
of distinct financial ratios presented to deci-
sion makers as well as the number of steps
required to classify the ratios. Their results
demonstrate that information level is only one
aspect of task complexity.
Apparently, no prior research has systemati-
cally compared the quality of individual and
group judgments across information levels. Sev-
eral studies have, however, compared the qual-
ity of the judgments reached by individuals and
groups at a single information level. Although
differences exist between these studies as to
the experimental tasks used, the decision con-
texts involved, and the type of individuals who
participated, their findings taken together sug-
gest that group judgments may be superior to
individual judgments in terms of accuracy
(Uecker, 1982; Chalos & Pickard, 1985), confi-
dence (Schultz & Reekers, 1981) and consen-
sus (Solomon, 1982).
Driver and Mock (1975) state, “Individuals
and groups use more information in decisions
with increasing information levels, up to a
point; then they begin to process less. . . .”
(p. 496). Although this suggests the informa-
tion processing of individuals and groups may
not differ, the findings of empirical studies sug-
gest that groups may encounter fewer informa-
tion processing difficulties than individuals at
high information levels. Chalos and Pickard
(1985) showed that when groups and indivi-
duals were supplied with a set of financial state-
ments to supplement three financial ratios used
in a financial distress task, the predictive accu-
racy of the groups increased while predictive
accuracy decreased for individuals. They sug-
gest that the groups may have benefited from
the added information, while the individuals
may have been unable to incorporate the addi-
tional information into their judgments. It is,
therefore, possible that individual and group
differences in judgment quality may be due in
part to a difference in the ability to process
large amounts of information.
Trotman et al . (1983) compared groups and
individuals at a single information level and
concluded that the groups outperformed indi-
ACCOUNTING INFORMATION LEVEL 689
viduals on all three measures (consensus, con-
sistency and cue usage). In discussing their
results, they suggest that, at higher information
levels, groups may be able to incorporate more
information into their decision processes than
individuals. While the present study examines
some of the propositions suggested by Trotman
et al. (1983), it differs from that study in several
important ways. First, this study investigates
judgment accuracy, an issue not considered
by Trotman et al. (1983). Second, this study
uses an experimental setting to examine their
suggestion that an increase in information level
should be expected to have a less adverse effect
on groups than on individuals. Finally, this
study employs a financial distress prediction
task while the Trotman et al. study examined
an internal control assessment.
In summary, the SDS theoretical model
implies that both individuals and groups will
encounter information processing difficulties
at higher information levels (Schroder et al.,
1967; Driver & Mock, 1975). Empirical stu-
dies, however, suggest that groups possess
greater information processing capabilities
than individuals and reach judgments of higher
quality than individuals. Accordingly, a gap in
judgment quality is predicted to exist between
groups and individuals. As suggested by Chalos
and Pickard (1985) and Trotman et al. (1983),
this judgment quality gap between groups and
individuals is predicted to become greater
with increases in information level. These pre-
dictions are summarized in the following
hypothesis:
The judgment quality of groups will be higher than the
judgment quality of individuals and will diverge posi-
tively from that of individuals as information level
increases.
The above hypothesis predicts the existence
of a positive ordinal interaction between the
type of participant (individual or group) and
the information level. The interaction is pre-
dicted to be ordinal in the sense that groups
are expected to outperform individuals at
both information levels. The interaction is pre-
dicted to be positive in that the difference in
judgment quality is expected to increase as
information level increases (Glass & Stanley,
1970). The following section describes the
methodology used to examine the above
hypothesis.
METHODOLOGY
An experiment was performed in a field set-
ting to examine the research hypothesis. The
participants, experimental design, data collec-
tion procedures, experimental task, and
research instrument are described below,
along with the judgment quality attributes
used to examine the hypothesis.
Participants
Experienced bank loan officers served as par-
ticipants in the experiment. Prior research has
suggested that bank loan officers are experi-
enced and appropriate subjects for studies
which require the prediction of financial dis-
tress (Casey, 1980; Libby, 1975a; Chalos,
1985). The participants were all currently
employed by the same large national bank.
The senior official who sponsored this study
stated that the bank used standard procedures
throughout its various regions, so all of the
participants were subject to the same person-
nel selection, training, and operating proce-
dures. Table 1 provides a summary of the
participants’ demographic information.
Design
As shown in Fig. 2, a 2 X 2 between-subjects
factorial design was used to compare the effect
of six-cue and nine-cue levels of information
(the LVL variable) on the quality of the judg-
ments reached by groups and individuals (the
TYPE variable). Accordingly, component com-
plexity was manipulated in this study by pro-
viding the participants with one of two
possible information levels, six distinct infor-
mation cues or nine distinct information cues.
As described below, the procedures used to
690 M. H. STOCKS and A. HARRELL
T
Y
P
E
Table 1
Summary of Participant’s Demographics Data
Attribute nean Between-Cell Predictor of Judgment
(Standard Deviation) Comparisons' Quality Indicants
Age
33 years Prob > F = .763 PI-& > F = .918
(8.7)
Lending 8.4 years Prob > F = .I95 Prob > F = -796
Experience
(7.5)
Annual $41,878 Prmb > F = .535 Prob > F = .634
Salary (15.1)
Education 91% Bachelor's Degree Prob > x2 = .153 Prob > F = .823
21% Master's Degree
Gender 79% Male
21% Female
Prob > 2' = .238 Prob > F = .456
'Two of the demographic measures, Education and Gender, were examined using
a x2 statistic due to the categorical nature of the data.
LVL
6 Cues 9 Cues
Individual
(16) (19)
n = 27 n = 28
Groups
Fig. 2 Experimental design.
identify the cues assured that each cue pos-
sessed unique information content. An
increase from six to nine information cues
therefore resulted in an actual increase in infor-
mation level. The participants were drawn
from four regions of a large national bank.
Each of the four regions was randomly
assigned to one of the four treatment cells
shown in Fig. 2, individual subjects receiving
six cues (16), individual subjects receiving nine
cues (I9), group subjects receiving six cues
(Gb), and group subjects receiving nine cues
(G9). According to the lending procedures of
the participating bank, these subjects were
often required to make financial judgments as
individuals and as members of small three- or
four-person groups. Accordingly, a group size
of three was selected, a suggested by Trotman
et al . (1983), Libby & Blashfield (1978), Ashton
(1986), and Einhorn et al . (1977).
Between-cell comparisons were made for
age, experience, salary, education and gender;
no significant differences were found. In addi-
tion, a series of MANOVA procedures demon-
strated that the demographic characteristics
had no impact on the judgment quality mea-
sures. Details of these results are provided in
Table 1.
Data col l ecti on procedures
The data were collected through the partici-
pating bank’s inter-office mail system. The
Director of Commercial Lending informed the
loan officers that the bank had agreed to parti-
ACCOUNTING INFORMATION LEVEL 691
cipate in the research project and urged each
participant to respond. Both groups and indivi-
duals received detailed instructions for the
judgment task. All participants were informed
their specific responses would remain anon-
ymous. The office of the Director of Commer-
cial Lending served as the collection point for
the research instruments. Of the 190 research
instruments distributed, 112 (58.9%) usable
responses were received. This included 55 indi-
vidual responses and 57 three-person group
responses (see Fig. 2). Accordingly, a total of
226 bank loan officers participated in the study.
In an effort to determine whether a non-
response bias existed, MANOVA was used to
compare the first 25 individual and first 25
group responses received with the last 25 of
each category received. No significant differ-
ences were found in either instance. These
findings, in conjunction with the relatively
high response rate, suggest that the sample
was not biased by non-responses.
Care was taken to assure that a single indivi-
dual didn’t complete the group instrument.
First, a check question was inchtded in the
research instrument which asked for the num-
ber of individuals that actively participated in
the completion of the instrument. In addition,
each group member was asked to complete a
separate individual information sheet. This
information sheet included a question which
asked whether he or she actively participated
in the group judgment process. These sheets
were then sealed by the individual and
included in the packet to be returned.
Further, the instructions emphasized the
importance of face-to-face group interaction
in completing the judgment task.
Previous research (Miner, 1984) has demon-
strated that the group process is most effective
when group members make individual judg-
ments prior to reaching a consensus judg-
ment. Chalos (1985) found that interacting,
face-to-face committees outperformed mec-
hanically aggregated committees. Accordingly,
group members were instructed to form indi-
vidual judgments for each case and then, face-
to-face, arrive at a group consensus. However,
a potential limitation of the present study is
that, although the manipulation questions
completed by the participants reported they
followed these procedures, the field setting
of the study meant the interactive group pro-
cess was not actually observed by the
researchers.’
J udgment task
The participants completed an experimental
task which involved using either six or nine
distinct information cues to arrive at profes-
sional judgments. Groups and individuals
were required to make a series of financial dis-
tress predictions (Chalos, 1985; Chewning &
Harrell, 1990). Financial distress was defined
as either bankruptcy, loan default, failure to
meet a preferred stock dividend payment or
asset liquidation.
In order to assure that each cue possessed
unique information content, the accounting
ratios the participants used as judgment cues
were derived through a factor analysis process
similar to that employed by Libby (1975b) and
Chewning & Harrell(l990). A set of forty finan-
cial ratios which have been shown to have
predictive ability in a bankruptcy prediction
was identified from the recent accounting lit-
erature (Zavgren, 1985; Hopwood et al., 1990;
Chewning & Harrell, 1990). The forty ratios
were then computed for each of the 1450
manufacturing firms listed on the Compustat
tapes for fiscal year 1986. These ratios were
factor analyzed, using a varimax rotation, to
reduce the redundancy within the ratios and
to identify common constructs. The nine fac-
tors accounting for the largest portion of the
variance within the data were selected. Each
’ However, since the group vs individual comparisons made in this and previous studies report the superiority of inter-
active group decisions, we believe the inclusion of a group response which was not the product of an interactive group
would bias the results of our study conservatively.
692 M. H. STOCKS and A. HARRELL
Firm No. 1
Given the following information:
1.
2.
3.
4.
5.
6.
7.
8.
9.
In
Firm Size. . . . . . . . . . . . .
Cash / Total Assets. . . . . . . .
Total Equity / Total Liabilities .
Earnings Before Interest and Taxes
Total Assets / Long Term.Debt. . .
Net Income / Net Worth . . . . . .
Sales / Fixed Assets . . . . . . .
Earnings Before Interest and Taxes
Working Capital / Sales. . . . . .
. . . . . . BOTTOM 113 l
. . . . . . TOP 113 *
. . . . . . BOTTOM 113
/ Sales . . TOP 113 l
. . . . . . BOTTOM 113 *
. . . . . . TOP 113
. . . . . . TOP 113
/ Assets. . BOTTOM 113 *
. . . . . . BOTTOM 113 *
your view, indicate below the likelihood this firm will
experience financial distress during the next three years.
Probably Will Not Experience Probably Will Experience
Financial Distress Financial Distress
VERY VERY
LOW 0 1 2 3 4 5 6 7 8 9 HIGH
CHANCE CHANCE
* - Judgment Cues provided in six-cue cases.
Fig. 3. Example of nine-cue case.
Position
Relative to
Industry
factor contained an eigenvalue greater than 1.4
and collectively they accounted for 80% of the
total variance. The ratio with the highest load-
ing on each factor was chosen as the cue to be
used in the task, as shown in Fig. 3.
Correhtion analysis indicates nine of the 36
pairs were significantly correlated. However,
most of these correlations were small. Only
two of these correlations were greater than
0.20 and the median correlation was 0.015,2
demonstrating that each of the financial ratios
provide the unique information content appro-
priate for distinct information cues. These cues
represent measures of profitability, liquidity
and capital structure and are similar to those
derived by Zavgren (1985), Libby (1975b) and
Chewning & Harrell ( 1990) . In addition, they
were reviewed for reasonableness by the Direc-
tor of Commercial Lending for the participating
institution. In summary, this process identified
relatively uncorrelated ratios which were use-
ful in predicting financial distress.
’ The highest correlation between any two cues, EBIT/Sales and EBIT/Assets, was 0.45, indicating the two variables had
about 20% common variance and about 80% independent variance. The only other correlation greater than 0.20 was
between Total Equity/Total Liabilities and CasWotaI Assets (r = 0.28). This supports the view that each cue provided
unique information content.
ACCOUNTING INFORMATION LEVEL 693
The research instrument
Each individual and group reached financial
distress predictions for 42 cases, with 32 cases
representing hypothetical firms and 10 cases
representing actual firms. A l/ 2 replicate of a
26 factorial design was incorporated into the 32
hypothetical cases completed by the partici-
pants who received six cues. For those who
received nine cues, a I/ I6 replicate of a 29
factorial design was incorporated into the 32
hypothetical cases. Both the six-cue and nine-
cue partial replicates conformed to an orthogo-
nal design, which eliminated any correlation
among the independent variables. In both
instances, the cases were presented in random
order, with the 10 cases representing actual
firms being randomly interspersed to make
them indistinguishable from the hypothetical
cases. Following Chewning & Harrell (1990)
the cues in the cases representing hypothetical
firms were manipulated at two levels (the top
l/ 3 and bottom l/ 3 of the firm’s industry). The
cues for the 10 cases representing actual firms
were presented at three levels (top l/ 3, middle
l/ 3, and bottom l/ 3), as some ratios appeared
in the middle of the industry as well as the top
and bottom. The six cues used in 16 and G6
were randomly selected from the nine cues
used in 19 and G9. The instructions informed
participants that each cue had been shown to
be useful in predicting financial distress and
emphasized that all of the cues should be
used in arriving at their judgments. An exam-
ple of the nine cue cases is shown in Fig. 3; the
asterisked judgment cues in Fig. 3 indicate
those used in the six-cue cases3
In order to provide a measure of accuracy,
five failed and five non-failed actual firms were
selected from the list provided by Zavgren
(1985). Obvious cases were avoided. For exam-
ple, if nearly all of the ratios of a firm fell within
either the “Bottom l/ 3” or within the “Top
l/ 3” of it’s industry, a financial distress predic-
tion for that firm is obvious, so such firms wer-
en’t used. The Wall Street Journal Index was
reviewed for the five non-failure firms for the
appropriate three-year period to assure they
did not experience any non-bankruptcy form
of financial distress. This was necessary
because financial distress was broadly defined
in the instructions to participants. The ratios
for the failed firms were computed for the
third year prior to bankruptcy. For non-failed
firms, the ratios were computed for the third
year prior to the bankruptcy of the correspond-
ing failed firm.
Previous research (Casey & Selling, 1986;
Houghton, 1984) has suggested that financial
distress judgments may be sensitive to the dis-
closure of the failure rate in the sample. The
50% failure rate (five failed, five non-failed) of
the ten actual firms used in the study exceeds
the failure rate of the overall population of
such firms. Therefore, subjects were informed
that the sample of firms used in the study was
not randomly selected and that the proportion
of failed firms was higher than for a random
sample of firms. The actual failure rate was
not disclosed in order to avoid judgment biases
which might result from “gaming” by the
participants.
Dependent variables
In this study, judgment quality refers to the
degree of judgment excellence, as measured by
four attributes; cues used, accuracy, consis-
tency, and consensus. All four of these attri-
butes have been used as indicants of
judgment quality in prior studies (e.g. Casey,
1980; Chewning & Harrell, 1990). Cues used
indicates the number of cues actually inte-
grated into the judgment process. This indi-
cant was computed by estimating a regression
model for each individual and each group,
’ In an effort to check the effectiveness of the information level manipulation, all participants were asked to indicate
which of the cues were useful in the prediction task. A comparison was made of the number of cues that were subjectively
rated as useful between participants in the two information level manipulations. The mean numbers of cues indicated
useful was 8.3 for nlnecue subjects and 5.6 for six-cue subjects. A two-sample C-test indicates that subjects in the nine-cue
manipulation perceived the availability of significantly more useful information than subjects who received six-cues (T =
13.49; Prob > T = 0.0001).
694
M. H. STOCKS and A. HARRELL
using the six or nine cues as independent vari-
ables and the financial distress predictions as
the dependent variable. Similar to Trotman et
al . (1983) and Chewning & Harrell (1990) the
number of cues that were integrated into the
judgment process was determined by counting
the number of significant main effects (at p c
0.05) in each of the regression models. Accu-
racy relates to how accurately the individual or
group was able to predict the future status of
the 10 actual firms. Following Chewning &
Harrell (1990) and Trotman et al . (1983), the
adjusted R2 value obtained from each indivi-
dual’s or group’s regression model was used
to indicate consistency. Consensus was
defined as the average paired correlation
between a participant’s judgments and the
judgments of all other participants from the
same experimental treatment. For example,
the judgments of an individual who received
Table 2
Summary of Univariate Analysis of Variance Results
JUDGMENT QUALITY F-VALUE PROBABILITY
INDICANTS
CUE8 USED
TYPE 31.44 .OOOl
LVL 3.31 .0715
TYPE x LVL 7.73 .0064
Contrasts
16 vs 19 .45 .5024
66 vs G9 10.78 .0014
16 vs 66 3.93 .0501
ACCURACY
TYPE
LVL
TYPE x LVL
Contrasts
16 vs 19
GC vs G9
16 vs 66
5.01 .0272
15.96 .OOOl
3.00 .0861
2.52 .1157
16.70 .OOOl
.13 .7229
CONSISTENCY
TYPE 35.18 .OOOl
LVL 11.79 .OOOl
TYPE X LVL 5.59 .0199
Contrasts
16 vs 19 16.51 -0001
66 vs G9 .58 .4467
16 vs 66 6.25 .0139
ACCOUNTING INFORMATION LEVEL 695
‘cue Usage: more cues were used in the judgments made by groups than by individuals at
both the six-cue @a = 5.3 vs 3~~s = 4.6; F = 3.93,~ = 0.0501) and nine-cue (Zos = 6.4 vs 4s
= 4.4; F = 33.82, p = 0.0001) levels.
bAcc~racy: the accuracy of the judgments reached by groups and individuals did not differ
at the sixcue level (2~ = 5.2 vs f IG = 5.1; F = 0.13, p =0.7229). Groups did, however,
reach more accurate judgments than individuals at the nine-cue level (32os = 6.4 vs 2,~ =5.6
F = 8.03, p = 0.0055).
‘Consistency: the judgment consistency of the groups was greater at both the six-cue (*a
= 0.83 vs _z’r = 0.76; F = 6.25, p =0.0139) and nine-cue levels (*os = 0.81 vs 4s = 0.64; F =
35.03, p =0.0001).
dConsensus: consensus between the judgments of groups was greater at both the sixcue
(& = 0.72 vs fib = 0.63; F = 53.39, p =0.0001) and nine-cue levels (& = 0.67 vs 4s =
0.52; F = 87.42, p = 0.0001)
six cues were correlated with the judgments of
all other individuals who received six cues. The
mean of these paired correlations served as a
measure of consensus (Trotman et al., 1983;
Chewning & Harrell, 1990).
ANALYSIS AND RESULTS
Several analysis steps. were required to exam-
ine the data provided by the experiment. Initi-
ally, MANOVA was used to examine the
relationship between the experimental manip-
ulations and the four indicants of judgment
quality. The MANOVA results indicate signifi-
cant main effects for both the TYPE (F-test, p
= 0.0001) and LVL (F-test,p =0.0001) variables
and a significant TYPE X LVL interaction (F-
test, p = 0.0001). While these results provide
preliminary support, they must be considered
carefully. First, the existence of the significant
two-way interaction makes interpretation of
the TYPE and LVL main effects difficult.* Sec-
ond, these results, alone, do not provide suffi-
cient information to examine the directional
predictions of the research hypothesis. Accord-
ingly, four univariate analyses of variance
(ANOVA), along with the appropriate con-
trasts of cell means, were performed to sepa-
rately examine the relationship between the
independent variables and each of the indi-
cants of judgment quality. The results of these
analyses are summarized in Tables 2 and 3 and
in Fig. 4.
Examination of hypothesis
In the context of the experiment, the
research hypothesis predicts the judgment
4 However, Kerlinger (1986, p.242) suggests that it is possible to interpret signiticant main effects when the interaction is
ordinal in nature. An examination of Fig. 4 indicates that such is the case in the present study.
696 M. H. STOCKS and A. HARRELL
Panel A
Cues Used
6 cu.* 9 c ues
Cues Available
Panel C
Consistency
CO~S‘~t.3~~~
I/
Cues Available
- Groups + Indlvlduolr
Fig.
quality attributes will be higher for groups than level (2~6 = 5.2 vs 216 = 5.1; F = 0.13, p =
for individuals at both the six-cue and nine-cue 0.7229). Groups did, however, reach more
levels. As predicted, more cues were used in accurate judgments than individuals at the
the judgments made by groups than by indivi- nine-cue level (2~9 = 6.4 vs x19 = 5.6; F =
duals at both the six-cue (3~6 =5.3 vs *I~=4.6; 8.03, p = 0.0055). As predicted, the judgment
F= 3.93,~ =0.0501) and nine-cue (%c9 =6.4 vs consistency of the groups was greater at both
S?t9 = 4.4; F = 33.82, p = 0.0001) levels. The the six-cue (2~6 =0.84 vs 216 =0.76; F = 6.25,~
accuracy of the judgments reached by groups =0.0139) and nine-cue levels (Zo9 =0.81 vs 219
and individuals did not differ at the six-cue = 0.64; F = 35.03, p = 0.0001). Consensus
Panel B
Accuracy
7
*ccurmy
6.4
by “....:...:....
56
5
6 cuss 9 c ues
Cues Available
Panel D
Consensus
consensus
0.8
0.5
t
_.-
6 C”es 9 c um
Cues Avallable
- Groupr + I ndi vi dual s
4.
ACCOUNTING INFORMATION LEVEL 697
between the judgments of groups was greater
at both the six-cue (266 =0.72 vs &6 =0.63; F =
27.24, p = 0.0001) and nine-cue levels (%09 =
0.67 vs %tIs = 0.52; F = 87.42, p = 0.0001).
Accordingly, although the accuracy of indivi-
duals and groups did not differ at the six-cue
level, the preponderance of the results support
the prediction that the judgment quality of
groups will be higher than that of individuals
at both information levels.
The research hypothesis also predicts the
existence of a positive ordinal interaction
between the type of judge (‘IYPE) and the
information level (LVL.) variables for each of
the judgment quality attributes. The mean
values of the four judgment quality indicants
for both groups and individuals are plotted at
the six-cue and nine-cue levels in Fig. 4. As
shown in Table 2, the results of the ANOVA
for cues used indicates the existence of a sig-
nificant TYPE X LVL interaction (F = 7.73, p =
0.0064) indicating that the cue usage pattern
of groups and individuals differed as informa-
tion level increased from the six-cue to nine-
cue level. In panel A of Fig. 4, the plotted
mean values for cues used indicate the pre-
dicted positive ordinal interaction. The second
ANOVA, using judgment accuracy as the
dependent variable, indicates the existence of
a weak TYPE X LVL interaction (F = 3.00, p =
0.0861>, providing limited support for the pre-
dicted relationships. The plotted mean values
for judgment accuracy indicate the increasing
gap in the accuracy of judgments implied by
the predicted positive ordinal interaction
(panel B of Fig. 4).
The ANOVA for judgment consistency indi-
cates the existence of a significant TYPE X LVL
interaction (F = 5.59, p = 0.0199). Panel C of
Fig. 4 again demonstrates the predicted posi-
tive ordinal interaction. Finally, the ANOVA
using consensus as the dependent variable
also indicates the existence of a significant
TYPE X LVL interaction (F = 8.00, p =
0.0056). When the cell means are plotted, it
is clear that the gap in judgment consensus
between groups and individuals increases as
the information level increases. The evidence
provided by the ANOVAs and the plotted mean
values of the indicants of judgment quality
appear to support the prediction that the judg-
ment quality of groups diverges positively from
the judgment quality of individuals as the infor-
mation level increases.
Additional @dings
The SDS model predicts that both individuals
and groups will experience information proces-
sing difficulties as the information level
increases. For individual participants in this
study, as shown in Tables 1 and 2, two indi-
cants of judgments quality (cues used and accu-
racy) did not change, while two indicants
(consistency and consensus) declined as the
information level increased. The individual par-
ticipants appear to have experienced moderate
information processing difficulties with
increased information level, as the quality of
their judgments does not appear to have bene-
fited from the availability of additional informa-
tion. For the group participants, as shown in
Tables 2 and 3, two judgment quality indicants
(cues used and accuracy) increased as informa-
tion level increased. One indicant (consis-
tency) did not change, and only one indicant
(consensus) declined. These results suggest
that groups experienced few information pro-
cessing difficulties with increased information
level since the quality of their judgments appar-
ently benefitted from the availability of addi-
tional information.
DISCUSSION
This study compares the effect of increasing
levels of information on the quality of the judg-
ments reached by individuals and groups. Sev-
eral findings can be reported. First, the
judgment quality of groups was higher than
that of individuals at both the six-cue and
nine-cue information levels. The only excep-
tion to this was the judgment accuracy mea-
sure at the six-cue level. ‘Second, as the
information level increased, the individual par-
ticipants appear to have experienced moderate
698
M. H. STOCKS and A. HARRELL
information processing difficulties, while the
group participants appear to have experienced
few information processing difficulties. As a
result, the gap in judgment quality which
existed between individuals and groups at the
six-cue information level became larger at the
ninecue information level, causing the judg-
ment quality of groups and individuals to
diverge.
Some limitations of this study should be
noted. As mentioned earlier, the empirical
data were gathered across several southeastern
states. The research instrument was forwarded
to the subjects through the participating bank’s
interoffice mail system and experienced loan
officers completed the experimental task in
their own work environment. Although this
approach allowed the data to be gathered in a
naturalistic field setting, it also means we were
unable to be present and personally observe
the group decision-making process. It is, there-
fore, difficult to speculate about particular
aspects of the group decision processes asso-
ciated with the superior group performance.
The results do, however, have potential impli-
cations for accountants, both as providers of
financial information and in their advisory role
to management. The findings imply that when
feasible, important business judgments should
be reached by groups, for groups can be
expected to consistently reach judgments of
higher quality that those reached by indivi-
duals. This is especially true for judgments
which are complex in terms of the number of
information inputs which should be considered.
In some business settings, the additional cost
of a group decision process may outweigh the
benefit derived from the process. In other situa-
tions, logistical considerations may suggest that
group decisions are impractical. When the
group judgment process isn’t feasible, an
attempt should be made to structure the judg-
ment process so as to limit to a relatively small
set the number of relevant information inputs
to be considered. Perhaps individuals should
be provided with only those information items
which possess the greatest predictive ability for
a particular judgment.
The evident importance of judgment quality
to the success of business organizations sug-
gests the need for further research of this
topic. A straightforward extension of this study
would be to examine group judgments at an
information level greater than the nine cues
considered in this study. It is conceivable
that, given the availability of more than nine
inputs, groups may be able to successfully inte-
grate more information into their judgments. A
second extension would be to examine the
impact on judgment quality of providing both
groups and individuals with a judgment aid in
the form of an appropriate judgment model. A
third extension of this study is to investigate
the effects of variations in information level
and presentation mode on group and indivi-
dual decision makers. Several studies have
examined the impact of presentation mode in
a multiple cue judgment task on individuals
(e.g. Amer, 1991). Since groups often make
these type of judgments, it appears appropri-
ate to consider the impact of these variables
on groups.
The results of this study suggest that, at
higher information levels, groups are able to
generate a judgment model that incorporates
more informational cues into a judgment and
to apply that model more consistently than
individuals. In our study, this led to more accu-
rate predictions which were in greater agree-
ment with other similar decision makers. A
final suggestion for extending this research is
to investigate why the judgment quality gap
between groups and individuals increases as
information level increases. The research
design and measurement approach employed
in this study does not appear to be appropriate
for such a study, which probably would
involve an examination of the reasoning pro-
cesses and procedures employed by groups
and individuals. We suggest that cognitive
science methodologies (e.g. Dillard, 1984) are
likely to be more appropriate for such an inves-
tigation. It is hoped that the present study will
stimulate others to a further examination of
these issues.
ACCOUNTING INFORMATION LEVEL
BIBLIOGRAPHY
699
Amer, T. S., An Experimental Investigation of Multi-cue Financial Information Display and Decision
Making, Journal oflnformation Systems (Fall 1991) pp. 18-33.
Ashton, R., Behavioural Implications of Information Overload in Managerial Accounting Reports, Cost
and Management (July/ August 1974) pp. 37-40.
Ashton, R., Combining the Judgments of Experts: How Many and Which Ones, Organi rati onaf Behavior
and Human Deci si on Processes (1986) pp. 405-414.
Casey, C. J., Jr., Variation in Accounting Information Load: The effect on Loan Officers’ Prediction of
Bankruptcy, The Accounti ng Revi ew (January 1980) pp. 36-49.
Casey, C. J., Jr. & Selling, T. I., The Effect of Task Predictability and Prior Probability Disclosure on
Judgment Quality and Confidence, The Accounti ng Revi ew (April 1986) pp. 302-317.
Chalos, P., Financial Distress: a Comparative Study of Individual, Model, and Committee Assessments,
J ournal of Accounti ng Research (Autumn 1985) pp. 527-543.
Chalos, P. & Pickard, S., Information Choice and Cue Use: an Experiment in Group Information Proces-
sing, J ournal of Appl i ed Psychofogy (1985) pp. 634-641.
Chewning, E. G., Jr. & Harrell, A. M., The Effect of Information Load on Decision Makers’ Cue Utilization
Levels and Decision Quality in a Financial Distress Decision Task, Accounti ng, Organi zati ons and
Soci ety (1990) pp. 527-542.
Dlllard, J. F., Cognitive Science and Decision Making Research in Accounting, Accounti ng, Organ&a-
ti ons and Soci ety (1984) pp. 343-354.
Driver, M. J. & Mock, T. J., Human Information Processing, Decision Style Theory, and Accounting
Information Systems, The Accounti ng Revi ew (July 1975) pp. 490-508.
Einhom, H., Hogarth, R. M. & Klempner, E., Quality of Group Judgment, Psychol ogi cal Buffeti n (1977)
pp. 158-172.
Fertakis, J. P., On Communication, Understanding, and Relevance in Accounting Reporting, 7Ire Account-
fng Revi ew (October 1969) pp. 680-691.
Glass, G. V. & Stanley, J. C., Stati sti cal Methods i n Educati onal Psychol ogy, pp. 406-411 (Englewood
Cliffs, NJ: Prentice Hall, Inc., 1970).
Haberlin, W. E., Systems Used by the Thirty Largest U.S. Banks to Approve Major Credits, The J ournal of
Commerci af Bank Lendi ng (June 1976) pp. 40-43.
Hopwood, W., McKeown, J. & Mutchler, J., The Sensitivity of Financial Distress Prediction Models to
Departures from Normality, Contemporary Accounti ng Research (1990) pp. 284-296.
Houghton, K. A., Accounting Data and the Prediction of Business Failure: the Setting of Priors and the Age
of Data, J ournal of Accounti ng Research (Spring 1984) pp. 361-368.
Iselin, E. R., The Effects of Information Load and Information Diversity on Decision Quality in a Struc-
tured Decision Task, Accounti ng, Organi zati ons and Soci ety (1988) pp. 147-164.
Kerlinger, F. N., Foundati ons of Behavi oral Research, Third Edition, (Forth Worth: Harcourt Brace
Jovanovich, 1986).
Libby, R., The Use of Simulated Decision Makers in Information Evaluation, The Accounti ng Revi ew (J ul y
1975a) pp. 475-489.
Libby, R., Accounting Ratios and the Prediction of Failure: Some Behavioral Evidence, J oumaf of
Accounti ng Research (Spring 1975b) pp. 150-161.
Libby, R. & Blashfield, R. K., Performance of a Composite as a Function of the Number of Judges,
Organi rati onal Behavi or and Human Performance (1978) pp. 121-129.
Libby, R. & Lewis, B. L., Human Information Processing in Accounting: the State of the Art in 1982,
Accounti ng, Organi zati ons and Soci ety (1982) pp. 231-285.
McNeil, J. H., Guiding Credit/ Loan Approval: a Comparison of the Committee System and Individual
Lending Authority, Credi t and Fi nanci al Management (lune 1986) pp. 13-16.
Miller, H., Environmental Complexity and Financial Reports, The Accounti ng Revi ew (January 1972) pp.
31-37.
700 M. H. STOCKS and A. HARRELL
Miller, D. & Gordon, L. A., Conceptual Levels and the Design of Accounting Information Systems,
Decision Sciences (April 1975) pp. 259-269.
Miner, F. C., Group versus Individual Decision Making: an Investigation of Performance Measure, Deci-
sion Strategies, and Process Losses/Gains, Organizational Behavior and Human Perfownance (1984)
pp. 112-124.
Revsine, L., Data Expansion and Conceptual Structure, The Accounting Review (October 1970) pp. 704-
711.
Schick, A. G., Gordon, L. A. & Haka, S., Information Overload: a Temporal Approach, Accounting,
Organizatfons and Society (1990) pp. 199-220.
Schroder, H. M., Driver, M. J. & Streufert, S., Human I nformation Processing (New York: Holt, Rinehart
and Winston, 1967).
Schultz, J, J. & Reekers, P. M. J., The Impact of Group Processing on Selected Audit Disclosure Decisions,
J ournal of Accounting Research (Autumn 1981) pp. 482-501.
Shields, M. D., Effects of Information Supply and Demand on Judgment Accuracy: Evidence from Corpo-
rate Managers, The Accounting Review (April 1983) pp. 429-442.
Simnett, R., The Effect of Variation of Information Load on Information Selection and Processing by
Auditors, Working Paper, University of New South Wales, Sydney, Australia (1993).
Solomon, I., Probability Assessment by Individual Auditors and Audit Teams: an Empirical Investigation,
J ournal of Accounting Research (Autumn 1982) pp, 689-710.
Snowball, D., Information Overload and Accounting Reports: Too Much, Too Little or Just Right, Cost
and Management (May/June 1979) pp. 22-28.
Stocks, M., Tuttle, B. & Waldrup, B., An Examination of Task Complexity Effects on Performance in a
Financial Distress Prediction Task, Working Paper, University of Mississippi (1995).
Trotman, K. T., Yetton, P. W. & Zimmer, I. R., Individual and Group Judgments of Internal Control
Systems, J ournal of Accounting Research (Spring 1983) pp. 286-292.
Uecker, W., The Quality of Group Performance in Simplified Information Evaluation, J ournal ofdccount-
ing Research (Autumn 1982) pp. 388-402.
Wood, R. E., Task Complexity: Definition of the Construct, Organizational Behavior and Human
Decision Processes (1986) pp. 60-82.
Zavgren, C. V., Assessing the Vulnerability of Failure of American Industrial Firms: a Logistic Analysis,
J ournal of Business Finance and Accounting (1985) pp. 19-45.
doc_498474030.pdf
A judgment modeling experiment which used bank loan officers as participants was performed in a field
setting to examine the impact of an increase in information level on the judgments of individuals and
groups. The results indicate that, as predicted, groups reached judgments of higher quality than
individuals at both information levels. Also as predicted, groups responded more positively
Pergamon Accountfng, Organizations and Socfety, Vol. 20, No. 7/8, pp. 685-700, 1995
Copyright 0 1995 FJsevier Science Ltd
Printed in Great Britain. Au rights reserved.
0361-3682/95 $9.50+0.00
THE IMPACT OF AN INCREASE IN ACCOUNTING INFORMATION LEVEL ON
THE JUDGMENT QUALITY OF INDMDUALS AND GROUPS
MORRIS H. STOCKS
University of Mississippi
and
ADRIAN HARRELL
University of South Carolina
Abstract
A judgment modeling experiment which used bank loan officers as participants was performed in a field
setting to examine the impact of an increase in information level on the judgments of individuals and
groups. The results indicate that, as predicted, groups reached judgments of higher quality than
individuals at both information levels. Also as predicted, groups responded more positively (or less
negatively) than individuals to increases in information level, resulting in an increase in the judgment
quality gap between groups and individuals. lndividuals experienced greater information processing
difficulties than groups as information level increased.
Individuals and groups often must integrate
many distinct pieces of information into their
judgments. Creditors, for example, typically
must incorporate numerous financial ratios,
industry standards and other financial informa-
tion into their lending decisions. Based on
Schroder et al., (1967), it has been proposed
that, as the number of relevant pieces of avail-
able information increases, the quality of a per-
son’s judgments will initially increase. At some
point, however, further increases in the num-
ber of relevant pieces of information will result
in a decline in the quality of these judgments.
Previous accounting research has provided
empirical support for the information utiliza-
tion pattern predicted by the Schroder et al.
(1967) model (e.g. Casey, 1980; Iselin, 1988;
Chewning & Harrell, 1990).
Prior accounting information level studies
have, to date, considered only individual deci-
sion makers. Despite this emphasis upon indi-
viduals, many important judgments using
accounting information are frequently reached
in a group setting. For example, a group pro-
cess is often involved in bankruptcy, credit,
and loan default judgments (Haberlin, 1976;
McNeil, 1986). Although no prior research
has examined the impact of increases in infor-
mation level on the quality of group judgments,
accounting researchers have recognized the
need to study both individual and group judg-
ments. In general, prior research indicates the
judgments reached by groups appear to be of
higher quality than the judgments reached by
individuals (Schultz & Reekers, 1981; Solomon,
1982; Trotman et al., 1983; Chalos, 1985).
Noting that prior studies have focused on
individuals and that many important judg-
ments are reached by groups. Snowball
(1979) called for accounting information level
research which compares individuals and
groups. The purpose of this study is, there-
685
686 M. H. STOCKS and A. HARRELL
fore, to compare the effect of two different
levels of accounting information, six distinct
information cues and nine distinct information
cues, on the quality of the judgments reached
by individuals and groups. It is expected that
the judgment quality of groups will be higher
than that of individuals at both information
levels. In addition, groups are expected to
respond more positively than individuals to
an increase in accounting information level.
Accordingly, it is proposed that, as the informa-
tion level increases, the expected gap between
the judgment quality of groups and individuals
will increase, so that the quality of group judg-
ments will diverge from that of individuals.
RELEVANT LITERATURE AND HYPOTHESIS
DEVELOPMENT
The remainder of this paper is organized into
four sections. First, a brief review of the rele-
vant prior research is provided, along with the
hypothesis which guided this study. Second,
the research methodology is presented. Third,
the data analysis procedures are described and
the results are shown. Finally, the last section
presents a discussion of the implications of the
findings.
Schroder et al . (1967) developed a concep-
tual model of human information processing.
The Schroder, Driver and Streufert (SDS)
model postulates an inverted-U curve when
the complexity of the information environ-
ment is plotted against the level of information
processing exhibited by an individual or group
(Fig. 1). This inverted-U curve suggests that, as
the information environment becomes more
complex, the level of information processing
by the processing unit increases up to some
maximum processing level. After this point,
additional environmental complexity results in
a lower level of information processing.
The SDS model has served as the conceptual
basis for most of the accounting research
which examines the effect of environmental
complexity on judgment quality. Accounting
researchers have generally operationalized
environmental complexity in terms of task
complexity. The widely recognized concep-
tual model of task complexity originated by
F
0
E
_
TASK COMPLEXITY
Fig. 1. Inverted-U curve.
ACCOUNTING INFORMATION LEVEL 687
Wood (1986) can be used to describe the com-
plexity of various tasks. Wood discusses three
aspects of task complexity: component com-
plexity, coordinative complexity, and dynamic
complexity. The component complexity of a
task is determined by the number of distinct
information cues that must be processed and
the number of distinct acts that must be exe-
cuted in performing the task. Coordinative
complexity is determined by the form and
strength of the relationships between informa-
tion cues and acts, including the timing, fre-
quency, intensity, and location requirements
for performances of required acts. Dynamic
complexity refers to the need to adapt to
changes which occur in the cause-effect chain
or means-ends task relationships during the
performance of the task. The experimental
tasks employed in the accounting information
level studies appear to be adequately described
by Wood’s component complexity construct.
Two distinct streams of accounting compo-
nent complexity research can be identified.
The first research stream is relatively well
established. It consists of those studies with
experimental tasks which require their partici-
pants to incorporate a number of distinct
(independent) information cues into a judg-
ment or decision (e.g. Trotman et al ., 1983;
Chewning & Harrell, 1990). In these studies,
the component complexity of the experimen-
tal task may be determined by counting the
number of distinct (independent) cues
involved in the task. Component complexity
may be manipulated directly by increasing or
decreasing the number of distinct information
cues provided to the participant. A second
research stream consists of studies with experi-
mental tasks which involve both elements of
component complexity. In these studies, com-
ponent complexity may be manipulated by
increasing or decreasing the number of dis-
tinct information cues, the number of distinct
acts that must be executed to perform the task,
or both. Only a few studies have been per-
formed which use tasks that appear to involve
both elements of component complexity (Ise-
lin, 1988; Simnett, 1993; Stocks et al ., 1995).
Soon after the introduction of the SDS
model, a series of conceptual papers appeared
in the accounting literature which suggested
that the availability of large volumes of account-
ing information might lead to sub-optimal finan-
cial judgments (Fertakis, 1969; Revsine, 1970;
Miller, 1972; Ashton, 1974; Driver & Mock,
1975; Miller & Gordon, 1975). The first related
empirical study to appear in the accounting
literature was conducted by Casey (1980). In
his study, bank loan officers were assigned to
one of three information levels and were asked
to predict bankruptcy for ten actual firms. The
accuracy of these predictions was used to indi-
cate the effect of information level on judg-
ment quality. Although the participants
perceived an increase in the information
level, the information level manipulation did
not affect the quality of the participants’ judg-
ments. Since the experimental task was not
described in terms of a conceptual model of
task complexity, it’s not clear whether the
manipulation actually increased task complex-
ity or whether task complexity was simply per-
ceived to have increased. Libby and Lewis
(1982, p. 271) suggest Casey’s (1980) manipu-
lation may have been ineffective because the
“high” information level may have lacked addi-
tional information content.
Shields (1983) also used judgment accuracy
as a measure of the impact of information level
on individuals’ information processing. Unlike
the between-persons design used by Casey
(1980) Shields used a within-persons manipu-
lation of information level. The results of this
study provide support for the SDS model. More
recently, Schick et al . (1990) proposed that
information load level (task complexity) can
be considered a function of time available,
rather than a function of cognitive processing
ability.
A study by Chewning and Harrell(l990) may
represent the first effort to relate the number of
distinct information cues used to the number
available, thus determining the effect of infor-
mation levels on information utilization. Their
subjects were asked to make repeated financial
distress predictions at three information levels
688 M. H. STOCKS and A. HARRELL
(four, six and eight distinct information cues).
Accordingly, the actual information level was
manipulated by increasing the number of dis-
tinct information cues to increase complexity.
By estimating a regression model for each sub-
ject at each cue level, it was possible to mea-
sure cue usage at each information level. The
results showed that when cue usage was
plotted against cue availability, approximately
one-third of the subjects exhibited the informa-
tion usage implied by the SDS model. Because
the cases employed in the study represented
hypothetical firms, it was impossible to mea-
sure judgment accuracy. Therefore, cue
usage, consistency and consensus served as
indicants of judgment quality. The results indi-
cate that the subjects who exhibited the
inverted-U pattern scored significantly lower
on these indicants of judgment quality at the
eight-cue level.
Iselin (1988) may have been the first to
employ an experimental task which involves
both facets of component complexity: distinct
information cues and distinct acts to be exe-
cuted in the performance of the task. His parti-
cipants completed an experimental task which
required them to perform relatively routine
computations to arrive at a mathematically
determined response. Thus Iselin’s study
might be described as an examination of the
impact of variations in the two elements of
component complexity on individuals’ perfor-
mance of a relatively routine, highly structured
task.
Two recent studies have attempted to
extend Iselin’s research to contexts which
involve professional judgment. Simnett (1993)
asked auditors to make going concern predic-
tions for a series of firms using either a low or a
high information level. Some subjects in the
high information level received additional
cues which were considered to be distinct
from those used in the low information level,
while others received additional cues which
were considered to be redundant. This manip-
ulation was designed to “. . . separately iden-
tify the effect of both information load and
additional information” (Simnett, 1993, p. 7).
Stocks et al . (1995) manipulated the number
of distinct financial ratios presented to deci-
sion makers as well as the number of steps
required to classify the ratios. Their results
demonstrate that information level is only one
aspect of task complexity.
Apparently, no prior research has systemati-
cally compared the quality of individual and
group judgments across information levels. Sev-
eral studies have, however, compared the qual-
ity of the judgments reached by individuals and
groups at a single information level. Although
differences exist between these studies as to
the experimental tasks used, the decision con-
texts involved, and the type of individuals who
participated, their findings taken together sug-
gest that group judgments may be superior to
individual judgments in terms of accuracy
(Uecker, 1982; Chalos & Pickard, 1985), confi-
dence (Schultz & Reekers, 1981) and consen-
sus (Solomon, 1982).
Driver and Mock (1975) state, “Individuals
and groups use more information in decisions
with increasing information levels, up to a
point; then they begin to process less. . . .”
(p. 496). Although this suggests the informa-
tion processing of individuals and groups may
not differ, the findings of empirical studies sug-
gest that groups may encounter fewer informa-
tion processing difficulties than individuals at
high information levels. Chalos and Pickard
(1985) showed that when groups and indivi-
duals were supplied with a set of financial state-
ments to supplement three financial ratios used
in a financial distress task, the predictive accu-
racy of the groups increased while predictive
accuracy decreased for individuals. They sug-
gest that the groups may have benefited from
the added information, while the individuals
may have been unable to incorporate the addi-
tional information into their judgments. It is,
therefore, possible that individual and group
differences in judgment quality may be due in
part to a difference in the ability to process
large amounts of information.
Trotman et al . (1983) compared groups and
individuals at a single information level and
concluded that the groups outperformed indi-
ACCOUNTING INFORMATION LEVEL 689
viduals on all three measures (consensus, con-
sistency and cue usage). In discussing their
results, they suggest that, at higher information
levels, groups may be able to incorporate more
information into their decision processes than
individuals. While the present study examines
some of the propositions suggested by Trotman
et al. (1983), it differs from that study in several
important ways. First, this study investigates
judgment accuracy, an issue not considered
by Trotman et al. (1983). Second, this study
uses an experimental setting to examine their
suggestion that an increase in information level
should be expected to have a less adverse effect
on groups than on individuals. Finally, this
study employs a financial distress prediction
task while the Trotman et al. study examined
an internal control assessment.
In summary, the SDS theoretical model
implies that both individuals and groups will
encounter information processing difficulties
at higher information levels (Schroder et al.,
1967; Driver & Mock, 1975). Empirical stu-
dies, however, suggest that groups possess
greater information processing capabilities
than individuals and reach judgments of higher
quality than individuals. Accordingly, a gap in
judgment quality is predicted to exist between
groups and individuals. As suggested by Chalos
and Pickard (1985) and Trotman et al. (1983),
this judgment quality gap between groups and
individuals is predicted to become greater
with increases in information level. These pre-
dictions are summarized in the following
hypothesis:
The judgment quality of groups will be higher than the
judgment quality of individuals and will diverge posi-
tively from that of individuals as information level
increases.
The above hypothesis predicts the existence
of a positive ordinal interaction between the
type of participant (individual or group) and
the information level. The interaction is pre-
dicted to be ordinal in the sense that groups
are expected to outperform individuals at
both information levels. The interaction is pre-
dicted to be positive in that the difference in
judgment quality is expected to increase as
information level increases (Glass & Stanley,
1970). The following section describes the
methodology used to examine the above
hypothesis.
METHODOLOGY
An experiment was performed in a field set-
ting to examine the research hypothesis. The
participants, experimental design, data collec-
tion procedures, experimental task, and
research instrument are described below,
along with the judgment quality attributes
used to examine the hypothesis.
Participants
Experienced bank loan officers served as par-
ticipants in the experiment. Prior research has
suggested that bank loan officers are experi-
enced and appropriate subjects for studies
which require the prediction of financial dis-
tress (Casey, 1980; Libby, 1975a; Chalos,
1985). The participants were all currently
employed by the same large national bank.
The senior official who sponsored this study
stated that the bank used standard procedures
throughout its various regions, so all of the
participants were subject to the same person-
nel selection, training, and operating proce-
dures. Table 1 provides a summary of the
participants’ demographic information.
Design
As shown in Fig. 2, a 2 X 2 between-subjects
factorial design was used to compare the effect
of six-cue and nine-cue levels of information
(the LVL variable) on the quality of the judg-
ments reached by groups and individuals (the
TYPE variable). Accordingly, component com-
plexity was manipulated in this study by pro-
viding the participants with one of two
possible information levels, six distinct infor-
mation cues or nine distinct information cues.
As described below, the procedures used to
690 M. H. STOCKS and A. HARRELL
T
Y
P
E
Table 1
Summary of Participant’s Demographics Data
Attribute nean Between-Cell Predictor of Judgment
(Standard Deviation) Comparisons' Quality Indicants
Age
33 years Prob > F = .763 PI-& > F = .918
(8.7)
Lending 8.4 years Prob > F = .I95 Prob > F = -796
Experience
(7.5)
Annual $41,878 Prmb > F = .535 Prob > F = .634
Salary (15.1)
Education 91% Bachelor's Degree Prob > x2 = .153 Prob > F = .823
21% Master's Degree
Gender 79% Male
21% Female
Prob > 2' = .238 Prob > F = .456
'Two of the demographic measures, Education and Gender, were examined using
a x2 statistic due to the categorical nature of the data.
LVL
6 Cues 9 Cues
Individual
(16) (19)
n = 27 n = 28
Groups
Fig. 2 Experimental design.
identify the cues assured that each cue pos-
sessed unique information content. An
increase from six to nine information cues
therefore resulted in an actual increase in infor-
mation level. The participants were drawn
from four regions of a large national bank.
Each of the four regions was randomly
assigned to one of the four treatment cells
shown in Fig. 2, individual subjects receiving
six cues (16), individual subjects receiving nine
cues (I9), group subjects receiving six cues
(Gb), and group subjects receiving nine cues
(G9). According to the lending procedures of
the participating bank, these subjects were
often required to make financial judgments as
individuals and as members of small three- or
four-person groups. Accordingly, a group size
of three was selected, a suggested by Trotman
et al . (1983), Libby & Blashfield (1978), Ashton
(1986), and Einhorn et al . (1977).
Between-cell comparisons were made for
age, experience, salary, education and gender;
no significant differences were found. In addi-
tion, a series of MANOVA procedures demon-
strated that the demographic characteristics
had no impact on the judgment quality mea-
sures. Details of these results are provided in
Table 1.
Data col l ecti on procedures
The data were collected through the partici-
pating bank’s inter-office mail system. The
Director of Commercial Lending informed the
loan officers that the bank had agreed to parti-
ACCOUNTING INFORMATION LEVEL 691
cipate in the research project and urged each
participant to respond. Both groups and indivi-
duals received detailed instructions for the
judgment task. All participants were informed
their specific responses would remain anon-
ymous. The office of the Director of Commer-
cial Lending served as the collection point for
the research instruments. Of the 190 research
instruments distributed, 112 (58.9%) usable
responses were received. This included 55 indi-
vidual responses and 57 three-person group
responses (see Fig. 2). Accordingly, a total of
226 bank loan officers participated in the study.
In an effort to determine whether a non-
response bias existed, MANOVA was used to
compare the first 25 individual and first 25
group responses received with the last 25 of
each category received. No significant differ-
ences were found in either instance. These
findings, in conjunction with the relatively
high response rate, suggest that the sample
was not biased by non-responses.
Care was taken to assure that a single indivi-
dual didn’t complete the group instrument.
First, a check question was inchtded in the
research instrument which asked for the num-
ber of individuals that actively participated in
the completion of the instrument. In addition,
each group member was asked to complete a
separate individual information sheet. This
information sheet included a question which
asked whether he or she actively participated
in the group judgment process. These sheets
were then sealed by the individual and
included in the packet to be returned.
Further, the instructions emphasized the
importance of face-to-face group interaction
in completing the judgment task.
Previous research (Miner, 1984) has demon-
strated that the group process is most effective
when group members make individual judg-
ments prior to reaching a consensus judg-
ment. Chalos (1985) found that interacting,
face-to-face committees outperformed mec-
hanically aggregated committees. Accordingly,
group members were instructed to form indi-
vidual judgments for each case and then, face-
to-face, arrive at a group consensus. However,
a potential limitation of the present study is
that, although the manipulation questions
completed by the participants reported they
followed these procedures, the field setting
of the study meant the interactive group pro-
cess was not actually observed by the
researchers.’
J udgment task
The participants completed an experimental
task which involved using either six or nine
distinct information cues to arrive at profes-
sional judgments. Groups and individuals
were required to make a series of financial dis-
tress predictions (Chalos, 1985; Chewning &
Harrell, 1990). Financial distress was defined
as either bankruptcy, loan default, failure to
meet a preferred stock dividend payment or
asset liquidation.
In order to assure that each cue possessed
unique information content, the accounting
ratios the participants used as judgment cues
were derived through a factor analysis process
similar to that employed by Libby (1975b) and
Chewning & Harrell(l990). A set of forty finan-
cial ratios which have been shown to have
predictive ability in a bankruptcy prediction
was identified from the recent accounting lit-
erature (Zavgren, 1985; Hopwood et al., 1990;
Chewning & Harrell, 1990). The forty ratios
were then computed for each of the 1450
manufacturing firms listed on the Compustat
tapes for fiscal year 1986. These ratios were
factor analyzed, using a varimax rotation, to
reduce the redundancy within the ratios and
to identify common constructs. The nine fac-
tors accounting for the largest portion of the
variance within the data were selected. Each
’ However, since the group vs individual comparisons made in this and previous studies report the superiority of inter-
active group decisions, we believe the inclusion of a group response which was not the product of an interactive group
would bias the results of our study conservatively.
692 M. H. STOCKS and A. HARRELL
Firm No. 1
Given the following information:
1.
2.
3.
4.
5.
6.
7.
8.
9.
In
Firm Size. . . . . . . . . . . . .
Cash / Total Assets. . . . . . . .
Total Equity / Total Liabilities .
Earnings Before Interest and Taxes
Total Assets / Long Term.Debt. . .
Net Income / Net Worth . . . . . .
Sales / Fixed Assets . . . . . . .
Earnings Before Interest and Taxes
Working Capital / Sales. . . . . .
. . . . . . BOTTOM 113 l
. . . . . . TOP 113 *
. . . . . . BOTTOM 113
/ Sales . . TOP 113 l
. . . . . . BOTTOM 113 *
. . . . . . TOP 113
. . . . . . TOP 113
/ Assets. . BOTTOM 113 *
. . . . . . BOTTOM 113 *
your view, indicate below the likelihood this firm will
experience financial distress during the next three years.
Probably Will Not Experience Probably Will Experience
Financial Distress Financial Distress
VERY VERY
LOW 0 1 2 3 4 5 6 7 8 9 HIGH
CHANCE CHANCE
* - Judgment Cues provided in six-cue cases.
Fig. 3. Example of nine-cue case.
Position
Relative to
Industry
factor contained an eigenvalue greater than 1.4
and collectively they accounted for 80% of the
total variance. The ratio with the highest load-
ing on each factor was chosen as the cue to be
used in the task, as shown in Fig. 3.
Correhtion analysis indicates nine of the 36
pairs were significantly correlated. However,
most of these correlations were small. Only
two of these correlations were greater than
0.20 and the median correlation was 0.015,2
demonstrating that each of the financial ratios
provide the unique information content appro-
priate for distinct information cues. These cues
represent measures of profitability, liquidity
and capital structure and are similar to those
derived by Zavgren (1985), Libby (1975b) and
Chewning & Harrell ( 1990) . In addition, they
were reviewed for reasonableness by the Direc-
tor of Commercial Lending for the participating
institution. In summary, this process identified
relatively uncorrelated ratios which were use-
ful in predicting financial distress.
’ The highest correlation between any two cues, EBIT/Sales and EBIT/Assets, was 0.45, indicating the two variables had
about 20% common variance and about 80% independent variance. The only other correlation greater than 0.20 was
between Total Equity/Total Liabilities and CasWotaI Assets (r = 0.28). This supports the view that each cue provided
unique information content.
ACCOUNTING INFORMATION LEVEL 693
The research instrument
Each individual and group reached financial
distress predictions for 42 cases, with 32 cases
representing hypothetical firms and 10 cases
representing actual firms. A l/ 2 replicate of a
26 factorial design was incorporated into the 32
hypothetical cases completed by the partici-
pants who received six cues. For those who
received nine cues, a I/ I6 replicate of a 29
factorial design was incorporated into the 32
hypothetical cases. Both the six-cue and nine-
cue partial replicates conformed to an orthogo-
nal design, which eliminated any correlation
among the independent variables. In both
instances, the cases were presented in random
order, with the 10 cases representing actual
firms being randomly interspersed to make
them indistinguishable from the hypothetical
cases. Following Chewning & Harrell (1990)
the cues in the cases representing hypothetical
firms were manipulated at two levels (the top
l/ 3 and bottom l/ 3 of the firm’s industry). The
cues for the 10 cases representing actual firms
were presented at three levels (top l/ 3, middle
l/ 3, and bottom l/ 3), as some ratios appeared
in the middle of the industry as well as the top
and bottom. The six cues used in 16 and G6
were randomly selected from the nine cues
used in 19 and G9. The instructions informed
participants that each cue had been shown to
be useful in predicting financial distress and
emphasized that all of the cues should be
used in arriving at their judgments. An exam-
ple of the nine cue cases is shown in Fig. 3; the
asterisked judgment cues in Fig. 3 indicate
those used in the six-cue cases3
In order to provide a measure of accuracy,
five failed and five non-failed actual firms were
selected from the list provided by Zavgren
(1985). Obvious cases were avoided. For exam-
ple, if nearly all of the ratios of a firm fell within
either the “Bottom l/ 3” or within the “Top
l/ 3” of it’s industry, a financial distress predic-
tion for that firm is obvious, so such firms wer-
en’t used. The Wall Street Journal Index was
reviewed for the five non-failure firms for the
appropriate three-year period to assure they
did not experience any non-bankruptcy form
of financial distress. This was necessary
because financial distress was broadly defined
in the instructions to participants. The ratios
for the failed firms were computed for the
third year prior to bankruptcy. For non-failed
firms, the ratios were computed for the third
year prior to the bankruptcy of the correspond-
ing failed firm.
Previous research (Casey & Selling, 1986;
Houghton, 1984) has suggested that financial
distress judgments may be sensitive to the dis-
closure of the failure rate in the sample. The
50% failure rate (five failed, five non-failed) of
the ten actual firms used in the study exceeds
the failure rate of the overall population of
such firms. Therefore, subjects were informed
that the sample of firms used in the study was
not randomly selected and that the proportion
of failed firms was higher than for a random
sample of firms. The actual failure rate was
not disclosed in order to avoid judgment biases
which might result from “gaming” by the
participants.
Dependent variables
In this study, judgment quality refers to the
degree of judgment excellence, as measured by
four attributes; cues used, accuracy, consis-
tency, and consensus. All four of these attri-
butes have been used as indicants of
judgment quality in prior studies (e.g. Casey,
1980; Chewning & Harrell, 1990). Cues used
indicates the number of cues actually inte-
grated into the judgment process. This indi-
cant was computed by estimating a regression
model for each individual and each group,
’ In an effort to check the effectiveness of the information level manipulation, all participants were asked to indicate
which of the cues were useful in the prediction task. A comparison was made of the number of cues that were subjectively
rated as useful between participants in the two information level manipulations. The mean numbers of cues indicated
useful was 8.3 for nlnecue subjects and 5.6 for six-cue subjects. A two-sample C-test indicates that subjects in the nine-cue
manipulation perceived the availability of significantly more useful information than subjects who received six-cues (T =
13.49; Prob > T = 0.0001).
694
M. H. STOCKS and A. HARRELL
using the six or nine cues as independent vari-
ables and the financial distress predictions as
the dependent variable. Similar to Trotman et
al . (1983) and Chewning & Harrell (1990) the
number of cues that were integrated into the
judgment process was determined by counting
the number of significant main effects (at p c
0.05) in each of the regression models. Accu-
racy relates to how accurately the individual or
group was able to predict the future status of
the 10 actual firms. Following Chewning &
Harrell (1990) and Trotman et al . (1983), the
adjusted R2 value obtained from each indivi-
dual’s or group’s regression model was used
to indicate consistency. Consensus was
defined as the average paired correlation
between a participant’s judgments and the
judgments of all other participants from the
same experimental treatment. For example,
the judgments of an individual who received
Table 2
Summary of Univariate Analysis of Variance Results
JUDGMENT QUALITY F-VALUE PROBABILITY
INDICANTS
CUE8 USED
TYPE 31.44 .OOOl
LVL 3.31 .0715
TYPE x LVL 7.73 .0064
Contrasts
16 vs 19 .45 .5024
66 vs G9 10.78 .0014
16 vs 66 3.93 .0501
ACCURACY
TYPE
LVL
TYPE x LVL
Contrasts
16 vs 19
GC vs G9
16 vs 66
5.01 .0272
15.96 .OOOl
3.00 .0861
2.52 .1157
16.70 .OOOl
.13 .7229
CONSISTENCY
TYPE 35.18 .OOOl
LVL 11.79 .OOOl
TYPE X LVL 5.59 .0199
Contrasts
16 vs 19 16.51 -0001
66 vs G9 .58 .4467
16 vs 66 6.25 .0139
ACCOUNTING INFORMATION LEVEL 695
‘cue Usage: more cues were used in the judgments made by groups than by individuals at
both the six-cue @a = 5.3 vs 3~~s = 4.6; F = 3.93,~ = 0.0501) and nine-cue (Zos = 6.4 vs 4s
= 4.4; F = 33.82, p = 0.0001) levels.
bAcc~racy: the accuracy of the judgments reached by groups and individuals did not differ
at the sixcue level (2~ = 5.2 vs f IG = 5.1; F = 0.13, p =0.7229). Groups did, however,
reach more accurate judgments than individuals at the nine-cue level (32os = 6.4 vs 2,~ =5.6
F = 8.03, p = 0.0055).
‘Consistency: the judgment consistency of the groups was greater at both the six-cue (*a
= 0.83 vs _z’r = 0.76; F = 6.25, p =0.0139) and nine-cue levels (*os = 0.81 vs 4s = 0.64; F =
35.03, p =0.0001).
dConsensus: consensus between the judgments of groups was greater at both the sixcue
(& = 0.72 vs fib = 0.63; F = 53.39, p =0.0001) and nine-cue levels (& = 0.67 vs 4s =
0.52; F = 87.42, p = 0.0001)
six cues were correlated with the judgments of
all other individuals who received six cues. The
mean of these paired correlations served as a
measure of consensus (Trotman et al., 1983;
Chewning & Harrell, 1990).
ANALYSIS AND RESULTS
Several analysis steps. were required to exam-
ine the data provided by the experiment. Initi-
ally, MANOVA was used to examine the
relationship between the experimental manip-
ulations and the four indicants of judgment
quality. The MANOVA results indicate signifi-
cant main effects for both the TYPE (F-test, p
= 0.0001) and LVL (F-test,p =0.0001) variables
and a significant TYPE X LVL interaction (F-
test, p = 0.0001). While these results provide
preliminary support, they must be considered
carefully. First, the existence of the significant
two-way interaction makes interpretation of
the TYPE and LVL main effects difficult.* Sec-
ond, these results, alone, do not provide suffi-
cient information to examine the directional
predictions of the research hypothesis. Accord-
ingly, four univariate analyses of variance
(ANOVA), along with the appropriate con-
trasts of cell means, were performed to sepa-
rately examine the relationship between the
independent variables and each of the indi-
cants of judgment quality. The results of these
analyses are summarized in Tables 2 and 3 and
in Fig. 4.
Examination of hypothesis
In the context of the experiment, the
research hypothesis predicts the judgment
4 However, Kerlinger (1986, p.242) suggests that it is possible to interpret signiticant main effects when the interaction is
ordinal in nature. An examination of Fig. 4 indicates that such is the case in the present study.
696 M. H. STOCKS and A. HARRELL
Panel A
Cues Used
6 cu.* 9 c ues
Cues Available
Panel C
Consistency
CO~S‘~t.3~~~
I/
Cues Available
- Groups + Indlvlduolr
Fig.
quality attributes will be higher for groups than level (2~6 = 5.2 vs 216 = 5.1; F = 0.13, p =
for individuals at both the six-cue and nine-cue 0.7229). Groups did, however, reach more
levels. As predicted, more cues were used in accurate judgments than individuals at the
the judgments made by groups than by indivi- nine-cue level (2~9 = 6.4 vs x19 = 5.6; F =
duals at both the six-cue (3~6 =5.3 vs *I~=4.6; 8.03, p = 0.0055). As predicted, the judgment
F= 3.93,~ =0.0501) and nine-cue (%c9 =6.4 vs consistency of the groups was greater at both
S?t9 = 4.4; F = 33.82, p = 0.0001) levels. The the six-cue (2~6 =0.84 vs 216 =0.76; F = 6.25,~
accuracy of the judgments reached by groups =0.0139) and nine-cue levels (Zo9 =0.81 vs 219
and individuals did not differ at the six-cue = 0.64; F = 35.03, p = 0.0001). Consensus
Panel B
Accuracy
7
*ccurmy
6.4
by “....:...:....
56
5
6 cuss 9 c ues
Cues Available
Panel D
Consensus
consensus
0.8
0.5
t
_.-
6 C”es 9 c um
Cues Avallable
- Groupr + I ndi vi dual s
4.
ACCOUNTING INFORMATION LEVEL 697
between the judgments of groups was greater
at both the six-cue (266 =0.72 vs &6 =0.63; F =
27.24, p = 0.0001) and nine-cue levels (%09 =
0.67 vs %tIs = 0.52; F = 87.42, p = 0.0001).
Accordingly, although the accuracy of indivi-
duals and groups did not differ at the six-cue
level, the preponderance of the results support
the prediction that the judgment quality of
groups will be higher than that of individuals
at both information levels.
The research hypothesis also predicts the
existence of a positive ordinal interaction
between the type of judge (‘IYPE) and the
information level (LVL.) variables for each of
the judgment quality attributes. The mean
values of the four judgment quality indicants
for both groups and individuals are plotted at
the six-cue and nine-cue levels in Fig. 4. As
shown in Table 2, the results of the ANOVA
for cues used indicates the existence of a sig-
nificant TYPE X LVL interaction (F = 7.73, p =
0.0064) indicating that the cue usage pattern
of groups and individuals differed as informa-
tion level increased from the six-cue to nine-
cue level. In panel A of Fig. 4, the plotted
mean values for cues used indicate the pre-
dicted positive ordinal interaction. The second
ANOVA, using judgment accuracy as the
dependent variable, indicates the existence of
a weak TYPE X LVL interaction (F = 3.00, p =
0.0861>, providing limited support for the pre-
dicted relationships. The plotted mean values
for judgment accuracy indicate the increasing
gap in the accuracy of judgments implied by
the predicted positive ordinal interaction
(panel B of Fig. 4).
The ANOVA for judgment consistency indi-
cates the existence of a significant TYPE X LVL
interaction (F = 5.59, p = 0.0199). Panel C of
Fig. 4 again demonstrates the predicted posi-
tive ordinal interaction. Finally, the ANOVA
using consensus as the dependent variable
also indicates the existence of a significant
TYPE X LVL interaction (F = 8.00, p =
0.0056). When the cell means are plotted, it
is clear that the gap in judgment consensus
between groups and individuals increases as
the information level increases. The evidence
provided by the ANOVAs and the plotted mean
values of the indicants of judgment quality
appear to support the prediction that the judg-
ment quality of groups diverges positively from
the judgment quality of individuals as the infor-
mation level increases.
Additional @dings
The SDS model predicts that both individuals
and groups will experience information proces-
sing difficulties as the information level
increases. For individual participants in this
study, as shown in Tables 1 and 2, two indi-
cants of judgments quality (cues used and accu-
racy) did not change, while two indicants
(consistency and consensus) declined as the
information level increased. The individual par-
ticipants appear to have experienced moderate
information processing difficulties with
increased information level, as the quality of
their judgments does not appear to have bene-
fited from the availability of additional informa-
tion. For the group participants, as shown in
Tables 2 and 3, two judgment quality indicants
(cues used and accuracy) increased as informa-
tion level increased. One indicant (consis-
tency) did not change, and only one indicant
(consensus) declined. These results suggest
that groups experienced few information pro-
cessing difficulties with increased information
level since the quality of their judgments appar-
ently benefitted from the availability of addi-
tional information.
DISCUSSION
This study compares the effect of increasing
levels of information on the quality of the judg-
ments reached by individuals and groups. Sev-
eral findings can be reported. First, the
judgment quality of groups was higher than
that of individuals at both the six-cue and
nine-cue information levels. The only excep-
tion to this was the judgment accuracy mea-
sure at the six-cue level. ‘Second, as the
information level increased, the individual par-
ticipants appear to have experienced moderate
698
M. H. STOCKS and A. HARRELL
information processing difficulties, while the
group participants appear to have experienced
few information processing difficulties. As a
result, the gap in judgment quality which
existed between individuals and groups at the
six-cue information level became larger at the
ninecue information level, causing the judg-
ment quality of groups and individuals to
diverge.
Some limitations of this study should be
noted. As mentioned earlier, the empirical
data were gathered across several southeastern
states. The research instrument was forwarded
to the subjects through the participating bank’s
interoffice mail system and experienced loan
officers completed the experimental task in
their own work environment. Although this
approach allowed the data to be gathered in a
naturalistic field setting, it also means we were
unable to be present and personally observe
the group decision-making process. It is, there-
fore, difficult to speculate about particular
aspects of the group decision processes asso-
ciated with the superior group performance.
The results do, however, have potential impli-
cations for accountants, both as providers of
financial information and in their advisory role
to management. The findings imply that when
feasible, important business judgments should
be reached by groups, for groups can be
expected to consistently reach judgments of
higher quality that those reached by indivi-
duals. This is especially true for judgments
which are complex in terms of the number of
information inputs which should be considered.
In some business settings, the additional cost
of a group decision process may outweigh the
benefit derived from the process. In other situa-
tions, logistical considerations may suggest that
group decisions are impractical. When the
group judgment process isn’t feasible, an
attempt should be made to structure the judg-
ment process so as to limit to a relatively small
set the number of relevant information inputs
to be considered. Perhaps individuals should
be provided with only those information items
which possess the greatest predictive ability for
a particular judgment.
The evident importance of judgment quality
to the success of business organizations sug-
gests the need for further research of this
topic. A straightforward extension of this study
would be to examine group judgments at an
information level greater than the nine cues
considered in this study. It is conceivable
that, given the availability of more than nine
inputs, groups may be able to successfully inte-
grate more information into their judgments. A
second extension would be to examine the
impact on judgment quality of providing both
groups and individuals with a judgment aid in
the form of an appropriate judgment model. A
third extension of this study is to investigate
the effects of variations in information level
and presentation mode on group and indivi-
dual decision makers. Several studies have
examined the impact of presentation mode in
a multiple cue judgment task on individuals
(e.g. Amer, 1991). Since groups often make
these type of judgments, it appears appropri-
ate to consider the impact of these variables
on groups.
The results of this study suggest that, at
higher information levels, groups are able to
generate a judgment model that incorporates
more informational cues into a judgment and
to apply that model more consistently than
individuals. In our study, this led to more accu-
rate predictions which were in greater agree-
ment with other similar decision makers. A
final suggestion for extending this research is
to investigate why the judgment quality gap
between groups and individuals increases as
information level increases. The research
design and measurement approach employed
in this study does not appear to be appropriate
for such a study, which probably would
involve an examination of the reasoning pro-
cesses and procedures employed by groups
and individuals. We suggest that cognitive
science methodologies (e.g. Dillard, 1984) are
likely to be more appropriate for such an inves-
tigation. It is hoped that the present study will
stimulate others to a further examination of
these issues.
ACCOUNTING INFORMATION LEVEL
BIBLIOGRAPHY
699
Amer, T. S., An Experimental Investigation of Multi-cue Financial Information Display and Decision
Making, Journal oflnformation Systems (Fall 1991) pp. 18-33.
Ashton, R., Behavioural Implications of Information Overload in Managerial Accounting Reports, Cost
and Management (July/ August 1974) pp. 37-40.
Ashton, R., Combining the Judgments of Experts: How Many and Which Ones, Organi rati onaf Behavior
and Human Deci si on Processes (1986) pp. 405-414.
Casey, C. J., Jr., Variation in Accounting Information Load: The effect on Loan Officers’ Prediction of
Bankruptcy, The Accounti ng Revi ew (January 1980) pp. 36-49.
Casey, C. J., Jr. & Selling, T. I., The Effect of Task Predictability and Prior Probability Disclosure on
Judgment Quality and Confidence, The Accounti ng Revi ew (April 1986) pp. 302-317.
Chalos, P., Financial Distress: a Comparative Study of Individual, Model, and Committee Assessments,
J ournal of Accounti ng Research (Autumn 1985) pp. 527-543.
Chalos, P. & Pickard, S., Information Choice and Cue Use: an Experiment in Group Information Proces-
sing, J ournal of Appl i ed Psychofogy (1985) pp. 634-641.
Chewning, E. G., Jr. & Harrell, A. M., The Effect of Information Load on Decision Makers’ Cue Utilization
Levels and Decision Quality in a Financial Distress Decision Task, Accounti ng, Organi zati ons and
Soci ety (1990) pp. 527-542.
Dlllard, J. F., Cognitive Science and Decision Making Research in Accounting, Accounti ng, Organ&a-
ti ons and Soci ety (1984) pp. 343-354.
Driver, M. J. & Mock, T. J., Human Information Processing, Decision Style Theory, and Accounting
Information Systems, The Accounti ng Revi ew (July 1975) pp. 490-508.
Einhom, H., Hogarth, R. M. & Klempner, E., Quality of Group Judgment, Psychol ogi cal Buffeti n (1977)
pp. 158-172.
Fertakis, J. P., On Communication, Understanding, and Relevance in Accounting Reporting, 7Ire Account-
fng Revi ew (October 1969) pp. 680-691.
Glass, G. V. & Stanley, J. C., Stati sti cal Methods i n Educati onal Psychol ogy, pp. 406-411 (Englewood
Cliffs, NJ: Prentice Hall, Inc., 1970).
Haberlin, W. E., Systems Used by the Thirty Largest U.S. Banks to Approve Major Credits, The J ournal of
Commerci af Bank Lendi ng (June 1976) pp. 40-43.
Hopwood, W., McKeown, J. & Mutchler, J., The Sensitivity of Financial Distress Prediction Models to
Departures from Normality, Contemporary Accounti ng Research (1990) pp. 284-296.
Houghton, K. A., Accounting Data and the Prediction of Business Failure: the Setting of Priors and the Age
of Data, J ournal of Accounti ng Research (Spring 1984) pp. 361-368.
Iselin, E. R., The Effects of Information Load and Information Diversity on Decision Quality in a Struc-
tured Decision Task, Accounti ng, Organi zati ons and Soci ety (1988) pp. 147-164.
Kerlinger, F. N., Foundati ons of Behavi oral Research, Third Edition, (Forth Worth: Harcourt Brace
Jovanovich, 1986).
Libby, R., The Use of Simulated Decision Makers in Information Evaluation, The Accounti ng Revi ew (J ul y
1975a) pp. 475-489.
Libby, R., Accounting Ratios and the Prediction of Failure: Some Behavioral Evidence, J oumaf of
Accounti ng Research (Spring 1975b) pp. 150-161.
Libby, R. & Blashfield, R. K., Performance of a Composite as a Function of the Number of Judges,
Organi rati onal Behavi or and Human Performance (1978) pp. 121-129.
Libby, R. & Lewis, B. L., Human Information Processing in Accounting: the State of the Art in 1982,
Accounti ng, Organi zati ons and Soci ety (1982) pp. 231-285.
McNeil, J. H., Guiding Credit/ Loan Approval: a Comparison of the Committee System and Individual
Lending Authority, Credi t and Fi nanci al Management (lune 1986) pp. 13-16.
Miller, H., Environmental Complexity and Financial Reports, The Accounti ng Revi ew (January 1972) pp.
31-37.
700 M. H. STOCKS and A. HARRELL
Miller, D. & Gordon, L. A., Conceptual Levels and the Design of Accounting Information Systems,
Decision Sciences (April 1975) pp. 259-269.
Miner, F. C., Group versus Individual Decision Making: an Investigation of Performance Measure, Deci-
sion Strategies, and Process Losses/Gains, Organizational Behavior and Human Perfownance (1984)
pp. 112-124.
Revsine, L., Data Expansion and Conceptual Structure, The Accounting Review (October 1970) pp. 704-
711.
Schick, A. G., Gordon, L. A. & Haka, S., Information Overload: a Temporal Approach, Accounting,
Organizatfons and Society (1990) pp. 199-220.
Schroder, H. M., Driver, M. J. & Streufert, S., Human I nformation Processing (New York: Holt, Rinehart
and Winston, 1967).
Schultz, J, J. & Reekers, P. M. J., The Impact of Group Processing on Selected Audit Disclosure Decisions,
J ournal of Accounting Research (Autumn 1981) pp. 482-501.
Shields, M. D., Effects of Information Supply and Demand on Judgment Accuracy: Evidence from Corpo-
rate Managers, The Accounting Review (April 1983) pp. 429-442.
Simnett, R., The Effect of Variation of Information Load on Information Selection and Processing by
Auditors, Working Paper, University of New South Wales, Sydney, Australia (1993).
Solomon, I., Probability Assessment by Individual Auditors and Audit Teams: an Empirical Investigation,
J ournal of Accounting Research (Autumn 1982) pp, 689-710.
Snowball, D., Information Overload and Accounting Reports: Too Much, Too Little or Just Right, Cost
and Management (May/June 1979) pp. 22-28.
Stocks, M., Tuttle, B. & Waldrup, B., An Examination of Task Complexity Effects on Performance in a
Financial Distress Prediction Task, Working Paper, University of Mississippi (1995).
Trotman, K. T., Yetton, P. W. & Zimmer, I. R., Individual and Group Judgments of Internal Control
Systems, J ournal of Accounting Research (Spring 1983) pp. 286-292.
Uecker, W., The Quality of Group Performance in Simplified Information Evaluation, J ournal ofdccount-
ing Research (Autumn 1982) pp. 388-402.
Wood, R. E., Task Complexity: Definition of the Construct, Organizational Behavior and Human
Decision Processes (1986) pp. 60-82.
Zavgren, C. V., Assessing the Vulnerability of Failure of American Industrial Firms: a Logistic Analysis,
J ournal of Business Finance and Accounting (1985) pp. 19-45.
doc_498474030.pdf