Professional Documents
Culture Documents
Decision-Making Procedure and Decision Quality
Decision-Making Procedure and Decision Quality
DECISION-MAKING PROCEDURE
AND DECISION QUALITY
BRANT R. BURLESON
Purdue University
BARBARA J. LEVINE
Mellon Bank,
Pittsburgh, Pennsylvania
WENDY SAMTER
Purdue University
557
558 HUMAN COMMUNICATION RESEARCH / Summer 1984
Several studies have found that the decisions produced through the
statistical pooling of individual judgments are just as good, if not
superior to, the decisions reached by groups engaging in natural social
interaction (e.g., Beran, Albert, Loiseaux, Mayfield, & Wright, 1958;
Kaplan, Skogstad, & Girshick, 1950; see the review of Hare, 1976).
Building on these results, Davis (1973) proposed that the central
determinants of group decisions are members’ initial preferences and a
decisional rule (usually not known consciously to members) for
combining these preferences. As Poole, McPhee, and Siebold (1982)
note, Davis’ “social decision scheme” model has the net effect of
eliminating social interaction as a meaningful force in the production of
group decisions.
A second noninteractive decision procedure permits only limited
communication among group members under highly restrictive con-
ditions. Referred to both as the “nominal group technique” (Delbecq,
Van der Ven, & Gustafson, 1975) and the “Delphi” technique (Helmer,
1967), this method permits communication among group members
(usually only in written form) through the mediation of a leader-
recorder or moderator. By carefully controlling both the content of
messages and the channels through which messages are conveyed, the
nominal group procedure seeks to maximize the rational process and
the substance of members’ contributions, and to eliminate problems
arising from open social interaction. Several studies have compared the
effectiveness of nominal groups and interacting groups in the context of
brainstorming tasks; these studies generally have found that groups
employing the nominal procedure generate both a greater number of
ideas and more good ideas than naturally interacting groups (e.g.,
Dunnette, Campbell, & Jaastad, 1963; Philipsen, Mulac, & Dietrich,
1979; Taylor, Berry, & Block, 1958; see the review of Lamm &
Trommsdorf, 1973). In particular, when groups contain individuals
with high levels of “communication apprehension,” more ideas have
been generated in nominal groups than in interacting groups (Jablin,
1981; rlablin, Seibold, & Sorenson, 1977). The superiority of the
nominal group procedure also has been documented with tasks other
than brainstorming; for example, Kanekar and Rosenbaum (1972)
found that nominal groups outperformed interacting groups on an
anagram task. (For a general review of research supporting the efficacy
of the nominal or Delphi technique, see Emmons & Kaplan, 197 1 .)
While recognizing the potentially debilitating effects of social
interaction and the empirical support for the merit of staticized and
560 HUMAN COMMUNICATION RESEARCH / Summer 1984
METHODS
SUBJECTS
GENERAL PROCEDURES
EXPERIMENTAL TREATMENTS
RESULTS
groups in the staticized condition, this was the average total deviation
score for each four-person group subsequent to discussion of the Island
Problem; for the nominal condition, this was the average total deviation
score for the group subsequent to the controlled information exchange
procedure; and for the interacting groups, posttest decision quality was
the total deviation score derived from the consensual rankings gener-
ated during the course of the open discussion. For analyses on group
decisions, statistical power (with alpha = .05) was .37 for a medium effect
size and .83 for a large effect size.
Prior to conducting the main APV analysis, it was first necessary to
ensure that there was no interaction between pretest decision quality
and membership in experimental condition. To assess this interaction,
the two dummy variables coding the nominal and interacting experi-
mental treatments were each multiplied by the pretest scores and the
resulting product terms were regressed hierarchically on the posttest
scores subsequent to both the pretest scores and the experimental
conditions. The results of this analysis indicated that the interaction
term was not significant, F (2,24) = .047, ns; thus the regression lines for
the experimental conditions on posttest scores can be assumed t o be
parallel.
In the main analysis, pretest decision quality was regressed on
posttest decision quality initially. Somewhat surprisingly, pretest and
postest decision quality were not related significantly [r = .27, r2 = .07; F
( 1,28) = 2.15, p = .15]. The effect of decision procedure was entered into
the regression next. The total regression explained a significant amount
of variance in posttest decision quality [R2 .60, R2= .36; F (3,26) 14.94,
p < ,011. More important, decision procedure was found to account
uniquely for a large and statistically significant amount of variance in
posttest decision quality [semipartial R = .54, semipartial R2 .29; F
(2,26) = 5.96, p < .02]. Since the primary concern here is with the residual
variance in posttest decision quality (i.e., with the variance in posttest
scores after partialling out the effect of pretest scores), a more
appropriate index of effect size for the experimental treatments is the
squared multiple partial correlation; in the present context, the partial
R2 .32.
Table 1 presents the pretest means, the unadjusted posttest means, and
the posttest means adjusted for pretest scores for groups in each of the
three experimental conditions. Of central concern are the adjusted
posttest means. Our first hypothesis predicted that nominal groups
would produce better decisions than staticized groups and that inter-
Burleson et al. / DECISION-MAKING 567
TABLE 1
Means and Standard Deviations for Quality of Group Decisions
Decision Procedure
Type of Mean Staticized Nominal Interacting
TABLE 2
Means and Standard Deviations for Quality of individual Decisions
~~
Decision Procedure
Type of Mean Staticized Nominal Interacting
significant, t (1 16) = 4.35, p < .001; this contrast accounted for 14% of
the residualized variance in posttest decision quality. Finally, subjects in
the interacting condition made significantly better individual decisions
on the posttest than those in the nominal condition, t (1 16) = 2.63, p <
.o 1.
DISCUSSION
The results of this study provide clear and consistent support for the
value of social interaction in small group decision-making. It was
hypothesized that groups employing an interacting decision procedure
would produce better decisions than groups employing either nominal
or staticized decision procedures; this hypothesis was confirmed
strongly. Moreover, the results of this investigation provide some
support for the existence of what Collins and Guetzkow (1964) term the
“assembly-effect bonus.” Not only were group decisions in the interact-
ing condition significantly better than those made in the nominal and
staticized conditions, but the consensual decisions reached in the
interacting groups were, on the average, significantly better than the
“pooled” decisions of these groups (which were created by averaging
individual preferences expressed subsequent to the achievement of
group consensus). This latter result emphasizes that the consensual
outcome generated through the give-and-take of social interaction can
transcend what would be expected only from knowledge of the average
individual contribution. It must be noted, however, that the results of
the study only partially support the existence of an assembly effect
570 HUMAN COMMUNICATION RESEARCH / Summer 1984
thus this hypothesis was not confirmed. Perhaps the flow of information
was so constrained by the nominal decision procedure that the quality of
decisions reached by groups in this condition was undermined. More-
over, although the nominal procedure permitted the “exchange” of
information about individual preferences and reasons for these pre-
ferences, it did not allow for the direct criticism of these preferences and
reasons. As Davis (1969) suggests, it may be the debating and critical
evaluation that occurs in group discussion that accounts for the
superiority of group decisions.
Previous research has found that group discussion exerts a positive
influence on the quality of individual decisions made subsequent t o
interaction. Thus in the present study, it was hypothesized that subjects
in the interacting condition would make better individual decisions
subsequent t o interaction than subjects in either the staticized or
nominal conditions. This hypothesis was confirmed, Social interaction
thus not only improves the quality of decisions made by groups, it also
benefits individuals in their private decision-making.
Obviously, the benefits accruing from social interaction are not
generalizable t o all tasks. As noted above, social interaction among
group members appear t o foster high quality group decisions when the
task facing the group is relatively complex, requires a considerable
amount of information processing, raises value questions, or needs to be
approached in multiple stages (see Hackman & Morris, 1975; Sorenson,
1971). But as Fisher (1980) notes, it is precisely such tasks on which
groups are most likely to work. Thus while social interaction may not
result in better decisions for all tasks, it is likely that social interaction
will result in higher quality decisions on “everyday” group decision-
making tasks.
Although the present findings demonstrate that the mean decision
quality achieved by interacting groups exceeds that of noninteracting
groups, it should not be assumed that interaction inevitably will result in
better decisions. In the present study, two of the ten groups in the
interacting condition produced consensual decisions lower in quality
than “decisions” created by averaging initial individual preferences.
This suggests that ineffective interaction may result in relatively poor
decisions. Clearly, research needs to examine the interaction patterns in
groups that lead to better and worse decisions. Although some research
has attempted t o document differences in the patterns of interaction for
effective and ineffective groups, much of this research has employed
interaction coding systems that d o not permit examination of relevant
572 HUMAN COMMUNICATION RESEARCH / Summer 1984
REFERENCES