ATTRIBUTION THEORYin the development of more systematic statements
about attributional processes.
Covariational model. Harold Kelley’s (1967,
1973) covariational model of attribution addresses
the question of whether a given behavior is caused
by an actor or, alternatively, by an environmental
stimulus with which the actor engages. According
to this model, the attribution of cause is based on
three types of information: consensus, distinctiveness,
and consistency. Consensus refers to the similarity
between the actor’s behavior and the behavior of
other people in similar circumstances. Distinctiveness
refers to the generality of the actor’s behavior:
Does she or he behave in this way toward stimuli in
general, or is the behavior specific to this stimulus?
Consistency refers to the actor’s behavior toward
this stimulus across time and modality. There are
many possible combinations of these three types
of information, but Kelley makes explicit predic-
tions about just three. The combination of high
consensus, high distinctiveness, and high consis-
tency supports an attribution to the environmen-
tal stimulus, whereas a profile of low consensus,
low distinctiveness, and high consistency supports
an attribution to the actor. When the behavior is
inconsistent, regardless of the level of consensus
or distinctiveness, an attribution to circumstances
is predicted.
Empirical tests of Kelley’s model have focused
either on the effects of a particular type of infor-
mation or, more in keeping with his formulation,
on the effects of particular patterns of information
(McArthur 1972). In an innovative analysis, Miles
Hewstone and Jos Jaspars (1987) proposed a dif-
ferent logic (although one consistent with Kelley’s
model), suggesting that potential attributers con-
sider whether different causal loci are necessary
and sufficient conditions for the occurrence of an
effect. They conclude that the notion of causality is
flexible and thus assert that there may be some
advantage to conceiving of situation-specific no-
tions of causality. More recent work has examined
the universality and external validity of Kelley’s
model. Irina Anderson and Geoffrey Beattie (1998),
for example, analyzed actual conversations be-
tween men and women talking about rape. They
found that men tended to use the reasoning out-
lined in Kelley’s model by making reference to
consensus, distinctiveness, and consistency, as well
as by using these types of information to formulate
attributions for behavior. Women, on the other
hand, made less use of these variables and intro-
duced the variable of ‘‘foreseeability’’ into their
analyses.Correspondent inference. The theory of cor-
respondent inference (Jones and Davis 1965; Jones
and McGillis 1976) addresses the attribution of
personality traits to actors on the basis of their
behavior and focuses on attributions about per-
sons in greater depth than does Kelley’s covariational
model. These two theories thus address different
questions. Kelley asks: When do we attribute an
event to an actor or to some stimulus in the
environment? Edward Jones asks: When do we
attribute a trait to an actor on the basis of her or his
behavior? The theory of correspondent inference
focuses more narrowly on the actor but also yields
more information about the actor in that it speci-
fies what it is about the actor that caused the
behavior. Jones and his coauthors predict that two
factors guide attributions: (1) the attributer’s prior
expectancies for behavior, specifically, expectancies
based either on knowledge of earlier behaviors of
the actor (target-based) or on the actor’s social
category memberships (category-based), and (2)
the profile of effects that follow from the behavior-
al choices available to the actor.Edward Jones and Daniel McGillis propose
that expectancies determine the degree of confi-
dence with which a particular trait is attributed;
the lower the expectancy of behavior, the more
confident the attribution. The profile of effects
helps the attributer identify what trait might have
produced the behavior in question. Noncommon
effects—effects that follow from only one of the
behavioral options—provide information about
the particular disposition. The fewer the noncommon
effects, the clearer the attribution. Thus, behavior
that contradicts prior expectancies and a profile of
behavioral choices with few noncommon effects
combine to maximize the possibility of attributing
a disposition to the actor (a correspondent infer-
ence). Empirical research generally has supported
these predictions.These two models share some attributional
principles. Expectancy variables (target-based and
category-based) are analogous to Kelley’s types of
information. Although the predicted effects of
consensus information and its analogue, category-
based expectancies, are compatible, the predicted