Good explanations are tailored to the questioner and they contrast the
given explanation with alternative explanations (Mills' method of
difference, Hilton & Slugoski, 1986).
Predictions provide beliefs about outcomes (typically future states),
they are sometimes verifiable, and they don't require knowledge of a
causal mechanism. Explanations, by contrast, provide understanding,
they are rarely directly verifiable, but they can be evaluated as
relatively convincing and clear. Predictions are often based on
explanations; conversely, an explanation can be tested by testing a
prediction derived from it. Explanations are in some sense easier to
produce (in hindsight, by fitting ideas to the data) but harder to
prove--prototypical examples of this asymmetry can be found in
psychoanalysis and evolutionary psychology.
Classical conditioning. Humans are extremely good at
representing co-occurrences of events. This conditioning process is
the core of associative learning and memory. But this ability makes
people somewhat biased in the detection of true correlations. The
bias is most crucial when the base rates of the events in question is
extreme and/or asymmetric. A pattern of data like the one below leads
to strong associative bonds for the co-occurrence of A and B, even
though A and B are not correlated with each other (they are just both
frequently present).
Plausibility. As discussed under the
representativeness heuristic, the plausibility or stereotypical fit of
an explanation is often mistaken as evidence for its truth. Examples
of the power of plausibility abound in the literature (e.g., Rorschach
ink blots, Draw-a-man test), and the most recent false accusations of
Islamic groups after the Oklahoma City bombing attack add a sad one to
the list. Plausibility can arise from semantic or conditioned
associations, from persuasive arguments, emotional appeals, vivid
experiences, wishful thinking, etc.
Normality. Kahneman and Miller (1986) propose a theory of
norms that describes how perceptions of objects and events recruit
"norms"--i.e., aggregates of past experiences, knowledge, and
expectations, which provide standards to judge the object or event at
hand.
Norms are often standards of comparison. Which standard you have in
mind when making a judgment is crucial: "I am assertive" can mean
"assertive relative to other people" or "assertive in most situations
I choose to enter." McGill (1989) showed that people's explanations
for a simple decision such as choosing psychology as one's major
differ depending on the standard of comparison (or norm) that is being
recruited. Compare "Why did you choose to major in
psychology?" with "Why did you choose to major in
psychology?"
In general, explanations of events are likely to focus on "mutable
elements" of the event--i.e., aspects that can easily be mentally
undone ("if only..."). This process of undoing is called
"counter-factual reasoning." Elements are more likely to be mutable
if they are (a) exceptions to a rule or (b) deviations from an ideal.
Emotions such as regret or anger also focus more often on mutable
events.
The framework of norms can be fruitfully applied to better understand
several phenomena in judgment and decision making, such as anchoring
and adjustment, non-regressive prediction, belief perseveration,
base-rate neglect, determinants of regret, and the conjunction
fallacy.
Hilton (1990) points out that discrepancies from norms evoke a
why question that contains the norm as a contrast cases (e.g.,
"Why X rather than non-X?", "Why X rather than Y?"). Explanations
themselves, understood as answers to questions and thus most often
uttered in conversation, must fill the questioner's knowledge gap in
light of those norms. Hilton describes in detail the communicative
rules (à la Grice) that speakers follow when providing
explanations, and he applies his conversational framework to several
past research findings (just like Schwarz, 1994, did in a paper
assigned last week).
Kelley (1967, 1972) formulated more extensive rules in his "ANOVA
model." People should explain a given behavior by reference to other
pieces of information about other people, other stimuli, and other
contexts. The pattern of covariation across these three factors
provides the normative basis for an explanation. Topics studied in
the wake of this model include the fundamental attribution error, the
actor-observer asymmetry, salience effects, ego-biases, and
attributional styles in relationships, depression, and achievement
situations.
As a model of how people do in fact explain behavior, Kelley's model
has major weaknesses, among them the sole focus on an individual
person's reasoning (rather than the communicative context of
explanations) and the neglect of people's own concept of human
behavior. The following two descriptive models try to correct these
weaknesses.
A descriptive model of behavior explanations. One of the major
weaknesses of classic attribution theory is its assumption that people
explain all behavior alike--namely by way of person causes and
situation causes. But this person-situation dichotomy of causal
explanations fits well only for unintentional behaviors; intentional
behaviors, by contrast, are explained by reason explanations, which
assume the actor's conscious choice in behaving that particular way
and cite the actor's reasons for that choice. Data from my own
research show that
1. What is an explanation?
An explanation is the answer to a why X? question (where X
is some event or state of affairs); it answers this question by
providing clarifying information in the form of facts, processes,
mechanisms, or principles in light of which X becomes
understandable to the questioner. 2. Bases of explanations
Explanations are given using a variety of constructs: theories,
models, stereotypes, dispositions, processes/mechanisms, event chains
(scripts). Where do these constructs come from? Event A
Many problems in human judgment discussed in previous lectures are
based on or related to this "trap of conditioning"--availability,
neglect of base rates, neglect of Bayes' theorem, SDT biases. But the
recognition of correlation /covariation can be very helpful for
finding a causal explanation for an event: Kelley's ANOVA model of
attribution is built on the assumption that people look for
covariation information (see Plous).
Event B present absent
present 16 4
absent 4 1
3. The Study of Behavior Explanation
Normative attribution models. Jones & Davis (1965)
analyzed rules that people use (or should use) when inferring
dispositions from other people's behavior (so called "correspondent
inferences"). Among these rules are the principle of noncommon
effects (to infer a person's disposition, you must look at those
aspects of the chosen behavior that are different from other possible
behaviors) and the principle of desirable effects (when inferring a
person's disposition, you can be more confident if the person behaves
in ways that do not bring about generally desirable effects).