when there is misleading information that fools people into thinking that they
have accurately reconstructed an event. In these latter two situations, people
may be as confident in the accuracy of an incorrectly reconstructed event as
they are of a correctly reconstructed event.
Consistent with the predictions of constructionist theory, a variety of experi-
ments have demonstrated that the correlation between confidence and accuracy
is typically quite low, especially in situations where eyewitnesses to crimes and
accidents must recollect details of those crimes and accidents (Wells & Murray,
1984; Donders, Schooler, & Loftus, 1987; Smith, Kassin, & Ellsworth, 1989).
Presumably most people have not had much practice developing learning and
reconstruction strategies for eyewitness information, and therefore have not
learned when such strategies produce accurate recollections (see Perfect, Wat-
son, & Wagstaff, 1993).
On the other hand, the correlation between confidence and accuracy is reli-
ably higher in situations where people have had such practice. For example, the
correlation between confidence and accuracy is moderately high when subjects
are asked to answer general knowledge questions, such as ‘‘Who wroteThe Mill
on the Floss?’’ (e.g., Hart, 1967; Perfect et al., 1993; see Nelson, 1988, for a re-
view). Presumably most people have learned how good they are at answering
general knowledge questions (Perfect et al., 1993). The correlation between
confidence and accuracy is also moderately high when subjects are asked to
answer questions about short texts they have recently read (e.g., Stephenson,
1984; Stephenson, Clark, & Wade, 1986). In this case, experience in academic
settings has presumably taught most people how good they are at answering
questions about texts.
Record-keeping theories of memory would predict that any variable that
decreases memory accuracy should also decrease confidence in the accuracy of
the memory. Contrary to the record-keeping prediction, Chandler (1994) re-
ported a series of studies in which accuracy was decreased but confidence
increased. Chandler had subjects study nature pictures, such as pictures of
lakes. Later, subjects were required to determine which of two related pictures
(e.g., two different lakes) had been previously displayed and to indicate their
confidence in their recognition judgment. When the subjects had also studied a
third related picture (e.g., a third lake), their recognition performance declined
but their confidence in their selection increased (compared to the case when
there was no third picture). The constructionist explanation is that subjects be-
come more familiar with the general theme (e.g., scenic lakes) of the pictures as
they study more of the related pictures. Both alternatives on the recognition test
fit the theme, making discrimination between them difficult, so recognition
memory performance declines. But because the selected picture fits the theme,
confidence in the selection is high.
Also consistent with constructionist theory is the finding that people become
confident of inaccurate recollections when those recollections are reconstructed
from misleading information supplied to them by an experimenter (e.g., Davis
& Schiffman, 1985; Spiro, 1977). Consider a study by Ryan and Geiselman
(1991). They presented subjects a film of a robbery and a week later had them
read a summary description of the film. For some of the subjects the summary
included a misleading detail, such as ‘‘The police car is at a brown house’’ (in
340 R. Kim Guenther