Marsh, E. J., Meade, M. L., & Roediger, H. L.
(2003). Learning Facts from
Fiction. Journal of Memory and Language, 49, 519-536.
Learning occurs in many
contexts. In addition to clearly educational
sources, we also learn from exposure to sources that contain information about
the world. Although we learn from them,
these fictional sources may not be as accurate as non-fiction sources. Research on learning from fiction focuses on
whether students use fictional sources to answer general knowledge questions
and how aware they are of reliance on fiction.
Integration
of information refers to linking new facts to pre-existing world knowledge; the
strongest form eliminates links to the fictional source. Compartmentalization
refers to keeping fictional knowledge separate from the rest of world
knowledge. A hybrid condition in which facts remain linked to their sources, but
are also associated with previous world knowledge has also been suggested. There has been evidence for
integration. A single reading of
correct and incorrect answers in a word list increased production of these
answers in later questions (Blaxton, 1989, Kelley & Lindsay, 1993, Thapar
& Rouder, 2001). Studying fantasy
(incorrect) facts slowed answers to true facts (Peterson & Potts, 1982).
Although there is evidence
that integration occurs, there are limits on integration. In eyewitness studies, once errors are
noticed, the suggestibility of the entire story is reduced (Loftus, 1979). There is less interference from facts that
contradict the state of the world (Gerrrig & Prentice, 1991).
The type of test matters,
which suggests compartmentalization.
Subjects verify true facts faster when the test does not contain any of
the incorrect study facts (Lewis & Anderson, 1976). Retaining the link to the story source is
more likely when studied facts seem contradictory or suspicious.
Marsh et al had two goals of
their research:
If the story source is
forgotten and the facts are related to world knowledge, integration would be supported. Compartmentalization would be supported if
the source is remembered and there is no increase in world knowledge
attributions. The hybrid position would
be supported if there is an increase in world knowledge attributions and the
source remains in memory.
Experiment 1
Participants were given five
minutes each to read nine stories. Each
story contained eight facts from general knowledge norms (Nelson & Narens,
1980). Four facts were deemed easy
(high prior knowledge) and four were defined as hard (low prior
knowledge). Half were framed Neutrally
(provided a reference, not an answer to a future general knowledge question),
and half were framed Correctly (provided an answer to the future question). After reading the stories, participants
answered four questions about each story.
They then were given a distractor task, and finally took a general
knowledge test which contained 72 critical questions and 72 filler
questions. At least four questions
separated questions related to the same story.
Participants were instructed to avoid guessing. In one condition, participants decided
whether their answer to the questions came from the story immediately after
answering each question. In the other
condition, participants answered all questions, then indicated whether their
answer came from the story or from prior world knowledge.
Participants correctly
answered more questions in the Correct frame than in the Neutral frame. They were aware of when they had actually
read the answers in the story, but their answers were increasingly attributed
to prior knowledge. This indicates that
the story source was retained, but facts were also integrated with previous
knowledge.
Experiment 2
Experiment two used
misleading information in the stories.
The researchers also manipulated how many times participants read the
story (0, 1, or 2). There were six
critical facts (3 easy, 3 hard) and were framed in Neutral, Correct, or
Misleading frames. Misleading frames
suggested incorrect answers. The
procedure was the same as in Experiment one, except that participants were also
asked questions that examined their awareness of the experiment’s purpose. They were asked to list the items of
misinformation that were presented, and rate how surprised they were at the corrected
answers.
The way the information was
framed affected ability to answer easy questions more than hard questions. The number of times they read the story had
a bigger effect on hard questions.
Reading the facts aided performance for easy and hard questions. Reading misinformation decreased ability to
answer easy questions. Hard questions
werre approaching floor, so it was difficult to see a decrease. Participants produced misinformation
slightly more for hard questions than for easy ones. They realized that facts were in the stories, but they attributed
answers to prior knowledge, even when the fact was in a Misleading frame. They also remembered reading answers that
were not presented in the stories in both the Neutral and Misleading frames; even
if they read misinformation, they thought the correct answer was in the
story. They were more surprised by
correct answers to difficult than easy questions. Most knew there was misinformation, but they could recall less
than one specific instance.
Experiment 3
Experiment three involved two
sessions that were separated by a week.
There were four critical facts (two easy, two hard, two Correctly
framed, two framed Misleadingly). A new
general knowledge test was used for the first session (84 items, 36 critical
items). The procedure for the first
session was the same as in Experiment two, except no source judgments were
made. When participants returned after
a week, they completed a filler task and then took a different general
knowledge test. They answered all
questions, then went back and made source judgments. They completed the survey about knowledge of misinformation and
were questioned whether they researched any questions during the delay.
Effects of story exposure
were weakened by the delay.
Participants could more easily answer easy questions after reading the
story once or twice, but this effect disappeared on hard questions. More misinformation was produced on the
immediate test, and they answered more questions correctly when they had not
read the misinformation. Misinformation
continued to be produced after the delay, but only when those items had been
previously tested. The largest effects
of story exposure were greater on the immediate test than on the delayed test
if the items had been previously tested.
If the items were not previously tested, participants did not produce
misinformation and story reading only benefited easy questions. Participants continued to be aware of the
story link after delay, and misinformation continued to be attributed to prior
knowledge. They were more surprised by
correct answers to the questions they read misinformation for. 81% recognized misinformation, but they
could only recall 1.2 instances.
General Discussion
More questions were answered
correctly when answers were read in the stories, and misinformation decreased
correct answers. Participants were
aware that their answers were in the stories, but answers were attributed to
prior knowledge. Subjects believed
their incorrect misinformation answers.
Reading the story led to an illusion of truth. They produced misinformation and then attributed it to prior
knowledge, which is similar to the knew it all along bias (Fischoff, 1977;
Wood, 1978).
Misinformation did not discredit
the entire story. Subjects were not
less likely to use facts from the story evern when they noticed
misinformation. The procedure was
similar to previous eyewitness studies like those of Loftus (1978), but there
was no “spillover effect” in this study.
Marsh et al. suggest this occurred because when answering the general
knowledge test, answers may be based on whatever comes to mind instead of
trying to remember the specific study phase.
This data supports the hybrid
view of fact reception. Integration is
represented in that participants were willing to use story facts (even
incorrect ones) to answer questions.
They believed that they had known these facts before they read them. Participants were able to remember that
facts were presented in the studies, even after a delay. This suggests compartmentalization.
Further research could
examine when and why subjects do or do not selectively use fiction for fact.