Can encoding differences explain the benefits of directed forgetting in the list method paradigm?

Abstract

We propose that the benefits of directed forgetting are explained by the differences in recall arising from individual strategy choices used to encode List 2. In Experiment 1, inducing participants to encode both lists using the same strategy (either shallow or deep) led to significant costs of directed forgetting but abolished the benefits. In Experiment 2, inducing a shallow encoding on List 1 and a deep encoding on List 2 produced similar results, abolishing the benefits but not the costs. Reanalysis of Sahakyan and Kelley’s (in press) Experiment 2 showed that the costs of directed forgetting could be detected irrespective of participants’ strategy choices. However, the benefits of directed forgetting are best explained by a more frequent use of deeper encoding of the second list by the forget group participants.

Full text

Page 1

Can encoding differences explain the benefits of directed forgetting in the list method paradigm?

By: Lili Sahakyan and Peter F. Delaney

Sahakyan, L., & Delaney, P. F. (2003). Can encoding differences explain the benefits of directed forgetting in
the list-method paradigm? Journal of Memory and Language, 48(1), 195-201.

Made available courtesy of Elsevier: http://www.elsevier.com/

*** Note: Figures may be missing from this format of the document

Abstract:
We propose that the benefits of directed forgetting are explained by the differences in recall arising from
individual strategy choices used to encode List 2. In Experiment 1, inducing participants to encode both lists
using the same strategy (either shallow or deep) led to significant costs of directed forgetting but abolished the
benefits. In Experiment 2, inducing a shallow encoding on List 1 and a deep encoding on List 2 produced
similar results, abolishing the benefits but not the costs. Reanalysis of Sahakyan and Kelley's (in press)
Experiment 2 showed that the costs of directed forgetting could be detected irrespective of participants' strategy
choices. However, the benefits of directed forgetting are best explained by a more frequent use of deeper
encoding of the second list by the forget group participants.

Article:
In this paper, we examine the phenomenon called directed forgetting in light of the encoding strategies
employed by participants while performing the task. The directed forgetting paradigm invented by Bjork, La-
Berge, and LeGrand (1968) is thought to have implications for understanding the mechanisms that are typically
involved in updating information in long-term memory. Others have shown that directed forgetting may have
implications for understanding the ways in which people segregate information according to particular mental
contexts (Sahakyan & Kelley, in press). In this paper, we will mainly focus on a particular version of the
paradigm called the list method of directed forgetting. A typical list method directed forgetting study involves
studying two lists of words followed by a recall test on both lists. The crucial manipulation involves inserting an
instruction to forget the first list before the study of the second list for only one group of participants the (forget
group), while the other group is told to continue remembering the first list of words (the remember group). The
second list is always followed by an instruction to remember the items. The directed forgetting effect is
typically reflected in (a) poorer memory for List 1 in the forget group compared to the remember group—a find-
ing called the costs of directed forgetting, and (b) better memory for List 2 in the forget group compared to the
remember group—called the benefits of directed forgetting (following Liu, Bjork, & Wickens, 1999). For a re-
view of directed forgetting studies and indices through which it is operationalized, see MacLeod (1998).

Theoretical explanations of directed forgetting
The dominant theory of directed forgetting has been the retrieval inhibition hypothesis (Bjork, 1989). The forget
instruction was said to invoke a process that at retrieval blocked or suppressed the access to the List 1 items,
producing the costs. The benefits of directed forgetting have been attributed to the forget group escaping from
proactive interference (PI) because inhibited List 1 items would not cause proactive interference.
The primary evidence for this comes from studies comparing performance in the forget group with performance
in a group that studies only a single list. Because the level of performance in the forget group and the no
proactive interference group did not statistically differ, the forget group was said to escape from proactive
interference (Bjork & Bjork, 1996; Bjork & Woodward, 1973).

Recently, Sahakyan and Kelley (in press) proposed an alternative explanation for the directed forgetting
phenomenon based on contextual change. Their theory proposes that the costs and benefits of directed forget-
ting result from an internal context change that occurs between the presentation of the two lists in response to
the forget instruction. They suggested that one way to intentionally forget a list of items would be to attempt to

Page 2

create a larger than normal change of context between the lists. This could happen, for example, if the forget
cue led people to deliberately think of something other than the experiment. Participants in the forget group
would then be more likely to treat the two lists as separate events and not maintain the context in which they
initially encoded the first list when they are encoding the second list. Instead, they would generate entirely new
contextual elements with which they encode the second list. The mismatch between the study context and the
testing context would produce the pattern of costs and benefits typically found in the directed forgetting studies.
The costs occur because the mental context at test mismatches the mental context in which List 1 was encoded.
The benefits are observed because proactive interference is reduced in the forget group as a result of encoding
the lists in separate contexts. Consistent with their hypothesis, experiments where a context change was deliber-
ately induced between the study of List 1 and List 2 in the absence of an instruction to forget mimicked the
pattern of List 1 and List 2 recall typical of participants in the forget group. In addition, reinstating the List 1
learning context at the time of the test reduced the costs and benefits associated with the instruction to forget,
suggesting further that context change provides a viable explanation for the effects of the forget instruction.

Encoding strategies in directed forgetting
The role of encoding strategies has been largely overlooked in directed forgetting studies. Both existing theories
predict a homogeneous effect of the forget instruction regardless of encoding strategy. Thus, our interest in
encoding strategies was motivated by two theoretical concerns. First, we wondered if the directed forgetting
effects would turn out to be contingent on the particular encoding strategy employed by a participant. For
example, shallow encoding strategies like rote rehearsal might lead to larger directed forgetting effects than
deeper encoding strategies like making up a story. This could happen because elaborative encoding provides a
greater number of retrieval routes or easily generated retrieval cues that are useful for free recall (Einstein &
Hunt, 1980; Masson & McDaniel, 1981) than maintenance rehearsal does. A recently conducted meta-analysis
of context-dependent memory studies revealed that the type of encoding task at input is a significant modulator
of the magnitude of the context effects for direct tests of memory such as recall (Smith & Vela, 2001). In
particular, they found that non-associative processing led to larger context-dependent memory effects than did
the associative inter-item processing of information. The authors speculated that forming inter-item associations
at learning could decrease the encoding of contextual information, because encouraging people to concentrate
more on the study items could lead to the exclusion of attention to the environment, and consequently the
changes in context will have smaller or no effects on memory. Smith and Vela refer to this as an overshadowing
hypothesis. Similarly they argued that using associative retrieval cues during retrieval could outshine or
overpower weaker contextual cues leading again to smaller context-dependent memory effects. This is known
as the outshining hypothesis. Overall, the context-dependent memory effects seem to be smaller when the use of
non-contextual cues either at learning or at test increases. Therefore, it is important to explore if the directed
forgetting effects, which Sahakyan and Kelley (in press) argued were similar to context-dependent memory
effects, turn out to be contingent on the particular encoding strategy employed by a participant.

Second, we were aware that by considering only aggregate group-level data, one might postulate a mechanism
based on group data when in fact no individual displays that pattern of performance (e.g., Delaney, Reder,
Staszewski, & Ritter, 1998). For example, directed forgetting benefits are said to be due to the reduction or
escape from proactive interference in the forget group (Bjork & Bjork, 1996; Bjork & Woodward, 1973;
Sahakyan & Kelley, in press). However, benefits might actually be due to a combination of shallow and deep
encoding strategies that produce a recall level that is comparable to the recall of a no-proactive interference
group—even though each individual member of the forget group suffers from proactive interference.

We first sought to explore the natural distribution of strategies in the directed forgetting task. We examined a
subset of the data collected by Sahakyan and Kelley (in press, Experiment 2) in which retrospective verbal re-
ports were collected immediately following recall. The subset includes 96 participants from the control condi-
tion of the context not-reinstated group from their experiment who were equally divided into the three
conditions with 32 participants in each group. We omit the details of the study that do not bear directly on our
current purposes; however, we briefly report the methods of the experiment. Participants studied two lists of 15
words at a rate of 5 s per word. Then, one-third of the participants received an instruction to forget the first list

Page 3

(the standard forget group), the remaining participants received an instruction to keep remembering the first list
of items (the remember group). However, half of the remember group participants upon the completion of study
of List 1 were given 40 s to imagine their childhood home, mentally walk through it, and describe it to the
experimenter—a task designed to create a context change. After the study of List 2, all groups were tested on
memory for both lists following a 90 s filler task. Recall was carried out on separate sheets of paper and List 1
was always tested before List 2.

The retrospective reports were elicited using the standard instructions from Ericsson and Simon (1993), and
involved asking participants to remember as much as they could about what they were thinking from the time
they saw the first word on the first list. This technique has been shown to provide reliable information about
processes that are too fast to report concurrently but that are potentially reportable (see Ericsson & Simon,
1993, for a review). Participants reported using a variety of strategies while memorizing the words. Because of
variability in individual strategy choices and the uneven distribution of observations per cell, verbal reports
were coded into three main categories (a similar approach to classifying verbal reports was taken by Perfect &
Dasgupta, 1997). Maintenance rehearsal of each individual item (usually accompanied by increasing the
number of items rehearsed together each time a new word was encountered), rhyming, and rehearsing only the
first letter of each word comprised the shallow encoding strategies. Creating individual sentences with the
words, using self-referencing, creating a story using all the items on the list, and picturing a scene that contained
all the items as interactive images were considered to be deep encoding strategies. Finally, visualizing
individual items or using a mixture of shallow and deep strategies on the same list were considered intermediate
encoding strategies.

When the verbal reports were coded into categories, it became apparent that the majority of participants (79%)
started out with one of the shallow encoding strategies. More interesting was the finding that participants
occasionally changed their encoding strategies between lists. In most cases, the strategy change involved using
a deeper, more elaborate way of encoding the List 2 items than the List 1 items. Despite the rather similar
distribution of encoding strategies on List 1 in all three conditions, the participants in the forget and remember
plus context change groups tended to switch to a deeper encoding strategy on List 2 more often than the
participants in the remember group did (38% of participants changed to a deeper encoding in the forget group,
44% in the remember plus context change group, and 22% in the remember group). Given the variability in
encoding strategies and the asymmetric rates of deep strategy use in different groups, we wondered whether a
particular combination of shallow and deep strategies could be responsible for the directed forgetting benefits.

In this paper, we report two experiments that address the role of encoding strategies in directed forgetting. Our
experiments directly controlled encoding strategy to assess the possibility of different directed forgetting effect
sizes connected with different strategies (Experiment 1), and the possibility that the observed differences in the
rate of strategy usage across conditions are responsible for the directed forgetting benefits (Experiment 2). We
will suggest that while contextual change provides the best explanation for the costs of directed forgetting, the
benefits appear to be due to some participants changing from a shallow to a deep encoding strategy. The change
appears to be initiated in response to a change in mental context.

Experiment 1
To determine the relationship between the encoding processes and the directed forgetting effect, the encoding
strategy on both lists was controlled in Experiment I. In particular, participants were told to use shallow and
deep encoding strategies most commonly reported in verbal reports, namely rehearsing the words and creating a
story with all the items. In retrospective verbal reports, most participants indicated that they engaged in
maintenance rehearsal and added each new word into the rehearsal cycle—so this was the instruction for the
shallow condition of the experiment. Other participants reported using more elaborate strategies like making up
a story using all of the items on the list—so this was the instruction for the deep condition of this study. There
were three types of instructions used in this study: the standard forget and remember conditions, and the re-
member plus context change condition. The latter group was included to provide a further test of the context
change hypothesis. In addition, no -proactive interference conditions were also included in the design as

Page 4

reference groups. The latter groups received only the second list to study and were engaged in irrelevant activity
while the other groups studied the first list. In previous studies, the performance on List 2 in the forget group
matched that of the no-proactive interference group, which suggested that the benefits of directed forgetting
could be attributed to the escape from proactive interference in the former group.

Method
Participants
The participants were 192 Florida State University undergraduates who participated to fulfill a course
requirement. There were 24 participants in each of the six experimental conditions. In addition, 48 participants
were added after the completion of the design. Half of them were tested in the shallow and half in the deep no-
proactive interference conditions.

Materials
Thirty unrelated English nouns of medium frequency were drawn from the Kucera and Francis (1967) norms.
Two lists of 15 words each were prepared that were followed either by the forget or remember cue resulting in
four possible combinations for counterbalancing.

Design
The design of the study was an encoding strategy (shallow vs. deep) by instruction type (forget, remember, and
remember plus context change) between-subjects design.

Procedure
Participants were tested individually. The procedure followed the list method of directed forgetting. The first
list of 15 words was presented at a rate of 5 s per word on computer screens. Half of the participants in each
experimental condition were instructed to learn the items using a shallow encoding strategy. Specifically, they
were instructed to repeat the words aloud several times, as they appeared, and increase the number of words
they repeated aloud each time by incorporating every new word that appeared on the screen into the rehearsal
cycle. They were told that if they forgot some words, then they should continue repeating the words that they
remembered without regard for order. The remaining half of the participants learned the items using a deeper
encoding strategy. Specifically, they were instructed to make up a story using all the items on the list starting
with the first word that appeared on the screen. They were told to tell the story aloud to ensure compliance with
the instructions. The participants were warned that their stories could be bizarre and incoherent, but that they
should nevertheless attempt to integrate each new word into the story.

Immediately after studying the first list, some participants were told to forget the items as they were only for
practice and to familiarize them with the task. The other participants were told to keep remembering the items,
as they were only the first half of the study list. The participants in the remember plus context change group
after receiving the remember instruction were given 45 s to imagine their childhood home and describe it aloud
to the experimenter by mentally walking through it. The participants in the forget and remember groups waited
for an equivalent time period before beginning the study of the second list. The second list was then presented
with the same encoding instruction that accompanied the first list encoding (either shallow or deep). The no-
proactive interference groups solved arithmetic problems throughout the time that the other groups studied the
first list. They only studied the second list with either shallow or deep instructions. The study was followed by
90 s of solving arithmetic problems for all groups before the free recall test. At test, participants were given two
minutes and asked to recall the first list, followed by the second list. The recall was carried out on separate
sheets of paper. The no-proactive interference groups were tested on recall of List 2 only.

Results and discussion
For all analyses, α was set at α = .05.

As in previous studies, the costs and benefits of directed forgetting were analyzed separately. For the analyses
of costs, we performed an ANOVA on proportion of correct List 1 recall by condition (forget, remember, and

Page 5

remember plus context change) and strategy (shallow, deep). We found a significant main effect of condition,
F(2, 138) = 9.26, MSE = .023, and a significant main effect of strategy F(1, 138) = 145.99, MSE = .023. The
interaction of condition and strategy was not significant, F < 1 (see Fig. 1). Post hoc tests confirmed by Tukey's
HSD showed that the remember group recalled significantly more items than did the forget and remember plus
context change groups, which did not differ from each other. This was true in both strategy groups although the
deep encoding resulted in higher levels of recall overall. Thus, the costs of directed forgetting were obtained
regardless of the quality of encoding.

For the analysis of benefits, we first performed an ANOVA on proportion List 2 recall by condition and
strategy, excluding the no-proactive interference groups. There was only a significant main effect of strategy,
F(1, 138) = 178.10, MSE = .018, confirming again that the deep encoding resulted in higher levels of recall. The
main effect of condition and the interaction of strategy and condition were not significant, F's < 1. To sum up,
the analyses showed that the benefits were eliminated in both strategy groups.

The absence of benefits can mean two things: either the remember group escaped proactive interference and
therefore performed as well as the no-proactive interference groups, and the remember plus context change and
forget groups, or the remember, forget, and remember plus context change groups all suffered proactive
interference and could not escape it. To determine which was the case, we included both shallow and deep no-
proactive interference groups in the analyses of benefits. The ANOVA on proportion List 2 recall by strategy
and condition including the no-proactive interference groups revealed a significant main effect of encoding
strategy, F(1, 184) = 230.82, MSE = .330, and a significant main effect of condition, F(3, 184) = 19.86, MSE =
.330. The interaction was not significant, F < 1 (see Fig. 2). Post hoc tests using Tukey's HSD showed that the
no-proactive interference groups recalled significantly more items than did the forget, remember, and remember
plus context change groups. The latter three groups did not differ from one another.

Overall, the results showed that the benefits were abolished when the encoding strategy was controlled. The
absence of the benefits in this study was due to all conditions suffering from proactive interference regardless of
the encoding quality. However, the costs were present regardless of the particular encoding strategy participants
were instructed to use. Thus, it appears that the type of encoding did not modulate the size of the directed
forgetting costs, because they were detected in both types of learning strategies. Smith and Vela's (2001) meta-
analysis revealed several important factors that contribute to the context-dependent memory effects apart from
the effects of the quality of encoding. Some of these variables include the study-test interval, the type of
paradigm used in a study, as well as experimenter-related factors. However, in this meta-analysis, the authors
identified only significant main effects, which gives no information about possible interactions. Perhaps some
combinations of these factors would decrease the differences in context-dependent memory effects between the
associative and the non-associative types of encoding.