Cued Memory Collaboration: Effects of Friendship ...

2 downloads 0 Views 159KB Size Report
group retrieval and that no such discrepancies were found for retrieval with ... towards successful retrieval to a greater extent than non-friends, and ``for.
EUR OPEA N JOUR NA L OF COGNITI VE PSY CHOLOG Y , 1997, 9 ( 3), 273± 287

C ued M emory Collaboration: E€ ects of Friendship and Type of Retrieval C ue Jan A ndersson and Jerker R oÈ nnberg D epartm ent of Education and Psychology, L ink oÈ ping University, L inkoÈ ping, S weden

The e€ ects of friendship and collaboration on memory retrieval were investigated in two experiments. The net negative e€ ect of collaboration in memory retrieval and the e€ ect of f riendship were studied within a cueing framework. The aim of the study was to investigate the reduced cue e€ ectiveness hypothesis. In Experiment 1, dyadic retrieval was compared to that of a nominal group ( a pooled score from two individuals working individually). The task was to encode and retrieve 90 words individually or dyadically. The retrieval phase was supported by cues produced by others or produced by the participants themselves ( individually or dyadically) . In Experiment 2, f riends were compared to non-f riends in a memory retrieval phase. A ll participants retrieved alone in Experiment 2. The results of Experiment 1 revealed that dyadic retrieval su€ ered compared to nominal group retrieval and that no such discrepancies were f ound f or retrieval with cues produced by others. In Experiment 2, f riends could help each other towards successful retrieval to a greater extent than non-f riends, and ``for others’ ’ instructions increased retrieval performance as well. These ® ndings replicate previous studies by A ndersson and R oÈ nnberg (1995, 1996), and advance the understanding of collaborative memory; that is, net negative e€ ects of collaboration can be explained in terms of the reduced cue e€ ectiveness hypothesis.

INTRODUCTION R eliance on episodic and explicit memory in a collaborative retrieval paradigm suggests that communication in dyads does not work with su cient precision to reach optimal memory performance. Dyadic memory productivity does not reach the potential predicted on the basis of pooled individual performances (A ndersson & R oÈ nnberg, 1995, 1996). Hence, R equests for reprints should be addresse d to Jan A ndersson , D epartment of Education and Psychology, LinkoÈ ping University, S-581 83 LinkoÈ ping, Sweden. E-mail: JanA n@ IPP. LiU.SE

€

1997 Psychology Press Ltd

274

ÈNNBERG ANDERSSON AND RO

dyads are not able to use the total memory potential of all members in the group; that is, they cannot cue each other with good e ciency (A ndersson & R oÈ nnberg, 1995; Meudell, Hitch, & Kirby, 1992; Meudell, Hitch, & Boyle, 1995). This net negative e€ ect of collaboration has been found for episodic and explicit memory retrieval of words, stories, taped lectures and pictures (A ndersson & R oÈ nnberg, 1995, 1996). However, the same performance decrement has not been shown for semantic and implicit retrieval (A ndersson & R oÈ nnberg, 1996). The net negative e€ ect of collaboration for episodic and explicit memory material was assumed to be a consequence of reduced cue e€ ectiveness. The overall aim of the present study was to investigate further this cognitive hypothesis and suggest how it can explain social in¯ uences on memory performance (i.e. in the collaborative context). Dyadic generation of retrieval cues is presumably less individual and less personal to both members in the dyad. This means that when one of the members in the dyad produces the association ``ketchup’ ’ for the target TOMA TO, the other may think it the perfect association. It is therefore chosen, or the partner might not understand the connection. However, even if all associations produced by the dyad are experienced as perfect by both members, each would have been generated initially by only one member of the dyad. Nor would they have been generated in a social vacuum. MaÈ ntylaÈ and Nilsson (1983) showed that self-generated cues worked better as retrieval cues than cues produced by others. A ll this suggests that the dyadic cue-generating (encoding) phase a€ ects the speci® city of cueing and therefore reduces the e€ ect of self-generated cueing (cf . MaÈ ntylaÈ , 1986; Wiseman & Tulving, 1976). Similarly, the dyadic retrieval phase would be less successful, since (1) the cue is not speci® c to both and (2) the communication among the members is not su ciently speci® c (cf . A ndersson & R oÈ nnberg, 1995, 1996). The aims of this study were to investigate the e€ ectiveness of cues in relation to how it a€ ects the mem ory performance of the collaborating dyad compared to the predicted mem ory potential for the sum of all members in the dyad Ð that is, the nominal group (Experiment 1) Ð and also how communication between friends a€ ects the probability of successful retrieval (Experiment 2; cf . A ndersson & R oÈ nnberg, 1995, 1996). Hence, the friendship e€ ect shown in earlier studies (i.e. friends did not su€ er in episodic retrieval tasks compared to the predicted potential as non-friends did) were couched in terms of increased cue e€ ectiveness. The speci® c aims were to (1) explicitly test if the dyad su€ ers compared to the nominal group in cued recall, (2) investigate if the loss is due to type of memory cue (semantic vs episodic) , (3) study if friends can cue one another more e ciently than non-friends, and (4) analyse how ``for

MEMORY, COLLABORATION AND FRIENDSHIP

275

others’ ’ and ``for themselves’ ’ instructions a€ ect the potential superiority of friends’ communication strategies. The predicted potential for the dyads is created from two individuals working individually. Their pooled score is treated as the predicted potential of the dyad (i.e. the nominal group). Di€ erent approaches can be used to trace the predicted potential of the group. R epeated designs where participants recall twice, on their own at ® rst and secondly in dyads, has been used (A ndersson & R oÈ nnberg, 1995; M eudell et al., 1992, 1995), as well as designs comparing one dyad, Persons A + B (working together), with the pooled performance of Persons C and D (working on their own) Ð that is, the nominal group (A ndersson & R oÈ nnberg, 1996). If Person C retrieved three items (a, d, g) and Person D retrieved three items (g, j, l), then the predicted potential or nominal group score would be ® ve items (a, d, g, j, l). The advantages of repeated designs are that it is possible to trace new items (mem ories) create d during collaboration, and to study whether old items (mem ories) were recalled once again or simply forgotten. The comparison of interest for the present paper was the total memory performance of dyads and nominal groups, and not to trace the exact amount of inhibition (i.e. forgotten items) or the exact amount of facilitation (i.e. new memories). The design chosen was to use the dyadic score from Persons A + B working collaboratively and compare that with the pooled score from Persons C and D working on their own (nominal group) . Semantic memory represents our shared general knowledge about the world and episodic mem ory represen ts our individual memories, dated in time and place (Tulving, 1983, 1985). The retrieval of words given self generated associations as retrieval cues to support memory performance would presumably be of a more episodic kind compared with the retrieval of the to-be-remembered words given cues produced by others (cf . Wiseman & Tulving, 1976). The cues produced by others contain semantic information, but no episodic information (i.e. personally encoded) connected to the speci® c target. Therefore, collaborative dyadic episodic retrieval performance should not su€ er to the same extent as that with nominal groups, when the two di€ erent constellations have to use cues produced by others only (cf . A ndersson & R oÈ nnberg, 1996). Friends or couples develop a speci® c and unique way of communicating to share and divide their mental potential to be successful in di€ erent tasks (Fleming & Darley, 1991; Fussell & Krauss, 1989; Wegner, Erber, & R aymond, 1991). Fleming and Darley (1991) suggest that friends (read: senders) use strategic communication processes, and Wegner et al. (1991) suggest that couples develop a transactive memory which represents implicit knowledge of the partner’ s way of thinking.

276

ÈNNBERG ANDERSSON AND RO

Even if the unique communication strategy that friends and couples develop is more or less implicit, it obviously a€ ects the possibilities for maximising memory performance. Friends would, then, by knowing whom the message was intended for (if the receiver is a friend), increase the possibility for successful retrieval compared to if the receiver is a non-friend. The participants who receive cues from a friend will outperform a non-friend receiver if the friends have developed a shared way of communication or an implicit knowledge of each other (cf . Wegner et al., 1991). Previous ® ndings suggest that there should be no di€ erence between friends and non-friends in semantic memory tasks (A ndersson & R oÈ nnberg, 1996). In the present study (Experiment 2), participants were to retrieve/guess the to-be-remem bered word from a ``new item’ ’ word-list (i.e. words not seen in the experimental context). R etrieval with cues generated by others in combination with a ``new item’ ’ word-list ful® ls the criteria of a semantic guessing task. But the combination of the knowledge of the sender (read: friend) and the semantic cue information increases the chances of making a correct guess (reduces possible alternatives), since the semantic cue is likely to activate shared memories (irrespective of their episodic/semantic status). For non-friends, this extra information does not exist. In the studies of Fleming and Darley (1991) and Fussell and Krauss (1989), friends knew that their messages or descriptions were to be used by a friend. ``For others’ ’ and ``for themselves’ ’ instructions were used in the present study to explore to what extent friends with ``for others’ ’ instructions can improve retrieval compared to non-friends and friends with ``for themselves’ ’ instructions. Friends are predicted to outperform non-friends regardless of instruction type, but friends who know that a message is intended for a friend may also perform better than friends that do not know that their message is intended for a friend. Summarising the arguments and predictions, based on previous collaborative retrieval data (A ndersson & R oÈ nnberg, 1995), it was predicted that dyads will su€ er from collaboration compared to the nominal group. Self-generated cues are better than cues produced by others (MaÈ ntylaÈ , 1986; Wiseman & Tulving, 1976), as episodic speci® city is increased (cf . A ndersson & R oÈ nnberg, 1996). When retrieval cues are produced by others, neither dyadic generation nor the dyadic retrieval condition will be negatively a€ ected by collaboration. Based on transactive memory (Wegner et al., 1991) and the fact that friends can intentionally send messages in a unique way to communicate more e€ ectively (Fleming & Darley, 1991), it was predicted that friends would outperform non-friends in the retrieval of words. Friends will outperform non-friends even if the retrieval phase includes a ``new item’ ’ word-list. ``For others’ ’ instructions

MEMORY, COLLABORATION AND FRIENDSHIP

277

during the generating phase will presumably increase the retrieval performance compared to participants that have ``for themselves’ ’ instructions.

EXPERIMENT 1 The aim of Experiment 1 was to investigate how dyads compared to nominal groups succee d in their memory retrieval given di€ erent types of cues, di€ erent encoding and di€ erent retrieval conditions (i.e. to test explicitly the reduced cue e€ ectiveness hypothesis in a collaborative context). The task was to generate associations of words from a word-list individually or dyadically and to retrieve the to-be-re membered words individually or in dyads. A nominal (cf . Faust, 1959) score was computed from two individuals working on their own and treate d as the predicted potential for the dyad. Thus Experiment 1 investigated (1) collaborative dyadic retrieval compared to the predicted potential of (nominal) group retrieval, (2) retrieval given self-generated cues compared to cues produced by others, and (3) the e€ ect on retrieval given dyadic or individual generation of cues.

Methods Participants. In total, 128 undergraduate students with a mean ( ± SD) age of 21.13 ± 2.75 years volunteered to participate. The participants were teste d at the Department of Education and Psychology, LinkoÈ ping University, Sweden and the Military A cademy in EnkoÈ ping, Sweden. Materials. One word-list comprising 90 Swedish words was presented in a to-be-remembered word booklet. The words in the word-list had a mean frequency of 10 words per million (A lle n, 1970). One word was presented on each page. Each participant constellation had a booklet with 90 blank pages to write down associations in, one association per page. A t retrieval of the to-be-remem bered words, the participants were either allowed to use their self-generated associations as cues or the cues produced by others. Cues produced by others were constructed from a pool of overlapping associations generated by 10 control students of the same age. The instructions given to the control participants were the same as for the participants in the experiment (see below). The associations used as ``cues produced by others’ ’ had a mean frequency of 2.63 (out of 10). The cues selected were the most overlapping associations generated (i.e. shared associations). The cues will therefore be denoted semantic cues because

278

ÈNNBERG ANDERSSON AND RO

they represen t, to some extent, more common/shared associations related to the target word. Procedure . The ® rst part of the whole session was devoted to the generation of one association to each word. The instructions were as follows: ``(1) Generate one association that is appropriate to you, and that you think of when you see the word, and (2) do not generate synonyms to the speci® c target word. Write down your association in the booklet with the blank pages (one word per page)’ ’ . The to-be-remembered words booklet with the 90 words (one word per page) was given to the participants who regulated the page turning by themselves. The maximum time allowed for the generating phase was 20 sec per word. The generating phase took no more than 30 min. Sixty-four participants were assigned to the dyadic generating condition and 64 to the individual generating condition. The dyads were instructed that they had to agree over the generated association. Each participant constellation generate d 90 associations and were then dismissed for 5 days. Thirty-two participants in the dyadic generating condition recalled the word-list alone and 32 in dyads. Thirty-two participants in the individual generating condition recalled alone and 32 recalled in dyads. Sixty-four participants recalled the word-list using cues produced by others and 64 used their self-generated cues. A t retrieval, all participant constellations received one cue per target, self-generated or generated by others, to accomplish the task requested. The instruction was to retrieve as many to-be-remembered words as possible with the help of the given cues. Thus the number of responses allowed per cue was not restricted . The participants were allowed to look at one cue at a time and only once. The selfpaced retrieval session took no more than 20 min. D esign. The experimental design was a 2 ´ 2 ´ 2 randomised factorial design. A ll factors were between-participant variables. The ® rst factor referred to the type of encoding condition (individual generation vs dyadic generation). The second factor referred to the type of retrieval condition (individually vs dyadically). The third factor referred to the type of associations the participants were allowed to use as a cue at retrieval (self-generated cues vs cues produced by others). A ll constellations received the same amount of information at the retrieval session (90 cues). In all, there were eight di€ erent conditions with eight dyads in each. Gender was balanced over condition. Two individuals, who generated associations by themselves initially, could in the retrieval session be recalling in dyads. They had initially produced two self-generated cues (one each) to each to-be-remem bered word. A t the retrieval session, the dyad received 45 cues produced by

279

MEMORY, COLLABORATION AND FRIENDSHIP

each participant (2 ´ 45 = 90) in a balanced order. A lso, eight dyads (who generated associations on their own initially as well) received both their self-generated associations as cues instead of one (2 ´ 90 = 180 cues; employed as an extra control group). This was done to provide for an equal amount of information in the relevant contrast with the ``pooled’ ’ nominal group. In the retrieval situation, however, each member of the pooled nominal group used one cue per target, whereas the extra controls used two cues per target. Thus nine conditions instead of eight were employed for control purposes.

Results and Discussion A ll answers that included the target word were given a score of 1 point. The maximum score was 90 points. A nominal group score was computed for two individuals working alone at retrieval and the pooled individual result was treate d as the predicted potential for the dyad. If only one of the individuals in the nominal group retrieved the correct word, the nominal group was still only given 1 point (cf . A ndersson & R oÈ nnberg, 1996). Table 1 shows the mean performances for all group conditions (nominal groups and collaborating dyads), and individual results for the four conditions that became, by way of scoring, the nominal group, and the extra control group that received two cues at retrieval (one self-generated cue from each participant). The scoring was done blind. A main e€ ect of retrieval [F(1,56) = 6.06, P < 0.02, MSe = 154.73] revealed that nominal group retrieval was, as expected , higher than TABLE 1 Mean (± SD) Retrieval Performance for all Conditions in Experiment 1

Ð

S elf-generated Cues Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Individual D yadic Generation Generation Ð

Ð

Cues Produced by O thers Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Ð Individual D yadic Generation Generation

Individua l retrieval

48.01

± 20.5

39.81

± 12.6

19.69

± 11.7

22.19

±

6.4

N ominal group retrieval

71.75

± 13.6

54.88

± 10.2

33.13

± 12.4

34.50

±

7.0

D yadic retrieval

37.75

±

60.38

± 19.9

32.50

± 12.4

33.00

± 10.1

D yadic retrieval with two cues

60.10

± 12.8

9.8

280

ÈNNBERG ANDERSSON AND RO

dyadic retrieval. A main e€ ect for type of cues [F(1,56) = 54.26, P < 0.001, MSe = 154.73] shows that self-generated cues were superior to cues produced by others (cf. MaÈ ntylaÈ , 1986). The interaction, retrieval ´ type of cues [F(1,56) = 4.50, P < 0.05, MSe = 154.73), suggests that both main e€ ects are quali® ed by the fact that dyadic retrieval only su€ ered from collaboration compared to nominal group retrieval performance, in the conditions where self-generated cues were used [t(56) = 3.24, P < 0.01, M Se = 154.73]. These e€ ects demonstrate that dyadic su€ ering became pronounced during retrieval of episodic traces by means of episodic cues, even if episodic cues were generally superior to semantic cues in all conditions. No such loss of collaboration existed for the retrieval of an episodic trace by a semantic cue (cues produced by others; P > 0.05). The interaction of encoding and retrieval [F(1,56) = 9.64, P < 0.01, MSe = 154.73] suggests that the retrieval condition e€ ect was quali® ed by the compatibility between encoding and retrieval, hence reducing loss of collaboration. The two interactions were quali® ed by the three-way interaction, encoding ´ retrieval ´ type of cues [F(1,56) = 10.54, P < 0.002, MSe = 154.73], suggesting the following: The two conditions that employed self-generated cues and a compatibility between encoding and retrieval di€ ered [t(56) = 1.83, P < 0.05, MSe = 154.73], implying that the nominal group (where the individuals encoded and retrieved on their own) outperformed the dyads (who encoded and retrieved in dyads). No di€ erences were found between the four retrieval conditions where cues were produced by others (semantic cues, all P > 0.05). A nother interesting result, quali® ed by the three-way interaction, was that the dyadic generating nominal group (collaborative encoding and individual retrieval) was a€ ected negatively compared with the individual generating nominal group (individual encoding and individual retrieval) [t(56) = 2.71, P < 0.01, MSe = 154.73]. This suggests that the dyadic encoding phase reduced the uniqueness of the cue. These results show that dyads suboptimised the potential of their collective memory ability (A ndersson & R oÈ nnberg, 1996) and that collaboration a€ ected the uniqueness of the cues negatively both at encoding and retrieval and therefore reduced memory performance. The individual generating nominal group also outperformed the dyads that used two self-generated cues [t (14) = 1.77, P < 0.05] at retrieval (the extra control group; Table 1). Thus these e€ ects reveal that episodic memory performance using self-generated cues was a€ ected not only by dyadic collaboration at encoding and retrieval, but also by the compatibility between encoding and retrieval (alone± alone or dyad ± dyad). In summary, episodic cues were superior to semantic cues in supporting memory performance, but given episodic cues, retrieval was

MEMORY, COLLABORATION AND FRIENDSHIP

281

negatively a€ ected by collaboration Ð that is, collaboration reduced cue e€ ectiveness. Experiment 2 was conducted to test further the reduced cue e€ ectiveness hypothesis, which suggests that friends can cue each other more e ciently than non-friends (A ndersson & R oÈ nnberg, 1995, 1996). These previous ® ndings suggest that friends reduce the net negative e€ ect of collaboration compared to non-friends. The assumption made was that friends communicated more e ciently (increased cue e€ ectiveness). Experiment 1 showed that the e€ ectiveness of the cue was reduced if the cue was ``produced by others’ ’ and if the individuals collaborated during encoding and retrieval. No di€ erences were found between the constellations that retrieve d with cues ``produced by others’ ’ . This ® nding is important, since nominal groups would outperform collaborating dyads if the superiority of the nominal group was due to the pooling procedure. To investigate the nature of intention in the communication superiority of friends, instruction type was varied. In Experiment 2, all participants received ``cues produced by others’ ’ as retrieval cues, but were explicitly instructed in the ``for others’ ’ condition to generate retrieval cues for their friend or non-friend. To investigate the friendship e€ ect further and to assess its impact relative to di€ erent kinds of memory tasks, participants either retrieve d an ``old item’ ’ word-list with cues ``produced by others’ ’ alone (episodic task with semantic cues), or generated/guessed a ``new item’ ’ word-list with cues ``produce d by others’ ’ alone (semantic task with semantic cues). Friends and non-friends were requested to communicate by means of cues in a similar task as in Experiment 1, but to increase the impact of communication (i.e. cueing), three cues for each target were generated at the encoding phase instead of one.

EXPERIMENT 2 The aim of Experiment 2 was to investigate how friends, compared to non-friends, succeeded in the retrieval of words from a word-list that they had worked with on their own initially (old items) during the encoding phase, or with a word-list never seen before (new items) . It was also of interest whether friends, compared to non-friends, can, in a more subtle way, help each other, using ``friendly’ ’ cues, to retrieve to-be-remem bered words more successfully. In all conditions, the participants were given their friends’ or non-friends’ generated cues. The participants never received their own generated cues at retrieval. During the encoding phase, participants were to generate three associations to each word. Participants were also given di€ erent types of instructions. One type of instruction was intended to make the participants believe that their cues were to be

282

ÈNNBERG ANDERSSON AND RO

used by themselves (``for themselves’ ’ instructions), and one instruction type was intended to make the participants believe that their cues were retrieval cues for their companion (``for others’ ’ instructions). A ll participants actually received their friend or non-friend cues at the retrieval phase (the ``for themselves’ ’ instructions were therefore not true). The aim was to investigate the importance of intention during the encoding phase and how it a€ ects mem ory retrieval. Thus Experiment 2 investigated (1) the memory performance of friends compared to non-friends, (2) the retrieval associated with the ``old item’ ’ word-list or the ``new item’ ’ word-list, (3) to what extent friends compared to non-friends can, given a ``new item’ ’ word-list at retrieval, help their partner to reach optimal retrieval performance, and (4) the importance of knowing whom the message is intended for using di€ erent instructions in the encoding phase.

Methods Participants. Ninety-six undergraduate students aged 23.79 ± 3.85 years volunteered to participate. Forty-eight participants were assigned to the friend partner condition and 48 to the non-friend partner condition. The friends had worked together for at least 1 year and socialised in their free time. The non-friend was unknown to the other person. They were drawn from di€ erent study programmes in the department. Gender was balanced over condition. Participants were tested at the Department of Education and Psychology, LinkoÈ ping University, Sweden. Materials. Two word-lists with 100 high-frequency Swedish words each were presented in two booklets. The words in the word-lists had a mean frequency of 24 and 26 words per million respectively (A lle n, 1970). The booklets included one word per page. Each participant received one word-list and four association booklets, giving a total of 100 blank pages to write down associations, three associations to each page. A t retrieval of the word-list, all participants were allowed to use their partners’ (friends/non-friends) associations as cues and never their own. A t the encoding phase, a tape-recorder was used to indicate time periods (30 sec) and at the retrieval phase an answering sheet with 114 lines was used. Procedure . The ® rst part of the whole session was devoted to the generation of three associations to each to-be-remembered word. Word-lists were balanced over conditions. The participants were allowed a maximum of 30 sec for each word and the tape-recorder was used to indicate the time periods. Forty-eight were instructed to generate three associations

MEMORY, COLLABORATION AND FRIENDSHIP

283

that their friend/non-friend was to use as a cue for success ful memory retrieval later on (``for others’ ’ instructions). The non-friends were also told that, ``your non-friend partner is unknown to you, but is studying behavioural science in the department’ ’ . Forty-eight participants were instructed that they themselves were to use their self-generated associations as retrieval cues (``for themselves’ ’ instructions). In fact, all participants used associations produced by others as cues in the retrieval session. A fter the generation phase, all participants were dismissed for 1 day. Pilot experiments revealed that a delay of 1 day was enough to avoid ceiling e€ ects. The next day, all participants received cues generated by friends/non-f riends taken from an ``old item’ ’ or ``new item’ ’ word-list. Twenty-four of the friends and 24 of the non-friends received three cues at retrieval for each to-beremembered word, given the ``old item’ ’ word-list. Twenty-four of the friends and 24 of the non-friends received three cues for each to-beremembered word, given the ``new item’ ’ word-list. In all, there were eight di€ erent conditions with 12 participants in each. A t retrieval, all participants were informed of the correct (true) circumstances. The non-friends in the ``new item’ ’ word-list condition with ``for themselves’ ’ instructions, for example, were told that their retrieval cues were produced by a student studying behavioural science in the department, and were generated from a new, non-studied (for them) word-list. They were given the four association booklets (three words per page ´ 25 pages) that their ``partner’ ’ produced the day before. R etrieval was selfpaced and all participants used an answering sheet to record their words/guesses. D esign. The experimental design was a 2 ´ 2 ´ 2 randomised factorial design. A ll factors were between-participant variables. The ® rst factor referred to friendship (friends vs non-friends), the second referred to word-list condition (``old item’ ’ vs ``new item’ ’ word-list) and the third factor referred to instruction condition (``for others’ ’ vs ``for themselves’ ’ ).

Results and Discussion Every answer including the target word was given 1 point. The maximum score was 100 points. Mean scores and standard deviations for all conditions are presented in Table 2. The scoring was done blind. The overall A NOVA revealed no interaction e€ ects but three main e€ ects. First, friends outperformed non-friends in all conditions [F(1,88) = 10.36, P < 0.002, MSe = 108.82]. Friends could, with the help of their partner’ s cues, retrieve and guess the correct word to a greater extent than non-friends. This ® nding supports the data discussed by

284

ÈNNBERG ANDERSSON AND RO TABLE 2 Mean (± SD) Retrieval Performance for all Conditions in Experiment 2 ``For O thers’ ’ Instructions ``O ld Item ’ ’ W ord-list

``New Item ’ ’ W ord-list

``For T hem selves’ ’ Instructions ``O ld Item ’ ’ W ord-list

``New Item ’ ’ W ord-list

Friends

81.00

± 7.5

47.75

± 9.8

74.25

± 10.9

41.17

± 11.1

N on-f riends

76.08

± 8.8

38.17

± 7.9

67.58

± 13.7

34.92

± 12.1

A ndersson and R oÈ nnberg (1995), in the sense that friends are able to cue one another more e ciently (i.e. friends increase cue e€ ectiveness compared to non-friends). It also replicates and further advances our understanding of the net negative e€ ects of collaboration. A nother type of control, not included in the design, would be participants receiving their own self-generated cues, but, based on the ® ndings of MaÈntylaÈ (1986), which clearly show that individuals with self-generated cues outperform individuals with cues produced by others, no such control was deemed necessary. Second, a main e€ ect of type of list [F(1,88) = 258.39, P < 0.001, MSe = 108.82] revealed that participants who received cues from the ``old item’ ’ word-list out-scored participants who received cues from a ``new item’ ’ word-list. The data in Experiment 1 suggest a compatibility superiority e€ ect between the encoded episodic trace and the retrieval cue. Experiment 2 suggests that the retrieval (guessing) of a semantic memory (``new item’ ’ list) by non-friends using a semantic cue (produced by a non-friend) was high in compatibility (in terms of memory system) but low in subsequent memory performance. Thus this result reveals a superiority e€ ect for episodic compatibility compared to semantic compatibility. Third, a main e€ ect of instruction was found [F(1,88) = 8.67, P < 0.01, MSe = 108.82], suggesting that explicit knowledge of the fact that associations were to be used by others as cues (friends or nonfriends) improved retrieval (cf . Fleming & Darley, 1991; Fussell & Krauss, 1989). Combined, these main e€ ects clearly demonstrate the end-points of the performance continuum: The friends that retrieved an ``old item’ ’ word-list with cues produced with ``for others’ ’ instructions, and the non-friends that retrieved/guessed a ``new item’ ’ word-list with cues produced with ``for themselves’ ’ instructions, were the two conditions that contributed to the best and the worst performances, respectively. These results suggest that friends have some kind of overlapping

MEMORY, COLLABORATION AND FRIENDSHIP

285

understanding or common ground (cf . Fussell & Krauss, 1989) that facilitates sending and utilising retrieval cues. This in itself is an important ® nding.

GENERAL DISCUSSION Dyadic memory retrieval was inferior to that of a nominal group using cues produced by themselves (episodic task). Experiment 1 replicated the net negative e€ ect of collaboration and provides further evidence for the interpretation that members of a dyad are unable to cue one another with su cient precision (cf . Meudell et al., 1992, 1995). The dyadic encoding phase, as well as the dyadic retrieval phase, reduced the e€ ectiveness of the cue. We also demonstrated that collaboration during retrieval with semantic cues (the four groups that received cues produced by others) was not a€ ected negatively among these groups. Thus an episodic task with episodic cues was a€ ected by the net negative e€ ect of collaboration, but not an episodic task with semantic cues. This result replicates A ndersson and R oÈ nnberg’ s (1996) ® ndings, but also advances the possible explanation for the net negative e€ ect, as the experiment explicitly supported the retrieval cue e€ ectiveness hypothesis (cf . M aÈntylaÈ , 1986; Tulving, 1979). It has been suggested that friends can communicate and send messages to each other in a su ciently e€ ective way (cf . Fleming & Darley, 1991), that they share a common ground (Fussell & Krauss, 1989), or that they even have a transactive memory (W egner et al., 1991). Experiment 2 strongly supports these suggestions, since friends performed better than non-friends in all conditions. Fussell and Krauss (1989) showed that it was easier to identify nonsense ® gures using a friend’ s description than the description of a stranger. Both instruction conditions showed similar results in the present experiment. Even in the ``for themselves’ ’ condition, friends performed better than non-friends. Fleming and Darley (1991) showed that friends can send messages to friends and mislead others. Their senders knew who their receivers were. Our data are in accord with their ® ndings, since ``for others’ ’ instructions promoted higher recall than ``for themselves’ ’ instructions. But knowledge of whom the sender was also a€ ected memory retrieval to a substantial degree. The experiment suggests that the semantic retrieval (guessing) task (``new item’ ’ wordlists) with semantic retrieval cues (produced by others) was a€ ected by friendship. Friends could reduce the memory loss by means of an overlapping understanding of each other. The status of the sender in combination with the semantic cue seems to create unique extra information for friends but not for non-friends.

286

ÈNNBERG ANDERSSON AND RO

CONCLUSIONS The main ® ndings suggest (1) that the net negative e€ ect of collaboration in episodic memory tasks can be explained by a lack of precision in the generation, retrieval and communication of cues, and (2) that friends can increase cue e€ ectiveness. The group productivity decrement for retrieval with episodic cues but not with semantic cues is interesting, since the two factors often used to explain loss of productivity in groups are loss of motivation and loss of coordination (Brown, 1988; Steiner, 1972). These factors, however, do not explain why groups performing cognitive tasks, such as memory retrieval, are a€ ected in the way A ndersson and R oÈ nnberg (1996) suggest and the present study has revealed. Due to the type of memory task, we can again di€ erentiate (cf . A ndersson & R oÈ nnberg, 1996) between loss and no loss in collaborative memory retrieval and, explicitly, show that collaboration a€ ects cue e€ ectiveness. Friends who communicate by means of cues increase the compatibility between memory and cue. Their shared memories in¯ uence the sender to produce more appropriate retrieval cues than non-friends. However, shared memories do not work e ciently enough to cue speci® c memories, since friends do not outperform non-friends in semantic tasks (cf . A ndersson & R oÈ nnberg, 1996). A tentative explanation might be that the communication between friends increases understanding, activates shared episodic and semantic information, and therefore reduces possible retrieval alternatives. But this e€ ect is less likely to occur in semantic tasks that do not trigger shared episodic or autobiographical components. Thus it is important for the purposes of collaborative retrieval to distinguish semantic tasks that are prone to trigger autobiographical experiences shared by friends, from semantic tasks of a more factual type (cf . Experiment 1; A ndersson & R oÈ nnberg, 1995). In the former case, a friendship e€ ect is to be expected , but not in the latter. Thus, as long as there is an episodic component in any memory task, a positive friendship e€ ect should be expected . Net negative e€ ects of collaboration can be explained in terms of the reduced cue e€ ectiveness hypothesis. M anuscrip t received February 1996 R evised m anuscrip t received A pril 1996

REFERENCES A lleÂn, S. (1970). Frequency dictionary of present-day S wedish. Stockhol m : A lm qvist & Wiksell. A ndersson , J., & R oÈ nnberg , J.R . (1995). R ecall su€ ers from collaboration: Joint recall e€ ects of friendship and task complexity. Applied Cognitive Psychology, 9, 199± 211.

MEMORY, COLLABORATION AND FRIENDSHIP

287

A ndersson , J., & R oÈ nnberg, J.R . (1996). Collabora tion and m em ory: E€ ects of dyadic retrieval on di€ erent m em ory tasks. Applied Cognitive Psychology, 10, 171± 181. Brown, R . (1988). Group processes: D ynam ics within and between groups. Oxford: Blackwell. Faust, W.L. (1959). Group versus individual problem solving . Journal of Abnorm al and S ocial Psychology, 59, 68 ± 72. Fleming, J.H ., & Darley, J.M. (1991). Mixed messages: The multiple audience proble m and strategic com munication. S ocial Cognition, 9, 25 ± 46. Fussell, S.R ., & Krauss , R .M . (1989). Understa nding friends and strangers: The e€ ects of audience design on m essage comprehension. European Journal of S ocial Psychology, 19, 509 ± 525. Meudell, P.R ., Hitch, G.J. , & K irby, P. (1992). A re two heads better than one? Experimental investigations of the social facilitation of memory . Applied Cognitive Psychology, 5, 525 ± 543. Meudell, P.R ., H itch, G.J. , & Boyle, M.M . (1995). Collaboration in recall: Do pairs of people cross-cu e each other to produc e new memorie s? Q uarterly Journal of Experim ental Psychology, 48A, 141 ± 152. MaÈ ntylaÈ , T. (1986). How do you cue? PhD thesis, U niversity of UmeaÊ , Sweden . MaÈ ntylaÈ , T., & Nilsson , L.G. (1983). A re m y cues better than your cues? S candinavian Journal of Psychology, 24, 303 ± 312. Steiner, I.D . (1972). Group process and productivity. New Y ork: A cademic Press. Tulving, E. (1979). R elation between encoding speci® city and levels of processin g. In L.S. Cerma k & F.I.M . Craik (Eds), L evels of processin g in hum an m em ory, pp. 405 ± 428. Hillsdale, NJ: Lawrenc e Erlbaum A ssociate s Inc. Tulving, E. (1983). Elem ents of episodic m em ory. New Y ork: Oxf ord U niversity Press. Tulving, E. (1985). H ow m any memory system s are there? Am erican Psychologist, 40, 385 ± 398. Wegner, D.M., Erber, R ., & R aymond, P. (1991). Transactive memory in close relationships. Journal of Personality and S ocial Psychology, 61, 923 ± 929. Wiseman, S., & Tulving, E. (1976). Encoding speci® city: The relation s between recall superiority and recognition failure. Journal of Experim ental Psychology: H um an L earning and Mem ory, 2, 349 ± 361.