Journal of Experimental Social Psychology
Journal of Experimental Social Psychology 38 (2002) 535–544 www.academicpress.com

Cognitive stimulation and interference in groups: Exposure effects in an idea generation task
Bernard A. Nijstad,a,* Wolfgang Stroebe,b and Hein F.M. Lodewijkxb a Department of Work and Organizational Psychology, University of Amsterdam, Roetersstraat 15, 1018 WB, Amsterdam, The Netherlands b Utrecht University, Utrecht, The Netherlands Received 5 December 2000; revised 21 January 2002

Abstract The effects of idea sharing on cognitive processes and performance were assessed in an idea exposure paradigm. Participants generated ideas while being exposed to stimulus ideas that were semantically homogeneous or diverse, and were offered in an organized or a random sequence. As compared to a control condition, participants generated more diverse ideas when exposed to ideas from a wide range of semantic categories, and they generated more ideas per category when exposed to many ideas from only a few categories. The semantic organization of ideas was higher when participants were exposed to ideas that were organized in semantic clusters than when participants were exposed to unorganized ideas. Idea exposure had positive effects in general, because it reduced response latencies for category changes. Implications for information processing in groups are discussed. Ó 2002 Elsevier Science (USA). All rights reserved.

Groups increasingly perform cognitive tasks, such as problem solving, decision making, inference, and idea generation. Recently, it has been argued that groups, much like individuals, can be conceptualized as information processors (Hinsz, Tindale, & Vollrath, 1997). Information processing in groups involves activities that occur at the individual as well as the group level. At the individual level, group members must process information, which leads to individual solutions, preferences, and ideas. At the group level, information processing involves the sharing of these solutions, preferences, and ideas during discussion (Tindale & Kameda, 2000). A key question regarding groups as information processors is how information or ideas suggested by one group member affect the mental processes of other members, and how this relates to group effectiveness. Hinsz et al. (1997) argued that the utterances of others may both stimulate and interfere with the mental processes of group members, and that this provides ‘‘a new area to address the relative impact of process gains (e.g., stimulated cognitive processes. . .) and process losses

*

Corresponding author. Fax: +31-20-639-0531. E-mail address: nijstad@psy.uva.nl (B.A. Nijstad).

(e.g., interference. . .) in small-group performance’’ (p. 49). When stimulation is stronger than interference, the result is an assembly bonus effect (Collins & Guetzkow, 1964), in which groups outperform their separate members. In this paper, we will focus on idea generation by groups and consider the question of how ideas suggested by one group member can affect the cognitive processes of the other members. Research on group idea generation has a long history in social and organizational psychology (for reviews see Lamm & Trommsdorff, 1973; Stroebe & Diehl, 1994). A very consistent finding in this literature is that a number of people working individually (nominal groups) can produce more ideas and more good ideas than can an equal number of people working in a group (Mullen, Johnson, & Salas, 1991). One important cause for this productivity loss in groups is mutual production blocking (Diehl & Stroebe, 1987). Usually, only one group member speaks at a given moment, so group members must often wait for their turn before they can express their ideas. It has been shown that group members cannot think effectively while waiting for their turns, and that the blocking effect is thus due to cognitive interference (Diehl & Stroebe, 1991; Nijstad, 2000).

0022-1031/02/$ - see front matter Ó 2002 Elsevier Science (USA). All rights reserved. PII: S 0 0 2 2 - 1 0 3 1 ( 0 2 ) 0 0 5 0 0 - 0

536

B.A. Nijstad et al. / Journal of Experimental Social Psychology 38 (2002) 535–544

Production blocking is one (unintended) consequence of idea sharing in groups. Another consequence, often presumed to have positive effects (e.g., Osborn, 1957), is that people have access to each otherÕs ideas. We propose that access to othersÕ ideas can have positive effects (cognitive stimulation), but that negative effects (cognitive interference) are also possible. In this paper, we first present a theoretical analysis of when positive or negative effects are likely to occur, and argue that an assembly bonus effect can be found when interference is prevented, while stimulation is maintained. Then we will discuss recent evidence pertaining to this issue, and present an experiment designed to test some hypotheses derived from our analysis.

Cognitive processes, stimulation, and interference To derive hypotheses regarding the way idea sharing affects the mental processes of group members, it is necessary to have some insight into these mental processes. To this end, we have developed a theory of idea generation, called SIAM (Search for Ideas in Associative Memory). SIAM is an application to idea generation of Raaijmakers and ShiffrinÕs (1981) Search of Associative Memory (SAM) model of memory retrieval. Idea generation differs from retrieval, because (new) ideas cannot be directly retrieved from memory. However, idea generation necessarily involves retrieval processes, because ideas cannot be generated ex nihilo, so previously stored knowledge must be used to generate ideas (cf. Amabile, 1983). Following SAM, SIAM distinguishes between a limited capacity short-term memory (STM), in which conscious operations are performed, and an unlimited capacity long-term memory (LTM), in which previously acquired knowledge is stored. LTM is partitioned into unitized images, or localized sets of strongly interconnected and semantically related features. These images are connected in a rich network, with many associations, levels, and categories. Semantically related images are presumed to have relatively strong mutual ties (cf. Collins & Loftus, 1975), and it is assumed that only one image can be active (in STM) at a given moment. SIAM proposes that idea generation is a two-stage process, in which a knowledge activation stage is followed by an idea production stage. To activate knowledge, a search cue is assembled in STM, which is used to probe LTM. This search cues consists of (elements of) the problem definition and/or other cues (e.g., previously generated ideas). A probe of LTM results in the activation of an image. Which image is activated is probabilistic and depends on the strength of the association between the search cue and the image. In the idea production stage, the features of the image are used to generate (new) ideas, by combining knowledge, forming

new associations, or applying knowledge to a new domain (Mednick, 1962).1 These ideas are added to the search cue to activate more images in memory, leading to the generation of additional ideas. Because semantically related images have strong mutual ties, successively activated images will often be semantically related. This leads to a Ôtrain of thought,Õ and a rapid accumulation of semantically related ideas. When a train of thought no longer leads to ideas (searches are unsuccessful), a new search cue must be assembled, which is a conscious process that takes some time. This new cue is used to probe memory, which results in the activation of new images, and the generation of additional ideas. This process continues until the session is terminated. At the individual level, one implication is that some degree of semantic clustering of ideas should be found, and that ideas within semantic clusters should be generated relatively quickly. Similar semantic and temporal clustering has been found in the free recall of categorized lists of words (e.g., Bousfield, 1953; Gruenewald & Lockhead, 1980). The ideas generated by others serve as external stimuli, which can be added to the search cue to probe LTM. We propose that these ideas are added to the search cue relatively automatically, provided that attention to the ideas is sufficiently high. In general, the ideas of others will stimulate idea generation (cognitive stimulation), because less time is needed to assemble search cues and search memory for problem-relevant knowledge. Depending on the semantic content of stimulus ideas, two types of positive effects are possible. First, the ideas of others can activate knowledge that otherwise would not be accessible (cf. Tulving & Pearlstone, 1966). This is likely to happen when stimulus ideas are semantically diverse. Diverse stimuli lead to the activation of a wide range of knowledge (including less accessible knowledge), which allows for the generation of semantically (more) diverse ideas and productivity gains. In contrast, stimulus ideas that are semantically homogeneous are unlikely to activate less accessible knowledge. Within the range of accessible knowledge, however, many ideas will be generated, because the (limited) range of knowledge remains highly accessible throughout the session. This will lead to productivity gains, as long as the opportunities to generate additional ideas within this limited range are not exhausted. To sum up, diverse stimulation will increase the breadth of idea production, whereas homogeneous stimulation will increase the depth of idea production.
1 SAM also assumes a two-stage process. The first stage in SAM is the activation of images, using a search cue that contains contextual information (information that was present during learning) and previously recalled items. In the second stage, SAM assumes a recovery process, in which the previously learned item (e.g., a word) is recovered from the image (see Raaijmakers & Shiffrin, 1981).

B.A. Nijstad et al. / Journal of Experimental Social Psychology 38 (2002) 535–544

537

Stimulus ideas can also interfere with a train of thought (cognitive interference). When stimulus ideas activate an image that is at odds with a personÕs train of thought, it may be prematurely aborted. This can lead to shorter trains of thought, the loss of potential ideas, and increased switching between semantic domains. This will reduce the depth of idea production, because rapid associations within a semantic domain are prevented. Cognitive interference is likely to occur when successive stimulus ideas activate semantically dissimilar images. When stimulus ideas are offered in clusters of semantically related ideas, cognitive interference may be less important, because stimuli successively activate semantically related images and a train of thought can be maintained. Thus, productivity gains should be larger when stimuli are offered in clusters than when they are offered at random, because idea stimulation is maintained and interference is prevented.

Evidence for stimulation and interference Recent evidence suggests that both cognitive stimulation and cognitive interference can occur in groups. Support for cognitive stimulation comes from studies showing productivity gains as a consequence of idea sharing. These gains have been found in paradigms where there is no turn-taking among participants. Because turn-taking causes production blocking, productivity losses are usually found (see Diehl & Stroebe, 1987, 1991). However, when ideas are not expressed aloud, but are recorded instead by the means of written notes (ÔbrainwritingÕ) or computers (Ôelectronic brainstormingÕ), production blocking is eliminated (see e.g., Gallupe, Cooper, Gris, & Base tianutti, 1994), and productivity gains become possible. For example, one brainwriting study showed that participants who could share ideas through written notes outperformed those who could not (Paulus & Yang, 2000). Productivity gains have also been observed in Ôelectronic brainstormingÕ (EBS). Relatively large EBS groups (n > 9) whose members shared ideas outperformed equivalent nominal groups without sharing (Dennis & Valacich, 1993; Valacich, Dennis, & Connolly, 1994). Leggett Dugosh, Paulus, Roland, and Yang (2001) used a paradigm in which individuals generated ideas while listening to a tape containing stimulus ideas. They found that the tape enhanced performance, but only if participants were instructed to remember the stimulus ideas, presumably because that increased their attention to the tape. This is consistent with our analysis, because only when attention to stimulus ideas was sufficiently high would stimuli be added to the search cue to probe memory. Leggett Dugosh et al. also found that productivity gains could be obtained in relatively small EBS groups (n ¼ 4), when attention to the ideas of others was increased by memory instructions. Unfortunately, in the

studies reporting productivity gains, the content of ideas was not assessed, so no conclusions regarding the effects of idea sharing on the content of ideas could be made. Idea content has been assessed in several other studies. Larey and Paulus (1999), for example, categorized the ideas generated by (orally) interactive groups and by nominal groups, and found that interactive groups surveyed fewer categories of ideas than did nominal groups. However, that effect may have been due to either production blocking or overhearing the ideas of others. Similarly, Ziegler, Diehl, and Zijlstra (2000) categorized the ideas of electronic brainstorming (EBS) groups. Four-person EBS groups that could share ideas surveyed fewer categories than did EBS groups that could not share ideas. However, the idea categories were surveyed in greater depth, which is consistent with our analysis. Finally, a study by Stroebe and Diehl (1994) suggests that the tendency of groups to survey fewer categories may be restricted to relatively homogeneous groups. They found that heterogeneous groups (in terms of dominant associations to the topic at hand) surveyed more categories than did homogeneous groups, and that heterogeneous groups outperformed homogeneous groups. This suggests that idea sharing was more stimulating in heterogeneous groups, presumably because a broader range of knowledge was accessible due to the sharing of more diverse ideas. There is some evidence for cognitive interference as well, both in free recall and idea generation research. Basden, Basden, Bryner, and Thomas III (1997) compared the performance of interactive and nominal groups in the free recall of categorized lists of words. They predicted and found that interactive groups recall fewer words than nominal groups. The degree of semantic organization (clustering) in recall was also much lower in interactive groups than in nominal groups, which suggests that overhearing the others interfered with the mental processes of group members and increased switching between semantic categories. When groups were forced to recall items category by category, thereby preventing switching between categories, no productivity loss in groups was found, perhaps because that procedure eliminated cognitive interference. Dennis, Valacich, Connolly, and Wynne (1996) reported a similar finding. They compared EBS groups that were instructed to generate ideas category by category with groups whose sessions were unstructured, and found that process structuring led to more ideas. Again, this is consistent with the idea that cognitive interference is prevented when there is less switching between semantic categories.

The present study Although the available evidence is consistent with our theoretical analysis, a direct test of our hypotheses re-

538

B.A. Nijstad et al. / Journal of Experimental Social Psychology 38 (2002) 535–544

quires manipulating the content of the ideas to which people are exposed. Because it is hard to manipulate the content of ideas in real interactive groups, we used an idea exposure paradigm instead (cf. Connolly, Routhieux, & Schneider, 1993; Leggett Dugosh et al., 2001). Participants generated and were exposed to stimulus ideas at a computer terminal. Stimulus ideas were either semantically diverse (representing a wide range of semantic categories) or semantically homogeneous (representing only a few semantic categories). In addition, ideas were offered in either a random order (random sequence), or in short clusters of several semantically related ideas (clustered sequence). A control condition was added in which participants were not exposed to ideas. The effects of exposure were assessed on productivity (the number of ideas generated), and all ideas were content-coded in semantic categories to establish the breadth (number of categories surveyed, or diversity) and depth (number of ideas per category, or within-category fluency) of production. The degree of switching between semantic categories was also assessed (clustering).

Hypotheses Productivity. Because stimulus ideas reduce the time needed to probe memory, we expected that exposure to ideas would be stimulating, leading to higher levels of productivity (number of non-redundant ideas) in the experimental conditions as compared to the control condition (H1). These productivity gains should be larger in the clustered sequence conditions than in the random sequence conditions, because cognitive interference in the latter conditions counteracts productivity gains (H2). This interference effect should be stronger with diverse and random stimulation than with homogeneous and random stimulation, because it is harder to maintain oneÕs own train of thought within a category when stimuli are more diverse. In addition to these predictions regarding stimulation effects on productivity, our model enables us to develop hypotheses about the distribution of ideas across categories. Stimulation effects. Due to the increased accessibility of a wide range of semantic categories, participants in the diverse stimuli condition should survey more categories than those in the control condition or those in the homogeneous stimuli conditions (H3; increased breadth or diversity). Participants in the homogeneous stimuli condition should generate more ideas within (fewer) categories than participants in the diverse stimuli conditions, because the stimulated categories were highly accessible throughout the session (H4; increased depth or within-category fluency). Interference effects. We expected that a random sequence of ideas would lead to cognitive interference

and interrupt a train of thought within a category. This should lead to fewer successive ideas in the same semantic category, and thus a lower level of production within semantic categories in the random stimulation condition compared to the condition with clustered stimulation (H5; reduced depth or withincategory fluency). These two effects on the production of ideas per category should counteract each other in the condition with homogeneous/random stimulation, so only small differences are expected between this condition and the control condition. However, participants in the condition with homogeneous/clustered stimulation should generate more ideas per category than those in the control condition, because categories will remain highly accessible and there will be no interference. Participants in the diverse/clustered condition should not differ from those in the control condition, but participants in the diverse/random condition should generate fewer ideas per category than participants in the control condition (because of interference). Finally, random stimulation should result in lower levels of semantic clustering compared to both the control condition and the conditions with clustered stimulation (H6, reduced clustering).

Method Participants and task Participants were 63 students of Utrecht University who received DFL 10 ($5) for their participation. For 20 min, they generated ideas individually at a computer terminal on the topic of what people can do to help preserve the environment. During the experiment, participants could not interact with one another. Design and materials The design was a 2 (stimulus diversity) Â 2 (stimulus sequence) factorial, with an added non-factorial control condition. In the control condition, no stimulus ideas were displayed on the monitors. In the experimental conditions, participants were exposed to a stimulus idea each time they entered an idea. Participants were exposed to stimuli selected from just two semantic categories (homogeneous stimuli), or to stimuli selected from 34 different semantic categories (diverse stimuli). Further, participants were either shown five successive ideas from the same semantic category (clustered sequence), or the ideas they saw were in a random, unclustered order (random sequence). Stimulus ideas were taken from a previous experiment (Nijstad, 2000). Two independent raters catego-

B.A. Nijstad et al. / Journal of Experimental Social Psychology 38 (2002) 535–544

539

rized these stimuli. CohenÕs j, a measure of coding agreement, was .87. The category system consisted of a matrix in which 10 goals were crossed with five means to reach those goals, resulting in 50 different categories (see Diehl, 1991). Examples of goals were Ôreduce waste production,Õ Ôreduce water use or pollution,Õ and Ôprotect animals and plants.Õ Examples of means were Ôconsumption,Õ Ôproduction,Õ and Ôorganization and action.Õ A total of 520 different stimulus ideas were randomly selected from the 34 categories that contained at least five different ideas. Using these ideas, four different computer files were created. One file consisted of all 520 selected ideas, representing all 34 categories. This file was used in the diverse stimuli condition. The other three files, used in the homogeneous stimulus condition, each contained the ideas from two large categories (with 50 or more different ideas). Three different sets of stimuli were used to ensure that the effects of diversity could not be attributed to specific categories. File 1 contained ideas from categories A and B, File 2 contained ideas from categories C and D, and File 3 contained ideas from categories E and F. Each file contained at least 100 different ideas. Procedure Upon arrival, participants were taken to separate small rooms, each containing a computer. They were told that they would be generating ideas, following the four brainstorming rules (generate many ideas, generate ÔwildÕ ideas, combine and improve, and no criticism) suggested by Osborn (1957). The session began with a three-minute practice period, during which participants generated ideas on how student housing could be improved. Next, they were told that we were interested in the effects of ideas from other people, and that some of them would be shown ideas on their monitors, while others would not be shown ideas. Participants were told that the ideas were drawn from a prepared file, containing ideas that had been generated by other students in an earlier experiment. To ensure that they paid attention to these ideas, participants in the experimental conditions were told that they would be tested for their recall of the ideas after the session ended (cf. Leggett Dugosh et al., 2001). Participants were reminded of the four brainstorming rules and the brainstorming topic. They were then given 20 min to generate as many ideas as possible on the environment problem. In the control condition, the topic was repeated on the monitor and space was provided to enter one idea. After someone pressed ‘‘enter,’’ the idea was stored and it became possible to enter a new idea. This was repeated until 20 min passed. In the experimental conditions, a stimulus idea appeared on the monitor each time the participants entered an idea. The idea was visible until

the participants entered a new idea. It became possible then to enter another idea, and a new stimulus was displayed on the screen. To ensure that the number of ideas displayed in the different conditions was approximately equal, no participant was shown more than 60 different ideas. Moreover, a maximum number of four different ideas were displayed per minute, and a new idea appeared on the screen after 20 s if a participant typed in no idea of his or her own. Ideas were drawn at random and without replacement from the previously prepared files. In the random sequence condition ideas were drawn and shown one at the time. In the random/diverse condition, where 34 different categories were used, it was unlikely that successive ideas would fall in the same category. In the random/homogeneous condition, where only two categories were used, ideas from the two categories alternated, with an average probability of .50 that successive ideas would fall in different categories. In the clustered sequence condition, clusters of five ideas from the same category were drawn at random. These ideas were then displayed one at the time. After showing five of them, the computer randomly drew another cluster of five ideas. No idea was shown more than once. After 20 min passed, participants in the experimental conditions were given a recall test. In 5 min, they had to write down as many ideas as possible from those that were displayed on their monitors. After this, the participants were debriefed, paid, and dismissed. Dependent variables Productivity was determined by counting the number of non-redundant ideas a person produced. These ideas were compared to the ideas offered as stimuli, and no credit was given for an idea that had appeared previously as a stimulus. Two independent raters categorized ideas as either new or old. CohenÕs j for their coding was .94. Next, all ideas were categorized in semantic categories by two independent raters. Their coding reliability was high as well (CohenÕs j ¼ :88). Based on their categorization of the ideas, three additional measures of performance were computed. First, diversity was the number of different semantic categories surveyed by a participant. Second, the average number of ideas per category (within-category fluency) was computed by dividing the number of non-redundant ideas by diversity (note that multiplying diversity by within-category fluency yields productivity). Third, the adjusted ratio of clustering (ARC, see Roenker, Thompson, & Brown, 1971) was computed. The ARC measures how often an idea is followed by an idea from the same category (a category repetition), corrected for chance. ARC scores usually fall between 0 (chance clustering) and 1 (maximal clustering), but negative scores (below

540

B.A. Nijstad et al. / Journal of Experimental Social Psychology 38 (2002) 535–544

Table 1 Productivity, diversity, within-category fluency, and clustering Measure Homogenous stimuli Random sequence Productivity Diversity Within-category fluency Clustering 41.42 (9.47) 13:50a (1.88) 3:08a (.58) .179 (.14) Clustered sequence 38.25 (10.80) 12:58a (2.91) 3:08a (.71) .240 (.11) Diverse stimuli Random sequence 39.33 (9.60) 17:50bc (3.97) 2:27b (.35) .186 (.13) Clustered sequence 41.25 (11.74) 17:83c (3.83) 2:29b (.36) .253 (.13) 32.40 (10.54) 14:27ab (2.84) 2:24b (.42) .301 (.13) Control

Note. Standard deviations are in parentheses. Different superscripts indicate significant differences on a Tukey post hoc test (p < :05). For each experimental condition n ¼ 12, for the control condition n ¼ 15.

chance clustering) are also possible.2 Finally, to measure recall, the number of correctly recalled stimulus ideas was counted. Two independent raters categorized the recalled stimuli as either correctly or incorrectly recalled. CohenÕs j was .89, again indicating good coding reliability.

Results Recall data The average number of correctly recalled stimulus ideas was 15.03, which was 30% of ideas that were displayed. Thus, at least to some degree, participants paid attention to the ideas. According to a 2 (stimulus diversity) Â 2 (stimulus sequence) ANOVA, participants in the homogeneous stimuli condition recalled somewhat more ideas (M ¼ 16:28) than did participants in the diverse stimuli condition (M ¼ 13:78), F ð1; 32Þ ¼ 3:43, p < :10.3 This may be due to the fact that semantically related stimuli are easier to recall than unrelated stimuli. The main effect of stimulus sequence, and the interaction between stimulus diversity and sequence, were not significant, F s < 1. Performance data Productivity. All of these hypotheses were tested using planned comparisons. According to Hypothesis 1, exposure to ideas should on average lead to productivity gains relative to the control condition. A planned comparison between the control condition (M ¼ 32:40)
2 The formula to compute the ARC is: ARC ¼ ½R À EðRފ= ½max R À EðRފ. In this formula, R is the number of observed category repetitions, EðRÞ is the expected number of category repetitions according to chance, and maxR is the maximum number of category repetitions. MaxR is equal to N À k, where k is the number of categories surveyed by a participant, and N is the total number of ideas generated. EðRÞ can be computed with the following formula, where ni is the number of ideas in category i: EðRÞ ¼ ðRn2 =N Þ À 1. i 3 Due to missing data (12 cases), the number of degrees of freedom in this analysis is lower than in the other analyses.

on one hand and the experimental conditions (M ¼ 40:06) on the other, confirmed this prediction, tð58Þ ¼ 2:48, p < :05. Hypothesis 2 predicted that productivity gains would be larger in the clustered sequence than in the random sequence conditions, because of cognitive interference in the latter conditions. However, a planned comparison between the clustered and random conditions was not significant, tð58Þ ¼ :21; ns. Tukey post hoc tests showed no differences between these conditions either. Thus, there were productivity gains overall, but no evidence for cognitive interference (see Table 1). Diversity. According to Hypothesis 3, the number of categories surveyed should have been higher in the diverse stimuli conditions than in the control condition. A planned comparison showed that this was the case (M ¼ 17:67 for the diverse stimuli conditions, and M ¼ 14:27 for the control condition), tð58Þ ¼ 3:27, p < :01. Diversity was also higher in the diverse stimuli conditions than in the homogeneous stimuli conditions (M ¼ 13:04), tð58Þ ¼ 5:07, p < :001 (see Table 1). Hypothesis 3 was thus confirmed—diverse stimulation increased the breadth of idea production. Within-category fluency. Hypotheses 4 and 5 involved within-category fluency (the average number of ideas per category). Hypothesis 4 predicted that within-category fluency would be higher in the homogeneous stimuli conditions than in the diverse stimuli conditions, because the stimulated categories would remain highly accessible throughout the session in the homogeneous conditions, but not in the diverse conditions. A planned comparison confirmed the predicted effect, tð58Þ ¼ 5:56, p < :001; within-category fluency was higher in the homogeneous stimuli conditions (M ¼ 3:08) than in the diverse stimuli conditions (M ¼ 2:28). Hypothesis 5 predicted that a random sequence of ideas would lead to cognitive interference and a premature interruption of someoneÕs train of thought. This should have led to lower levels of production within categories in the random sequence conditions versus the clustered sequence conditions. However, the planned comparison testing this prediction was not significant, tð58Þ ¼ :08; ns. Within-category fluency was as high in the random sequence conditions (M ¼ 2:68) as in the clustered

B.A. Nijstad et al. / Journal of Experimental Social Psychology 38 (2002) 535–544 Table 2 Response latencies (in s) for category repetitions and category changes Type of idea Homogenous stimuli Random sequence Category change Category repetition 23:16a (2.83) 21:52a (5.04) Clustered sequence 24:55a (4.85) 24:05a (7.09) Diverse stimuli Random sequence 24:37a (4.91) 25:13a (7.88) Clustered sequence 25:55a (8.44) 27:38a (11.92) 41:53b (12.06) 29:33a (8.78) Control

541

Note. Response latencies measure the time to generate and enter an idea. Standard deviations are in parentheses. Duplicates (ideas previously displayed on the screens) were left out of the analysis. Different superscripts within each row indicate significant differences on a Tukey post hoc test (p < :001).

sequence conditions (M ¼ 2:69). Thus, although Hypothesis 4 was supported, Hypothesis 5 was not. We also compared within-category fluency in the control condition with fluency in each of the other conditions. To this end, we used Dunnett t tests (see e.g., Winer, 1971). The results showed that within-category fluency in the homogeneous/random stimuli condition was higher than in the control condition, d ¼ :84, t ¼ 4:38, p < :001, and the same was true in the homogeneous/clustered condition, d ¼ :84, t ¼ 4:37, p < :001. There were no differences between diverse/random condition and the control condition, d ¼ :03, t < 1; ns, or between the diverse/clustered condition and the control condition, d ¼ :06, t < 1; ns. Thus, regardless of stimulus sequence, homogeneous stimuli led to higher levels of within-category fluency, whereas diverse stimuli did not. These results are consistent with Hypothesis 4, but inconsistent with Hypothesis 5. There was clear evidence for cognitive stimulation and productivity gains in the homogeneous stimuli conditions (due to increased depth), but no evidence for cognitive interference. It is likely that the positive effect of homogeneous stimuli on within-category fluency was due to the fact that participants in that condition generated many ideas in the two categories to which they were exposed. To explore this possibility, we calculated the percentage of ideas in the two stimulated categories for each of the three different versions of the homogeneous stimuli condition. The percentage of ideas in these categories was indeed higher in the homogeneous stimuli conditions than in the other conditions (File 1: 38.70% versus 16.22%, v2 ð1Þ ¼ 94:57, p < :001; File 2: 55.04% versus 27.05%, v2 ð1Þ ¼ 317:57, p < :001; File 3: 30.51% versus 13.88%, v2 ð1Þ ¼ 42:52, p < :001). Thus, withincategory fluency was higher in the homogeneous stimuli conditions because participants in that condition generated many ideas in the two stimulated categories. Clustering. Hypothesis 6 predicted that stimuli offered in a random sequence would interfere with idea generation, and lead to more switching between categories and thus to lower levels of clustering. A planned comparison between the control condition (M ¼ :301) and the two random sequence conditions (M ¼ :182) confirmed this hypothesis, tð58Þ ¼ À2:76, p < :01.

Moreover, there was a tendency for clustering to be higher when stimuli were offered as a clustered sequence (M ¼ :247) than when they were offered as a random sequence, tð58Þ ¼ 1:71, p < :10. These results were consistent with Hypothesis 6. Response latencies The previous analyses confirmed that exposure to ideas has positive effects (cognitive stimulation), and that these effects are due to either surveying more categories (in the diverse stimuli conditions) or surveying categories in greater depth (in the homogeneous stimuli conditions). However, there was no evidence for cognitive interference. Although clustering was lower in the random sequence conditions than in the other conditions, this did not lead to lower levels of within-category fluency or productivity. The prediction that low levels of clustering would be associated with low levels of withincategory fluency was based on the reasoning that category changes (the next idea is from a different category) take more time than category repetitions (the next idea is from the same category), because category repetitions are based on rapid associations within a semantic domain. To evaluate this reasoning, we analyzed response latencies. The average response latencies per participant for category repetitions and category changes were computed. These latencies measured the time a participant needed to generate and enter an idea. We analyzed these latencies in a 5 (experimental condition) Â 2 (category repetition or change) mixed model ANOVA, with the former variable as a between-participants factor and the latter as a within-participants factor (see Table 2). This ANOVA revealed a significant within-participants main effect and of category repetition or change, F ð1; 54Þ ¼ 6:78, p < :05.4 Category repetitions (M ¼ 25:67 s) were faster than category changes (M ¼ 28:47). The main effect of conditions was significant as well, F ð4; 54Þ ¼ 6:44, p < :001. Tukey post hoc tests showed that overall response latencies were longer in the control condition than in all four experimental conditions. Thus, response
There were four outliers. These outliers had response latencies for category repetitions of three standard deviations or more above the means in their conditions, and thus were not included in this analysis.
4

542

B.A. Nijstad et al. / Journal of Experimental Social Psychology 38 (2002) 535–544

latencies were longer in the control condition (M ¼ 35:43) than in the homogeneous/random (M ¼ 22:34, p < :001), the homogeneous/clustered (M ¼ 24:30, p < :01), the diverse/random (M ¼ 24:75, p < :01) and the diverse/clustered conditions (M ¼ 26:47, p < :05). No differences among the experimental conditions were significant (all ps > :65). This result, of course, is consistent with the productivity findings, and with our claim that participants in the control condition generated fewer ideas because their average response latency was longer. However, these two main effects were qualified by a significant interaction, F ð4; 54Þ ¼ 9:01, p < :001. Paired associate t tests showed that only in the control condition were category repetitions faster than changes, tð13Þ ¼ À5:23, p < :001. The difference in response latencies between category repetitions and changes was not significant in the homogeneous/random, tð10Þ ¼ À1:40; ns, the homogeneous/clustered, tð10Þ ¼ À:29; ns, the diverse/random, tð11Þ ¼ :38; ns, or the diverse/clustered conditions, tð10Þ ¼ :82; ns. Thus, our prediction that semantically related ideas are generated faster than semantically unrelated ideas was confirmed only for the control condition.5 When response latencies for category repetitions and category changes were analyzed in two separate ANOVAs, we found no differences between conditions in response latencies for category repetitions, F ð4; 58Þ ¼ 1:54; ns. However, the effect of conditions on response latencies for category changes was significant, F ð4; 58Þ ¼ 13:41, p < :001. This was due to a significant difference between the control condition and all the experimental conditions (Tukey post hoc test, all ps < :001). Again, there were no other differences among conditions (see Table 2). Thus, it appears that stimulus ideas reduced response latencies for category changes, whereas response latencies for category repetitions were not affected. In the experimental conditions, the result was that category changes were as fast as category repetitions. Productivity gains in the experimental conditions thus seemed to occur because response latencies for category changes were reduced to the level for category repetitions.

Discussion In this paper, we examined the effects of idea sharing on cognitive processes and performance in an idea generation paradigm. Hypotheses were derived from a new theory of idea production, which we refer to as SIAM. SIAM conceptualizes idea generation as a twostage process, in which a cue-based knowledge activa5 Related to this finding, the correlation between clustering (as a corrected measure of the number of category repetitions) and productivity was positive in the control condition (r ¼ :35), but lower or even negative in the experimental conditions (À:26 < r < :15).

tion stage is followed by an idea generation stage within a semantic domain. Based on SIAM, we hypothesized three different effects of idea sharing on cognitive processes and performance. First, when the ideas exchanged are semantically diverse, they increase the range of accessible knowledge and allow for the generation of semantically more diverse ideas (increased breadth of production). Second, when the ideas exchanged are semantically homogeneous, they allow for the generation of many new ideas within a semantic domain (increased depth). However, we also argued that idea sharing interferes with idea production when exposure to ideas causes more switching between semantic categories and prevents rapid associations within a semantic domain. These hypotheses were tested in an idea exposure paradigm, where individuals were exposed to stimulus ideas that systematically varied in content. In a control condition, participants were not exposed to such ideas. Evidence was found for both types of cognitive stimulation. Participants who were exposed to ideas from a wide array of semantic categories surveyed more categories of ideas than either participants who were not exposed to ideas or participants who were exposed to ideas from just a few semantic categories. Participants exposed to homogeneous ideas, in contrast, surveyed the categories in greater depth. Evidence for cognitive interference, however, was mixed. As expected, exposure to a random sequence of ideas, with a lot of switching between categories of stimulus ideas, led to lower levels of semantic clustering, compared to a condition in which ideas were offered in short clusters of semantically related ideas. Although this suggests that participants were unable to follow their own Ôtrain of thoughtÕ within a category, this did not lead to lower production.6 The result was that there were productivity gains in all experimental conditions compared to the control condition. Based on SIAM, we further predicted that fast associations are possible within semantic categories, and that category repetitions would be faster than category changes. However, this was only true in the control condition, not in any of the experimental conditions. Instead, response latencies for category changes in the experimental conditions were shorter than in the control condition, and were just as fast as category repetitions. Thus, stimuli were effective because they reduced the response latencies for category changes. Because category changes were as fast as category repetitions, more

Our sample size was not particularly small (n ¼ 12 for the experimental conditions; n ¼ 15 for the control condition), so it is not likely that the lack of evidence for cognitive interference effects on our measures of productivity and within-category fluency was due to low statistical power. Moreover, a power analysis indicated that the sizes for these effects were small (< :10), and thus unlikely to be of substantive interest.

6

B.A. Nijstad et al. / Journal of Experimental Social Psychology 38 (2002) 535–544

543

switching among semantic categories (and thus fewer category repetitions) in the random sequence condition did not lead to productivity losses. In retrospect, these results are not inconsistent with SIAM. SIAM assumes that when a train of thought within a category no longer leads to new ideas, a search cue is needed to probe memory for problem-relevant knowledge. This takes some time, and generally results in a new train of thought in another semantic category (a category change). Stimulus ideas are added to the search cue relatively automatically, which reduces the time needed to produce other search cues. Because a search cue is only needed when a train of thought within a category no longer leads to new ideas (i.e., before a category change), category changes should become faster when participants are exposed to stimuli. This, of course, is exactly what we found. One issue that needs to be discussed is whether our results generalize to other tasks. We would argue that one factor that may be important in this respect is the Ôsolution spaceÕ of the idea generation topic. Topics with a relatively large solution space, with many categories of solutions and many possible ideas per category (such as our topic), are more likely to show stimulation effects than topics with a small solution space. When a problem does not have many solution categories, a stimulus idea is less likely to lead to a category that would otherwise be inaccessible. Moreover, when relatively few ideas per category are possible, the category can become ÔexhaustedÕ quickly. This will decrease opportunities to generate additional ideas within the category, and so cognitive stimulation is less likely. Thus, both the positive effect of diverse stimuli on the number of categories surveyed, and the effect of homogeneous stimuli on the number of ideas per category, may disappear when a topic with a relatively small solution space is used (see Fiore, 2001, for a discussion of problem space in group versus individual problem solving). Interestingly, this line of reasoning may also explain why Basden et al. (1997) found evidence for cognitive interference in group recall of categorized lists of words, whereas we found no such evidence. In free recall, the number of possible items is necessarily restricted to the previously learned list, whereas in idea generation there is no a priori limitation to the number of possible ideas. In free recall, both the number of categories and the number of items per category is therefore restricted, which makes cognitive stimulation unlikely to occur. Instead, any item that is presented during free recall (either by the experimenter or by another group member) reduces the number of remaining items. And when more items are mentioned, it becomes more likely that an item someone recalls has already been mentioned by another group member. This will interfere with the recall of new items, which may explain the Basden et al. (1997) findings (also see Raaijmakers & Shiffrin, 1981). In idea

generation, these processes are less important, because the number of possible ideas is often quite large. This makes it less likely that a previously offered idea is generated again, so cognitive interference is less important. In this study, we used an idea exposure paradigm to study the effects of idea sharing on cognitive processes and performance. An important question is whether results from this paradigm generalize to real interactive groups. We suggest that they do. Recent evidence has shown that idea sharing in groups can lead to an assembly bonus effect, and that these productivity gains are probably due to cognitive stimulation (e.g., Leggett Dugosh et al., 2001; Paulus & Yang, 2000). Thus, productivity gains as a consequence of access to stimulus ideas are not restricted to idea exposure paradigms, but can also be found in paradigms involving actual idea sharing among group members. The methodology of our study allowed us to go beyond previous results and show that the underlying mechanism is the increased accessibility of problem relevant knowledge, and the reduction of response latencies of category changes. Moreover, our study showed that cognitive stimulation can increase both the breadth and depth of idea generation, depending on the content of the stimulus ideas. Future research may test whether the same mechanisms operate in real interactive groups. However, a limitation of our idea exposure paradigm is that it cannot be used to assess the dynamic properties of idea sharing in groups. In interactive groups, members mutually affect each other, whereas in our paradigm there was no mutual influence. One avenue for future research would be to gain more insight into the dynamic properties of group process and performance. In general terms, group members share information or ideas, which in part affects what information is generated and subsequently shared. We suggest that research within the Ôgroup as information processorsÕ framework would benefit from attention to the dynamic qualities of group level cognitive processes. We see this as one of the most exiting areas for future research on group process and performance.

Acknowledgments This research was conducted while Bernard Nijstad was working at Utrecht University. The authors thank Pepijn Visscher for his help with data coding, and Carsten de Dreu and Daan van Knippenberg for their helpful comments on an earlier draft of this paper. This research was funded by Grant 575-31.007 of The Netherlands Organization for Scientific Research (NWO) awarded to Wolfgang Stroebe. Part of this research was presented during a conference on Group Creativity, sponsored by the National Science Founda-

544

B.A. Nijstad et al. / Journal of Experimental Social Psychology 38 (2002) 535–544 Hinsz, V. B., Tindale, R. S., & Vollrath, D. A. (1997). The emerging conceptualization of groups as information processors. Psychological Bulletin, 121, 43–64. Lamm, H., & Trommsdorff, G. (1973). Group versus individual performance on a task requiring ideational proficiency (brainstorming): A review. European Journal of Social Psychology, 3, 362–388. Larey, T. S., & Paulus, P. B. (1999). Group preference and convergent tendencies in small groups: A content analysis of group brainstorming performance. Creativity Research Journal, 12, 175–184. Leggett Dugosh, K., Paulus, P. B., Roland, E. J., & Yang, H. C. (2001). Cognitive stimulation in brainstorming. Journal of Personality and Social Psychology, 79, 722–735. Mednick, S. A. (1962). The associative basis of the creative process. Psychological Review, 69, 220–232. Mullen, B., Johnson, C., & Salas, E. (1991). Productivity loss in brainstorming groups: A meta-analytic integration. Basic and Applied Social Psychology, 12, 3–24. Nijstad, B. A. (2000). How the group affects the mind: Effects of communication in idea generating groups. Unpublished doctoral dissertation. The Netherlands: Utrecht University. Osborn, A. F. (1957). Applied Imagination (second ed.). New York: Scribner. Paulus, P. B., & Yang, H. C. (2000). Idea generation in groups: A basis for creativity in organizations. Organizational Behavior and Human Decision Processes, 82, 76–87. Raaijmakers, J. G. W., & Shiffrin, R. M. (1981). Search of associative memory. Psychological Review, 88, 93–134. Roenker, D. L., Thompson, C. P., & Brown, S. C. (1971). Comparison of measures for the estimation of clustering in free recall. Psychological Bulletin, 76, 45–48. Stroebe, W., & Diehl, M. (1994). Why groups are less effective than their members: On productivity losses in idea-generating groups. In W. Stroebe & M. Hewstone (Eds.), European review of social psychology (vol. 5, pp. 271–303). London: Wiley. Tindale, R. S., & Kameda, T. (2000). ÔSocial sharednessÕ as a unifying theme for information processing in groups. Group Processes and Intergroup Relations, 3, 123–140. Tulving, E., & Pearlstone, Z. (1966). Availability versus accessibility of information in memory for words. Journal of Verbal Learning and Verbal Behavior, 5, 381–391. Valacich, J. S., Dennis, A. R., & Connolly, T. (1994). Idea- generation in computer-based groups: A new ending to an old story. Organizational Behavior and Human Decision Processes, 57, 448–467. Winer, B. J. (1971). Statistical principles in experimental design. New York: McGraw Hill. Ziegler, R., Diehl, M., & Zijlstra, G. (2000). Idea production in nominal and virtual groups: Does computer-mediated communication improve group brainstorming? Group Processes and Intergroup Relations, 3, 141–158.

tion and the University of Texas at Arlington, Arlington, TX, April 2000.

References
Amabile, T. M. (1983). Social psychology of creativity: A componential conceptualization. Journal of Personality and Social Psychology, 45, 357–376. Basden, B. H., Basden, D. R., Bryner, S., & Thomas III, R. L. (1997). A comparison of group and individual remembering: Does collaboration disrupt retrieval strategies? Journal of Experimental Psychology: Learning, Memory, and Cognition, 23, 1176–1189. Bousfield, W. A. (1953). The occurrence of clustering in the recall of randomly arranged associates. Journal of General Psychology, 36, 67–81. Collins, B. E., & Guetzkow, H. (1964). A social psychology of group processes for decision making. New York: Wiley. Collins, A. M., & Loftus, E. F. (1975). A spreading-activation theory of semantic processing. Psychological Review, 82, 407–428. Connolly, T., Routhieux, R. L., & Schneider, S. K. (1993). On the effectiveness of group brainstorming: Test of one underlying mechanism. Small Group Research, 24, 490–503. Dennis, A. R., & Valacich, J. S. (1993). Computer brainstorms: More heads are better than one. Journal of Applied Psychology, 78, 531– 537. Dennis, A. R., Valacich, J. S., Connolly, T., & Wynne, B. E. (1996). Process structuring in electronic brainstorming. Information Systems Research, 7, 268–277. Diehl, M. (1991). Kollektive Kreativit€t: Zur Quantit€t und Qualit€t der a a a Ideenproduktion in Kleingruppen [Collective creativity: On quantity and quality of idea production in small groups]. Habilitationschrift, University of T€bingen, Germany. u Diehl, M., & Stroebe, W. (1987). Productivity loss in brainstorming groups: Toward the solution of a riddle. Journal of Personality and Social Psychology, 53, 497–509. Diehl, M., & Stroebe, W. (1991). Productivity loss in idea-generating groups: Tracking down the blocking effect. Journal of Personality and Social Psychology, 61, 392–403. Fiore, S. M. (2001). Problem space and problem solving failure: Cognitive mechanisms in the explanation of process loss in group problem solving. Unpublished doctoral dissertation. University of Pittsburgh. Gallupe, R. B., Cooper, W. H., Gris, M. L., & Bastianutti, L. M. e (1994). Blocking electronic brainstorms. Journal of Applied Psychology, 79, 77–86. Gruenewald, P. J., & Lockhead, G. R. (1980). The free recall of category examples. Journal of Experimental Psychology: Human Learning and Memory, 6, 225–240.