Promoting the drawing of inferences in collaboration: insights from two experimental studies

Collaborative problem-solving tasks often require learners to co-construct new knowledge by drawing inferences from distributed information. We investigated the impact of information distribution and instructional support on university students’ collaboration in such tasks. Two experiments using specifically designed tasks show a robust negative effect of distributed information on inferences. Detailed analyses of collaborators’ discussions in Study 1 highlight the importance of specific inference patterns. Study 2 aimed to support these patterns of collaborative reasoning in a training phase, and tested the effects of support in a test phase. Two support measures were realized: written information on task difficulties and collaboration strategies, and an inference tutoring tool providing feedback and prompts based on an online assessment of students’ collaboration. Results show that reflected experience with the training task improved subsequent unsupported collaboration. The best performance was achieved when the tutoring tool had been available during the training task. In groups, the knowledge relevant for solving a given task is often distributed across group members. In fact, complementary knowledge is seen as one reason why groups are more effective than individuals: It is generally assumed that “two heads are better than one” in solving complex problems, and thus that groups holding complementary knowledge will outperform individual problem solvers (Kerr & Tindale, 2004). In pedagogical contexts, it is expected that students will learn from collaborating on joint tasks that require them to exchange and integrate complementary knowledge or viewpoints (Aronson, Balney, Stephan, Sikes, & Snapp, 1987; Kneser & Ploetzner, 2001). An important group process in these situations is the collaborative coconstruction of new ideas or concepts through which students integrate and expand their knowledge (Jeong & Chi, 2007; Suthers, 2006; Webb & Palincsar, 1996), reaching insights and eventually problem solutions that go beyond what each individual problem solver would have been able to achieve (Rummel & Spada, 2005). However, collaborative knowledge creation is not without problems. A first challenge is the necessity to pool the unshared knowledge held by individual group members (Stasser & Titus, 1985). A large literature consistently shows that groups often come up with suboptimal decisions because they fail to take into account unshared information and instead focus primarily on the shared knowledge all group members knew from the beginning (“information pooling effect”; for a recent review, see Brodbeck, Kerschreiter, Mojzisch, & SchulzHardt, 2007). However, it has not yet been studied whether similar biases can be found in groups whose task requires them to collaboratively co-construct new knowledge, which would be more interesting from a learning sciences perspective. This is the kind of task we study in our experiments. Our interest, specifically, is in the inferences collaborators draw from both shared and unshared information during collaborative problem-solving. We use the term “inference” in a broad sense, encompassing instances were at least two pieces of information are combined and, on the background of general knowledge, transformed into a new piece of information (i.e. plausible inferences, Collins & Michalski, 1989). The main questions we ask are: How does the initial distribution of information in a group influence the inferences drawn in collaborative problem-solving? And: How can the collaborative drawing of inferences be supported? Studying Inferences from Distributed Information Three Types of Inferences In the simplest case of only two persons between whom the relevant knowledge is distributed, three types of inferences can be distinguished (Table 1): A collaborative inference results from unshared information that is distributed between collaborators, i.e. an individual person holds only an unconnected piece of information, while matching information is known only to his or her partner. Thus, this inference type can only be drawn in collaboration. An individual inference results from unshared information that is located with the same person (“undistributed”). An inference of this type can therefore be drawn individually by that person. During collaboration, partners need to inform each other about their individual inferences. A shared inference results from shared information. An inference of this type can, in principle, be drawn individually by both partners, and therefore does not require collaboration between them. Table 1: Visualization of collaborative, individual, and shared inferences (adapted from Härder & Spada, 2004). Information distribution Person A Person B Inference type unshared distributed collaborative unshared undistributed individual