A Working ABM Example:
The Dynamic Information Sampling
Model of Group Discussion
(DISM-GD)

At the bottom of this page is a link to a simple ABM that that you can explore by varying its input parameters and observe how these changes affect its output (i.e., its predictions for real human groups).  It will run in your browser—there is no need to download anything.  Before running it, however, you should read the description that follows of the real-world phenomenon being modeled and how the model works.  A quick-start guide to running the model is also provided below.  But there is also a good deal of useful information about model, its controls, and how to interpret its output in the Model Info drop-down box that is within the model itself.  You should look at that information after you have run the model a few times.

The Target Phenomenon

When people come together as a group to discuss a problem, they may each individually hold some decision-relevant information that others in the group do not possess.  This uniquely held, unshared information stands in contrast to the shared information known to all group members.  Group discussion gives members an opportunity to exchange their unshared information, thereby helping the group as a whole become better informed and so better positioned to make a high-quality decision.

Surprisingly, however, the unshared information people hold is often relatively neglected during discussion.  Other things being equal, decision-making groups frequently discuss more of their shared than unshared information overall, and consider their shared information earlier in discussion (for detailed summaries of the empirical evidence supporting these claims, see Larson & Egan, 2020; Lu, Yuan, & McLeod, 2012; Sohrab, Waller, & Kaplan, 2015; Stasser & Titus, 2003).  These patterns are important because the content and sequencing of discussion can significantly impact the quality of the decisions that are made.
A Common (But Unnecessary) Explanation

A commonly suggested explanation for these two tendencies (i.e., to discuss more shared than unshared information overall, and to consider shared information earlier in discussion) is that they are a consequence of people’s desire to establish “common ground” at the outset of a group discussion.  According to this explanation, group members intentionally delay bringing up issues and information that others in the group might not be aware of until after they have first had a chance to discuss what they all know in common, as doing so makes for a more comfortable, pleasant, and agreeable interaction.  By itself, this would account for the tendency of shared information to be brought out earlier in discussion than unshared information.  Further, if groups were to conclude their discussions before all of their decision-relevant information was mentioned (e.g., because they feel that they have already discussed enough information to make an informed choice), then they will have discussed more of their shared than unshared information overall.

However, this explanation makes two critical assumptions. First, it assumes that members know what information they hold is also held by others (shared information) and what information is held by them alone (unshared information). Second, it assumes that members make biased choices about what to discuss and when to discuss it—that they prefer to talk about their shared information before talking about their unshared information.

But what happens if, in a given situation, one or both of these assumptions does not hold? After all, members do not always know prior to discussion what information they hold is shared and what is unshared.  And even when they do know, they may not always have a clear preference for talking about one type before the other.  The Dynamic Information Sampling Model of Group Discussion (DISM-GD) explores these questions.  DISM-GD is a reinstantiation of the model described in Larson (1997).  It predicts that even when group members make completely unbiased choices about what information to discuss and when to discuss it, the group discussion itself often will be, paradoxically, quite biased in favor of the group’s shared information.
How The Model Works

Group discussion is conceptualized in DISM-GD as a cyclic process of members recalling facts from the pool of decision-relevant information they each hold, and then speaking those facts aloud to the rest of the group.  However, because only one member need recall and mention a given fact in order to bring it to the attention of the group as a whole, the more members there are who potentially can mention that fact, the more likely it is that it will actually be discussed. An item of shared information should therefore have a higher probability of being discussed than an otherwise comparable item of unshared information, simply because more members hold—and so potentially can recall and mention—the shared item. Said differently, there are more opportunities for the group to sample an item of shared information than there are for them to sample an item of unshared information.

There are two phases to each simulated group discussion in DISM-GD. The first is a pre-discussion phase in which each group member (agent) is endowed with a memory containing a subset of the available decision-relevant information, some of which is also in the memories of the other group members (shared information), and some of which is in their memory alone (unshared information).

Groups discussion is simulated in the second phase, which is organized as a sequence of speaking turns. Prior to each turn, members sample (recall) at random one fact from among the not-yet-mentioned facts they hold, and then compete for an opportunity to “speak” that fact. It is important to note that this sampling process is completely unbiased—each piece of information that a member holds at the outset of discussion has an equal probability of being sampled at any point during discussion, regardless of its shared/unshared status, as long as it has not already been mentioned during discussion by anyone (Note: DISM-GD is designed to model only the initial entry of information into discussion, not repetitions or elaborations of that information).  A speaker is then chosen from among those agents who have been able to recall something that is relevant and has not already mentioned.  Speakers are chosen in proportion to their “contribution propensity.” (In the real world, group members seldom contribute equally to discussion; typically, some members contribute more than others.  The relative equality/inequality of member contributions is a parameter that can be set in the model.  See below). “Speaking” is represented simply by having the speaker add its recalled fact to a list of information discussed by the group. The speaker then attempts to recall another not-yet-mentioned fact, and, if successful, competes for another opportunity to speak. Members who were prepare to speak (i.e., they had successfully recalled a fact that had not already been mentioned) but who were not chosen to speak, simply compete again to speak their fact (i.e., they do not try to recall another fact until either they or someone else speaks the fact they currently are prepare to speak). This process continues until either none of the agents can recall any information that has not already been discussed, or until a pre-set limit on the maximum amount of information discussed is reached.

The model keeps track of the temporal order in which shared and unshared information was spoken, and displays graphically the proportion of groups in which shared and (separately) unshared information were mentioned in each successive speaking turn (i.e., the proportion in which shared facts were spoken during the first, second, third, etc., speaking turn.  Likewise for unshared facts). The average proportion of shared and (separately) unshared information discussed overall is also displayed.
Quick-Start Guide to Running the Model
1.
 Click here to launch the model.
 DISM-GD(Web) Image
2.
 Adjust the sliders to the desired values for GroupSize (the number of members in each simulated group), nShared (the number of shared facts each group has—every member gets all of this), and nUnshared (the number of unshared facts each group has—these are divided equally among members).
3.
 Adjust the sliders to desired values for pRecall (members' ability to recall during discussion the various facts they were given beforehand—higher values imply better recall), cRatio (the degree to which members contribute equally to discussion—higher values imply more equal contribution), and discLength (the desired length of the group discussions—defined as the average number of facts to be spoken before concluding discussion).
4.
 To run the model, click Setup, then Go.  The model will run one group discussion after another, cumulating the results as it goes.  Click Go again to stop the model.
5.
 Adjust the model's speed.  There is a slider at the top of the page to control the speed of the model.  The model will run fastest when (a) that slider is set all the way to the right and (b) the ShowHistory? checkbox is UNCHECKED.  To obtain relatively stable results, the model should be allowed to run until 1,000 or more group discussions have been simulated.  This should not take very long when the model runs as full speed.  NOTE: Moving the speed slider all the way to the right will temporarily disable the updating of the two figures.  However, when the model is stopped (by clicking Go again) the cumulative results will be displayed.  
6.
 More detailed information about the model's controls, output, and general operation can be found in the Model Info drop-down box near the bottom of the model page.