DOING EXPERIMENTS
A considerable amount of our knowledge about negotiation, mediation, and other small-group conflict processes come from laboratory experiments. Spurred on by the path-breaking experiments conducted on levels of aspiration by Siegel and Fouraker (1960) and on the development of social norms and trust by Deutsch and Krauss (1962), negotiation researchers have found this approach useful for isolating important variables.
(See Rubin and Brown, 1975, for a review of the early experimental studies on bargaining.) Examples are self-other orientations, distributive or problem-solving approaches, pre-negotiation experience, time pressure, constraints on representations, the role played by values and interests, alternative power configurations, alternatives to negotiated agreements, and task framing. With regard to mediation, we have learned about effects of different mediator approaches, the timing and functions of mediation, and the use of electronic aids in performing diagnoses and analyses of conflicts. (See Carnevale and De Dreu, 2006, for a recent review of laboratory findings.). In addition to constructing and refining theories, the experimental findings have been used in training contexts for developing negotiating skills (e.g. Druckman, 2006). Few researchers would deny that these studies have made important contributions. At the same time, few would claim that the laboratory findings encompass the complexities of many real-world conflicts. For methodologists, this is the trade-off between internal and external validity. I now turn to a discussion of this issue.Experiments have the distinct advantage of allowing researchers to infer causation with some degree of confidence. Referred to as internal validity, this is due essentially to three features of experimental design: random assignment of experimental subjects to conditions (groups, treatments), control by the experimenter over independent (or causal) variables, and replication of the experiments many times.
Random assignment assures comparability of the groups (experimental and control conditions) being compared. Control groups decrease the chances that alternative (uncontrolled) variables explain the findings. Replication allows a researcher to assess variability or to judge whether the same finding re-occurs in repeated administrations of the experiment. These features also enable a researcher to use statistical tests in order to discover relationships between independent and dependent variables and to ascertain whether the relationships are due largely to chance.On the flip side of these advantages are problems that result from the very features essential to inferring causation. Highly controlled and compressed-time situations, needed to increase confidence in causal inference, create artificial settings. The more controls instituted to reduce the plausibility of alternative explanations - or to increase internal validity - the less the experimental setting resembles other situations. The relatively homogeneous populations used in most experiments combined with contrived tasks and limited time periods pose threats to generalizability or external validity. The interesting challenge for researchers is to conduct studies in realistic settings (field or simulation) while retaining the analytical advantages of experimentation. This challenge calls for balancing the two validities in the design of studies. Examples of several balancing strategies are offered in the paragraphs to follow.
Three strategies are comparing laboratory with field data, designing realistic laboratory simulations, and conducting randomized experiments in field settings. The comparison strategy is illustrated by the work of Hopmann and Walcott (1977) and Beriker and Druckman (1996). The former investigators explored the relationship between external stress and negotiating behavior in two settings, a simulation of the Partial Nuclear Test Ban Treaty of 1963 and in the actual negotiations. Convergences were found between the experimental and field results, indicating that stresses were dysfunctional for the negotiations: high stress produced greater hostility, harder bargaining strategies, and fewer agreements than low stress.
The experimental findings showed that a high- stress condition produced significantly fewer solutions than both a low-stress and neutral condition. This kind of controlled comparison is a strength of experiments; unlike the actual talks, the simulation provided for variation in outcomes. The comparison showed that low stress does not improve outcomes, but high stress hinders attempts to reach agreements.A similar type of comparison was made in the Beriker and Druckman study of the Lausanne peace conference in 1922-1923. An additional condition created for the simulation provided insights not likely to be obtained from an analysis of the actual talks. Negotiators in a weak power-symmetry condition were more satisfied with the outcome, achieved faster resolutions, disagreed less often, and made fewer competitive statements during the discussions than those in the other conditions. These studies illustrate an advantage of experiments: they allow for exploration of new conditions not present in the setting being simulated. Further, similarities found between the simulation and field bolster support for the external validity of the laboratory simulation.
The realistic simulation strategy is illustrated by my study of situational levers. An attempt was made in that study to create “packages” of variables for each of four phases of a multilateral conference over environmental issues: pre-negotiation planning, setting-the-stage, the give-and-take, and the endgame. The variables were drawn from a well-known framework of factors that influence negotiating processes (see Sawyer and Guetzkow, 1965). Hypotheses about flexible (or inflexible) negotiating behavior were evaluated by gearing the variables in the direction of more or less flexibility. Many of the hypotheses were confirmed. More importantly, however, the study shows how complexity can be incorporated in simulations. Comparisons were also made between different kinds of role-players in the simulation, scientists, and diplomats.
Correspondences between the findings from these samples contribute further to the external validity of the experiment (see Druckman, 1993).An example of the randomized field experiment strategy is the study by McGillicuddy et al. (1987). This is a rare example of a field study that randomly assigned 36 disputant pairs and mediators to one of three experimental conditions: mediation without arbitration, mediators become arbitrators if the dispute is not settled (med/arb same), and a new party is appointed to be an arbitrator if the dispute is not settled (med/arb different). Strong differences were found between the conditions on process but not outcomes. The med/arb same condition disputants were less hostile, made more new proposals and more concessions, agreed more often with the other disputant, and were more satisfied with the outcome than disputants in the other two third- party configurations. The cooperative motivation shown by disputants in this condition may have been an attempt to discourage the mediator from arbitrating the competing claims. However, subject morality posed a threat to the internal validity of the experiment: 68% of the participants dropped out of the study before it was concluded. By showing that the distribution of dropouts was random across the three conditions, the threat to validity was not serious. This study is a good example of increasing the relevance of findings for real- world settings while retaining the analytical advantages of classical experimental designs. (See Pruitt, 2005b, for a review of field experiments on social conflict.)2