1.1 Why Use Laboratory Experiments?
Operations management (OM) is a field with strong tradition of analytical modeling. Most of the early analytical work in OM was primarily optimization based and dealt with central planning for such problems as jobāshop scheduling, lot sizing, and queuing. Starting in the 1980s, OM researchers became interested in modeling strategic settings that involve interactions between firms. Today, OM models tackle problems that deal with supply chain coordination, competition, and cooperation, which examine incentives and objectives of firms as well as individual decision makers. This type of work requires a model of decisionāmaking at individual and/or firm level.
Supply chains are not centralized, but consist of individual selfāinterested firms ā original equipment manufacturers (OEMs), different tiers of suppliers, transportation vendors, and retailers. These firms face uncertainty from the environment, such as production yield, processing times, and customer demand, as well as strategic uncertainty, which comes from the uncertainty about the actions of the other supply chain members. Traditionally, OM models assumed that firms are expected profit maximizers and are fully rational, meaning that they correctly anticipate the actions of the other supply chain members.
Behavioral operations management (BOM) started in order to first test, and then improve, modeling assumptions about decisionāmaking. Schweitzer and Cachon (2000) is the seminal BOM paper that tested how individuals solve the ānewsvendor problem.ā It turned out that individuals generally do not solve the problem correctly, but are rather systematic and predictable in how their decisions deviate from optimal. Schweitzer and Cachon (2000) finding, and numerous studies that followed (see Chapter 11), has major implications for OM models, because the newsvendor problem is a building block for much of the inventory theory.
BOM work lives at the boundary of analytical and behavioral disciplines. It is aimed at developing models of decisionāmaking to better explain, predict, and improve analytical models in OM. There are many empirical methods for studying human behavior in general and human judgment and decisionāmaking in particular. Laboratory experiment, the topic of this chapter, is one of the empirical methods we use in BOM. Similar methods have been employed in a number of other social science fields, including psychology and sociology (social networks), law (jury behavior), political science (coalition formation), anthropology, biology (reciprocity), and especially experimental economics, that have a long and rich tradition of studying problems that are similar to the ones of interest to the OM community.
Laboratory experiments can be designed to test analytical models in a way that gives the theory the best possible shot to work. This is done by carefully controlling the environment, especially information available to the participants, to match theoretical assumptions. Parameters can be selected in a way that treatment effects predicted by the model are large enough to be detected in the laboratory, given appropriate sample sizes and the level of general ānoiseā in human behavior. If the theory fails to survive such a test, a conclusion can be made that the model is likely to be missing some important behavioral aspect. If a theory survives such a test, we can conclude that that the model qualitatively captures enough of the behavioral factors to organize the data, and further robustness tests can be performed by manipulating parameters.
The ability to cleanly establish causality is a relative advantage of laboratory experiments, compared with other empirical methods. In the laboratory, causality is established by directly manipulating treatment variables at desired levels and randomly assigning participants to treatments. Random assignment ensures that treatment effects can be attributed to the treatment variables and not be confounded by any other, possibly unobservable, variables. Other empirical methods rely on existing field data, so neither random assignment nor direct manipulation of treatment conditions is possible, so causality cannot be directly established.
Another advantage of laboratory experiments is that they lend themselves well to being replicated by researchers in different laboratories. Replicating results is important because any single laboratory result can be an artifact of the protocols or settings in the specific laboratory.
Results that have been replicated in different contexts and by different research teams can be considered reliable. A recent article published in Science (Open Science Collaboration 2015) highlighted the importance of replicating experimental results. It reported that only 36% of psychology studies published in three important psychology journals and selected as part of a largeāscale replication project had statistically significant results when replicated. Replications done in the Science article showed that while in the original studies most reported r...
