To get a clearer idea of the complex interplay between our tests and the re- liance on background information, it will be helpful to look in some detail at actual applications of these tests. For this purpose, we will examine an at- tempt to find the cause of a particular phenomenon, an outbreak of what came to be known as Legionnaires’ disease. The example not only shows how causal reasoning relies on background assumptions, it has another interesting feature as well: In the process of discovering the cause of Legion- naires’ disease, the investigators were forced to abandon what was previ- ously taken to be a well-established causal generalization. In fact, until it was discarded, this false background principle gave them no end of trouble.
The story began at an otherwise boring convention:
The 58th convention of the American Legion’s Pennsylvania Department was held at the Bellevue-Stratford Hotel in Philadelphia from July 21 through 24, 1976. . . . Between July 22 and August 3, 149 of the conventioneers developed what appeared to be the same puzzling illness, characterized by fever, coughing and pneumonia. This, however, was an unusual, explosive outbreak of pneumo- nia with no apparent cause. . . . Legionnaires’ disease, as the illness was quickly named by the press, was to prove a formidable challenge to epidemiologists and laboratory investigators alike.2
Notice that at this stage the researchers begin with the assumption that they are dealing with a single illness and not a collection of similar but different illnesses. That assumption could turn out to be wrong; but, if the symptoms of the various patients are sufficiently similar, this is a natural starting as- sumption. Another reasonable starting assumption is that this illness had a single causative agent. This assumption, too, could turn out to be false, though it did not. The assumption that they were dealing with a single disease with a single cause was at least a good simplifying assumption, one to be held onto until there was good reason to give it up. In any case, we now have a clear specification of our target feature, G: the occurrence of a carefully described illness that came to be known as Legionnaires’ disease. The situation concerning it was puzzling because people had contracted a disease with symptoms much like those of pneumonia, yet they had not tested positive for any of the known agents that cause such diseases.