What Are Data About?

Empiricism is no longer a tenable position in the philosophy of science. As a result, it is worth examining what it was that made empiricism such an attractive position for so long and if anything valuable can be salvaged from that tradition. The debates about empiricism usually contrasted knowledge obtained from observation and experiment on the one hand with knowledge obtained from theories on the other. Because computer simulations are firmly entrenched as a third mode of pursuing scientific inquiry, one way to explore what made empiricism important is by contrasting data that is provided by experiments and observations with data that is generated by computer simulations. In doing so, my paper will indirectly address one of the original philosophical issues about computer simulations: in what ways, if at all, do computer simulations differ from scientific theories on the one hand and experiments on the other? 1 Early in the discussions, claims were made that simulations had some kind of intermediate status between theory and experiment, while also standing as sui generis methods. More recently, claims have been made that simulations can be used in place of material experiments under certain circumstances. 2 Although it is true that there are similarities between certain aspects of simulations and experiments, pointing out analogies between laboratory experiments and computer simulations, such as the ability to manipulate variables and control for confounders, does not address one of the central epistemological questions that arises once simulations are brought into the picture. This question is: Are the data produced by computer simulations different in kind and in content from experimental and observational data, and from data generated by traditional scientific or mathematical theories? If we have reasons to agree that they are different in a scientifically relevant way, that is one dimension along which simulations occupy a distinct scientific niche. Empiricists have usually treated the issue of the empirical source of data within a methodological context such as confirmation, verification, or falsification. I shall focus instead on content. One important question is this: what is it for a datum to have empirical content? 3 An answer to that question is provided in (Humphreys forthcoming); here we can make additional progress by addressing a related question: what are various kinds of data about? 1. What are Data? In order to remain as ontological neutral as possible about data, I take a datum to be the value of a variable. The term `variable’ will be used here in a way that is neutral between items such as a mathematical function that represents a property and the property itself. 4 This dual use carries with it certain dangers because the role of representations in computer simulations is crucial, but where appropriate I shall explicitly note which use is in play. The variable can be scalar and discrete valued, which covers situations in which the datum concerns a qualitative monadic property such as “is red”, or it can be vector and continuous valued, capturing relational features such as “has velocity v with respect to frame F”. Other possibilities can be accommodated. I shall not distinguish between atomic and non-atomic data because nothing that follows depends upon making that distinction. Finally, although the expression “data” often carries the force of something given, something fundamental, those connotations must be rejected. Data can be the result of processing, transformations, and interpretation, and we can and often do question the data. 5