Theory Development

9. Two Approaches

Two Approaches to Confounding or Underdetermination

The idea of mechanism (Hedström and Swedberg, 1996) has recently been suggested as an alternative account of what social scientist do and should do when they explain to the model of theory that was popularized in the middle part of the twentieth century. But the differences between the two strategies are not always apparent, and the two overlap in many cases. Nevertheless, there are some important differences, and these are also important in relation to minimally-theoretical statistical modeling, to be discussed in the next section. To understand the differences we need to return to the basic problems posed by complexity.

As we have seen, one problem, apparent at the level of individual action explanations, is underdetermination. Many possible reasons and explanations may apply to a given case of action. The same problem arises for theoretical explanations generally. More than one theory may fit with the facts. If the goal is to find, out of all the possible theories, the real causes, there are a variety of options. The standard option available to the physical sciences, to measure more precisely and to see which theory predicts more successfully, is available in some contexts, such as randomized experiments. But for the cases of concern to the rest of social science, which involves at best natural experiments or data collected in non-random settings, where causes are confounded with one another, the normal situation is this:

A variety of theories with different advantages fit the data reasonably well, and there is no single means of decisively settling the question of which is correct.

The theoretical and the mechanism strategies take somewhat different approaches to this fundamental difficulty. The simplest case is this: two variables are correlated in the data, and there are a variety of possible interpretations of the relationship. The “theory” approach is to argue for the interpretation that generalizes most successfully, and which can be made to fit with other generalizations. The mechanism approach is to further specify the causal details of the hypothesized explanation of the connection, which in turn allows for the introduction of new forms of evidence about these details.