In The Landscape of History: How Historians Map the Past, John Gaddis discusses the methods that historians use, comparing them to methods of sciences such as astronomy, paleontology, and geology. The book is a great exploration of different approaches to examining evidence.  Here are a few key lessons.

Since we’re historians, not novelists , we’re obliged to tier our narrative as closely as possible to the evidence that has survived: that’s an inductive process.  But we have no way of knowing, until we begin looking for evidence with the purposes of our narrative in mind, how much of its going to be relevant: that’s a deductive calculation. Composing the narrative will then produce places where more research is needed, and we’re back to induction again.  But that new evidence will still have to fit within the modified narrative so we’re back to deduction.  And so on until, as I earlier quoted William H. McNeill, “it feels right, and then I write it up and ship it off to the publisher.”  That’s why the distinction between induction and deduction is largely meaningless for the historian seeking to establish causation.  The verb “to fit” which implies both procedures, is much better.  Its not just tailors who look at what they have to cover, and then what they have with which to cover it, and then back and forth, again and again, until the fit is as good as its going to get.

Biography, like the larger sphere of history within which it resides is at once a deductive and an inductive exercise.  Patterns of human behavior extending across time and space can alert us to the kinds of questions we should be asking about the particular individual we’re dealing with:  that’s where deduction comes in.

 …

But these patterns alone can’t determine the answers, for it’s all to easy to to find what you’re looking for when you’ve already decided a head of time what it is.  The evidence of particular experience in biography has got to discipline what we know from collective experience: induction is how we do that.

Theories like relativity, plate tectonics, and natural selection emphasize relationships among variables, some of them continuous and others contingent.  Regularity and randomness coexist within such theories: they allow for punctuations that upset equilibria such as asteroid impacts, earthquakes, or the outbreak of new and lethal diseases.  Nor do they require singling out certain variables as more important than others.Nor do they require singling out certain variables as more important than others: what would the independent variables be for the Andromeda galaxy or the Norwegian coastline, or the Darwin finch? Reductionism in these realms is only a stepping stone towards synthesis.

Historians,however, reject the doctrine of immaculate causation, which seems to be implied in the idea that one can identify, without reference to all that has preceded it, such a thing as an independent variable.  Causes always have antecedents.  We may rank their relative significance, but we’d think it irresponsible to seek to isolate or “tease out” single causes for complex events.  We see history as proceeding instead from multiple causes and their intersections.  Interconnections matter more to us than does the enshrinement of particular variables.