“Every theory is a program of perception.” (Pierre Bourdieu)
I recently stumbled on the DoctoralWritingSIG blog (HT Julia Molinari). “What does it mean to ‘theorise’ research?” asks Cally Guerin in a recent post . This is something I happen to have a lot of ideas about, so I posted a quick comment about what I tell students and researchers: “Theory” is really just the expectations you share with your reader about your object, or at least the expectations you shared with them before you analyzed your data. In your theory section, then, you are setting your reader up for an artful disappointment. You are reminding the reader of what they expect, well aware that what you have found will challenge those expectations and therefore occasion learning.
In classical hypothesis-testing approaches the theory would serve as the basis for constructing the null. The hypotheses that are constructed are normally sought to be confirmed, not disappointed. That is, the “artful disappointment” comes from the rejection of the null, not the hypotheses. Though I don’t pretend to be an expert here, theory can, as I understand it, be used in a similar way by Bayesians (like Andrew Gelman) to construct the “prior”. In both cases, “theory” includes the empirical results of past studies, which allows to estimate effect sizes. That is, theory is not just a set of causal laws, but also some generalized initial conditions.
But a “softer” approach to theory-as-expectation can also be taken to qualitative research. The theory provides a schema of expectations. So, by announcing that you are deploying, say, Genette’s theory of narrative, you are fostering an expectation that the analysis will identify the order, frequency, and duration of events, and provide an account of the voice and mood of the telling. The theory section will be written with a presumption that the reader is familiar with Genette’s narratology, and with what past applications have found “works” (and does not work) in particular narratives. As in statistical analysis, the theory gets us to anticipate anticipate “effects” in the material to be analyzed.
Given only the theory and description of the data (in the methods section) the properly trained (“peer”) reader should be able to form a qualified opinion about what the analysis will show. That opinion should, preferably, be shown to be incomplete in the paper. But I should stress that the importance of publishing null results is becoming increasingly clear in many fields.
In fact, I need to rethink this view of nulls, priors and expectations in light of the growing “replication crisis” in social science. An important guiding insight here, which has been with me for a long time, has come from Ezra Zuckerman, who emphasizes the importance of constructing a “compelling null”. And here we have to keep in mind that what we find compelling is always changing. So, in the early days of research into so-called “priming” and “implicit bias”, the null was supposed to be that such effects did not exist. Today, however, the orthodoxy (albeit one that is somewhat besieged) is that such biases do affect our thinking. Now, the question is: how much and under what conditions?
The general point, I dare say, still holds: our theory structures our expectations of our object. And those expectations are shared, communal. Our research, likewise, should be designed to challenge those expectations, but they should probably be published even where their challenge fails and the expectations hold. I guess there’s a sort of meta-theoretical twist here: although our theories tell us what to expect, we also expect that all theories leave something to be desired. In a sense, we expect to be disappointed. Some people, and especially students, sometimes forget that and let the disappointment, when it inevitably comes, frustrate them. What has actually happened, like I say, is that they learned something.