Note: This post is based on a lecture that was given as part of OxERN’s Hillary Term Seminar Series on Tuesday, January 26th at 2pm at the Oxford Internet Institute. Click here to learn more about OxERN’s seminar series.
For anyone trying to do social scientific research, typologies and definitions are very useful constructs. For example, if you want to research political parties then having a definition will allow you to group different parties into distinctive ‘families’ (enabling comparative analysis), or to identify the defining feature of a single party, which can help justify why that party is worth studying. Definitions are endemic to the academic enterprise; pretty much every bit of social research will – implicitly or explicitly – make use of one.
Normally, when we define a political party we make a descriptive claim. This description is typically articulated as a direct statement. For example, “UKIP is a populist radical right party”. Or, in other terms, “Party A is Type X”: Or, most simply of all, “A is X”. This descriptive formulation is widely used, yet it is precisely what I am concerned by, and what I hope a probabilistic approach will resolve. So, “A is X”. What’s the problem?
The problem is that it is very unlikely that we are going to have enough data to decide with sufficient confidence what Type a party belongs to. Indeed, we might never have enough data to make such statements with any definitiveness – perhaps, given the complexity of the social world, such certainty is impossible! And maybe it would be useful to capture the fact that really existing parties are too multi-faceted to map perfectly onto pre-given typologies. Maybe it is worth embracing uncertainty and fuzziness, rather than trying to find a ‘best fit’ definition that inevitably fits pretty terribly.
So, what to do? One solution is to build an ever more complex and granular typology, with a growing number of criteria and/or categories. Though this is intuitively appealing (and what most people have done so far), it is an unwieldy and unsatisfactory solution that is hard to implement. Another option is to use a mix of definitions; parties could be defined using hybrids, and so could belong to several different types at the same time. Again, this is a very inelegant solution that is hard to apply and doesn’t resolve the underlying uncertainty about what Type a party should belong to.
My response to the problem of defining parties is to abandon the formulation “A is X” and move instead to “the probability that A is X is [value]”. Rather than just assigning A to a single X, this approach calculates the probability that A is X, as well as the probability that A is Y or A is Z… (and so on for as many types as there are in the typology).
So, if we want to define a party like UKIP we would start first with a relevant political typology, with as many categories as we think appropriate. Then we would assign probabilities to each category. Such as:
|UKIP Party Type||Probability|
These probabilities are – of course – made up (and how they would be calculated in a proper study is most certainly a topic for another time!) Nonetheless, this table illustrates the basic output of a probabilistic typology. Rather than saying that UKIP definitively is any of the categories, we say how likely it is that the party belongs to any of them. And, importantly, at no point do we shift from probabilities to certainty: we don’t say, ‘UKIP is most likely (0.55) to be Populist-Right, so let’s just call it that’ or, ‘UKIP is 0.85 likely to be a hybrid of Conservative and Populist-Right, so let’s say it is a mix of the two’. Instead, we remain with the probabilities: we don’t know what UKIP really is, and shouldn’t pretend that we do.
The main advantage of a probabilistic approach is that it highlights the uncertainty inherent to any definition, and then quantifies that uncertainty. In this way, we can – however imperfectly – start to have definitions that are commensurate with the complexity of reality. There is much to do in fleshing out this approach – the methodological challenges are significant, as are the theoretical difficulties of drawing on what is an essentially Bayesian framework. But, the applications are many; whilst here I am using this approach to understand populist/extremist groups, it could easily be applied to other types of political phenomena.
If you have any thoughts on the ideas presented here (whether practical or theoretical) email me at Bertram.firstname.lastname@example.org
Bertram Vidgem is a DPhil student at the Oxford Internet Institute researching far-right political extremism in online contexts. He is interested in how the beliefs and ideologies of extremist groups are related to how they act, both online and offline. Situated in the computational social sciences, his research project will use large quantities of unstructured textual data, collected from online platforms such as Twitter, to map contemporary extremism. Prior to studying at Oxford he completed a BA in History and Politics at the University of Warwick (2013) and an MA in Ideology and Discourse Analysis at the University of Essex (2014).