Writing Armageddon

Writing Armageddon
Furious writing or writing furiously?

Wednesday, April 13, 2016

Old Ideas: A Note on Periodisation in Conflict Studies

(This is from an old exchange I had with Peter Wallensteen, unfortunately I do not have access to his answer. Pretty much a discussion on whether periodisation in conflict studies creates a selection bias.)

Or conflict studies for that matter? Good Question.

I would argue that when we periodisize the temporal domain of our theories we are essentially including another variable in our models. This is because periodization is always done on the basis of some variable and the values it takes. Thus a lot of the concerns that come with including variables in a model also apply to including periods. First of all one must choose which variable out off many possible is going to be the criterion on which periods will be demarcated. The choice of demarcation is crucial for not all possible demarcations are mutually inclusive or mutually exclusive. The first case, of mutually inclusive demarcations, can create ambiguity of results, while the second, of mutual exclusive demarcations, can lead to the muting of the effect of some important variable X that our demarcation variable leaves out.

Secondly   the same proscription that applies to selecting cases on the dependent variable, also applies to selecting a variable as the basis for periodisation that is highly correlated with our dependent variable.  Additionally, while selecting on the independent variable is not proscribed in research design, it may be advisable to avoid selecting periodisation on the basis of a variable that is highly correlated with the dependent variables of interest. The reason is that it could either muddle the clarity of the causal effect, or lead to seeing a strong relation were a weaker exists due to the presence of an effect twice in a model; in the level of analysis, and  in the demarcation variable.
So, taking into consideration the above, what should be the guidelines for choosing periodisation? Is there a variable that is expected to have an effect on both the dependent and independent variables which is not correlated with the two. This should provide the basis on which one should separate periods of the international system. Secondly one should ask if periodisation is the proper way to account for that variable, rather than some other form of research design decision.

Edited: Mark Nieman from the University of Alabama has a new article out in Conflict Management and Peace Science  that provides a way to potential resolve these issues.

1 comment:

  1. Mark Nieman(Iowa State University) gave me permission to fully post his comment here

    "I have a forthcoming article in CMPS that deals with periodisation using a Bayesian change-point approach (http://cmp.sagepub.com/.../03/24/0738894215570427.abstract). Rather than the scholar selecting arbitrary time periods, the Bayesian change-point approach essentially lets the data identify time periods/strictness of demarcation. The approach lets the scholar relax the strict demarcation by running MCMCs to identify the degree of demarcation (strict/gradual); that is, temporal regimes are considered to be latent variables and are treated probabilistically when estimating parameters. As for your selection critique, the change-point approach treats the DV as endogenous with change-points, allowing the effect of IVs to change over time."