Background. Coincidence Analysis (CNA) is a configurational comparative method of causal data analysis that was first introduced in (Baumgartner 2009a, 2009b), substantively re-worked and generalized in (Baumgartner and Ambühl 2020), and implemented in a software library of the R environment for statistical computing in (Ambühl and Baumgartner 2020). In recent years, CNA was applied in numerous studies in public health as well as in the social and political sciences. For example, Dy et al. (2020) used CNA to investigate how different implementation strategies influence patient safety culture in medical homes. Yakovchenko et al. (2020) applied the method to data on the factors affecting the uptake of innovation in the treatment of hepatitis C virus infection, while Haesebrouck (2019) drew on CNA to search for factors influencing EU member states’ participation in the military operations in Libya and against the Islamic State. In contrast to more standard methods of data analysis, which primarily quantify effect sizes, CNA belongs to a family of methods designed to group causal influence factors conjunctively (i.e. in complex bundles) and disjunctively (i.e. on alternative pathways). It is firmly rooted in a so-called regularity theory of causation and it is the only method of its kind that can process data generated by causal structures with multiple outcomes (effects), for example, causal chains.
Main goals. The development of CNA is not finished. The AdCNA project will address four remaining weaknesses and limitations of the method. First, CNA’s applicability, which is currently limited to data on a maximum of about 15 factors, shall be extended to data of significantly higher dimensionality. Second, we will develop CNA-specific inference tests to further improve the quality of the method’s output—at present, that quality is not high enough when the data have small sample sizes and high noise levels. Third, instruments will be devised for reducing model ambiguities, which are particularly severe when the data are heavily fragmented. Fourth, by applying CNA in studies on auditory hallucinations and infant mortality, we will extend the scope of CNA applications to psychology and epidemiology. Overall, CNA has proven its value in some disciplines. But to establish itself in the methodological toolbox of the special sciences, more algorithmic power and flexibility, more output reliability, and wider dissemination are needed. The AdCNA project sets out to deliver exactly that.
Collaborators on this project:
Since the late 1980ies, configurational comparative methods (CCMs) have gradually been added to the methodological toolkit in disciplines as diverse as political science, sociology, business administration, management, environmental science, evaluation science, and public health. The most prominent CCM is Qualitative Comparative Analysis (QCA) (Ragin 2008). QCA, however, is unsuited to analyze causal structures with more than one endogenous variable, e.g. structures with common causes or causal chains. To overcome that restriction, Coincidence Analysis (CNA) has been first introduced in Baumgartner (2009a, 2009b). It has meanwhile been generalized in Baumgartner & Ambuehl (2020) and is available as software package for the R environment (Ambuehl & Baumgartner 2020).
This project has three objectives. The first is to fill all remaining gaps in the methodological protocol of CNA and to complement the CNA R-package accordingly. In particular, tools for robustness tests of CNA models shall be developed. The second objective is to systematically test the inferential potential of CNA by applying it to real-life studies from varying disciplines and, thereby, to explore the applicability of CNA outside of the standard domain of CCMs. The third objective is to analyze the relationship between CNA and methods from other theoretical traditions—in particular Bayes-nets methods (cf. Spirtes et al. 2000; Pearl 2000) and regression-analytical methods (Gelman and Hill 2007). Are there substantive points of contact between these methodological traditions? Are there ways to fruitfully integrate them in multi-method studies? What are the conditions that determine what method is best suited to investigate a given phenomenon or to answer a given research question?
Collaborators on this project:
The goal of this project is to develop a Bayesian theory of constitution that identifies as constituents those spatiotemporal parts of a phenomenon whose causal roles contain the phenomenon’s causal role. By drawing on the conceptual resources of Bayesian networks, the project should pave the way for a Bayesian methodology for constitutional discovery. Collaborator: Lorenzo Casini.
In recent years, numerous non-reductive physicalists (e.g. Shapiro, Sober, Raatikainen, Menzies) have argued that, by adopting a variant of Woodward’s (2003) popular interventionist theory of causation, it becomes possible to provide empirical evidence in favor of the existence of macro-to-micro downward causation. This projects intends to show that all of these proposals are bound to fail, for it is impossible, in principle, to generate evidence for downward causation. The question as to the existence of macro-to-micro causation is of inherently pragmatic nature.