Induced Dependence, Factor Interaction, and Discriminating Between Causal Structures
Properties of a collider, i.e., a model in which two independent variables (causes or factors) exert influence on a third variable (effect) are analyzed. It is shown that, as a result of conditioning on the effect, its causes usually become mutually dependent. Induced (provoked) dependence is studied quantitatively, a close relationship between induced dependence and factor interaction is revealed, and a new index of interaction of two factors is proposed. The efficiency of applying induced dependence to the identification of orientations of links and revealing of masked links (edges) of the model is shown.
Keywordsconditional independence induced (provoked) dependence collider factor interaction masked edge identification of links orientation of links
Unable to display preview. Download preview PDF.
- 5.E. Bareinboim, J. Tian, and J. Pearl, “Recovering from selection bias in causal and statistical inference,” in: Proc. 28th AAAI Conference on Artificial Intelligence, Québec City, Québec, Canada (2014), pp. 2419–2416.Google Scholar
- 12.Ph. I. Andon and O. S. Balabanov, “On revealing a latent binary factor in categorical statistical data,” Reports of the National Academy of Sciences of Ukraine, No. 9, 37–43 (2008).Google Scholar
- 13.L. Badea, “Determining the direction of causal influence in large probabilistic networks: A constraint-based approach,” in: Proc. 16th European Conf. on Artificial Intelligence (ECAI-2004) (Valencia, Spain, 2004), IOS Press, Amsterdam (2004), pp. 263–267.Google Scholar
- 15.J. Ramsey, P. Spirtes, and J. Zhang, “Adjacency-faithfulness and conservative causal inference,” in: Proc. 22nd Conf. on Uncertainty in Artificial Intelligence, AUAI Press, Oregon (2006), pp. 401–408.Google Scholar