Cybernetics And Systems Analysis logo
Editorial Board Announcements Abstracts Authors Archive
Cybernetics And Systems Analysis
International Theoretical Science Journal
UDC 004.855:519.216
O.S. Balabanov1


1 Institute of Software Systems of the National Acedemy of Sciences of Ukraine, Kyiv, Ukraine

bas@isofts.kiev.ua

UPPER BOUND ON CORRELATION SUM FOR THREE INDICATORS
UNDER ABSENCE OF COMMON FACTOR

Abstract. We demonstrate that in linear model with three indicator variables, where each pair of indicators is associated by separate hidden pairwise factor, the sum of correlations is upper bounded. The inequality constraint violation manifests that the underlying causal structure differs from the supposed one. In the case of the inequality violation, one may suggest that there is a common cause of the three indicators, or one indicator variable causally influences the other one. The inequality constraint remains efficient under partial observability (for instance, when two indicators only are observed).

Keywords: correlation, inequality constraint, cycle with three colliders, hidden common cause, linear SEM.



FULL TEXT

REFERENCES

  1. Kang Ch., Tian J. Inequality constraints in causal models with hidden variables. Proc. of 22nd Conf. on Uncertainty in Artificial Intelligence (UAI-06). AUAI Press, 2006. P. 233–240.

  2. Steeg G.V., Galstyan A. A sequence of relaxations constraining hidden variable models. Proc. of 27th Conf. on Uncertainty in Artif. Intel. (UAI-11). AUAI Press, 2011. P. 717–726.

  3. Steudel B., Ay N. Information-theoretic inference of common ancestors. Entropy. 2015. Vol. 17. P. 2304–2327.

  4. Chen B., Tian J., Pearl J. Testable implications of linear structural equation models. Proc. 28th AAAI Conf. on Artif. Intel. (AAAI’14). QuÁbec, Canada, AAAI Press, 2014. P. 2424–2430.

  5. van Ommen T., Mooij J.M. Algebraic equivalence class selection for linear structural equation models. Proc. of 33rd Conf. on Uncertainty in Artif. Intel. (UAI-17). Vol. 1. Curran Associates, Inc., 2017. P. 763–773.

  6. Spirtes P., Richardson T. Ancestral graph Markov models. Ann. Statist. 2002. Vol. 30, N 4. P. 962–1030.

  7. Balabanov O.S. Structurally determined inequality constraints on correlations in the cycle of linear dependencies. Cybernetics and Systems Analysis. 2018. Vol. 54, N 2. P. 173–184.

  8. Gill R.D. Statistics, causality and Bell’s theorem. Statistical Science. 2014. Vol. 29, N 4. P. 512–528.

  9. Suppes P., de Barros J.A., Oas G. A collection of probabilistic hidden-variable theorems and counterexamples. In: Waves, Information and Foundations of Physics. Conf. Proceeding. Pratesi R., Ronchi E. (Eds.). Vol. 60. Bologna: Societ× Italiana Di Fisica, 1998. P. 267–291 (arXiv:quant-ph/9610010).

© 2019 Kibernetika.org. All rights reserved.