Choice of Units and the Causal Markov Condition

Jiji ZHANG, Peter SPIRTES

Research output: Book Chapters | Papers in Conference ProceedingsConference paper (refereed)Research

Abstract

Elliott Sober's well-known challenge to the principle of the common cause — and to its generalization, the causal Markov condition — appeals to the apparent positive correlation between two causally unconnected quantities: Venetian sea levels and British bread prices. In this paper we examine Kevin Hoover's and Daniel Steel's opposite evaluations of Sober's case. We argue that the difference in their assessments results from a difference in their choice of units and populations for statistical modeling. Our analysis suggests yet another diagnosis of Sober's counterexample: the failure of the causal Markov condition in the population chosen by Sober and Steel is due to the presence of causal relations that hold between the relevant properties across units. Such inter-unit causation is left unrepresented in causal models congenial to statistical analysis, because statistics does not deal with inter-unit relationships once the units are fixed. Accordingly, the causal Markov condition is formulated in terms of causal structures that depict intra-unit causal relations only. It is therefore worth highlighting a methodological principle for causal inference: the units should be so chosen that they do not interfere with each other, a principle that, fortunately, is often observed in practice.
Original languageEnglish
Title of host publicationScientific explanation and methodology of science : selected papers from the International Conference on SEMS 2012
EditorsGuichun GUO, Chuang LIU
PublisherWorld Scientific Publishing Company
Pages240-251
Number of pages12
ISBN (Electronic)9789814596640, 9789814596657
ISBN (Print)9789814596633
DOIs
Publication statusPublished - May 2014
EventThe International Conference on Scientific Explanation and Methodology of Science - Shanxi University, Taiyuan, Shanxi, China
Duration: 17 Sep 201222 Sep 2012

Conference

ConferenceThe International Conference on Scientific Explanation and Methodology of Science
CountryChina
CityTaiyuan, Shanxi
Period17/09/1222/09/12

Fingerprint

Causal Markov Condition
Causal Relation
Causal
Causal Inference
Sea Level
Steel
Statistical Analysis
Statistics
Evaluation
Causes
Counterexample
Causation
Bread
Elliott Sober
Causal Model
Modeling

Cite this

ZHANG, J., & SPIRTES, P. (2014). Choice of Units and the Causal Markov Condition. In G. GUO, & C. LIU (Eds.), Scientific explanation and methodology of science : selected papers from the International Conference on SEMS 2012 (pp. 240-251). World Scientific Publishing Company. https://doi.org/10.1142/9789814596640_0018
ZHANG, Jiji ; SPIRTES, Peter. / Choice of Units and the Causal Markov Condition. Scientific explanation and methodology of science : selected papers from the International Conference on SEMS 2012. editor / Guichun GUO ; Chuang LIU. World Scientific Publishing Company, 2014. pp. 240-251
@inproceedings{50ed6c511dc54f479e0d4d5d7d4bfa0c,
title = "Choice of Units and the Causal Markov Condition",
abstract = "Elliott Sober's well-known challenge to the principle of the common cause — and to its generalization, the causal Markov condition — appeals to the apparent positive correlation between two causally unconnected quantities: Venetian sea levels and British bread prices. In this paper we examine Kevin Hoover's and Daniel Steel's opposite evaluations of Sober's case. We argue that the difference in their assessments results from a difference in their choice of units and populations for statistical modeling. Our analysis suggests yet another diagnosis of Sober's counterexample: the failure of the causal Markov condition in the population chosen by Sober and Steel is due to the presence of causal relations that hold between the relevant properties across units. Such inter-unit causation is left unrepresented in causal models congenial to statistical analysis, because statistics does not deal with inter-unit relationships once the units are fixed. Accordingly, the causal Markov condition is formulated in terms of causal structures that depict intra-unit causal relations only. It is therefore worth highlighting a methodological principle for causal inference: the units should be so chosen that they do not interfere with each other, a principle that, fortunately, is often observed in practice.",
author = "Jiji ZHANG and Peter SPIRTES",
year = "2014",
month = "5",
doi = "10.1142/9789814596640_0018",
language = "English",
isbn = "9789814596633",
pages = "240--251",
editor = "Guichun GUO and Chuang LIU",
booktitle = "Scientific explanation and methodology of science : selected papers from the International Conference on SEMS 2012",
publisher = "World Scientific Publishing Company",

}

ZHANG, J & SPIRTES, P 2014, Choice of Units and the Causal Markov Condition. in G GUO & C LIU (eds), Scientific explanation and methodology of science : selected papers from the International Conference on SEMS 2012. World Scientific Publishing Company, pp. 240-251, The International Conference on Scientific Explanation and Methodology of Science, Taiyuan, Shanxi, China, 17/09/12. https://doi.org/10.1142/9789814596640_0018

Choice of Units and the Causal Markov Condition. / ZHANG, Jiji; SPIRTES, Peter.

Scientific explanation and methodology of science : selected papers from the International Conference on SEMS 2012. ed. / Guichun GUO; Chuang LIU. World Scientific Publishing Company, 2014. p. 240-251.

Research output: Book Chapters | Papers in Conference ProceedingsConference paper (refereed)Research

TY - GEN

T1 - Choice of Units and the Causal Markov Condition

AU - ZHANG, Jiji

AU - SPIRTES, Peter

PY - 2014/5

Y1 - 2014/5

N2 - Elliott Sober's well-known challenge to the principle of the common cause — and to its generalization, the causal Markov condition — appeals to the apparent positive correlation between two causally unconnected quantities: Venetian sea levels and British bread prices. In this paper we examine Kevin Hoover's and Daniel Steel's opposite evaluations of Sober's case. We argue that the difference in their assessments results from a difference in their choice of units and populations for statistical modeling. Our analysis suggests yet another diagnosis of Sober's counterexample: the failure of the causal Markov condition in the population chosen by Sober and Steel is due to the presence of causal relations that hold between the relevant properties across units. Such inter-unit causation is left unrepresented in causal models congenial to statistical analysis, because statistics does not deal with inter-unit relationships once the units are fixed. Accordingly, the causal Markov condition is formulated in terms of causal structures that depict intra-unit causal relations only. It is therefore worth highlighting a methodological principle for causal inference: the units should be so chosen that they do not interfere with each other, a principle that, fortunately, is often observed in practice.

AB - Elliott Sober's well-known challenge to the principle of the common cause — and to its generalization, the causal Markov condition — appeals to the apparent positive correlation between two causally unconnected quantities: Venetian sea levels and British bread prices. In this paper we examine Kevin Hoover's and Daniel Steel's opposite evaluations of Sober's case. We argue that the difference in their assessments results from a difference in their choice of units and populations for statistical modeling. Our analysis suggests yet another diagnosis of Sober's counterexample: the failure of the causal Markov condition in the population chosen by Sober and Steel is due to the presence of causal relations that hold between the relevant properties across units. Such inter-unit causation is left unrepresented in causal models congenial to statistical analysis, because statistics does not deal with inter-unit relationships once the units are fixed. Accordingly, the causal Markov condition is formulated in terms of causal structures that depict intra-unit causal relations only. It is therefore worth highlighting a methodological principle for causal inference: the units should be so chosen that they do not interfere with each other, a principle that, fortunately, is often observed in practice.

U2 - 10.1142/9789814596640_0018

DO - 10.1142/9789814596640_0018

M3 - Conference paper (refereed)

SN - 9789814596633

SP - 240

EP - 251

BT - Scientific explanation and methodology of science : selected papers from the International Conference on SEMS 2012

A2 - GUO, Guichun

A2 - LIU, Chuang

PB - World Scientific Publishing Company

ER -

ZHANG J, SPIRTES P. Choice of Units and the Causal Markov Condition. In GUO G, LIU C, editors, Scientific explanation and methodology of science : selected papers from the International Conference on SEMS 2012. World Scientific Publishing Company. 2014. p. 240-251 https://doi.org/10.1142/9789814596640_0018