Skip to main content
Log in

Intervening on structure

  • Published:
Synthese Aims and scope Submit manuscript

Abstract

Some explanations appeal to facts about the causal structure of a system in order to shed light on a particular phenomenon; these are explanations which do more than cite the causes X and Y of some state-of-affairs Z, but rather appeal to “macro-level” causal features—for example the fact that A causes B as well as C, or perhaps that D is a strong inhibitor of E—in order to explain Z. Appeals to these kinds of “macro-level” causal features appear in a wide variety of social scientific and biological research; statements about features such as “patriarchy,” “healthcare infrastructure,” and “functioning DNA repair mechanism,” for instance, can be understood as claims about what would be different (with respect to some target phenomenon) in a system with a different causal structure. I suggest interpreting counterfactual questions involving structural features as questions about alternative parameter settings of causal models, and propose an extension of the usual interventionist framework for causal explanation which enables scientists to explore the consequences of interventions on “macro-level” structure.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Institutional subscriptions

Fig. 1

Similar content being viewed by others

Notes

  1. This relates to a point raised by Cartwright (1999, pp. 16–17) and others: policies may produce violations of faithfulness (Spirtes et al. 2000). That is because policies which engineer structural parameters on converging paths to exactly cancel make variables appear statistically independent even when they are in fact causally dependent. This can pose problems for applying causal search algorithms that rely on faithfulness as an assumption.

  2. Note that Hoover would object to my characterization here, because we use different terminology, viz., different definitions of “parameter,” “variable,” and “structure.” I will return to Hoover’s alternative view later in the paper, after my own view has been more formally spelled out. In this discussion, I use “variable” and “parameter” as is common in classical statistics and the literature on causal modeling (e.g., Pearl, Woodward, Spirtes et al.): variables are measured quantities and parameters are population-level quantities inferred or estimated from the data.

  3. If we leave the world of SEMs and instead model a system with a parameterized, discrete-valued Bayes net, then the variables can each take on a finite number of “states” and parameters are just state transition probabilities.

  4. \(\varepsilon \) is inside the function because the authors do not assume that the causal mechanism is necessarily additive in the noise variable. They also allow for causal feedback; more on that below.

  5. Note that some variables may not be causally related and some may have no causes at all among the measured variables, only error terms. This is all easy to accomodate with functions defined accordingly: functions can be independent of some (or all) of their arguments, and so we can assume every variable is an argument of every function to be maximally general.

  6. Steel’s focus in (2006) is invariance and explanation across macro/micro descriptions of social systems. He is not concerned with learning the truth-values of structure-altering counterfactual claims, so he does not provide any procedure for doing so. The examples he uses to motivate his definition of structure-altering interventions may be quite plausible, and it would be illuminating to work out a technique for predicting the outcomes of structure-altering interventions to compare with the proposal in the next section.

  7. See Bright et al. (2016). Note that some counterfactual questions may be ambiguous, especially in the context of non-linear models. For example, to ask “What would be the distribution of Y if the causal effect of \(X_1\) was twice as strong?” relative to the model \(Y = 0.4 X_1 -1.3 X_2 + 0.2 X_1X_2\), is ambiguous because I could have in mind the coefficient on \(X_1\), the coefficient on the interaction term \(X_1 X_2\), or some combination of these. Each option corresponds to a different intervention.

  8. I omit equations for distance and the other exogenous variables from the model. Of course it is quite likely that a better model is not linear and includes many more factors, but this model is for illustrative purposes only. See Bhargava et al. (2005) for an actual empirical study which partially inspired this example.

  9. I am indebted to Greg Gandenberger for raising this point.

  10. More technically, I am not assuming that researchers will have a joint probability distribution over all the variables in the expanded model which includes P, nor even that such a joint probability distribution would be well-defined in complex, non-recursive cases. My only claim is that one can always write down an expanded system of equations with moderator variables which is equivalent to the actual observed system when the policy is “off” and which is describes the counterfactual system when the policy is “on.”

  11. See Woodward (1999) comments on invariance, and also Cartwright (2003). They have in mind invariance with respect to interventions on variables, but many of the same considerations carry over to invariance with respect to interventions on structural features.

  12. The eigenvalues of the coefficient matrix of the non-recursive structural equations must have modulus less than or equal to 1.

  13. If the quantity of interest is only the post-manipulation mean of Y, and not its entire density function, then it may be prudent to simply add the mean of the relevant error variable at each step. The errors are assumed to be Gaussian with mean zero in many typical models, in which case this is trivial. However, more generally if one is interested in the distribution of Y and the error variables are not normally distributed, i.e., if there is important information in the error distribution, then simply adding the mean of \(\varepsilon \) will not suffice. When the errors are significantly non-Gaussian, then the full distribution of \(\varepsilon \) is likely relevant. Thanks to Jonathan Livengood for raising this issue.

  14. Thanks to Conor Mayo-Wilson for discussion on this point.

References

  • Aguirregabiria, V., & Ho, C. Y. (2012). A dynamic oligopoly game of the US airline industry: Estimation and policy experiments. Journal of Econometrics, 168(1), 156–173.

    Article  Google Scholar 

  • Bhargava, A., Chowdhury, S., & Singh, K. K. (2005). Healthcare infrastructure, contraceptive use and infant mortality in Uttar Pradesh, India. Economics and Human Biology, 3(3), 388–404.

    Article  Google Scholar 

  • Bishop, J. G., & Schemske, D. W. (1998). Variation in flowering phenology and its consequences for lupines colonizing Mount St. Helens. Ecology, 79(2), 534–546.

    Article  Google Scholar 

  • Bright, L. K., Malinsky, D., & Thompson, M. (2016). Causally interpreting intersectionality theory. Philosophy of Science, 83(1), 60–81.

    Article  Google Scholar 

  • Cartwright, N. (1999). Causal diversity and the Markov condition. Synthese, 121(1), 3–27.

    Article  Google Scholar 

  • Cartwright, N. (2003). Two theorems on invariance and causality. Philosophy of Science, 70(1), 203–224.

    Article  Google Scholar 

  • Casini, L., Illari, P. M., Russo, F., & Williamson, J. (2011). Models for prediction, explanation and control. Theoria. Revista de Teora, Historia y Fundamentos de la Ciencia, 26(1), 5–33.

    Google Scholar 

  • Cooley, T. F., & LeRoy, S. F. (1985). Atheoretical macroeconometrics: A critique. Journal of Monetary Economics, 16(3), 283–308.

    Article  Google Scholar 

  • Fisher, F. M. (1970). A correspondence principle for simultaneous equation models. Econometrica, 38(1), 73–92.

    Article  Google Scholar 

  • Glymour, C. (2004). Review of James Woodward, Making things happen: A theory of causal explanation. British Journal for the Philosophy of Science, 55(4), 779–790.

    Article  Google Scholar 

  • Glymour, C., & Glymour, M. R. (2014). Commentary: Race and sex are causes. Epidemiology, 25(4), 488–490.

    Article  Google Scholar 

  • Haslanger, S. (2016). What is a (social) structural explanation? Philosophical Studies, 173(1), 113–130.

    Article  Google Scholar 

  • Hoover, K. D. (2001). Causality in macroeconomics. Cambridge: Cambridge University Press.

    Book  Google Scholar 

  • Hoover, K. D. (2011). Counterfactuals and causal structure. In P. M. Illari, F. Russo, & J. Williamson (Eds.), Causality in the sciences (pp. 338–360). Oxford: Oxford University Press.

    Chapter  Google Scholar 

  • Hoover, K. D. (2012). Causal structure and hierarchies of models. Studies in History and Philosophy of Science Part C: Studies in History and Philosophy of Biological and Biomedical Sciences, 43(4), 778–786.

    Article  Google Scholar 

  • Hoover, K. D. (2013). Identity, structure, and causal representation in scientific models. In H. Chao, S. Chen, & R. Millstein (Eds.), Towards the methodological turn in the philosophy of science: mechanism and causality in biology and economics (pp. 35–60). Dordrecht: Springer.

    Google Scholar 

  • Hoover, K. D., & Sheffrin, S. M. (1992). Causation, spending, and taxes: Sand in the sandbox or tax collector for the welfare state? The American Economic Review, 82, 225–248.

    Google Scholar 

  • Jackson, F., & Pettit, P. (1992). Structural explanation in social theory. In D. Charles & K. Lennon (Eds.), Reduction, explanation and realism (pp. 97–131). Oxford: Oxford University Press.

    Google Scholar 

  • Kaufman, J. S. (2014). Commentary: Race: Ritual, regression, and reality. Epidemiology, 25(4), 485–487.

    Article  Google Scholar 

  • Lange, M. (2016). Because without cause: Non-causal explanations in science and mathematics. Oxford: Oxford University Press.

    Book  Google Scholar 

  • List, C., & Spiekermann, K. (2013). Methodological individualism and holism in political science: A reconciliation. American Political Science Review, 107(4), 629–643.

    Article  Google Scholar 

  • MacKinnon, D. P. (2008). Introduction to statistical mediation analysis. New York: Routledge.

    Google Scholar 

  • Mooij, J. M., Janzing, D., Heskes, T., & Schölkopf, B. (2011). On causal discovery with cyclic additive noise models. In Advances in neural information processing systems (pp. 639–647).

  • Mooij, J., & Heskes, T. (2013). Cyclic causal discovery from continuous equilibrium data. In Proceedings of the 29th annual conference on uncertainty in artificial intelligence (pp. 431–439). AUAI Press.

  • Morgan, S., & Winship, C. (2012). Bringing context and variability back into causal analysis. In H. Kincaid (Ed.), Oxford handbook of the philosophy of the social sciences (pp. 319–354). Oxford: Oxford University Press.

    Google Scholar 

  • Pearl, J. (2009). Causality: Models, reasoning and inference (2nd ed.). Cambridge: Cambridge University Press.

    Book  Google Scholar 

  • Perron, P. (2006). Dealing with structural breaks. Palgrave Handbook of Econometrics, 1, 278–352.

    Google Scholar 

  • Richardson, T. (1996). A discovery algorithm for directed cyclic graphs. In Proceedings of the twelfth international conference on uncertainty in artificial intelligence (pp. 454–461). Morgan Kaufmann Publishers Inc.

  • Rodrik, D. (2008). One economics, many recipes: Globalization, institutions, and economic growth. Princeton: Princeton University Press.

    Google Scholar 

  • Spirtes, P. (1995). Directed cyclic graphical representations of feedback models. In Proceedings of the eleventh conference on uncertainty in artificial intelligence (pp. 491–498). Morgan Kaufmann Publishers Inc.

  • Spirtes, P., Glymour, C. N., & Scheines, R. (2000). Causation, prediction, and search. Cambridge: MIT Press.

    Google Scholar 

  • Steel, D. (2006). Methodological individualism, explanation, and invariance. Philosophy of the Social Sciences, 36(4), 440–463.

    Article  Google Scholar 

  • Strevens, M. (2008). Depth: An account of scientific explanation. Cambridge: Harvard University Press.

    Google Scholar 

  • Tian, J., & Pearl, J. (2001). Causal discovery from changes. In Proceedings of the seventeenth conference on uncertainty in artificial intelligence (pp. 512–521). Morgan Kaufmann Publishers Inc.

  • Witz, A. (1990). Patriarchy and professions: The gendered politics of occupational closure. Sociology, 24(4), 675–690.

    Article  Google Scholar 

  • Woodward, J. (1999). Causal interpretation in systems of equations. Synthese, 121(1), 199–247.

    Article  Google Scholar 

  • Woodward, J. (2003). Making things happen: A theory of causal explanation. Oxford: Oxford University Press.

    Google Scholar 

  • Zhang, K., Zhang, J., & Schölkopf, B. (2015). Distinguishing cause from effect based on exogeneity. In R. Ramanujam (Ed.), Proceedings of the fifteenth conference on theoretical aspects of rationality and knowledge, TARK (pp. 261–271).

Download references

Acknowledgements

I would like to thank Liam Kofi Bright, David Danks, Greg Gandenberger, Clark Glymour, Kevin Hoover, Jonathan Livengood, Conor Mayo-Wilson, Jim Woodward, and two anonymous referees for comments on earlier drafts.

Author information

Authors and Affiliations

Authors

Corresponding author

Correspondence to Daniel Malinsky.

Rights and permissions

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Malinsky, D. Intervening on structure. Synthese 195, 2295–2312 (2018). https://doi.org/10.1007/s11229-017-1341-z

Download citation

  • Received:

  • Accepted:

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1007/s11229-017-1341-z

Keywords

Navigation