Graphical Causal Models

Abstract

This chapter discusses the use of directed acyclic graphs (DAGs) for causal inference in the observational social sciences. It focuses on DAGs' main uses, discusses central principles, and gives applied examples. DAGs are visual representations of qualitative causal assumptions: They encode researchers' beliefs about how the world works. Straightforward rules map these causal assumptions onto the associations and independencies in observable data. The two primary uses of DAGs are (1) determining the identifiability of causal effects from observed data and (2) deriving the testable implications of a causal model. Concepts covered in this chapter include identification, d-separation, confounding, endogenous selection, and overcontrol. Illustrative applications then demonstrate that conditioning on variables at any stage in a causal process can induce as well as remove bias, that confounding is a fundamentally causal rather than an associational concept, that conventional approaches to causal mediation analysis are often biased, and that causal inference in social networks inherently faces endogenous selection bias. The chapter discusses several graphical criteria for the identification of causal effects of single, time-point treatments (including the famous backdoor criterion), as well identification criteria for multiple, time-varying treatments.

Are Smart People Less Racist? Verbal Ability, Anti-Black Prejudice, and the Principle-Policy Paradox

“…They are also more consistent with prior research on racial policy attitudes among whites, in which concerns about policy efficacy have not emerged as an important explanatory factor (Harrison et al 2006; Kluegel and Smith 1983; Krysan 2000; Sears, Hensler, and Speer 1979; Sears, Sidanius, and Bobo 2000; Tuch and Martin 1997). Nevertheless, these results should be interpreted cautiously because they are not based on direct measurement or manipulation of group threat and because both parental and employment status are themselves affected by verbal ability, meaning that a causal interpretation of these estimates requires more stringent assumptions (Elwert 2013). …”

“…I avoid this practice because these variables are mediators, rather than confounders, for the effect of verbal ability on racial attitudes (Hodson and Busseri 2012; Kanazawa 2010), and controlling for mediators may induce bias in effect estimates due to over control of intermediate pathways and endogenous selection (Elwert 2013). Respondent education is likely both a mediator and a confounder because verbal ability simultaneously affects success in school and is also affected by schooling.…”

Reflections on Partial Least Squares Path Modeling

“…One approach that can be used to conduct local tests of PLS-PM models involves the vanishing partial correlations implied by the model (Elwert, 2013;Hayduk et al, 2003;Shipley, 2000, Pearl, 2009). To illustrate, consider the basic mediational model: A→B→C, which implies that A and C are conditionally independent given B; more formally, A ⊥ C | B.…”

simcausal R Package: Conducting Transparent and Reproducible Simulation Studies of Causal Effect Estimation with Complex Longitudinal Data

“…We also undertake an approach similar to the one described in (Elwert 2013) and use the following examples to highlight some of the differences between the non-parametric structural equation models (Pearl 2009) and the traditional linear structural equation models based on the LISREL framework (Bollen 1989). Many traditional applications of structural equation modeling are devoted to addressing the problem of the measurement in the exposure, and more precisely, to address problems in which the true exposure of interest is a latent variable, such as talent, motivation or political climate that cannot be observed directly, but that is instead measured via some noisy and correlated proxies.…”

“…As a result, an NPSEM enforces the separation of the notion of a causal “effect” from its algebraic representation in a particular parametric family (i.e., a coefficient in a linear causal model), and redefines an effect as a ‘general capacity to transmit changes among variables’ (Pearl 2010b, 2012). In particular, the NPSEM framework allows the extension of the capabilities of traditional SEM methods to problems that involve discrete variables, nonlinear dependencies, and heterogeneous treatment effects (Elwert 2013). The interventions can then be defined by replacing some of the equations in NPSEM with their intervened values, which then defines the counterfactual data.…”