Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable

By Felix Elwert, Christopher Winship. The Annual review of sociology, 2014

1,060 citations recorded by the source corpus.

Endogenous selection bias is a central problem for causal inference. Recognizing the problem, however, can be difficult in practice. This article introduces a purely graphical way of characterizing endogenous selection bias and of understanding its consequences (Hernán et al. 2004). We use causal graphs (direct acyclic graphs, or DAGs) to highlight that endogenous selection bias stems from conditioning (e.g., controlling, stratifying, or selecting) on a so-called collider variable, i.e., a variable that is itself caused by two other variables, one that is (or is associated with) the treatment and another that is (or is associated with) the outcome. Endogenous selection bias can result from direct conditioning on the outcome variable, a post-outcome variable, a post-treatment variable, and even a pre-treatment variable. We highlight the difference between endogenous selection bias, common-cause confounding, and overcontrol bias and discuss numerous examples from social stratification, cultural sociology, social network analysis, political sociology, social demography, and the sociology of education.

Subfields: Management Science and Operations Research, Statistics and Probability

View Paper: https://www.semanticscholar.org/paper/9fcc6fe93c9ce0c0b4773c18074c01bdf71db52d

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Management Science and Operations Research

In Statistics and Probability

Citations in the corpus

How Conditioning on Posttreatment Variables Can Ruin Your Experiment and What to Do about It (2018, background): “…T and u, which unbalances the experiment with respect to u and makes accurately estimating the causal e↵ect impossible without further assumptions (Elwert and Winship 2014).11 In our example, conditioning on political interest unbalances the treatment and control groups on SES, which in turn…”
Equalization or Selection? Reassessing the “Meritocratic Power” of a College Degree in Intergenerational Income Mobility (2019, background): “From this DAG, we can see that college graduation is not only an intervening variable in intergenerational income reproduction, but also a collider variable—a common consequence of parental income (X ) and other individual attributes (Z) (see Elwert and Winship 2014).”
Text and Causal Inference: A Review of Using Text to Remove Confounding from Causal Estimates (2020, background): “By contrast, conditioning on a collider can create dependence between treatment-collider-outcome5 (Pearl, 2009a) potentially introducing more bias (Montgomery et al., 2018; Elwert and Winship, 2014).”
Control variables and causal inference: a question of balance (2018, background): “…and there are no other confounding factors.1 More complicated structures such as having more than three independent variables or complex causal pathways will only make the issues I raise here more problematic (Elwert & Winship, 2014; Morgan & Winship, 2015; Pearl et al., 2016; Rohrer, 2018).”
The Statistics of Causal Inference: A View from Political Methodology (2015, neutral): “See Elwert and Winship (2014) for an excellent discussion of how threats to identiﬁcation of this form can arise.”
Does school spending matter? The new literature on an old question. (2018, neutral): “To complicate this approach further, adding additional controls can in some cases exacerbate any underlying biases and influence the result in unpredictable and counter-intuitive ways (see Elwert and Winship 2014, Lechner 2008).”
Cross‐Country Determinants of Early‐Stage Necessity and Opportunity‐Motivated Entrepreneurship: Accounting for Model Uncertainty (2018, neutral): “Controls can also leverage backward causal links with the outcome variable, producing reverse causation or selection bias (Elwert and Winship 2014).”
Collider scope: when selection bias can substantially influence observed associations (2016, background): “There are other ways in which ascertainment can generate biases.(31) For example, Figure 3 (panel B) shows a situation in which entry into a study is conditional upon the value of the phenotype (but not the outcome of interest) and where the phenotype does not cause the outcome, but the phenotype an”
Thinking Clearly About Correlations and Causation: Graphical Causal Models for Observational Data (2017, methodology): “Elwert and Winship (2014) provided a succinct summary of the many ways in which endogenous selection bias can arise.”
Moving Forward: Challenges and Directions for Psychopathological Network Theory and Methodology (2017, background): “Another concern when including many nodes is conditioning on a collider (Elwert & Winship, 2014).”