The DOI matches its Crossref record; 120 of the 120 references it cites were found in a registry or an index; 7 of the 75 sources its claims cite were checked for support.
Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable
Open accessAbstract
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.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
60 Own interpretation · 27 No full text of the sources · 18 No basis given · 9 Sources not found · 7 Not checked yet · 4 Supported · 2 Partly supported · 1 Supported so far · accuracy not yet measured
Every source the paper cites for this claim was checked and supports it.
In the paper
“This is exactly what Gangl & Ziefle’s (2009) selectivity-corrected analysis of the motherhood wage penalty finds.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 5. EXAMPLES OF ENDOGENOUS SELECTION BIAS IN SOCIOLOGY > Conditioning on a (Post-) Outcome Collider
What the sources say
- Supportedcites Gangl & Ziefle 2009 source
The paper states that previous research tended to underestimate the motherhood wage penalty because it did not systematically account for dynamic selection into employment. Its Summary and Conclusions passage also reports that its selectivity-corrected estimates often exceed earlier estimates. (LLM, conf=0.94)
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The sources with a verdict support this claim; the rest weren’t checked yet, have no full text we could read, or their verdicts aren’t rated yet.
In the paper
“Over time, the number of cases available for analysis decreases, sometimes drastically so (e.g., Behr et al. 2005, Alderman et al. 2001).”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 5. EXAMPLES OF ENDOGENOUS SELECTION BIAS IN SOCIOLOGY > Conditioning on an Intermediate Variable
What the sources say
- Not checked — no full text of the sourcecites Behr et al. 2005 source
- Supportedcites Alderman et al. 2001 source
The Introduction reports that longitudinal datasets lose cases between survey rounds, with attrition ranging from 6% to 50%; it says there is often considerable attrition. This supports the claim that the number of cases available for analysis can decrease drastically over time. (LLM, conf=0.93)
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on an assumption: Most prospective longitudinal studies experience attrition over time.
Every source the paper cites for this claim was checked and supports it.
In the paper
“Shalizi & Thomas (2011) recently showed that latent homophily bias constitutes a previously unrecognized example of endogenous selection bias, where the social tie itself plays the role of the pre-treatment collider.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 5. EXAMPLES OF ENDOGENOUS SELECTION BIAS IN SOCIOLOGY > Conditioning on a Pre-Treatment Collider
What the sources say
- Supportedcites Shalizi & Thomas 2011 source
In “Identifying Contagion From Non-Neighbors” (page unavailable), the paper explicitly says that selecting or conditioning on the social tie Aij activates a collider. Its account of network formation before observable outcomes also supports describing the tie as pre-treatment. “Endogenous selection bias” is a paraphrase of this mechanism, not the paper’s terminology. (LLM, conf=0.91)
boosteddirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“If the analyst is willing to assume that controlling for some combination of observed pre-treatment variables is sufficient to identify the total causal effect of treatment, then it is sufficient to control for all variables that (directly or indirectly) cause treatment or outcome or both (VanderWeele & Shpitser 2011).”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 5. EXAMPLES OF ENDOGENOUS SELECTION BIAS IN SOCIOLOGY > Conditioning on a Pre-Treatment Collider
What the sources say
- Supportedcites VanderWeele & Shpitser 2011 source
Theorem 1 (Section 3) states that if any subset of measured pretreatment covariates blocks all backdoor paths from treatment to outcome, the subset comprising all observed causes of treatment or outcome does too. The Introduction states the same conditional result for confounding control. (LLM, conf=0.95)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- The authors state this as an assumption: The analyst assumes that controlling for some combination of observed pre-treatment variables is sufficient to identify the total causal effect of treatment.
- Supported
A claim is checked only against the cited paper’s full text, never its abstract, and none is available for its sources.
In the paper
“Surveys of sample selection bias in sociology tend to emphasize examples where selection is affected by the treatment or the outcome, i.e., where sample selection occurs after treatment”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 1. INTRODUCTION
What the sources say
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“all nonparametric identification problems can be classified as one of three underlying problems: overcontrol bias, confounding bias, and endogenous selection bias.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 1. INTRODUCTION
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: unclear — It is unclear why the paper says it holds.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Identification analysis requires a theory (i.e., a model or assumptions) of how the data were generated.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 2. IDENTIFICATION VERSUS ESTIMATION
plain- Asserted: passes — The authors put it forward as true.
- Checkable: unclear — It is unclear whether it could be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Identification (assuming ideal data) is a necessary precondition for unbiased estimation (with real data).”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 2. IDENTIFICATION VERSUS ESTIMATION
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: unclear — It is unclear why the paper says it holds.
In the paper
“there is, at present, no universal solution to estimating causal effects in the presence of endogenous selection”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 2. IDENTIFICATION VERSUS ESTIMATION
What the sources say
- Not checked — no full text of the sourcecites Stolzenberg & Relles 1997 source
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
In the paper
“the analyst’s theory of data generation can never be fully tested empirically”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 2. IDENTIFICATION VERSUS ESTIMATION
What the sources say
- Not matchedcites Robins & Wasserman 1999
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: unclear — It is unclear whether it could be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Not matched
- Probably not a substantial claim. 27
- It may state two separate claims as one. 34
- The quote only partly states this claim. 1
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“One solution to this problem is to model tie formation (and dissolution) explicitly (Shalizi & Thomas 2011). In the model of Figure 12, this can be accomplished by measuring and conditioning on the tie-forming characteristics of individual i, Ui, or individual j, Uj, or both.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 5. EXAMPLES OF ENDOGENOUS SELECTION BIAS IN SOCIOLOGY > Conditioning on a Pre-Treatment Collider
What the sources say
- Partly supportedcites Shalizi & Thomas 2011 sourceExperimental — accuracy not yet measured
In “Contagion Effects Are Nonparametrically Unidentifiable” (page unavailable), the paper says that making the traits influencing tie formation observable can remove the confounding path, supporting the proposed conditioning on tie-forming characteristics. But it does not present explicitly modeling tie formation and dissolution alone as a solution: it says allowing ties to vary over time does not help, and describes successful explicit models as also requiring strong assumptions and adequate observed covariates. (LLM, conf=0.84)
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Greenland (2003), in particular, has derived useful analytic results and approximations for the relative size of endogenous selection bias and confounding bias in DAGs with binary variables (see also Kaufman et al. 2009).”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 6. DISCUSSION AND CONCLUSION
What the sources say
- Not checked — no full text of the sourcecites Greenland 2003 source
- Partly supportedcites Kaufman et al. 2009 sourceExperimental — accuracy not yet measured
Section 2.2 describes analytic bounds for causal risk differences in DAGs with binary variables, supporting the general connection made by “see also Kaufman et al. 2009” (page unavailable). The paper does not report approximations comparing the relative size of endogenous selection bias and confounding bias; the searched passages address causal-effect bounds instead. (LLM, conf=0.84)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
Every source the paper cites for this claim was checked and supports it.
In the paper
“In short, there exists no set of pre-treatment controls that is sufficient for identifying the causal effect of T on Y, and hence the VanderWeele & Shpitser (2011) rule does not apply.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 5. EXAMPLES OF ENDOGENOUS SELECTION BIAS IN SOCIOLOGY > Conditioning on a Pre-Treatment Collider
What the sources say
- Supportedcites VanderWeele & Shpitser 2011 source
Theorem 1 (Section 3) guarantees that the selected controls suffice only if some subset of measured pretreatment covariates blocks all backdoor paths. Section 4 explicitly says that when no subset suffices, the controls selected by the criterion will not suffice either. Thus the rule's identification guarantee does not apply to the situation described; the source does not itself assess Figure 13c. (LLM, conf=0.92)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- Supported
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“First, bias can also arise if sample selection occurs prior to treatment.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 1. INTRODUCTION
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
The quote only partly states this claim.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Missing arrows are essential for inferring causal information from data.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 3. DAGs IN BRIEF3
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
None of the sources cited for this claim have been checked yet.
In the paper
“Confounding originates from common causes, whereas endogenous selection originates from common outcomes.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 1. INTRODUCTION
What the sources say
- Not checked — no full text of the sourcecites Pearl 1995 source
- Not matchedcites Greenland et al. 1999a
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Not matchedcites Hernán et al. 2002, 2004
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Collider variables may occur before or after the treatment and before or after the outcome.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 1. INTRODUCTION
hedged- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“We highlight the difference between endogenous selection bias, common-cause confounding, and overcontrol bias”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > Abstract
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
In the paper
“Endogenous selection can result from conditioning on an outcome, but it can also result from conditioning on variables that have been affected by the outcome, variables that temporally precede the outcome, and even variables that temporally precede the treatment (Greenland 2003).”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 1. INTRODUCTION
What the sources say
- Not checked — no full text of the sourcecites Greenland 2003 source
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
None of the sources cited for this claim have been checked yet.
In the paper
“a generalization of conventional linear path diagrams (Wright 1934, Blalock 1964, Duncan 1966, Bollen 1989) that is entirely nonparametric”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 1. INTRODUCTION
What the sources say
- Not matchedcites Pearl 1995, 2009
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Not checked — no full text of the sourcecites Elwert 2013 source
- Not matchedcites Wright 1934, Blalock 1964, Duncan 1966, Bollen 1989
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may state two separate claims as one.
- Not matched
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Recognizing the problem, however, can be difficult in practice.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > Abstract
plain- Asserted: passes — The authors put it forward as true.
- Checkable: fails — It may be too vague to check against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. It may be too vague to check against evidence. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
In the paper
“Endogenous selection bias results from conditioning on an endogenous variable that is caused by two other variables, one that is (or is associated with) the treatment and one that is (or is associated with) the outcome”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 1. INTRODUCTION
What the sources say
- Not matchedcites Hernán et al. 2002, 2004
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
- Not matched
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“missing arrows represent the sharp assumption of no causal effect between variables.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 3. DAGs IN BRIEF3
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
In the paper
“DAGs encode the analyst’s qualitative causal assumptions about the data-generating process in the population”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 3. DAGs IN BRIEF3
What the sources say
- Not checked — no full text of the sourcecites Pearl 2009 source
plain- Asserted: passes — The authors put it forward as true.
- Checkable: fails — It may be too vague to check against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may be too vague to check against evidence.
In the paper
“Sample selection bias occurs when the selection of observations into a sample is not independent of the outcome”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 1. INTRODUCTION
What the sources say
- Not checked — no full text of the sourcecites Winship & Mare 1992 source
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
None of the sources cited for this claim have been checked yet.
In the paper
“Confounding bias results from failing to control for a common cause variable that one should have controlled for, whereas endogenous selection bias results from controlling for a common outcome variable that one should not have controlled for.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 1. INTRODUCTION
What the sources say
- Not checked — no full text of the sourcecites Pearl 1995 source
- Not matchedcites Greenland et al. 1999a
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Not matchedcites Hernán et al. 2002, 2004
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Recognizing endogenous selection bias can be difficult, especially when sample selection occurs prior to treatment or when no selection (in the sense of discarding observations) takes place.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 1. INTRODUCTION
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Because class size is randomized, T and Y share no common cause, and the total effect of treatment, T, on the outcome, Y, is identified by their marginal association.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 5. EXAMPLES OF ENDOGENOUS SELECTION BIAS IN SOCIOLOGY > Conditioning on an Intermediate Variable
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“If this assumption is correct, then analyzing complete cases only, i.e., conditioning on not having attrited, is entirely unproblematic; the causal effect P→D is identified without any adjustments for attrition.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 5. EXAMPLES OF ENDOGENOUS SELECTION BIAS IN SOCIOLOGY > Conditioning on an Intermediate Variable
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- The authors state this as an assumption: In Figure 9a, attrition is completely random with respect to both treatment and outcome.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“overcontrol bias, confounding bias, and endogenous selection bias are distinct phenomena: They arise from different causal structures, and they call for different remedies.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 4. SOURCES OF ASSOCIATION: CAUSATION, CONFOUNDING, AND ENDOGENOUS SELECTION > The Building Blocks: Causation, Confounding, and Endogenous Selection
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
None of the sources cited for this claim could be matched to a paper in Crossref, so there is nothing to check it against.
In the paper
“these three configurations correspond exactly to the sources of causation, confounding bias, and endogenous selection bias.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 4. SOURCES OF ASSOCIATION: CAUSATION, CONFOUNDING, AND ENDOGENOUS SELECTION > The Building Blocks: Causation, Confounding, and Endogenous Selection
What the sources say
- Not matchedcites Pearl 1988, 2009
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Not matchedcites Verma & Pearl 1988
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- Not matched
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“conditioning on a noncollider blocks the flow of association along a path, whereas conditioning on a collider opens the flow of association.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 4. SOURCES OF ASSOCIATION: CAUSATION, CONFOUNDING, AND ENDOGENOUS SELECTION > Identification in General DAGs
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Nevertheless, under the assumptions of Figure 10b, controlling for Q will remove at least some of the confounding bias exerted by U and not introduce any new biases.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 5. EXAMPLES OF ENDOGENOUS SELECTION BIAS IN SOCIOLOGY > Conditioning on an Intermediate Variable
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“outright selection on an outcome, or selection on a variable affected by the outcome, is always potentially problematic.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 5. EXAMPLES OF ENDOGENOUS SELECTION BIAS IN SOCIOLOGY > Conditioning on a (Post-) Outcome Collider
hedgedconditional- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“If, however, there are unmeasured factors, such as marital distress, U, that jointly affect attrition and divorce, as in Figure 9c, then C becomes a collider along the noncausal path between poverty and divorce, P→C←U→D, and conditioning on C will unblock this path and distort the association between poverty and divorce.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 5. EXAMPLES OF ENDOGENOUS SELECTION BIAS IN SOCIOLOGY > Conditioning on an Intermediate Variable
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
In the paper
“Furthermore, they found that although controlling for additional pre-treatment variables generally reduced bias, sometimes including additional controls increased bias.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 5. EXAMPLES OF ENDOGENOUS SELECTION BIAS IN SOCIOLOGY > Conditioning on a Pre-Treatment Collider
What the sources say
- Not checked — no full text of the sourcecites Steiner et al. 2010 source
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Figure 10a starts by positing the usual assumption that innate ability, U, confounds the effect of schooling on wages via the unblocked noncausal path S←U→W. Because true innate ability is unobserved, the path cannot be blocked, and the effect S→Y is not identified.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 5. EXAMPLES OF ENDOGENOUS SELECTION BIAS IN SOCIOLOGY > Conditioning on an Intermediate Variable
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- The authors state this as an assumption: Innate ability confounds the effect of schooling on wages.
- Rests on an assumption: True innate ability is unobserved.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“We hypothesize that the finding is due to endogenous selection bias.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 5. EXAMPLES OF ENDOGENOUS SELECTION BIAS IN SOCIOLOGY > Conditioning on a (Post-) Outcome Collider
hedgeddirect- Asserted: fails — The authors may not put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The authors may not put it forward as true.
- The authors state this as an assumption: In the music study example, additional explanatory variables are not considered.
- Rests on an assumption: In the music example, topping the Billboard charts has a positive effect on inclusion in the Rolling Stone 500 in the population of all albums.
- The authors state this as an assumption: In the music example, there is no common-cause confounding.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In Figure 9b, poverty is assumed to affect the risk of dropping out, P→C. Nevertheless, the causal effect of P→D remains identified. As in Figure 9a, conditioning on C in Figure 9b does not lead to bias because conditioning on C does not open any noncausal paths between P and D.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 5. EXAMPLES OF ENDOGENOUS SELECTION BIAS IN SOCIOLOGY > Conditioning on an Intermediate Variable
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- The authors state this as an assumption: In Figure 9b, poverty affects the risk of dropping out.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“If the analyst restricts the analysis to employed women, he will thus condition on the collider E, unblock the non-causal path from motherhood to offer wages, M→WR→E←WO, and induce endogenous selection bias.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 5. EXAMPLES OF ENDOGENOUS SELECTION BIAS IN SOCIOLOGY > Conditioning on a (Post-) Outcome Collider
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- Rests on an assumption: Many social science data sets include information on motherhood but not on women's reservation wage.
- Rests on an assumption: Data typically include information on offer wages only for women who are employed.
None of the sources cited for this claim could be matched to a paper in Crossref, so there is nothing to check it against.
In the paper
“does not transmit association and is said to be blocked, closed, or d-separated if the path contains a noncollider, C, that has been conditioned on”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 4. SOURCES OF ASSOCIATION: CAUSATION, CONFOUNDING, AND ENDOGENOUS SELECTION > Identification in General DAGs
What the sources say
- Not matchedcites Pearl 1988, 1995
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Not matchedcites Verma & Pearl 1988
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- Not matched
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“It is the source of informative censoring and attrition bias in longitudinal studies, and it is a central problem of mediation analysis in the search for social mechanisms and the estimation of direct and indirect effects.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 5. EXAMPLES OF ENDOGENOUS SELECTION BIAS IN SOCIOLOGY > Conditioning on an Intermediate Variable
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Note that this second—but not the first—endogenous selection problem persists even if the noncollider WR is measured and controlled.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 5. EXAMPLES OF ENDOGENOUS SELECTION BIAS IN SOCIOLOGY > Conditioning on a (Post-) Outcome Collider
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Either practice can lead to endogenous selection bias.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 5. EXAMPLES OF ENDOGENOUS SELECTION BIAS IN SOCIOLOGY > Conditioning on an Intermediate Variable
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“To the extent that U strongly affects Q, Q is a valid proxy for U. But to the extent that Q does not perfectly measure U, the effect of S on W remains partially confounded.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 5. EXAMPLES OF ENDOGENOUS SELECTION BIAS IN SOCIOLOGY > Conditioning on an Intermediate Variable
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The odds ratio from the logistic regression of R on B in this selected sample would then be downward biased compared with the true population odds ratio and possibly flip directions, ORS = aSd/bc < ad/bc = ORP.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 5. EXAMPLES OF ENDOGENOUS SELECTION BIAS IN SOCIOLOGY > Conditioning on a (Post-) Outcome Collider
hedgeddirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- The authors state this as an assumption: In the music example, there is no common-cause confounding.
- Rests on an assumption: Most albums ever released neither topped the charts nor won inclusion in the Rolling Stone 500.
None of the sources cited for this claim have been checked yet.
In the paper
“One class of pre-treatment variables that analysts should approach with caution is pretreatment colliders (Pearl 1995, Greenland et al. 1999b, Hernán et al. 2002, Greenland 2003, Hernán et al. 2004, Elwert 2013).”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 5. EXAMPLES OF ENDOGENOUS SELECTION BIAS IN SOCIOLOGY > Conditioning on a Pre-Treatment Collider
What the sources say
- Not checked — no full text of the sourcecites Pearl 1995 source
- Not matchedcites Greenland et al. 1999b
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Not matchedcites Hernán et al. 2002
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: fails — It may be too vague to check against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may be too vague to check against evidence.
None of the sources cited for this claim have been checked yet.
In the paper
“It is well known that outright selection on the outcome, as well as conditioning on a variable affected by the outcome, can lead to bias”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 5. EXAMPLES OF ENDOGENOUS SELECTION BIAS IN SOCIOLOGY > Conditioning on a (Post-) Outcome Collider
What the sources say
boosteddirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Using DAGs, sociologists can formally analyze the consequences of their theoretical assumptions and determine whether causal inference is possible.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 6. DISCUSSION AND CONCLUSION
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: fails — It may be too vague to check against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may be too vague to check against evidence. It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Here, X is both a confounder (on the noncausal paths T←X←U1→Y and T←X←U2→Y) and a collider (on the noncausal path T←U2→X←U1→Y). Controlling for X would remove confounding but induce endogenous selection, and not controlling for X would do the opposite.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 5. EXAMPLES OF ENDOGENOUS SELECTION BIAS IN SOCIOLOGY > Conditioning on a Pre-Treatment Collider
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“By contrast, in Figure 13b, X is a collider that blocks the noncausal path between treatment and outcome, T←U2→X←U1→Y. Conditioning on X would open this noncausal path and induce endogenous selection bias.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 5. EXAMPLES OF ENDOGENOUS SELECTION BIAS IN SOCIOLOGY > Conditioning on a Pre-Treatment Collider
hedgeddirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
In the paper
“The most general advice to be derived from the first two groups—post-outcome and intermediate collider examples—is straightforward: When estimating total causal effects, avoid conditioning on post-treatment variables (Rosenbaum 1984).”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 6. DISCUSSION AND CONCLUSION
What the sources say
- Not checked — no full text of the sourcecites Rosenbaum 1984 source
plain- Asserted: passes — The authors put it forward as true.
- Checkable: fails — It may be too vague to check against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may be too vague to check against evidence.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In Figure 13a, X is indeed a common-cause confounder as it sits on an open noncausal path between treatment and outcome, T←X→Y. Conditioning on X will block this noncausal path and identify the causal effect of T on Y.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 5. EXAMPLES OF ENDOGENOUS SELECTION BIAS IN SOCIOLOGY > Conditioning on a Pre-Treatment Collider
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
In the paper
“Hudson et al. (2008) apply these results productively in a study of familial coaggregation of diseases.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 6. DISCUSSION AND CONCLUSION
What the sources say
- Not checked — no full text of the sourcecites Hudson et al. 2008 source
boosted- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Of course, this advice can be hard to follow in practice when conditioning is implicit in the data-collection scheme or the result of selective nonresponse and attrition.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 6. DISCUSSION AND CONCLUSION
plain- Asserted: passes — The authors put it forward as true.
- Checkable: fails — It may be too vague to check against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may be too vague to check against evidence. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“We caution, however, that these rules of thumb for binary-variable DAGs do not necessarily generalize to DAGs with differently distributed variables.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 6. DISCUSSION AND CONCLUSION
hedged- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Given that assumption-free causal inference is impossible, sociologists must rely on theory, prudent substantive judgment, and well-founded prior knowledge to assess the threat of bias.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 6. DISCUSSION AND CONCLUSION
plain- Asserted: passes — The authors put it forward as true.
- Checkable: fails — It may be too vague to check against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may be too vague to check against evidence.
- Rests on an assumption: Assumption-free causal inference is impossible.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In this review, we have argued that understanding endogenous selection bias is central for understanding the identification of causal effects in sociology.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 6. DISCUSSION AND CONCLUSION
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: fails — It may be too vague to check against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may be too vague to check against evidence.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The virtue of this advice is that it prevents the analyst from conditioning on a pretreatment collider.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 5. EXAMPLES OF ENDOGENOUS SELECTION BIAS IN SOCIOLOGY > Conditioning on a Pre-Treatment Collider
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Such analyses will strengthen confidence in a causal claim when it is warranted and explicitly indicate sources of potential bias— including endogenous selection bias—when a causal claim is not appropriate.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 6. DISCUSSION AND CONCLUSION
hedgeddirect- Asserted: passes — The authors put it forward as true.
- Checkable: fails — It may be too vague to check against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may be too vague to check against evidence. It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“It is possible that these increases in bias resulted from controlling for pre-treatment colliders.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 5. EXAMPLES OF ENDOGENOUS SELECTION BIAS IN SOCIOLOGY > Conditioning on a Pre-Treatment Collider
hedgedconditional- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
In the paper
“Interrogating the simple steps necessary for separating causal from noncausal associations, we next saw that endogenous selection bias is logically on par with confounding bias and overcontrol bias in the sense that all nonparametric identification problems can be reduced to confounding (i.e., not conditioning on a common cause), overcontrol (i.e., conditioning on a variable on the causal pathway”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 6. DISCUSSION AND CONCLUSION
What the sources say
- Not matchedcites Verma & Pearl 1988
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- Not matched
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In both models, the pre-treatment variable X meets the commonsense, purely associational, definition of confounding: (a) X temporally precedes treatment, T; (b) X is associated with the treatment; and (c) X is associated with the outcome, Y.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 5. EXAMPLES OF ENDOGENOUS SELECTION BIAS IN SOCIOLOGY > Conditioning on a Pre-Treatment Collider
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may state two separate claims as one.
In the paper
“Furthermore, if the pre-treatment variable is both a collider and a confounder (as in Figure 13c), then conditioning on it may remove more bias than it induces, even though nonparametric point identification is still not feasible.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 6. DISCUSSION AND CONCLUSION
What the sources say
- Not checked — no full text of the sourcecites Greenland 2003 source
hedgedconditional- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“If it is known that the pre-treatment collider is not also a confounder, then the solution is not to condition on it when possible”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 6. DISCUSSION AND CONCLUSION
plain- Asserted: passes — The authors put it forward as true.
- Checkable: fails — It may be too vague to check against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may be too vague to check against evidence.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“(this is not possible in social network analysis, where conditioning on the pretreatment social tie is implicit in the research question).”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 6. DISCUSSION AND CONCLUSION
plain- Asserted: passes — The authors put it forward as true.
- Checkable: fails — It may be too vague to check against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. It may be too vague to check against evidence. The paper may give no reason why it holds.
In the paper
“Second, conditioning on an intermediate collider may lead to bias that is larger than the effects pointing into the collider, although the bias will usually be smaller than those effects if the intermediate collider does not itself exert a causal effect on the outcome.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 6. DISCUSSION AND CONCLUSION
What the sources say
- Not checked — no full text of the sourcecites Greenland 2003 source
hedgedconditional- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“We believe that DAGs provide such a language because they focus the analyst’s attention on what matters for nonparametric identification of causal effects: qualitative assumptions about the causal relationships between variables.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 6. DISCUSSION AND CONCLUSION
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: fails — It may be too vague to check against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may be too vague to check against evidence.
The authors state this as an assumption.
The DAGs in this review display all observed and unobserved common causes in the process under investigation.
“we assume that the DAGs in this review display all observed and unobserved common causes in the process under investigation.”
The authors state this as an assumption.
In the Hollywood example, beauty and talent are unassociated in the general population.
“Suppose, for argument’s sake, that beauty and talent are unassociated in the general population”
The authors state this as an assumption.
In the Hollywood example, beauty and talent are separately sufficient for becoming a successful Hollywood actor.
“Suppose further that beauty and talent are separately sufficient for becoming a successful Hollywood actor.”
The authors state this as an assumption.
In the Hausman and Wise example, education and the error term U are independent in the population.
“Assume, for clarity and without loss of generality, that E and U are independent in the population”
The authors state this as an assumption.
In the Lin et al. example, income and child support payments are treated as sharing no common causes.
“Without loss of generality, we neglect that I and P may share common causes.”
The authors state this as an assumption.
In the music study example, additional explanatory variables are not considered.
“For clarity, we do not consider additional explanatory variables in the study.”
The paper relies on this without support.
In the music example, topping the Billboard charts has a positive effect on inclusion in the Rolling Stone 500 in the population of all albums.
“The treatment is topping the Billboard charts, B, which presumably has a positive effect on R in the population of all albums.”
The authors state this as an assumption.
In the music example, there is no common-cause confounding.
“In the absence of common-cause confounding (as encoded in Figure 7)”
The paper relies on this without support.
Most albums ever released neither topped the charts nor won inclusion in the Rolling Stone 500.
“most albums ever released neither topped the charts nor won inclusion in the Rolling Stone 500.”
The authors state this as an assumption.
Motherhood affects a woman's reservation wage.
“Following standard labor market theory, assume that motherhood will affect a woman’s reservation wage, WR, i.e., the wage that would be necessary to draw her out of the home and into the workforce.”
The paper relies on this without support.
A woman accepts employment only if the wage offer meets or exceeds her reservation wage.
“Employment, E, is a function of both the wage offer and the reservation wage because a woman will only accept the job if the wage offer meets or exceeds her reservation wage.”
The authors state this as an assumption.
Motherhood and offer wages share no common cause.
“(For simplicity, assume that M and WO share no common cause.)”
The authors state this as an assumption.
In Figure 9a, attrition is completely random with respect to both treatment and outcome.
“attrition is assumed to be completely random with respect to both treatment and outcome.”
The authors state this as an assumption.
In Figure 9b, poverty affects the risk of dropping out.
“In Figure 9b, poverty is assumed to affect the risk of dropping out, P→C.”
The paper relies on this without support.
Many social science data sets include information on motherhood but not on women's reservation wage.
“Many social science data sets include information on motherhood, M, but they typically do not include information on women’s reservation wage, WR.”
The paper relies on this without support.
Data typically include information on offer wages only for women who are employed.
“Importantly, the data typically include information on offer wages, WO, only for those women who are actually employed.”
The paper relies on this without support.
Most prospective longitudinal studies experience attrition over time.
“Most prospective longitudinal (panel) studies experience attrition over time.”
The paper relies on this without support.
Schooling is not randomly assigned, and individuals with higher innate ability are more likely to obtain more schooling and favorable later outcomes regardless of schooling.
“Schooling, of course, is not randomly assigned to students: On average, individuals with higher innate ability are both more likely to obtain more schooling and more likely to achieve favorable outcomes later in life regardless of their level of schooling.”
The authors state this as an assumption.
Innate ability confounds the effect of schooling on wages.
“Figure 10a starts by positing the usual assumption that innate ability, U, confounds the effect of schooling on wages via the unblocked noncausal path S←U→W.”
The paper relies on this without support.
True innate ability is unobserved.
“Because true innate ability is unobserved, the path cannot be blocked, and the effect S→Y is not identified.”
The authors state this as an assumption.
Civic engagement does not spread between friends.
“For expositional clarity, assume that civic engagement does not spread between friends [i.e., no arrow Yi(t)→Yj(t+1)].”
The authors state this as an assumption.
Each person's civic engagement is caused by their own characteristics, which are at least partially unobserved.
“Instead, each person’s civic engagement is caused by their own characteristics, U, such as altruism, which are at least partially unobserved [giving arrows Yi(t)←Ui→Yi(t+1) and Yj(t)←Uj→Yj(t+1)].”
The authors state this as an assumption.
Tie formation is homophilous, such that altruistic people form preferential attachments.
“Tie formation is homophilous, such that altruistic people form preferential attachments, Ui→Fi,j←Uj.”
The authors state this as an assumption.
The analyst assumes that controlling for some combination of observed pre-treatment variables is sufficient to identify the total causal effect of treatment.
“If the analyst is willing to assume that controlling for some combination of observed pre-treatment variables is sufficient to identify the total causal effect of treatment, then it is sufficient to control for all variables that (directly or indirectly) cause treatment or outcome or both (VanderWeele & Shpitser 2011).”
The authors state this as an assumption.
There exists some sufficient set of observed controls.
“Nevertheless, the advice is still conditional on some theory of data generation, however coarsely articulated, specifically that there exists some sufficient set of observed controls.”
The paper relies on this without support.
Assumption-free causal inference is impossible.
“Given that assumption-free causal inference is impossible, sociologists must rely on theory, prudent substantive judgment, and well-founded prior knowledge to assess the threat of bias.”
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In the citing paper
“The direct line between two variables in a network graph appears to suggest a direct (causal) relationship between them, but there are several other possibilities:”
From Using person-specific networks in psychotherapy: challenges, limitations, and how we could use them anyway (2020) · cites this paper as [47]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 5. EXAMPLES OF ENDOGENOUS SELECTION BIAS IN SOCIOLOGY > Conditioning on an Intermediate Variable
“Because true innate ability is unobserved, the path cannot be blocked, and the effect S→Y is not identified.”
§ Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable > 5. EXAMPLES OF ENDOGENOUS SELECTION BIAS IN SOCIOLOGY > Conditioning on an Intermediate Variable
“To be concrete, consider the effect of schooling, S, on wages, W.”
Citations
Cites
Baron, R. M., & Kenny, D. A. (1986) · Journal of Personality and Social Psychology · doi:10.1037/0022-3514.51.6.1173 · Cited by 91,532
Sample Selection Bias as a Specification Error
Heckman, J. J. (1979) · Econometrica · doi:10.2307/1912352 · Cited by 28,779
(2001) · RePEc: Research Papers in Economics · Cited by 27,577
Structural Equations with Latent Variables
Bollen, K. A. (1989) · Wiley series in probability and statistics · doi:10.1002/9781118619179 · Cited by 17,944
Structural Equations with Latent Variables.
Clogg, C. C., & Bollen, K. A. (1991) · Contemporary Sociology A Journal of Reviews · doi:10.2307/2072165 · Cited by 17,475
Causality: Models, Reasoning, and Inference
Pearl, J. (2009) · doi:10.1017/cbo9780511803161 · Cited by 11,653
Estimating causal effects of treatments in randomized and nonrandomized studies
Rubin, D. B. (1974) · Journal of Educational Psychology · doi:10.1037/h0037350 · Cited by 9,436
Statistics and Causal Inference
Holland, P. W. (1986) · Journal of the American Statistical Association · doi:10.1080/01621459.1986.10478354 · Cited by 5,293
(1976) · Annals of Economic and Social Measurement · Cited by 5,035
Causation, Prediction, and Search
Spirtes, P., Glymour, C., & Scheines, R. (1993) · Lecture Notes in Statistics · doi:10.1007/978-1-4612-2748-9 · Cited by 4,617
Research works printed but not linked: 12
- Angrist JD, 1996, J. Am. Stat. Assoc., 8, 328
- Bareinboim E, 2012, Proc. 15th Int. Conf. Artif. Intell. Stat.(AISTATS), April 21–23, 2012, La Palma, Canary Islands, 22, 100
- Farr W, 1858, Transaction National Association Promotion Social Science, 504
- Glymour MM, 2008, Modern Epidemiology, 183
- Hausman JA, 1981, The Econometrics of Discrete Data, 365
- Heckman JJ, 1976, Ann. Econ. Soc. Meas., 5, 475
- Kim JH, 1983, Proc. 8th Int. Jt. Conf. Artif. Intell. (IJCAI-83), Karlsruhe, FRG, Aug. 8–12, 1983, 190
- Lin I, 1999, J. Off. Stat., 15, 143
- Pearl J, 2001, Proc. 17th Conf. Uncertain. Artif. Intell., Aug. 2–5, 2001, Seattle, WA, 411
- VanderWeele TJ, 2009, J. Mach. Learn., 10, 699
- Ver Steeg G, 2011, Proc. 27th Conf. Uncertain. Artif. Intell.(UAI2011), July 14–17, 2011, Barcelona, Spain, 717
- Verma T, 1988, Proc. 4th Workshop Uncertain. Artif. Intell., 352
Cited by
Sorted by citations
Scott, A. J., Webb, T. L., Martyn-St James, M., Rowse, G., & Weich, S. (2021) · Sleep Medicine Reviews · doi:10.1016/j.smrv.2021.101556 · Cited by 1,094
Thinking Clearly About Correlations and Causation: Graphical Causal Models for Observational Data
Rohrer, J. M. (2018) · Advances in Methods and Practices in Psychological Science · doi:10.1177/2515245917745629 · Cited by 1,002
“Elwert and Winship (2014) provided a succinct summary of the many ways in which endogenous selection bias can arise.”
BackgroundMethodMoving Forward: Challenges and Directions for Psychopathological Network Theory and Methodology
Fried, E. I., & Cramer, A. O. J. (2017) · Perspectives on Psychological Science · doi:10.1177/1745691617705892 · Cited by 1,015
“Another concern when including many nodes is conditioning on a collider (Elwert & Winship, 2014).”
BackgroundCollider scope: when selection bias can substantially influence observed associations
Munafò, M. R., Tilling, K., Taylor, A. E., Evans, D. M., & Davey Smith, G. (2018) · International Journal of Epidemiology · doi:10.1093/ije/dyx206 · Cited by 813
“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 and outcome are corre…”
BackgroundCollider bias undermines our understanding of COVID-19 disease risk and severity
Griffith, G. J., Morris, T. T., Tudball, M. J., Herbert, A., Mancano, G., Pike, L., Sharp, G. C., Sterne, J., Palmer, T. M., Davey Smith, G., et al. (2020) · Nature Communications · doi:10.1038/s41467-020-19478-2 · Cited by 826
How Conditioning on Posttreatment Variables Can Ruin Your Experiment and What to Do about It
Montgomery, J. M., Nyhan, B., & Torres, M. (2018) · American Journal of Political Science · doi:10.1111/ajps.12357 · Cited by 800
“…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…”
BackgroundA Crash Course in Good and Bad Controls
Cinelli, C., Forney, A., & Pearl, J. (2020) · Sociological Methods & Research · doi:10.1177/00491241221099552 · Cited by 789
“Homophily bias in social network analysis (Elwert and Winship, 2014).”
Moshontz, H., Campbell, L., Ebersole, C. R., IJzerman, H., Urry, H. L., Forscher, P. S., Grahe, J. E., McCarthy, R. J., Musser, E. D., Antfolk, J., et al. (2018) · Advances in Methods and Practices in Psychological Science · doi:10.1177/2515245918797607 · Cited by 377
“In particular, the composition and insufficient size of typical samples in psychological research introduces uncertainty about the veracity (Anderson & Maxwell, 2017; Cohen, 1992; Maxwell, 2004) and generalizability of findings (Elwert & Winship, 2014; Henrich, Heine, & Norenzayan, 2010).”
BackgroundWhat Is Your Estimand? Defining the Target Quantity Connects Statistical Evidence to Theory
Lundberg, I., Johnson, R., & Stewart, B. (2021) · American Sociological Review · doi:10.1177/00031224211004187 · Cited by 422
Keep it real: rethinking the primacy of experimental control in cognitive neuroscience
Nastase, S. A., Goldstein, A., & Hasson, U. (2020) · NeuroImage · doi:10.1016/j.neuroimage.2020.117254 · Cited by 380
“…(Wells and Windschitl, 1999; Judd et al., 2012), and neuroimaging (Bedny et al., 2007; Westfall et al., 2016); or endogenous selection bias, where a spurious relationship between variables of interest is induced by biased sampling along another collider variable (Elwert and Winship, 2014).”
BackgroundThe Taboo Against Explicit Causal Inference in Nonexperimental Psychology
Grosz, M. P., Rohrer, J. M., & Thoemmes, F. J. (2020) · Perspectives on Psychological Science · doi:10.1177/1745691620921521 · Cited by 358
“Adjusting for a collider can introduce a spurious (i.e., noncausal) association between the treatment and the outcome (e.g., Elwert & Winship, 2014; Rohrer, 2018).”
Administrative Records Mask Racially Biased Policing
KNOX, D., LOWE, W., & MUMMOLO, J. (2020) · American Political Science Review · doi:10.1017/s0003055420000039 · Cited by 314
“At the fi rst glance, the problem of race-based selection into policing data may appear a classic case of sample selection bias (Elwert and Winship 2014; Heckman 1979) for which numerous remedies already exist.”
What You Can—and Can’t—Do With Three-Wave Panel Data
Vaisey, S., & Miles, A. (2014) · Sociological Methods & Research · doi:10.1177/0049124114547769 · Cited by 301
Marsman, M., Borsboom, D., Kruis, J., Epskamp, S., van Bork, R., Waldorp, L., van der Maas, H. L. J., & Maris, G. (2017) · Multivariate Behavioral Research · doi:10.1080/00273171.2017.1379379 · Cited by 290
“…though the observables may be marginally independent (Blalock, 1971; Bollen & Lennox, 1991), conditioning on the common effect results in associations between the observables (Elwert & Winship, 2014; Greenland et al., 1999; Greenland, 2003; Heckman, 1979; Hernán, Hernández-Diaz, & Robins, 2004).”
Wang, T., Huang, J., Zhang, H., & Sun, Q. (2020) · 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr42600.2020.01077 · Cited by 291
Digital well-being theory and research
Büchi, M. (2024) · New Media & Society · doi:10.1177/14614448211056851 · Cited by 277
“…questions or hedging in non-causal language like “X is associated with Y.” Researchers need to explicate the assumed mechanisms that generated the observed data—directed acyclic graphs can be employed to clearly state such a data-generating model (Elwert and Winship, 2014; Grosz et al., 2020).”
BackgroundKim, S. J., Wang, R. J.-H., & Malthouse, E. C. (2015) · Journal of Interactive Marketing · doi:10.1016/j.intmar.2015.05.004 · Cited by 285
That’s a Lot to Process! Pitfalls of Popular Path Models
Rohrer, J. M., Hünermund, P., Arslan, R. C., & Elson, M. (2022) · Advances in Methods and Practices in Psychological Science · doi:10.1177/25152459221095827 · Cited by 267
“If we statistically adjust for a collider variable, we introduce spurious associations between its causes (Elwert & Winship, 2014; for more explanation geared toward psychologists, see also Rohrer, 2018).”
BackgroundCausal inference with observational data: the need for triangulation of evidence
Hammerton, G., & Munafò, M. R. (2021) · Psychological Medicine · doi:10.1017/s0033291720005127 · Cited by 262
“This bias is (usually) a consequence of conditioning (i.e. stratifying, adjusting or selecting) on a common effect of an exposure and an outcome (or a common effect of a cause of the exposure and a cause of the outcome), known as collider bias (Elwert & Winship, 2014; Hernan et al., 2004).”
BackgroundMethodInfluentialModel Uncertainty and Robustness
Young, C., & Holsteen, K. (2015) · Sociological Methods & Research · doi:10.1177/0049124115610347 · Cited by 257
“Controlling for race (with the dummy variable ‘‘black’’) also increases the effect size of gender by 1.91, a full 83 percent higher than the mean estimate.”
BackgroundInfluentialStatistical Control Requires Causal Justification
Wysocki, A. C., Lawson, K. M., & Rhemtulla, M. (2022) · Advances in Methods and Practices in Psychological Science · doi:10.1177/25152459221095823 · Cited by 275
“For example, both IQ and hard work can result in getting accepted to college, so collegestudent status is a collider between IQ and hard work (Elwert & Winship, 2014; Rohrer, 2018).”
BackgroundDesJardine, M. R., Marti, E., & Durand, R. (2020) · Academy of Management Journal · doi:10.5465/amj.2019.0238 · Cited by 240
Genetic analysis of over half a million people characterises C-reactive protein loci
Said, S., Pazoki, R., Karhunen, V., Võsa, U., Ligthart, S., Bodinier, B., Koskeridis, F., Welsh, P. I., Alizadeh, B. Z., Chasman, D. I., et al. (2022) · Nature Communications · doi:10.1038/s41467-022-29650-5 · Cited by 213
VanderWeele, T. J., Chen, Y., Long, K. N. G., Kim, E. S., Trudel‐Fitzgerald, C., & Kubzansky, L. D. (2019) · Epidemiology · doi:10.1097/ede.0000000000001147 · Cited by 198
The Statistics of Causal Inference: A View from Political Methodology
Keele, L. J. (2015) · Political Analysis · doi:10.1093/pan/mpv007 · Cited by 175
“See Elwert and Winship (2014) for an excellent discussion of how threats to identification of this form can arise.”
Growing Wealth Gaps in Education
Pfeffer, F. T. (2018) · Demography · doi:10.1007/s13524-018-0666-7 · Cited by 178
“…a consequence of wealth; see Schneider 2011) and, with that, of endogenous selection bias induced by conditioning on a collider (e.g. when marital status a consequence of wealth and also related to further unobserved determinants of educational attainment; see Elwert and Winship 2014: pp. 44-45).”
BackgroundNikolaev, B., Boudreaux, C. J., & Palich, L. E. (2018) · Journal of Small Business Management · doi:10.1111/jsbm.12400 · Cited by 185
“Controls can also leverage backward causal links with the outcome variable, producing reverse causation or selection bias (Elwert and Winship 2014).”
Job Displacement among Single Mothers: Effects on Children’s Outcomes in Young Adulthood
Brand, J. E., & Thomas, J. S. (2014) · American Journal of Sociology · doi:10.1086/675409 · Cited by 146
The health consequences of child marriage: a systematic review of the evidence
Fan, S., & Koski, A. (2022) · BMC Public Health · doi:10.1186/s12889-022-12707-x · Cited by 219
Earnings and first birth probability among Norwegian men and women 1995-2010
Hart, R. K. (2015) · Demographic Research · doi:10.4054/demres.2015.33.38 · Cited by 146
“org 1073 Selecting the sample on union status implies conditioning on an endogenous variable, which could both net out part of the total impact of earnings on fertility and introduce further selection bias in the model (Winship and Elwert 2014).”
Lopez, K. N., Baker-Smith, C. M., Flores, G., Gurvitz, M. Z., Karamlou, T., Gallegos, F. N., Pasquali, S. K., Patel, A., Peterson, J. K., Salemi, J. L., et al. (2022) · Journal of the American Heart Association · doi:10.1161/jaha.122.025358 · Cited by 148
Psychological networks in clinical populations: investigating the consequences of Berkson's bias
de Ron, J., Fried, E. I., & Epskamp, S. (2019) · Psychological Medicine · doi:10.1017/s0033291719003209 · Cited by 121
“Throughout the paper, we will use the term Berkson’s bias to refer to an estimation error due to selecting a subpopulation, but a variety of terms have been used to describe this bias: conditioning on a collider (Pearl, 2000), endogenous selection bias (Elwert and Winship, 2014), collider-stratification bias (Cole et…”
MethodThe Cambridge Handbook of Personality Psychology
(2020) · Cambridge University Press eBooks · doi:10.1017/9781108264822 · Cited by 148
Reflection on modern methods: selection bias—a review of recent developments
Infante‐Rivard, C., & Cusson, A. (2018) · International Journal of Epidemiology · doi:10.1093/ije/dyy138 · Cited by 143
Common Concerns with MTurk as a Participant Pool: Evidence and SolutionsPreprint
Hauser, D., Paolacci, G., & Chandler, J. (2018) · doi:10.31234/osf.io/uq45c · Cited by 162
Thompson, A. J., & Pickett, J. T. (2019) · Journal of Quantitative Criminology · doi:10.1007/s10940-019-09436-7 · Cited by 132
Let’s Talk About Fixed Effects: Let’s Talk About All the Good Things and the Bad Things
Collischon, M., & Eberl, A. (2020) · KZfSS Kölner Zeitschrift für Soziologie und Sozialpsychologie · doi:10.1007/s11577-020-00699-8 · Cited by 152
Walters, G. D., & Espelage, D. L. (2018) · Journal of School Psychology · doi:10.1016/j.jsp.2017.12.003 · Cited by 132
Zhou, X. (2019) · American Sociological Review · doi:10.1177/0003122419844992 · Cited by 115
“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).”
BackgroundCausal Empiricism in Quantitative Research
Samii, C. (2016) · The Journal of Politics · doi:10.1086/686690 · Cited by 127
Socioeconomic status and school absenteeism: A systematic review and narrative synthesis
Sosu, E., Dare, S., Goodfellow, C., & Klein, M. (2021) · Review of Education · doi:10.1002/rev3.3291 · Cited by 155
Sundaram, M. E., Calzavara, A., Mishra, S., Kustra, R., Chan, A. K., Hamilton, M. A., Djebli, M., Rosella, L. C., Watson, T., Chen, H., et al. (2021) · Canadian Medical Association Journal · doi:10.1503/cmaj.202608 · Cited by 118
Atanasov, V. A., & Black, B. S. (2020) · Management Science · doi:10.1287/mnsc.2019.3510 · Cited by 123
Sequeira, M., Lewis, S. J., Bonilla, C., Smith, G. D., & Joinson, C. J. (2016) · The British Journal of Psychiatry · doi:10.1192/bjp.bp.115.168617 · Cited by 118
“46 Unfortunately, no published research that we are aware of has identified SNPs distinctly associated with ‘pre-and post-menarche’ rather than ‘childhood’ BMI.”
Age and Life Satisfaction: Getting Control Variables under Control
Bartram, D. (2020) · Sociology · doi:10.1177/0038038520926871 · Cited by 111
“There is an obvious need for greater clarity in this regard, in particular to avoid what is sometimes called ‘overcontrol bias’ (e.g. Elwert and Winship, 2014).”
Scientist’s guide to developing explanatory statistical models using causal analysis principles
Grace, J. B., & Irvine, K. M. (2019) · Ecology · doi:10.1002/ecy.2962 · Cited by 113
A Crash Course in Good and Bad ControlsPreprint
Cinelli, C., Forney, A., & Pearl, J. (2020) · SSRN Electronic Journal · doi:10.2139/ssrn.3689437 · Cited by 109
Accounting for Confounding in Observational Studies
D'Onofrio, B. M., Sjölander, A., Lahey, B. B., Lichtenstein, P., & Öberg, A. S. (2020) · Annual Review of Clinical Psychology · doi:10.1146/annurev-clinpsy-032816-045030 · Cited by 112
Text and Causal Inference: A Review of Using Text to Remove Confounding from Causal Estimates
Keith, K. A., Jensen, D. D., & O’Connor, B. (2020) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2020.acl-main.474 · Cited by 127
“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).”
BackgroundHamaker, E. L., Mulder, J. D., & van IJzendoorn, M. H. (2020) · Developmental Cognitive Neuroscience · doi:10.1016/j.dcn.2020.100867 · Cited by 133
“Additionally, DAGs are also a useful tool in understanding phenomena such as selection bias, nonresponse bias, and measurement bias (e.g., Elwert and Winship, 2014; Hernán and Robins, 2020).”
BackgroundMethodRecent Developments in Causal Inference and Machine Learning
Brand, J. E., Zhou, X., & Xie, Y. F. (2023) · Annual Review of Sociology · doi:10.1146/annurev-soc-030420-015345 · Cited by 132
“Like the NIE and NDE, researchers can estimate these PSEs via regression-simulation (Miles et al. 2017), regression-imputation (Zhou & Yamamoto 2022), IPW (VanderWeele et al. 2014), or multiply robust methods that are amenable to machine learning estimation of its nuisance functions (Miles et al. 2020, Zhou 2022b).”
BackgroundInfluentialSchoretsanitis, G., Weiler, S., Barbui, C., Raschi, E., & Gastaldon, C. (2024) · JAMA Network Open · doi:10.1001/jamanetworkopen.2024.23385 · Cited by 107
Srinivasan, R., & Chander, A. (2021) · Communications of the ACM · doi:10.1145/3464903 · Cited by 132
Integrating Biomarkers in Social Stratification and Health Research
Harris, K. M., & Schorpp, K. (2018) · Annual Review of Sociology · doi:10.1146/annurev-soc-060116-053339 · Cited by 107
“Identifying a causal relationship requires consideration of the degree to which the social exposure is exogenous or endogenous to the biological stress process measured by the biomarker (Elwert & Winship 2014, Wagner et al. 2013).”
MethodFritz, J., Fried, E. I., Goodyer, I. M., Wilkinson, P. O., & van Harmelen, A. (2018) · Scientific Reports · doi:10.1038/s41598-018-34130-2 · Cited by 102
“Therefore, enhancing RFs in the CA network may not be sufficient to effectively reduce distress levels, as higher levels of RFs may even further hamper other RFs.”
InfluentialInattentive responding can induce spurious associations between task behaviour and symptom measures
Zorowitz, S., Solis, J., Niv, Y., & Bennett, D. (2023) · Nature Human Behaviour · doi:10.1038/s41562-023-01640-7 · Cited by 107
“One major concern with performing rigorous screening and exclusion of participants based on C/IE detection methods is that we might inad-vertently introduce an overcontrol bias 34 .”
BackgroundIntergenerational Mobility and Equality of Opportunity
Torche, F. (2015) · European Journal of Sociology · doi:10.1017/s0003975615000181 · Cited by 100
“This problem is known as endogenous selection bias [Elwert and Winship 2014]. valuable descriptive devices that help researchers assess the different pathways for the transmission of advantage.”
BackgroundCrutzen, R., Peters, G.-J., & Noijen, J. (2017) · Frontiers in Public Health · doi:10.3389/fpubh.2017.00165 · Cited by 101
Uses and abuses of statistical control variables: Ruling out or creating alternative explanations?
Li, M. (2021) · Journal of Business Research · doi:10.1016/j.jbusres.2020.12.037 · Cited by 100
Morris, T., Manley, D., & Sabel, C. E. (2016) · Progress in Human Geography · doi:10.1177/0309132516649454 · Cited by 106
The Mechanics of Omitted Variable Bias: Bias Amplification and Cancellation of Offsetting Biases
Steiner, P. M., & Kim, Y. (2016) · Journal of Causal Inference · doi:10.1515/jci-2016-0009 · Cited by 88
Pollet, T. V., & Saxton, T. K. (2019) · Evolutionary Psychological Science · doi:10.1007/s40806-019-00192-2 · Cited by 89
“From a statistical point of view, restricted sampling can lead to selection bias which in turn can lead to confounding (Bareinboim and Pearl 2012; Elwert and Winship 2014; Fiedler 2000; Rohrer 2018).”
BackgroundAgile work practices and employee proactivity: A multilevel study
Junker, T. L., Bakker, A. B., Gorgievski, M. J., & Derks, D. (2021) · Human Relations · doi:10.1177/00187267211030101 · Cited by 117
“Thus, the observed relationships are unlikely to be owing to collider bias or Berkson’s paradox introduced by “over-controlling” for the differences arising from the agile transformation (see Elwert and Winship, 2014).”
BackgroundThree representations of the Ising model
Kruis, J., & Maris, G. (2016) · Scientific Reports · doi:10.1038/srep34175 · Cited by 108
“To that end, we recognise p(x) from equation (34) as p(x|e = 1), the conditional probability of the collective cause (X) given that all effects are present (E = 1), proportional to the product of the marginal distribution for the collective cause δ ∝ ∑ p x x [ ( ) exp( )] i i i , and the probability of observing the e…”
BackgroundTemperament and brain networks of attention
Posner, M. I., & Rothbart, M. K. (2018) · Philosophical Transactions of the Royal Society B Biological Sciences · doi:10.1098/rstb.2017.0254 · Cited by 95
Mishra, G. S., Clewlow, R., Mokhtarian, P. L., & Widaman, K. F. (2015) · Research in Transportation Economics · doi:10.1016/j.retrec.2015.10.010 · Cited by 94
Why are mediation effects so small?
Walters, G. D. (2018) · International Journal of Social Research Methodology · doi:10.1080/13645579.2018.1517232 · Cited by 96
“Conditioning on a precursor can create endogenous selection bias or a collider effect capable of artificially inflating the a and b path coefficients (Elwert & Winship, 2014).”
BackgroundApplying Moral Foundations Theory to the Explanation of Capital Jurors’ Sentencing Decisions
Vaughan, T. J., Holleran, L. B., & Silver, J. R. (2019) · Justice Quarterly · doi:10.1080/07418825.2018.1537400 · Cited by 90
“Here, we used the degree of attenuation as an indicator of the extent to which the spurious relationships between political ideology and sentencing decisions could be explained by a common cause – particularly, endorsement of the binding foundations (see Elwert & Winship, 2014, p. 360).”
MethodMoral Foundations, Intuitions of Justice, and the Intricacies of Punitive Sentiment
Silver, J. R. (2017) · Law & Society Review · doi:10.1111/lasr.12264 · Cited by 87
The earnings payoff from attending a selective college
Witteveen, D., & Attewell, P. A. (2017) · Social Science Research · doi:10.1016/j.ssresearch.2017.01.005 · Cited by 82
Interactions between Polygenic Scores and Environments: Methodological and Conceptual Challenges
Domingue, B. W., Trejo, S., Armstrong‐Carter, E., & Tucker–Drob, E. M. (2020) · Sociological Science · doi:10.15195/v7.a19 · Cited by 86
Family Background and Earnings Inequality among College Graduates
Witteveen, D., & Attewell, P. A. (2017) · Social Forces · doi:10.1093/sf/sow109 · Cited by 79
We Ran 9 Billion Regressions: Eliminating False Positives through Computational Model Robustness
Muñoz, J., & Young, C. (2018) · Sociological Methodology · doi:10.1177/0081175018777988 · Cited by 76
“For further discussion, see Elwert and Winship (2014) on the problem of conditioning on a collider variable.”
BackgroundMund, M., Johnson, M. D., & Nestler, S. (2021) · Frontiers in Psychology · doi:10.3389/fpsyg.2021.666928 · Cited by 80
“Finally, when possible, as many assumptions as possible about the variables included in a model should be tested (Elwert and Winship, 2014; Brick and Bailey, 2020).”
BackgroundInfluentialGene-environment dependencies lead to collider bias in models with polygenic scores
Akimova, E. T., Breen, R., Brazel, D. M., & Mills, M. C. (2021) · Scientific Reports · doi:10.1038/s41598-021-89020-x · Cited by 86
Rosoff, D. B., Mavromatis, L. A., Bell, A. S., Wagner, J., Jung, J., Marioni, R. E., Smith, G. D., Horvath, S., & Lohoff, F. W. (2023) · Nature Aging · doi:10.1038/s43587-023-00455-5 · Cited by 89
Multigenerational aspects of social stratification: Issues for further research
Mare, R. D. (2014) · Research in Social Stratification and Mobility · doi:10.1016/j.rssm.2014.01.004 · Cited by 77
Manzoni, A., & Streib, J. (2018) · Research in Higher Education · doi:10.1007/s11162-018-9523-1 · Cited by 77
Network Diffusion Under Homophily and Consolidation as a Mechanism for Social Inequality
Zhao, L. H., & Garip, F. (2021) · Sociological Methods & Research · doi:10.1177/00491241211014237 · Cited by 76
“Individuals might self-select into a network in anticipation of engaging in a practice (e.g., a student looking to improve academic performance might join a study group and start working harder; Elwert and Winship 2014).”
BackgroundA Unification of Mediator, Confounder, and Collider Effects
MacKinnon, D. P., & Lamp, S. J. (2021) · Prevention Science · doi:10.1007/s11121-021-01268-x · Cited by 78
Auspurg, K., & Brüderl, J. (2021) · Socius Sociological Research for a Dynamic World · doi:10.1177/23780231211024421 · Cited by 74
“When estimating direct effects, the general rule is to control for both confounders and mediators but to not control for colliders (Elwert and Winship 2014; Young and Holsteen 2017:11).”
BackgroundMethodInfluentialDifferential Registration Bias in Voter File Data: A Sensitivity Analysis Approach
Nyhan, B., Skovron, C., & Titiunik, R. (2017) · American Journal of Political Science · doi:10.1111/ajps.12288 · Cited by 68
CHIELD: the causal hypotheses in evolutionary linguistics database
Roberts, S. G., Killin, A., Deb, A., Sheard, C., Greenhill, S. J., Sinnemäki, K. T.-K., Segovia‐Martín, J., Nölle, J., Berdičevskis, A., Humphreys-Balkwill, A., et al. (2020) · Journal of Language Evolution · doi:10.1093/jole/lzaa001 · Cited by 70
“This property of colliders means that spurious correlations can sometimes be introduced by statistical controls (see Elwert & Winship, 2014; Ding & Miratrix, 2015; Westfall & Yarkoni, 2016; Rohrer, 2018; Middleton et al., 2016; York, 2018).”
BackgroundSmith, B. B., & Waldner, D. (2021) · Cambridge University Press eBooks · doi:10.1017/9781108776837 · Cited by 75
Gain Scores Revisited: A Graphical Models Perspective
Kim, Y., & Steiner, P. M. (2019) · Sociological Methods & Research · doi:10.1177/0049124119826155 · Cited by 74
“The paths are “naturally” (i.e., without any other conditioning) open because they do not contain a collider (Elwert and Winship 2014; Pearl et al. 2016).”
BackgroundMari, G. (2019) · Journal of Marriage and the Family · doi:10.1111/jomf.12600 · Cited by 77
“Second, parsimony shields estimates from the risk of overcontrol bias (Elwert & Winship, 2014), that is, of muting the effect fatherhood has on wages by accounting for the channels through which the effect could manifest in the first place.”
BackgroundInfluentialWessel, T., Andersson, R., Kauppinen, T. M., & Andersen, H. S. (2016) · Urban Affairs Review · doi:10.1177/1078087416638448 · Cited by 76
Ethnic and social class discrimination in education: Experimental evidence from Germany
Wenz, S. E., & Hoenig, K. (2019) · Research in Social Stratification and Mobility · doi:10.1016/j.rssm.2019.100461 · Cited by 82
von Klipstein, L., Riese, H., van der Veen, D. C., Servaas, M. N., & Schoevers, R. A. (2020) · BMC Medicine · doi:10.1186/s12916-020-01818-0 · Cited by 63
“variable that affects both variables (a common cause); and (iii) it might even be spurious, if the two variables both have an effect on a variable that is included in the network (a so-called collider structure) [47].”
BackgroundDevelopment of subjective well‐being and its relationship with self‐esteem in early adolescence
Katsantonis, I., McLellan, R., & Marquez, J. (2022) · British Journal of Developmental Psychology · doi:10.1111/bjdp.12436 · Cited by 87
Hajek, A., Brettschneider, C., Ernst, A., Posselt, T., Wiese, B., Prokein, J., Weyerer, S. B., Werle, J., Fuchs, Â., Pentzek, M., et al. (2015) · Social Psychiatry and Psychiatric Epidemiology · doi:10.1007/s00127-015-1138-7 · Cited by 68
Bartram, D. (2021) · Social Indicators Research · doi:10.1007/s11205-020-02586-3 · Cited by 74
Lundberg, I. (2022) · Sociological Methods & Research · doi:10.1177/00491241211055769 · Cited by 72
Third generation algae biofuels in Italy by 2030: A scenario analysis using Bayesian networks
Gambelli, D., Alberti, F., Solfanelli, F., Vairo, D., & Zanoli, R. (2017) · Energy Policy · doi:10.1016/j.enpol.2017.01.013 · Cited by 64
Katikireddi, S. V., Green, M. J., Taylor, A. E., Smith, G. D., & Munafo, M. R. (2017) · Addiction · doi:10.1111/add.14038 · Cited by 63
“This threat to causal inference is perhaps less intuitive than either of the above, but can impact upon the strength and direction of associations observed [12,15,16].”
BackgroundSong, X. (2016) · Demography · doi:10.1007/s13524-016-0515-5 · Cited by 60
“, that all confounders of the association between the exposure A and the outcome Y are included in the model) is violated (Elwert and Winship 2014).”
BackgroundInfluentialEras, L. (2022) · Nationalities Papers · doi:10.1017/nps.2021.100 · Cited by 62
“Since some of the variables might be mediators and thus lead to overcontrol bias (Elwert and Winship 2014), I first estimate models without any control variables.”
BackgroundGraphical Models for Quasi-experimental Designs
Steiner, P. M., Kim, Y., Hall, C. E., & Su, D. (2015) · Sociological Methods & Research · doi:10.1177/0049124115582272 · Cited by 56
Control variables and causal inference: a question of balance
York, R. (2018) · International Journal of Social Research Methodology · doi:10.1080/13645579.2018.1468730 · Cited by 62
“…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).”
BackgroundHow fear of crime affects punitive attitudes
Armborst, A. (2017) · European Journal on Criminal Policy and Research · doi:10.1007/s10610-017-9342-5 · Cited by 60
“By the same token the problem of collider variables and over-control bias (Elwert and Winship 2014) cannot be ruled out.”
Background
Metadata
- Authors
- Elwert, F., & Winship, C.
- Affiliations
- University of Wisconsin–Madison (US)
- Harvard University (US)
- Venue
- Annual Review of Sociology
- Publisher
- Annual Reviews
- Published
- 2014-07-30
- Volume
- 40
- Issue
- 1
- Pages
- 31-53
- ISSN
- 0360-0572 (ISSN-L)
- 1545-2115
- PMC
- PMC6089543
- OpenAlex
- W2097829006
- Semantic Scholar
- 9fcc6fe93c9ce0c0b4773c18074c01bdf71db52d
- Cited by
- 1,168
- References
- 120
- Citations by year
- 2026: 70 · 2025: 102 · 2024: 156 · 2023: 156 · 2022: 122 · 2021: 125 · 2020: 150 · 2019: 94 · 2018: 82 · 2017: 52 · 2016: 29 · 2015: 23 · 2014: 7
- FWCI
- 4.69
- Type
- journal article
- Language
- en
- Open access
- bronze
- doi.org
- Locations
- Annual Review of Sociology (publishedVersion)
- PubMed (publishedVersion)
- Europe PMC (PubMed Central) (acceptedVersion)
- PubMed Central (acceptedVersion)
- Funders
- National Institute on Aging (P30AG017266)
- Eunice Kennedy Shriver National Institute of Child Health and Human Development (P2CHD047873)
- Topics
- Physical Sciences › Mathematics › Statistics and Probability › Advanced Causal Inference Techniques
- Physical Sciences › Computer Science › Artificial Intelligence › Bayesian Modeling and Causal Inference
- Social Sciences › Arts and Humanities › History and Philosophy of Science › Philosophy of Biology and Scientific Explanation
- Keywords
- collider bias, causal inference, causal graphs, social stratification, sociology of culture, social network analysis, political sociology, social demography, sociology of education
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
A language model drew the claims from the paper on 2026-09-28, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-09-28 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
