TC1-08 Confounder identification with directed acyclic graphs (adjustment sets, backdoor criterion)
Draws assumed causal links as a directed acyclic graph and produces a diagram and an adjustment set of variables to include and to leave out.
What it does
Draws assumed causal links to decide which variables to adjust for and which to leave alone. Artefact: a diagram and an adjustment set.
When to use it
Use when you must decide which variables to control for before estimating an effect and the causal links can be defended from subject knowledge.
When it does not fit
Important common causes are unmeasured, or a collider or mediator is adjusted for, which can add bias.
Result format
The result has a diagram of the assumed causal links between the variables and an adjustment set naming the variables to adjust for and the variables to leave alone. It also has a register of the causal assumptions the diagram depends on.
Requirements on the material
| Property | Requirement type | Requirement |
|---|---|---|
| Content type | Always applies | requires=text |
| Unit of analysis | Always applies | requires=variable |
| Appraises source credibility | Applies in some cases | no |
Not restricted on: Structure, Variable types, Time structure, How the sample was chosen, Level of detail available, Number of sources, Whether sources can be joined, Amount of material, Known data problems, Producer interest, Source dependence, Origin, Document genre, Language, Personal information, Security classification, Licence.
| Minimum material | A named exposure and outcome, and subject knowledge strong enough to defend each assumed link. |
|---|---|
| Additional sources needed | none |
| Not for (rules) | none |
| Not for (text) | Wrong choice when important common causes are unmeasured, or when a collider or mediator is adjusted for, which can add bias. The graph encodes assumptions and does not estimate anything by itself. |
Choosing and ordering
| Kind of technique | Primary technique, runs in step 5.3 |
|---|---|
| Phase tag | analyse |
| Other phase tags | frame |
| Kind of finding | cause |
| Question answered | why |
| Strongest statement it supports | estimate the effect of an action |
| Needs as input |
|
| Makes as output |
|
| Must exist first | None |
| Loop target | return to frame |
Library checks by stakes
Each stakes level includes the checks of the level below it. Checks that need a person apply at high stakes only, so the process leaves them out at lower stakes (decision 38). Repeatability: judgement dependent.
| Stakes | Check | Person involved | Counts as a different method |
|---|---|---|---|
| exploratory | assumption and limitation statement | none | no |
| decision-informing | assumption and limitation statement | none | no |
| decision-informing | alternative explanations check | none | no |
| high-stakes | assumption and limitation statement | none | no |
| high-stakes | alternative explanations check | none | no |
| high-stakes | expert review | needed | no |
| high-stakes | independent review | needed | no |
Failure modes and limits
| Model suitability | human judgement required |
|---|---|
| Typical model failure | Draws links to match the data at hand and not prior knowledge, or includes a collider or mediator in the adjustment set. |
| Error sources | input data; expert opinion; coder or rater judgement |
| Certainty limits | Certainty drops when links are untested, when relevant common causes are unmeasured, or when experts disagree on the graph (risk of bias). The ceiling is estimate an intervention effect because the graph states what must be adjusted for, and the estimate comes from another technique. |
Related techniques
- pairs with DM-19 Causal loop diagram Also serves: Trace cause and test claims.: a causal loop diagram shows feedback that an acyclic graph cannot represent
- pairs with TC1-07 Quasi-experimental designs (natural experiment, instrumental variables, regression discontinuity): the chosen adjustment set feeds the estimate
Evidence and tag basis
Unconfirmed. The book lists the topic, but detailed assumptions were not read. (HernĂ¡n and Robins, Causal Inference: What If)
Tag basis: input profile inference; selection inference; sequencing inference; robustness inference. All tag values are inference unless a source is named.