Questions tagged [causality]
The relationship between cause and effect.
1,990 questions
6
votes
1
answer
87
views
What happens when predictor variables influence other predictors?
I'm fitting a regression of Y on predictors X₁, X₂, X₃ using pre-intervention data, then plugging in observed post-intervention X values to generate a counterfactual Y and compare it to the observed Y....
- 101
3
votes
1
answer
87
views
On the impossibility to define causality in purely statistical terms
Context
This is his statement, from "The book of why" that Judea Pearl co-authored with Dana Mackenzie:
Probabilities like $P(Y|X)$, lie on the first rung of the Ladder of Causation and ...
- 71
1
vote
0
answers
26
views
How to determine exactly why variable A is no longer a predictor of Y when B is added?
Am I right that these are the three possibilities?
1 A and B predict overlapping variance in Z, and so when B is added A no longer has any unique variance to explain.
2 B mediates the relationship ...
- 2,761
7
votes
1
answer
182
views
Does including a predictor derived from the outcome process itself cause problems?
I am modeling a process where subjects transition between states over time. I want to include a time-varying covariate in my model that is not externally measured but is instead a deterministic ...
6
votes
2
answers
214
views
Does perfect deterministic dependence of an IV on controls cause multicollinearity in 2SLS first stage?
Context
I am estimating the causal effect of mosquito net use on dengue risk using an IV strategy. A government programme provides free mosquito nets to households satisfying both of the following ...
- 61
10
votes
2
answers
755
views
The assumptions for causal inference seem obvious - why state them?
In this paper Combining counterfactual outcomes and ARIMA models for policy [doi], the authors listed 3 assumptions for estimates to be considered as causal:
Single Persistent Intervention. There ...
- 169
10
votes
3
answers
669
views
Causal effects without a control group?
I found this paper https://www.personal.soton.ac.uk/cz1y20/Reading_Group/mlts-2023/week9/utac024.pdf which discusses a method in which the causal effect of an intervention can be estimated in a time ...
- 169
1
vote
0
answers
52
views
Where do randomization blocks go in a causal directed acyclic graphs?
In a block randomized experiment, with the blocks $B$ created from gender $G$, age $A$, and education $E$ all treated as categories. With a binary treatment indicator $T$ what would a
directed acyclic ...
- 1,126
3
votes
0
answers
103
views
A question on Technical Point 16.3 of the What If book on causal inference
I have some confusion on the Technical Point 16.3 of the What If book on causal inference, by Hernan and Robins.
The set up is $Z$ is the treatment that a patient was assigned, $A$ is the treatment he ...
- 359
5
votes
1
answer
188
views
How to explain the collider bias from treating a post randomization variable as a moderator?
I have been struggling with explaining and understanding the impact of collider bias from intuitive standpoint particularly when using a post randomization variable as a moderator in a longitudinal ...
- 1,126
2
votes
0
answers
66
views
What are the differences between doing causal inference in general and target trial emulation?
I am statistician self-training in causal inference. By now, I think I have developed a good grasp of several methodologies related to causality, in particular applied to public health research.
...
- 954
1
vote
0
answers
38
views
Is g-computation using mixed-effects models with time varying confounders causally valid?
I am trying to clarify the conditions under which g-computation based solely on an outcome regression is a valid causal estimator in longitudinal data; specifically, a mixed-effects (or more generally,...
- 350
1
vote
0
answers
18
views
Time-varying “remission” is a function of time-varying biomarkers a(t), b(t), c(t): modeling choices for death outcome?
I have a longitudinal cohort with repeated biomarker measurements a(t),b(t),c(t) over follow-up and a time-to-event outcome (death).
I want a prognostic (...
- 530
4
votes
0
answers
50
views
G-computation for the ATT, difference between using observed outcome under treatment and model predicted outcome to calculate E[Y|A = 1]?
Assume we are using g-comp to estimate the ATT of a binary treatment A where 1 is treated 0 is control on outcome Y controlling for confounders W.
The ATT is defined as:
ATT = E[Y|A = 1] - E[Y0|A = 1],...
- 41
5
votes
1
answer
175
views
(How) Are techniques for analysing/adjusting for confounding applicable to biomarkers?
There is extensive literature on confounding in epidemiological context, where one analyses the effect of an exposure (E) on incidence of a disease (D). A confounder is then a variable that may be the ...
- 5,324