Questions tagged [causality]
The relationship between cause and effect.
1,611
questions
3
votes
1
answer
211
views
Why use causal inference if coefficients are same in an OLS?
I was reading this amazing article about FWL theorem and it's application to causal inference.
In the article, there are some examples showing that the coefficients of an OLS estimator is the same ...
- 155
0
votes
0
answers
22
views
Confounding variables are nested with treatment, not able to be measured, how to address the influence from confounding factors?
We gathered driving data from two cohorts of drivers belonging to the same age group. The first cohort, Group A, utilized System A (treatment group), whereas Group B drove vehicles without this system ...
- 51
3
votes
0
answers
8
views
Competitions/datasets fit for exploring Pearl's graphical causal models
Are there any competitions/challenges/datasets fit for testing Pearl's graphical causal inference methods? I do not necessarily mean live competitions.
I would expect these setups to be different than ...
1
vote
1
answer
17
views
In what cases is identification not possible in causal inference?
In step one of judea pearls causal inference book it is to define your graphical causal model. The second step is identification of the estimand for estimation in step 3. Are there any cases where ...
- 469
0
votes
0
answers
8
views
Testing sum of dummy variables
Say that you had data for the point spread of a basketball game: Team A points - Team B points (where team A is the home team and team B the away, if there is no home and away team, then team A is ...
2
votes
1
answer
28
views
According to DAG theory, why controlling for this variable doesn't close the backdoor path opened by controlling for collider?
I have made the following model in DAGitty:
Where $X_2$ is controlled for.
DAGitty says:
The total effect cannot be estimated due to adjustment for an intermediate or a descendant of an intermediate....
- 1,928
1
vote
1
answer
31
views
Synthetic control method based on several treated units
I am trying to causally evaluate an impact of a specific labor policy that was implemented in three U.S. states. I wonder if constructing a synthetic/artificial control method (SCM) for those states ...
- 71
0
votes
1
answer
26
views
What does extrapolation mean in the context of regression weights and what are its downsides?
In numerous articles I have read (Abadie 2021, Samii 2016, basically any Abadie piece that talks about the synthetic control method), the authors cite regression's reliance on extrapolation for ...
- 519
1
vote
0
answers
17
views
Tuning a Synthetic Control
I am working on a project that is seeking to use the synthetic control method to estimate the effect of Costa Rica's military abolition on levels of democracy (0-1 interval). I'll attach my code here ...
- 519
0
votes
1
answer
35
views
How can you rewrite the estimand in terms of propensity scores? Dowhy question
I am going through the backdoor criterion and how we get from an expression involving do to one which doesn't as below.
What i don't quite get is how to rewrite this estimand in terms of propensity ...
- 469
0
votes
0
answers
17
views
Instrumental variable identifiability in a linear setting in the presence of unobserved confounders
Long story short, I'm seeing in the literature that linear instrumental variables models are identifiable, even in the presence of unobserved confounders. The unobserved confounding aspect befuddles ...
- 317
0
votes
0
answers
16
views
Understanding Optimizer Options Using tidysynth
Under the generate_weights() command using the {tidysynth} package for executing the synthetic control method in R, users are given a number of optimization options. However, I cannot find ...
- 519
0
votes
1
answer
26
views
Interpreting Mixed Model ANOVA
In a 3 (condition) X 2 (time) mixed model ANOVA. If I hypothesised that anxiety in group A will increase from time 1 to time 2 and my results found no significant interaction but a significant main ...
1
vote
2
answers
41
views
DAG - why is the path open?
I have this DAG
As I understand it, the paths D <- Ed -> St -> P -> Su and D <- A -> P -> Su are both closed because the contain the collider P. If I condition on P, both these ...
1
vote
1
answer
44
views
Understanding the Difference Between Independent and Dependent Variables
I'm new to statistics and I'm struggling to grasp the distinction between an independent and a dependent variable. For instance, if I want to examine the correlation between daily COVID-related deaths ...
- 13
0
votes
0
answers
15
views
Estimating variance of estimated mean in leaf when using Honest splitting
In Recursive Partitioning for Heterogeneous Causal Effects by Susan Athey, Guido W. Imbens, under section 2.5 Honest Splitting, two different datasets (called tr and est) are used for (a) creating the ...
- 43
0
votes
0
answers
19
views
Interpreting Non-Parallel Pre-Treatment Trends in Difference-in-Difference Analysis: Coincidential Failure of Common Trend Assumption?
I'm conducting a Difference-in-Difference (DiD) analysis where I included clusters at the state level. To fulfill the crucial common trend assumption of DiD, I examined the pre-treatment period to see ...
- 1
1
vote
0
answers
20
views
An Alternative In-sample-goodness-of-fit Measure in Machine Learning Methods for Estimating Heterogeneous Causal Effects Paper
In 3.5.2 of Machine Learning Methods for Estimating Heterogeneous Causal
Effects by Susan Athey and Guido Imbens it is stated:
"If the models include an intercept, as they usually do, most ...
- 43
2
votes
1
answer
32
views
How to understand the second rule of front door criterion?
In the Definition 3.4.1 of Pearl's causal inference book (Primer), the second rule for the front door criterion is "There is no backdoor path from $X$ to $Z$". But from my understanding, ...
- 23
2
votes
0
answers
57
views
Generalized Difference in Differences model: time*group interaction contradicts lift
I've constructed a Bayesian Generalized Difference in Differences model. I model an intercept as well as three coefficients; one for treatment_group assignment, one for post-start period, and one for ...
- 2,183
0
votes
0
answers
7
views
How to choose features to condition on in DiD causal model?
I’m looking at CausalPy documentation and the algorithm of interest ir Difference in Differences. I see that they build two different models.
In general, how should one select the features and/or ...
- 2,183
1
vote
1
answer
24
views
How to handle intermittent missing visits in marginal structural model?
The reference papers for marginal structural models only talked about handling monotone censoring using IPCW.
How to deal with intermittent missing visits? Does it make sense to use available visits ...
- 579
3
votes
1
answer
26
views
In regression, should we adjust for variables only associated with the independent or dependent variable?
I have recently been reading more about causal inference so am trying to conceptually think about model specification in more detail. From reading (e.g. this paper), we adjust for confounders which, ...
- 133
1
vote
1
answer
35
views
Bang and Robins doubly-robust estimator biased and with large variance?
In their 2005 paper (also see the correction here) Bang and Robins describe a doubly robust estimator of the average treatment effect.
In short, the procedure is:
Estimate inverse probability of ...
- 768
0
votes
1
answer
32
views
Biased sample same like conditioning on a collider?
I am studying chapter 5 "The many variables & the spurious waffles" of the book Rethinking and trying to answer the following question:
How is biased sample like conditioning on a ...
- 387
2
votes
0
answers
36
views
Reasoning weighted regression via causal inference
I want to follow a similar approach to what was done in the paper of Bornkamp et al. in 2021 (DOI: 10.1002/pst.2104). More specifically they added a link to a R markdown file (https://oncoestimand....
- 21
8
votes
2
answers
1k
views
Incorrectly Using the Word "Causal" to Describe a Regression Model?
Suppose we take the classical linear regression model:
$$y_i = \beta_0 + \beta_1 x_i + \epsilon_i$$
Over the years, I have heard so many people say that such an interpretation can be drawn from this ...
- 7,150
0
votes
0
answers
10
views
diff-in-diff regression - parallel assumptions
I want to run a diff-in-diff regression for a pre-post rollout experiment. Assume for a moment that I have subjects impacted by the experiments.
To validate the trends, I AGGREGATE subjects to a daily ...
- 127
1
vote
1
answer
35
views
Invalid use of Propensity Score Matching?
I wonder if using a propensity score in the following situation is wrong. Imagine I have the next causal model
$$X = N_x$$
$$Y = f(X, N_y)$$
$$ Z = g(X, Y, N_z)$$
Where $N_z, N_y, N_x$ are ...
1
vote
2
answers
107
views
Can controlling for a variable block the backdoor path opened by controlling for a collider?
I have made the following model in DAGitty:
Where X2 is controlled for.
DAGitty says:
The total effect cannot be estimated due to adjustment for an intermediate or a descendant of an intermediate.
...
- 1,928
2
votes
1
answer
46
views
Causal inference for intervention with no data in pre-intervention period
This question is a generalization of another question that I asked here.
Suppose that Walmart has 1,000 stores. It has a 20% coupon for cereal, and it hypothesizes that the coupon will increase the ...
- 111
1
vote
0
answers
35
views
Difference-in-difference estimation of average treatment effect with no pre-intervention period
Suppose that Walmart has 1,000 stores. It has a 20% coupon for cereal, and it hypothesizes that the coupon will increase the sales of cereal by 3%.
Walmart put the coupon in 100 stores on 2022-05-01; ...
- 111
3
votes
3
answers
73
views
How should control candidates be decided for causal inference?
I’m getting a bit confused about who to include in the control and treatment pools and would appreciate the help. I need to estimate the effect of treatment where 100% of the population was assigned a ...
- 31
1
vote
1
answer
148
views
Causal counterfactual inference model comparison
When refuting two causal models, model 1 has a bigger p-value and an estimated effect closer to the new effect (compared to model 2). Both can't be refuted because they have a p-value above 0.05.
Is ...
- 81
0
votes
0
answers
17
views
Interpret CausalImpact summary
What is Prediction(s.d.) in the summary output calculated based on? My understanding is that there are iterations done to get predictions of the counterfactual in the post-intervention period at each ...
- 1
1
vote
1
answer
31
views
How to Evaluate and Visualize the Positivity Assumption for the ATT
I am trying to formally evaluate and visualize the satisfaction of the positivity assumption when estimating the ATT using R and I am having a tricky time figuring out how to do so. As Greifer and ...
- 519
1
vote
1
answer
42
views
Noise abduction for computing counterfactuals
Given observational data $X$ and knowledge of the true causal graph structure $\mathcal{G}$. How does abduction of the exogenous noise ($U$) for computing counterfactuals work?
We don't have data ...
- 363
1
vote
1
answer
56
views
Linear regression assumptions for causal Inference
When I'm using linear regression as an estimation method to infer the average treatment effect (between treatment and outcome), am I assuming there is a linear relation only between treatment and ...
- 81
1
vote
0
answers
28
views
Difference between NIE and TE - NDE in causal inference?
I have a question regarding mediation analysis. I'm wondering why oftentime the fraction of the total effect which corresponds to the indirect effect, i.e. 1 - NDE/TE, is different from the natural ...
- 11
1
vote
0
answers
22
views
Minimal sample size for difference-in-difference estimation of causal effect with time-series data
Suppose that Walmart has 1,000 stores. It has a 20% coupon for cereal, and it hypothesizes that the coupon will increase the sales of cereal by 3%.
Walmart wants to put the coupon in $n$ stores on ...
- 111
1
vote
0
answers
29
views
Is it necessary to observe a significant mean difference in post/pre for the treated group in DiD?
I have a staggered adoption of a policy across states. I expected the mean of my outcome in post to increase for all treated states, however, I do not observe it consistently. Furthermore, the t-stat ...
- 11
1
vote
1
answer
32
views
Using predicted outcomes to adress selection bias in causal inference
Can I use predicted outcomes from one model as the dependent variable in another model to make causal claims? Put differently, is there something equivalent to the Frish-Waugh-Lovell theorem for ...
- 113
3
votes
1
answer
107
views
Estimating and Interpreting the ATT with Regression Adjustment and Marginal Effects
I am beginning a project that will employ regression/covariate adjustment to estimate the average effect of treatment on the treated (ATT) and I realize that I have two questions concerning how one ...
- 519
3
votes
1
answer
58
views
Why is covariate balance fundamental for causal inference in observational data but unnecessary in experimental data?
The observational case
For observational data, Hernán & Robins (2023, p. 49) state:
In the absence of marginal randomization, [computing the average causal effect in the entire population] ...
- 1,241
2
votes
0
answers
36
views
Create a model with non-DAG relationship
I want to model the relationship between time spent watching video ads and purchases. There is a correlation between viewing time and purchase. The longer the viewing time, the higher the purchase ...
- 31
1
vote
0
answers
19
views
Non-intervenable variables in causal directed acyclic graphs
Consider a very simple causal DAG representing a randomized experiment of $Treatment \rightarrow Death$ where $Treatment$ is randomly assigned and $Death$ is the outcome of interest. A causal DAG ...
- 337
1
vote
1
answer
82
views
Computing counterfactual query given an SCM and how it differs from computing interventional query?
Assuming we have the following structural causal model (SCM), with a confounder DAG structure, as follows:
Noise variables:
$$U_1 \sim \mathcal{N}(0,\,1)$$
$$U_2 \sim \mathcal{N}(0,\,1)$$
$$U_3 \sim \...
- 363
3
votes
2
answers
49
views
Causal modeling in the presence of a latent variable
Suppose that four variables of $X$, $Y$, $L$, and $C$ have the following relationships in the form of directed acyclic graph.
$X$, $Y$, and $C$ are observable variables while $L$ is a latent (...
- 2,518
0
votes
0
answers
17
views
Should instrumental variables abide by the effect hierarchy principle?
People sometimes include the interactions between their instruments and region/year or the interactions between different instruments in the first stage of a 2SLS regression. I wonder if the effect ...
0
votes
0
answers
19
views
Statistical inference in A/B testing: is it enough to compare observed test outcome to control distribution?
I want to run an A/B test to examine the effect of a marketing campaign on revenue. I’m using a synthetic control setup, hence I have to compare the revenue generated by treated units to the ...
- 1