Questions tagged [causality]
The relationship between cause and effect.
1,728
questions
0
votes
0
answers
8
views
Regression with bounded dependent variable
I'm using interrupted time series analysis to estimate the impact of an intervention in the same group. However, my target variable is a satisfaction index that goes from -1 to 1.
How can I model a ...
- 1
1
vote
0
answers
17
views
PC algorithm in "Causation, Prediction, and Search"
In the page 120 in "Causation, Prediction, and Search", why the edge B-D will not be removed if the edge E-D is mistakenly removed from the initial complete graph. I think B-D is D-connected ...
- 11
1
vote
0
answers
13
views
Is There a Standard Metric for Evaluating Treatment Impact Considering Action Cost in Uplift Models?
I'm currently exploring Uplift modeling, specifically the use of the Conditional Average Treatment Effect (CATE) metric:
$$ \tau(t', t, x) := \mathbb{E}[Y | X=x, T=t'] - \mathbb{E}[Y | X=x, T=t] $$
...
- 125
0
votes
0
answers
15
views
Fuzzy or sharp RDD?
I am looking at the causal effect of a Danish retirement reform, where individuals born after 1954 have an increased normal retirement age. Therefore, the cutoff of the treatment is clear. However, ...
- 1
9
votes
3
answers
388
views
For what reasons should regression models not adjust for variables comprising a composite score?
Medicine has many examples where some score is derived from other clinical measures. Examples include BMI (which is a function of weight and height) and MELD (which appears to be based on blood tests)...
- 33.4k
0
votes
0
answers
12
views
Popular Causal Inference Methods? [closed]
I want to be able to list/bucket different types of causal inference methods.
Please correct me and place them in the appropriate bucket if the following isn't correct:
Causal inference methods:
A/B ...
- 831
11
votes
2
answers
305
views
What is the stopping criterion for adding nodes to a causal DAG?
I'm currently involved in creating a causal DAG (Directed Acyclic Graph) to map causal relationships of a real industry problem. The more I think, the more ancestor nodes I add to the DAG which is ...
- 213
2
votes
1
answer
69
views
Could classical astronomic discoveries be made using ML?
Imagine that prior to the time of Copernicus, scientists could run machine learning algorithms (somehow).
Would it be possible to discover the truth of the geocentric model with their use?
Would the ...
- 511
1
vote
0
answers
29
views
Is it possible to estimate effects using Bayesian modelling after matching?
I am following [Greifer 2023][1] to estimate the effect size after (genetic) matching, where I am using bootstrapping to estimate the confidence intervals. Since I have a hierarchical setup with ...
- 41
3
votes
1
answer
42
views
How to report the effect of covariates in post-matching analysis?
I have conducted genetic matching using the MatchIt package in R with the ultimate aim to compare the effect of the transition to a new form of land management. I ...
- 41
3
votes
2
answers
273
views
Why do we need a consistency assumption in causal inference?
Why do we need a consistency assumption in causal inference? I think the consistency assumption is quite obvious and it is more like a definition for the observed outcome.
- 51
0
votes
1
answer
35
views
Inferring A given C and chain ABC
We have a claimed causal chain $A\to B\to C$. We know that $P(C) < 0.01$ and we know that $P(B|A) > 0.9$ and $P(C|B) > 0.9,$ how can we find $P(A)?$
We might lack information here but can we ...
- 263
1
vote
1
answer
67
views
Causal inference on time-series data: is intervention needed?
I'm working on the topic of causal inference, I use time-series data. I have two scenarios in front of me and I don't understand the difference:
Given X and Y "time" features. I would like ...
- 23
0
votes
0
answers
13
views
Conditional Average Treatment Effects: What is the logic underlying using the covariates to predict the residualized scores?
I've consulted a few references, including this online book, and I've got a good grasp on using double machine learning to estimate ATE after debiasing/denoising the treatment variable (T) and the ...
2
votes
0
answers
17
views
Semi-parameteric estimation
I am interested in the effect of certain interventions $T$ on my value of interest $Y$, my model is,
$$Y = \tau f(T, X, Z) + g(X, Z), $$
where $f(T, X, Z) = T \times X + T \times Z$ , that is all the ...
- 1,323
3
votes
1
answer
91
views
The use of p-values in tests of d-separation
In his very helpful online book on structural equation modeling, Jon Lefcheck writes the following concerning d-separation tests for SEMs:
Once the model is fit, statistical independence is assessed ...
- 339
3
votes
1
answer
60
views
Is running P(T=1|Y,X) a recommended way to help diagnose the unconfoundedness assumption in causal inference?
I've always heard that there isn't a lot of ways to diagnose the assumption that you've found the correct set of confounders in causal inference in confounder-control studies, and the best we can do ...
- 87
2
votes
0
answers
28
views
Unbiased estimation of treatment regime contrast with time-varying treatment and outcome
I have some troubles finding a strategy to identify the causal effect I am looking for from my observed data. I am assuming the following DAG:
where $Z$ is the result of a coin toss (randomization), $...
- 352
0
votes
0
answers
18
views
Derivation of the formula for E(MSE) in Recursive Partitioning for Heterogeneous Causal Effects (Athey and Imben 2015)
This post answers a question from the Recursive Partitioning for Heterogeneous Causal Effects paper by Athey and Imbens.
However, I am blocked at an even earlier stage from the previous post. I don't ...
0
votes
0
answers
27
views
Causality for cross sectional data
I have been reading and reading about causality for cross-sectional datasets (different variables including only ONE-YEAR information) and I am still confused...
CAN WE analyze causality in one-year ...
- 1
6
votes
2
answers
167
views
Causal Inference When Treatment Assignment is Fully-Known But Not Randomized
Consider a situation where a public formula specifies the amount of treatment (the dollar amount of financial aid) students receive and I am interested in estimating the causal effect of treatment on ...
- 633
0
votes
0
answers
8
views
Controlling for Price Elasticity Bias in Causal Model
I am trying to study the treatment effect of lowering prices on sales (demand). My treatment includes 3 difference price points ($10, $12, $14) and sales is a continuous $ variable. All buyers have ...
- 203
1
vote
1
answer
17
views
How to recover total distance in of optimal full matching in MatchIt?
It is easy to recover the total distance using optmatch::fullmatch, e.g.,
...
- 187
0
votes
0
answers
17
views
How does the heteroskedasticity-robust SE equal the conservative estimator of SE for sample average treatment effect?
I have been told that the heteroskedasticity-robust standard error of $\hat{\beta}_1$ from an OLS regression with a binary $X_i$: $Y_i = \beta_0 + \beta_1 X_i + u$ should be the same as the ...
1
vote
0
answers
12
views
Intervention in pgmpy causal given evidence
The .query method in pgmpy computes the effect of one variable X on Y given evidence Z. I'm not sure Z is a set of observed covariates; if that is the case, isn't ...
- 126
0
votes
0
answers
22
views
What is the correct specification of covariates/matching variables for exact matching when estimating effects with MatchIt in R?
First, I understand that including covariates in the outcome model after matching is optional based on reading the matchit vignette and similar question here. However, I'm a bit confused on what is ...
10
votes
3
answers
2k
views
Are directed acyclic graphs (DAGs) only used for visualization?
I see people using these DAGs a lot in articles (e.g. this vignette). Are these kinds graphs only serving purposes of aesthetics and visualizations representations? Or do these graphs actually have ...
0
votes
0
answers
23
views
Effect modification on the absolute scale
Usually when talking about effect (measure) modification or treatment effect heterogeneity, we are talking about heterogeneity in the relative effect under two interventions in some subset of the ...
- 507
2
votes
0
answers
19
views
How to determine impact of a Visiting Team on Attendance at a Sporting Event? [Personal Thought Exercise]
I am trying to figure out the impact of a visiting team on the attendance draw of a sporting event.
This is a personal inquiry exercise to see if the team that I root for has any impact on the other ...
- 21
1
vote
0
answers
16
views
Converting PanelMatch Estimates to Marginal Effects in Non-Linear Outcome Models
I am interested in using the {PanelMatch} package for a project that I am working on where the conventional matching/weighting framework is expanded to account for the complexities of the panel data ...
- 633
7
votes
3
answers
590
views
Should You Specify a Curvilinear/Non-Linear Effect If You Suspect It is Spurious?
Consider the following (simplified) example of a project I am working on:
I assume that $X$ has a linear effect on $Y$. However, after plotting the relationship on a scatter plot, it looks like the ...
- 633
4
votes
1
answer
52
views
What is an intuitive explanation for a lay-person why controlling for a collider is bad practice while controlling for a confounder is good practice?
Cinelli et al., 2022 and Wysnocki et al., 2022 describe in technical terms how controlling for a collider can lead to biased estimates. If one needs to explain why one should not control for a ...
- 838
0
votes
0
answers
24
views
Stepped wedge design to analyze healthcare intervention rolled out at different times to members?
I'm working on a pre/post analysis of longitudinal healthcare data where an intervention program was implemented on a member by member basis over a period of several years. Some eligible members ...
- 5,572
0
votes
1
answer
42
views
DAGitty Question: Testable Implications only describes independencies, but not dependencies
Why does the DAGitty "Testable Dependencies" function only describe independencies, but not spurious correlations?
E.g, if I have B->A<-C,
DAGitty just tells me that
B ⊥ C is the only ...
- 101
1
vote
1
answer
34
views
Two-way fixed effects model: Is non-random treatment a problem here?
Consider the following panel regression model, were we try to estimate whether Photovoltaic Panels (PV) installations on apartment blocks have a driving effect on adoption speed of surrounding ...
- 131
3
votes
1
answer
47
views
How to pick the winner in the "Play the Winner" treatment assignment scheme (Urn model)
"Play the winner" is an intutitive procedure whereby successful treatments in earlier trials have a higher probability of being assigned in later trials. Quoting https://online.stat.psu.edu/...
- 5,127
2
votes
1
answer
151
views
When controlling for confounders in a causal study, should we always expect a decrease in the treatment effect estimate?
In a causal study, can treatment effect go up upon applying causal techniques when comparing with the naive / simple difference in means? I think this question is method-agnostic, but for transparency ...
- 87
0
votes
0
answers
18
views
Given a known Bayesian Belief Network or Causal DAG, how can we score the value of different pieces of evidence?
Assume we know the BBN (Bayesian Belief Network), how can we score the value of each additional piece of evidence that the system ingests?
In simpler terms, here are two example use cases:
Make an app ...
- 931
0
votes
0
answers
19
views
Identification of average treatment effect in Pearl's introductory article
I'm trying to understand the identification of the average treatment effect in Pearl's introductory paper on causal inference.
We have the following structural equation model:
\begin{align}
z &= ...
- 259
0
votes
0
answers
66
views
Synthetic Control using CausalPy
I am using CausalPy (https://causalpy.readthedocs.io/en/latest/) to implement synthetic controls for Bayesian Geo-Lift.
Goal is to test a business initiative/feature (on website) in let's say 1 EU ...
5
votes
1
answer
105
views
Correcting for selection bias with standardisation/g-computation
Two sets of methods for correcting for selection bias are g-computation (standardisation) and inverse probability of censoring weighting (IPCW). I'm having a difficult time understanding how to apply ...
- 1,172
2
votes
1
answer
72
views
After obtaining a minimally sufficient set from a DAG, which variables should I include in the logistic regression model?
I have created a DAG using daggity, and from this DAG, two variables need to be controlled to evaluate the unbiased total effect of the exposure on the outcome. However, I'm confused about whether I ...
- 23
1
vote
0
answers
72
views
A question about "do" operator under unobservable confounder
The original question is here. Suppose we have a DAG in the figure.
If there is no confounder $U$, as chang_trenton point out since $S$ and $W$ happen before $X$, we have
$$
P(Y \mid do(X), S) = ...
- 203
2
votes
3
answers
74
views
Baffled by Rubin's Potential Outcomes RE: What Would Have Happened?
Background
Seeking a clear and authoritative explanation of a key concept of Rubin's Potential Outcomes Framework that is causing this hapless OP enormous grief. While the necessity to distinguish ...
- 121
0
votes
0
answers
19
views
Applied statistical propensity score matching with panel data. (Economics)
Im am working on a university project that tries to estimate the kausal impact of childbirth and and labour income. The problem can be described as follows. In theory you would need childbirth to be ...
3
votes
2
answers
123
views
Does Pr(Y | X=x) equal Pr(Y | do(X=x)) in a randomized experiment?
On page 435 of Cosma Shalizi's advanced data analysis book (link: https://www.stat.cmu.edu/~cshalizi/ADAfaEPoV/ADAfaEPoV.pdf), he states the following about randomized experiment for causal inference:
...
- 31
1
vote
0
answers
27
views
Do operator in a given DAG
The Original question is here Suppose we have a DAG in the figure.
The question to is find the decomposition for $P(Y \mid do(X), S)$. If the backdoor criteria can be applied here, then the following ...
- 11
0
votes
0
answers
41
views
What are the current guidelines for performing sensitivity analysis on matched data that is able to compare bias in reference to other covariates?
I am trying to perform sensitivity analysis on a causal inference (observational study) problem in a dataset that has a binary outcome and binary treatment. I've applied matching and g-computation ...
- 87
4
votes
1
answer
94
views
A question about do calculus
Suppose we have a DAG in the figure.
I want to find the formula for $P(Y \mid do(X), S)$. What I think is that: Since $W$ is the parent of $X$ then we should have
$$
P(Y \mid do(X), S) = \sum_{W} ...
- 203
2
votes
0
answers
39
views
Understanding the Intersection Between Causal and Statistical Inference
Assume a simple example motivating a causal research design. Say that I collect a data set on rural counties in Texas and I wish to understand if rainfall causes a change in crop sales. Working with ...
- 633