Skip to main content
51 votes
Accepted

Propensity score matching - What is the problem?

It's true that there are not only other ways of performing matching but also ways of adjusting for confounding using just the treatment and potential confounders (e.g., weighting, with or without ...
Noah's user avatar
  • 35k
32 votes
Accepted

Strong ignorability: confusion on the relationship between outcomes and treatment

I'll try to break it down a bit.. I think most of the confusion when studying potential outcomes (ie $Y_0,Y_1$) is to realize that $Y_0,Y_1$ are different than $Y$ without bringing in the covariate $X$...
doubled's user avatar
  • 4,977
29 votes
Accepted

Which Theories of Causality Should I know?

Strictly speaking, "Granger causality" is not at all about causality. It's about predictive ability/time precedence, you want to check whether one time series is useful to predict another ...
Carlos Cinelli's user avatar
22 votes

Strong ignorability: confusion on the relationship between outcomes and treatment

Doubled has a fantastic answer, but I wanted to follow up with some intuitions that have helped me. First, think of potential outcomes as pre-treatment covariates. I know this seems like a strange ...
Noah's user avatar
  • 35k
18 votes

Unconfoundedness in Rubin's Causal Model- Layman's explanation

How would you describe the uncoundedness/ignorability assumption to somebody who has not studied the RCM? Regarding intuition to somebody not versed in causal inference, I think this is where you ...
Carlos Cinelli's user avatar
17 votes
Accepted

Panel data diff-in-diff and the pattern of the binary treatment indicator

I will assume you have a thorough grasp of the two group/two period difference-in-differences (DD) design and you now want to extend your intuition of the method to the multi-group/multi-period case. ...
Thomas Bilach's user avatar
16 votes
Accepted

When is it valid to use race/ethnicity in causal inference?

Race and ethnicity are variables that cannot be "controlled" in experiments, since it is not possible for the researcher to assign or change this characteristic of the study participant.$^\...
Ben's user avatar
  • 129k
16 votes
Accepted

What are the use cases for Propensity Score Matching?

You need to distinguish between uses of propensity scores for matching of cases versus for more general adjustments. The discussion on this page suggests that there isn't much of a use case for ...
EdM's user avatar
  • 95.8k
14 votes

Propensity score matching - What is the problem?

@Noah's answer is superb and qualifies as a mini review article. To me, the severe problems with PS matching are topped off by (1) it does not represent reproducible research in that the choice of ...
Frank Harrell's user avatar
14 votes
Accepted

What are some examples when the Average Treatment Effect on the Treated/Control (ATT,ATC) is more sought after than the ATE?

I'm writing a paper about this very topic, so I'll just summarize here and update with a link to the paper when it's ready. (Edit: Here is the arxiv version.) In short, the ATE, ATT, and ATC can be ...
Noah's user avatar
  • 35k
13 votes

What are the use cases for Propensity Score Matching?

Propensity score (PS) analysis has many problems in general, and matching is especially problematic. I prefer covariate adjustment for a spline function of the logit of PS if you need propensity ...
Frank Harrell's user avatar
11 votes
Accepted

In causal inference in statistics, how do you interpret the consistency assumption in mathematical terms?

Let me use $X$ for the treatment, $Y$ for the observed outcome and $Y(x)$ for the potential outcome under $X = x$. Consistency means that for an individual $i$, his observed outcome $Y_i$ when $X_i =...
Carlos Cinelli's user avatar
11 votes
Accepted

How exactly to evaluate Treatment effect after Matching?

The documentation for Matching is sadly fairly incomplete, leaving what it does quite mysterious. What is clear is that it takes a different approach from Stuart (...
Noah's user avatar
  • 35k
10 votes

Difference between marginal and conditional treatment effect? Relating to regression vs. propensity score methods

What do marginal and conditional relate to? Assuming the treatment effects are accurately estimated, the conditional treatment effect relates to the estimated effect on an individual whereas the ...
John G's user avatar
  • 101
10 votes
Accepted

Is Coarsened Exact Matching superior to other matching methods in case-control studies?

CEM does not allow you to estimate the ATE. This is because the matched units in each treatment group will not resemble the overall sample. If no treated units are unmatched, you can estimate the ...
Noah's user avatar
  • 35k
9 votes
Accepted

Is parallel trends assumption necessary in difference-in-differences analysis?

The contrast (i.e., estimand) of interest in diff-in-diff is $\color{red}{E[Y^1_{post}|A=1]} - \color{blue}{E[Y^0_{post}|A=1]}$, which relies on the unobserved quantity $\color{blue}{E[Y^0_{post}|A=1]}...
Noah's user avatar
  • 35k
9 votes
Accepted

Isn't strong ignorability an incorrect assumption in complex causal structures?

The assumption of strong ignorability is that there exists a set of variables $W$, possibly a subset of all measured variables $V$, such that $Y^X \perp X \mid W$. It does not say that $Y^X \perp X \...
Noah's user avatar
  • 35k
8 votes
Accepted

Analysis strategy for rare outcome with matching

Removing good data from an analysis is scientifically suspect in my humble opinion, and naive matching methods are inefficient. It may be very easy to adjust for patient characteristics using ...
Frank Harrell's user avatar
7 votes
Accepted

Adjusting for baseline as covariate in observational studies

Adding baseline as a covariate is statistically acceptable - or in fact advisable - in observational studies, as well as RCTs. It is typically just not sufficient to ensure valid inference, and ...
Björn's user avatar
  • 33.6k
7 votes
Accepted

Dynamic treatment timing in a panel-DiD framework

You construct the policy dummy the way you first describe it, i.e. create a column of zeroes. Then for each firm you replace this with ones if a firm is in the treatment group AND it is in the post-...
Andy's user avatar
  • 19.3k
7 votes

Analysis strategy for rare outcome with matching

Based on the comments and the availability of such a large control group, I would probably advise to do in a step first exact matching on age groups and sex, and perhaps common disease groups. Hereby, ...
Arne Jonas Warnke's user avatar
7 votes
Accepted

IPTW for multiple treatments

You'll want to check out McCaffrey et al. (2013) for advice on this, not Austin & Stuart (2015), which is for binary treatments only. It's not clear to me which causal estimand you want, so I'll ...
Noah's user avatar
  • 35k
7 votes
Accepted

Is Bayesian estimation useful for causal analyses?

While you say we want unbiased estimators of the causal effect, generally we are interested in obtaining an accurate/precise estimate of a quantity of interest. When offered a range of estimators to ...
Richard Hardy's user avatar
7 votes
Accepted

When controlling for confounders in a causal study, should we always expect a decrease in the treatment effect estimate?

Yes, the treatment effect can go up after adjusting for confounding. Take the simple example of a continuous treatment ($A$), continuous confounder ($Z$), and continuous outcome ($Y$). If $Z$ is ...
Lachlan's user avatar
  • 1,192
6 votes
Accepted

For treatment assignment, what is the difference between Bernoulli assignment vs. completely randomized assignment?

Let's imagine you have a group of $n$ people and you want to separate them between treatment and control groups. Bernoulli trials In a bernoulli assignment, you consider each person individually ...
Carlos Cinelli's user avatar
6 votes
Accepted

Conditional treatment effect and average treatment effect under no unmeasured confounders (ignorability)

CATE = ATE when there is no treatment effect modification. If you conduct a study in a rural area and attempt to generalize to a population of rural people, then conduct another study in an urban area ...
Noah's user avatar
  • 35k
6 votes
Accepted

Propensity Scores: What is this estimator?

Let's switch from $r_t$ to the more standard notation $Y_t$, which is the potential outcome corresponding to setting treatment $z$ to $t$. The whole point of propensity score analysis is to estimate a ...
Noah's user avatar
  • 35k
6 votes
Accepted

How to adjust for the confounder of a confounder and how to call the confounder of a confounder within treatment effect estimation?

Thank you for including the DAG! The answer here is pretty straight-forward: you simply condition on both $C$ and $B.$ Neither $C$ nor $B$ is part of a collider, so you're not opening up new paths by ...
Adrian Keister's user avatar
6 votes

In clinical trials, what is the benefit of using a composite rather than individual outcome

Suppose you have a new treatment that you expect to reduce the risk of heart attack (acute MI), but that you worry might cause an increase in deaths from non-cardiovascular causes. This was exactly ...
Thomas Lumley's user avatar

Only top scored, non community-wiki answers of a minimum length are eligible