Iván Díaz
@ildiazm
Followers
2K
Following
1K
Media
24
Statuses
1K
Statistician. Associate prof. at NYU Grossman Department of Population Health. Causal inference, machine learning, and semiparametric estimation.
New York, USA
Joined July 2012
New paper and software alert! . Interested in modern mediation analysis methods with machine learning and multivariate mediators?. Take a look at this joint work with Richard Liu, @nickWillyamz , and @kara_rudolph. Short 🧵. .
1
13
64
RT @LarsvanderLaan3: Excited to share that our paper "Self-Calibrating Conformal Prediction" with @_ahmedmalaa is accepted at #NeurIPS2024!….
0
24
0
RT @pangramble_com: 👋 Hello, Word Explorers! 👋. Introducing get 3 new words every day and write a pangram using al….
0
2
0
Our Division is hosting its inaugural yearly Biostatistics Symposium, and this year the topic is Causal Inference! We have an exciting lineup of speakers listed below. If you are in the NYC area, please join us! Link to register in the QR below.
0
12
53
RT @KellyVanLancker: Interested in data-driven covariate adjustment? I’m presenting some recent work with @ildiazm and @SVansteelandt next….
0
6
0
We addressed those challenges by (1) rewriting integrals as conditional expectations using permuted versions of the original random variables, and (2) parameterizing the EIF in terms of Riesz representers, using methods from a novel strand of the literature on "Riesz learning".
0
0
2
Developing these estimators required addressing a couple of challenges, namely that (1) we required to integrate out mediators w.r.t their conditional densities, and (2) that the efficient influence function required estimation of ratios of those conditional densities.
1
0
1
Here is a link to the software on GitHub. We'd love to hear your thoughts.
1
0
2
What can we do now that we could not before? For six definitions of direct/indirect or path-specific effects, the software allows you to do mediation analysis using ML, providing valid (under assumptions) inference, and incorporating multivariate mediators and complex exposures.
1
0
3
We also discuss generalizations of modified treatment policies and stochastic interventions to define mediational effects, and propose estimators for them. This allows us to tackle problems with continuous or multivariate treatments.
1
0
1
We show that the relevant identifying functionals for six well known mediation effects can be reduced to two main statistical parameters. We then develop one-step estimators that can be coupled with ML à la DML/TMLE.
1
0
1
In this paper we take a look at some well-known and some new definitions for mediational parameters (natural and randomized direct/indirect, path-specific effects using recanting twins, organic effects, separable effects, and more).
1
0
1
Also, there are few frameworks out there that can be used to define mediational effects when the exposure is continuous or multivariate (e.g., pollutant concentrations, mixtures, doses, etc).
1
0
1
The literature on the definition and identification of parameters for mediation analysis with binary exposures is rich, but the development of estimators for cases where the mediators are multivariate or high-dimensional (e.g. omics data) has lagged behind.
1
0
1
RT @kat_hoffman_: A tutorial on Longitudinal Modified Treatment Policies-- a flexible method for defining, identifying, and estimating caus….
0
10
0
RT @StanfordAILab: arXiv -> alphaXiv. Students at Stanford have built alphaXiv, an open discussion forum for arXiv papers. @askalphaxiv. Yo….
0
2K
0
RT @prof_joe_: Sorry to miss #SER2024 but we ♥️ the lmtp package and can’t wait to see what @nickWillyamz @kara_rudolph and @ildiazm have i….
0
1
0
If you are coming to #SER2024 and are interested in learning how to define and estimate causal effects for complex exposures (continuous, multivariate, ordinal, etc) in longitudinal studies using off-the-shelf software, join us in this workshop!.
@societyforepi pals! @ildiazm @nickWillyamz &I are leading a workshop Tues pm at #SER2024 on estimating causal effects of multiple or nonbinary exposures. @nickWillyamz is an absolute wizard and made a beautiful, user-friendly workshop with lots of examples in R. Register & join!.
0
1
23
I have been for a long time trying to understand Frank’s views but every day I am more baffled. If there is rarely any treatment effect heterogeneity, why the insistence on conditional treatment effects? Shouldn’t they be equal to marginal in that case?.
@DanMarkMD Just from reading the abstract, there is nothing to budge my belief in the rarity of ACTIONABLE heterogeneity of treatment effect. Sure you can mimic data with models that allow for HTE but identifying beforehand pts likely to have large benefit is another thing.
2
0
21
RT @LarsvanderLaan3: What are the differences between one-step estimation, Double ML, and Targeted ML? . This commentary (@ildiazm) and blo….
0
22
0