@yiqingxu.bsky.social
We also highlighted the need to use the leave-one-out approach to obtain pretrend estimates, thanks to Zikai and Anton's cautionary note: osf.io/ngr3d_v1/ @astrezh.bsky.social
August 25, 2025 at 8:56 PM
We also highlighted the need to use the leave-one-out approach to obtain pretrend estimates, thanks to Zikai and Anton's cautionary note: osf.io/ngr3d_v1/ @astrezh.bsky.social
In this new version, we fixed many bugs and introduced more features.
Check it out: yiqingxu.org/packages/fect/
Special thanks to Ziyi, Rivka, and Tianzhu for their incredible work and dedication. More features are on the way!
Check it out: yiqingxu.org/packages/fect/
Special thanks to Ziyi, Rivka, and Tianzhu for their incredible work and dedication. More features are on the way!
August 25, 2025 at 8:56 PM
In this new version, we fixed many bugs and introduced more features.
Check it out: yiqingxu.org/packages/fect/
Special thanks to Ziyi, Rivka, and Tianzhu for their incredible work and dedication. More features are on the way!
Check it out: yiqingxu.org/packages/fect/
Special thanks to Ziyi, Rivka, and Tianzhu for their incredible work and dedication. More features are on the way!
Thrilled to share that **fect** has won the 2025 Best Statistical Software Award from the Society of Political Methodology. We're honored!
polmeth.org/statistical-...
To celebrate, we've just released fect v2.0.5 on CRAN & Github 🎉
polmeth.org/statistical-...
To celebrate, we've just released fect v2.0.5 on CRAN & Github 🎉
August 25, 2025 at 8:56 PM
Thrilled to share that **fect** has won the 2025 Best Statistical Software Award from the Society of Political Methodology. We're honored!
polmeth.org/statistical-...
To celebrate, we've just released fect v2.0.5 on CRAN & Github 🎉
polmeth.org/statistical-...
To celebrate, we've just released fect v2.0.5 on CRAN & Github 🎉
7/ We have prototypes for all the algorithms used in the Element and will soon roll them out in **interflex** for R.
Comments and suggestions are more than welcome!
Comments and suggestions are more than welcome!
April 3, 2025 at 5:25 AM
7/ We have prototypes for all the algorithms used in the Element and will soon roll them out in **interflex** for R.
Comments and suggestions are more than welcome!
Comments and suggestions are more than welcome!
6/ Through simulations & many empirical examples, we find AIPW-Lasso & PO-Lasso outperforming competitors with political science data, while DML requires much larger samples to do better.
April 3, 2025 at 5:25 AM
6/ Through simulations & many empirical examples, we find AIPW-Lasso & PO-Lasso outperforming competitors with political science data, while DML requires much larger samples to do better.
5/ We then turn to double machine learning (DML), incorporating modern learners such as neural nets, random forests, and HGB to estimate nuisance parameters and construct signals.
April 3, 2025 at 5:25 AM
5/ We then turn to double machine learning (DML), incorporating modern learners such as neural nets, random forests, and HGB to estimate nuisance parameters and construct signals.
4/ The core of the Element is robust estimation strategies.
First, we adapt AIPW-Lasso & Partialing-Out Lasso, both w/ basis expansion & Lasso double selection.
We walk through signal construction step-by-step to aid intuition. For smoothing, we support both kernel and B-spline regressions.
First, we adapt AIPW-Lasso & Partialing-Out Lasso, both w/ basis expansion & Lasso double selection.
We walk through signal construction step-by-step to aid intuition. For smoothing, we support both kernel and B-spline regressions.
April 3, 2025 at 5:25 AM
4/ The core of the Element is robust estimation strategies.
First, we adapt AIPW-Lasso & Partialing-Out Lasso, both w/ basis expansion & Lasso double selection.
We walk through signal construction step-by-step to aid intuition. For smoothing, we support both kernel and B-spline regressions.
First, we adapt AIPW-Lasso & Partialing-Out Lasso, both w/ basis expansion & Lasso double selection.
We walk through signal construction step-by-step to aid intuition. For smoothing, we support both kernel and B-spline regressions.
3/ We start by defining the estimand, the CME, and presenting main identification results in the discrete-covariate case.
We then review & improve the semiparametric kernel estimator. The improvements include fully moderated models, adaptive kernels, and uniform confidence intervals.
We then review & improve the semiparametric kernel estimator. The improvements include fully moderated models, adaptive kernels, and uniform confidence intervals.
April 3, 2025 at 5:25 AM
3/ We start by defining the estimand, the CME, and presenting main identification results in the discrete-covariate case.
We then review & improve the semiparametric kernel estimator. The improvements include fully moderated models, adaptive kernels, and uniform confidence intervals.
We then review & improve the semiparametric kernel estimator. The improvements include fully moderated models, adaptive kernels, and uniform confidence intervals.
2/ We prepared it for Cambridge Element & previewed parts of it in our response to a blog post last month.
Scholars are often interested in how treatment effects vary with a moderating variable (example below).
We hope this will serve as a useful reference for this common task down the road.
Scholars are often interested in how treatment effects vary with a moderating variable (example below).
We hope this will serve as a useful reference for this common task down the road.
April 3, 2025 at 5:25 AM
2/ We prepared it for Cambridge Element & previewed parts of it in our response to a blog post last month.
Scholars are often interested in how treatment effects vary with a moderating variable (example below).
We hope this will serve as a useful reference for this common task down the road.
Scholars are often interested in how treatment effects vary with a moderating variable (example below).
We hope this will serve as a useful reference for this common task down the road.
Draft “A Practical Guide to Estimating Conditional Marginal Effects: Modern Approaches” is on arXiv: arxiv.org/pdf/2504.01355
w/ two amazing grad students, Jiehan_Liu & Ziyi Liu 🧵
w/ two amazing grad students, Jiehan_Liu & Ziyi Liu 🧵
April 3, 2025 at 5:25 AM
Draft “A Practical Guide to Estimating Conditional Marginal Effects: Modern Approaches” is on arXiv: arxiv.org/pdf/2504.01355
w/ two amazing grad students, Jiehan_Liu & Ziyi Liu 🧵
w/ two amazing grad students, Jiehan_Liu & Ziyi Liu 🧵
5/ We then turn to double machine learning (DML), incorporating modern learners such as neural nets, random forests, and HGB to estimate nuisance parameters and construct signals.
April 3, 2025 at 3:14 AM
5/ We then turn to double machine learning (DML), incorporating modern learners such as neural nets, random forests, and HGB to estimate nuisance parameters and construct signals.
4/ The core of the Element is robust estimation.
First, we adapt AIPW-Lasso & partialing-out Lasso (PO-Lasso), both w/ basis expansion & Lasso double selection.
We walk through signal construction step by step to aid intuition. For smoothing, we support both kernel and B-spline regressions.
First, we adapt AIPW-Lasso & partialing-out Lasso (PO-Lasso), both w/ basis expansion & Lasso double selection.
We walk through signal construction step by step to aid intuition. For smoothing, we support both kernel and B-spline regressions.
April 3, 2025 at 3:14 AM
4/ The core of the Element is robust estimation.
First, we adapt AIPW-Lasso & partialing-out Lasso (PO-Lasso), both w/ basis expansion & Lasso double selection.
We walk through signal construction step by step to aid intuition. For smoothing, we support both kernel and B-spline regressions.
First, we adapt AIPW-Lasso & partialing-out Lasso (PO-Lasso), both w/ basis expansion & Lasso double selection.
We walk through signal construction step by step to aid intuition. For smoothing, we support both kernel and B-spline regressions.
3/ We start by defining the estimand, the CME, and presenting the main identification result.
We then review & improve the semiparametric kernel estimator by introducing fully moderated models, adaptive kernels, and uniform confidence intervals.
We then review & improve the semiparametric kernel estimator by introducing fully moderated models, adaptive kernels, and uniform confidence intervals.
April 3, 2025 at 3:14 AM
3/ We start by defining the estimand, the CME, and presenting the main identification result.
We then review & improve the semiparametric kernel estimator by introducing fully moderated models, adaptive kernels, and uniform confidence intervals.
We then review & improve the semiparametric kernel estimator by introducing fully moderated models, adaptive kernels, and uniform confidence intervals.
We prepared this for Cambridge Element and previewed parts of it in our response to a blog post last month.
Scholars are often interested in how treatment effects vary with a moderating variable (example below). We hope this Element will serve as a useful reference for such tasks down the road.
Scholars are often interested in how treatment effects vary with a moderating variable (example below). We hope this Element will serve as a useful reference for such tasks down the road.
April 3, 2025 at 3:14 AM
We prepared this for Cambridge Element and previewed parts of it in our response to a blog post last month.
Scholars are often interested in how treatment effects vary with a moderating variable (example below). We hope this Element will serve as a useful reference for such tasks down the road.
Scholars are often interested in how treatment effects vary with a moderating variable (example below). We hope this Element will serve as a useful reference for such tasks down the road.
We streamlined six new DID-like estimators and created this tutorial for implementation in R.
yiqingxu.org/packages/fec...
Hope you no longer need to spend months figuring out what these estimators are and how to use them.
yiqingxu.org/packages/fec...
Hope you no longer need to spend months figuring out what these estimators are and how to use them.
February 21, 2025 at 4:41 AM
We streamlined six new DID-like estimators and created this tutorial for implementation in R.
yiqingxu.org/packages/fec...
Hope you no longer need to spend months figuring out what these estimators are and how to use them.
yiqingxu.org/packages/fec...
Hope you no longer need to spend months figuring out what these estimators are and how to use them.
7/ Based on the discussion, we propose a set of recommendations, starting from clearly stating the quantity of interest and the key ID & modeling assumptions.
We provide software support for various estimation strategies: yiqingxu.org/packages/int...
We provide software support for various estimation strategies: yiqingxu.org/packages/int...
February 11, 2025 at 4:14 PM
7/ Based on the discussion, we propose a set of recommendations, starting from clearly stating the quantity of interest and the key ID & modeling assumptions.
We provide software support for various estimation strategies: yiqingxu.org/packages/int...
We provide software support for various estimation strategies: yiqingxu.org/packages/int...
6/ In contrast, Simonsohn’s proposed implementation of GAM is not well suited for estimating the CME. The method estimates the CAPE given a specific value of the treatment D = d, and we show that the estimated CAPE varies depending on the choice of d.
February 11, 2025 at 4:14 PM
6/ In contrast, Simonsohn’s proposed implementation of GAM is not well suited for estimating the CME. The method estimates the CAPE given a specific value of the treatment D = d, and we show that the estimated CAPE varies depending on the choice of d.
5/ (3) We review recent advancements in the causal inference literature and highlight doubly robust and double/debiased machine learning estimators as appealing strategies for estimating the CME. Most of these methods are already supported by the *interflex* package in R.
February 11, 2025 at 4:14 PM
5/ (3) We review recent advancements in the causal inference literature and highlight doubly robust and double/debiased machine learning estimators as appealing strategies for estimating the CME. Most of these methods are already supported by the *interflex* package in R.
4/ (2) We show that the kernel estimator accurately recovers the true CME in the scenarios presented in the critique. The binning estimator, as a diagnostic tool, also performs reasonably well.
February 11, 2025 at 4:14 PM
4/ (2) We show that the kernel estimator accurately recovers the true CME in the scenarios presented in the critique. The binning estimator, as a diagnostic tool, also performs reasonably well.
3/ The critique compares the estimated CME to the wrong benchmark, leading to misleading conclusions. For example, in its simulation, the benchmark CME is not *zero* (as incorrectly stated) but increases with the moderator X because X and the treatment D are positively correlated.
February 11, 2025 at 4:14 PM
3/ The critique compares the estimated CME to the wrong benchmark, leading to misleading conclusions. For example, in its simulation, the benchmark CME is not *zero* (as incorrectly stated) but increases with the moderator X because X and the treatment D are positively correlated.
2/ While the critique is intriguing, it is fundamentally flawed for three reasons. (1) the causal estimand & ID assumptions are not clearly defined.
Most applied research focuses on estimating the CME, whereas the critique centers on CAPE, an estimand rarely used in applied work.
Most applied research focuses on estimating the CME, whereas the critique centers on CAPE, an estimand rarely used in applied work.
February 11, 2025 at 4:14 PM
2/ While the critique is intriguing, it is fundamentally flawed for three reasons. (1) the causal estimand & ID assumptions are not clearly defined.
Most applied research focuses on estimating the CME, whereas the critique centers on CAPE, an estimand rarely used in applied work.
Most applied research focuses on estimating the CME, whereas the critique centers on CAPE, an estimand rarely used in applied work.
Looks like everyone is here 😆
1/ Recently, Professor Uri Simonsohn critiqued Hainmueller, Mummolo & Xu (2019), arguing that the proposed methods fail to recover the conditional marginal effect (CME): datacolada.org/121
We appreciate the critique and offer this response: arxiv.org/pdf/2502.05717 🧵
1/ Recently, Professor Uri Simonsohn critiqued Hainmueller, Mummolo & Xu (2019), arguing that the proposed methods fail to recover the conditional marginal effect (CME): datacolada.org/121
We appreciate the critique and offer this response: arxiv.org/pdf/2502.05717 🧵
February 11, 2025 at 4:14 PM
Looks like everyone is here 😆
1/ Recently, Professor Uri Simonsohn critiqued Hainmueller, Mummolo & Xu (2019), arguing that the proposed methods fail to recover the conditional marginal effect (CME): datacolada.org/121
We appreciate the critique and offer this response: arxiv.org/pdf/2502.05717 🧵
1/ Recently, Professor Uri Simonsohn critiqued Hainmueller, Mummolo & Xu (2019), arguing that the proposed methods fail to recover the conditional marginal effect (CME): datacolada.org/121
We appreciate the critique and offer this response: arxiv.org/pdf/2502.05717 🧵