Modeling Covariate Transition for Efficient Estimation of Longitudinal Treatment Effects in Randomized Experiments

This repository contains the implementation of the ICML 2026 paper, "Modeling Covariate Transition for Efficient Estimation of Longitudinal Treatment Effects in Randomized Experiments ," by Naoki Chihara, Tatsushi Oka, Yasuko Matsubara, Yasushi Sakurai, and Shota Yasui.

Usage

1. Clone this repository.

   git clone https://github.com/C-Naoki/dynamic-adjustment.git

2. Construct a virtual environment and install the required packages.

   make install

   • Note that it requires pyenv and uv.
   • If you prefer not to use them, you can also use requirements.txt created based on pyproject.toml.

   Specifically, the above command performs the following steps:

   1. if necessary, install Python 3.11.13 using pyenv, and then switch to this version.
   2. create a virtual environment based on Python 3.11.13
   3. install packages in pyproject.toml.
   4. attach the path file (i.e., *.pth) in the site-packages/ for extending module search path.

   Please check the Makefile for more details.

3. Run quick demo.

   bash bin/demo.sh

   If you want the command to continue running after logging out, you use -n option as shown below (using nohup).

   bash bin/demo.sh -n

   • The execution log is saved in nohup/ directory.

Minimal Example

import numpy as np

from src.models.dynamicra.model import DynamicRAEstimator
from src.utils import build_arrays


X, Y, W, T = build_arrays(data)
estimator = DynamicRAEstimator(
    method='dynamicra',
    regmodel_name='ols',
    transmodel_name='var',
)

results = estimator.estimate(
    X=X,
    Y=Y,
    W=W,
    times=list(range(T)),
    treatment_path=np.ones(T),
    control_path=np.zeros(T),
    variance_type='moment',
)

for result in results:
    print(
        f"t={result['time']}: "
        f"ATE={result['mu']:.3f}, "
        f"95% CI=({result['ci_lower']:.3f}, {result['ci_upper']:.3f})"
    )

Input Data Format

It expects a long-format pandas.DataFrame with the following columns:

Column	Default name	Type	Description
Unit ID	id	int	Experimental unit identifier.
Time	date	int	Integer time index from 0 to T - 1.
Feature	feature	str	Variable name, e.g., Y, W, X1, ...
Value	value	float	Numeric value of that feature.

Required features are:

• Y: outcome.
• W: treatment assignment, encoded as integer arms such as 0 and 1.
• Covariates: all non-Y/W features are used.

Example:

id,date,feature,value
1,0,Y,2.31
1,0,W,0
1,0,X1,0.42
1,0,X2,-1.10
1,1,Y,2.76
1,1,W,1
1,1,X1,0.50
1,1,X2,-0.87

Citation

If you use this code for your research, please consider citing our paper.