Evaluating a Cash Transfer Program (RCT) with Panel Data

Downloads

Each dataset is available as a labeled Stata .dta and its source file.

⇩ Download all data (ZIP)stata_codebook.do

Run stata_codebook.do in Stata once to attach long-form per-variable notes to the .dta files.

Load directly in code

Every file loads straight from GitHub (raw URLs). Swap the file name to load any dataset.

Stata

* Stata 14+ : `use` reads an https URL directly
global BASE "https://raw.githubusercontent.com/cmg777/starter-academic-v501/master/content/post/stata_rct/data/"
use "${BASE}dataSIM4RCT.dta", clear
describe
notes

Python

!pip install -q pyreadstat
import pandas as pd
BASE = "https://raw.githubusercontent.com/cmg777/starter-academic-v501/master/content/post/stata_rct/data/"
df = pd.read_stata(BASE + "dataSIM4RCT.dta")

# load every dataset at once
files = ["dataSIM4RCT"]
data = {f: pd.read_stata(BASE + f + ".dta") for f in files}

# pyreadstat (richest metadata) reads LOCAL files -> download first
import pyreadstat, urllib.request
urllib.request.urlretrieve(BASE + "dataSIM4RCT.dta", "dataSIM4RCT.dta")
df, meta = pyreadstat.read_dta("dataSIM4RCT.dta")

Copy and paste this snippet in Google Colab app. https://colab.research.google.com/notebooks/empty.ipynb

R

# R : haven::read_dta auto-downloads an https URL
library(haven)
BASE <- "https://raw.githubusercontent.com/cmg777/starter-academic-v501/master/content/post/stata_rct/data/"
df <- read_dta(paste0(BASE, "dataSIM4RCT.dta"))

Overview & sources

Companion data for a hands-on Stata tutorial that evaluates the causal effect of a cash transfer program on household consumption. The data are fully synthetic: 2,000 households in a developing country, observed in a balanced panel across a 2021 baseline and a 2024 endline (4,000 observations). The outcome is log monthly consumption and the program raises it by a known true effect of 12% (0.12 log points). The tutorial walks from baseline-balance checks through three cross-sectional estimators — regression adjustment (RA), inverse probability weighting (IPW), and doubly robust AIPW/IPWRA — then difference-in-differences and doubly robust DiD on the panel, and an endogenous-treatment IV model for imperfect compliance. Because the ground truth is known, every estimate can be checked against 0.12.

Data sources

Cite this data

Please cite this dataset as follows.

APA

Mendez, C. (2026). Evaluating a Cash Transfer Program (RCT) with Panel Data in Stata [Data set]. https://carlos-mendez.org/post/stata_rct/

Sant'Anna, P. H. C., & Zhao, J. (2020). Doubly robust difference-in-differences estimators. Journal of Econometrics, 219(1), 101–122. https://doi.org/10.1016/j.jeconom.2020.06.003 Imbens, G. W., & Rubin, D. B. (2015). Causal Inference for Statistics, Social, and Biomedical Sciences. Cambridge University Press. StataCorp. teffects — Treatment-effects estimation. Stata Treatment-Effects Reference Manual.

BibTeX

@misc{mendez2026statarct,
  author       = {Mendez, Carlos},
  title        = {Evaluating a Cash Transfer Program (RCT) with Panel Data in Stata},
  year         = {2026},
  howpublished = {\url{https://carlos-mendez.org/post/stata_rct/}},
  note         = {Data set}
}

@article{santanna2020doubly,
  author  = {Sant'Anna, Pedro H. C. and Zhao, Jun},
  title   = {Doubly robust difference-in-differences estimators},
  journal = {Journal of Econometrics},
  volume  = {219}, number = {1}, pages = {101--122}, year = {2020},
  doi     = {10.1016/j.jeconom.2020.06.003}
}
@book{imbens2015causal,
  author    = {Imbens, Guido W. and Rubin, Donald B.},
  title     = {Causal Inference for Statistics, Social, and Biomedical Sciences},
  publisher = {Cambridge University Press}, year = {2015}
}

Variable explorer search & filter all 14 variables

Type to filter by name or label, or use the chips to filter by type. Each row shows a mini distribution. Click a header to sort.

Cross-file variable index

Which file each variable appears in (● = present).

Construction & formulas

The outcome y is log monthly consumption. The causal target is the program's average effect, with a known true value of 0.12 log points (≈12%). Two estimands recur throughout:

• ATE (policymaker's quantity): ATE = E[Y(1) − Y(0)] — the average effect if the program were scaled to everyone.
• ATT (evaluator's quantity): ATT = E[Y(1) − Y(0) | T=1] — the average effect among the treated. DiD estimates the ATT only.

Five estimation strategies are applied to these data:

• Regression adjustment (RA): fit outcome models on treated and control, impute both potential outcomes, average the difference: τ_RA = (1/N) Σ [μ̂_1(X_i) − μ̂_0(X_i)]. Consistent if the outcome model is correct (teffects ra).
• Inverse probability weighting (IPW): reweight by the inverse propensity score p̂(X)=Pr(T=1|X): τ_IPW = (1/N) Σ [T_i Y_i / p̂ − (1−T_i) Y_i / (1−p̂)]. Consistent if the treatment model is correct (teffects ipw).
• Doubly robust (AIPW / IPWRA): RA plus an IPW-weighted bias-correction term, τ_DR = τ_RA + (1/N) Σ [T_i(Y_i−μ̂_1)/p̂ − (1−T_i)(Y_i−μ̂_0)/(1−p̂)]. Consistent if either model is correct (teffects aipw / teffects ipwra).
• Difference-in-differences (DiD): the difference of pre/post changes, τ_DiD = (Ȳ_tr,post − Ȳ_tr,pre) − (Ȳ_ct,post − Ȳ_ct,pre), netting out time-invariant unobservables; estimates the ATT (xtdidregress).
• Doubly robust DiD (DR-DiD): Sant'Anna & Zhao (2020) extend the DR logic to changes ΔY, consistent if the outcome or the propensity model holds (drdid, xthdidregress aipw).

Imperfect compliance is handled by using random assignment treat as an instrument for receipt D (endogenous-treatment regression, etregress), separating the effect of the offer (ITT) from the effect of receipt.

Data-generating process. For household i, log consumption is built from a baseline level plus a household component alpha, an idiosyncratic shock eps, a common time trend, and the 0.12 treatment bump applied at endline to households that receive the transfer. The baseline value is stored as y0 and carried to both rows; wave/year/post encode the two-period time axis.

The datasets

Switch datasets with the tabs. Each shows the full variable dictionary plus a sortable statistics table with mini distributions and data coverage.

Expand all datasets expand to search (Ctrl/⌘+F) or print across all datasets

Known limitations & caveats

• Synthetic data. There is no real program behind this tutorial; values are simulated with a known 0.12 log-point effect. Results are internally consistent with the calibration but are not empirical evidence about real cash-transfer programs.
• Two periods only. The panel has a single baseline (2021) and endline (2024), so it cannot test pre-treatment (parallel) trends or estimate dynamic/staggered effects.
• Homogeneous effect by construction. The treatment effect is the same for all households, so ATE and ATT nearly coincide; real programs typically show heterogeneity worth exploring by subgroup.
• Imperfect compliance. Random assignment treat ≠ actual receipt D (85% take-up among the offered, 5% among controls); offer-based (ITT) and receipt-based estimates differ — use treat for the policy offer and the IV/DR-receipt models for the effect of receipt.
• DGP internals. alpha and eps are simulation components (household and idiosyncratic terms), and y0 is the baseline outcome carried to both rows; they are exposed for transparency and are not analysis covariates in the tutorial.