This study explores public perceptions of Bitcoin prices and the factors shaping them using data from the Bitcoin omnibus survey conducted by the Bank of Canada from 2017 to 2021. Through regression analysis and Oaxaca–Blinder decomposition, I examine differences in price expectations between Bitcoin owners and nonowners. Additionally, I investigate how demographic characteristics, Bitcoin knowledge, and financial literacy influence these views. My findings reveal significant disparities, with owners consistently more optimistic about future prices than nonowners. The Oaxaca–Blinder decomposition shows that only a small portion of this gap is explained by observable characteristics, suggesting the presence of unobserved influences. Bitcoin knowledge emerges as a key explanatory variable, accounting for much of the explained difference, while demographic factors—such as age, gender, and education—also play important roles.

This presentation extends the literature on firm dynamics by incorporating ownership networks and financing in the study of firm growth. I observe co-ownership connections for the universe of privately owned Ecuadorian manufacturing firms between 2000 and 2019. The structure of my data allows to construct ownership network variables and determine their impact on firm growth in a quantile fixed-effect dynamic regression framework. This approach uncovers the heterogeneous impact of firm age on firm growth across the entire conditional firm-growth distributions and statistically significant leverage and network effects. The relationship between firm growth and leverage remains positive with the inclusion of ownership networks. For young firms, the results indicate that there is no significant relationship between age and growth. This result suggests that financial variables continue to matter and that ownership networks capture alternative aspects of firm dynamics that have not been previously acknowledged.

Most Stata users' programming skills are focused on the development of do-files that make use of Stata's many features for automation of data management, statistics, and graphics. Do-file programming is a valuable skill because it is the foundation of reproducible research—an important issue in every discipline. Many Stata users may not have taken advantage of Stata's development tools to take the next step: the construction of ado-files, or Stata programs, to further automate frequent tasks. A key feature of the Stata programming language is its integration with Mata. Many users say that some day, they will figure out what Mata might do for them. That day should be today, given the many important capabilities that Mata provides, including easy handling of matrices, speed improvements for computationally demanding tasks, and the ease of integration between Stata and Mata workspaces. This presentation will provide several worked examples of ado-file and Mata programming.

Causal inference aims to identify and quantify a causal effect. With traditional causal inference methods, we can estimate the overall effect of a treatment on an outcome. When we want to better understand a causal effect, we can use causal mediation analysis to decompose the effect into a direct effect of the treatment on the outcome and an indirect effect through another variable, the mediator. Causal mediation analysis can be performed in many situations—the outcome and mediator variables may be continuous, binary, or count, and the treatment variable may be binary, multivalued, or continuous.

In this presentation, I will introduce the framework for causal mediation analysis and demonstrate how to perform this analysis with the mediate command. Examples will include various combinations of outcome, mediator, and treatment types.

I introduce a Stata package for the difference-in-differences estimator for unpoolable data (UN-DID), designed for settings where data from treatment and control units are partitioned across silos that cannot be pooled because of legal, technical, or institutional constraints. UN-DID estimates the average treatment effect on the treated (ATET) by computing within-silo pre- and post differences and aggregating these across silos. The estimator is unbiased under parallel trends across silos and accommodates both common and staggered adoption. For inference, I implement a randomization-based procedure and a jackknife standard-error estimator that remains valid under treatment-effect heterogeneity and variation in treatment timing. To support implementation in applied work, I developed a three-stage Stata interface (undid) for executing the UN-DID procedure. Users begin by specifying unit identifiers and treatment timings in a simple initialization file. The software then prepares the required difference calculations within each silo, followed by a final estimation stage that calculates ATETs and computes p-values. The interface supports optional covariate adjustment and produces descriptive statistics and identification diagnostics. The Stata package is designed for collaborative environments where direct pooling of data across treatment and control units is not possible, such as multijurisdictional research settings or contexts with confidentiality constraints.

Monte Carlo simulations in Stata are often constrained by the software’s memory architecture, particularly when the total number of replications required for inference or robustness checks is large. As memory consumption accumulates over the course of a simulation, performance can degrade severely, with many replications failing because of insufficient available RAM. This poster presents a procedure that bypasses these constraints by dividing the full simulation task into smaller, memory-manageable batches, which are executed independently in separate Stata sessions. The method relies on partitioning the total number of replications, \(R\), into \(B\) batches of \(r\) replications each, where \(R=B×r\). Each batch is encoded in a distinct Stata do-file, generated automatically via a short Python script. These batch files are then executed sequentially or in parallel using a Bash shell script. Because each batch runs in its own instance of Stata, memory usage is reset between runs, preventing the accumulation of data across replications. This approach allows simulations that were previously infeasible because of RAM limitations to run to completion. In addition to resolving memory constraints, the method enables embarrassingly parallel computation on multicore machines without requiring any specialized parallel-processing software. By assigning different batch files to different processor cores via concurrent shell calls, total run time can be substantially reduced. After a brief setup phase involving preprocessing and batch generation, the entire simulation can be launched with a single command. The proposed workflow improves the feasibility and efficiency of large-scale Monte Carlo experiments in Stata, especially in environments with modest hardware and limited software support for parallelization.

I introduce didint, a Stata wrapper for the Julia-based DiDInt.jl package, which implements a recent extension of the interaction difference-in-differences estimator to account for covariates. The method, proposed by Karim and Webb (2024), addresses bias that can arise when adjusting for covariates in staggered adoption settings—especially when the common causal covariates (CCC) assumption is violated. didint estimates the average treatment effect on the treated (ATET) by applying a regression-based residualization approach that allows for state-level, time-level, or fully interacted (state-by-time) varying controls. The command supports multiple specifications via the ccc() option, enabling researchers to flexibly compare assumptions about the role of covariates in the data-generating process. In addition to covariate handling, didint implements both a cluster jackknife and a randomization inference procedure, allowing users to construct cluster–robust p-values, especially when few units are treated or treatment timing is concentrated. The interface is designed for flexibility in applied research: users specify outcome, time, and state variables, treatment timing, and optional control variables. Additional options include custom time frequencies, automatic cohort length adjustment, and full compatibility with panel data. Estimates and inference results (ATETs, standard errors, and p-values) are returned directly to the active Stata dataset for immediate use. This presentation will briefly describe the underlying methodology and demonstrate its application to empirical examples. didint equips Stata users with a robust and principled framework for conducting difference-in-differences analyses with covariates in staggered treatment designs.

This project aims to examine the relationship between women's household bargaining power and their adoption of modern contraception in Zambia, using the 2018 DHS survey data. Relying on direct measures of women's bargaining power (as indicated by the preexisting literature), which include a woman's ability to make decisions about her own healthcare, large household purchases, small household purchases, visits to her family and friends, and contraceptive use, as well as measures of her autonomous financial capability. This measure of financial capability is then interacted with a woman's ability to make healthcare decisions solely or jointly with her husband, to shed additional light on the influence that bargaining power has on the uptake of modern contraceptive methods. Having both financial capability and the sole ability to make healthcare decisions for herself increases a woman's probability of adopting modern contraception.

The bta2score command (short for “beta to score”) is a Stata utility designed to transform postestimation regression coefficients into a practical, interpretable scoring system for immediate clinical decision-making, particularly in time-sensitive or urgent settings. The command accepts outputs from commonly used regression procedures—namely, regress, glm, logistic, logit, and stcox. Its primary function is to generate rounded item scores and a total score from multivariable model coefficients. This transformation enables clinicians to apply the scoring system without needing computational tools, making it suitable for bedside use. Importantly, bta2score is also compatible with ordinal logistic regression models by requiring only one odds ratio per predictor, avoiding the complexity of category-specific estimates. The command handles various variable types, including continuous, binary, indicator (i.), and categorical predictors, by converting their coefficients into discrete item scores—typically as integers or scores with one decimal precision (for example, 0.5). These item scores are then summed into a single total score. The algorithm consists of three main steps: (1) division—each coefficient is divided by the smallest coefficient in absolute value, preserving the sign of negative coefficients; (2) rounding—the resulting values are rounded to the nearest user-specified unit (for example, 1 or 0.5); and (3) summation—the rounded scores are summed into a new variable representing the total score. This scoring system mirrors the simplicity and practicality of established clinical tools such as the Alvarado Score, APGAR Score, Glasgow Coma Scale, and Ottawa Ankle Rule. By creating transparent, interpretable clinical scores from regression models, bta2score facilitates evidence-based, real-time decisions without the need for digital devices—enhancing clinical workflow and supporting resource-limited environments.

This study investigates the empirical relevance of the Phillips Curve in Liberia from 2001 to 2023, addressing a critical research gap in fragile, low-income economies. Despite extensive global literature, Liberia’s inflation–unemployment dynamics remain understudied amid persistent macroeconomic volatility and structural labor market weaknesses. This study employs a multimethod econometric framework, including OLS, robust and quantile regressions, and vector autoregressive (VAR) models, to evaluate the interplay between inflation, unemployment, money supply, and exchange rate. Granger causality, impulse–response, and variance decomposition techniques reinforce the analysis, revealing strong, bidirectional feedback between exchange rate movements and inflation. Findings show a weak but negative short-run relationship between inflation and unemployment, broadly validating the Phillips Curve hypothesis. Exchange rate depreciation consistently emerges as the primary driver of inflation, while money supply exhibits an unexpected but statistically significant deflationary effect. Interaction terms suggest the inflation–unemployment relationship is conditioned by macroeconomic context. The study concludes that inflation control in Liberia requires exchange rate stabilization, targeted structural reforms, and employment-sensitive policies. These findings challenge monetarist orthodoxy and highlight the need for integrated, context-specific macroeconomic strategies in postconflict settings.

Although cluster–robust standard errors are widely used, they can sometimes yield very unreliable inferences. Tests and confidence intervals based on the usual (CV1) standard errors are known to work poorly in certain circumstances, such as when there are few clusters, few treated clusters, or clusters that vary greatly in size or other features. Numerous methods have been proposed to obtain more reliable inferences. These include alternative standard errors, such as ones based on the cluster jackknife (CV3), nonstandard critical values, and bootstrap methods. I discuss what we have learned from the recent literature and attempt to provide some guidance for how to deal with cases where alternative methods yield conflicting results. The talk focuses on linear regression models, but logit models will also be discussed.

This presentation introduces two Stata packages: twowayjack and csdidjack, which implement jackknife-based inference for models with clustered data. The twowayjack command provides robust inference for OLS regression models with two-way clustering, such as by unit and time. It implements the CV3 standard errors from MacKinnon, Nielsen, and Webb (2024), which are designed to remain valid even with few clusters in one or both dimensions. These estimators omit one-cluster resampling to account for dependence across both clustering dimensions. The csdidjack command applies these ideas to the csdid estimator of Callaway and Sant'Anna (2021), providing improved inference for average treatment effects on the treated (ATET) in staggered adoption settings. It also supports jackknife-based CV3 inference for calendar-time, cohort-based, and ATET_gt effects. The underlying methodology is described in MacKinnon, Nielsen, Webb, and Karim (2025), which extends jackknife and bootstrap inference to this setting. Both tools offer practical solutions for empirical researchers facing clustered data and limited numbers of clusters. The packages are freely available on GitHub as community-contributed Stata commands: twowayjack and csdidjack.

aaaft_rils implements a simulation-based rank inference procedure to estimate the effect of a treatment or risk variable using projected residuals and rank statistics. The method is robust to nonlinearity and does not require classical distributional assumptions such as normality or homoskedasticity.

The mapsm command (multiple arms propensity-score matching) is a Stata utility designed to perform stratified propensity-score matching for studies with two or more treatment arms. For two-arm studies, it uses binary logistic regression to generate a propensity score (0–1), which is divided into 10 strata. Subjects are matched 1:1 within each stratum between treatment groups. For studies with three or more arms, mapsm applies multinomial logistic regression to estimate a set of propensity scores per subject, reflecting the likelihood of assignment to each treatment. Each set is stratified into 10 intervals, and matched sets are formed across arms in equal ratios (for example, 1:1:1). To enhance postmatching comparability, mapsm iterates the matching process with varying random seeds and selects the best result based on balance diagnostics such as standardized mean differences. The best seed is reported for reproducibility, ensuring optimal balance across treatment groups.

TBD

Contribute to the Stata community by sharing your feedback with StataCorp's developers. From feature improvements to bug fixes and new ways to analyze data, we want to hear how Stata can be made better for our users.