This talk provides an introduction to batched bandit experiments. I will discuss how to simulate, interactively run, and analyze batched bandit experiments using the Stata program bbandits. We will discuss results from Monte Carlo simulations and study how to obtain valid statistical inference and correct coverage and discuss a wide range of statistics and illustrations to analyze adaptively collected data. The objective is to learn how to implement you're own batched bandit experiments.

In this presentation, we will present approaches that can be used by Stata users to customize both Stata and community-contributed commands to suit their specific needs. We have been helping customers with Stata at SDAS and have seen many customers interested in community-contributed commands or changing approaches to Stata commands but not quite able to.

The challenge of reproducing economics research has gained increased attention with the growing advocacy for open science in the field. Economics journals and research institutions are quickly adopting reproducibility guidelines, requiring authors to provide code and data for reproducing results and ensuring the trustworthiness of their findings.

This presentation delves into the intricacies of achieving reproducibility in Stata works. Since the launch of the World Bank's Reproducible Research Repository, the team has conducted reproducibility verifications and curated reproducibility packages for almost 200 working papers and reports from diverse research teams in the organization, building up a valuable and novel experience into addressing common issues that break reproducibility in Stata analyses. I will present an overview of the workflows and tools the team has developed in response to identified reproducibility challenges in typical Stata works, covering key topics such as controlling the versions of external dependencies and appropriately handling randomness in Stata code. The presentation will include practical strategies for enhancing the transparency and reliability of Stata-based research.

This presentation will demonstrate the use of Stata to visualize and diagnose spillover within randomized controlled trials. In the past, techniques such as the L’abbe plot might have been used, but the plots available with diagnostic test assessment methods in Stata (community-contributed commands) are better. Spillover is crucial for the inference from RCT’s but difficult to demonstrate without use of information from outside the RCT. The data (plots and Stata code) are available in Hurley (2024).

Reference:

Hurley, J. C. 2024. Visualizing and diagnosing spillover within randomized concurrent controlled trials through the application of diagnostic test assessment methods. BMC Medical Research Methodology 24: 182.

This presentation explores perspectives on Jeff Wooldridge's DID approach, incorporating his latest flex method. Additionally, this presentation includes modifications developed for gravity models.

Causal inference studies are designed to identify and quantify the effect of a treatment (T) on an outcome of interest (Y). Sometimes, the treatment has an effect on a third variable, called a mediating variable (M), which also influences the outcome. So the treatment may have both a direct effect on the outcome (T -> Y) and an indirect effect on the outcome through its influence on the mediating variable (T -> M -> Y). The goal of causal mediation analysis is to identify and quantify these direct and indirect effects. This talk will introduce the concepts and jargon of causal mediation analysis, demonstrate how to analyze these kinds of data using Stata's mediate command, and show how to interpret and visualize these kinds of relationships.

Learning and sharing Stata knowledge online can be a challenging endeavor, especially when it comes to data visualization. In this presentation, I will cover some existing resources for doing so, including the notable advantages offered by the website Medium. I will demonstrate how Stata users can use Medium—and its popular “Stata Gallery”—to learn, or share their own, valuable insights for making more effective visualizations, communicating key statistical concepts, and doing better analysis.

This study looks at the likelihood that a country that defaults once would default again by testing the statistical significance between sovereign default as the dependent variable against lags of itself as the independent variable. When we use panelized probit models with Stata, the results show that a sovereign country that has defaulted is very likely to default again in the next eight years following the initial default.

This presentation highlights another legacy issue that characterizes the current versions of Stata. The issue is that, by default, Stata stores data in single-precision data format but performs all the calculations in double precision. When one handles noninteger data, this gives rise to unexpected behaviors that undermine the presumed functions of basic logical operators (;==, !=, <, and >). I will give several examples to illustrate the ways in which this behavior can fundamentally alter the conclusions of statistical models. Similar to my earlier presentation, this one is intended to start the conversation on whether it is time to move away from the single-precision storage default and fully embrace the double-precision format.