I turn to a development few applied researchers can now choose to ignore: the arrival of AI coding agents in the empirical workflow. Data cleaning, model estimation, visualization, and replication are all tasks increasingly open to delegation. In this presentation, I discuss where these tools genuinely help, where they mislead, and what their use implies for the reliability and reproducibility of published research.

I describe the Stata implementation of unit-root and stationarity tests with flexible Fourier forms as in Enders and Lee (2012a,b) and Becker, Enders, and Lee (2006).

Threshold regression provides a flexible framework for capturing regime-dependent relationships and has become a widely used tool for uncovering structural breaks, nonlinearities, and state-dependent patterns in data across diverse fields. This presentation introduces a new community-contributed command called xtthreshold, which provides researchers with a complete toolbox for analyzing threshold regression in panel data with interactive fixed effects. The new command is able to estimate both discontinuous threshold regression and kink regression, both with interactive fixed effects. It further accommodates three alternative specifications of slope heterogeneity: fully homogeneous, fully heterogeneous, and semihomogeneous models. Model selection is facilitated by a modified information criterion capable of discriminating among competing specifications, and inference is supported through confidence intervals and hypothesis testing for all model parameters, in a addition to bootstrap-based tests for the existence of nonlinearity.

Researchers typically collect experimental data sequentially, allowing early outcome observations and adaptive treatment assignment to reduce exposure to inferior treatments. This presentation reviews multi-armed-bandit adaptive experimental designs that balance exploration and exploitation. Because adaptively collected experimental data through bandit algorithms violate standard asymptotics, inference is challenging. We implement an estimator that yields valid heteroskedasticity-robust confidence intervals in batched bandit designs and compare coverage in Monte Carlo simulations. We introduce bbandits for Stata, a tool for designing experiments via simulation, running interactive bandit experiments, and implementing and analyzing adaptively collected data. bbandits includes three common assignment algorithms—ε-first, ε-greedy, and Thompson sampling—and supports estimation, inference, and visualization.

We present the rdlasso command, which enables the inclusion of high-dimensional covariates in regression discontinuity design (RDD) settings. This command is based on the paper “Inference in regression discontinuity designs with high-dimensional covariates” by Kreiss and Rothe (2023). The command automates covariate selection using lasso-based procedures, supports both sharp and fuzzy settings, and integrates seamlessly with rdrobust for bandwidth selection and inference. The command relies on Stata’s native implementation of lasso for high-dimensional covariate selection and on rdrobust for bandwidth selection, estimation, and inference, making the methodology both accessible to Stata users and computationally feasible for applied researchers.

Stata’s gsem command fits generalized structural equation models. This generalization allows users to fit structural equation models with continuous, binary, ordinal, categorical, count, and survival-time outcomes. Users can also fit models with continuous or categorical latent variables. Multilevel models are also supported. In this presentation, I will demonstrate a wide variety of models that can be fit with gsem, from multilevel confirmatory factor models to latent class models. Along the way, participants will see both typical SEM applications and unique models that can be fit in the generalized SEM framework.

Meta-analysis is traditionally used to synthesize evidence across independent studies. This presentation illustrates a nonstandard application of Stata’s multivariate meta-analysis framework to investigate parameter heterogeneity estimated from firm-level dynamic models. The empirical motivation comes from corporate finance, where heterogeneous adjustment dynamics make pooled panel specifications with numerous interactions difficult to interpret. I first estimate fully heterogeneous error-correction models for individual firms, obtaining firm-specific parameters measuring the speed of leverage adjustment, the sensitivity to free cash flow, and debt-maturity interaction, together with their estimated standard errors. Rather than treating these parameters as final estimates, I use Stata’s multivariate random-effects meta-regression to explain their cross-sectional heterogeneity through firm characteristics, contractual features, and institutional changes.

I demonstrate how Stata’s meta commands can be extended beyond their conventional role of evidence synthesis to provide a flexible second-stage modeling framework for heterogeneous parameter estimates. The approach accommodates multiple correlated outcomes, accounts for estimation uncertainty through inverse-variance weighting, and avoids the overparameterization that often arises in pooled interaction models. Although illustrated using a novel dataset on corporate debt covenants extracted from SEC filings, the methodology is applicable to any context in which unit-specific parameters are estimated in a first stage and subsequently related to observed characteristics.

This presentation presents a bibliometric analysis workflow that combines Stata with its integrated Python environment to map the thematic structure of a large OpenAlex corpus and a dataset of abstracts in Scopus on the theme of energy policy. Relevant keywords are extracted from article titles through transparent linguistic preprocessing, domain-specific filtering, and frequency-based selection. The resulting article-by-keyword matrix is analyzed through principal component analysis to reduce dimensionality and identify the main latent thematic dimensions of the literature. K-means clustering is then applied to the retained component scores, assigning each article to a homogeneous research cluster. Stata manages data preparation, descriptive statistics, graphical outputs, and result export, while Python provides scalable text mining and machine learning procedures. The workflow produces an interpretable keyword dictionary, PCA loadings and scores, cluster memberships for individual articles, and cluster-level bibliometric profiles. A different approach based on the analysis of abstracts classifies them using cosine distance in an unsupervised framework. Finally, the co-occurrence matrix is analyzed using network analysis to identify the most relevant “cores” in the literature. The contribution is a practical workflow for conducting transparent, computationally efficient bibliometric analyses entirely within a Stata-centered research environment.

Stata is one of the most widely used environments for statistical analysis, but many workflows still require users to interact directly with the command line or do-files. This presentation illustrates how Python can be used to build lightweight desktop graphical user interfaces (GUIs) that interact seamlessly with Stata, allowing users to launch complex analyses through intuitive point-and-click applications. The presentation will discuss the architecture of the communication between Python and Stata, practical implementation details, and possible extensions.

adoadd is a package developed to control Stata’s ado-environment.

Weighted quantile sum (WQS) regression is a statistical method for quantifying the association between a set of possibly correlated predictors and a health outcome, estimating the joint effect of the predictors as well as their individual contributions to the total effect. We present wqsreg, the first Stata command for WQS regression, implemented for continuous, binary, and count outcomes. The execution of the command involves two sequential steps: 1) estimating the weights and constructing the WQS index under specific constraints; and 2) modeling its association with the outcome. wqsreg integrates several flexible components of the framework such as bootstrap, training and validation, and repeated holdout procedures; it returns regression estimates as well as graphical displays of the individual weights. We present an application of the command on exposome data exploring the association between 38 exposures and a continuous outcome while adjusting for a set of covariates. To the best of our knowledge, wqsreg provides the first command to conduct WQS regression in Stata. We anticipate that our contribution will further promote the use of appropriate statistical methods for handling multiple correlated predictors.

We introduce stfform, a postestimation command for stcox that allows users to assess the functional form of continuous covariates in Cox proportional hazards regression models using cumulative sums of martingale residuals. By complementing graphical diagnostics with a formal statistical test, stfform provides a principled assessment of the adequacy of the specified functional form.

This talk introduces catmetrics, a new Stata command for computing various measures of association, similarity, and deterministic and probabilistic forecast evaluation for categorical data. The command accommodates binary, nominal, and ordinal multicategory data. It reports over 300 distinct measures, providing their alternative names across various disciplines and their attainable ranges. Furthermore, in addition to overall metrics, catmetrics calculates class-specific quantities for each category, as well as their macro and weighted averages.

Treatment effects estimate the causal effects of a treatment on an outcome. The effect may be heterogeneous. Average advanced procedures.

This presentation develops a counterfactual microsimulation framework for evaluating the fiscal consequences of guaranteed minimum income (GMI) schemes under adverse labor supply responses. The framework combines a theoretical cost-effectiveness measure based on the Lorenz curve and the Gini coefficient with an empirical implementation using EU-SILC microdata in Stata. It enables a systematic comparison of alternative GMI architectures while quantifying their fiscal implications under identical behavioral assumptions.

The analysis considers an extreme counterfactual scenario in which all households eligible for GMI—defined as those with disposable incomes below 60% of the relevant median disposable income—withdraw completely from the labor market. The scenario is intended as a fiscal stress test rather than a behavioral forecast. Two institutional arrangements are compared: (i) decentralized national GMI schemes, where eligibility is defined relative to national median income; and (ii) a centralized EU-wide GMI system, where eligibility is defined relative to the EU-wide median income.

The proposed framework is implemented in Stata through a reproducible computational workflow that automates the identification of eligible households, construction of counterfactual income distributions, computation of population-weighted cost-effectiveness indicators, and decomposition of the cost-effectiveness ratio into contributions stemming from between-country and within-country redistributive effects. The implementation provides a flexible computational environment for evaluating alternative GMI designs under different eligibility thresholds and behavioral assumptions and can be readily adapted to other comparative redistribution analyses based on EU-SILC microdata.

The empirical application shows that complete labor market withdrawal increases total GMI expenditures by 89% under national schemes and by 104% under a centralized EU-wide scheme relative to baseline estimates. Although beneficiary coverage remains unchanged and the redistributive impact of both systems is preserved, fiscal efficiency deteriorates substantially. The proposed framework illustrates how Stata can be used to integrate theoretical distributional analysis with large-scale EU-SILC microsimulation, providing a transparent and reproducible approach for stress-testing alternative social protection policies.

The coexistence of evolving chronic illnesses and financial constraints represents a major challenge for households, particularly in countries where social health coverage remains incomplete. This presentation proposes an application of latent transition analysis (LTA) to dynamically model individuals’ chronic health trajectories and assess their influence on household financial solvency. Using panel data (health status, healthcare expenditures, income, and debt), we identify latent health states (for example, “stable,” “moderate deterioration,” “severe deterioration”) and estimate the transition probabilities between these states over time. These trajectories are then linked to indicators of financial vulnerability (debt ratio, savings capacity, credit reliance, insurance premiums) in order to assess the extent to which health deterioration increases the risk of household financial fragility.

The expected results will highlight typical at-risk trajectory profiles and estimate the magnitude of the relationship between chronic health status and solvency. This approach offers practical implications for actuarial pricing in health insurance, the design of adapted microinsurance products, and the orientation of public prevention policies. Methodologically, all estimations will be carried out using Stata, employing its dedicated commands for latent class models and their longitudinal extension (latent transitions), thereby illustrating Stata’s capabilities for analyzing longitudinal health data applied to economics and insurance.

Geographical analysis is a powerful tool for public health surveillance because it allows the visualization of spatial patterns, territorial inequalities, and population groups at greater risk. Maps provide an immediate and intuitive representation of epidemiological indicators, supporting the identification of geographic clusters and helping policy makers prioritize prevention strategies and resource allocation. Within the Italian PASSI surveillance system, referred to as adult population aged 18–69 years, geographic visualization can enhance the interpretation of behavioral risk factors by highlighting regional differences that may not emerge from national averages alone.

This talk presents a geographic analysis of two PASSI indicators, smoking status and high-risk alcohol consumption, using data from the 2024–2025 biennium. Regional estimates were calculated as weighted prevalences through the svy command in Stata, ensuring that the results account for the complex sampling design and are representative of the resident adult population. The indicators were then displayed through thematic maps at the regional level and local level realized through Stata maps visualization commands.

To improve the interpretability of territorial differences, we classified prevalence estimates using quintiles rather than fixed thresholds or comparisons with the national average. Quintile-based classification offers several advantages: it distributes regions more evenly across categories, enhances visual contrast, facilitates the identification of relative geographic gradients, and reduces the risk of masking meaningful variability when indicator distributions are skewed. Unlike classifications centered on a national benchmark, quintiles emphasize the relative position of each region within the overall distribution, providing a clearer picture of territorial inequalities.

The use of weighted prevalence estimates combined with quintile-based thematic mapping represents an effective approach for communicating PASSI surveillance data and supporting evidence-based public health planning. Furthermore, this methodology can be easily extended to finer geographic levels, such as local health authorities (LHAs) and municipalities, allowing the identification of local patterns and inequalities that may be hidden in regional-level analyses.