Panel vector autoregression (panel VAR) generalizes dynamic panel models and time-series VAR estimation by modeling the time dynamics while accounting for unobserved cross-sectional heterogeneity. This presentation provides an end-to-end empirical workflow for implementing a panel VAR model with real-world data. I will discuss model formulation, lag selection, and managing instrument proliferation from excessive moment conditions in GMM estimation. I will then demonstrate how to assess model stability and interpret impulse–response functions (IRFs) to evaluate the impact of structural shocks on the endogenous variables in the system.

ages is a new community-contributed module that estimates the relationship between different population-age groups (cohorts) and a dependent variable of choice, based on the polynomial transformation suggested by Fair and Dominguez (1991) and applied later by Higgins (1998), Arnott and Chaves (2012), and Juselius and Takas (2021). Capturing the effect of the age structure in a regression involves different methodological issues, such as weak precision if the number of population cohorts is large compared with the number of time periods, large correlation between consecutive cohorts’ shares and perfect collinearity with respect to the constant. The solution consists of restricting the age groups' coefficients to lie on a Pth degree polynomial and restricting its sum to be equal to zero to remove the perfect multicolinearity. The ages command performs all the necessary transformations and mathematical calculations to estimate and report the final effect by age cohort, together with its standard error and confidence interval, as an output matrix and through an automated graph that displays the impact of age cohort on the dependent variable. It allows testing for different age structure effects based on different groups captured by an interaction with a dummy variable. It also allows estimating a fully dynamic model, allowing the user to define the desired number of lags of the age structure groups, reporting the age structure by each lag and as sum of all them. ages also allows the user to choose the preferred time-series or panel-data method. The user can easily choose different options for the graph with the impact by age group, as well as several other options to save and use the results.

This presentation will demonstrate, through an example in Stata, how to conduct a target trial emulation. Target trial emulation is a practical framework that uses observational data to answer causal questions by mimicking a hypothetical randomized clinical trial. In this context, it is essential to control for potential biases and confounding through causal inference methods in order to correctly estimate the causal effect of treatments on the event of interest. Some of the methods that will be discussed and implemented in Stata during the presentation include inverse probability of treatment weighting (IPTW) and the g-formula. Using the comparison of two treatments as an example, the necessary steps to implement an emulated clinical trial will be presented. First, the data will be structured and expanded into a weekly format. Subsequently, different methods will be applied to estimate the causal effect, including a pooled logistic regression model, IPTW, the estimation of adjusted survival curves, and the g-formula. The results will be evaluated under different scenarios to assess how changes in the analytical assumptions affect the findings. The ultimate aim is for attendees to understand how to translate a clinical question into a reproducible, transparent Stata analysis that is consistent with the principles of causal inference.

Background: With the increasing use of clinical prediction models in medical research, evaluating model performance has become an essential component of the development and validation process. However, most studies still rely primarily on the area under the ROC curve (AUC) to assess model performance, paying less attention to key indicators such as calibration, incremental model improvement, clinical value, and external generalizability. Guidelines for predictive model evaluation, such as TRIPOD, recommend a multidimensional assessment; however, in Stata, these analyses typically require multiple independent commands.

Objective: To develop a comprehensive Stata workflow for evaluating clinical prediction models and demonstrate its application using a real-world multicenter dataset, thereby providing a standardized and reproducible analytical framework for predictive model evaluation.

Methods: An analysis workflow was constructed covering the main dimensions of predictive model evaluation, including discrimination (ROC curve and AUC), calibration (calibration plots, calibration in the large [CITL], and calibration slope), incremental predictive value (net reclassification improvement [NRI] and integrated discrimination improvement [IDI]), clinical utility (decision curve analysis [DCA]), and model validation. A multicenter cohort comprising 3,169 patients from 5 European countries was used to build a perioperative cardiovascular risk prediction model based on the revised cardiac risk index (RCRI) combined with NT-proBNP; this workflow was then applied to systematically evaluate the model's performance.

Results: The entire evaluation process was conducted within the Stata environment, achieving a standardized and reproducible assessment of model performance. Following the inclusion of NT-proBNP, the model's discriminative ability improved compared with the base model (AUC increased from 0.58 to 0.67). The NRI was 0.38 and the IDI was 0.02, suggesting an improvement in patient risk reclassification. Calibration analysis, clinical decision curve analysis, and external validation provided additional relevant information that cannot be captured by the AUC alone.

Conclusion: This study establishes a comprehensive workflow in Stata for evaluating clinical prediction models, integrating discriminative ability, calibration, incremental improvement, clinical utility, and model validation into a unified analytical framework. This approach enables a more standardized and reproducible assessment of predictive models, facilitates the practical application of TRIPOD recommendations, and provides a useful methodological tool for multicenter studies of clinical prediction models.

Graphs are widely used to represent social structures and to study relationships between variables. This presentation proposes an approach that extends their analytical potential by combining regression modeling with interactive network visualization. The method fits generalized linear models (including multinomial models) and their mixed-effects extensions, then identifies the categories whose predicted proportions or means differ significantly from the weighted population mean. The resulting graphs highlight these categories, making the most relevant effects immediately visible. Interactivity adds a further layer: users can filter nodes by size or attributes and explore the most central nodes and strongest links within the network. We present two new commands, dime and netreg, that compute marginal effects from the fitted models and generate the interactive graphs. Their use is illustrated with a variety of applied examples drawn from statistical studies.

Exploratory graphical analysis remains essential in an era of abundant data and increasingly sophisticated data science methods. Since Tukey's seminal work, tools for visualizing distributions and relationships between variables have become standard practice in data analysis. This presentation introduces two community-contributed commands, netcorr and netcoin, that extend this tradition by combining statistical measures of correlation and association with social network analysis. Besides examining coefficients in a matrix, the analyst obtains a network representation of the variables, where nodes are variables and edges reflect the strength of their relationships, making the underlying structure of the data immediately visible. Both commands handle quantitative and qualitative data, making them well suited for the exploratory analysis of questionnaires, textual networks, and general database screening. The presentation illustrates their syntax, options, and graphical output with applied examples.

Stata’s mixed command offers three common random-effects covariance structures (independent, exchangeable, unstructured) with little guidance on which to choose or what a wrong choice actually costs. This study analyzes results from a systematic Monte Carlo study varying the true correlation among random effects, the heterogeneity of their variances, cluster count and size, and the number of random effects to answer that question directly. Four findings carry direct implications for practice. Fixed-effect point estimates are robust to structure choice, but exchangeable can miscalibrate confidence-interval coverage even when its coefficients look unaffected, so a check comparing only point estimates can miss the problem. Independent biases the covariances by construction but leaves variances close to unbiased even under strong true correlation, so its cost is often smaller than assumed. Exchangeable can be actively misleading under heterogeneity, in one scenario reversing a covariance's sign entirely. Unstructured convergence degrades sharply as random effects grow relative to cluster count. The study derives concrete decision rules for when the extra unstructured parameters are worth fitting.

Most commonly used parametric regression models share a common foundation: maximum likelihood estimation. Viewing them this way lets researchers build from simple models to complex specifications using a single statistical principle. We begin with linear regression recast as Gaussian regression, then extend it to generalized linear models for binary, count, and positive continuous outcomes. Next, we add latent variables, including hierarchical random effects, to obtain generalized multilevel models. Throughout, we illustrate each step with Stata’s gsem command for generalized structural equation modeling.

A robust literature documents that education is one of the strongest individual-level predictors of European identification (EUID), yet the mechanisms underlying this relationship remain incompletely understood. Standard cross-national surveys capture EUID but lack the detailed measures needed to test key theorized pathways, particularly the transactionalist expectation that transnational travel experiences shape identification. Drawing on YOEDER (youth education and European attitudes), a 4-wave panel survey of Spanish youth aged 18–29 (N = 16,073), this presentation estimates a multiple-indicators multiple-causes (MIMIC) structural equation model to decompose the education–EUID link through 7 mediators: EU travel and satisfaction, EU institutional knowledge, universalist values, income, press consumption, peers' opinions of the EU, and foreign language ability. Results show that education is positively associated with EUID and six of the seven mediators and that all mediators are positively linked to EUID. Satisfactory travel experience within the EU emerges as the dominant channel, accounting for roughly 44% of education's total effect, and this pattern holds under an alternative sequential specification and after accounting for travel to other world regions. Disentangling frequency from satisfaction shows that the affective quality of travel, not mere exposure, drives the mediation. These findings extend transactionalist accounts of European identity by highlighting the concrete experiential pathway through which education fosters identification with Europe.

surveye converts a Survey Solutions or SurveyCTO questionnaire HTML file and the corresponding Stata dataset into a polished, standalone, interactive HTML dashboard. It automatically uses the questionnaire structure, wording, response categories, and section order to organize the output. The command supports filters, highlights, custom variables, related-variable groups, subgroup comparisons, weighted and unweighted estimates, local-currency/USD switching, profile tables, numeric summaries, optional confidence intervals, customizable themes, right-to-left Arabic and Urdu interfaces, and optional GPS maps. The dataset in memory is preserved, and all required JavaScript, styles, and selected data are embedded in the generated dashboard. The package includes its Java engine and requires no separate Stata dependencies.

Coca leaf chewing (acullico) is typically viewed as a cultural custom. This presentation—which extends the evidence documented for Bolivia by Gutierrez Miranda (2023) to Peru and a two-country comparative framework—argues, based on two decades of microdata, that it is also an economic phenomenon: coca functions as a subsistence good for poor households, and its satiating effect—physiologically documented yet lacking real nutritional value (Bolton, 1976; Penny et al., 2009)—crowds out spending on actual, more expensive food. The analysis is framed within the economics of addictive consumption and temptation goods (Becker and Murphy 1988; Banerjee and Mullainathan 2010). A harmonized two-country panel was constructed (Bolivia household survey and Peru's ENAHO, 2006–2024; ˜1.13 million household-years), made comparable regarding poverty and income through relative position and PPP-adjusted lines. The empirical strategy combines a pooled probit model (participation), a two-part/double-hurdle model (intensity), and the actual reported ordinal consumption frequency—derived from expenditure modules to account for self-supply bias. Results show that coca prevalence falls monotonically with income (Bolivia: 48% → 23%; Peru: 17% → 2%), whereas alcohol consumption does not; the indigenous effect remains robust when controlling for income and education. The subsistence gradient operates at the extensive margin (who consumes) rather than the frequency margin; consumption frequency is habitual and consistent across consumers—unlike alcohol consumption, which is social and sporadic. The methodological contribution consists of a reproducible R–Stata workflow (estimation in R; validation, margins, twopm, and graphing in Stata, ensuring exact replication) and a community-contributed command (cruceandino) that generates both the table (table and collect) and the figure for a multiway cross-tabulation in a single step—designed for consumption studies using household surveys and proposed for the SSC Archive repository.

This presentation constructs synthetic sustainability indicators based on the 2030 Agenda using harmonized Eurostat data and fully reproducible procedures implemented entirely in Stata. I consider time series for the 169 targets of the sustainable development goals (SDGs) over the period 2000–2023, covering 27 European countries: Austria, Belgium, Bulgaria, Croatia, Cyprus, Czech Republic, Denmark, Estonia, Finland, France, Greece, Germany, Hungary, Ireland, Italy, Latvia, Lithuania, Luxembourg, Malta, the Netherlands, Poland, Portugal, Romania, Slovenia, Slovak Republic, Spain, and Sweden. The indicators are downloaded directly from Eurostat using the community-contributed eurostatuse command, ensuring transparency, traceability, and easy data updates. Based on these series, synthetic indices are constructed following the methodology proposed by Rivero and Fernández (2008). The SDG targets are normalized on a scale from 0 to 1, distinguishing between direct and inverse variables, so that higher values reflect a greater probability of achieving the corresponding goal. Subsequently, the normalized values are aggregated to obtain partial indices for each SDG, calculated as the average of the targets included in each goal. Following the classification of the SDGs into five pillars proposed by Tremblay et al. (2020), the goal-level indices are weighted and aggregated to construct synthetic indicators for each pillar, in line with the institutional structure of the 2030 Agenda. The entire process of normalization, weighting, and aggregation is implemented in Stata, allowing for a coherent treatment of panel data and facilitating the replicability of the analysis.

Building on these synthetic indicators, the presentation conducts two complementary empirical applications. First, this systematic treatment allow us to examine the causal interactions among the 5 pillars of the 2030 Agenda for 28 countries over the period 2000–2021. The results reveal a leadership role of the planet pillar over the remaining pillars, while the planet, peace, prosperity, and partnership pillars jointly exhibit leadership over the people pillar. These findings highlight the central importance of environmental sustainability and international cooperation as prerequisites for social well-being and economic prosperity. Second, this novel dataset also assesses the efficiency of public spending in achieving the SDGs in 27 European countries during the period 1995–2023 by applying data envelopment analysis (DEA) models. The results show efficiency scores ranging from 0.70 to 0.94 for inputs and from 0.80 to 0.94 for outputs, indicating potential improvement margins of up to 30.6% in resource use or 20.3% in outcomes. Denmark, Ireland, Finland, and Sweden emerge as consistent benchmarks of sustained efficiency over time. Overall, the combined use of eurostatuse and customized Stata routines provides a flexible and fully reproducible framework for the comparative analysis of sustainability performance and policy efficiency in Europe.

Mozambique has recorded a decline in its overall fertility rate, an important element of the demographic transition and the exploitation of the demographic dividend. In fact, while in 1997, the country average was estimated at 5.2 children per woman, in 2017 available estimates indicated 4.9 children per woman. The reduction in the fertility rate has, however, not been homogeneous across the country. While in urban areas women have on average 3.6 children, in rural areas this rate remains high at 5.8 children per woman. By region and province, the disparity in the fertility rate is even more pronounced: the southern region registers the lowest rates, with the extreme case of Maputo City, where each woman has an average of only 2.8 children. Conversely, the northern region has the highest fertility rates, with the extreme case of the provinces of Niassa and Cabo Delgado, where the fertility rate is 6.8 and 6.2 children per woman, respectively. The fertility rate in the province of Nampula is even higher than the rest of the country's provinces, at 5.8 children per woman. It can therefore be stated that one of the greatest challenges to achieving a sustainable demographic dividend in the country lies in influencing adequate changes in demographic behavior outside the south of the country, with an even more intense focus on the northern region. The communities in the north and center of the country are essentially traditional. This study aims to understand the socioeconomic factors that influence the difference in fertility between the three regions of the country in such a way that the most important elements on which to focus for a more accelerated reduction can be identified.

Household shocks are a pervasive form of disruption to children’s education in developing countries, yet the extent to which their consequences depend on the nature of the shock remains poorly understood. This presentation studies two of the most common shocks to rural households—parental illness and harvest loss—both plausibly income-reducing but theoretically expected to operate through different mechanisms. Using four rounds of the learning and educational achievement in Punjab schools (LEAPS) data from rural Punjab, Pakistan, I find that these different channels appear to drive the observed differential effects of parental illness and harvest loss on children’s standardized test scores. Father’s illness is consistently associated with lower test scores: a one-standard-deviation increase in the father’s health impairment predicts a decline in mathematics and Urdu performance, equivalent to approximately one month of learning progress documented in the LEAPS sample (Bau et al. 2021). The effects are concentrated at moderate illness severity and are greater for boys than girls. I find no such corresponding effects for the mother’s illness.

In contrast, negative harvest shocks are not associated with a detectable effect on test scores across two-way fixed-effects, instrumental-variable, and staggered difference-in-differences specifications. I find no suggestive evidence that this average null result conceals heterogeneous subgroup effects: an exploratory causal forest vulnerability analysis finds no subpopulation of children experiencing larger learning losses from harvest shocks. For father’s illness, however, the evidence points to a vulnerability pattern—children in low-asset households with thin social networks experience estimated effects nearly 1.5 times the sample average, while low-vulnerability children exhibit no significant effect. These findings suggest that the nature of the shock matters: parental illness appears more strongly linked to children’s education than harvest shocks, and its educational costs are disproportionately concentrated among already disadvantaged households.

Assessing the quality of social research often requires combining structured indicators with the rich, contextual evidence gathered through interviews, yet workflows linking quantitative software like Stata with qualitative platforms such as ATLAS.ti remain largely manual and poorly documented. This presentation introduces a reproducible pipeline that connects Stata with ATLAS.ti to support mixed-methods research on the quality of social research practices, using semi-structured interviews with researchers as the empirical case.

The workflow shows how Stata can (1) organize and preprocess interview metadata (sample characteristics, coding frequencies, respondent profiles) exported from ATLAS.ti for descriptive and inferential analysis; (2) import code cooccurrence tables and thematic network outputs generated in ATLAS.ti to quantify patterns across interviews, complementing qualitative interpretation with statistical summaries; and (3) generate reproducible reports (do-files, dynamic documents) that integrate quotations, code frequencies, and quantitative indicators of research quality into a single analytical narrative.

Drawing on an applied case examining perceptions of methodological rigor and questionable research practices among social scientists, this work illustrates how this Stata–ATLAS.ti integration strengthens triangulation, improves traceability, and reduces the friction typically involved in moving between qualitative and quantitative environments. The approach is adaptable to other disciplines combining interview-based and structured data.