Introduction to Categorical and Panel Data Analysis

A large share of quantative research in the health and social sciences depends on understanding how categorical variables - such as gender, race, education, occupation, or politcal affilliation - influencer and interact with key outcomes. This workshop offers a comprehensive introduction to techniques for analysing categorical data in Stata, providing participants with the conceptual understanding and practical skills needed to go beyond descriptive comparisons and toward more rigorous, explanatory analysis.

Participants will laern to estimate and interpret binary, ordinal, and nominal outcome models (logit, probit, ologit, mlogit), as well as panel data models (fixed and random effects, xtlogit, xtmlogit). Special attention will be given to the causal insights afforded by panel data methods, which allow researchers to control for unobserved heterogeneity, explore within-individual or within-group changes over time, and make stronger inferences about cause-and effect relationships.

Through an integrated mix of theoretical explanation, guided exercises, and hands-on data analysis, the workshop emphasises not only how to estimate these models, but also how to interpet results meaningfully, assess model fit and assumptions, and communicate findings effectively to both technical and applied audiences.

Key Skills Accquired:

Participants will learn to:

• Structure and prepare datasets for categorical and panel analysis in Stata

• Specify and estimate logit, probit, ordered logit, and multinominal logit models

• Apply fixed and random effects models for panel data (xtlogit, xtmlogit)

• Interpret Stata output tables and model diagnostics

• Visualise results using marginal effects plots and post-estimation tools (margins, marginsplot)

• Validate model adequacy and communicate findings effectively in academic research

Learning Outcomes:

By the end of the workshop, participants will be able to:

• Identify  suitable statistical models for different types of categorical outcomes

• Estimate and interpret model parameters and marginal effects

• Understand the theoretical assumptions behind discrete choice and panel models

• Apply these techniques to real-world datasets from the health and social sciences using Stata

• Produce publication-quality tables and graphics for categorical and panel data analyses

Particpants will have a practical toolkit for applying categorical and panel data techniques to their own research - enabling them to produce analyses that are not only statistically sound but also causally informed and policy-relevant.

Prerequisites:

• This workshop will be delivered in Portuguese

• Basic knowledge of regression analysis (OLS or GLM)

• Interest in quantative research methods in the health or social sciences

• Laptop with Stata installed (trial versions will be provided for particpants)

Recommended Readings:

• Agresiti, Alan & Finlay, Barbara. 2021. Cap. 15 - Regressão logística: modelando respostas categóricas, in Métodos Estatísticos para as Ciências Sociais, 4a ed. Porto Alegre: Penso, pp. 533-571.

• Long, Scott J, & Freese, Jeremy. 2006. Regression Models for Categorical Dependent Variables Using Stata. 2a ed. Collge Station, TX: Stata Press.

• Powers, Daniel A, & Xie, Yu. 2008. Statistical Methods for Categorical Data Analysis 2a ed. San Diego: Emerald Group Publishing.

• StataCorp. 2023. xtmlogit - Multinomial logit models for panel data. In: Stata Longitudinal-Data/ Panel-Data Reference Manual: Release 18. College Station, TX: Stata Press. https://www.stata.com/manuals/xtxtmlogit.pdf

• StataCorp. 2023. xtlogit - Random-effects, fixed-effects, and population averaged logit models. In: Stata Longitudinal-Data/ Panel-Data Reference Manual: Release 18. College Station, TX: Stata Press. https://www.stata.com/manuals/xtxtlogit.pdf

Categorical Data Analysis

2pm: Introduction to Binary Models (logit, probit)

3pm: Multinomial and Ordinal Models (mlogit, ologit)

Panel Data Models

4pm Introduction to Panel Data Structures (xtset) and Estimation

5pm: Categorical Response Panel Models (xtlogit, xtmlogit) and Interpretation with Margins