Lectures

Prof. Hudson Golino (University of Virginia)

Revolutionizing Psychometrics with Automatic Item Development and Validation with LLMs and Network Psychometrics: Structural Validity and Item Difficulty Estimation in silica. 

Brief Abstract: TBA

Prof. David Goretzko (University of Utrecht) 

Introduction to Supervised Machine Learning in Psychology

In recent years, (supervised) machine learning (ML) has gained immense popularity in psychology and other social sciences. While ML promises high predictive power, its application can be challenging, and researchers have to make sure to evaluate their models properly to avoid overoptimistic performance estimates. In this course, participants learn about the basic concepts, common pitfalls, important algorithms, and applications of supervised ML modeling (e.g., in the area of psychological assessment).

Prof. Alessio Farcomeni (Tor Vergata University of Rome)

A brief overview of methods for population size estimation in the social sciences

We discuss several methods for estimation of a population size based on repeated sampling of subjects from it. We admit heterogeneity of capture probabilities in four respects:  measurement occasions, observed subject-specific characteristics, unobserved subject-specific characteristics, and  behavioral response to capture. The material will mostly, but not exclusively, be focused to closed populations in discrete time. We illustrate with several examples in the social sciences and epidemiology. 

Prof. Livio Finos (University of Padova)

Reliable inference by permutation based methods 

The inferential analysis of complex data requires the use of appropriate models that often require unrealistic assumptions. Permutation tests (and conditional resampling) offer an approach that requires fewer assumptions than parametric models and guarantees more reliable results. In this unit we will present the general principles of permutation tests, showing their advantages and limitations with particular attention to practical aspects in the analysis of real data. Among the methods presented, we will show methods for comparing two or more samples with quantitative, ordinal and categorical responses. We will then extend the methods to the case of quantitative predictors (i.e. linear models and generalised linear models) in the presence of covariates (i.e. confounders) and multivariate responses.

Prof. Massimiliano Pastore (University of Padova)

Dealing with small samples in Bayesian framework:  The force (and the dark side) of the priors

The presentation deals with the analysis of small samples within the Bayesian framework, highlighting the importance of prior distributions.  Two examples (a binomial and a correlation case) are used to illustrate how priors can be very useful but at the same time can significantly affect posterior estimates in small samples. The presentation warns of the dangers of over-informative priors and emphasises the need for sensitivity analysis to identify optimal priors.

Prof. Angela Andreella (University of Trento)

Advances in statistical inference for longitudinal generalized linear models

This course examines approaches to testing fixed effects in longitudinal data and binary dependent variables. We will review generalized linear mixed models (GLMM), generalized estimating equations (GEE), two-stage summary statistics (2sss), and a recent robust approach called flip2sss. The sessions will critically evaluate the advantages and limitations of these approaches from both theoretical and applicative perspectives, analyzing real-data applications using R.

Prof. Ottavia M. Epifania (University of Trento)

So simple, yet so effective: Rasch-like pamatetrizations of accuracies and responses times of complex data structure with a (Generalized) Linear Mixed-Effects Models approach

Brief Abstract: TBA

Prof. Massimo Stella (University of Trento)

Text analysis with Explainable AI and psychologically validated data: The EmoAtlas tool

This short hands-on course introduces the audience to novel tools for extracting psychological insights from texts. Several techniques will be covered, ranging from lexicon-based approaches (VADER) to vector-based approaches (GloVE, BERT), briefly touching GPT architectures and their biases. Tasks like valence estimation, keyword identification, semantic frame analysis and emotion detection will be demonstrated pratically with a Python tool merging AI and psychological lexicons, EmoAtlas in Python. Examples will include the analysis of LLMs' responses, genuine suicide notes and social media data.