Seminar

Martina Boschi

Martina Boschi
(University of Svizzera Italiana)
Bio I am a first-year PhD candidate at the Università della Svizzera italiana, supervised by Professor Ernst C. Wit. My educational background has been centered around Statistical Sciences. When writing my Master's thesis at the Institute of Computing, I delved into network science, which is now my primary research focus, particularly in the domain of relational event models. Currently, my research is directed towards an ecological empirical application, with plans to shortly initiate a financial application aimed to anomaly detection.

Abstract

The invasion of alien species into non-native environments is a critical issue due to its potential harm to biodiversity, economies, and human health. The several forces at play have so far made it challenging to fully understand the relative importance of different causes behind these invasions. Our proposal entails developing a generalized mixed additive relational event model (REM) to analyze plant and insect co-invasions. We particularly focus on introductions occurred between 1880 and 2005 and documented in the First Record Database. In this context, a “first record” is considered a relational event where a species (sender) first lands in a region (receiver) during a specific year (time). In addition to possibly time-varying exogenous and endogenous variables, our proposed REM includes time-varying and random effects to explain the invasion rates. Our inference approach employs a case- control sampling technique, which enables efficient computation. Nevertheless, assessing if REMs properly fit the data is still mostly an ongoing research challenge, especially for REMs that involve time-varying and random effects. Present methodologies often depend on comparing observed and simulated events using specific statistics. However, this can be computationally intensive and necessitate various assumptions. We present a flexible framework for testing the goodness of fit of REMs by means of cumulative martingale-residuals. We particularly focus on a test of the Kolmogorov-Smirnov type, intended to determine if covariates are properly modeled. Nevertheless, our approach can easily be extended to determine whether any other network dynamics features have been adequately incorporated into the model. Implementation is performed through the R package mgcv.

Keywords

  • relational event models, time-varying effects, random effects, generalised additive models, alien species invasions, goodness of fit, martingale residuals

Schedule

  • Monday 23 October 2023, 11:30-12:30 room seminar 1 @ Povo 0

Details

Material (Restricted access, user: MAM2023)