Partially constrained group variable selection to adjust for complementary unit performance in American college football

Given the importance of accurate team rankings in American college football (CFB) - due to heavy title and playoff implications - strides have been made to improve metrics for team performance evaluation, going from basic averages (e.g. points scored per game) to metrics that adjust for a team's strength of schedule, but one aspect that's yet to be accounted for is the ability of team's offense and defense to complement one another, termed 'complementary football'. American football is unique because the same team's offensive and defensive units typically consist of separate player sets that don't share the field simultaneously, which tempts one to evaluate them independently. Yet, some aspects of your team's defensive (offensive) performance may directly impact the complementary unit, e.g. turnovers forced by your defense could lead to easier scoring chances for your offense. Our main goal is to identify the most consistently influential features of complementary football in a data-driven way, subsequently adjusting each team's offensive (defensive) performance for that of their complementary unit. To achieve that, for the 2009-2019 CFB seasons, we incorporate natural splines with group penalty approaches, conducting partially constrained optimization to guarantee the full adjustment for the strength of schedule and home-field factor.

Using localized Twitter activity to assess harmful algal bloom impacts of Karenia brevis in Florida, USA

Red tide blooms of the dinoflagellate Karenia brevis (K. brevis) produce toxic coastal conditions that can impact marine organisms and human health, while also affecting local economies. During the extreme Florida red tide event of 2017-2019, residents and visitors turned to social media platforms to both receive disaster-related information and communicate their own sentiments and experiences. This was the first major red tide event since the ubiquitous use of social media, thus providing unique crowd-sourced reporting of red tide impacts. We evaluated the spatial and temporal accuracy of red tide topic activity on Twitter, taking tweet sentiments and user types (e.g. media, citizens) into consideration, and compared tweet activity with reported red tide conditions, such as K. brevis cell counts, levels of dead fish and respiratory irritation on local beaches. The analysis was done on multiple levels with respect to both locality (e.g., entire Gulf coast, county-level, city-level, zip code tabulation areas) and temporal frequencies (e.g. daily, every three days, weekly), resulting in strong correlations between local per-capita Twitter activity and the actual red tide conditions observed in the area. Moreover, an association was observed between proximity to the affected coastal areas and per-capita counts for relevant tweets. Results show that Twitter presents a trustworthy reflection of the red tide's local impacts and development over time, and can potentially augment the already existing tools for efficient assessment and a more coordinated response to the disaster.

Regularized Joint Estimation of Related Vector Autoregressive Models

In a number of applications, one has access to high-dimensional time series data on several related subjects. A motivating application area comes from the neuroimaging field, such as brain fMRI time series data, obtained from various groups of subjects (cases/controls) with a specific neurological disorder. The problem of regularized joint estimation of multiple related Vector Autoregressive (VAR) models is discussed, leveraging a group lasso penalty in addition to a regular lasso one, so as to increase statistical efficiency of the estimates by borrowing strength across the models. A modeling framework is developed that it allows for both group-level and subject-specific effects for related subjects, using a group lasso penalty to estimate the former. An estimation procedure is introduced, whose performance is illustrated on synthetic data and compared to other state-of-the-art methods. Moreover, the proposed approach is employed for the analysis of resting state fMRI data. In particular, a group-level descriptive analysis is conducted for brain inter-regional temporal effects of Attention Deficit Hyperactive Disorder (ADHD) patients as opposed to controls, with the data available from the ADHD-200 Global Competition repository.