2016-Statistics in Action


Statistics in Action 
Chair: Yi Yang (McGill University) 
[PDF]

DEVAN BECKER, University of Western Ontario
Aerial Firefighting Fatigue Project  [PDF]
 
In this ongoing project, we are attempting to find indicators of pilot fatigue in aerial firefighters using raw flight data (location, altitude, speed, yaw, pitch, roll, etc. recorded each second). We are applying Shewhart control charts to detect unusual flights based on summaries of the data. Each flight is different, but control charts are meant for homogeneous data. We must either find homogenous processes within the heterogeneous data or extend the methodology. This is still an open problem, but successful completion might impact fatigue regulations for non-typical flights and aid future applications of control charts with non-homogeneous data. 
 
PIERRE-JÉRÔME BERGERON, PJB Consulting
Valar Morghulis: A Survival Analysis of 'A Song of Ice and Fire'  [PDF]
 
George R. R. Martin's {\it A Song of Ice and Fire} is an ongoing series of fantasy novels with a large cast of characters, and numerous, sometimes shocking deaths. Readers do not know what to expect in terms of the eventual resolution of the story's many threads. We apply here the methods of survival analysis to see if we can discern patterns that could inform the literary analysis of the text. We discuss data selection, the choice of variables and units of measurements in defining event times in fiction. We then compare survival between different groups of characters. 
 
DEBBIE DUPUIS, HEC Montreal
Realized Peaks over Threshold  [PDF]
 
Recent contributions to the financial econometrics literature exploit high-frequency data to improve models for daily asset returns. In this talk, we present a new class of dynamic extreme value models that profit from high-frequency data when estimating the tails of daily asset returns. Our Realized Peaks-Over-Threshold approach provides estimates for the tails of the time-varying conditional return distribution, and remains applicable even when data are non-stationary. An in-sample fit to S\&P 500 index returns shows that high-frequency data contain information on daily extreme returns beyond that included in low frequency data. 
 
NICHOLAS MITSAKAKIS, University of Toronto
Investigating Structural Independence in Health Utility Instruments Using Graphical Models  [PDF]
 
In health economics cost effectiveness analyses rely on accurate estimation of health utility, a single global measure of health related quality of life (HRQoL). Health utilities are often measured with the use of questionnaire-type instruments, containing a number of items describing specific domains of HRQoL. The construction of these instruments relies on multi-attribute utility theory, assuming structural independence among the attributes. This property is rarely tested empirically. Here we applied discrete graphical models to multiple patient datasets to test the structural independence of a prostate cancer-specific instrument, Patient-Oriented Prostate Utility Scale (PORPUS-U). Results reveal undesired interdependence among items in PORPUS-U. 
 
DAVID RIEGERT, Queen's University
Statistical Estimation of Transfer Functions to Examine the Relationship Between Geomagnetism and Power Systems  [PDF]
 
Geomagnetic disturbances disrupt ground-based systems in a profound manner; increasing the rate of corrosion in pipelines, damaging equipment in power transmission systems, and interfering with communications. We will focus specifically on the relationship between Earth's magnetic field and induced currents in power lines. In order to describe this relationship, transfer functions were estimated using current methods which are robust to outliers. We discuss the shortcomings of such current methods in the context of a Northwest US power system. 
 
PAVEL SLAVCHEV, McGill University
The Analysis of Live Births as a Competing Event to Stillbirths  [PDF]
 
Understanding the causal risk factors for stillbirths should help develop more effective prevention programs. The Institut National de la Santé Public du Québec (INSPQ) birth registry data includes all births over the last 25 years in the province of Quebec. Classical time-to-event analyses (e.g. using Cox-proportional hazards models) for the INSQP registry data on stillborn infants are complicated by two problems: clustering of births within mothers and live births that censor the time to stillbirth. In this presentation, we show how modelling live births as competing events to stillbirths can address both of these problems.