Marc Scott, (NYU)

Sensitivity Analysis for Grouped Data: Bounding the Bias in a Multilevel Setting

Abstract:We are concerned with the unbiased estimation of a treatment effect in the context of observational studies with grouped or multilevel data. When analysing such data with this goal, practitioners typically include as many predictors (controls) as possible, in an attempt to satisfy ignorability. In one approach, the econometric “within-group" treatment estimator, one includes indicator variables to control for unobserved between-group variation. However, depending on the mathematical properties of the underlying stochastic process, adding such predictors can actually increase bias in the targeted treatment effect.

Exploiting information contained in successive multilevel model specifications and estimates, we generate bounds on the potential differences in bias for several competing estimators, informing model selection. Our approach relies on a parametric model for grouped data and omitted confounders and establishes a framework for sensitivity analysis in the multilevel modelling context. We characterize the strength of the confounding and corresponding bias amplification using easily interpretable parameters and graphical displays.

We apply this approach to data from a multinational educational evaluation study. We demonstrate the extent to which different treatment effect estimators are robust to potential unobserved individual- and group-level confounding. Time permitting, we re-frame the sensitivity analysis for these data using a Bayesian approach. The latter reveals simultaneously the uncertainty we have about unobserved confounders and the implied treatment effects.

Joint work with Jennifer Hill (NYU), Joel Middleton (UC/Berkeley) & Ronli Diakow (NYC DOE & NYU).

Simon Spencer, (University of Warwick)

Bayesian methods for source attribution and spatio-temporal modelling of campylobacteriosis.

Abstract: Campylobacteriosis is a common form of food poisoning with a complex epidemiology. Despite the large number of cases, the dominant pathways to infection were until recently poorly understood. In this talk I will outline how novel statistical methodology has helped to develop our understanding of this pathogen as part of a multidisciplinary approach spanning epidemiology, genetics, public health and evolutionary biology.

In particular I will outline how to estimate the probability that a human infection has emanated from a potential host species; how to identify risk factors from the spatial distribution of cases; and how to detect outbreaks from a background of sporadic cases.

Joint work with Petra Müllner and Nigel French.

14.00 - 15.00: Marc Scott, Sensitivity Analysis for Grouped Data: Bounding the Bias in a Multilevel Setting
15.00 - 15.30: Coffee break
15.30 - 16.30: Simon Spencer, Bayesian methods for source attribution and spatio-temporal modelling of campylobacteriosis

Alan Gelfand (Duke, Dept of Statistical Science)

Title: Space and circular time log Gaussian Cox processes with application to crime event data

Abstract: We view the locations and times of a collection of crime events as a space-time point pattern. So, with either a nonhomogeneous Poisson process or with a more general Cox process, we need to specify a space-time intensity. For the latter, we need a random intensity which we model as a realization of a spatio-temporal log Gaussian process. In fact, we view time as circular, necessitating valid separable and nonseparable covariance functions over a bounded spatial region crossed with circular time. In addition, crimes are classified by crime type.

Furthermore, each crime event is marked by day of the year which we convert to day of the week. We present models to accommodate such data. Then, we extend the modeling to include the marks. Our specifications naturally take the form of hierarchical models which we t within a Bayesian framework. In this regard, we consider model comparison between the nonhomogeneous Poisson process and the log Gaussian Cox process. We also compare separable vs. nonseparable covariance specifications. 

Our motivating dataset is a collection of crime events for the city of San Francisco during the year 2012. Again, we have location, hour, day of the year, and crime type for each event. We investigate a rich range

of models to enhance our understanding of the set of incidences.

Andrew Walden, (Imperial College London)

Problems and Solutions in Partial Coherence Estimation
Abstract: Partial coherence is an important quantity in multivariate time series and is used to build time series graphs in areas such as neuroscience. Estimation of this quantity is fraught with difficulties and unless these issues are carefully treated a "garbage-in, garbage-out" scenario is possible. On the pre-processing side we will consider misalignment bias, the removal of spectral lines and filtering. We will then consider statistical problems such as ill-conditioned spectral matrices, spectral leakage, and the statistical distribution of the estimated partial coherence and its consequences.

Mark Fiecas, (University of Warwick)

Two Problems for Multivariate Time Series
Abstract: In this talk, I will discuss two related problems for multivariate time series data. In the first problem, I will consider high-dimensional time series in a regime-switching context. Whenever a state is rarely visited, the estimates of the parameters for that state will be highly unstable. I will discuss a modification of the EM algorithm that will yield more stable estimates of covariance matrices, and consequently, a more reliable filter for reconstructing the hidden state sequence. In the second problem, I will discuss spectral analysis for multivariate time series data. Motivated by the analysis of fMRI data, I will discuss shrinkage estimators for obtaining regularised estimates of the spectral density matrix. The proposed approach includes the development of the multivariate time-frequency toggle bootstrap for time series data, which we use to obtain estimates of the regularisation parameters.

My collaborators for these works are Rainer von Sachs (Universite catholique de Louvain), Jürgen Franke (TU Kaiserslautern), and Joseph Tadjuidje (TU Kaiserslautern).

14.00 - 15.00: Andrew Walden, "Problems and Solutions in Partial Coherence Estimation"
15.00 - 15.30: Coffee break
15.30 - 16.30: Mark Fiecas, "Two Problems for Multivariate Time Series"

Sebastian Ebert (Tilburg) - Measuring Multivariate Risk Preferences

We measure risk preferences for decisions that involve more than a single, monetary attribute. According to theory, correlation aversion, cross- prudence and cross-temperance determine how risk preferences over two single attributes co-vary and interact. We obtain model-free measurements of these cross-risk attitudes in three economic domains, viz., time preferences, social preferences, and preferences over waiting time. This first systematic empirical exploration of multivariate risk preferences provides evidence for assumptions made in economic models on inequality, labor, time preferences, saving, and insurance. We observe non-neutrality of cross-risk attitudes in all domains which questions the de- scriptive accuracy of economic models that assume that utility is additively separable in its arguments.

See: https://sites.google.com/site/ebertecon/home

Petros Dellaporta (UCL)

Scalable inference for a full multivariate stochastic volatility model

Abstract: We introduce a multivariate stochastic volatility model for asset returns that imposes no restrictions to the structure of the volatility matrix and treats all its elements as functions of latent stochastic processes. When the number of assets is prohibitively large, we propose a factor multivariate stochastic volatility model in which the variances and correlations of the factors evolve stochastically over time. Inference is achieved via a carefully designed feasible andscalable Markov chain Monte Carlo algorithm that combines two computationally important ingredients: it utilizes invariant to the prior Metropolis proposal densities for simultaneously updating all latent paths and has quadratic, rather than cubic, computational complexity when evaluating the multivariate normal densities required. We apply our modelling and computational methodology to 571 stock daily returns of Euro STOXX index for data over a period of 10 years.