arXiv subjects

Emma J. McCoy

Publications and source records attributed to Emma J. McCoy.

9 recordsLinked to original sources

arXiv2026

Non-parametric Bayesian inference via loss functions under model misspecification

In the usual Bayesian setting, a full probabilistic model is required to link the data and parameters, and the form of this model and the inference and prediction mechanisms are specified via de Finetti's representation. In general, such a formulation is not robust to model misspecification of its component parts. An alternative approach is to draw inference based on loss functions, where the quantity of interest is defined as a minimizer of some expected loss, and to construct posterior distributions based on the loss-based formulation; this strategy underpins the construction of the Gibbs posterior. We develop a Bayesian non-parametric approach; specifically, we generalize the Bayesian bootstrap, and specify a Dirichlet process model for the distribution of the observables. We implement this using direct prior-to-posterior calculations, but also using predictive sampling. We also study the assessment of posterior validity for non-standard Bayesian calculations. We show that the developed non-standard Bayesian updating procedures yield valid posterior distributions in terms of consistency and asymptotic normality under model misspecification. Simulation studies show that the proposed methods can recover the true value of the parameter under misspecification.

stat.ME↗

arXiv2023

Causal Analysis at Extreme Quantiles with Application to London Traffic Flow Data

Transport engineers employ various interventions to enhance traffic-network performance. Quantifying the impacts of Cycle Superhighways is complicated due to the non-random assignment of such an intervention over the transport network. Treatment effects on asymmetric and heavy-tailed distributions are better reflected at extreme tails rather than at the median. We propose a novel method to estimate the treatment effect at extreme tails incorporating heavy-tailed features in the outcome distribution. The analysis of London transport data using the proposed method indicates that the extreme traffic flow increased substantially after Cycle Superhighways came into operation.

stat.ME↗

arXiv2023

An Evaluation Framework for Personalization Strategy Experiment Designs

Online Controlled Experiments (OCEs) are the gold standard in evaluating the effectiveness of changes to websites. An important type of OCE evaluates different personalization strategies, which present challenges in low test power and lack of full control in group assignment. We argue that getting the right experiment setup -- the allocation of users to treatment/analysis groups -- should take precedence of post-hoc variance reduction techniques in order to enable the scaling of the number of experiments. We present an evaluation framework that, along with a few simple rule of thumbs, allow experimenters to quickly compare which experiment setup will lead to the highest probability of detecting a treatment effect under their particular circumstance.

stat.ME↗

arXiv2023

Measuring e-Commerce Metric Changes in Online Experiments

Digital technology organizations routinely use online experiments (e.g. A/B tests) to guide their product and business decisions. In e-commerce, we often measure changes to transaction- or item-based business metrics such as Average Basket Value (ABV), Average Basket Size (ABS), and Average Selling Price (ASP); yet it remains a common pitfall to ignore the dependency between the value/size of transactions/items during experiment design and analysis. We present empirical evidence on such dependency, its impact on measurement uncertainty, and practical implications on A/B test outcomes if left unmitigated. By making the evidence available, we hope to drive awareness of the pitfall among experimenters in e-commerce and hence encourage the adoption of established mitigation approaches. We also share lessons learned when incorporating selected mitigation approaches into our experimentation analysis platform currently in production.

stat.AP↗

arXiv2022

Datasets for Online Controlled Experiments

Online Controlled Experiments (OCE) are the gold standard to measure impact and guide decisions for digital products and services. Despite many methodological advances in this area, the scarcity of public datasets and the lack of a systematic review and categorization hinder its development. We present the first survey and taxonomy for OCE datasets, which highlight the lack of a public dataset to support the design and running of experiments with adaptive stopping, an increasingly popular approach to enable quickly deploying improvements or rolling back degrading changes. We release the first such dataset, containing daily checkpoints of decision metrics from multiple, real experiments run on a global e-commerce platform. The dataset design is guided by a broader discussion on data requirements for common statistical tests used in digital experimentation. We demonstrate how to use the dataset in the adaptive stopping scenario using sequential and Bayesian hypothesis tests and learn the relevant parameters for each approach.

stat.AP↗

arXiv2021

Analysing the causal effect of London cycle superhighways on traffic congestion

Transport operators have a range of intervention options available to improve or enhance their networks. Such interventions are often made in the absence of sound evidence on resulting outcomes. Cycling superhighways were promoted as a sustainable and healthy travel mode, one of the aims of which was to reduce traffic congestion. Estimating the impacts that cycle superhighways have on congestion is complicated due to the non-random assignment of such intervention over the transport network. In this paper, we analyse the causal effect of cycle superhighways utilising pre-intervention and post-intervention information on traffic and road characteristics along with socio-economic factors. We propose a modeling framework based on the propensity score and outcome regression model. The method is also extended to the doubly robust set-up. Simulation results show the superiority of the performance of the proposed method over existing competitors. The method is applied to analyse a real dataset on the London transport network. The methodology proposed can assist in effective decision making to improve network performance.

stat.AP↗

arXiv2017

Multiply robust dose-response estimation for multivalued causal inference problems

This paper develops a multiply robust (MR) dose-response estimator for causal inference problems involving multivalued treatments. We combine a family of generalised propensity score (GPS) models and a family of outcome regression (OR) models to achieve an average potential outcomes estimator that is consistent if just one of the GPS or OR models in each family is correctly specified. We provide proofs and simulations that demonstrate multiple robustness in the context of multivalued causal inference problems.

stat.ME↗

arXiv2015

Doubly robust dose-response estimation for continuous treatments via generalized propensity score augmented outcome regression

This paper constructs a doubly robust estimator for continuous dose-response estimation. An outcome regression model is augmented with a set of inverse generalized propensity score covariates to correct for potential misspecification bias. From the augmented model we can obtain consistent estimates of mean average potential outcomes for distinct strata of the treatment. A polynomial regression is then fitted to these point estimates to derive a Taylor approximation to the continuous dose-response function. The bootstrap is used for variance estimation. Analytical results and simulations show that our approach can provide a good approximation to linear or nonlinear dose-response functions under various sources of misspecification of the outcome regression or propensity score models. Efficiency in finite samples is good relative to minimum variance consistent estimators.

math.ST↗

arXiv2007

Non-Regular Likelihood Inference for Seasonally Persistent Processes

The estimation of parameters in the frequency spectrum of a seasonally persistent stationary stochastic process is addressed. For seasonal persistence associated with a pole in the spectrum located away from frequency zero, a new Whittle-type likelihood is developed that explicitly acknowledges the location of the pole. This Whittle likelihood is a large sample approximation to the distribution of the periodogram over a chosen grid of frequencies, and constitutes an approximation to the time-domain likelihood of the data, via the linear transformation of an inverse discrete Fourier transform combined with a demodulation. The new likelihood is straightforward to compute, and as will be demonstrated has good, yet non-standard, properties. The asymptotic behaviour of the proposed likelihood estimators is studied; in particular, $N$-consistency of the estimator of the spectral pole location is established. Large finite sample and asymptotic distributions of the score and observed Fisher information are given, and the corresponding distributions of the maximum likelihood estimators are deduced. A study of the small sample properties of the likelihood approximation is provided, and its superior performance to previously suggested methods is shown, as well as agreement with the developed distributional approximations.

stat.ME↗