My research has focussed on methodological, computational and foundational issues in Bayesian statistics, asymptotic theory of statistical inference and reliability analysis. I have also carried out work on limit theory in probability. Bayesian ideas, or at least notions of conditionality, have motivated much of my work in methodological and theoretical statistics, even if this is not outwardly Bayesian.
Bayesian inference has been a major research interest for many years. Bayesian asymptotic theory is reviewed below. Current research interests include objective Bayesian inference based on coverage probability bias and probability matching priors (Biometrika, 2001, 2005, Handbook of Statistics, 2005, Festschrift for J. K. Ghosh, 2008), relative entropy loss for Bayesian prediction (Ann. Statist.., 2006) and nonparametric Bayesian methods. An invited paper on Bayesian computation based on directed likelihood appeared in Bayesian Statistics 5 (1996). Other papers on this topic and related hybrid computational schemes include Test (2003), Statistics & Computing (2005) and Bayesian Anal. (2010).
Other research has concerned various aspects of generalised regression models, such as approximate analysis of data from location-scale regression models (Biometrika, 1987, for example). Applications of this work include the analysis of censored survival data arising in the medical and biological sciences, and of failure-time data arising in reliability studies. Work on likelihood and Bayesian inference for ratios of regression coefficients in linear models appeared in Ann. Inst. Stat. Math. (2006).
The analysis of transformations in regression models (for example, Bayesian Statistics 2, 1985) has been another topic of interest. This work provides some insight into stable parameterisations and prior specifications for statistical problems, and there are interesting links with parameter orthogonality (c.f. my discussion of D. R. Cox and N. Reid in J. R. Statist. Soc. B, 1987). Parameterisation issues more generally are addressed in a number of my papers.
Asymptotic theory of statistical inference
This has been a major research topic over the years. Some of this work involved the extension of classical asymptotic theory for independent observations to dependent data arising from stochastic processes, including asymptotic normality of maximum likelihood estimators (Ann. Statist., 1980). Subsequent work involved asymptotic conditional inference and asymptotic ancillarity in nonergodic models. The case of a branching process was analysed in Ann. Statist. (1986) and a general approach appeared in Ann. Statist. (1992).
Results on asymptotic theory of Bayesian inference for stochastic processes appeared in J. R. Statist. Soc. B (1987) and some very general results for multiparameter models for dependent data were obtained in Bayesian Statistics 4 (1992). In particular, this paper addressed the problems that arise when information on different parameters grow at different rates.
Research on higher-order asymptotic theory has focussed on connections between Bayesian and (conditional) frequentist inferences. See, for example, my discussions of the papers by A. P. Dawid (J. R. Statist. Soc. B, 1991) and D. A. Pierce and D. Peters (J. R. Statist. Soc. B, 1992) and the papers Biometrika (1995 x 2), J. R. Statist. Soc. B (1999) and Ann. Statist. (2000), Biometrika (2005) and IMS Collections (2008).
Other research carried out in this area includes (i) asymptotic behaviour of spacings, which has application to multiple outlier detection (Ann. Statist., 1986), (ii) asymptotic problems associated with the prequential approach (Scand. J. Statist., 1992) and (iii) asymptotic theory based on a suitable function of time and the unknown parameters (Biometrika, 1992).
A variety of reliability problems have been pursued with statistics colleagues and a book on the analysis of reliability data was published by Chapman and Hall (1991). Collaborative industrial research involving non-destructive testing (NDT) led to some new results on parsimonious models for NDT reliability data, based on post-inspection flaw size distribution and safe cyclic life (Int. Statist. Review, 1995), presented at the ISI, 1993.
An EPSRC-funded research project (SMST) completed in 2002 concerned the modelling and analysis of degradation data.
Limit theory in probability
My early work in this area involved investigation of the rate of convergence in the multivariate central limit theorem and associated asymptotic expansions. The approach via convolution operators, rather than the more usual approach via characteristic functions, enabled known results to be improved and extended (Ann. Probability, 1977, 1980). In particular, Berry-Esséen-type bounds were shown to be valid for arbitrary Borel sets.
Extreme value theory has been another interest. In Ann. Probability (1985) identification of a certain slowly varying function provided the key to a number of characterisations which simplified and extended existing results in local limit theory. Suitable expansions of regularly varying functions lead to asymptotic expansions for extreme quantiles (Lecture Notes in Statist, 1989). Local and conditional limit theory more generally has been a topic of interest, with papers appearing in Ann. Statist. (1986) and J. Theoret. Prob. (1989).
I have also been involved with a number of collaborative projects, for example Midhurst Medical Research Institute (investigation of the magnitude of a possible direct effect of beta-blockage on myocardial oxygen consumption), Civil Engineers (relationship between the porosity of a granular material and its particle size distribution), Royal Aerospace Establishment/DRA (in-service inspection requirements for airworthiness, construction of safe S-N curves), Rolls-Royce (design and analysis of non-destructive testing methods aimed at gas turbine discs), Medics (clearance kinetics for patients with Crohn’s disease), MRC Prion Unit (quantification of prion infected titres), Computer Scientists (information retrieval systems).
Some recent publications
Murugiah, S. and Sweeting, T. J. (2012). Selecting the precision parameter prior in Dirichlet process mixture model. Journal of Statistical Planning and Inference, 142, 1947-1959. doi: 10.1016/j.jspi.2012.02.013
Castro-Seoane, R., Hummerich, H., Sweeting, T., Tattum, H., Sandberg, M., Collinge, J. and Klöhn, P. (2012). Plasmacytoid dendritic cells sequester high prion titres at early stages of prion infection. PLOS Pathog., 8(2): e1002538. DOI: 10.1371/journal.ppat.1002538
Sweeting, T. J. (2011). Invited discussion of M. Ghosh : Objective priors: an introduction for frequentists. Statistical Science 26, 206-209. doi: 10.1214/11-STS338B
Hosseini, M., Cox, I., Milic-Frayling, N., Vinay, V., Sweeting, T. (2011). Selecting a subset of queries for acquisition of further relevance judgements. Advances in Information Retrieval Theory, Lecture Notes in Computing Science, 6031/2011, 113–124. DOI: 10.1007/978-3-642-23318-0_12
Kharroubi, S.A. and Sweeting, T.J. (2010). Posterior simulation via the signed root log-likelihood ratio. Bayesian Anal., 5, 787-815.
Smith, A.M., F.Z. Rahman, B.H. Hayee, S.J. Graham, D.J.B. Marks, G.W. Sewell, C.D. Palmer, J. Wilde, B.M.J. Foxwell, I.S. Gloger, T. J. Sweeting, M. Marsh, A. P. Walker, Stuart L. Bloom and A. W. Segal (2009). Disordered macrophage cytokine secretion underlies impaired acute inflammation and bacterial clearance in Crohn's disease. J. Exp. Med. 206:1883–1897 (doi:10.1084/jem.20091233)
Sweeting, T. J. (2008). On predictive probability matching priors. IMS Collections: Pushing the Limits of Contemporary Statistics: Contributions in Honor of Jayanta K. Ghosh (B. Clarke and S. Ghosal, eds.), 3, 46-59.