Search

Traditional methods of model diagnostics may include a plethora of graphical techniques based on residual analysis, as well as formal tests (e.g. Shapiro-Wilk test for normality and Bartlett test for homogeneity of variance). In this paper we derive a new distance metric based on the half-normal plot with a simulation envelope, a graphical model evaluation method, and investigate its properties through simulation studies. The proposed metric can help to assess the fit of a given model, and also act as a model selection criterion by being comparable across models, whether based or not on a true likelihood. More specifically, it quantitatively encompasses the model evaluation principles and removes the subjective bias when closely related models are involved. We validate the technique by means of an extensive simulation study carried out using count data, and illustrate with two case studies in ecology and fisheries research.

This study investigates the impact of Education Endowment Foundation (EEF)-funded trials on pupils eligible for free school meals. Although similar analysis is conducted during each individual evaluation, this report conducts a meta-analysis using data from 88 trials and over half a million pupils to reach conclusions. The report contributes to the evidence about what type of approaches may be effective at reducing the attainment gap, which influences the choice of interventions that EEF chooses to trial and scale up. This study investigates the immediate impact of EEF-funded interventions on attainment in literacy and mathematics based on the following research questions: (1) Do EEF trials improve literacy/mathematics attainment of pupils eligible for FSM? (2) What broad types of interventions are more beneficial for mathematics and literacy outcomes of FSM pupils? and (3) Do FSM pupils improve their literacy/mathematics attainment more or less from EEF-funded interventions compared to their non-FSM peers? [This report was written in partnership with the Durham Research Methods Centre (DRMC).]

We propose a cautious Bayesian variable selection routine by investigating the sensitivity of a hierarchical model, where the regression coefficients are specified by spike and slab priors. We exploit the use of latent variables to understand the importance of the co-variates. These latent variables also allow us to obtain the size of the model space which is an important aspect of high dimensional problems. In our approach, instead of fixing a single prior, we adopt a specific type of robust Bayesian analysis, where we consider a set of priors within the same parametric family to specify the selection probabilities of these latent variables. We achieve that by considering a set of expected prior selection probabilities, which allows us to perform a sensitivity analysis to understand the effect of prior elicitation on the variable selection. The sensitivity analysis provides us sets of posteriors for the regression coefficients as well as the selection indicators and we show that the posterior odds of the model selection probabilities are monotone with respect to the prior expectations of the selection probabilities. We also analyse synthetic and real life datasets to illustrate our cautious variable selection method and compare it with other well known methods., Comment: 33 pages, 5 figures, 2 tables

Proteins are essential for maintaining life. For example, knowing the structure of a protein, cell regulatory mechanisms of organisms can be modeled, supporting the development of disease treatments or the understanding of relationships between protein structures and food attributes. However, discovering the structure of a protein can be a difficult and expensive task, since it is hard to explore the large search to predict even a small protein. Template-based methods (coarse-grained, homology, threading etc) depend on Prior Knowledge (PK) of proteins determined using other methods as X-Ray Crystallography or Nuclear Magnetic Resonance. On the other hand, template-free methods (full-atom and ab initio) rely on atoms physical-chemical properties to predict protein structures. In comparison with other approaches, the Estimation of Distribution Algorithms (EDAs) can require significant less PK, suggesting that it could be adequate for proteins of low-level of PK. Finding an EDA able to handle both prediction quality and computational time is a difficult task, since they are strong inversely correlated. We developed an EDA specific for the ab initio Protein Structure Prediction (PSP) problem using full-atom representation. We developed one univariate and two bivariate probabilistic models in order to design a proper EDA for PSP. The bivariate models make relationships between dihedral angles $\phi$ and $\psi$ within an amino acid. Furthermore, we compared the proposed EDA with other approaches from the literature. We noticed that even a relatively simple algorithm such as Random Walk can find the correct solution, but it would require a large amount of prior knowledge (biased prediction). On the other hand, our EDA was able to correctly predict with no prior knowledge at all, characterizing such a prediction as pure ab initio., Comment: 45 pages, 13 figures

Shape information is of great importance in many applications. For example, the oil-bearing capacity of sand bodies, the subterranean remnants of ancient rivers, is related to their cross-sectional shapes. The analysis of these shapes is therefore of some interest, but current classifications are simplistic and ad hoc. In this paper, we describe the first steps towards a coherent statistical analysis of these shapes by deriving the integrated likelihood for data shapes given class parameters. The result is of interest beyond this particular application., Comment: 4 pages, 1 figure, In proceedings 29th International Workshop on Statistical Modelling, 14-18 July 2014, Gottingen, Germany. Amsterdam: Statistical Modelling Society, pp. 161-164

Often the relation between the variables constituting a multivariate data space might be characterized by one or more of the terms: ``nonlinear'', ``branched'', ``disconnected'', ``bended'', ``curved'', ``heterogeneous'', or, more general, ``complex''. In these cases, simple principal component analysis (PCA) as a tool for dimension reduction can fail badly. Of the many alternative approaches proposed so far, local approximations of PCA are among the most promising. This paper will give a short review of localized versions of PCA, focusing on local principal curves and local partitioning algorithms. Furthermore we discuss projections other than the local principal components. When performing local dimension reduction for regression or classification problems it is important to focus not only on the manifold structure of the covariates, but also on the response variable(s). Local principal components only achieve the former, whereas localized regression approaches concentrate on the latter. Local projection directions derived from the partial least squares (PLS) algorithm offer an interesting trade-off between these two objectives. We apply these methods to several real data sets. In particular, we consider simulated astrophysical data from the future Galactic survey mission Gaia., Comment: 25 pages. In "Principal Manifolds for Data Visualization and Dimension Reduction", A. Gorban, B. Kegl, D. Wunsch, and A. Zinovyev (eds), Lecture Notes in Computational Science and Engineering, Springer, 2007, pp. 180--204, http://www.springer.com/dal/home/generic/search/results?SGWID=1-40109-22-173750210-0