Sally Wood

Modelling the Impact of Personality on Individual Performance Behavior with a Time-Varying Mixture of Monotonic Random Effects

Sally A. Wood et al. — Thu, 24 Nov 2011 16:14:07 PST

A method is presented for flexibly modelling longitudinal data that provides insight to a central question in psychology theory: the dependency between personality clas- sification and individual performance behavior. Flexibility is achieved by assuming the regression coefficients of random effects models are generated from a time-varying mixture of an unknown but finite number of processes, where the weights attached to the number of processes are parameterised to depend upon an individual’s personality classification. For a given number of mixture components the component processes are constrained distributions and the weights attached to them depend upon time. The method is made robust to outliers and we demonstrate this is an important addition when making inference at the individual level. The frequentist properties of the ap- proach are examined via simulation. The results support the hypothesis in psychology that individuals who believe abilities are inherited traits are much more likely to exhibit sustained periods of failing performance than other individuals.

Locally Adaptive Nonparametric Binary Regression

Sally A. Wood et al. — Sat, 28 Nov 2009 23:17:03 PST

A Bayesian approach to ordinal outcomes for neurosurgical clinical research.

Sally A. Wood — Sat, 28 Nov 2009 22:57:54 PST

The objective of this study is to demonstrate a Bayesian approach for the statistical analysis of neurosurgical data where the investigators have used an ordinal scale for the outcome. A Bayesian approach that uses data augmentation and Gibbs sampling to perform ordinal probit regression is demonstrated in a neurosurgical context. The statistical approach is applied to a regression analysis to examine the relationship between female gender and the outcome of severe traumatic brain injury measured with the Glascow Outcome Scale. The approach is applied to a hierarchical meta-analysis to examine the relationship between age and the outcome from subarachnoid haemorrhage where the outcome is measured using ordinal scales. A model selection procedure is described for deriving the probability of statistical association between a regression covariates and the outcome variable. In the selection procedure the likelihood of the model is compared with alternative models that have random ordering of the covariate of interest. Results and Conclusions. The method had the flexibility and intuitive appeal of the Bayesian approach. The models provided a good fit to the data and useful predictive probabilities. The method has an underlying assumption that a normally distributed outcome variable has been categorized. The intuitive appeal of this assumption should make the method suitable to a wide range of neurosurgical studies including observational studies, controlled trials, and meta-analyses.

Mixture of random effects for individual learning curves

Sally A. Wood et al. — Sat, 28 Nov 2009 22:45:36 PST

In the pyschology literature individuals are often classified as entity theorists or incrementalists. In this paper we explore the different learning behaviours over time of these two groups. To assess learning an individual is assigned a task and their performance on the task is measured over a number of trials. Learning behaviour is modelled as a mixture of two random effects, where the random effects components of the mixture correspond to increased learning and spiralling behaviour. We find significant differences in the learning behaviours of the two groups. Specifically those individuals who are categorized as entity theorists are more likely to exhibit spiralling behaviour and have narrower search strategies than those categorized as incremnatalists.

Trans-dimensional Metropolis-Hastings Using Parallel Chains

Sally A. Wood et al. — Sat, 28 Nov 2009 22:27:22 PST

A general Bayesian sampling method is developed that uses parallel chains to select between models and to average the predictive density over such models. The method applies to both non-nested models and to nested models, and is particularly useful for mixtures of complex component models, where a novel approach to overcome the label-switching problem is used. The method is illustrated with real and simulated data in model-averaging over alternative financial time series models, mixtures of normal distributions, and mixtures of smoothing spline models.

Priors for a Bayesian Analysis of Extreme Values

Sally A. Wood et al. — Sat, 28 Nov 2009 22:20:56 PST

This article proposes a new prior specification for a Bayesian analysis of the k largest order statistics model. We show that using Jeffreys priors for the end-point and shape parameters of the k largest order statistics model leads to biased estimates of the shape parameter for small to medium sample sizes and to the posterior mode of the end-point being equal to the most extreme observed value. We propose a conjugate prior for the shape parameter and a prior for the end-point which removes the posterior mode at the most extreme observed value while remaining uninformative for values of the end-point away from this value. We show by simulation that the proposed priors perform well even when the sample sizes are small and/or when the shape parameter is less than one. We illustrate the performance of our method by comparing the tail distributions of female and male lifespans and by analysing the improvement in time of the men's 100m sprint from the 68 &72 Olympic games to the 1988 &1992 Olympics.

Bayesian Mixtures of Autoregressive Models

Sally A. Wood et al. — Sat, 28 Nov 2009 22:14:05 PST

In this paper we propose a class of time-domain models for analyzing possibly nonstationary time series. This class of models is formed as a mixture of time series models, whose mixing weights are a function of time. We consider specifically mixtures of autoregressive models with a common but unknown lag. The model parameters, including the number of mixture components, are estimated via Markov chain Monte Carlo methods. The methodology is illustrated with simulated and real data.

Local spectral analysis via a Bayesian mixture of smoothing splines” Journal of the American Statistical Association

Sally A. Wood et al. — Sat, 28 Nov 2009 22:04:56 PST