Estimation of Gaussian process regression model using probability distance measures

A new class of parameter estimation algorithms is introduced for Gaussian process regression (GPR) models. It is shown that the integration of the GPR model with probability distance measures of (i) the integrated square error and (ii) Kullback–Leibler (K–L) divergence are analytically tractable. An efficient coordinate descent algorithm is proposed to iteratively estimate the kernel width using golden section search which includes a fast gradient descent algorithm as an inner loop to estimate the noise variance. Numerical examples are included to demonstrate the effectiveness of the new identification approaches.

A parametric model for the interval censored survival times of Acacia mangium plantation in a spacing trial

Survival times for the Acacia mangium plantation in the Segaliud Lokan Project, Sabah, East Malaysia were analysed based on 20 permanent sample plots (PSPs) established in 1988 as a spacing experiment. The PSPs were established following a complete randomized block design with five levels of spacing randomly assigned to units within four blocks at different sites. The survival times of trees in years are of interest. Since the inventories were only conducted annually, the actual survival time for each tree was not observed. Hence, the data set comprises censored survival times. Initial analysis of the survival of the Acacia mangium plantation suggested there is block by spacing interaction; a Weibull model gives a reasonable fit to the replicate survival times within each PSP; but a standard Weibull regression model is inappropriate because the shape parameter differs between PSPs. In this paper we investigate the form of the non-constant Weibull shape parameter. Parsimonious models for the Weibull survival times have been derived using maximum likelihood methods. The factor selection for the parameters is based on a backward elimination procedure. The models are compared using likelihood ratio statistics. The results suggest that both Weibull parameters depend on spacing and block.

Power properties if invariant tests for spatial autocorrelation in linear regression

This paper derives some exact power properties of tests for spatial autocorrelation in the context of a linear regression model. In particular, we characterize the circumstances in which the power vanishes as the autocorrelation increases, thus extending the work of Krämer (2005). More generally, the analysis in the paper sheds new light on how the power of tests for spatial autocorrelation is affected by the matrix of regressors and by the spatial structure. We mainly focus on the problem of residual spatial autocorrelation, in which case it is appropriate to restrict attention to the class of invariant tests, but we also consider the case when the autocorrelation is due to the presence of a spatially lagged dependent variable among the regressors. A numerical study aimed at assessing the practical relevance of the theoretical results is included

Investigation of the factors influencing the survival of Bifidobacterium longum in model acidic solutions and fruit juices

The survival of Bifidobacterium longum NCIMB 8809 was studied during refrigerated storage for 6 weeks in model solutions, based on which a mathematical model was constructed describing cell survival as a function of pH, citric acid, protein and dietary fibre. A Central Composite Design (CCD) was developed studying the influence of four factors at three levels, i.e., pH (3.2–4), citric acid (2–15 g/l), protein (0–10 g/l), and dietary fibre (0–8 g/l). In total, 31 experimental runs were carried out. Analysis of variance (ANOVA) of the regression model demonstrated that the model fitted well the data. From the regression coefficients it was deduced that all four factors had a statistically significant (P < 0.05) negative effect on the log decrease [log10N0 week−log10N6 week], with the pH and citric acid being the most influential ones. Cell survival during storage was also investigated in various types of juices, including orange, grapefruit, blackcurrant, pineapple, pomegranate and strawberry. The highest cell survival (less than 0.4 log decrease) after 6 weeks of storage was observed in orange and pineapple, both of which had a pH of about 3.8. Although the pH of grapefruit and blackcurrant was similar (pH ∼3.2), the log decrease of the former was ∼0.5 log, whereas of the latter was ∼0.7 log. One reason for this could be the fact that grapefruit contained a high amount of citric acid (15.3 g/l). The log decrease in pomegranate and strawberry juices was extremely high (∼8 logs). The mathematical model was able to predict adequately the cell survival in orange, grapefruit, blackcurrant, and pineapple juices. However, the model failed to predict the cell survival in pomegranate and strawberry, most likely due to the very high levels of phenolic compounds in these two juices.

A unified neurofuzzy model for classification

This work proposes a unified neurofuzzy modelling scheme. To begin with, the initial fuzzy base construction method is based on fuzzy clustering utilising a Gaussian mixture model (GMM) combined with the analysis of covariance (ANOVA) decomposition in order to obtain more compact univariate and bivariate membership functions over the subspaces of the input features. The mean and covariance of the Gaussian membership functions are found by the expectation maximisation (EM) algorithm with the merit of revealing the underlying density distribution of system inputs. The resultant set of membership functions forms the basis of the generalised fuzzy model (GFM) inference engine. The model structure and parameters of this neurofuzzy model are identified via the supervised subspace orthogonal least square (OLS) learning. Finally, instead of providing deterministic class label as model output by convention, a logistic regression model is applied to present the classifier’s output, in which the sigmoid type of logistic transfer function scales the outputs of the neurofuzzy model to the class probability. Experimental validation results are presented to demonstrate the effectiveness of the proposed neurofuzzy modelling scheme.

Tensor regression based on linked multiway parameter analysis

Classical regression methods take vectors as covariates and estimate the corresponding vectors of regression parameters. When addressing regression problems on covariates of more complex form such as multi-dimensional arrays (i.e. tensors), traditional computational models can be severely compromised by ultrahigh dimensionality as well as complex structure. By exploiting the special structure of tensor covariates, the tensor regression model provides a promising solution to reduce the model’s dimensionality to a manageable level, thus leading to efficient estimation. Most of the existing tensor-based methods independently estimate each individual regression problem based on tensor decomposition which allows the simultaneous projections of an input tensor to more than one direction along each mode. As a matter of fact, multi-dimensional data are collected under the same or very similar conditions, so that data share some common latent components but can also have their own independent parameters for each regression task. Therefore, it is beneficial to analyse regression parameters among all the regressions in a linked way. In this paper, we propose a tensor regression model based on Tucker Decomposition, which identifies not only the common components of parameters across all the regression tasks, but also independent factors contributing to each particular regression task simultaneously. Under this paradigm, the number of independent parameters along each mode is constrained by a sparsity-preserving regulariser. Linked multiway parameter analysis and sparsity modeling further reduce the total number of parameters, with lower memory cost than their tensor-based counterparts. The effectiveness of the new method is demonstrated on real data sets.

Observed trends and teleconnections of the Siberian high: A recently declining center of action

This study investigates variability in the intensity of the wintertime Siberian high (SH) by defining a robust SH index (SHI) and correlating it with selected meteorological fields and teleconnection indices. A dramatic trend of -2.5 hPa decade(-1) has been found in the SHI between 1978 and 2001 with unprecedented (since 1871) low values of the SHI. The weakening of the SH has been confirmed by analyzing different historical gridded analyses and individual station observations of sea level pressure (SLP) and excluding possible effects from the conversion of surface pressure to SLP. SHI correlation maps with various meteorological fields show that SH impacts on circulation and temperature patterns extend far outside the SH source area extending from the Arctic to the tropical Pacific. Advection of warm air from eastern Europe has been identified as the main mechanism causing milder than normal conditions over the Kara and Laptev Seas in association with a strong SH. Despite the strong impacts of the variability in the SH on climatic variability across the Northern Hemisphere, correlations between the SHI and the main teleconnection indices of the Northern Hemisphere are weak. Regression analysis has shown that teleconnection indices are not able to reproduce the interannual variability and trends in the SH. The inclusion of regional surface temperature in the regression model provides closer agreement between the original and reconstructed SHI.

Individual differences in the effects of perceived controllability on pain perception: critical role of the prefrontal cortex

The degree to which perceived controllability alters the way a stressor is experienced varies greatly among individuals. We used functional magnetic resonance imaging to examine the neural activation associated with individual differences in the impact of perceived controllability on self-reported pain perception. Subjects with greater activation in response to uncontrollable (UC) rather than controllable (C) pain in the pregenual anterior cingulate cortex (pACC), periaqueductal gray (PAG), and posterior insula/SII reported higher levels of pain during the UC versus C conditions. Conversely, subjects with greater activation in the ventral lateral prefrontal cortex (VLPFC) in anticipation of pain in the UC versus C conditions reported less pain in response to UC versus C pain. Activation in the VLPFC was significantly correlated with the acceptance and denial subscales of the COPE inventory [Carver, C. S., Scheier, M. F., & Weintraub, J. K. Assessing coping strategies: A theoretically based approach. Journal of Personality and Social Psychology, 56, 267–283, 1989], supporting the interpretation that this anticipatory activation was associated with an attempt to cope with the emotional impact of uncontrollable pain. A regression model containing the two prefrontal clusters (VLPFC and pACC) predicted 64% of the variance in pain rating difference, with activation in the two additional regions (PAG and insula/SII) predicting almost no additional variance. In addition to supporting the conclusion that the impact of perceived controllability on pain perception varies highly between individuals, these findings suggest that these effects are primarily top-down, driven by processes in regions of the prefrontal cortex previously associated with cognitive modulation of pain and emotion regulation.

The income requirements of marine protected areas

Given the growing impact of human activities on the sea, managers are increasingly turning to marine protected areas (MPAs) to protect marine habitats and species. Many MPAs have been unsuccessful, however, and lack of income has been identified as a primary reason for failure. In this study, data from a global survey of 79 MPAs in 36 countries were analysed and attempts made to construct predictive models to determine the income requirements of any given MPA. Statistical tests were used to uncover possible patterns and relationships in the data, with two basic approaches. In the first of these, an attempt was made to build an explanatory "bottom-up" model of the cost structures that might be required to pursue various management activities. This proved difficult in practice owing to the very broad range of applicable data, spanning many orders of magnitude. In the second approach, a "top-down" regression model was constructed using logarithms of the base data, in order to address the breadth of the data ranges. This approach suggested that MPA size and visitor numbers together explained 46% of the minimum income requirements (P < 0.001), with area being the slightly more influential factor. The significance of area to income requirements was of little surprise, given its profile in the literature. However, the relationship between visitors and income requirements might go some way to explaining why northern hemisphere MPAs with apparently high incomes still claim to be under-funded. The relationship between running costs and visitor numbers has important implications not only in determining a realistic level of funding for MPAs, but also in assessing from where funding might be obtained. Since a substantial proportion of the income of many MPAs appears to be utilized for amenity purposes, a case may be made for funds to be provided from the typically better resourced government social and educational budgets as well as environmental budgets. Similarly visitor fees, already an important source of funding for some MPAs, might have a broader role to play in how MPAs are financed in the future. (C) 2007 Elsevier Ltd. All rights reserved.

A revised Tobit procedure for mitigating bias in the presence of non-zero censoring with an application to milk-market participation in the Ethiopian highlands

Fixed transactions costs that prohibit exchange engender bias in supply analysis due to censoring of the sample observations. The associated bias in conventional regression procedures applied to censored data and the construction of robust methods for mitigating bias have been preoccupations of applied economists since Tobin [Econometrica 26 (1958) 24]. This literature assumes that the true point of censoring in the data is zero and, when this is not the case, imparts a bias to parameter estimates of the censored regression model. We conjecture that this bias can be significant; affirm this from experiments; and suggest techniques for mitigating this bias using Bayesian procedures. The bias-mitigating procedures are based on modifications of the key step that facilitates Bayesian estimation of the censored regression model; are easy to implement; work well in both small and large samples; and lead to significantly improved inference in the censored regression model. These findings are important in light of the widespread use of the zero-censored Tobit regression and we investigate their consequences using data on milk-market participation in the Ethiopian highlands. (C) 2004 Elsevier B.V. All rights reserved.

Domestically produced food: Consumer perceptions of origin, safety and the issue of trust

It is becoming increasingly difficult for the public to attempt to assess risks using traditional methods such as smell, taste or other physical attributes of food. The existence of extrinsic cues such as the country of origin (COO) of food can help to make food purchase decisions easier for consumers. However, the use of extrinsic cues depends heavily on the extent to which consumers trust such signals to be indicative of quality or safety, which in turn depends on the credibility behind that cue. This paper aims to examine consumers association of domestically produced food with increased food safety standards and the association of COO and food safety information with socio-demographics and other aspects of consumer psychology such as attitudes, risk perception and trust. Using an ordered probit model, domestic production is examined as an extrinsic cue for food safety by looking at the relationship with trust in food safety information provided by national food standards agencies (NFSAs) and other socio-demographic characteristics, based on nationally representative data from 2725 face-to-face interviews across five European countries. Results suggest that domestic production of food is an extrinsic cue for food safety and as consumers place increasing importance on food safety they are more interested in food produced in their own country. This, coupled with consumer trust in a strong, and independent national food standards agency, suggests the potential exists for the increased consumption of domestically produced foods.

Modelling risk perception and trust in food safety information within the theory of planned behaviour

This study suggests a statistical strategy for explaining how food purchasing intentions are influenced by different levels of risk perception and trust in food safety information. The modelling process is based on Ajzen's Theory of Planned Behaviour and includes trust and risk perception as additional explanatory factors. Interaction and endogeneity across these determinants is explored through a system of simultaneous equations, while the SPARTA equation is estimated through an ordered probit model. Furthermore, parameters are allowed to vary as a function of socio-demographic variables. The application explores chicken purchasing intentions both in a standard situation and conditional to an hypothetical salmonella scare. Data were collected through a nationally representative UK wide survey of 533 UK respondents in face-to-face, in-home interviews. Empirical findings show that interactions exist among the determinants of planned behaviour and socio-demographic variables improve the model's performance. Attitudes emerge as the key determinant of intention to purchase chicken, while trust in food safety information provided by media reduces the likelihood to purchase. (C) 2006 Elsevier Ltd. All rights reserved.

Meta-analysis: a unifying meta-likelihood approach framing unobserved heterogeneity, study covariates, publication bias, and study quality

OBJECTIVES: This contribution provides a unifying concept for meta-analysis integrating the handling of unobserved heterogeneity, study covariates, publication bias and study quality. It is important to consider these issues simultaneously to avoid the occurrence of artifacts, and a method for doing so is suggested here. METHODS: The approach is based upon the meta-likelihood in combination with a general linear nonparametric mixed model, which lays the ground for all inferential conclusions suggested here. RESULTS: The concept is illustrated at hand of a meta-analysis investigating the relationship of hormone replacement therapy and breast cancer. The phenomenon of interest has been investigated in many studies for a considerable time and different results were reported. In 1992 a meta-analysis by Sillero-Arenas et al. concluded a small, but significant overall effect of 1.06 on the relative risk scale. Using the meta-likelihood approach it is demonstrated here that this meta-analysis is due to considerable unobserved heterogeneity. Furthermore, it is shown that new methods are available to model this heterogeneity successfully. It is argued further to include available study covariates to explain this heterogeneity in the meta-analysis at hand. CONCLUSIONS: The topic of HRT and breast cancer has again very recently become an issue of public debate, when results of a large trial investigating the health effects of hormone replacement therapy were published indicating an increased risk for breast cancer (risk ratio of 1.26). Using an adequate regression model in the previously published meta-analysis an adjusted estimate of effect of 1.14 can be given which is considerably higher than the one published in the meta-analysis of Sillero-Arenas et al. In summary, it is hoped that the method suggested here contributes further to a good meta-analytic practice in public health and clinical disciplines.

Bayesian decision procedures for binary and continuous bivariate dose-escalation studies

In this paper, Bayesian decision procedures are developed for dose-escalation studies based on binary measures of undesirable events and continuous measures of therapeutic benefit. The methods generalize earlier approaches where undesirable events and therapeutic benefit are both binary. A logistic regression model is used to model the binary responses, while a linear regression model is used to model the continuous responses. Prior distributions for the unknown model parameters are suggested. A gain function is discussed and an optional safety constraint is included. Copyright (C) 2006 John Wiley & Sons, Ltd.

A practical approach to a multi-level analysis with a sparse binary outcome within a large surgical trial

Objectives: To assess the potential source of variation that surgeon may add to patient outcome in a clinical trial of surgical procedures. Methods: Two large (n = 1380) parallel multicentre randomized surgical trials were undertaken to compare laparoscopically assisted hysterectomy with conventional methods of abdominal and vaginal hysterectomy; involving 43 surgeons. The primary end point of the trial was the occurrence of at least one major complication. Patients were nested within surgeons giving the data set a hierarchical structure. A total of 10% of patients had at least one major complication, that is, a sparse binary outcome variable. A linear mixed logistic regression model (with logit link function) was used to model the probability of a major complication, with surgeon fitted as a random effect. Models were fitted using the method of maximum likelihood in SAS((R)). Results: There were many convergence problems. These were resolved using a variety of approaches including; treating all effects as fixed for the initial model building; modelling the variance of a parameter on a logarithmic scale and centring of continuous covariates. The initial model building process indicated no significant 'type of operation' across surgeon interaction effect in either trial, the 'type of operation' term was highly significant in the abdominal trial, and the 'surgeon' term was not significant in either trial. Conclusions: The analysis did not find a surgeon effect but it is difficult to conclude that there was not a difference between surgeons. The statistical test may have lacked sufficient power, the variance estimates were small with large standard errors, indicating that the precision of the variance estimates may be questionable.