Point-wise confidence interval estimation by neural networks: A comparative study based on automotive engine calibration.

In developing neural network techniques for real world applications it is still very rare to see estimates of confidence placed on the neural network predictions. This is a major deficiency, especially in safety-critical systems. In this paper we explore three distinct methods of producing point-wise confidence intervals using neural networks. We compare and contrast Bayesian, Gaussian Process and Predictive error bars evaluated on real data. The problem domain is concerned with the calibration of a real automotive engine management system for both air-fuel ratio determination and on-line ignition timing. This problem requires real-time control and is a good candidate for exploring the use of confidence predictions due to its safety-critical nature.

Conditional Confidence Interval for the Scale Parameter of a Weibull Distribution

2000 Mathematics Subject Classification: 62F25, 62F03.

Minimising speaker verification utterance length through confidence based early verification decisions

This paper presents a novel approach of estimating the confidence interval of speaker verification scores. This approach is utilised to minimise the utterance lengths required in order to produce a confident verification decision. The confidence estimation method is also extended to address both the problem of high correlation in consecutive frame scores, and robustness with very limited training samples. The proposed technique achieves a drastic reduction in the typical data requirements for producing confident decisions in an automatic speaker verification system. When evaluated on the NIST 2005 SRE, the early verification decision method demonstrates that an average of 5–10 seconds of speech is sufficient to produce verification rates approaching those achieved previously using an average in excess of 100 seconds of speech.

Confidence to cook vegetables and the buying habits of Australian households

Cooking skills are emphasized in nutrition promotion but their distribution among population subgroups and relationship to dietary behavior is researched by few population-based studies. This study examined the relationships between confidence to cook, sociodemographic characteristics, and household vegetable purchasing. This cross-sectional study of 426 randomly selected households in Brisbane, Australia, used a validated questionnaire to assess household vegetable purchasing habits and the confidence to cook of the person who most often prepares food for these households. The mutually adjusted odds ratios (ORs) of lacking confidence to cook were assessed across a range of demographic subgroups using multiple logistic regression models. Similarly, mutually adjusted mean vegetable purchasing scores were calculated using multiple linear regression for different population groups and for respondents with varying confidence levels. Lacking confidence to cook using a variety of techniques was more common among respondents with less education (OR 3.30; 95% confidence interval [CI] 1.01 to 10.75) and was less common among respondents who lived with minors (OR 0.22; 95% CI 0.09 to 0.53) and other adults (OR 0.43; 95% CI 0.24 to 0.78). Lack of confidence to prepare vegetables was associated with being male (OR 2.25; 95% CI 1.24 to 4.08), low education (OR 6.60; 95% CI 2.08 to 20.91), lower household income (OR 2.98; 95% CI 1.02 to 8.72) and living with other adults (OR 0.53; 95% CI 0.29 to 0.98). Households bought a greater variety of vegetables on a regular basis when the main chef was confident to prepare them (difference: 18.60; 95% CI 14.66 to 22.54), older (difference: 8.69; 95% CI 4.92 to 12.47), lived with at least one other adult (difference: 5.47; 95% CI 2.82 to 8.12) or at least one minor (difference: 2.86; 95% CI 0.17 to 5.55). Cooking skills may contribute to socioeconomic dietary differences, and may be a useful strategy for promoting fruit and vegetable consumption, particularly among socioeconomically disadvantaged groups.

Confidence intervals for half-life deviations from purchasing power parity

Existing point estimates of half-life deviations from purchasing power parity (PPP), around 3-5 years, suggest that the speed of convergence is extremely slow. This article assesses the degree of uncertainty around these point estimates by using local-to-unity asymptotic theory to construct confidence intervals that are robust to high persistence in small samples. The empirical evidence suggests that the lower bound of the confidence interval is between four and eight quarters for most currencies, which is not inconsistent with traditional price-stickiness explanations. However, the upper bounds are infinity for all currencies, so we cannot provide conclusive evidence in favor of PPP either. © 2005 American Statistical Association.

Substitution of (R,S)-methadone by (R)-methadone: Impact on QTc interval.

Methadone is administered as a chiral mixture of (R,S)-methadone. The opioid effect is mainly mediated by (R)-methadone, whereas (S)-methadone blocks the human ether-à-go-go-related gene (hERG) voltage-gated potassium channel more potently, which can cause drug-induced long QT syndrome, leading to potentially lethal ventricular tachyarrhythmias. To investigate whether substitution of (R,S)-methadone by (R)-methadone could reduce the corrected QT (QTc) interval, (R,S)-methadone was replaced by (R)-methadone (half-dose) in 39 opioid-dependent patients receiving maintenance treatment for 14 days. (R)-methadone was then replaced by the initial dose of (R,S)-methadone for 14 days (n = 29). Trough (R)-methadone and (S)-methadone plasma levels and electrocardiogram measurements were taken. The Fridericia-corrected QT (QTcF) interval decreased when (R,S)-methadone was replaced by a half-dose of (R)-methadone; the median (interquartile range [IQR]) values were 423 (398-440) milliseconds (ms) and 412 (395-431) ms (P = .06) at days 0 and 14, respectively. Using a univariate mixed-effect linear model, the QTcF value decreased by a mean of -3.9 ms (95% confidence interval [CI], -7.7 to -0.2) per week (P = .04). The QTcF value increased when (R)-methadone was replaced by the initial dose of (R,S)-methadone for 14 days; median (IQR) values were 424 (398-436) ms and 424 (412-443) ms (P = .01) at days 14 and 28, respectively. The univariate model showed that the QTcF value increased by a mean of 4.7 ms (95% CI, 1.3-8.1) per week (P = .006). Substitution of (R,S)-methadone by (R)-methadone reduces the QTc interval value. A safer cardiac profile of (R)-methadone is in agreement with previous in vitro and pharmacogenetic studies. If the present results are confirmed by larger studies, (R)-methadone should be prescribed instead of (R,S)-methadone to reduce the risk of cardiac toxic effects and sudden death.

Point estimates and confidence regions for sequential trials involving selection

A number of authors have proposed clinical trial designs involving the comparison of several experimental treatments with a control treatment in two or more stages. At the end of the first stage, the most promising experimental treatment is selected, and all other experimental treatments are dropped from the trial. Provided it is good enough, the selected experimental treatment is then compared with the control treatment in one or more subsequent stages. The analysis of data from such a trial is problematic because of the treatment selection and the possibility of stopping at interim analyses. These aspects lead to bias in the maximum-likelihood estimate of the advantage of the selected experimental treatment over the control and to inaccurate coverage for the associated confidence interval. In this paper, we evaluate the bias of the maximum-likelihood estimate and propose a bias-adjusted estimate. We also propose an approach to the construction of a confidence region for the vector of advantages of the experimental treatments over the control based on an ordering of the sample space. These regions are shown to have accurate coverage, although they are also shown to be necessarily unbounded. Confidence intervals for the advantage of the selected treatment are obtained from the confidence regions and are shown to have more accurate coverage than the standard confidence interval based upon the maximum-likelihood estimate and its asymptotic standard error.

Quantitative evaluation in health promotion : evaluation with confidence

Over the past 10 years or so, confidence intervals have become increasingly recognised in program evaluation and quantitative health measurement generally as the preferred way of reporting the accuracy of statistical estimates. Statisticians have found that the more traditional ways of reporting results - using P-values and hypothesis tests - are often very difficult to interpret and can be misleading. This is particularly the case when sample sizes are small and results are 'negative' (ie P>0.05); in these cases, a confidence interval can communicate much more information about the sample and, by inference, about the population. Despite this trend among statisticians and health promotion evaluators towards the use of confidence intervals, it is surprisingly difficult to find succinct and reasonably simple methods to actually compute a confidence interval. This is particularly the case for proportions or percentages. Much of the data which are analysed in health promotion are binary or categorical, rather than the quantities and continuous variables often found in laboratories or other branches of science, so there is a need for health promotion evaluators to be able to present confidence intervals for percentages or proportions. However, the most popular statistical analysis computer package among health promotion professionals, SPSS does not have a routine to compute a simple confidence interval for a proportion! To address this shortcoming, I present in this paper some fairly simple strategies for computing confidence intervals for population percentages, both manually and using the right computer software.

Sequential fixed-width confidence bands for kernel regression estimation

We consider a random design model based on independent and identically distributed (iid) pairs of observations (Xi, Yi), where the regression function m(x) is given by m(x) = E(Yi|Xi = x) with one independent variable. In a nonparametric setting the aim is to produce a reasonable approximation to the unknown function m(x) when we have no precise information about the form of the true density, f(x) of X. We describe an estimation procedure of non-parametric regression model at a given point by some appropriately constructed fixed-width (2d) confidence interval with the confidence coefficient of at least 1−. Here, d(> 0) and 2 (0, 1) are two preassigned values. Fixed-width confidence intervals are developed using both Nadaraya-Watson and local linear kernel estimators of nonparametric regression with data-driven bandwidths.

The sample size was optimized using the purely and two-stage sequential procedure together with asymptotic properties of the Nadaraya-Watson and local linear estimators. A large scale simulation study was performed to compare their coverage accuracy. The numerical results indicate that the confidence bands based on the local linear estimator have the best performance than those constructed by using Nadaraya-Watson estimator. However both estimators are shown to have asymptotically correct coverage properties.

Sample size determination for kernel regression estimation using sequential fixed-width confidence bands

We consider a random design model based on independent and identically distributed pairs of observations (Xi, Yi), where the regression function m(x) is given by m(x) = E(Yi|Xi = x) with one independent variable. In a nonparametric setting the aim is to produce a reasonable approximation to the unknown function m(x) when we have no precise information about the form of the true density, f(x) of X. We describe an estimation procedure of non-parametric regression model at a given point by some appropriately constructed fixed-width (2d) confidence interval with the confidence coefficient of at least 1−. Here, d(> 0) and 2 (0, 1) are two preassigned values. Fixed-width confidence intervals are developed using both Nadaraya-Watson and local linear kernel estimators of nonparametric regression with data-driven bandwidths. The sample size was optimized using the purely and two-stage sequential procedures together with asymptotic properties of the Nadaraya-Watson and local linear estimators. A large scale simulation study was performed to compare their coverage accuracy. The numerical results indicate that the confi dence bands based on the local linear estimator have the better performance than those constructed by using Nadaraya-Watson estimator. However both estimators are shown to have asymptotically correct coverage properties.

Non-asymptotic confidence bounds for the optimal value of a stochastic program

We discuss a general approach to building non-asymptotic confidence bounds for stochastic optimization problems. Our principal contribution is the observation that a Sample Average Approximation of a problem supplies upper and lower bounds for the optimal value of the problem which are essentially better than the quality of the corresponding optimal solutions. At the same time, such bounds are more reliable than “standard” confidence bounds obtained through the asymptotic approach. We also discuss bounding the optimal value of MinMax Stochastic Optimization and stochastically constrained problems. We conclude with a small simulation study illustrating the numerical behavior of the proposed bounds.

Electrocardiographic PR interval and adverse outcomes in older adults: the Health, Aging, and Body Composition study

BACKGROUND The electrocardiographic PR interval increases with aging, differs by race, and is associated with atrial fibrillation (AF), pacemaker implantation, and all-cause mortality. We sought to determine the associations between PR interval and heart failure, AF, and mortality in a biracial cohort of older adults. METHODS AND RESULTS The Health, Aging, and Body Composition (Health ABC) Study is a prospective, biracial cohort. We used multivariable Cox proportional hazards models to examine PR interval (hazard ratios expressed per SD increase) and 10-year risks of heart failure, AF, and all-cause mortality. Multivariable models included demographic, anthropometric, and clinical variables in addition to established cardiovascular risk factors. We examined 2722 Health ABC participants (aged 74±3 years, 51.9% women, and 41% black). We did not identify significant effect modification by race for the outcomes studied. After multivariable adjustment, every SD increase (29 ms) in PR interval was associated with a 13% greater 10-year risk of heart failure (95% confidence interval, 1.02-1.25) and a 13% increased risk of incident AF (95% confidence interval, 1.04-1.23). PR interval >200 ms was associated with a 46% increased risk of incident heart failure (95% confidence interval, 1.11-1.93). PR interval was not associated with increased all-cause mortality. CONCLUSIONS We identified significant relationships of PR interval to heart failure and AF in older adults. Our findings extend prior investigations by examining PR interval and associations with adverse outcomes in a biracial cohort of older men and women.