Accurate Coverage Probabilities Research Articles

Analyzing injury crashes using random-parameter bivariate regression models

ABSTRACTThis paper proposes a random-parameter bivariate zero-inflated negative binomial (RBZINB) regression model for analyzing the effects of investigated variables on crash frequencies. A Bayesian approach is employed as the estimation method, which has the strength of accounting for the uncertainties related to models and parameter values. The modeling framework has been applied to the bivariate injury crash counts obtained from 1000 intersections in Tennessee over a five-year period. The results reveal that the proposed RBZINB model outperforms other investigated models and provides a superior fit. The proposed RBZINB model is useful in gaining new insights into how crash occurrences are influenced by the risk factors. In addition, the empirical studies show that the proposed RBZINB model has a smaller prediction bias and variance, as well as more accurate coverage probability in estimating model parameters and crash-free probabilities.

Bootstrapping a Universal Pivot When Nuisance Parameters are Estimated

In complete samples from a continuous cumulative distribution with unknown parameters, it is known that various pivotal functions can be constructed by appealing to the probability integral transform. A pivotal function (or simply pivot) is a function of the data and parameters that has the property that its distribution is free of any unknown parameters. Pivotal functions play a key role in constructing confidence intervals and hypothesis tests. If there are nuisance parameters in addition to a parameter of interest, and consistent estimators of the nuisance parameters are available, then substituting them into the pivot can preserve the pivot property while altering the pivot distribution, or may instead create a function that is approximately a pivot in the sense that its asymptotic distribution is free of unknown parameters. In this latter case, bootstrapping has been shown to be an effective way of estimating its distribution accurately and constructing confidence intervals that have more accurate coverage probability in finite samples than those based on the asymptotic pivot distribution. In this article, one particular pivotal function based on the probability integral transform is considered when nuisance parameters are estimated, and the estimation of its distribution using parametric bootstrapping is examined. Applications to finding confidence intervals are emphasized. This material should be of interest to instructors of upper division and beginning graduate courses in mathematical statistics who wish to integrate bootstrapping into their lessons on interval estimation and the use of pivotal functions.[Received November 2014. Revised August 2015.]

Confidence intervals based on some weighting functions for the difference of two binomial proportions

In this paper, we propose two new methods for computing confidence intervals for the difference of two independent binomial proportions in small sample cases. Several test-based exact confidence intervals have been developed to guarantee the nominal coverage probability in small sample cases. However, these methods are sometimes unnecessarily too conservative because they use the exact p-value for constructing confidence intervals by maximizing the tail probability to account for the worst configuration. In order to reduce conservatism, our new methods adopt the p-value weighted by two types of functions instead of the maximum p-value. Our proposed methods can be regarded as quasi-exact methods. The performance evaluation results showed that our methods are much less conservative than the exact method. Compared with other existing quasi-exact methods, generally, our methods possess coverage probabilities closer to the nominal confidence level and shorter expected confidence widths. In particular, the beta weighing method provides the most reasonable balance between accurate coverage probability and short interval width in small sample cases.

Interval estimation for the mean of lognormal data with excess zeros

This paper considered interval estimations for the mean of lognormal distribution with excess zeros. We proposed two methods for interval estimation based on an approximate generalized pivotal quantity and a fiducial quantity. Simulation results show that the fiducial approach has highly accurate coverage probability and fairly low bias.

Confidence regions for repeated measures ANOVA power curves based on estimated covariance

BackgroundUsing covariance or mean estimates from previous data introduces randomness into each power value in a power curve. Creating confidence intervals about the power estimates improves study planning by allowing scientists to account for the uncertainty in the power estimates. Driving examples arise in many imaging applications.MethodsWe use both analytical and Monte Carlo simulation methods. Our analytical derivations apply to power for tests with the univariate approach to repeated measures (UNIREP). Approximate confidence intervals and regions for power based on an estimated covariance matrix and fixed means are described. Extensive simulations are used to examine the properties of the approximations.ResultsClosed-form expressions are given for approximate power and confidence intervals and regions. Monte Carlo simulations support the accuracy of the approximations for practical ranges of sample size, rank of the design matrix, error degrees of freedom, and the amount of deviation from sphericity. The new methods provide accurate coverage probabilities for all four UNIREP tests, even for small sample sizes. Accuracy is higher for higher power values than for lower power values, making the methods especially useful in practical research conditions. The new techniques allow the plotting of power confidence regions around an estimated power curve, an approach that has been well received by researchers. Free software makes the new methods readily available.ConclusionsThe new techniques allow a convenient way to account for the uncertainty of using an estimated covariance matrix in choosing a sample size for a repeated measures ANOVA design. Medical imaging and many other types of healthcare research often use repeated measures ANOVA.

Finite-sample properties of the adjusted empirical likelihood

Empirical likelihood-based confidence intervals for the population mean have many interesting properties [Owen, A.B. (1988), ‘Empirical Likelihood Ratio Confidence Intervals for a Single Functional’, Biometrika, 75, 237–249]. Calibrated by χ2 limiting distribution, however, their coverage probabilities are often lower than the nominal when the sample size is small and/or the dimension of the data is high. The application of adjusted empirical likelihood (AEL) is one of the many ways to achieve a more accurate coverage probability. In this paper, we study the finite-sample properties of the AEL. We find that the AEL ratio function decreases when the level of adjustment increases. Thus, the AEL confidence region has higher coverage probabilities when the level of adjustment increases. We also prove that the AEL ratio function increases when the putative population mean moves away from the sample mean. In addition, we show that the AEL confidence region for the population mean is convex. Finally, computer simulations are conducted to further investigate the precision of the coverage probabilities and the sizes of the confidence regions. An application example is also included.

Empirical likelihood inferences for the semiparametric additive isotonic regression

We consider the (profile) empirical likelihood inferences for the regression parameter (and its any sub-component) in the semiparametric additive isotonic regression model where each additive nonparametric component is assumed to be a monotone function. In theory, we show that the empirical log-likelihood ratio for the regression parameters weakly converges to a standard chi-squared distribution. In addition, our simulation studies demonstrate the empirical advantages of the proposed empirical likelihood method over the normal approximation method in Cheng (2009) [4] in terms of more accurate coverage probability when the sample size is small. It is worthy pointing out that we can construct the empirical likelihood based confidence region without the hassle of tuning any smoothing parameter due to the shape constraints assumed in this paper.

Bayesian semiparametric regression models for evaluating pathway effects on continuous and binary clinical outcomes.

Many statistical methods for microarray data analysis consider one gene at a time, and they may miss subtle changes at the single gene level. This limitation may be overcome by considering a set of genes simultaneously where the gene sets are derived from prior biological knowledge. Limited work has been carried out in the regression setting to study the effects of clinical covariates and expression levels of genes in a pathway either on a continuous or on a binary clinical outcome. Hence, we propose a Bayesian approach for identifying pathways related to both types of outcomes. We compare our Bayesian approaches with a likelihood-based approach that was developed by relating a least squares kernel machine for nonparametric pathway effect with a restricted maximum likelihood for variance components. Unlike the likelihood-based approach, the Bayesian approach allows us to directly estimate all parameters and pathway effects. It can incorporate prior knowledge into Bayesian hierarchical model formulation and makes inference by using the posterior samples without asymptotic theory. We consider several kernels (Gaussian, polynomial, and neural network kernels) to characterize gene expression effects in a pathway on clinical outcomes. Our simulation results suggest that the Bayesian approach has more accurate coverage probability than the likelihood-based approach, and this is especially so when the sample size is small compared with the number of genes being studied in a pathway. We demonstrate the usefulness of our approaches through its applications to a type II diabetes mellitus data set. Our approaches can also be applied to other settings where a large number of strongly correlated predictors are present.

Applying Bayesian Model Averaging for Quantile Estimation in Accelerated Life Tests

In an accelerated life test, inferences on extreme quantiles of the lifetime distribution at the use condition are obtained via extrapolation in two directions: in time, and in stress levels. Extrapolation is known to highly depend on the working model, and ignoring model uncertainty can result in over-confidence. This paper explores the use of Bayesian model averaging for estimating quantiles in an accelerated life test. Two of the most commonly used lifetime regression models, lognormal, and Weibull log-location-scale regression models, are considered in this paper as candidate models. To illustrate, we analyse complementary metal-oxide semiconductor integrated circuit data. We also construct a simulation study to compare the performance of the Bayesian model averaging -credibility intervals with other exiting interval estimators. The simulation study shows that, for estimating extreme quantiles, both the standard Bayesian and the maximum likelihood approaches can lead to an over-confident result, and Bayesian model averaging provides a -credibility interval with a wider average length, but a more accurate coverage probability.

Confidence interval of the jump activity index based on empirical likelihood using high frequency data

It is widely accepted that jumps exist in the asset price process. The jump activity index is a natural measure of how frequent the jumps are. Statistical inference of the jump activity index is of importance in determining the type of process that underlies the dynamics of the log price process. In this paper, we implement the empirical likelihood approach to construct the confidence interval of the jump activity index of a pure jump model using high frequency data. Wilks' theorem is established. We also extend the result on Zhao and Wu (2009)'s estimator to the more general framework in this paper. Simulation studies demonstrate the good performance of the empirical likelihood approach. Compared with the existing non-parametric estimator proposed by Zhao and Wu (2009), the empirical likelihood approach gives more accurate coverage probabilities in the simulation studies.

Estimation and testing of the relative risk of disease in case-control studies with a set of k matched controls per case with known prevalence of disease.

The analysis of case-control studies with matched controls per case is well documented in the medical literature. Of primary interest is the estimation of the relative risk of disease. Matched case-control studies fall into two scenarios: the probability of exposure is constant within each of the case and control groups, or the probability of exposure varies within each group. Numerous estimation procedures have been developed for both scenarios. Often these procedures are developed under the rare disease assumption, where the relative risk of disease is approximated by the odds ratio. In this paper, without making the rare disease assumption, we develop consistent estimators of the relative risk of disease for both scenarios. Exact derivations of the relative risk of disease are provided. Estimators, confidence intervals, and test statistics for the relative risk of disease are developed. We then make the following observations based on extensive simulations. First, our estimators are as close or closer to the relative risk of disease than other estimators. Second, our estimators produce mean square errors for the relative risk of disease that are as good as or better than these other estimators. Third, our confidence intervals provide accurate coverage probabilities. Therefore, these new estimators, confidence intervals, and test statistics can be used to either estimate or test the relative risk of disease in matched case-control studies.

Strong approximations for resample quantile processes and application to ROC methodology

The receiver operating characteristic (ROC) curve is defined as true positive rate versus false positive rate obtained by varying a decision threshold criterion. It has been widely used in medical sciences for its ability to measure the accuracy of diagnostic or prognostic tests. Mathematically speaking, ROC curve is the composition of survival function of one population with the quantile function of another population. In this paper, we study strong approximation for the quantile processes of the Bayesian bootstrap (BB) resampling distributions, and use this result to study strong approximations for the BB version of the ROC process in terms of two independent Kiefer processes. The results imply asymptotically accurate coverage probabilities for the confidence bands for the ROC curve and confidence intervals for the area under the curve functional of the ROC constructed using the BB method. Similar results follow for the bootstrap resampling distribution.

On Lower Tolerance Limits With Accurate Coverage Probabilities for the Normal Random Effects Model

A method for constructing accurate lower tolerance limits for the balanced one-way normal random-effects model is derived by conditioning on an estimator of the unknown expected mean square ratio. Simulation studies indicate that the present procedure is less conservative than that of several existing methods and gives more accurate coverage rates and smaller standard deviations. Numerical examples are given to illustrate the use of the new procedure. Tables needed to implement the procedure are also included.

Confidence intervals for the amount of heterogeneity in meta‐analysis

Effect size estimates to be combined in a systematic review are often found to be more variable than one would expect based on sampling differences alone. This is usually interpreted as evidence that the effect sizes are heterogeneous. A random-effects model is then often used to account for the heterogeneity in the effect sizes. A novel method for constructing confidence intervals for the amount of heterogeneity in the effect sizes is proposed that guarantees nominal coverage probabilities even in small samples when model assumptions are satisfied. A variety of existing approaches for constructing such confidence intervals are summarized and the various methods are applied to an example to illustrate their use. A simulation study reveals that the newly proposed method yields the most accurate coverage probabilities under conditions more analogous to practice, where assumptions about normally distributed effect size estimates and known sampling variances only hold asymptotically.

Bootstrap standard error and confidence intervals for the correlation corrected for range restriction: a simulation study.

The standard Pearson correlation coefficient is a biased estimator of the true population correlation, rho, when the predictor and the criterion are range restricted. To correct the bias, the correlation corrected for range restriction, rc, has been recommended, and a standard formula based on asymptotic results for estimating its standard error is also available. In the present study, the bootstrap standard-error estimate is proposed as an alternative. Monte Carlo simulation studies involving both normal and nonnormal data were conducted to examine the empirical performance of the proposed procedure under different levels of rho, selection ratio, sample size, and truncation types. Results indicated that, with normal data, the bootstrap standard-error estimate is more accurate than the traditional estimate, particularly with small sample size. With nonnormal data, performance of both estimates depends critically on the distribution type. Furthermore, the bootstrap bias-corrected and accelerated interval consistently provided the most accurate coverage probability for rho.

Weighted empirical likelihood inference

A weighted empirical likelihood approach is proposed to take account of the heteroscedastic structure of the data. The resulting weighted empirical likelihood ratio statistic is shown to have a limiting chisquare distribution. A limited simulation study shows that the associated confidence intervals for a population mean or a regression coefficient have more accurate coverage probabilities and more balanced two-sided tail errors when the sample size is small or moderate. The proposed weighted empirical likelihood method also provides more efficient point estimators for a population mean in the presence of side information. Large sample resemblances between the weighted and the unweighted empirical likelihood estimators are characterized through high-order asymptotics and small sample discrepancies of these estimators are investigated through simulation. The proposed weighted approach reduces to the usual unweighted empirical likelihood method under a homogeneous variance structure.

Bootstrap confidence intervals for relative risk parameters in affected-sib-pair data.

In affected-sib-pair (ASP) studies, parameters such as the locus-specific sibling relative risk, lambda(s), may be estimated and used to decide whether or not to continue the search for susceptibility genes. Typically, a maximum likelihood point estimate of lambda(s) is given, but since this estimate may have substantial variability, it is of interest to obtain confidence limits for the true value of lambda(s). While a variety of methods for doing this exist, there is considerable uncertainty over their reliability. This is because the discrete nature of ASP data and the imposition of genetic "possible triangle" constraints during the likelihood maximization mean that asymptotic results may not apply. In this paper, we use simulation to evaluate the reliability of various asymptotic and simulation-based confidence intervals, the latter being based on a resampling, or bootstrap approach. We seek to identify, from the large pool of methods available, those methods that yield short intervals with accurate coverage probabilities for ASP data. Our results show that many of the most popular bootstrap confidence interval methods perform poorly for ASP data, giving coverage probabilities much lower than claimed. The test-inversion, profile-likelihood, and asymptotic methods, however, perform well, although some care is needed in choice of nuisance parameter. Overall, in simulations under a variety of different genetic hypotheses, we find that the asymptotic methods of confidence interval evaluation are the most reliable, even in small samples. We illustrate our results with a practical application to a real data set, obtaining confidence intervals for the sibling relative risks associated with several loci involved in type 1 diabetes.

Multivariate Zero-Inflated Poisson Models and Their Applications

The zero-inflated Poisson (ZIP) distribution has been shown to be useful for modeling outcomes of manufacturing processes producing numerous defect-free products. When there are several types of defects, the multivariate ZIP (MZIP) model can be useful to detect specific process equipment problems and to reduce multiple types of defects simultaneously. This article proposes types of MZIP models and investigates distributional properties of an MZIP model. Finite-sample simulation studies show that, compared to the method of moments, the maximum likelihood method has smaller bias and variance, as well as more accurate coverage probability in estimating model parameters and zero-defect probability. Real-life examples from a major electronic equipment manufacturer illustrate how the proposed procedures are useful in a manufacturing environment for equipment-fault detection and for covariate effect studies.

Multivariate Zero-Inflated Poisson Models and Their Applications

The zero-inflated Poisson (ZIP) distribution has been shown to be useful for modeling outcomes of manufacturing processes producing numerous defect-free products. When there are several types of defects, the multivariate ZIP (MZIP) model can be useful to detect specific process equipment problems and to reduce multiple types of defects simultaneously. This article proposes types of MZIP models and investigates distributional properties of an MZIP model. Finite-sample simulation studies show that, compared to the method of moments, the maximum likelihood method has smaller bias and variance, as well as more accurate coverage probability in estimating model parameters and zero-defect probability. Real-life examples from a major electronic equipment manufacturer illustrate how the proposed procedures are useful in a manufacturing environment for equipment-fault detection and for covariate effect studies.

Likelihood‐Weighted Confidence Intervals for the Difference of Two Binomial Proportions

AbstractTwo new methods for computing confidence intervals for the difference δ = p1 — p2 between two binomial proportions (p1, p2) are proposed. Both the Mid‐P and Max‐P likelihood weighted intervals are constructed by mapping the tail probabilities from the two‐dimensional (p1, p2)‐space into a one‐dimensional function of δ based on the likelihood weights. This procedure may be regarded as a natural extension of the CLOPPER‐PEARSON (1934) interval to the two‐sample case where the weighted tail probability is α/2 at each end on the δ scale. The probability computation is based on the exact distribution rather than a large sample approximation. Extensive computation was carried out to evaluate the coverage probability and expected width of the likelihood‐weighted intervals, and of several other methods. The likelihood‐weighted intervals compare very favorably with the standard asymptotic interval and with intervals proposed by HAUCK and ANDERSON (1986), COX and SNELL (1989), SANTNER and SNELL (1980), SANTNER and YAMAGAMI (1993), and PESKUN (1993). In particular, the Mid‐P likelihood‐weighted interval provides a good balance between accurate coverage probability and short interval width in both small and large samples. The Mid‐P interval is also comparable to COE and TAMHANE'S (1993) interval, which has the best performance in small samples.