Percentage points of the statistics for testing hypotheses on mean vectors of multivariate normal distributions with missing observations

A problem of testing of hypotheses on the mean vector of a multivariate normal distribution with unknown and positive definite covariance matrix is considered when a sample with a special, though not unusual, pattern of missing observations from that population is available. The approximate percentage points of the test statistic are obtained and their accuracy has been checked by comparing them with some exact percentage points which are calculated for complete samples and some special incomplete samples. The approximate percentage points are in good agreement with exact percentage points. The above work is extended to the problem of testing the hypothesis of equality of two mean vectors of two multivariate normal distributions with the same, unknown covariance matrix     

On the Optimality and Limitations of Buehler Bounds

In 1957, R.J. Buehler gave a method of constructing honest upper confidence limits for a parameter that are as small as possible subject to a pre‐specified ordering restriction. In reliability theory, these ‘Buehler bounds’ play a central role in setting upper confidence limits for failure probabilities. Despite their stated strong optimality property, Buehler bounds remain virtually unknown to the wider statistical audience. This paper has two purposes. First, it points out that Buehler's construction is not well defined in general. However, a slightly modified version of the Buehler construction is minimal in a slightly weaker, but still compelling, sense. A proof is presented of the optimality of this modified Buehler construction under minimal regularity conditions. Second, the paper demonstrates that Buehler bounds can be expressed as the supremum of Buehler bounds conditional on any nuisance parameters, under very weak assumptions. This result is then used to demonstrate that Buehler bounds reduce to a trivial construction for the location‐scale model. This places important practical limits on the application of Buehler bounds and explains why they are not as well known as they deserve to be.     

A modified Kolmogorov-Smirnov test for the inverse gaussian density with unknown parameters

The Kolmogorov-Smirnov (KS) test is an empirical distribution function (EDF) based goodness-of-fit test that requires the underlying hypothesized density to be continuous and completely specified. When the parameters are unknown and must be estimated from the data, standard tables of the KS test statistic are not valid. Approximate upper tail percentage points of the KS statistic for the inverse Gaussian (IG) distribution with unknown parameters are tabled in this paper.

A study of the power of the KS test for the IG distribution indicates that the test is able todiscriminate between the IG distribution and distributions such as the uniform and exponentialdistributions that are very different in shape, but is relatively unable to discriminate between the IG distribution and distributions that are similar in shape such as the lognormal and Weibull distributions. In modeling settings the former distinction is typically more important to make than the latter distinction.     

Empirical bayes estimation of reliability characteristics for an exponential family

This paper considers empirical Bayes (EB) squared-error-loss estimations of mean lifetime, variance and reliability function for failure-time distributions belonging to an exponential family, which includes gamma and Weibull distributions as special cases. EB estimators are proposed when the prior distribution of the lifetime parameter is completely unknown but has a compact (known or unknown) support. Asymptotic optimality and rates of convergence of these estimators are investigated. The rates established here under the compact support restriction are better than the polynomial rates of convergence obtained previously.     

Prior distributions expressing ignorance about convex increasing failure rates

This paper deals with the specification of probability distributions expressing ignorance concerning annual or otherwise discretized failure or mortality rates, when these rates can safely be assumed to be increasing and convex, but are completely unknown otherwise. Such distributions can be used as noninformative priors for Bayesian analysis of failure data. We demonstrate why a uniform distribution used in earlier work is unsatisfactory, especially from the point of view of insensitivity with respect to the time scale that is chosen for the problem at hand. We suggest alternative distributions based on Dirichlet distributed weights for the extreme points of relevant convex sets, and discuss which consequences a requirement for scale neutrality has for the choice of Dirichlet parameters.     

Regular mles for nonregular distributions

Distributions whose extremity values of the support depend on unknown pa¬rameters are usually known as nonregular distributions. In most cases, the MLEs for these parameters cannot be obtained by differentiation. Familiar examples are the uniform distribution on the interval (0,0) and the truncated exponential distribution with truncation parameter 0. However, there exist distributions whose extremity points of the support depend on unknown pa¬rameters, which nevertheless are regular in the sense that the MLEs can be obtained by differentiation. This note provides a method of constructing such nonregular distributions with regular MLEs.     

Alternate test statistics for a functional errors-in-variables model

Some test statistics for the structural coefficients of simultaneous equations model often referred to as the multivariate linear functional relationship model are proposed in this article. The following cases are considered: the covariance matrix of errors is either unknown, known up to a proportionality factor, or completely known. The exact and approximate distributions of the proposed test statistics, as well as those of some that are known, are also given.     

Nonparametric maximum likelihood estimators for ar and ma time series

The problem of estimating the parameters of moving average or autoregressive time series is studied when the error distribution is completely unknown. Four nonparametric maximum likelihood estimators (NPMLE) are presented for this purpose. These estimators are compared with the classical moment and least squares estimators in a simulation study. The behavior of these NPMLEs is much better than the classical ones, suggesting that they should be used extensively when no parametric information is known in advance about the error distribution. An application of these estimators to coal mining accidents data is also included.     

Modified S-charts for controlling process variability

To help in the detection of variance increases and decreases, three modified versions of traditional Shewhart S-charts are evaluated in terms of their average run length values. One scheme uses control limits based on equal tail chi-square distribution probabilities. The second uses control limits based on unequal tail probabilities. The third uses warning limits based on equal tail probabilities, but requires two successive points beyond the warning limit to give an out-of-control signal. They all result in better average run length values than the traditional S-chart. Also, if the only concern is the detection of variance increases, then both S-charts and warning limit charts without lower control limits are shown to have better average run length values than those of the traditional charts.     

Testing for heteroscedasticity occuring at unknown points

This paper considers procedures for the detection of heteroscedasticity when it occurs at unknown points. First the number of points is assumed known apriori and then this assumption is relaxed so that both the position and the number of possible aberrant observations is unknown. Monte Carlo evidence is provided on the performance of both tests and they are found to perform reasonably.     

Improved simultaneous confidence intervals for linear contrasts and regression parameters

A well known method for obtaining conservative simultaneous confidence intervals for the K parameters in a linear regression model, or for K linear contrasts, is based on the percentage points of the Studentized maximum modulus distribution. From an inequality due to Sidak, conservative yet uniformly shorter confidence intervals would be possible if the percentage points of a particular form of the multivariate t distribution were available. The purpose of this paper is to provide the required percentage points. For K<8 the resulting confidence intervals can be substantially shorter.     

Non-parametric Estimation for NHPP Software Reliability Models

The non-homogeneous Poisson process (NHPP) model is a very important class of software reliability models and is widely used in software reliability engineering. NHPPs are characterized by their intensity functions. In the literature it is usually assumed that the functional forms of the intensity functions are known and only some parameters in intensity functions are unknown. The parametric statistical methods can then be applied to estimate or to test the unknown reliability models. However, in realistic situations it is often the case that the functional form of the failure intensity is not very well known or is completely unknown. In this case we have to use functional (non-parametric) estimation methods. The non-parametric techniques do not require any preliminary assumption on the software models and then can reduce the parameter modeling bias. The existing non-parametric methods in the statistical methods are usually not applicable to software reliability data. In this paper we construct some non-parametric methods to estimate the failure intensity function of the NHPP model, taking the particularities of the software failure data into consideration.     

A New Procedure to Monitor the Mean of a Quality Characteristic

The Shewhart, Bonferroni-adjustment, and analysis of means (ANOM) control charts are typically applied to monitor the mean of a quality characteristic. The Shewhart and Bonferroni procedure are utilized to recognize special causes in production process, where the control limits are constructed by assuming normal distribution for known parameters (mean and standard deviation), and approximately normal distribution regarding to unknown parameters. The ANOM method is an alternative to the analysis of variance method. It can be used to establish the mean control charts by applying equicorrelated multivariate non central t distribution. In this article, we establish new control charts, in phases I and II monitoring, based on normal and t distributions having as a cause a known (or unknown) parameter (standard deviation). Our proposed methods are at least as effective as the classical Shewhart methods and have some advantages.     

A note on mse coverage intervals in a partial spline model

A partial spline model is used to estimate an unknown function which is smooth except for some break points. Assuming the break points are known, a Generalized Cross-Validated smoothing spline estimation method is proposed. Some interval estimation methods for the magnitude of the discontinuities based on the mean square error are introduced and investigated.     

A Continuous Representation of the Family of Stable Law Distributions

Conventional parametric representations of stable law distributions do not allow all members of the family to be obtained as continuous limits of the parameters. Model building (or simulation) using such representations will be numerically unstable near such limits in consequence. Existing tables are not satisfactory near such limits as interpolation cannot be carried out. We show that these difficulties are overcome by using a new shifted Cartesian representation which characterizes the entire stable law family in a completely continuous way. Standardization is still possible with this representation so that tabulation, using just two bounded parameters, can be carried out. Its use is illustrated in a non-regular threshold estimation problem involving stable distributions which are discontinuous limits in conventional representations.     

Revisiting the Berger location model: Fallacious confidence interval or a rigged example?

Since the 1960s the Bayesian case against frequentist inference has been partly built on several “classic” examples which are devised to show how frequentist inference procedures can give rise to fallacious results; see Berger and Wolpert (1988) [2]. The primary aim of this note is to revisit one of these examples, the Berger location model, that is supposed to demonstrate the fallaciousness of frequentist Confidence Interval (CI) estimation. A closer look at the example, however, reveals that the fallacious results stem primarily from the problematic nature of the example itself, since it is based on a non-regular probability model that enables one to (indirectly) assign probabilities to the unknown parameter. Moreover, the proposed confidence set is not a proper frequentist CI in the sense that it is not defined in terms of legitimate error probabilities.     

The Anderson–Darling Goodness-of-Fit Test Statistic for the Three-Parameter Lognormal Distribution

Anderson–Darling goodness-of-fit test percentage points are given for the three-parameter lognormal distribution for both the cases of positive skewness and a lower bound and negative skewness and an upper bound. The focus is on the most practical case when all parameters are unknown and must be estimated from the sample data. Fitted response functions for the critical values based on the shape parameter and sample size are reported to avoid using a vast array of tables.     

The delta-corrected Kolmogorov-Smirnov test for the two-parameter Weibull distribution

SUMMARY Monte Carlo simulation techniques are used to create tables of critical values for the delta-corrected Kolmogorov-Smirnov statistic-a modification of the classical Kolmogorov-Smirnov statistic-for the Weibull distribution with known location parameter and unknown shape and scale parameters. The power of the proposed test is investigated relative to values of delta in the unit interval and relative to a wide variety of alternative distributions. The results indicate that using the delta-correction can lead to as many as 8.4 percentage points more power than can be achieved with the classical Kolmogorov-Smirnov test, with no change in the size of the test. Furthermore, carrying out the delta-corrected test involves no more steps or calculations than for the classical Kolmogorov-Smirnov test. In general, it is shown that a slight modification-or correction-in the definition of the empirical distribution function of the Kolmogorov-Smirnov test can lead to power enhancement without changing the type I error rate of the test. Two examples clearly show the effectiveness of the delta-corrected test. The delta-corrected Kolmogorov-Smirnov test is recommended for testing the goodness of fit to the twoparameter Weibull distribution.     

A Note on the Use of Confidence Limits following Rejection of a Null Hypothesis

The equivalence of some tests of hypothesis and confidence limits is well known. When, however, the confidence limits are computed only after rejection of a null hypothesis, the usual unconditional confidence limits are no longer valid. This refers to a strict two-stage inference procedure: first test the hypothesis of interest and if the test results in a rejection decision, then proceed with estimating the relevant parameter. Under such a situation, confidence limits should be computed conditionally on the specified outcome of the test under which estimation proceeds. Conditional confidence sets will be longer than unconditional confidence sets and may even contain values of the parameter previously rejected by the test of hypothesis. Conditional confidence limits for the mean of a normal population with known variance are used to illustrate these results. In many applications, these results indicate that conditional estimation is probably not good practice.     

WEEKDAY-ONLY POLLING AND PARTISAN SUPPORT LEVELS: EVIDENCE FROM NEW ZEALAND

Some scholars have argued that opinion polls conducted only on weekday evenings are likely to be biased in favour of right-wing parties, but there is no scholarly consensus on this. This paper presents evidence from New Zealand suggesting that weekday-only polling consistently increases the reported support levels for right-wing parties. Polls conducted only on weekdays in New Zealand give estimated support levels for the major right-leaning party approximately five percentage points higher than do concurrent polls that survey voters both on weekdays and during the weekend. Results for left-leaning parties are also broadly consistent with this effect.