Search

A kappa-like classification statistic is proposed for assessing the fit of GEE regression models with a categorical response. The proposed statistic is a summary measure depicting how well categorical responses are predicted from the fitted GEE model. The statistic takes on a value of 1 if prediction is perfect and a value of 0 if the fitted model fares no better than random chance, i.e., fitting the repeated categorical responses with an intercept-only model. To demonstrate the usefulness of the classification statistic, we present simulation results as well as two examples from biomedical studies.

We develop a likelihood ratio test statistic, based on the beta- binomial distribution, for comparing a single treated group with dichoto- mous data to dual control groups. This statistic is useful in cases where there is overdispersion or extra-binomial variation. We apply the statis- tic to data from a two year rodent carcinogenicity study with dual control groups. The test statistic we developed is similar to others that have been developed for incorporation of historical control groups with rodent carcino- genicity experiments. However, for the small sample case we considered, large sample theory used by the other test statistics did not apply. We de- termined the critical values of this statistic by enumerating its distribution. A small Monte Carlo study shows the new test statistic controls the signifi- cance level much better than Fisher's exact test when there is overdispersion and that it has adequate power.

Summary Consider a high-dimensional linear regression problem, where the number of covariates is larger than the number of observations and the interest is in estimating the conditional variance of the response variable given the covariates. A conditional and an unconditional framework are considered, where conditioning is with respect to the covariates, which are ancillary to the parameter of interest. In recent papers, a consistent estimator was developed in the unconditional framework when the marginal distribution of the covariates is normal with known mean and variance. In the present work, a certain Bayesian hypothesis test is formulated under the conditional framework, and it is shown that the Bayes risk is a constant. This implies that no consistent estimator exists in the conditional framework. However, when the marginal distribution of the covariates is normal, the conditional error of the above consistent estimator converges to zero, with probability converging to one. It follows that even in the conditional setting, information about the marginal distribution of an ancillary statistic may have a significant impact on statistical inference. The practical implication in the context of high-dimensional regression models is that additional observations where only the covariates are given are potentially very useful and should not be ignored. This finding is most relevant to semi-supervised learning problems where covariate information is easy to obtain.

In comparing two treatments via a randomized clinical trial, the analysis of covariance (ANCOVA) technique is often utilized to estimate an overall treatment effect. The ANCOVA is generally perceiv...

For the analysis of clinical trials, the study participants are usually assumed to be representative sample of a target population. This assumption is rarely fulfilled in clinical trials, and particularly not if the sample size is small. In addition, covariate imbalances may affect the trial. Randomization tests provide a nonparametric analysis method of the treatment effect that does not rely on population-based assumptions. We propose a nonparametric statistical model that yields a formal basis for randomization tests. We adapt the model for the presence of covariate imbalance in the form of selection bias and investigate the effects of bias on the rejection probability of the randomization test using Monte Carlo simulations. Finally, we show that ancillary statistics can be used to control for the influence of bias. We show that covariate imbalance leads to an inflation of the type I error probability. The proposed nonparametric model allows for the use of ancillary statistics that yield an unbiased adjusted randomization test.

For a two-group comparative study, a stratified inference procedure is routinely used to estimate an overall group contrast to increase the precision of the simple two-sample estimator. Unfortunately, most commonly used methods including the Cochran-Mantel-Haenszel statistic for a binary outcome and the stratified Cox procedure for the event time endpoint do not serve this purpose well. In fact, these procedures may be worse than their two-sample counterparts even when the observed treatment allocations are imbalanced across strata. Various procedures beyond the conventional stratified methods have been proposed to increase the precision of estimation when the naive estimator is consistent. In this paper, we are interested in the case when the treatment allocation proportions vary markedly across strata. We study the stochastic properties of the two-sample naive estimator conditional on the ancillary statistics, the observed treatment allocation proportions and/or the stratum sizes, and present a biased-adjusted estimator. This adjusted estimator is asymptotically equivalent to the augmentation estimators proposed under the unconditional setting. Moreover, this consistent estimation procedure is also equivalent to a rather simple procedure, which estimates the mean response of each treatment group first via a stratum-size weighted average and then constructs the group contrast estimate. This simple procedure is flexible and readily applicable to any target patient population by choosing appropriate stratum weights. All the proposals are illustrated with the data from a cardiovascular clinical trial, whose treatment allocations are imbalanced.

The concept of pattern arises in many applications comprising experimental trials with two or more possible outcomes in each trial. A binary scan of type r / k is a special pattern referring to success–failure strings of fixed length k that contain at least r-successes, where r, k are positive integers with $$r\le k$$ . The multiple scan statistic $$W_{t,k,r}$$ is defined as the enumerating random variable for the overlapping moving windows occurring until trial t which include a scan of type r / k. In the present work, we consider a sequence of independent binary trials with not necessarily equal probabilities of success and develop upper bounds for the probability of the event that the multiple scan statistic will perform a jump from $$\ell $$ to $$\ell +1$$ (where $$\ell $$ is a nonnegative integer) in a finite time horizon.

The continuous time Markov process is commonly used to model state transitions in a multi-state system. However, it is hard to verify the Markov assumption by traditional goodness-of-fit techniques due to the multiple states. In this article, a goodness-of-fit test conditioning on a sufficient statistic is developed to test the Markov assumption. We propose a test statistic whose conditional distribution given the sufficient statistic is free of any unknown parameters. Therefore, the conditional distribution of the test statistic together with the corresponding percentiles could be obtained by Monte Carlo simulation. This conditional test outperforms two conventional goodness-of-fit techniques for the Markov assumption. Further, we consider a multi-state degradation system in which the operating conditions are available. An acceleration relation is often used to model the effects of the operating conditions on the degradation system. In this scenario, a goodness-of-fit test based on the Greenwood statistic is developed to test both the Markov assumption and the acceleration relation. The performance of the proposed goodness-of-fit tests in multi-state systems is demonstrated by extensive simulations and a numerical example.

Summary We offer an exposition of modern higher order likelihood inference and introduce software to implement this in a quite general setting. The aim is to make more accessible an important development in statistical theory and practice. The software, implemented in an R package, requires only that the user provide code to compute the likelihood function and to specify extra-likelihood aspects of the model, such as stopping rule or censoring model, through a function generating a dataset under the model. The exposition charts a narrow course through the developments, intending thereby to make these more widely accessible. It includes the likelihood ratio approximation to the distribution of the maximum likelihood estimator, that is the p∗ formula, and the transformation of this yielding a second-order approximation to the distribution of the signed likelihood ratio test statistic, based on a modified signed likelihood ratio statistic r∗. This follows developments of Barndorff-Nielsen and others. The software utilises the approximation to required Jacobians as developed by Skovgaard, which is included in the exposition. Several examples of using the software are provided.

Detection of multiple outliers or subset of influential points has been rarely considered in the linear measurement error models. In this paper a new influence statistic for one or a set of observations is generalized and characterized based on the corrected likelihood in the linear measurement error models. This influence statistic can be expressed in terms of the residuals and the leverages of linear measurement error regression. Unlike Cook’s statistic, this new measure of influence has asymptotically normal distribution and is able to detect a subset of high leverage outliers which is not identified by Cook’s statistic. As an illustrative example, simulation studies and a real data set are analysed.

As a submodel of generalized order statistics with two unknown model parameters, m-generalized order statistics may serve as a simple model for ordered quantities in a given application. It is shown that the joint distribution of m-generalized order statistics has a representation as a regular exponential family in the model parameters, as it is the case for the comprising model. Utilizing this finding, a minimal sufficient and complete statistic is obtained along with distributional properties. Joint maximum likelihood estimation of the parameters is considered, and strong consistency and asymptotic efficiency of the estimator are established. A test is provided to decide whether a restriction to the submodel is reasonable.

New statistics are proposed to estimate and test the structural change when the data dimension is comparable to or larger than the sample size. Consistency of the new statistic in estimating the change point position is established under the alternative hypothesis. The asymptotic distribution of the new statistic in testing the existence of a change point is obtained under the null hypothesis. Some simulation results are presented which show that the numerical performance of our method is satisfactory. The method is illustrated via the analysis of the house price index of US.

In this study, a simple diagnostic statistic for determining the size of random forest is proposed. This method is based on MV (margin of victory), a scaled difference in the votes at the infinite forest between the first and second most popular categories of the current random forest. We can note that if MV is negative then there is discrepancy between the current and infinite forests. More precisely, our method is based on the proportion of cases that -MV is greater than a fixed small positive number (say, 0.03). We derive an appropriate diagnostic statistic for our method and then calculate the distribution of the statistic. A simulation study is performed to compare our method with a recently proposed diagnostic statistic.

A test for ordered categorical variables is of considerable importance, because they are frequently encountered in biomedical studies. This paper introduces a simple ordering test approach for the two-way r × c contingency tables with incomplete counts by developing six test statistics, i.e., the likelihood ratio test statistic, score test statistic, global score test statistic, Hausman-Wald test statistic, Wald test statistic and distance-based test statistic. Bootstrap resampling methods are also presented. The performance of the proposed tests is evaluated with respect to their empirical type I error rates and empirical powers. The results show that the likelihood ratio test statistic based on the bootstrap resampling methods perform satisfactorily for small to large sample sizes. A real example from a wheeze study in six cities is used to illustrate the proposed methodologies.

In practice, it is often not possible to find an appropriate family of distributions which can be used for fitting the sample distribution with high precision. In these cases, it seems to be opportune to search for the best approximation by a family of distributions instead of an exact fit. In this paper, we consider the Anderson–Darling statistic with plugged-in minimum distance estimator for the parameter vector. We prove asymptotic normality of the Anderson–Darling statistic which is used for a test of goodness of approximation. Moreover, we introduce a measure of discrepancy between the sample distribution and the model class.

The statistician often needs to test whether or not a time series has a seasonal first moment. The problem often arises in environmental series, where most time-ordered data display some type of periodic structure. This paper reviews the problem, proposing new statistics in both the time and frequency domains. Our new time domain statistic has an analysis of variance form that is based on the one-step-ahead prediction errors of the series. This statistic inherits the classic traits of the F-distribution arising in one-way analysis of variance tests, is easy to use, and is asymptotically equivalent to the likelihood ratio test. The statistics asymptotic distribution is quantified when time series parameters are estimated. In the frequency domain, a statistic modifying Fisher's classical test for a sinusoidal mean superimposed on independent and identically distributed Gaussian noise is devised. The performance and comparison of these statistics are studied via simulation. Implementation of the methods merely requires sample means, autocovariances, and periodograms of the series. Application to a data set of monthly temperatures from Tuscaloosa, Alabama, is given. Copyright © 2016 John Wiley & Sons, Ltd.

A new method is proposed for identifying clusters in continuous data indexed by time or by space. The scan statistic we introduce is derived from the well-known Mann–Whitney statistic. It is completely non parametric as it relies only on the ranks of the marks. This scan test seems to be very powerful against any clustering alternative. These results have applications in various fields, such as the study of climate data or socioeconomic data.

This article reexamines ANOVA problems for autocorrelated data. Using linear prediction techniques for stationary time series, a new test statistic that assesses a null hypothesis of equal means is proposed and investigated. Our test statistic mimics the classical F-type ratio form used with independent data, but substitutes estimated prediction residuals in for the errors. This simple tactic departs from past studies that adjust the quadratic forms in the numerator and denominator in the F ratio for autocorrelation. One of the advantages is that our statistic retains the classical null hypothesis F distribution (now as a limit) with the customary degrees of freedom. The statistic is shown to perform well in simulations. Asymptotic proofs are given in the case of autoregressive random errors; a sports application is supplied.[Received December 2014. Revised August 2015.]

In this paper, we consider the consistency of Cp-type statistics for selecting variables in a normality-assumed linear regression with multiple responses when the dimension of the vector of the response variables may be large. We propose a new consistent Cp-type statistic for which consistency can be achieved whenever the dimension of the response variables vector is fixed or goes to infinity. A high probability of selecting the true subset of explanatory variables can be expected under a moderate sample size when the proposed Cp-type statistic is used to select variables, even when there is a high-dimensional response variables vector.

The C statistic is a commonly reported measure of screening test performance. Optimistic estimation of the C statistic is a frequent problem because of overfitting of statistical models in small data sets, and methods exist to correct for this issue. However, many studies do not use such methods, and those that do correct for optimism use diverse methods, some of which are known to be biased. We used clinical data sets (United Kingdom Down syndrome screening data from Glasgow (1991-2003), Edinburgh (1999-2003), and Cambridge (1990-2006), as well as Scottish national pregnancy discharge data (2004-2007)) to evaluate different approaches to adjustment for optimism. We found that sample splitting, cross-validation without replication, and leave-1-out cross-validation produced optimism-adjusted estimates of the C statistic that were biased and/or associated with greater absolute error than other available methods. Cross-validation with replication, bootstrapping, and a new method (leave-pair-out cross-validation) all generated unbiased optimism-adjusted estimates of the C statistic and had similar absolute errors in the clinical data set. Larger simulation studies confirmed that all 3 methods performed similarly with 10 or more events per variable, or when the C statistic was 0.9 or greater. However, with lower events per variable or lower C statistics, bootstrapping tended to be optimistic but with lower absolute and mean squared errors than both methods of cross-validation.

Barndorff-Nielsen’s celebrated p*-formula and variations thereof have amongst their various attractions the ability to approximate bimodal distributions. In this paper we show that in general this requires a crucial adjustment to the basic formula. The adjustment is based on a simple idea and straightforward to implement, yet delivers important improvements. It is based on recognizing that certain outcomes are theoretically impossible and the density of the MLE should then equal zero, rather than the positive density that a straight application of p* would suggest. This has implications for inference and we show how to use the new p**-formula to construct improved confidence regions. These can be disjoint as a consequence of the bimodality. The degree of bimodality depends heavily on the value of an approximate ancillary statistic and conditioning on the observed value of this statistic is therefore desirable. The p**-formula naturally delivers the relevant conditional distribution. We illustrate these results in small and large samples using a simple nonlinear regression model and errors in variables model where the measurement errors in dependent and explanatory variables are correlated and allow for weak proxies.

Now, we are ready to discuss the relationship between probability and statistical inference. The two key facts to statistical inference are (a) the population parameters are fixed numbers that are usually unknown and (b) sample statistics are known for any sample. For different samples, we get different values of the statistics and hence this variability is accounted for identifying distributions called sampling distributions. In this chapter, we discuss certain distributions that arise in sampling from normal distribution. The other topics covered in this chapter are as follows: unbiasedness, mean square error, consistency, relative efficiency, sufficiency, minimum variance, Fisher information for a function of a parameter, Cramer–Rao lower bound, efficiency, method of moments, maximum likelihood estimation to find estimators analytically and numerically, and asymptotic distributions of maximum likelihood estimators.

Tests of goodness of fit of sparse multinomial models with non-canonical links is proposed by using approximations to the first three moments of the conditional distribution of a modified Pearson Chi-square statistic. The modified Pearson statistic is obtained using a supplementary estimating equation approach. Approximations to the first three conditional moments of the modified Pearson statistic are derived. A simulation study is conducted to compare, in terms of empirical size and power, the usual Pearson Chi-square statistic, the standardized modified Pearson Chi-square statistic using the first two conditional moments, a method using Edgeworth approximation of the p-values based on the first three conditional moments and a score test statistic. There does not seems to be any qualitative difference in size of the four methods. However, the standardized modified Pearson Chi-square statistic and the Edgeworth approximation method of obtaining p-values using the first three conditional moments show power advantages compared to the usual Pearson Chi-square statistic, and the score test statistic. In some situations, for example, for small nominal level, the standardized modified Pearson Chi-square statistic shows some power advantage over the method using Edgeworth approximation of the p-values using the first three conditional moments. Also, the former is easier to use and so is preferable. Two data sets are analyzed and a discussion is given.

In this article, several scan statistics are discussed for detecting a local change in variance for one dimensional normal data. When the length of the scanning window is known, a fixed window scan statistic based on moving sum of squares is proposed. Two approximations for the distribution of this scan statistic are investigated. When the length of the scanning window is unknown, a variable window scan statistic based on a generalized likelihood ratio test and a multiple window minimum P-value scan statistic are proposed for detecting the local change in variance. For a moderate or large shift in variance, numerical results indicate that both the variable and multiple window scan statistics perform well. For large data sets, considering the detection power and computing efficiency, the multiple window scan statistic is recommended.

Pearson's Chi-square statistic for frequency tables depends on what is hypothesized as the expected frequencies. Its partitions also depend on the hypothesis. Lancaster (J R Stat Soc B 13:242---249, 1951) proposed ANOVA-like partitions of Pearson's statistic under several representative hypotheses about the expected frequencies. His expositions were, however, not entirely clear. In this paper, we clarify his method of derivations, and extend it to more general situations. A comparison is made with analogous decompositions of the log likelihood ratio statistic associated with log-linear analysis of contingency tables.

This paper proposes a new simple panel unit-root test by extending the cross-sectionally augmented panel unit-root test (CIPS) developed by Pesaran et al. (2013) to allow for smoothing structural changes in deterministic terms, approximated by a Fourier series. The proposed statistic is the simple average of the individual statistics constructed from the breaks and cross-sectional dependence augmented Dickey-Fuller (BCADF) regression and is called the BCIPS statistic. We initially develop the tests by assuming that the number of factors in the model is known and show that the limiting distribution of the BCADF statistic is free of nuisance parameters. The nonstandard limiting distribution of the (truncated) BCIPS statistic is also shown to exist and its critical values are tabulated. Monte-Carlo experiments point out that the sizes and powers of the BCIPS statistic are generally satisfactory as long as T is greater than or equal to fifty and a hundred, respectively. By using two different methods to determine the number of factors, both the BCIPS and CIPS tests are applied to examine the validity of long-run purchasing power parity. The proposed test complements the panel unit-root tests with breaks using dummy variables.

A new characterization of the Pareto distribution is proposed, and new goodness-of-fit tests based on it are constructed. Test statistics are functionals of U-empirical processes. The first of these statistics is of integral type, it is similar to the classical statistics $$\omega _n^1$$ . The second one is a Kolmogorov type statistic. We show that the kernels of our statistics are non-degenerate. The limiting distribution and large deviations asymptotics of the new statistics under null hypothesis are described. Their local Bahadur efficiency for parametric alternatives is calculated. This type of efficiency is mostly appropriate for the solution of our problem since the Kolmogorov type statistic is not asymptotically normal, and the Pitman approach is not applicable to this statistic. For the second statistic we evaluate the critical values by using Monte-Carlo methods. Also conditions of local optimality of new statistics in the sense of Bahadur are discussed and examples of such special alternatives are given. For small sample size we compare the power of those tests with some common goodness-of-fit tests.

As an important extension of the regular kappa statistic, the weighted kappa statistic has been widely used to assess the agreement between two procedures for independent matched-pair ordinal data. For clustered matched-pair ordinal data, based on the delta method and sampling techniques, a non-parametric variance estimator for the weighted kappa statistic is proposed without within-cluster correlation structure or distributional assumptions. The results of an extensive Monte Carlo simulation study demonstrate that the proposed weighted kappa statistic provides consistent estimation, and the proposed variance estimator behaves reasonably well for at least a moderately large number of clusters (e.g., K ≥ 50 ). Compared with the variance estimator ignoring dependence within a cluster, the proposed variance estimator performs better in maintaining the nominal coverage probability when the intra-cluster correlation is fair ( ρ ≥ 0.3 ), with more pronounced improvement when ρ is further increased. Moreover, under the general analysis of variance setting with systematic variability between procedures and clusters being included as a component of total variation, the equivalence between weighted kappa statistic and intraclass correlation coefficient is established. To illustrate the practical application of the proposed estimator, two real medical research data examples of clustered matched-pair ordinal data are analyzed, including an agreement study to compare two methods for assessing cervical ectopy, and a physician–patients data example from the Enhancing Communication and HIV Outcomes study.

In normal theory exploratory factor analysis, likelihood ratio (LR) statistic plays an important role in evaluating the goodness-of-fit of the model. In this paper, we derive an approximation of the LR statistic. The approximation is then used to show explicitly that the expectation of the LR statistic agrees with the degrees of freedom of the asymptotic chi-square distribution.

A U-statistic for the tail index of a multivariate stable random vector is given as an extension of the univariate case introduced by Fan (2006). Asymptotic normality and consistency of the proposed U-statistic for the tail index are proved theoretically. The proposed estimator is used to estimate the spectral measure. The performance of both introduced tail index and spectral measure estimators is compared with the known estimators by comprehensive simulations and real datasets.

Our new method uses order statistics to judge the fit of a distribution to data. A test-statistic based on quantiles of order-statistics compares favorably with the Kolmogorov-Smirnov (K-S) and Anderson-Darling (A-D) test statistics. The performance of this new goodness-of-fit test statistic is examined with simulation experiments. For certain hypothesis tests, the test statistic is more powerful than the K-S and A-D test statistics. The new test statistic is calculated using a computer algebra system because of the need to compute exact distributions of order statistics.