摘要:

This article presents a method for analyzing repeated measures with categorical responses based on maximizing a conditional likelihood. Many existing methods for analyzing repeated categorical responses, such as those proposed by Duncan (1980, 1985), Bishop, Fienberg, and Holland (1975), and Cochran (1950) are special cases of the method based on the conditional likelihood approach. This article explores the relationships among the various methods and extends several of the existing methods. The conditional likelihood approach is used to analyze data from a four-wave panel study with trichotomous responses. The data, taken from Fienberg, Bromet, Follmann, Lambert, and May (1985), are from an investigation of the psychological effects of the accident at the Three Mile Island nuclear power plant in the spring of 1979. In this example, methods are presented for checking the assumption of local independence in the model used for the conditional likelihood analysis. With an extension of the results of Tjur (1982), it is demonstrated that the conditional estimates can be obtained with standard statistical software.

ISSN:0162-1459    

DOI:10.1080/01621459.1989.10478738

出版商:Taylor & Francis Group    

年代:1989

数据来源: Taylor

摘要:

Using current and retrospective school-enrollment data from October Current Population Surveys (CPS's) together with demographic accounts for the U.S. civilian noninstitutional population, this study specifies new methods for the estimation of school-life tables. The new methods combine multistate (increment—decrement) methods for those ages for which CPS data on enrollment flows are available together with prevalence-rate methods for other ages for which CPS data are available on enrollment stocks alone. Applying an algorithm to construct tables of school life for the 1969–1970, 1974–1975, 1979–1980, and 1984–1985 school years, it is found that, between the ages of 14 and 34, most sex-, race-, and school-year-specific groups have an average of 1.5 to 2.0 exits per person, from enrolled to not enrolled in school. These exits are interspersed by an average of .5 to 1.0 exits per person, from not enrolled to enrolled in school. Since prevalence-rate life-table methods arbitrarily constrain individuals to only one exit from school enrollment to not enrolled over the life course, this finding verifies the need for increment—decrement methods for the teenage and young-adult years. Several other findings regarding differentials in schooling life by sex and race and trends over time are discussed.

ISSN:0162-1459    

DOI:10.1080/01621459.1989.10478739

出版商:Taylor & Francis Group    

年代:1989

数据来源: Taylor

摘要:

In theory, a flexible functional form can approximate an arbitrary elasticity of substitution; however, when faced with an actual finite data set, a parsimonious parametric functional form that imposes more structural restrictions may yield a more accurate approximation about underlying substitution. Therefore, for uncovering the underlying degree of input substitution a data-exploration strategy is suggested that combines the parametric family of a system of quadratic log-ratio demand equations and conventional flexible functional forms. This allows testing for more restrictive hypotheses about substitution. These issues are examined in the context of interfuel substitution at the two-digit SIC level for manufacturing industries in Ontario and Quebec, Canada. A basic tenet of the approach of this article is that although letting data speak for itself is preferred, economic theory should be used wherever possible, and a simple model is preferable to complicated models unless the evidence suggests otherwise. Given the variations of technologies, managerial behavior, and speed of adjustment across manufacturing sectors, the existence of a single optimal model is questioned. In fact, out of 39 industries examined, suitable models are found for 30 industries. Out of the 30, 18 can be modeled by the relatively more restricted quadratic log-ratio specifications and 12 by the more general flexible functional forms. By using a combined structural and flexible functional approach, this article obtains more precise estimates of the elasticities of substitution and models more likely to satisfy monotonicity and concavity conditions than those obtained by considering the conventional search procedures within a flexible functional approach. It is also interesting to note that, unlike previous studies of interfuel substitution, some industries display biased technological changes, and some appear to possess a dynamic adjustment process consistent with an error-correction mechanism composed of both derivative and proportional control mechanisms.

ISSN:0162-1459    

DOI:10.1080/01621459.1989.10478740

出版商:Taylor & Francis Group    

年代:1989

数据来源: Taylor

摘要:

This article revises and updates through 1984 an annual and quarterly series originally constructed from 1952–1968 on the average workweek of capital in manufacturing. The series provides estimates of the input of capital services into production and uses data on production-worker employment, shift patterns, and weekly working hours to estimate the number of hours that the capital stock is operated each week. The updated series indicates that the capital stock is working longer and the hours of operation are more cyclically variable than in the earlier period. Shift work is found to be an important source of variation in the hours of operation. The findings imply a change in the relationship of the capital stock to manufacturing production.

ISSN:0162-1459    

DOI:10.1080/01621459.1989.10478741

出版商:Taylor & Francis Group    

年代:1989

数据来源: Taylor

摘要:

We first discuss the construction of bootstrap confidence bands for the distribution functionFof a population for simple random sampling but do not assume thatFis continuous. The known alternative approach is to use the quantiles from the tabled Kolmogorov distribution. This approach is known to be conservative. It has already been shown in Bickel and Freedman (1981) that the bootstrap confidence band has the correct coverage probability asymptotically. We show by simulation that the bootstrap works well for small samples and outperforms the conservative approach, particularly for distributions that have small carriers. We also investigate the analogous problem of finding bootstrap confidence bands for the distribution functionFof a population for a more complicated situation, stratified random sampling. The conservative approach in the previous situation is extended to this case when sampling is with replacement (we expect that it holds for sampling without replacement) and the supports of the conditional distributions in each stratum are not overlapping. If the strata overlap there seems to be little alternative to the bootstrap. Use of the bootstrap in setting confidence bands or curves in this way should prove widely applicable, particularly when we leave the simple random sampling context as we have done. Asymptotic theory for the bootstrap confidence band is presented, and the conservative and bootstrap approaches are compared for small samples by simulation.

ISSN:0162-1459    

DOI:10.1080/01621459.1989.10478742

出版商:Taylor & Francis Group    

年代:1989

数据来源: Taylor

摘要:

In choosing a model, there are important considerations beyond the goodness of fit, namely that the subsequent inferences should not rely on questionable assumptions implicit in the model, because otherwise they will be suspect. The two-parameter lognormal distribution provides an example. The symmetry on the log scale leads to the anomaly that relative frequencies of classes in the lower tail can substantially affect inferences on the upper tail. This is contrary to the realities of the alcohol consumption example considered in this article, as well as other practical situations. A mathematical analysis of assumptions is made, using the fact that maximum likelihood estimation is asymptotically equivalent to least squares regression. It is shown that, for inferences regarding the upper tail of the two-parameter lognormal distribution, the introduction of the usual third parameter, or, almost equivalently, censoring the lower tail, can remove the anomaly. The theoretical results are illustrated by a numerical example, using data from an alcohol consumption survey at Busselton, Western Australia, showing how the specification affects the estimates of certain functions of the class probabilities.

ISSN:0162-1459    

DOI:10.1080/01621459.1989.10478743

出版商:Taylor & Francis Group    

年代:1989

数据来源: Taylor

摘要:

Hellinger distance analogs of likelihood ratio tests are proposed for parametric inference. The proposed tests are based on minimized Hellinger distances between nonparametric density estimates and densities corresponding to null and unconstrained parametric models. If the parametric model is correct, the Hellinger deviance test is asymptotically equivalent to the likelihood ratio test. A simulation study examines the relative performance of these tests in finite samples. Breakdown properties of the Hellinger deviance tests and likelihood ratio tests are compared. The Hellinger tests are generally high breakdown-point tests, whereas, in many instances, the likelihood ratio tests have breakdown points of 0. Two numerical examples illustrate the implementation and performance of the proposed tests. Likelihood-based methods are widely used in applications. They provide a routine method for generating efficient estimates and inference statements. In many instances, however, these methods are known to be sensitive to anomalous data points, and the careful data analyst will screen for outliers prior to a likelihood-based analysis. In large-scale data analysis, for instance, when processing the results of many similar experiments in a data base, the outlier screening procedure needs to be automated. This type of data processing is common in mutation research, where numerous chemicals are subjected to a battery of mutagenicity and carcinogenicity trials. An automated outlier screen is one method for obtaining robust inferences in this setting. The Hellinger test studied here is proposed as a more direct method for obtaining robust inferences. The test statistic simply measures how much farther the data are from the null model than from the unconstrained model. Regions of low predicted density receive very little weight, so the Hellinger deviance test is able to cope with anomalous data. The method adapts to a variety of parametric hypothesis testing situations.

ISSN:0162-1459    

DOI:10.1080/01621459.1989.10478744

出版商:Taylor & Francis Group    

年代:1989

数据来源: Taylor

摘要:

The problem considered is the computation reduction for general delete-djackknife variance estimators. The delete-djackknife estimator was proved consistent (Shao and Wu 1986), and in this article its mean squared error is shown to have ordero(n–2), wherenis the sample size. These properties are not shared by the traditional delete-1 jackknife in some situations. Use of the delete-djackknife, however, requires (nd) recomputations of a point estimate θ, which increases rapidly asnanddincrease. Using techniques from survey sampling, a shortcut can be taken with θ evaluated onlymtimes,m≪ (nd). The efficiency and consistency of the resulting jackknife-sampling (hybrid) variance estimators (JSVE's) are studied. Ifmis chosen so thatn/m→ 0, the increase in mean squared error by using the JSVE is relatively negligible. For the consistency of JSVE,m→ ∞ is sufficient. Hence the JSVE withm<ncan also be used to alleviate the computational burden for the delete-1 jackknife in the case wherenis large and evaluating θ needs large computations. The performance of JSVE is also studied in a simulation study.

ISSN:0162-1459    

DOI:10.1080/01621459.1989.10478745

出版商:Taylor & Francis Group    

年代:1989

数据来源: Taylor

摘要:

Expressions are derived for lower bounds on Bayes factors for interval null hypotheses. These are then studied for varying interval lengths and are related to two interesting results—one for point null hypotheses and another for one-sided hypotheses. These results are illustrated with two of the most frequently used hypothesis tests—the normal test and Student'sttest.

ISSN:0162-1459    

DOI:10.1080/01621459.1989.10478746

出版商:Taylor & Francis Group    

年代:1989

数据来源: Taylor

摘要:

A new two-parameter family of continuous univariate distributions on the interval (– 1, 1) is introduced, and some properties are given. It is shown that parameters θ andvare globally orthogonal in the Fisher information sense and that, to some extent, they have the properties of a location parameter and a precision parameter, respectively. A pivotal statistic is constructed whose distribution is independent of the location parameter. Finally, the connection with ultraspherical functions and Brownian motion is described.

ISSN:0162-1459    

DOI:10.1080/01621459.1989.10478747

出版商:Taylor & Francis Group    

年代:1989

数据来源: Taylor