論文

We propose a propensity score weighted M-estimation method for marginal parametric models for groups in which all subjects are missing. It is shown that the propose estimator is consistent and its limiting distribution is normal. We also propose a hypothetsis testing method using the proposed estimator. Using the proposed method, we estimate the intervention effect of maternal smoking during pregnancy on the development of the child's cognitive functioning from data obtained from the US National Longitudinal Survey of Children and Youth.

In behavioral sciences, it is often difficult to execute an experimental study with random assignment. Therefore researchers usually do a quasi-experiment or a survey study without random assignment. However, under these studies the distributions of the covariates that would affect dependent variables usually differ with the values of the independent variables. To eliminate the influence of the covariates, various adjustment methods such as analysis of covariance have been applied to these data. Recently new adjustment methods using the propensity score proposed by Rosenbaum & Rubin (1983) ...

The purpose of the present study was to confirm the structure of scales measuring dominant psychopathologies, such as depression, anxiety, delusional ideation, and obsessional-compulsive disorder, in the context of the view of personality as a predispositional factor. Scales were administered to 362 university students. In Study 1, 14 subscales were subjected to a principal axis factor analysis with oblique rotation. Four factors were obtained. Factor 1 was mainly delusional ideations with negative emotion; Factor 2, depression, anxiety, and 2 psychopathologies of obsessional compulsive disorder; Factor 3, delusional ideation with positive emotion; and Factor 4, 2 psychopathologies of obsessional compulsive disorder. In Study 2, we calculated the correlations among the psychopathology and personality scales, and found that the variables in each factor showed the same pattern of correlation with personality. In Study 3, we used structural equation modeling in order to investigate the correlations among latent variables and personality. Each latent variable showed a different pattern of correlation with personality. We obtained a structure of psychopathology that reflected personality.

The present article proposes Bayesian inference for multiple group factor analysis, via a Gibbs sampling algorithm. When this method is used, a significance test for the difference of factor means between groups and a test for linear contrast can be done to determine whether a point hypothesis is in the Bayesian canfidemce interval of the posterior distribution. Baysian inference is not used in research in educational psychology because of its arbitrariness in the selection of the prior distribution. However, all the methods described in the present article can be done with a noninformative prior distribution, thus excluding subjectivity. As a result, researchers will be able to use Bayesian inference easily and objectively. In order to compare the proposed method with a x2 asymptotic likelihood test in terms of their respective power, 1400 simulation data sets for N = 160 and N = 80 were generated. The proposed method has some advantages over the x2 test in that control of Type-1 errors is complete with small-sized samples, and also in that the Haywood case problem does not occur. The proposed method was used to analyze the relationship of age and intelligence as measured by the WAIS-R, and a composite hypothesis was evaluated.

In educational evaluation and psychological measurement, we often measure traits that contain several sub-traits. When tests, such as entrance examinations, are used for decision-making, often test users do not evaluate the sub-traits directly, but rather evaluate the linear combination of sub-traits in which weights were pre-specified by the test users. In the present research, we propose a new test information function and item information that enable us to evaluate linear combinations of sub-traits in which the traits are correlated. Nuisance dimensions of linear combinations are integrated out in the proposed information function. This will enable test editing in which linear combinations of sub-traits can be evaluated prior to executing the test. We also propose a method for comparing values of information functions that have different weights and/or different correlations. A numerical simulation study showed the validity of our methods. We also applied our methods to the item pool of Geometry and Statistics of the 1996 NAEP test.

When two groups are present, they are said to form an allometric model, if one group is the extension of the other group along the main axis of variation. The model is widely used in the context of principal component analysis, especially for the description of growth processes of creatures. In this paper, the notion of allometric extension model is applied to conditional distributions. More specifically, we derive a sufficient condition, for which the two conditional distributions given the sign of the first principal component form an allometric extension model. © 2008 Elsevier Inc. All rights reserved.

This article proposes a Bayesian method to directly evaluate and test hypotheses in multiple comparisons. Transformation and integration over the coordinates relevant to the hypothesis are shown to enable us to directly test the hypotheses expressed as a linear equation of a parameter vector, given a linear constraint. When the conditional posterior distribution of the parameter vector we are interested in is the multivariate normal distribution, the proposed method can be applied to calculate the p-value of hypotheses pertaining to the parameters in any complex model such as generalized linear mixed effect models with latent variables, by using outputs from Markov chain Monte Carlo (MCMC) methods. Further, the proposed testing can be implemented without prior information. Some applications are presented, and the simulation results are provided to compare the powers of this method with those of other methods of conventional multiple comparisons. Simulation studies have shown that the proposed method is valid for multiple comparisons under nonequivalent variances and mean comparisons in latent variable modeling with categorical variables. © 2007 Elsevier Ltd. All rights reserved.

The estimation of the differences among groups in observational studies is frequently inaccurate owing to a bias caused by differences in the distributions of covariates. In order to estimate the average treatment effects when the treatment variable is binary, Rosenbaum and Rubin [1983. The central role of the propensity score in observational studies for causal effects. Biometrika 70, 41-55] proposed an adjustment method for pre-treatment variables using propensity scores. Imbens [2000. The role of the propensity score in estimating dose-response functions. Biometrika 87, 706-710] extended the propensity score methodology for estimation of average treatment effects with multivalued treatments. However, these studies focused only on estimating the marginal mean structure. In many substantive sciences such as the biological and social sciences, a general estimation method is required to deal with more complex analyses other than regression, such as testing group differences on latent variables. For latent variable models, the EM algorithm or the traditional Monte Carlo methods are necessary. However, in propensity score adjustment, these methods cannot be used because the full distribution is not specified. In this paper, we propose a quasi-Bayesian estimation method for general parametric models that integrate out the distributions of covariates using propensity scores. Although the proposed Bayes estimates are shown to be consistent, they can be calculated by existing Markov chain Monte Carlo methods such as Gibbs sampler. The proposed method is useful to estimate parameters in latent variable models, while the previous methods were unable to provide valid estimates for complex models such as latent variable models. We also illustrated the procedure using the data obtained from the US National Longitudinal Survey of Children and Youth (NLSY1979-2002) for estimating the effect of maternal smoking during pregnancy on the development of the child's cognitive functioning. © 2007 Elsevier B.V. All rights reserved.