• Communications for Statistical Applications and Methods
• /
• v.10 no.2
• /
• pp.277-289
• /
• 2003

We propose modified exact inferential methods in logistic regression model. Exact conditional distribution in logistic regression model is often highly discrete, and ordinary exact inference in logistic regression is conservative, because of the discreteness of the distribution. For the exact inference in logistic regression model we utilize the modified P-value. The modified P-value can not exceed the ordinary P-value, so the test of size $\alpha$ based on the modified P-value is less conservative. The modified exact confidence interval maintains at least a fixed confidence level but tends to be much narrower. The approach inverts results of a test with a modified P-value utilizing the test statistic and table probabilities in logistic regression model.

• Journal of Information Technology Services
• /
• v.9 no.3
• /
• pp.123-144
• /
• 2010

As knowledge becomes a critical success factor of companies nowadays, lots of rule-based systems have been and are being developed to support their activities. Large number of rule-based systems serve as Web sites to advise, or recommend their customers. They usually use a backward chaining inference algorithm based on backtrack to implement those interactive Web-enabled rule-based systems. However, when the users like customers are using these systems interactively, it happens frequently where the users do not know some of the answers for the questions from the rule-based systems. We are going to design a backward chaining inference methodology considering unknown facts based on backtrack technique. Firstly, we review exact and inexact reasoning. After that, we develop a backward chaining inference algorithm for exact reasoning based on backtrack, and then, extend the algorithm so that it can consider unknown facts and reduce its search space. The algorithm speeded-up inference and decreased interaction time with users by eliminating unnecessary questions and answers. We expect that the Web-enabled rule-based systems implemented by our methodology would improve users' satisfaction and make companies' competitiveness.

• The Korean Journal of Applied Statistics
• /
• v.25 no.5
• /
• pp.837-846
• /
• 2012

Monte Carlo methods have been used in exact inference for contingency tables for a long time; however, they suffer from ergodicity and the ability to achieve a desired proportion of valid tables. In this paper, we apply the stochastic approximation Monte Carlo(SAMC; Liang et al., 2007) algorithm, as an adaptive Markov chain Monte Carlo, to the exact test of mutual independence in a multiway contingency table. The performance of SAMC has been investigated on real datasets compared to with existing Markov chain Monte Carlo methods. The numerical results are in favor of the new method in terms of the quality of estimates.

• Journal of the Korean Data and Information Science Society
• /
• v.10 no.1
• /
• pp.29-36
• /
• 1999

We propose a new method of asymptotic inference on the odds ratio (or cross-product ratio) in $2{\times}2$ contingency table. Saddlepoint approximations to the conditional tail probability we used in this procedure. We assess the accuracy of the suggested method by comparing with the exact one. To obtain the exact values, we need very complicated calculations containing the cumulative probabilities of non-central hypergeometric distribution. The suggested method in this paper is very accurate even for small or moderate sample sizes as well as simple and easy to use. Example with a real data is also considered.

• Communications for Statistical Applications and Methods
• /
• v.15 no.2
• /
• pp.147-159
• /
• 2008

We consider using ranked set sampling methods to draw inference about the three well-known measures of overlap, namely Matusita's measure $\rho$, Morisita's measure $\lambda$ and Weitzman's measure $\Delta$. Two exponential populations with different means are considered. Due to the difficulties of calculating the precision or the bias of the resulting estimators of overlap measures, because there are no closed-form exact formulas for their variances and their exact sampling distributions, Monte Carlo evaluations are used. Confidence intervals for those measures are also constructed via the bootstrap method and Taylor series approximation.

• Journal of the Korean Data and Information Science Society
• /
• v.24 no.4
• /
• pp.911-918
• /
• 2013

This study deals with the small sample likelihood based inference for the ratio of two normal variances. The small sample likelihood inference is an approximation method. The signed log-likelihood ratio statistic and the modified signed log-likelihood ratio statistic, which converge to standard normal distribution, are proposed for the normal variance ratio. Through the simulation study, the coverage probabilities of confidence interval and power of the exact, the signed log-likelihood and the modified signed log-likelihood ratio statistic will be compared. A real data example will be provided.

• The Korean Journal of Applied Statistics
• /
• v.22 no.5
• /
• pp.1097-1102
• /
• 2009

Classification of subjects with unknown distribution in small sample size setup may involve order-restricted constraints in multivariate parameter setups. Those problems makes optimality of conventional likelihood ratio based statistical inferences not feasible. Fortunately, Roy (1953) introduced union-intersection principle(UIP) which provides an alternative avenue. Redescending M-estimator along with that principle yields a considerably appropriate robust testing procedure. Furthermore, conditionally distribution-free test based upon exact permutation theory is used to generate p-values, even in small sample. Applications of this method are illustrated in simulated data and read data example (Lobenhofer et al., 2002)

• Communications for Statistical Applications and Methods
• /
• v.15 no.5
• /
• pp.655-666
• /
• 2008

Small sample likelihood based inference for the shape parameter of the inverse Gaussian distribution is the purpose of this paper. When shape parameter is of interest, the signed log-likelihood ratio statistic and the modified signed log-likelihood ratio statistic are derived. Hsieh (1990) gave a statistical inference for the shape parameter based on an exact method. Throughout simulation, we will compare the statistical properties of the proposed statistics to the statistic given by Hsieh (1990) in term of confidence interval and power of test. We also discuss a real data example.

• Communications for Statistical Applications and Methods
• /
• v.25 no.5
• /
• pp.559-568
• /
• 2018

Gene classification can involve complex order-restricted inference. Examining gene expression pattern across groups with order-restriction makes standard statistical inference ineffective and thus, requires different methods. For this problem, Roy's union-intersection principle has some merit. The M-estimator adjusting for outlier arrays in a microarray study produces a robust test statistic with distribution-insensitive clustering of genes. The M-estimator in conjunction with a union-intersection principle provides a nonstandard robust procedure. By exact permutation distribution theory, a conditionally distribution-free test based on the proposed test statistic generates corresponding p-values in a small sample size setup. We apply a false discovery rate (FDR) as a multiple testing procedure to p-values in simulated data and real microarray data. FDR procedure for proposed test statistics controls the FDR at all levels of ${\alpha}$ and ${\pi}_0$ (the proportion of true null); however, the FDR procedure for test statistics based upon normal theory (ANOVA) fails to control FDR.

• 제어로봇시스템학회:학술대회논문집
• /
• 2004.08a
• /
• pp.631-634
• /
• 2004

In this paper, the Bayesian recurrent neural network (BRNN) is proposed to predict time series data. Among the various traditional prediction methodologies, a neural network method is considered to be more effective in case of non-linear and non-stationary time series data. A neural network predictor requests proper learning strategy to adjust the network weights, and one need to prepare for non-linear and non-stationary evolution of network weights. The Bayesian neural network in this paper estimates not the single set of weights but the probability distributions of weights. In other words, we sets the weight vector as a state vector of state space method, and estimates its probability distributions in accordance with the Bayesian inference. This approach makes it possible to obtain more exact estimation of the weights. Moreover, in the aspect of network architecture, it is known that the recurrent feedback structure is superior to the feedforward structure for the problem of time series prediction. Therefore, the recurrent network with Bayesian inference, what we call BRNN, is expected to show higher performance than the normal neural network. To verify the performance of the proposed method, the time series data are numerically generated and a neural network predictor is applied on it. As a result, BRNN is proved to show better prediction result than common feedforward Bayesian neural network.