• Journal of the Korean Data and Information Science Society
• /
• v.23 no.4
• /
• pp.627-641
• /
• 2012

We consider a model estimation and model selection methods for the multi-way contingency table data with non-response or missing values. We also consider hierarchical Bayesian model in order to handle a boundary solution problem that can happen in the maximum likelihood estimation under non-ignorable non-response model and we deal with a model selection method to find the best model for the data. We utilized Bayes factors to handle model selection problem under Bayesian approach. We applied proposed method to the pre-election survey for the 2004 Korean National Assembly race. As a result, we got the non-ignorable non-response model was favored and the variable of voting intention was most suitable.

• The Korean Journal of Applied Statistics
• /
• v.18 no.2
• /
• pp.395-410
• /
• 2005

For the analysis of large tables formed by many categorical variables, we suggest a method to group the variables into several disjoint groups in which the variables are completely associated within the groups. We use a simple function of Kullback-Leibler divergence as a similarity measure to find the groups. Since the groups are complete hierarchical sets, we can identify the association structure of the large tables by the marginal log-linear models. Examples are introduced to illustrate the suggested method.

• The Korean Journal of Applied Statistics
• /
• v.21 no.1
• /
• pp.141-149
• /
• 2008

Recently we frequently analyze multivariate data with quite large number of variables. In such data sets, virtually duplicated variables may exist simultaneously even though they are conceptually distinguishable. Duplicate variables may cause problems such as the distortion of principal axes in principal component analysis and factor analysis and the distortion of the distances between observations, i.e. the input for cluster analysis. Also in supervised learning or regression analysis, duplicated explanatory variables often cause the instability of fitted models. Since real data analyses are aimed often at multiple purposes, it is necessary to reduce the number of variables to a parsimonious level. The aim of this paper is to propose a practical algorithm for selection of a subset of variables from a given set of p input variables, by the criterion of minimum trace of partial variances of unselected variables unexplained by selected variables. The usefulness of proposed method is demonstrated in visualizing the relationship between selected and unselected variables, in building a predictive model with very large number of independent variables, and in reducing the number of variables and purging/merging categories in categorical data.

• The Korean Journal of Applied Statistics
• /
• v.16 no.2
• /
• pp.407-426
• /
• 2003

We consider the missing covariates problem in generalized estimating equations(GEE) model. If the covariate is partially missing, GEE can not be calculated. In this paper, we study the performance of 7 imputation methods to handle missing covariates in GEE models, and the properties of GEE estimators are investigated after missing covariates are imputed for ordinal data of repeated measurements. The 7 imputation methods include i) Naive Deletion ii) Sample Average Imputation iii) Row Average Imputation iv) Cross-wave Regression Imputation v) Carry-over Imputation vi) Bayesian Bootstrap vii) Approximate Bayesian Bootstrap. A Monte-Carlo simulation is used to compare the performance of these methods. For the missing mechanism generating the missing data, we assume ignorable nonresponse. Furthermore, we generate missing covariates with or without considering wave nonresp onse patterns.

• The Korean Journal of Applied Statistics
• /
• v.30 no.5
• /
• pp.615-632
• /
• 2017

Unlike randomized trial, statistical strategies for inferring the unbiased causal relationship are required in the observational studies. The matching with the propensity score is one of the most popular methods to control the confounders in order to evaluate the effect of the treatment on the outcome variable. Recently, new methods for the causal inference in latent class analysis (LCA) have been proposed to estimate the average causal effect (ACE) of the treatment on the latent discrete variable. They have focused on the application study for the real dataset to estimate the ACE in LCA. In practice, however, the true values of the ACE are not known, and it is difficult to evaluate the performance of the estimated the ACE. In this study, we propose a method to generate a synthetic data using the propensity score in the framework of LCA, where treatment and outcome variables are latent. We then propose a new method for estimating the ACE in LCA and evaluate its performance via simulation studies. Furthermore we present an empirical analysis based on data form the 'National Longitudinal Study of Adolescents Health,' where puberty as a latent treatment and substance use as a latent outcome variable.

• The Korean Journal of Applied Statistics
• /
• v.14 no.1
• /
• pp.39-57
• /
• 2001

Categorical data has sometimes misclassification errors. If this data will be analyzed, then estimated cell probabilities could be biased and the standard Pearson X2 tests may have inflated true type I error rates. On the other hand, if we regard wellclassified data with misclassified one, then we might spend lots of cost and time on adjustment of misclassification. It is a necessary and important step to ask whether categorical data is misclassified before analyzing data. In this paper, when data is misclassified at one of two variables for two-dimensional contingency table and marginal sums of a well-classified variable are fixed. We explore to partition marginal sums into each cells via the concepts of Bound and Collapse of Sebastiani and Ramoni (1997). The double sampling scheme (Tenenbein 1970) is used to obtain informations of misclassification. We propose test statistics in order to solve misclassification problems and examine behaviors of the statistics by simulation studies.

• The Korean Journal of Applied Statistics
• /
• v.14 no.1
• /
• pp.25-37
• /
• 2001

The marginal model is proposed for the analysis of data which have c(2: 3) categories in the 3 x 3 cross-over trials with three periods and three treatments. This model could be used for the counterpart of the Kenward-Jones' joint probability one and should be the generalization of Balagtas et ai's univariate marginal logits one, which analyze the treatment effects in the 3 x 3 cross-over trials with binary response variables[Kenward and Jones(1991), Balagtas et al(1995)]. The model equations for the marginal probability are constructed by the three types of link functions. The methods would be given for making of the link function matrices and model ones, and the estimation of parameters shall be discussed. The proposed model is applied to the analysis of Kenward and Jones' data.

• Journal of the Korean Data and Information Science Society
• /
• v.20 no.5
• /
• pp.819-827
• /
• 2009

This study was conducted to investigate the relationship between the sociodemographic factors and the Korean consumers palatability evaluation grades with Hanwoo sensory evaluation data. The dichotomy logistic regression model and the multinomial logistic regression model are fitted with the independent variables such as the consumer living location, age, gender, occupation, monthly income, and beef cut and the the palatability grade as the dependent variable. Stepwise variable selection procedure is incorporated to find the final model and odds ratios are calculated to find the associations between categories.

• The Korean Journal of Applied Statistics
• /
• v.19 no.2
• /
• pp.361-368
• /
• 2006

There are lots of two and three dimensional graph representing two dimensional categorical data. Among them, Li, et al. (2003) proposed Diamond Graph that projects three dimensional graph into two dimension whereby the third dimension is replaced with a diamond shape whose area and middle and vertical and horizontal lengths represent the outcome. In this paper, we use the Diamond graph to test the independence of two predictor variables for two dimensional data. And this graph could be applied for finding the best fitted log-linear model to three dimensional data.

• The Korean Journal of Applied Statistics
• /
• v.25 no.2
• /
• pp.363-376
• /
• 2012

Excessive zeroes are often observed in ordinal categorical response variables. An ordinary ordered Probit model is not appropriate for zero-inflated data especially when there are many different sources of generating 0 observations. In this paper, we apply a two-stage zero-inflated ordered Probit (ZIOP) model which incorporate the zero-flated nature of data, propose a Bayesian analysis of a ZIOP model, and apply the method to alcohol consumption data collected by the National Bureau of Statistics, Korea. In the first stage of a ZIOP model, a Probit model is introduced to divide the non-drinkers into genuine non-drinkers who do not participate in drinking due to personal beliefs or permanent health problems and potential drinkers who did not drink at the time of the survey but have the potential to become drinkers. In the second stage, an ordered probit model is applied to drinkers that consists of zero-consumption potential drinkers and positive consumption drinkers. The analysis results show that about 30% of non-drinkers are genuine non-drinkers and hence the Korean alcohol consumption data has the feature of zero-inflated data. A study on the marginal effect of each explanatory variable shows that certain explanatory variables have effects on the genuine non-drinkers and potential drinkers in opposite directions, which may not be detected by an ordered Probit model.