• Communications for Statistical Applications and Methods
• /
• v.11 no.2
• /
• pp.399-411
• /
• 2004

A common method of handling nonresponse in sample survey is to delete the cases, which may result in a substantial loss of cases. Thus in certain situation, it is of interest to create a complete set of sample values. In this case, a popular approach is to impute the missing values in the sample by the mean or the median of responders. The difficulty with this method which just replaces each missing value with a single imputed value is that inferences based on the completed dataset underestimate the precision of the inferential procedure. Various suggestions have been made to overcome the difficulty but they might not be appropriate for public-use files where the user has only limited information for about the reasons for nonresponse. In this note, a multiple imputation method is considered to create complete dataset which might be used for all possible inferential procedures without misleading or underestimating the precision.

• Communications for Statistical Applications and Methods
• /
• v.9 no.2
• /
• pp.479-489
• /
• 2002

The case which applies two-phase sampling to stratification and nonresponse problem, it is a poweful and effective technique. In this paper we study the calibration estimator and its variance estimator for the population total using two-phase sampling method according to the of auxiliary information for population and sample having strong correlation with an interested variable in unit nonresponse situation. The auxiliary information that available both at first-phase and second-phase sampling can be used to improve weights by the calibration procedure. A weight which corresponds to the product of sampling weights and response probability is calculated at each phase of sampling.

• Journal of the Korean Statistical Society
• /
• v.32 no.2
• /
• pp.121-150
• /
• 2003

The National Health Interview Survey (NHIS) is one of the surveys used to assess the health status of the US population. One indicator of the nation's health is the total number of doctor visits made by the household members in the past year, There is a substantial nonresponse among the sampled households, and the main issue we address here is that the nonrespones mechanism should not be ignored because respondents and nonrespondents differ. It is standard practice to summarize the number of doctor visits by the binary variable of no doctor visit versus at least one doctor visit by a household for each of the fifty states and the District of Columbia. We consider a nonignorable nonresponse model that expresses uncertainty about ignorability through the ratio of odds of a household doctor visit among respondents to the odds of doctor visit among all households. This is a hierarchical model in which a nonignorable nonresponse model is centered on an ignorable nonresponse model. Another feature of this model is that it permits us to "borrow strength" across states as in small area estimation; this helps because some of the parameters are weakly identified. However, for simplicity we assume that the hyperparameters are fixed but unknown, and these hyperparameters are estimated by the EM algorithm; thereby making our method Bayes empirical Bayes. Our main result is that for some of the states the nonresponse mechanism can be considered non-ignorable, and that 95% credible intervals of the probability of a household doctor visit and the probability that a household responds shed important light on the NHIS.

• Journal of Preventive Medicine and Public Health
• /
• v.35 no.2
• /
• pp.136-140
• /
• 2002

Objectives : A common problem with analyzing survey data involves incomplete data with either a nonresponse or missing data. The mail questionnaire survey conducted for collecting lifestyle variables on the members of the Korean Elderly Phamacoepidemiologic Cohort(KEPEC) in 1996 contains some nonresponse or missing data. The proper statistical method was applied to evaluate the missing pattern of a specific KEPEC data, which had no missing data in the independent variable and missing data in the response variable, BMI. Methods : The number of study subjects was 8,689 elderly people. Initially, the BMI and significant variables that influenced the BMI were categorized. After fitting the log-linear model, the probabilities of the people on each category were estimated. The EM algorithm was implemented using a log-linear model to determine the missing mechanism causing the nonresponse. Results : Age, smoking status, and a preference of spicy hot food were chosen as variables that influenced the BMI. As a result of fitting the nonignorable and ignorable nonresponse log-linear model considering these variables, the difference in the deviance in these two models was 0.0034(df=1). Conclusion : There is a lot of risk if an inference regarding the variables and large samples is made without considering the pattern of missing data. On the basis of these results, the missing data occurring in the BMI is the ignorable nonresponse. Therefore, when analyzing the BMI in KEPEC data, the inference can be made about the data without considering the missing data.

• The Korean Journal of Applied Statistics
• /
• v.21 no.3
• /
• pp.361-375
• /
• 2008

We propose proper methods to impute the item nonresponse in 4-8-4 rotation sample survey. We consider nonignorable nonresponse mechanism that can happen when survey deals with sensitive question (e.g. income, labor force). We utilize modeling imputation method based on Bayesian approach to avoid a boundary solution problem. We also estimate a interview time bias using imputed data and calculate cell expectation and marginal probability on fixed time after removing estimated bias. We compare the mean squared errors and bias between maximum likelihood method and Bayesian methods using simulation studies.

• 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.

• Survey Research
• /
• v.11 no.1
• /
• pp.1-18
• /
• 2010

We compare various imputation methods for compensating wave nonresponse that are commonly adopted in many panel surveys. Unlike the cross-sectional survey, the panel survey is involved a time-effect in nonresponse in a sense that nonresponse may happen for some but not all waves. Thus, responses in neighboring waves can be used as powerful predictors for imputing wave nonresponse such as in longitudinal regression imputation, carry-over imputation, nearest neighborhood regression imputation and row-column imputation method. For comparison, we carry out a simulation study on a few income data from the Korean Welfare Panel Study based on two performance criteria: predictive accuracy and estimation accuracy. Our simulation shows that the ratio and row-column imputation methods are much more effective in terms of both criteria. Regression, longitudinal regression and carry-over imputation methods performed better in predictive accuracy, but less in estimation accuracy. On the other hand, nearest neighborhood, nearest neighbor regression and hot-deck imputation show higher performance in estimation accuracy but lower predictive accuracy. Finally, the mean imputation shows much lower performance in both criteria.

• Journal of applied mathematics & informatics
• /
• v.8 no.3
• /
• pp.975-987
• /
• 2001

We treat the problem of variance estimation for the estimator of population total, which is derived from the calibration estimation procedure corresponding to the levels of auxiliary information under nonresponse situation. We develop the calibrated variance estimation procedure using the fact that the population total and variance as well as the sample total and variance of the auxiliary variable are known. We show that the proposed variance estimation procedure improves the $Lundst\ddot{o}rm$ and $S\ddot{a}rndal's$ (1999) procedure with respect to the variance and nonresponse bias reduction through the simulation study.

• Proceedings of the Korean Statistical Society Conference
• /
• 2005.11a
• /
• pp.59-65
• /
• 2005

We propose five Bayesian methods to estimate the cell expectation in an incomplete multi-way categorical table with nonignorable nonresponse mechanism. We study 3 Bayesian methods which were previously applied to one-way categorical tables. We extend them to multi-way tables and, in addition, develop 2 new Bayesian methods for multi-way categorical tables. These five methods are distinguished by different priors on the cell probabilities: two of them have the priors determined only by information of respondents; one has a constant prior; and the remaining two have priors reflecting the difference in the response mechanisms between respondent and non-respondent. We also compare the five Bayesian methods using a categorical data for a prospective study of pregnant women.

• Communications for Statistical Applications and Methods
• /
• v.9 no.3
• /
• pp.665-674
• /
• 2002

The advantage of internet survey compared to the traditional survey methods are speedy in data collection, cost-effective, high performed design and able to data process and analysis at the same time. The other side are difficult to select sample, come from serious nonresponse errors. We suggest the new internet survey method to the questionnaire design that have the high response rate, enough to advanced preparations and system stability.