00000ctm a22000004a 4500 UP-1685523046125428681 Buklod 20230503092606.0 m o j ta 071031s xx d r |||| u| (iLib)UPMNL-00000056100 UPM-CPH eng LG995 1999 P91 L67 Lopez, Charito Marie T. Comparison of properties of estimators of the mean using different imputation methods under different patterns of missing data a demonstration using the Vitamin A Expansion (VITEX) Project data Charito Marie Tuason-Lopez. 111 leaves. Thesis (Master of Science in Public Health, Biostatistics)--University of the Philippines Manila. The study is basically a simulation study which focused on nonsampling errors particularly item nonresponse which occurs when the unit or individual cooperates but fails to give a response to an item when the questionnaire is being completed. This study was undertaken to demonstrate the effect of the available case method, class mean, hot-deck and regression imputation techniques on the estimates of the mean under different patterns of missing data as applied to the VITEX project data. To simulate patterns of missing data, the concept of leverage points wa used. Ten thousand replications of seven different sample sizes were done. For each sample size, four different levels of probabilities of being missing based on leverage points were used. Also, four different cases were applied wherein individual observations were given equal probabilities of being a missing observation. The variable of interest for this study is the weight (kg.) of children 6 to 83 months.Generally, results showed that the computed biases are small and its effect on the estimates negligible as reflected by the Cochran's Criterion being 0.1. An exception to this result is the bias under the available case which is no longer considered negligible as the unequal probability of being missing is increased. The variances of the means decreases as the sample size increases, implying an increase in the precision of estimates as the sample size increases. Due to the nature of the effect of biases being negligible, the mean square errors of the means behave in exactly the same way as the variances of the mean. It is expected that as the sample size increases there is a corresponding increase in the confidence coverage because of the reduction in a, the Type 1 error however, under the presence of nonresponse, the confidence coverage became smaller as the sample size is increased. Among the imputation procedures, the regression imputation procedure seems to perform well compared to the class mean and hot deck imputation procedures. This is reflected by the small MSE, variance and bias that was exhibited by the regression imputation method.To conclude, this study does not recommend the use of the available case method especially when nonresponse rates are high since characteristics of the missing observations may differ significantly with the characteristics of respondents. It is better to do imputation on the data set before further analysis since this may possibly reduce bias due to nonresponse especially when nonrespondent' characteristics differ from respondent' characteristics. Sampling (Statistics) Multiple imputation (Statistics) Imputation methods. UP UPMNL CPH LG995 1999 P91 L67 Thesis