<p> Many medical studies are performed to investigate the effectiveness of new treatments (such as new drugs, new surgery) versus traditional (or placebo) treatments. In many cases, researchers measure a continuous variable at baseline and again as an outcome assessed at follow up. The baseline measurement usually has strong relationship with post treatment measurement. Consequently, the ANCOVA model using baseline as covariate may provide more powerful and precise results than the ANOVA model.</p> <p> However, most epidemiologic studies will encounter the problem of missing covariate data. As a result, the patients with missing baseline measurements will be excluded from the data analysis. Hence, there exists a tradeoff between the ANOVA with full data set and the ANCOVA with partial data set.</p> <p> This study focuses on the variance of the estimator of treatment means difference. In practical situation, the standard error of the estimator obtained from the ANCOVA model with partially missing baseline relative to the standard error obtained form the ANOVA with full data relies on the correlation between baseline and follow-up outcome, the proportion of the missing baseline, and the difference of the group means on the baseline. In moderate sample size studies, it is also affected by the sample size.</p> <p> The theoretically required minimum correlations for the ANCOVA model were calculated to obtain the same precision with the ANOVA model assuming the missing proportion, sample size and difference of group means on covariate are available. The minimum correlation can be obtained through checking the reference table or figures.</p> <p> The figures of asymptotic relative efficiencies provide the asymptotic variance and the length of the confidence intervals of the estimated difference obtained from the ANCOVA model relative to the ANOVA model for all the range of the correlation between baseline and follow up.</p> / Thesis / Master of Science (MSc)

This dissertation addresses problems found in observational cohort studies where the repeated outcomes of interest are truncated by both death and by dropout. In particular, we consider methods that make inference for the population of survivors at each time point, otherwise known as 'partly conditional inference'. Partly conditional inference distinguishes between the reasons for missingness; failure to make this distinction will cause inference to be based not only on pre-death outcomes which exist but also on post-death outcomes which fundamentally do not exist. Such inference is called 'immortal cohort inference'. Investigations of health and cognitive outcomes in two studies - the 'Origins of Variance in the Old Old' and the 'Health and Retirement Study' - are conducted. Analysis of these studies is complicated by outcomes of interest being missing because of death and dropout. We show, first, that linear mixed models and joint models (that model both the outcome and survival processes) produce immortal cohort inference. This makes the parameters in the longitudinal (sub-)model difficult to interpret. Second, a thorough comparison of well-known methods used to handle missing outcomes - inverse probability weighting, multiple imputation and linear increments - is made, focusing particularly on the setting where outcomes are missing due to both dropout and death. We show that when the dropout models are correctly specified for inverse probability weighting, and the imputation models are correctly specified for multiple imputation or linear increments, then the assumptions of multiple imputation and linear increments are the same as those of inverse probability weighting only if the time of death is included in the dropout and imputation models. Otherwise they may not be. Simulation studies show that each of these methods gives negligibly biased estimates of the partly conditional mean when its assumptions are met, but potentially biased estimates if its assumptions are not met. In addition, we develop new augmented inverse probability weighted estimating equations for making partly conditional inference, which offer double protection against model misspecification. That is, as long as one of the dropout and imputation models is correctly specified, the partly conditional inference is valid. Third, we describe methods that can be used to make partly conditional inference for non-ignorable missing data. Both monotone and non-monotone missing data are considered. We propose three methods that use a tilt function to relate the distribution of an outcome at visit j among those who were last observed at some time before j to those who were observed at visit j. Sensitivity analyses to departures from ignorable missingness assumptions are conducted on simulations and on real datasets. The three methods are: i) an inverse probability weighted method that up-weights observed subjects to represent subjects who are still alive but are not observed; ii) an imputation method that replaces missing outcomes of subjects who are alive with their conditional mean outcomes given past observed data; and iii) a new augmented inverse probability method that combines the previous two methods and is doubly-robust against model misspecification.

In this work we model the historical development of best performances in men's 100, 200, 400 and 800m running events. We suppose that the years best performances are independent random variables with generalized extreme value distribution for minima and that there is a decreasing trend in location. Parameters of the models are estimated by using maximum likelihood techniques. The data of years best performances are missing for some years, we treat them as right censored data that are censored by value of world record valid at that time. Graphic tools used for models diagnostics are adjusted to the censoring. The models we get are used to estimate the ultimate records and to predict new records in next years. At the end of the work we estimate several models describing historical development of years best performances for more events at one time.

The longitudinal feature of measurements and counting process of responses motivate the regression models for longitudinal count data (LCD) to take into account the phenomenons such as within-subject association and overdispersion. One common problem in longitudinal studies is the missing data problem, which adds additional difficulties into the analysis. The missingness can be handled with missing data techniques. However, the amount of missingness in the data and the missingness mechanism that the data have affect the performance of missing data techniques. In this thesis, among the regression models for LCD, the Log-Log-Gamma marginalized multilevel model (Log-Log-Gamma MMM) and the random-intercept model are focused on. The performance of the models is compared via a simulation study under three missing data mechanisms (missing completely at random, missing at random conditional on observed data, and missing not random), two types of missingness percentage (10% and 20%), and four missing data techniques (complete case analysis, subject, occasion and conditional mean imputation). The simulation study shows that while the mean absolute error and mean square error values of Log-Log-Gamma MMM are larger in amount compared to the random-intercept model, both regression models yield parallel results. The simulation study results justify that the amount of missingness in the data and that the missingness mechanism that the data have, strictly influence the performance of missing data techniques under both regression models. Furthermore, while generally occasion mean imputation displays the worst performance, conditional mean imputation shows a superior performance over occasion and subject mean imputation and gives parallel results with complete case analysis.