- YANG JIAN, University of Calgary
Minimum Hellinger Distance Estimation for Linear Regression Model [PDF]
- Minimum Hellinger Distance estimation (MHDE) has been shown an appealing method of estimation for discrete data when the assumed model is suspected to be true. In this presentation, we first introduce Minimum Hellinger Distance (MHD) as well as the MHDE. Then, we will derive the MHDE for linear regression model. Some properties of the estimator are discussed and a comparison with MLE is carried out through Monte Carlo simulation studies. Lastly, we will apply this method to a breast cancer data set and demonstrate its implementation and efficiency in estimation.
- ERIN LUNDY, University of Western Ontario
Analyzing Heaped Counts and Longitudinal Presence/Absence Data in Joint Zero-inflated Poisson Regression Models [PDF]
- Recurrent event data where a fraction of subjects are not at-risk for an event are frequently seen in longitudinal studies. In many settings, an aggregate count of the number of self-reported events over an observation period is recorded. Self-reported counts are often subject to heaping which yields a distorted distribution for the observed counts and therefore may bias estimation. Alternatively, the presence/absence of events between shorter periodic assessments may be recorded. Motivated by a major study of criminal behaviour, we compare the analysis of aggregate heaped count data and longitudinal presence/absence data using joint zero-inflated Poisson regression models.
- RACHID BENTOUMI, Université d'Ottawa
Information Gain under Length-biased Sampling [PDF]
- In epidemiological studies, subjects with disease (prevalent cases) differ from newly diseased (incident) cases. Methods for regression analyses have recently been proposed to measure the potential effects of covariates on survival. We propose to extend the measure of dependence based on information gain in the context of length-biased sampling. This will require development of an estimator for the covariate distribution. We will assess the asymptotic properties of the measure of dependence and estimated information gain and illustrate the methods using both simulation and application to the Canadian Study on Health.
- ELNAZ GHADIMI, Concordia University
Multivariate Cure Rate Estimation under Random Censoring [PDF]
- We considered a non-parametric multivariate cure rate estimator under random censoring. Individuals can be cured or fail. The event of interest is defined as death. Our data have censoring at the end of follow-up. Sen and Stute proposed a multivariate Kaplan-Meier estimator via mass-shifting method. We use the tail of the Sen-Stute estimator as an estimator of cure rate under random censoring. Taking a cue from Maller and Zhou without using martingale theory, the asymptotic normality of univariate and multivariate cure rate estimator is established and its variance estimator is obtained. The results are illustrated using simulation and real data.
- REGINA S. KAMPO, McMaster University
Assessing the Influence of Non-adherence on Fixed-Effect Meta-Analysis for a Continuous Outcome: A Simulation Study [PDF]
- A traditional meta-analysis assumes complete adherence for an intervention. For some interventions such as exercise, change of diet etc., adherence might be a challenge that may compromise inferences. We investigated the influence of non-adherence on meta-analysis of a continuous outcome through simulations. We varied several key parameters and assessed the effect of non-adherence on estimation and hypothesis test properties. The findings from the simulation studies show that, the properties of estimation and hypothesis perform well under complete adherence but are not optimal as non-adherence is observed. Thus, researchers must treat non-adherence cautiously before using results in decision making.
- YUYING XIE, University of Waterloo
A Model Averaging Approach for Estimating Propensity Scores by Optimizing Balance [PDF]
- Many approaches, including traditional parametric modelling and machine learning techniques, have been proposed to estimate propensity scores. We proposes a new model averaging approach to propensity score estimation in which parametric and nonparametric estimates are combined to achieve covariate balance. Simulation studies are conducted across different scenarios varying in the degree of misspecification in the treatment model. It shows that the proposed method produces less bias and smaller standard error than existing approaches. This approach is applied to a real data set in evaluating the causal effect of formula or mixed feeding versus exclusive breastfeeding on a child's BMI Z-score.