Norbert David Yanez Iii
Advanced-level topics in biostatistics offered by regular and visiting faculty. Prerequisite: permission of instructor. Offered: jointly with STAT 578; AWSpS.
Covariate measurement error is present in almost all statistical analyses. However, its effects are typically dismissed as bias towards the null, with little understanding of whether this statement is actually appropriate. The truth is often more complicated. This course aims to equip the student with a more accurate intuition about when these situations arise and where they may have an important analytic impact.
After acknowledging that covariate measurement error can substantially affect inference in many situations, it is clearly important to appropriately adjust results to reflect this problem. This course will teach classic and more modern methods to assess and adjust for such errors, chiefly in the context of regression-based analyses. Examples and methods covered will be targeted for a biostatistical audience.
The core of the material will be
1. Impact of measurement error on continuous covariates in routine linear, generalized linear models 2. Impact of misclassification on categorical covariates in routine linear, generalized linear models 3. Classical & Berkson errors 4. Regression calibration 5. The SIMEX algorithm 6. Likelihood-based methods for adjustment
Time allowing, there will also be discussion of novel methods and more field-specific issues;
7. Moment reconstruction 8. Matched case-control studies and other outcome-dependent sampling 9. Survival analysis
Student learning goals
General method of instruction
While the course will emphasize problems of practicality and applied relevance, theoretic results will also be covered. In particular students should be familiar with the application of likelihood as per BIOSTAT/STAT 570.
Class assignments and grading
Grades will be based on 20% class participation, 20% exercises, 20% in-class midterm exam and 40% term paper. The term paper should be written with the aim of publication in a statistics journal; for example Statistics in Medicine. Suggested paper titles will be available, but the only restriction is that the paper relate to the subject matter of the course, i.e. to topics either discussed in class or mentioned in bibliographies that will be made available on the class website. Presentation of the paper is mandatory, either orally during class sessions, or (dependent on student numbers) in a final poster session.
Possible topics for the term paper include;
Data Analysis: A comprehensive and innovative data analysis, preferably from a research project in which you have been involved. Adjustment of an existing or novel analysis for realistic covariate measurement error is suggested, with comparisons of any similarities or differences due to the methods of adjustment considered.
Simulation Study: A comprehensive simulation study evaluating and/or comparing the performance of different adjustment methods to a well-specified problem, for which covariate measurement adjustment is not routine. Simulation studies assessing the impact of covariate measurement error are also acceptable if the effect seen is novel, and divergent from known results addressing previously-studied situations
Review Article: A comprehensive but focused review of a well-defined scientific area where it is appropriate to consider covariate measurement error. The strengths and weaknesses of various adjustment methods should be considered relative to the sources of data likely to be available. Examples contrasting these should be given. The target audience here should be considered more clinical than Statistics in Medicine.
Theory/Methodology (Students are advised not to select this option unless they already have dissertation or other work-in-progress that could be suitably modified or extended). Development of new statistical theory and/or methodology related to one of the topics that would be suitable for publication, e.g. in the Theory and Methods Section of JASA, or another methods journal.
Papers will be evaluated on the basis of originality, scholarship (including appropriate literature citations), clarity, organization and relevance to class goals.