This course is only offered in English. Ce cours est offert en anglais.
Spaces are limited. Please register early to avoid disappointment.
- Session 1: September 25, 2013, Wednesday from 1:00 to 4:00 PM EST
- Session 2: September 26, 2013, Thursday from 1:00 to 4:00 PM EST
Course Level: Intermediate
Course Type: Interactive instruction and discussion. Total 6 hours.
Keith O’Rourke Senior Epidemiologist and Biostatististian O’Rourke Consulting firstname.lastname@example.org
For More Information: Contact email@example.com.
Direct Visualizations of the Mechanics of Bayesian and Frequentist Methods: A Virtual Meccano Set to Directly Represent and Analyse Observations
The Canadian Society for Epidemiology and Biostatistics is proud to offer an interactive online short course on the “mechanics” of Bayesian and Frequentist methods. This series of webinar workshops is designed for those who wish to enhance their knowledge and especially understanding of advanced quantitative statistical methods that can be used in epidemiology.
Course Background: The course presents physical and simulation based representations of statistical methods as an alternative to mathematical representations to grasp how statistical methods actually work. Pragmatic understanding involves adequately (and concretely) grasping how individual observations and assumptions collaborate to provide intervals or tests for unknowns. This is arguably best achieved by being able to comfortably (even playfully) manipulate the representations or models. Almost all representations in statistics assume each observation was a random outcome drawn from a (class of) probability model(s). Until recently, comfortable manipulation required advanced mathematical training and talent. Today, simulation based representations can successfully handle many realistic applications until the number of unknowns parameters becomes too large (e.g. nine plus regression covariates). Participants will be shown how the representations can be constructed for a given set of assumptions, advanced analyses then conducted and how these can be modified for different assumptions or to handle complications such as missing data, confounding, selection bias, etc. The course will be primarily Bayesian but informative for both Bayesian and Frequentist methods. The simple physical and simulation based representations are related to Approximate Bayesian Computation (ABC) very recently developed for almost impossible to do Bayesian analyses. ABC is not required for standard Bayesian analyses but slight modifications of ABC algorithms provide exact Bayesian computation from simple simulations – a virtual Meccano set to directly represent and analyse observations. When complications arise from missing data, confounding, selection bias, etc. that complicate this, a small number of observations can still be handled.
- Provide a physical allegory (Galton’s orange tree) for Bayesian methods showing how observations revise prior to posterior probabilities and why these are useful (in what sense) to construct intervals that also can have good Frequentist confidence coverage.
- Confirm objective 1 using computer simulations for very simple but real examples.
- Clarify challenges of treatment effect parameters in the presence of multiple parameters (e.g. odds ratio versus relative risks).
- Show how confounding, selection and misclassification can be addressed.
- Show how to clarify what is learned from observations versus priors in both Bayesian and Frequentist methods (multiple bias analysis).
Format and Requirements: The course is designed for those who want to understand rather than just do statistics. Some computational skill/experience (e.g. simulations in Excel) would be helpful. The slides will start with a fictionalised historical account of how Francis Galton first thought of making a Bayesian analysis machine while “sitting under an orange tree”. It will then use computer simulations starting with simple examples. The first session will primarily focus on randomised ideally conducted experiments with the second session focussing on observational studies that epidemiologists must deal with. The schema of the programs will be presented along with graphs of their outputs. Programs will be written in R with a few in Excel and will be available for use during/after the course. Although the course is trying to get by without any calculus, a calculus based statistics course may be required to immediately appreciate what is being represented. The required probability background simply consists of one slide of basic probability formulas.
|CSEB member||$200 +hst|
|Student CSEB member||$75+hst|
|Groups of 10 participants or more||Special rates. Please contact us at firstname.lastname@example.org.|
How to Register: Please use our online registration system to register for this course. A PayPal invoice will be sent to you electronically to complete the registration. To register a group of 10 participants or more, please email@example.com.
Cancellation Policy: Refunds will be given for requests received in writing postmarked no later than September 20th, less an administration fee of $20.00. After that date, we regret no refunds will be issued; however, we will accept substitute delegates.
Certificate of Completion: Participants will be issued an electronic certificate of completion issued by the Canadian Society for Epidemiology and Biostatistics upon completion of the course.
- A computer/laptop
- Speakers/headphones/headset (microphone and headphone combo)
- A reliable internet connection
- Microsoft Excel
- A headset or microphone is required should the participant wish to ask questions or participate orally to the session
Webinar recording: The webinar will be recorded and made available to those attendees who have paid but are unable to participate in the live presentation.
Keith O’Rourke Speaker Bio: After working for 10 years in clinical research as an applied statistician specializing in meta-analysis of clinical trials (both randomized and non-randomized), he presented the mathematical basis for his likelihood based approach to meta-analysis (L’Abbe, Detsky and O’Rourke, 1987) in two ASA CE courses in 1998 and 1999. Afterwards he completed a DPhil in Statistics at the University of Oxford in 2007, where his thesis developed and extended this likelihood based approach to meta-analysis. This material was presented in a graduate level Statistics course in the Statistics department at Duke University in 2008 and then a SAMSI summer program on meta-analysis that he organised that summer. Subsequent to this, he has focussed on developing the techniques for the more usual (non-meta-analysis) statistical applications and making them more fully Bayesian, sharable and teachable in a transparent manner. He currently works at a Canadian regulatory agency.
Selected Peer-reviewed Publications:
- Greenland S, O’ ROURKE K: Meta-Analysis. In Modern Epidemiology, 3rd ed. Edited by Rothman KJ, Greenland S, Lash T. Lippincott Williams and Wilkins; 2008.
- O’ROURKE K., Altman, D. Bayesian random effects meta-analysis of trials with binary outcomes: methods for the absolute risk difference and relative risk scales. Statistics in Medicine 2005 Sep 15;24(17):2733-42.
- O’ROURKE K.Two Cheers for Bayes. Controlled Clinical Trials 1996 Aug;17(4):350-2.
- L’Abbe KA, Detsky AS, O’ROURKE K. Meta analysis in clinical research. Annals of Internal Medicine 107:224 33, 1987.