Wednesday 20 Jan 2016: Making Valid Inferences in Observational Studies using Propensity Score Analysis
Dr Michael A Posner - Villanova University
Room 115, Rolle Building, Plymouth University 15:00-16:00
The International Centre for Statistical Education based in the School of Computing, Electronics and Mathematics has invited Dr Michael A Posner - Associate Professor of Statistics, Villanova University, USA to give a talk entitled 'Making Valid Inferences in Observational Studies using Propensity Score Analysis'.
Randomised controlled trials are considered the gold standard in research studies. However, situations often arise which make them unfeasible, unethical, too restrictive in their generalisability, or just too time consuming and expensive.
A common alternative is using observational or natural studies where subjects self-select into modalities. In the era of data science and big data, some have termed this “found data.” However, observational data presents challenges in making valid inferences due to the presence of selection bias and confounding variables. The propensity score method is frequently used for analysis of observational data in fields including medicine, psychology, education, and survey research.
It is essentially stratification on multiple factors using a single summary measure and is performed by calculating the conditional probability (propensity) of being in the treated group given a set of covariates, weighting (or sampling) the data based on these propensity scores, and then making statistical inferences using the weighted data.
In this colloquium, Michael provides an overview of propensity score analysis and review methods of data selection or allocation of weights, including proposing an alternative weighting method – weighting within strata. This new method is compared to existing ones using empirical analysis and via an application on sending patients to respite care. Simulations are then described and discussed to compare the existing and new methods.
All are welcome to this talk. Please book your place in advance by emailing the ICSE on email@example.com.