The title quote is from Blaise Pascal (1623-1662), considered the father of probability statistics. There are two main types of statistics, Frequentist and Bayesian.In a gross simplification, frequentist is probably what you learned before you had a professional need to apply statistical thinking to a complex problem. Bayesian describes epistemological uncertainty (taking into account limits and validity) using mathematical probability.
A Bayesian is one who, vaguely expecting a horse, and catching a glimpse of a donkey, strongly believes he has seen a mule
A recent article describing Rapid-Cycle Evaluation has direct application to those of us needing actionable evidence. It is used in policy analysis and in the financial industry as well as medicine.
As I continue to work in health policy research and economics, I am exposed to interesting applications of measuring behavioral change and outcomes. What Works for Whom? A Bayesian Approach to Channeling Big Data Streams for Policy Analysis is an article examining a Bayesian approach that "efficiently estimates heterogeneous treatment effects, identifying what works for whom."
The authors are using a randomized control design applied to students in an online course--easy to see direct applicability to online learning in medical education for example. I use this approach to allow dynamic matching of individual study subjects (participants) with interventions most likely to address a need or benefit the subject.
The standard design (fig 2) we observe equal randomization probabilities of 20 percent for each arm contrasted with the Bayesian approach where information from each prior cohort informs the randomization of the next increasing the probability of randomization to the treatment arm that is most effective.
The authors are using a randomized control design applied to students in an online course--easy to see direct applicability to online learning in medical education for example. I use this approach to allow dynamic matching of individual study subjects (participants) with interventions most likely to address a need or benefit the subject.
The standard design (fig 2) we observe equal randomization probabilities of 20 percent for each arm contrasted with the Bayesian approach where information from each prior cohort informs the randomization of the next increasing the probability of randomization to the treatment arm that is most effective.
Different "nudging" strategies are most effective in different countries for example--or for our purposes, different specialties in medicine or a host of other demographic drivers based on our selected research question.
We now have big data in medicine. Claims data, EHR data, in addition to wearables, and CMS there is big demand for accessible and affordable but meaningful analyses. In particular, there is demand for medical education partners to mobilize data routinely collected to address the unique needs of a heterogeneous population of learners as well as the patient population.
Send an email if you are interested in applying these principles to research methodologies in value-based measurement or outcomes.
bonny@graphemeconsulting.info
Thoughtful discussions about content development and outcomes analytics that apply the principles and frameworks of health policy and economics to persistent and perplexing health and health care problems.
bonny@graphemeconsulting.info
Thoughtful discussions about content development and outcomes analytics that apply the principles and frameworks of health policy and economics to persistent and perplexing health and health care problems.
