Hi all,
Is there a standard / commonly accepted Bayesian approach to calculating sample numbers? I’ve always been taught to use a pre-determined sampling strategy such as the '77 NIOSH tables or a caluclation with t-stat, GM, & SD.
I’m reading “Doing Bayesian Data Anaylsis” by John Kuschke.
It explains to estimate desired sample numbers:
-
Create a hypothetical informed prior.
(In the practice, I understood that this prior would be informed in the same why you inform any prior - exposure results collected from the same task last year, results from a preliminary test (EN689), literature etc. etc.) -
Generate representative data set based on a particular plan - i.e. if N samples were collected
-
Tally if the goal was reached by this data set
(In the practice, is the overexposure risk below the overexposure risk threshold) -
Repeat many times to approximate the power if N samples were collected
I guess you then play around with how large N is in this simulation until the desired power is reached. (What level of power there should be when calculating the overexposure risk… I don’t know.)
Am I way off? Is there reading anyone can recommend on this topic in the context of occupational hygiene?
Diagram of power analysis taken from the mentioned textbook.