D-error of designs from non-informative vs. Bayesian priors

[sam197902]

Hi Michiel,

Can a design generated from non-informative priors (I used small priors close to 0) be more efficient than a design derived from Bayesian priors sourced from a pilot study? In our study, the design using non-informative priors had a D-error of 0.1632, whereas the design with Bayesian priors reported a D-error of 0.3698. The pilot study had a sample size of 30.
Does this suggest that the design created with non-informative priors is more efficient and should be used for the primary data collection?

Thanks
Sameera

[Michiel Bliemer]

Your comparison is flawed. You can only compare D-errors assuming the same priors. If priors are different, then D-errors are no longer comparable.

Design 1: generated using uninformative priors
Design 2: generating using informative Bayesian priors with D-error is 0.3698

You need to evaluate Design 1 under the exact same informative Bayesian priors to compare efficiency. You can use the command ;eval in Ngene to do this.

Michiel

[sam197902]

Thank you very much for the clarification!

S