Fixed and Bayesian designs

[Sumudu_Hewage]

Hi moderators,

I used Bayesian priors from my pilot data analysis in my Ngene design to develop the experimental design for the next pilot study.
I ran the code for 36 hours and the result for the final design is as below.

Fixed Bayesian mean
D error 0.242769 0.503778
A error 0.261961 0.667213
B estimate 75.89114 0.233881
S estimate 2521.819 17.57238

Prior b_training(e0) b_training(e1) b_monitoring(e0) b_monitoring(e1) b_he(e0) b_he(e1) b_typing(e0) b_typing(e1) asc_n
Fixed prior value -0.01929 -0.27081 0.60503 0.38115 -0.08118 -0.20622 0.25112 0.03074 -0.21856
Sp estimates 2521.819 11.73375 3.596624 6.632094 148.2556 19.36763 17.24057 853.2737 26.87951
Sp t-ratios 0.03903 0.572187 1.033495 0.761081 0.160972 0.445367 0.472042 0.067098 0.378047
Sb mean estimates 3.798151 15.05214 2.651051 2.245271 4.408886 7.469981 2.929282 2.759574 2.899802
Sb mean t-ratios 1.019325 0.721261 1.460671 1.45942 0.951293 0.829117 1.25715 1.19842 1.199566

Can the 'B estimates' and the 'S estimates' be so different in the fixed and the Bayesian designs?

Many thanks in advance.
Sumudu

[Michiel Bliemer]

If your standard errors from your pilot study are quite large, then there can be large differences yes.
For example, your fixed prior value of -0.01929 will result in a very large sample size estimate because it is essentially 0 and will be difficult to estimate.
But if your standard error for this prior value is 0.8, then draws further away from 0 will be used and the sample size estimate changes.

So please check your standard errors that you used and see if your priors are sufficiently reliable.

Michiel