Timeline for Parameter values fall outside the prior range after post-hoc adjustments in the context of Approximate Bayesian Computation?
Current License: CC BY-SA 3.0
7 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jan 4, 2018 at 7:46 | comment | added | Xi'an | Your question is hard to answer without further details. The prior may be incompatible with the data, the non-parametric regression can be poor, &tc. I would suggest running several ABC with different priors and with/without the postprocessing step to assess the impact of the different factors. | |
| Nov 17, 2017 at 17:24 | comment | added | Dave Harris | If you have zero mass on the prior range, however, you then have a problem. It is no different than being sure the Earth is flat and allowing a zero percent probability that it is round. That is a problem prior because you can never learn the truth. | |
| Nov 17, 2017 at 17:22 | comment | added | Dave Harris | The prior is your prior belief. It appears that your prior belief was wrong. You do not recalibrate the prior because the data does not like it. Besides, you could just have a weird sample, in which case, the prior is protective. There should be at least some prior mass everywhere that is possible. The analysis is not invalid, but you will want to disclose the strength of the effect the prior had on the posterior. It may have little effect in practice. The data can overwhelm the prior. This is just a disclosure issue. | |
| Nov 17, 2017 at 16:41 | history | edited | SimonLL | CC BY-SA 3.0 |
edited title
|
| Nov 17, 2017 at 6:34 | history | edited | SimonLL | CC BY-SA 3.0 |
edited title
|
| Nov 16, 2017 at 16:53 | history | edited | SimonLL | CC BY-SA 3.0 |
edited title
|
| Nov 16, 2017 at 15:42 | history | asked | SimonLL | CC BY-SA 3.0 |