Page 1 of 1

A question of mode_check plot

PostPosted: Sun Sep 28, 2014 6:37 am
by Econbee
Dear all,

I recently encountered a problem when finding the posterior mode. I hope anyone who experienced similar situations could give me some hints or suggestions. The plot has been attached.

Usually, the mode_check plot after finding the mode looks like a smooth concave curve as the middle one in the attached plot. But I also got some cases like the left one and the right one, in which the curve is not a concave one. The highest point is the intersection of two convex curves. I did a very comprehensive search for the mode and always got this result. I am wondering if this situation is normal?

Thanks in advance!

Re: A question of mode_check plot

PostPosted: Sun Sep 28, 2014 8:21 am
by jpfeifer
I cannot say much about this, but it looks your prior is relatively wide and the posterior around +- 10% of the mode value seems to concave. Posteriors can be really irregular.

More generally, I am not aware of any argument that a cut through the posterior should be concave for the whole range. It should be concave in an open neighbourhood of the mode, but this seems to be the case.

Re: A question of mode_check plot

PostPosted: Sun Sep 28, 2014 8:55 am
by Econbee
Thanks for your prompt reply, jpfeifer!