Wrong priors in Ngene

This forum is for posts that specifically focus on Ngene.

Moderators: Andrew Collins, Michiel Bliemer, johnr

Wrong priors in Ngene

Postby Benjamin » Fri Nov 04, 2016 7:56 pm

Hi!

What happens if we use wrong priors in Ngene (only sign => example.: -0,000001), in other words: wrong sign!

And this happens on half of the attributes in the model..

Will the data be totally unusable? Or what is possible to use? How does Ngene use take use of the priors we give?

-Ben
Benjamin
 
Posts: 13
Joined: Tue Sep 27, 2016 9:04 pm

Re: Wrong priors in Ngene

Postby Michiel Bliemer » Sat Nov 05, 2016 4:51 pm

The priors are used im Ngene to generate an efficient experimental design in which (Fisher) information is maximised. If the priors are different from the estimated parameter values (which is inevitable), then this simply means that the data looses some efficiency/information. The closer the priors are to the estimated parameter values, the smaller is the efficiency/information loss.

Note that loss of information does not mean that you cannot use the data for model estimation. For example, orthogonal designs make an implicit assumption that all priors are equal to zero, and all over the world people have been able to estimate models using data from orthogonal designs. It simply means that you will likely need a larger sample size to compensate for the loss of information.

If the priors are unknown, setting priors equal to (or close to) zero is a safe choice. Priors that are too large (positive or negative) can be problematic and could possibly lead to a rather inefficient design and should be avoided. The best approach is to conduct a pilot study (using an orthogonal design or an efficient design with priors equal to zero), estimate parameters, and use these to create normally distributed Bayesian priors (to account for uncertainty about the priors amd hence to make the design more robust against prior misspecification).

So there is no reason to worry if your priors were wrong, the data is still usable.
Michiel Bliemer
 
Posts: 1705
Joined: Tue Mar 31, 2009 4:13 pm


Return to Choice experiments - Ngene

Who is online

Users browsing this forum: No registered users and 9 guests

cron