Multi-Season Analysis - zero detection in first season

questions concerning analysis/theory using program PRESENCE

Re: Multi-Season Analysis - zero detection in first season

Postby jhines » Thu Dec 01, 2016 4:18 pm

Hi Leyla,

The log-likelihood should always be higher (negative log-likelihood should be lower) when you add covariates to a model. Sometimes, the initial values of the parameters lead the optimization process in the wrong direction and it finds a local maximum log-like value, which is lower than a simpler model. For this reason, there is an option in PRESENCE to supply your own initial values for the beta parameters, so it will find the overall maximum log-like value. In the Run model dialog box, there are a bunch of check-boxes on the right-hand side. One of those is for supplying initial values. Check that box, then when you click 'Run', a new dialog box will appear with an empty box for the initial beta values. I suggest entering the final beta values from the simpler model, plus an initial value of zero for the new beta associated with the covariate that you're adding to the model. For example, if the simpler model is just psi(.),p(.), you'll have 2 beta values, b1 and b2. If you want to try a model, psi(.),p(timeofday), you'll have 3 beta's. In the initial value box, enter the b1 and b2 from the output of the simple model (one per line), then 0 for the 3rd beta. In this case, the log-likelihood value for the 2nd model will start with the final log-like value from the simple model, then go higher from there.

Jim
jhines
 
Posts: 599
Joined: Fri May 16, 2003 9:24 am
Location: Laurel, MD, USA

Previous

Return to analysis help

Who is online

Users browsing this forum: No registered users and 6 guests