www.phidot.org

Hi Murray,

I'm helping out a colleague with a secr analysis and cannot determine the cause of the following error reported for calculating fitted real parameters when printing a secr.fit object for a model with structural parameters modeled as functions of covariates. In our case, the only parameter modeled with covariates is density. Can you tell me possible sources of the error?

Thanks!
Jared

Code: Select all

Fitted (real) parameters evaluated at base levels of covariates
Error in object$details$scalesigma & scaled :
  operations are possible only for numeric, logical or complex types


Hi Jared
The message seems to be coming from predict.secr, but I've never seen it before. You seem to be using an old version of secr (that code does not appear in the current version) so the first thing to do is probably to update and report the current error message, if any, with code. The handling of alternative parameterisations (a0, k etc.) was revamped about a year ago.
Murray

Hi Murray,

Thanks for the quick response. The error code was produced with version 2.10.2. I refit the model on my workstation that has version 2.9.5 and was able to obtain real parameter estimates without the error message occurring. Perhaps more alarming is that some beta estimates were quite different between versions including a beta coefficient for session-specific density switching signs. If it would help, I can send you the secr.fit objects offline so you can take a look at them. Given the error occurred with version 2.10.2, would you still recommend updating to 2.10.3 and refitting? Let me know if you would still like to see the code and I can put together a workspace of relevant R objects and a script to fit the model.

Thanks
Jared

I do not see how 2.10.2 can have produced the error message as the words do not appear in the R code and the 'scaled' argument of predict was explicitly removed in version 2.10.0 (see news). I suggest you check again. Updating from 2.10.2 to 2.10.3 won't affect this problem, but keeping up to date is recommended. I am not yet convinced this is something I should spend time on.
Murray