You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Feedback by Franziska Meissner concerning the ReaL model:
The MPTinR no-pooling approach often threw errors that some participants did not converge successfully. This is something that puzzled me because I did not observe these problems in multitree (as long as the number of iterations was changed to a reasonably high value). To me, it seems that some adjustments in the estimation process of MPTinR could be necessary for complex models in order to achieve a higher rate of convergence.
The text was updated successfully, but these errors were encountered:
This refers to warnings given out during fitting such as:
In fit.mptinr(data = data, objective = llk.model, param.names = param.names, ... :
Error code(s) in final results: 0 1. The following dataset(s) did not converge succesfully in the best fitting optimization run:
20 40
These warnings are passed on from the numerical optimizer used for model fitting in MPTinR, nlminb().
They indicate that the (numerical) optimizer cannot confirm that it has really reached an optimum or maxima. It usually indicates that for those data sets (here number 20 and 40) the likelihood surface is somehow problematic. Most likely the issue is either a non-identified parameters or a relatively flat likelihood surface near the optmimum. The usual cause of this is a relatively low number of data points relative to the number of parameter or (more likely) many empty cells (which can easily lead to some parameters being non-identifiable).
Because MPTmultiverse contains additional checks on-top of what is provided by MPTinR and replaces parameters that are empirically non-identified with NA, these warnings can usually be ignored. However, it seems also wrong to remove them as they provide valuable information of the data quality.
Please be sure to run ckeck_mpt() to ensure that there are not additional problems with these data sets. This will also list all non-identified parameters.
Feedback by Franziska Meissner concerning the ReaL model:
The MPTinR no-pooling approach often threw errors that some participants did not converge successfully. This is something that puzzled me because I did not observe these problems in multitree (as long as the number of iterations was changed to a reasonably high value). To me, it seems that some adjustments in the estimation process of MPTinR could be necessary for complex models in order to achieve a higher rate of convergence.
The text was updated successfully, but these errors were encountered: