[R-sig-ME] Suggestion for Model convergence problem:non-positive-definite Hessian matrix.

Ben Bolker bbolker at gmail.com
Wed May 24 03:34:12 CEST 2017

 [cc'ing r-sig-mixed-models at r-project.org]

 Without a repeatable example this is hard.  The only general advice I
can give is:

 - try centering and scaling any continuous predictors
 - you can probably extract some information about the fit (via
VarCorr(), fixef(), etc.) even though the model may not have converged
properly, but it will be hard to trust it
 - visualize your data (look for outlier individual observations or groups)
 - do you have combinations of predictor variables (especially
combinations of fixed effects) that lead to all-zero values?
 - if necessary, try simplifying the model

On 17-05-22 12:17 PM, Zohora Sultana (zsultana) wrote:
> Hi Dr. Bolker,
> My data set contains large amount of zeros and total number of
> observations is 52172. I am trying to run zero inflated negative
> binomial random parameter model by glmmTMB. But I am getting the
> following error: 
> /Model convergence problem; non-positive-definite Hessian matrix./
> /
> /
> Can you please give suggestions  how can I solve this problem? 
> Thanks,
> Zohora

More information about the R-sig-mixed-models mailing list