That you are using the Poisson family indicates that the values are counts... if this is correct you should better ask how negative counts can arise. This does not seem to make any sense. Are you wildly playing around or doe you have some idea about the data generatiing process?

My data came from a meta-analysis output, so that is the reason why there are negative values. I indicated Poisson family because the model requests an indication but it was only made to complete the code since there seems to be an issue to be solved before I really run it. I edited the question to avoid misinterpretation.

My data distribution does not look to fit in any of the options the model accepts, so it is apparently also a part of the problem.

What are your observations? What does the data matrix look like? There is not enough information about the nature of the data to draw conclusions. What do you data look like?

I am adding the needed information from now. My observations result from a meta-analysis. yi is mean effect size and ID are the ids of the studies. I did not manage to model the significance of treatment (turbidity on freshwater plankton) through any of the "metafor" R package functions since my model has a random factor and is built over non-normally distributed data and residuals, so it was indicated to use lmer4 glmm to do so. I am including a subset of the matrix with mean effect (yi), effect variation (vi) which is not used in the model but is a part of the data and ID, as the identification of the referent study.

Any help would be appreciated.