I’m trying to fit a negative binomial GLM for a count response variable (stems per hectare). The data were significantly overdispersed. Thus I chose glm.nb() from MASS and then checked model fit using the DHARMa package. However, the DHARMa residual density plot shows a clear “two-hump”/bimodal pattern and significant quantile deviation, suggesting some form of nonlinearity or heteroskedasticity. The QQ plot did not show any substantial dispersion or outliers, and the KS test results were not significant. Can someone help me with approaches to deal with such an issue

Full article content could not be extracted automatically. Read the original below.