We currently use Poisson GLM for the estimation of the growth rate. This likely underestimate uncertainty, definitely for the variance in predictions (but we don't care), but also possibly for the mean prediction (confidence interval). The reason NB GLM is not used currently is that it is too unreliable and does not fit some of the countries. We may consider using quasiPoisson but this model needs implementing in i2extras::fit_curve().
We currently use Poisson GLM for the estimation of the growth rate. This likely underestimate uncertainty, definitely for the variance in predictions (but we don't care), but also possibly for the mean prediction (confidence interval). The reason NB GLM is not used currently is that it is too unreliable and does not fit some of the countries. We may consider using quasiPoisson but this model needs implementing in
i2extras::fit_curve()
.