Author: Samuel Soubeyrand; Melina Ribaud; Virgile Baudrot; Denis Allard; Denys Pommeret; Lionel Roques
Title: The current COVID-19 wave will likely be mitigated in the second-line European countries Document date: 2020_4_22
ID: dh3cgd48_45
Snippet: We used the negative-binomial distribution because the increments in mortality are counting variables and because we observed a relatively large dispersion with respect to the predictors. The Poisson distribution could not cope with this over-dispersion and led therefore to poor results in terms of forecast performance as measured in Figure 5 . Thus, the negative binomial distribution embedded in the mixture model offer larger flexibility. It mus.....
Document: We used the negative-binomial distribution because the increments in mortality are counting variables and because we observed a relatively large dispersion with respect to the predictors. The Poisson distribution could not cope with this over-dispersion and led therefore to poor results in terms of forecast performance as measured in Figure 5 . Thus, the negative binomial distribution embedded in the mixture model offer larger flexibility. It must be however noted that in an hypothetical case without overdispersion, the estimation of dispersal parameters η in the negative-binomial distributions would lead to identifiability issues. Other distributional hypotheses could be made, e.g. discrete versions of the exponential and gamma distributions defined over N, for which EM algorithms are well suited in a mixture-model context.
Search related documents:
Co phrase search for related documents- forecast performance and mixture model: 1
Co phrase search for related documents, hyperlinks ordered by date