Quasi-Poisson vs. negative binomial regression: how should we model overdispersed count data?

By Jay M. Ver Hoef, Peter L. Boveng. Ecology, 2007

1,143 citations recorded by the source corpus.

Subfields: Ecology, Water Science and Technology

View Paper: https://www.semanticscholar.org/paper/01d13ec2187a90742ebcf288437e3d0a1ca28dd8

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Ecology

1. J. Linnell
2. J. Gaillard
3. J. Odden
4. R. Andersen
5. Heiko Schmaljohann
6. R. Fox
7. Harlan Campbell
8. Jonas Knape

In Water Science and Technology

1. Ferran Páramo
2. Jonas Knape
3. Anthony R. Ives
4. C. Stefanescu
5. Harlan Campbell
6. X. Harrison

Citations in the corpus

• For testing the significance of regression coefficients, go ahead and log‐transform count data (2015, background): “For example, a negative binomial GLM assumes that the variance scales according to l (1 + l / h ), while the quasi-Poisson GLM assumes the variance scales according to /l where / is a dispersion parameter; this contrast can give different results between these two GLMs (Ver Hoef & Boveng 2007).”
• Multi‐generational long‐distance migration of insects: studying the painted lady butterfly in the Western Palaearctic (2013, methodology): “However, because the variance greatly exceeded the mean in all datasets, we 270 also applied generalized linear regression models (GLM), in particular quasi-Poisson and 271 negative binomial regression models (Ver Hoef and Boveng 2007).”
• Mixed Effects Models and Extensions in Ecology with R (2009, neutral): “Further details can be found on page 111 in West et al. (2006), Verbeke and Molenberghs (2000), or Pinheiro and Bates (2000).”
• The Political Economy of Deforestation in the Tropics (2012, positive): “The only difference is that Equation (6) is weighted by initial volumes in production (deforestwp0), whereas the Poisson model implicity uses contemporaneous volumes for weights (deforestwpt) (see VerHoef and Boveng 2007).”
• What shapes Eurasian lynx distribution in human dominated landscapes: selecting prey or avoiding people? (2009, methodology): “Finally, to assess the relationship between roe deer abundance and human-related variables, we fitted a generalized linear model with a negative binomial error and a log link function to handle overdispersed count data (Ver Hoef and Boveng 2007).”
• Using observation-level random effects to model overdispersion in count data in ecology and evolution (2014, background): “Distinguishing between these two mean–variance relationships can be critical, as model mis-specification can lead to different parameter estimates in either case (Ver Hoef & Boveng, 2007), although this is not always true (O’Hara & Kotze, 2010, this study).”
• Decomposing trends in Swedish bird populations using generalized additive mixed models (2016, methodology): “For each of 500 simulated data sets for each scenario, I fitted smooth trend models with (5) and without (4) random year effects and with and without automatic selection of d.f., assuming a quasi-Poisson response (ver Hoef & Boveng 2007; the simulated data were not overdispersed).”
• The consequences of checking for zero‐inflation and overdispersion in the analysis of count data (2019, methodology): “Ver Hoef, J. M., & Boveng, P. L. (2007).”
• Sensitivity of binomial N‐mixture models to overdispersion: The importance of assessing model fit (2018, background): “Specifically, overdispersion is a common feature of population count data (ver Hoef & Boveng, 2007; Linden & Mäntyniemi, 2011), and the focus of this paper.”