I wish to model the number of bugs caused by software development. This is intuitively sort of a Poisson process, however it is overdispersed. One thing we can do in this case is to use a negative binomial distribution (because negative binomial approaches Poisson as r gets larger, or because we might think the parameter $\lambda$ of the poisson is itself gamma-distributed.)

I'm not sure how to do this though. For example, we have that $$\lim_{r\to\infty}NB\left(r,\frac{\lambda}{\lambda+r}\right)=\text{Poisson}(\lambda)$$ Given that a poisson process of duration $t$ can be modeled as $\text{Poisson}(\lambda t)$ I guess we could look at $NB\left(r,\frac{\lambda t}{\lambda t+r}\right)$ - is that correct? Given that I know $t$, it seems like I should be setting $r=t$.

At a more technical level, glm.nb from the MASS package seems to fit $r$ not the dispersion parameter and I don't see an obvious parameter to change this.

Any insight at the theoretical or technical level would be appreciated.


1 Answer 1


Several stochastic processes lead to marginal counts having a Negative Binomial (NB) distribution and can therefore be called NB processes. Among them, the NB Lévy Process is of special interest since increments (counts) over non-overlapping time intervals are independent, a property shared with the Poisson Process, a Gamma process and the Wiener Process. The count $N_t$ on an interval of length $t$ has the NB distribution $$ N_t \sim \textrm{NB}(r,\,p), \quad r = \gamma t $$ so the process depends on the two parameters $\gamma >0$ (with the dimension of an inverse time) and the probability $p$ ($0 < p < 1$). The expectation is proportional to the interval length, and so is its variance $$ \mathbb{E}(N_t) = \gamma t \, (1-p)/p \qquad \textrm{Var}(N_t) = \gamma t \, (1-p)/p^2. $$ The variance is greater than the mean (overdispersion), and the index of dispersion $\textrm{Var}(N_t)/\mathbb{E}(N_t) = 1/p$ does not depend on $t$. When $p$ is close to $1$ and $\gamma (1-p)$ is close to $\lambda >0$, the process behaves like a Poisson Process with rate $\lambda$. An explanation for overdispersion is that several events can happen at the same time, so a small interval can contain more than one event.

It is easy to fit such a process by Maximum Likelihood when the intervals have different lengths. In this case we face a NB regression with a link function differing from the default link in NB GLMs. A special likelihood maximisation is useful.

The article by T.J. Kozubowski and K. Podgorski provide theoretical results as well as an illustration.

Curiously enough, this process does not seem to be frequently used as such by statisticians.

  • $\begingroup$ Can you say what you mean by: "A special likelihood maximisation is useful."? $\endgroup$
    – AdamO
    Feb 9, 2018 at 16:04
  • $\begingroup$ For a given value of $\gamma$, the value $\widehat{p}(\gamma)$ of $p$ maximising the log-likelihood $\log L(\gamma, p)$ can be given in closed form. Thus we can maximise the concentrated (or profile) log-likelihood $\log L[\gamma, \widehat{p}(\gamma)]$, a function of the single parameter $\gamma$. Moreover the second order derivative of the concentrated log-likelihood can be obtained in closed form and it can proved that the profile log-likelihood is concave, so a global maximum exists. $\endgroup$
    – Yves
    Feb 11, 2018 at 16:36

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.