Cross Validated is a question and answer site for people interested in statistics, machine learning, data analysis, data mining, and data visualization. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

How does one approach the problem of modeling a "birth-death process" where the arrivals are dependent on the current state in the following way: if the population is above a certain point, the probability of an arrival decreases.

Basically, I'm interested in complicating (slightly) an existing model of "births" that just has Poisson distributed arrivals and thinking of adding in the idea that there's a "saturation point" above which arrivals are less likely (waiting until the population drops back below the point).

Should I be reading about nth-order Markov processes? Or should I be looking at queueing theory?

share|improve this question
Your example still sounds like a first order Markov chain. – vqv Dec 20 '10 at 4:20
I think my use of the word "state" probably does imply that -- although I did want to think about a "continuous" situation, where the "birth rate" of the process declines gradually (perhaps precipitousness) as the population grows. – ariddell Jan 6 '11 at 0:29
up vote 3 down vote accepted

If this is for a particular application, I would first consider the question: "do you need to fit such a complex model?" which is code for, what other, more simple methods have you tried prior to this one? Have you looked at a plot of the "births & deaths" data that you have?

I would advise that you look up some of the work on "state space" theory, and kalman filtering as a starter. It sounds like you basically have an autoregressive process that isn't a simple random walk. These can usually be dealt with via the Kalman filter (as long as you are willing to assume normality of the errors).

A simple way (I think) is to consider a "regression" of the births/deaths against the current state. You could just use OLS as a simple start (to figure out what's going on), but this ignores that the errors from the regression are correlated (and not independent as in OLS). This will have the effect of OLS standard errors being either too small or too big, depending on the direction of the correlation (the OLS estimates are still "unbiased")

share|improve this answer
This is helpful. I've added more details in the question description. Any thoughts? – ariddell Dec 23 '10 at 15:49

Frankly, what you have is still a Markov Chain, as vqv has very rightly pointed out. (PS I tried to add this as a comment but where is the button??)

share|improve this answer
Unfortunately, you have to have at least 50 reputation in order to post comments. The purpose of that restriction has never been clear to me, as commenting seems like a pretty low-cost way for new users to start participating. – Matt Parker Dec 23 '10 at 17:56
I was somewhat frustrated with that restriction myself. I remember having something along the lines of 49 reputation and wanted to make a comment and I couldn't but I definitely knew it wasn't appropriate to be a standalone answer... – Dason Dec 24 '10 at 5:35

You might find Scott and Smyth '03 of interest:

They discuss markov modulated poisson processes, which varies the rate parameter based on which markov state it is in. You could have two states, one below and one above the given population of interest.

share|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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