0

I need help in calculating the goodness of fit for a Poisson model. The online literature on it is quite scarce, so if you have any useful articles and links I would appreciate it very much.

Here is how the table looks like from SPSS. The outcome variable is a count of patents a company has.

model http://img827.imageshack.us/img827/1655/screenshot20130606at124.png

IvLi
  • 631
  • 1
  • 8
  • 12
  • What software are you using? – dimitriy Jun 05 '13 at 23:00
  • What is the response? Is it counts? What of? If the Poisson model is appropriate, the Deviance should be chi-square with the indicated df. 5873 is not a plausible value for a chi-square with 61 df. – Glen_b Jun 05 '13 at 23:29
  • I am using SPSS and yes the response in is a count variable (noumber of patents a company has). – IvLi Jun 06 '13 at 03:33
  • 1
    Three possibilities: (1) the data is hugely overdispersed relative to the Poisson; (2) your model is not picking up the a major cause of variation in the mean, so your conditional distribution is a mixture of very different nearly-Poissons, making it *look* overdispersed compared to a better model; or (3) it's actually overdispersed relative to the Poisson *and* you're not picking up the mean well, making it worse – Glen_b Jun 06 '13 at 03:41
  • 1
    @Scortchi I am not convinced that's necessarily a duplicate. Here we have (at least potentially) a Poisson regression model - what may effectively be a mix of Poissons with different parameters; a univariate goodness of fit test would no use for that. The OP should clarify, but I don't want to close too hastily. – Glen_b Jun 21 '13 at 00:45
  • 2
    @Glen_b You're quite right - it looks a lot more like [this](http://stats.stackexchange.com/questions/37732/when-someone-says-residual-deviance-df-should-1-for-a-poisson-model-how-appro). – Scortchi - Reinstate Monica Jun 21 '13 at 08:18
  • 1
    @Scortchi ... yes, I think that's basically a duplicate. – Glen_b Jun 21 '13 at 08:27
  • 1
    The image link is now dead, making this question impossible to answer. – Sycorax Aug 30 '18 at 16:06
  • Other possible targets: https://stats.stackexchange.com/questions/263084/quasi-poisson-glm-cant-make-sense-of-the-p-values-for-goodness-of-fit, https://stats.stackexchange.com/questions/266688/poisson-regression-residuals-diagnostic, https://stats.stackexchange.com/questions/25068/interpreting-plot-of-residuals-vs-fitted-values-from-poisson-regression – kjetil b halvorsen Aug 31 '18 at 08:31

0 Answers0