Timeline for Forecasting count data after fitting a Poisson regression model
Current License: CC BY-SA 4.0
6 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Dec 13, 2021 at 6:33 | comment | added | Ben | Other than the cited reference, I'm not really across that literature. I recommend reading some textbooks on GLMs, with particular focus on the sections discussing prediction intervals. | |
| Dec 13, 2021 at 3:48 | comment | added | user551504 | Hi Ben, could you please provide some references for the more sophisticated approaches? As far as I know the multivariate t is only reasonable when there are known correlations, and I know nothing (but am very interested) in the case of unknown correlations. | |
| Nov 24, 2021 at 7:00 | comment | added | Ben | As to difference with PIs, I would think that if you just plug in the estimator to the distribution then the interval will be too narrow, since it won't take account of the uncertainty in the coefficient. As to how much difference it makes, it depends on the sample size and corresponding level of uncertainty in the coefficient estimator --- if there is low variance in the coefficient estimator then failure to account for the uncertainty in the coefficients might be okay. | |
| Nov 24, 2021 at 6:59 | comment | added | Ben |
The ciTools package has some functionality to compute prediction intervals using bootstrap resampling. I haven't looked into the underlying method, so can't really comment on its quality or give a recommendation. There is a discussion of prediction intervals that uses this package. here.
|
|
| Nov 24, 2021 at 6:42 | comment | added | Stephan Kolassa | Nice, thank you for adding this answer. Can you recommend any packages in R to do this (possibly also for a negbin)? Also, do you know whether accounting for the parameter estimation uncertainty makes a difference in practice for the PIs over just using point estimates and plugging them into a Poisson distribution (which is much faster)? | |
| Nov 24, 2021 at 1:53 | history | answered | Ben | CC BY-SA 4.0 |