Closed Spatial-R closed 2 years ago
The question of "weighting" the data is equivalent to the question of what form the measurement model takes. The larger the variance of the measurement model, the smaller the weight that is associated with the data. In other words, if you don't believe that the low values in the data are precise measurements, this should be reflected in your measurement model as a large variance, which in turn will cause these data to be down-weighted in the likelihood calculation. If you do believe that the measurements are precise, then why would you want to ignore them?
So the first question is: Is the evident lack of fit a problem with the data or with the model (or both)? In particular, what is the precision of the various measurements? Is this properly reflected in your measurement model? If it is, then the next question is, Why is my model capable of fitting the low periods, or the high periods, but not both?
Also, may I ask why you are using such an out-of-date version of pomp?
Thanks!I will try the newest version of pomp and reset the measurement model.
I will close this issue for now. Feel free to re-open if more discussion is warranted.
I wanted to ask for your suggestion as to what is the most effective way of fitting the surevillance data. Currently, I want to use the mathematical model to fit the long-term time-series data. However, the vast majority of time points are in the non-epidemic period, when the number of monitored cases is at a low value. Thus, the model seems to focus on fitting these low values, and the fit of the epidemic surveillance data is poorer.
Therefore, is it possible to give greater weight to the surveillance data in the epidemic period when calculating the likelihood values in pomp package? Because I would prefer that the model fit these epidemic surveillance data well, even if the non-epidemic fits are mediocre.
Following is my codes and the link to the data. I appreciate your insightful comments.
The version of pomp package is 2.8.